Special Issue - Department of Ecology and Evolutionary Biology [PDF]

Special Issue Integrating Phylogenies into Community Ecology1 The organisms that live together in a community do so both because they are present in the larger regional pool and because they have characteristics that permit their existence at that locality and their coexistence with other species in the community. Neither a species’ regional presence nor its characters can be fully understood without taking the species’ history into account. That history is contingent on chance events and on deterministic interactions with other species in historical communities. As this historical approach gains favor in ecology, and as our understanding of the tree of life expands, ecologists and systematists are increasingly working together. However, this new partnership often requires synthesizing ideas across disciplines. Our goal with this Special Issue is to explore the practical interchange of concepts between evolutionary biology and community ecology, highlighting studies that both use phylogenetic information and consider the community context of individual organisms, and that represent a range of disciplines, from microbiology and parasitology to ornithology. Several common threads weave through the papers. The first concerns the importance and definition of the local community itself. As one steps back and takes a historical and biogeographic view of a species, averaging over variation in local community composition across its range, local interspecific interactions appear less influential for the species’ evolution. Ricklefs, in considering the causes of variation in the emergent property of community species richness, goes so far as to say that ‘‘ecologists [must] abandon the idea of the local community.’’ At the same time, however, there is abundant evidence that inter-individual interactions do influence which particular taxa co-occur (e.g., Webb et al.), alter their ranges, and under certain circumstances, lead to evolutionary adaptation that reduces negative interspecific interaction. Separating the effects of local processes on regional patterns and regional processes on local patterns will always be hard. However, with large and numerous samples we can come to understand variation in community structure over wide areas. Lovette and Hochachka draw on the vast Breeding Bird Survey data set to examine both local and regional composition in warbler communities, and Kembel and Hubbell examine phylogenetic community structure of trees at varying scales within the 50-ha BCI Forest Dynamics Plot. As well as drawing spatial boundaries around communities, we are forced to define their taxonomic bounds. Cavender-Bares et al. demonstrate how increasing the ‘‘phylogenetic scale’’ of communities influences our understanding of their phylogenetic structure. Brooks et al. and Weiblen et al. address the complex question of the phylogenetic structure of compound communities, with platyhelminth parasites of anurans, and insect herbivores of plants, respectively. The second key thread dealt with by many authors is, ‘‘what exactly are ecological characters, and how do they evolve?’’ For ecologists, it is obvious to ask how an organism’s niche has evolved and to treat it as a character to be reconstructed on a phylogeny. However, systematists often argue that because the habitats and realized niches that we can observe are influenced by interspecific interactions and community composition, they are not actually evolvable entities. Instead we should decompose overall niches into directly heritable, morphological, and physiological characters. For example, Agrawal and Fishbein show how defensive syndromes involve combinations of numerous characters of plants, and Fine et al. show that defensive traits evolve in a trade-off with growth. Different components of the overall niche may also be subject to different ecological interactions: an organism might occupy a habitat that conforms to a niche on one environmental axis, while competition within a habitat might lead to resource partitioning on another axis. Silvertown et al. 1 Reprints of this 166-page Special Issue are available for $20.00 each. Prepayment is required. Order reprints from the Ecological Society of America, Attention: Reprint Department, 1707 H Street, N.W., Suite 400, Washington, DC 20006. Costs for this Special Issue were defrayed by NSF grant DEB-0408432 to C. O. Webb and J. B. Losos. NCEAS funded a workshop on ‘‘Phylogenies and Community Ecology’’ in 2002, which a number of authors in this Special Issue attended.

S1

and Ackerly et al. attempt to detect which characters diverged earlier during the course of plant diversification. Knouft et al. use the new methods of GIS-based niche modeling to examine the evolution of multidimensional niches in a clade of lizards, by associating specimens’ collection locations with layers of environmental variation in space. Inferring evolutionary process from the pattern of character evolution we observe requires models of character change under known mechanisms. Unbounded continuous characters evolving under Brownian drift (slow relative to speciation rate) will tend to show conservatism (closely related species tend to be similar). Stabilizing selection will tend to increase that conservatism, while, conversely, a reduced number of potential character states, a long time between speciation events, and divergent selection will tend to increase convergence. Using various methods, the authors found generally more conservatism in ecological characters (e.g., Ackerly et al., Silvertown et al.) than convergence (e.g., Knouft et al.), with some studies finding no clear relationship of traits with relatedness (e.g., Agrawal and Fishbein). Overall, the results are consistent with the action of divergent selection in some systems, overlaid on a null expectation of some level of phylogenetic conservatism in all systems. The third major thread in these papers is closely linked to the second: ‘‘what is the phylogenetic relatedness of co-occurring taxa in communities, and what does this tell us about community assembly?’’ Authors used a number of methods to combine community lists with phylogenies to answer this question. Cavender-Bares et al. and Lovette and Hochachka correlated taxon cooccurrence rates (across many samples) with phylogenetic distance (e.g., with Mantel tests). HornerDevine and Bohannan, Kembel and Hubbell, Weiblen et al., Silvertown et al., and Webb et al. tested the observed distribution of intra-sample, inter-taxon phylogenetic distances against null models of community assembly (so raising many of the perennial questions of community null models). Several authors pointed out that methods using inter-taxon phylogenetic distance rather than ancestral state reconstruction are less prone to bias introduced by the sampling of taxa that are very widely distributed on the tree of life. Several authors found that taxa in their communities were more closely related than expected, indicating a common role of habitat choice and evolutionarily conserved characters (e.g., Horner-Devine and Bohannan, Weiblen et al.). Cavender-Bares et al. and Kembel and Hubbell found cases where taxa were less closely related than expected. Taking the results of trait evolution and community phylogenetic structure together, the importance of community interactions does appear to be diminished, and a long-term regional view of taxa more justified. However, Lovette and Hochachka found both conservatism of habitat specialization in warblers at regional scales and evidence for competitive ‘‘repulsion’’ among close relatives at local sites. Because different combinations of trait evolution pattern and ecological interaction (competition vs. habitat choice) can give similar community phylogenetic structure, trait data, community data, and phylogenies are all needed for a full understanding of the evolution and assembly of communities. The discussion of the nature of communities and niche evolution is decades old; note that a similarly titled Special Feature appeared in this journal ten years ago. However, the vast number of species that have been sequenced, and for which phylogenies have been generated, means that ecologists can now often infer the phylogenetic relationships of their taxa from publications and databases without further systematics work. We hope that this Special Issue will inspire readers to take advantage of these opportunities, to phrase their questions in a more evolutionary way, and as Westoby anticipates, to participate in the new Natural History. —CAMPBELL O. WEBB Guest Editor Harvard University —JONATHAN B. LOSOS Guest Editor Washington University —ANURAG A. AGRAWAL Special Features Editor Cornell University Key words: character evolution; community structure; competition; habitat filtering; multidimensional niche; null models. Ó 2006 by the Ecological Society of America

Ecology, 87(7) Supplement, 2006, pp. S3–S13 Ó 2006 by the Ecological Society of America

EVOLUTIONARY DIVERSIFICATION AND THE ORIGIN OF THE DIVERSITY–ENVIRONMENT RELATIONSHIP ROBERT E. RICKLEFS1 Department of Biology, University of Missouri–St. Louis, 8001 Natural Bridge Road, St. Louis, Missouri 63121-4499 USA

Abstract. Global patterns in species richness have resisted explanation since they first caught the attention of ecologists in the 1960s. The failure of ecology to fully integrate the diversity issue into its core of accepted wisdom derives from an inappropriate concept of community and the rejection of history and region as formative contexts for ecological systems. Traditionally, ecologists have held that the pervasive relationship between species richness and conditions of the physical environment reflects the influence of environment on the ability of populations to coexist locally. However, many ecologists now recognize that this relationship can also develop historically from the evolutionary diversification of lineages within and between ecological zones. To assess the relative roles of local ecological constraint vs. regional and historical unfolding of diversity–environment relationships, we must abandon localized concepts of the community and adopt historical (particularly phylogenetic) and geographic methods to evaluate the evolution of diversity within large regions and its influence on diversity at local scales. This integrated perspective opens new research directions for ecologists to explore the formation of species, adaptive diversification, and the adjustment of ecological distributions of species on regional scales. Key words: adaptation; community; diversity gradient; history; local determinism; phylogeny; saturation; speciation; species packing; species richness.

which the relative contributions of local and regional processes can be weighed, and sketch out some implications of diversity patterns for our concepts of ecological systems. All biological systems have general properties, which are governed by the pervasive influences of thermodynamics, evolutionary adaptation, and other ubiquitous processes. They also have special properties, which reflect the unique history and present-day circumstances of every species and location. These special properties form the foundation of systematics and biogeography. Ecologists traditionally have been concerned with general properties of systems arising from universal processes with deterministic outcomes. Ideas about the distributions of organisms, population regulation, population interactions, community succession, and ecosystem energetics have stemmed from such thinking (Kingsland 1985, McIntosh 1985, Brown et al. 2004). Global patterns of species richness occupy an uncertain position between the special and the general. Early treatments regarded such patterns as the special outcome of history and, therefore, outside the realm of ecology (Wallace 1878, Matthew 1915, Willis 1922, Fischer 1960). Tropical diversity was thought to reflect the greater age, area, and climatic stability of equatorial, compared to temperate and boreal, environments, which allowed ample time and opportunity for the evolution of diverse forms of life (Dobzhansky 1950, Pianka 1966).

INTRODUCTION The number of species at all spatial scales varies widely over the surface of the earth. Ecologists have sought explanations for patterns of diversity in the varied expression of ecological processes under different local physical conditions of the environment. However, patterns in diversity also can be explained plausibly as the outcome of large-scale processes that control the production and extinction of species within regions, which in turn influence the number of species in local assemblages. These alternatives are not mutually exclusive, and ecologists are beginning to join features of both into a unified theory for the origin and maintenance of patterns of diversity. The task is more difficult than it would seem, because it requires the integration of dissimilar properties of biological systems: deterministic properties that are molded by local ecological processes, and the outcomes of evolutionary and biogeographic processes that depend on unique features of the history and physiography of regions (Ricklefs 2004, 2005a). Because local determinism has been so prominent in ecological thinking, I first consider how local determinism has achieved its current status, then briefly review some of the conflicting data that weaken this central ecological paradigm, suggest ways in Manuscript received 17 January 2005; revised 22 August 2005; accepted 7 September 2005. Corresponding Editor (ad hoc): C. O. Webb. For reprints of this Special Issue, see footnote 1, p. S1. 1 E-mail: [email protected]

THE RISE

OF

LOCAL DETERMINISM

With the development of community ecology as a mature discipline in the 1960s, ecologists began to regard S3

S4

ROBERT E. RICKLEFS

diversity as a general feature of biological systems regulated locally by processes with deterministic outcomes (MacArthur 1965, 1972). Moreover, because local population interactions achieve equilibrium within tens of generations, they were thought to be fast enough to override more ponderous regional and evolutionary processes (Ricklefs 1989). These considerations led to a theoretical construct wherein population interactions limited membership in a community to species that are ecologically compatible (MacArthur 1968, May 1975, Case 1990, Morton et al. 1996). Accordingly, differences in the number of species between communities reflected the different outcomes of species interactions under particular environmental conditions. This implied that diversity would be correlated with variation in the physical environment, which has been borne out, to greater or lesser degree, by empirical studies (e.g., Mittelbach et al. 2001, Hawkins et al. 2003b). Ironically, the rise of local determinism in community ecology occurred almost simultaneously with two other developments that contradicted its basic tenets. The first was the acceptance by most ecologists of an ‘‘open’’ community structure (Gleason 1926), which is to say that communities lack boundaries and that locally coexisting species have more or less independent distributions over spatial and ecological gradients within regions (Whittaker 1967). The second was the colonization–extinction steady state in MacArthur and Wilson’s equilibrium theory of island biogeography (MacArthur and Wilson 1967), in which local (island) diversity is responsive to an external driver (colonization). Referring to the number of species on islands of different size close to the source of colonization, MacArthur and Wilson (1963) introduced the idea of saturation. In their meaning of the word, saturation was not a local property, but reflected the influence of the size of an area on sampling properties and rates of extinction of populations. Abbott and Grant (1976) also used this concept of saturation to represent diversity within a particular continental source area for colonists, against which the diversity of islands could be measured. These empirical concepts were transferred to local communities through theoretical analyses of limiting similarity among species (e.g., MacArthur and Levins 1967), which led to the idea of species packing and its corollary that diversity (species coexistence) was constrained by the capacity of the local environment to support interacting species. During the following decade, ecologists conducted comparative studies of communities largely in the context of species packing and the partitioning of niche space, presuming that environment constrained diversity (e.g., Pianka 1973, Cody 1974). In fact, community theory sets no upper ‘‘saturated’’ limit to diversity in a particular environment. Referring to a community of competing species, MacArthur (1970) emphasized that ‘‘. . . in a constant environment there is almost no limit to the number of species which can improve the fit and hence be packed in . . . .’’ However,

Ecology Special Issue

the flexible filling of niche space was too complicated to be handled by theory. Eventually, local community ecology was insulated from such external drivers as colonization and regional species production. Many ecologists accepted local saturation, and they reconciled differences between local and regional diversity by differential turnover of species between habitats (beta diversity; Cody 1975). Thus, while recognizing the influence of large-scale processes on regional diversity, most ecologists still regarded patterns of local diversity as the consequence of ecological sorting of species available within a region (MacArthur 1965, Diamond 1975, Zobel 1997, Weiher and Keddy 1999). NEW CONCEPTS

OF

ECOLOGICAL COMMUNITIES

Recent excitement over Hubbell’s (2001) neutral theory of communities suggests that many ecologists were willing, after struggling for decades without resolving the diversity problem, to entertain theories that discount ecological interactions and niche specialization completely. Hubbell’s theory is historical and geographical (geohistorical). Diversity depends solely on the area of suitable habitat within a region (that is, proportional to the number of individuals in the metacommunity) and the rate of species production, assuming that systems have had sufficient time to attain equilibrium. Species extinction is purely stochastic, depending only on the size of a population. The theory includes ecology only in the sense that the total number of individuals within the regional metacommunity is fixed. Thus, all individuals compete on a homogeneous ecological landscape, but on equal footing irrespective of their identity. As species richness within the metacommunity increases, average population size decreases, as does the time to extinction. This balances species production to arrive at the equilibrium diversity. Although a purely neutral theory does not withstand scrutiny on a number of counts (Chave 2004), particularly because neutral drift is too slow to account for the development of ecological patterns (Ricklefs 2006a), a purely local, ecological theory of diversity also cannot account for certain empirical patterns. These include correlations between local and regional diversity (Cornell and Lawton 1992, Srivastava 1999), which contradict the idea of local saturation of species richness, and incomplete convergence in diversity between areas of similar environment in different regions with independently evolved biotas (e.g., Latham and Ricklefs 1993a, Qian and Ricklefs 2000). A geohistorical, evolutionary alternative to both local determinism and neutral theory is the generation of gradients of species richness through diversification within ancestral ecological zones of origin combined with occasional adaptive shifts associated with invasion of new ecological zones (Farrell et al. 1992, Latham and Ricklefs 1993b, Wiens and Donoghue 2004). The idea presumes that species are best adapted to the conditions in the ecological zone of origin of their lineage;

July 2006

THE DIVERSITY–ENVIRONMENTAL RELATIONSHIP

transitions to other ecological zones require evolutionary change. This evolutionary model establishes gradients of diversity, producing greater species richness in environments that are older, more widespread, or less stressful. The idea originated with biogeographers (e.g., Darlington 1957, Axelrod 1966), but was slow to be adopted by ecologists. Terborgh (1973) was the first to articulate a comprehensive theory of community diversity that explicitly included historical and geographic influences on local species richness. Applied, for example, to the species richness of forest trees, the latitudinal gradient could be explained by the origination and diversification of most flowering plant lineages in the extensive tropics of the early Tertiary (Burnham and Johnson 2004, Davis et al. 2005), followed by diversification of some lineages across adaptive barriers into high-latitude frost zones (Latham and Ricklefs 1993b). This process is shown diagrammatically in Fig. 1. A second evolutionary alternative to local determination is that rates of evolutionary proliferation of species (speciation minus extinction) are higher in regions or ecological zones of high diversity (Farrell and Mitter 1993, Jablonski 1993, Cardillo 1999). This mechanism could be called a general process if certain ecological conditions, such as temperature, promoted or retarded proliferation (Rohde 1992, Allen et al. 2002), or a special process if proliferation depended on unique physiography or geographic configurations of regions or if extinction depended on unique history. For example, Qian and Ricklefs (2000) speculated that the high diversity of plants in temperate eastern Asia could be related, in part, to the complex physiography of the region, which is both mountainous and features land masses (China mainland, Korean peninsula, and Japan) that have alternately been connected and separated as sea levels have risen and fallen during the Tertiary. Molecular phylogenetic analysis of plant clades distributed in eastern Asia and North America supports both an older age (Asian taxa paraphyletic to North American taxa) and more rapid diversification in eastern Asia as underlying causes of diversity differences (Xiang et al. 2004). Extinction also can play a role (Vermeij 1987, Jablonski 1991). For example, Latham and Ricklefs (1993a) and Svenning (2003) concluded from comparisons of fossil and modern taxa that the impoverished tree flora of Europe resulted from differential extinction of species within predominantly tropical and subtropical groups caused by cooling climates and glaciation during the late Tertiary. THE INFLUENCE

OF

LOCAL

AND

REGIONAL PROCESSES

If one accepts the premise that local diversity represents a balance between the constraining influence of local population interactions and the augmenting influence of regional species production, then one can hope to estimate only the relative balance of these factors in determining patterns of species richness. Moreover, general and special processes often make

S5

FIG. 1. Evolutionary diversification of a clade within its ecological zone of origin, with occasional adaptive shifts (stars) to different ecological zones. The adaptive shifts might be difficult and occur infrequently, potentially establishing a gradient in contemporary diversity that favors the ecological zone of origin. Modified from Ricklefs (2005a); see also Wiens and Donoghue (2004).

the same predictions, which cannot then be used to distinguish between them. For example, if species richness were determined locally by ecological constraints on species packing, the phylogenetic history of the species in regions having different diversity might also be consistent with a diversification constraint. A simple classification of the mechanisms that influence diversity is presented in Table 1. These are divided into local and regional/historical processes, and the latter are further subdivided into the influences of the ecological zone of origin, of environment on the rate of diversification, and of diversity itself in promoting or retarding further diversification. Community saturation Taken to its extreme, local determinism predicts that communities are saturated with species and that species richness is directly related to factors in the physical environment that determine the number of species that can coexist locally (Table 1: Panel A). Community saturation implies an upper limit to the number of species in a local assemblage, regardless of the diversity within the surrounding region (Cody 1966). In principle, this hypothesis can be confirmed in comparisons of species assemblages in the same habitats between areas having different regional diversity, i.e., the principle of community convergence (e.g., Orians and Paine 1983, Schluter and Ricklefs 1993), particularly when local diversity levels off with increasing regional diversity. Terborgh and Faaborg (1980) were the first to apply a saturation test explicitly, with a positive result. Most subsequent studies have failed to find evidence for a hard upper limit to species number, but Srivastava (1999) called attention to the shortcomings of such analyses, including inconsistent definitions of local and

S6

ROBERT E. RICKLEFS

Ecology Special Issue

TABLE 1. A classification of influences on large-scale patterns in local diversity. Mechanism A) Local determinism a) Limiting similarity and saturation b) Diversity increases resistance to invasion B) Regional/historical processes a) Ecological zone of origin i) Age and area ii) Adaptive diversification b) Net rate of diversification i) Physiography and history promote speciation ii) Climate change and catastrophes cause extinction iii) Diversity promotes or retards diversification

Comments Explanations based on local factors require explicit models of how the physical environment influences coexistence. This extreme form of local determinism predicts community convergence, which is rejected in many comparative studies. Ecological compression and release demonstrate the influence of population interactions on local community membership. These models predict diversity within larger regions, but also provide external drivers for local species richness. Phylogenetic analysis permits the reconstruction of ancestral ecological positions within the environmental landscape. Phylogenetic analysis provides an estimate of relative age; area effects are apparent in contemporary biotas, but the relative roles of species production and extinction should be distinguished. Zones of origin can be identified by phylogenetic analysis, and these should support the highest species richness regardless of the depth of the clade stem. Differences in rates of diversification can be seen in lineage-through-time plots and inferred, to some degree, from genetic distances between sister taxa. Assuming allopatric speciation, regional analyses of geographic heterogeneity at appropriate scales would be informative, as would studies of incipient species formation (e.g., genetic differentiation of populations) This can be judged primarily through analysis of fossil material and the geological record, hence applications are limited to groups with good fossil records. Diversity might be self-accelerating or self-limiting. The pattern is accessible through analysis of lineages through time and analysis of biological and physical aspects of niche structure in contemporary biotas.

regional scales and lack of independence between regions. Ricklefs (2000) emphasized the balance between local and regional factors, pointing out that increased ‘‘regional’’ diversity among West Indian island avifaunas was accommodated equally by increases in local species richness and turnover of species between habitats. Consistent with this ecological flexibility, introduced species apparently have not caused extinctions on islands through competitive exclusion (Sax et al. 2002; but see Gurevitch and Padilla (2004)), and establishment of new colonists in the avifauna of the Lesser Antilles appears to be independent of local diversity (Table 1, [Panel B, row b(iii)]; Ricklefs and Bermingham 2001). Over longer periods observable in the marine fossil record of the Paleozoic Era, for example, diversity does appear to have constrained diversification (e.g., Foote 2000). Local constraint Local constraint predicts that local species richness should be strongly correlated with environmental conditions, irrespective of the mechanisms regulating coexistence. Under local determinism, diversity patterns must ultimately be related to physical aspects of the environment (e.g., Ricklefs 1977, Currie et al. 2004). Many studies have demonstrated strong correlations between local species richness, using various definitions of the scale of ‘‘local’’ (Rahbek and Graves 2001, Willis and Whittaker 2002), and climate (Currie 1991, O’Brien 1998, Hawkins et al. 2003b) or other variables, such as habitat structure (Cody 1974) and productivity (Rosenzweig 1995, Mittelbach et al. 2001). Coefficients of

determination commonly fall in the range of 60–70%, or more (Hawkins et al. 2003a). However, such correlations might also be predicted by evolutionary theories of species richness patterns, where diversity reflects either environmental history combined with evolutionary inertia, or the influence of environment on net species proliferation. Unique history and geography Local constraints predict convergence of local community properties in similar environments regardless of the evolutionary and geographic history of a region, within which species richness might vary widely. Conversely, differences in species richness in the same habitat between regions suggest that special factors influence local assemblages. Such differences have been noted frequently in community comparisons. Among the most conspicuous of these diversity anomalies in plants, for example, are the difference in species richness between mangrove communities in the Indo-West Pacific and the Atlantic/Caribbean regions (Duke 1992, Ricklefs and Latham 1993), the depauperate tree flora of Europe compared to eastern Asia (Latham and Ricklefs 1993a), and the greater diversity within disjunct genera of angiosperms in eastern Asia compared to eastern North America (Qian and Ricklefs 1999). Such diversity anomalies have been explained by differences between regions in (a) the frequency of adaptive shifts of new lineages to the stressful mangrove environment, (b) extinction in the case of the European tree flora, and (c) a combination of species production, species invasion from the tropics, and late-Tertiary extinction in the case

July 2006

THE DIVERSITY–ENVIRONMENTAL RELATIONSHIP

of the disjunct temperate floras of Asia and North America. Although special explanations for diversity anomalies make good sense, they are suggested by the data themselves rather than emerging from the use of data to test contrasting predictions. Special explanations are difficult to place in an experimental or hypothesistesting framework (Francis and Currie 1998), and the anomalies themselves might reflect unmeasured differences in local environments between regions (Pianka 1975, Morton 1993, Ricklefs et al. 2004). Can diversity anomalies be used other than to provide anecdotal support for the idea of special influences on diversity? Some special geographic circumstances are repeated themes in many regions of the world. Among terrestrial environments, for example, mountainous areas coincide with diversity hotspots (e.g., Barthlott et al. 1996, Orme et al. 2005), providing an opportunity to test the consistency of some special relationships. Even where these relationships are unique, geographic/historical models suggest mechanisms, such as enhancement of allopatric speciation or restriction of evolutionary diversification, which can be assessed in multiple, independently evolving lineages. Thus, although empirical observations might not provide statistical support for a particular historical scenario, one might obtain statistically valid appraisals of mechanisms postulated to be at work within a unique geohistorical framework. Speciation rate The role of speciation rate in creating patterns of diversity (Table 1, [Panel B, row b]) has received considerable attention, particularly with respect to identifying the influence of ‘‘key innovations’’ on diversification (Slowinski and Guyer 1993, Heard and Hauser 1995, Bennett and Owens 2002). Among the more successful of these tests have been the association of rapid diversification in insects with the switch to herbivorous diets and the association of rapid diversification in plants with the evolution of certain defenses against herbivory (Farrell and Mitter 1994). The physiography of particular regions, such as eastern Asia, might promote species production. General characteristics of the environment, such as the benign nature of tropical climates, could also influence speciation rate (Dobzhansky 1950, Schemske 2002). Several authors have suggested that thermal energy can accelerate speciation (e.g., Rohde 1992, Allen et al. 2002). Attempts to test hypotheses on the rate of speciation have focused on comparisons of sister (same-age) clades in tropical and temperate regions (Farrell and Mitter 1993, Cardillo 1999, Cardillo et al. 2005, Ricklefs 2005b, 2006b), but these have been relatively inconclusive because suitable data are difficult to assemble. Such comparisons depend on well-supported phylogenies. Because these are rapidly increasing in number, testing the effect of various environmental or regional factors on rates of species production will become more commonplace and will show whether and how speci-

S7

ation rate and the ages of lineages influence diversity within regions (e.g., Xiang et al. 2004). Ecological zone of origin Every lineage of organism has an ecological zone of origin (Table 1 [Panel B, row a]), which, under some circumstances, can be identified by tracing character (i.e., ecological zone) evolution on a phylogenetic tree (e.g., Schluter et al. 1997). This is similar to inferring the geographic area of origin, which often can be ambiguous (Ronquist 1997, Brown and Lomolino 1998:Chapter 12, Sanmartin et al. 2001). When an entire clade is restricted to a particular ecological zone, one can infer that it originated within that zone, even though this parsimonious conclusion might be incorrect (Davis et al. 2005). When a clade is distributed over several ecological zones, identifying the origin in the absence of fossil evidence depends on the paraphyletic distribution of zones over the clade. Diversification across adaptive barriers Some of the strongest diversity gradients are clearly associated with particular environmental stressors (Table 1 [Panel B, row a(ii)]), including high salt and substrate anoxia in the case of mangrove vegetation and freezing in the case of temperate vegetation. The roughly 20 present-day lineages of mangrove trees and shrubs were independently derived from terrestrial lineages of plants over a period of more than 60 million years (Ricklefs and Latham 1993, Ellison et al. 1999). This suggests that the evolutionary transition from terrestrial to mangrove environments is difficult and that this adaptive barrier is crossed infrequently. The same is likely true of evolutionary transitions from mesic tropical to temperate habitats (Terborgh 1973). Phylogenetic analysis reveals that the lineages of trees in north temperate latitudes are mostly nested within clades having deep tropical origins (Ricklefs 2005a). Indeed, more than half the families of flowering plants are restricted to tropical latitudes and evidently have not been able to cross the adaptive barrier into temperate latitudes (Ricklefs and Renner 1994). During most of the early evolution and diversification of the flowering plants, the planet’s environments were predominately tropical (Behrensmeyer et al. 1992, Graham 1999). Models of the establishment of regional patterns of species richness, based on origins vs. rates of proliferation, can be distinguished by phylogenetic analysis (compare Figs. 1 and 2). Whether either of these evolutionary models were to shape gradients in local diversity, in contrast to merely reflecting the gradient in numbers of contemporary coexisting species permitted by ecological interactions, is more difficult to resolve. That is, all species have an evolutionary history, including relationships to other lineages, and one can reconstruct phylogenies for the members of a local community regardless of the mechanisms that control local diversity. A role for adaptive barriers can be

S8

ROBERT E. RICKLEFS

Ecology Special Issue

FIG. 2. When species proliferate more rapidly in a novel environment than in the ecological zone of origin, diversity need not coincide with the ancestral environment. In this example, the zone of origin can be identified through phylogenetic analysis, because species within the ancestral zone are paraphyletic with respect to species in the zone of highest diversity.

inferred from the partitioning of evolutionary lineages between ecological zones (Webb 2000, Webb et al. 2002). If species can cross between these zones easily, ecological zone itself would be labile, varying at a shallow depth within a phylogeny, as in the case of oak species across a moisture gradient in Florida (CavenderBares et al. 2004) or evolutionary transitions to chaparral environments in western North America (Ackerly 2004). However, if these adaptive shifts were difficult, as in the case of entering the mangrove environment, ecological zone would appear to be a conservative trait. For example, the four genera and perhaps 17 species of mangrove Rhizophoraceae represent a single lineage that entered the mangrove environment early in the evolution of the family and diversified there without any lineages crossing back to the terrestrial ecological zone (Schwarzbach and Ricklefs 2000). PROSPECTS Ecological regions of high diversity and ecological regions of origin likely coincide, although rates of lineage proliferation also appear to differ between regions and might produce a contradictory pattern (Fig. 2). In general, ecological and evolutionary models for the origin and maintenance of diversity gradients cannot be distinguished by examining patterns of diversity. Because of this, ecologists must expand their inquiries to include evolutionary hypotheses (Webb et al. 2002, Ackerly 2004), and they must develop mechanistic, testable, ecological models that connect diversity to the physical environment (Currie et al. 2004). Special explanations for species richness require that special processes have strong enough influence to leave a distinctive imprint on ecological pattern. Evolution and

adaptation are slow compared to local ecological dynamics. However, when the interactions between species play out within regions rather than local communities, general processes can adjust ecological and geographic distributions of species so that all populations come into demographic balance within the region (Ricklefs 2004). Ironically, this deterministic outcome is independent of special aspects of regions and lineages, including the number of species that have been generated within a region. As more species are added, ecological distributions are compressed, beta diversity (species turnover with respect to ecology and distance) increases, and equilibrium is maintained (Terborgh 1973). This concept requires that ecologists consider populations as regional ecological entities made integral by the movements of individuals (Lennon et al. 1997, Ricklefs 2004, Case et al. 2005, Holt et al. 2005). Evolutionary and geographic history—special components of ecological systems—can be revealed through phylogenetic analysis, among other approaches, which provides insight into the development of diversity patterns and unique aspects of biological communities in different regions. Speciation and extinction also reflect the intimate connections between ecology, geography, and evolution. Although many areas of ecological inquiry are independent of evolution and history, interactions between species played out within a large regional context bring ecological and evolutionary processes onto a continuum of scale that intimately links local ecology, history, and geography. Within this regional context, ecologists can characterize species distributions, including sampling on a variety of scales that comprise local individual movements, population interactions, habitat heterogeneity, and regional pro-

July 2006

THE DIVERSITY–ENVIRONMENTAL RELATIONSHIP

cesses of population subdivision and species formation (Rahbek 1997, Rahbek and Graves 2001). Islands, particularly archipelagoes, will continue to be important laboratories for studying mechanisms of species origination (Grant 1998, Mayr and Diamond 2001) and the response of ecological distributions to the pressure of regional diversity (Cox and Ricklefs 1977). Analyses of the niche structure of assemblages, particularly by defining the multivariate dimensionality of niche axes from ecological, morphological, and life history data (Ricklefs and Miles 1994), will assess the contribution of niche space to diversification and the complementary contribution of niche diversification to the evolutionary development of species assemblages. Most of these techniques are familiar to ecologists; new directions for the future will arise from their application in a novel framework. Within this framework, I can make several specific recommendations. Of course, this list reflects my own perspective and would be broadened considerably by others. First, it is important to understand that the scale of analysis must be appropriate to the scale of the process in both time and space. Thus, if one accepts the premise that regional and historical processes can influence the development of local assemblages of species, then analyses of patterns of local diversity must include regional distribution and both environmental and evolutionary history. These might be accommodated, in part, in the following ways. 1) It is essential that ecologists abandon the idea of the local community. Interactions occur between populations over large regions, and we should therefore characterize the distributions of species on the geographic and ecological gradients over which they interact (Case et al. 2005). What ecologists have traditionally regarded as a ‘‘community’’ comprises the populations that co-occur together at a particular point in these regional continua of conditions. Whittaker (1967) and others used this approach in gradient analysis of vegetation. Sampling is conducted in plots along environmental gradients, and species populations can be characterized by their mean position, dispersion, and modal abundance along these gradients. When gradients are made comparable between regions, differences in the overlap and partitioning of species along these gradients can be compared as an approach to understanding how population responses accommodate variation in region diversity (e.g., Nekola and White 1999, Qian et al. 2005). Although this will be straightforward in principle, it may be difficult in practice because ecologists typically consider too few large-scale gradients within regions to separate hundreds or thousands of species. Ordination and canonical correspondence analysis typically explain a modest proportion of the variance in species distributions, leaving the rest to biological complexities in the environment and chance, both of which have traditionally been ignored by ecologists (cf. Condit et al. 2002).

S9

2) The history of distribution of clades over environmental gradients involves adaptive changes and may constrain the diversity of species. Accordingly, it is important to examine patterns of species richness across environmental gradients in a phylogenetic framework. The general context for this is presented in Figs. 1 and 2. Ecologists should focus on strong barriers (e.g., frost tolerance, salinity gradients) as well as less imposing temperature and moisture gradients within regions (Ackerly 2004, Cavender-Bares et al. 2004). Such studies should seek factors that influence multiple lineages and that presumably impose the strongest constraints. These analyses of habitat shifts should be based on wellsupported phylogenies, preferably in which branch points can be dated by fossils or calibrations developed for other lineages (e.g., Kishino et al. 2001, Renner and Meyer 2001) to relate adaptive shifts to climate or geographic change, and they must also use proper estimates of ancestral environments (Schluter et al. 1997). Although this approach is primarily descriptive rather than hypothesis testing, it can provide support for the prediction from adaptive constraint that ancestral lineages might change abruptly at a deep level across strong environmental gradients (Westoby 2006). 3) Building upon the analysis of population distributions within regions and phylogenetic analysis across environmental gradients, it seems reasonable to examine local assemblages and distributions of species across particularly important gradients of diversity. Among these, one that has been important in development of ideas about species richness patterns is the difference in tree species diversity between tropical and temperate environments. The transitions between these ends of an environmental continuum for broad-leaved trees are located in areas (eastern Mexico, southern China) for which few floristic data exist, but where we might expect the greatest information concerning diversity gradients. Although altitudinal gradients do not have the same seasonal dimension as latitudinal gradients, they are also informative to the extent that species turn over in parallel fashion. 4) Several studies have examined the depth of the clades that comprise a community as a way to assess the relationship between the ages of lineages and their diversity. Simply put, if the net rate of diversification (speciation minus extinction) is homogeneous, older lineages will leave more descendants. That is, the logarithm of the number of species is a function of the net rate of diversification and time. Ricklefs and Schluter (1993) showed that the clades that comprised avian assemblages in a tropical locality in Panama were roughly twice as old as the clades that made up an assemblage in Illinois, based on field data in Karr (Karr 1971) and lineage ages from Sibley and Ahlquist (1990). The implication is that time is an important factor in both regional and local diversity. Hawkins (2005) has provided a more comprehensive analysis that shows the same pattern in the birds of Australia, and Ricklefs

S10

ROBERT E. RICKLEFS

(2005a) confirmed the conventional botanical wisdom (Judd et al. 1994) that temperate trees belong to clades that are nested within deeper clades of tropical lineages. This topology is illustrated in Fig. 1, where the species in the more diverse ecological region belong to clades that have diversified within that region for longer periods. 5) A second facet of the historical dimension is the rate of diversification of species. All other things being equal, diversity increases as a function of the rate of diversification and time (Ricklefs 2006b). Thus, faster diversification leads to more species over a given period. Although few studies have made this comparison to date, and these have been inconclusive, the increasing availability of phylogenetic reconstructions will change this prospect rapidly. Appropriate comparisons include sister clades that occur in different regions or environments (Farrell and Mitter 1993, Cardillo 1999), which, by definition, are of equal age, or samples of clades from different regions or environments for which age and number of taxa are known (e.g., Cardillo et al. 2005, Ricklefs 2005b). Another approach is the lineagethrough-time plot (Nee et al. 1992, Harvey et al. 1994), whose slope can be used, with well-sampled phylogenetic reconstructions of large clades, to estimate both speciation and extinction rates (e.g., Ricklefs 2006a). Another index to the rate of species proliferation is the genetic divergence between sister species, which in general decreases as the rate of speciation increases and extinction decreases. In all aspects of analyses that address species properties, results depend on the way in which species are defined, which should be as uniform as possible across comparisons. 6) Speciation as a process contributing to regional diversity can be examined phylogeographically by searching for incipient speciation among geographically separated populations. Incipient species might be recognized by large genetic differences between populations (Avise 2000), but, in any event, these can be compared easily among regions in terms of the number and geographic distribution of populations or subpopulations at a particular phylogenetic depth (e.g., Brumfield and Capparella 1996, Bates et al. 1999, Weir and Schluter 2004). This type of analysis could identify regions of rapid population differentiation that might lead to high rates of species production. Diversity is built up locally only when geographically isolated, evolutionarily independent populations achieve secondary sympatry. Thus, comparison of sister taxa in allopatry and sympatry (assuming an allopatric model of species formation), would permit an analysis of the time required for the local accumulation of species and whether the ecological differentiation that permits coexistence evolves in allopatry, or primarily by divergent selection after the initiation of secondary sympatry. Island groups, with their discrete geographic organization, provide ideal opportunities for exploring these issues (Grant 1998, Ricklefs and Bermingham 2001).

Ecology Special Issue

7) Finally, extinction can play an important role in the establishment of patterns of diversity. Where an adequate fossil record exists, one has a direct estimate of diversity, at some resolution, through time. The extinction of many clades of animals at the Cretaceous– Tertiary boundary, as well as the disappearance of many plants from Europe during the late-Tertiary period of climate cooling (Sauer 1988, Latham and Ricklefs 1993b, Svenning 2003), provide instructive examples of the importance of this factor. A large literature in paleontology addresses rate of extinction and its relationship to diversification, environmental changes, catastrophic events, and changes in the configurations of landmasses and ocean basins (Jablonski 1989, 1991, 1993, Vermeij 1991, Jackson et al. 1993). Unfortunately, the fossil record is rarely resolved to the taxonomic level of species and hardly exists for many groups. This should not prevent those of us interested in living forms from examining what is known of the fossil history of a group. At some level of resolution, this will provide at least an envelope around possible historical scenarios. Where fossil data are not available, information can be extracted from models of lineage diversification (e.g., Magallo´n and Sanderson 2001, Ricklefs 2006a) and extinction inferred, for example, from gaps in the distribution of species through archipelagoes (Ricklefs and Cox 1972, Ricklefs and Bermingham 1999). The better we develop the space and time contexts of ecological systems, the more we shall appreciate the variety of factors that have influenced their composition and diversity. We should view populations as evolved entities, with unique histories and adaptations, that are distributed geographically according to their tolerance of ecological conditions and interactions with other populations, and, within the limitations of dispersal, over regional landscapes. Inevitable consequences of these processes are the co-occurrence of a unique assemblage of species at any particular point and geographic patterns in the number of species observed over many such points over the surface of the earth. ACKNOWLEDGMENTS I am grateful to Cam Webb for discussion, encouragement, and helpful comments on the manuscript. The National Geographic Society, Smithsonian Institution, National Science Foundation, and the University of Missouri Board of Curators have supported research related to this paper. LITERATURE CITED Abbott, I., and P. R. Grant. 1976. Nonequilibrial bird faunas on islands. American Naturalist 110:507–528. Ackerly, D. D. 2004. Adaptation, niche conservatism, and convergence: comparative studies of leaf evolution in the California chaparral. American Naturalist 163:654–671. Allen, A. P., J. H. Brown, and J. F. Gillooly. 2002. Global biodiversity, biochemical kinetics, and the energetic-equivalence rule. Science 297:1545–1548. Avise, J. 2000. Phylogeography. Harvard University Press, Cambridge, Massachusetts, USA. Axelrod, D. I. 1966. Origin of deciduous and evergreen habits in temperate forests. Evolution 20:1–15.

July 2006

THE DIVERSITY–ENVIRONMENTAL RELATIONSHIP

Barthlott, W., W. Lauer, and A. Placke. 1996. Global distribution of species diversity in vascular plants: towards a world map of phytodiversity. Erdkunde 50:317–326. Bates, J. M., S. J. Hackett, and J. M. Goerck. 1999. High levels of mitochondrial DNA differentiation in two lineages of antbirds (Drymophila and Hypocnemis). Auk 116:1093–1106. Behrensmeyer, A. K., J. D. Damuth, W. A. DiMichele, R. Potts, H.-D. Sues, and S. L. Wing. 1992. Terrestrial ecosystems through time. Evolutionary paleoecology of terrestrial plants and animals. University of Chicago Press, Chicago, Illinois, USA. Bennett, P. M., and I. P. F. Owens. 2002. Evolutionary ecology of birds. Life histories, mating systems, and extinction. Oxford University Press, Oxford, UK. Brown, J. H., J. F. Gillooly, A. P. Allen, V. M. Savage, and G. B. West. 2004. Toward a metabolic theory of ecology. Ecology 85:1771–1789. Brown, J. H., and M. V. Lomolino. 1998. Biogeography. Second edition. Sinauer Associates, Sunderland, Massachusetts, USA. Brumfield, R. T., and A. P. Capparella. 1996. Historical diversification of birds in northwestern South America: a molecular perspective on the role of vicariant events. Evolution 50:1607–1624. Burnham, R. J., and K. R. Johnson. 2004. South American palaeobotany and the origins of neotropical rainforests. Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences 359:1595–1610. Cardillo, M. 1999. Latitude and rates of diversification in birds and butterflies. Proceedings of the Royal Society of London B 266:1221–1225. Cardillo, M., C. D. L. Orme, and I. P. F. Owens. 2005. Testing for latitudinal bias in diversification rates: an example using New World birds. Ecology 86:2278–2287. Case, T. J. 1990. Invasion resistance arises in strongly interacting species-rich model competition communities. Proceedings of the National Academy of Sciences (USA) 87:9610–9614. Case, T. J., R. D. Holt, M. A. McPeek, and T. H. Keitt. 2005. The community context of species’ borders: ecological and evolutionary perspectives. Oikos 108:28–46. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Chave, J. 2004. Neutral theory and community ecology. Ecology Letters 7:241–253. Cody, M. L. 1966. The consistency of intra- and intercontinental grassland bird counts. American Naturalist 100:371– 376. Cody, M. L. 1974. Competition and the structure of bird communities. Princeton University Press, Princeton, New Jersey, USA. Cody, M. L. 1975. Towards a theory of continental species diversities. Pages 214–257 in M. L. Cody and J. M. Diamond, editors. Ecology and evolution of communities. Harvard University Press, Cambridge, Massachusetts, USA. Condit, R., et al. 2002. Beta-diversity in tropical forest trees. Science 295:666–669. Cornell, H. V., and J. H. Lawton. 1992. Species interactions, local and regional processes, and limits to the richness of ecological communities: a theoretical perspective. Journal of Animal Ecology 61:1–12. Cox, G. W., and R. E. Ricklefs. 1977. Species diversity, ecological release, and community structuring in Caribbean land bird faunas. Oikos 29:60–66. Currie, D. J. 1991. Energy and large scale patterns of animal and plant species richness. American Naturalist 137:27–49. Currie, D. J., G. G. Mittelbach, H. V. Cornell, R. Field, J. F. Guegan, B. A. Hawkins, D. M. Kaufman, J. T. Kerr, T. Oberdorff, E. O’Brien, and J. R. G. Turner. 2004. Predictions

S11

and tests of climate-based hypotheses of broad-scale variation in taxonomic richness. Ecology Letters 7:1121–1134. Darlington, P. J., Jr 1957. Zoogeography. The geographical distribution of animals. Wiley, New York, New York, USA. Davis, C. C., C. O. Webb, K. J. Wurdack, C. A. Jaramillo, and M. J. Donoghue. 2005. Explosive radiation of Malpighiales supports a mid-Cretaceous origin of modern tropical rain forests. American Naturalist 165:E36–E65. Diamond, J. M. 1975. Assembly of species communities. Pages 342–444 in M. L. Cody and J. M. Diamond, editors. Ecology and evolution of communities. Harvard University Press, Cambridge, Massachusetts, USA. Dobzhansky, T. 1950. Evolution in the tropics. American Scientist 38:209–221. Duke, N. C. 1992. Mangrove floristics and biogeography. Pages 63–100 in A. I. Robertson and D. M. Alongi, editors. Tropical mangrove ecosystems. American Geophysical Union, Washington, D.C., USA. Ellison, A. M., E. J. Farnsworth, and R. E. Merkt. 1999. Origins of mangrove ecosystems and the mangrove biodiversity anomaly. Global Ecology and Biogeography Letters 8: 95–115. Farrell, B. D., and C. Mitter. 1993. Phylogenetic determinants of insect/plant community diversity. Pages 253–266 in R. E. Ricklefs and D. Schluter, editors. Species diversity in ecological communities. Historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Farrell, B. D., and C. Mitter. 1994. Adaptive radiation in insects and plants: time and opportunity. American Zoologist 34:57–69. Farrell, B., C. Mitter, and D. J. Futuyma. 1992. Diversification at the insect–plant interface. BioScience 42:34–42. Fischer, A. G. 1960. Latitudinal variation in organic diversity. Evolution 14:64–81. Foote, M. 2000. Origination and extinction components of taxonomic diversity: Paleozoic and post-Paleozoic dynamics. Paleobiology 26:578–605. Francis, A. P., and D. J. Currie. 1998. Global patterns of tree species richness in moist forest: another look. Oikos 81:598– 602. Gleason, H. A. 1926. The individualistic concept of the plant association. Torrey Botanical Club Bulletin 53:7–26. Graham, A. 1999. Late Cretaceous and Cenozoic history of North American vegetation. Oxford University Press, New York, New York, USA. Grant, P. R., editor. 1998. Evolution on islands. Oxford University Press, Oxford, UK. Gurevitch, J., and D. K. Padilla. 2004. Are invasive species a major cause of extinctions? Trends in Ecology and Evolution 19:470–474. Harvey, P. H., R. M. May, and S. Nee. 1994. Phylogenies without fossils. Evolution 48:523–529. Hawkins, B. A., J. A. F. Diniz, and S. A. Soeller. 2005. Water links the historical and contemporary components of the Australian bird diversity gradient. Journal of Biogeography 32:1035–1042. Hawkins, B. A., R. Field, H. V. Cornell, D. J. Currie, J. F. Guegan, D. M. Kaufman, J. T. Kerr, G. G. Mittelbach, T. Oberdorff, E. M. O’Brien, E. E. Porter, and J. R. G. Turner. 2003a. Energy, water, and broad-scale geographic patterns of species richness. Ecology 84:3105–3117. Hawkins, B. A., E. E. Porter, and J. A. F. Diniz. 2003b. Productivity and history as predictors of the latitudinal diversity gradient of terrestrial birds. Ecology 84:1608–1623. Heard, S. B., and D. L. Hauser. 1995. Key evolutionary innovations and their ecological mechanisms. Historical Biology 10:151–173. Holt, R. D., T. H. Keitt, M. A. Lewis, B. A. Maurer, and M. L. Taper. 2005. Theoretical models of species’ borders: single species approaches. Oikos 108:18–27.

S12

ROBERT E. RICKLEFS

Hubbell, S. P. 2001. The unified neutral theory of biodiversity and biogeography. Princeton University Press, Princeton, New Jersey, USA. Jablonski, D. 1989. The biology of mass extinction: a palaeontological view. Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences 325:357–368. Jablonski, D. 1991. Extinctions: a paleontological perspective. Science 253:754–757. Jablonski, D. 1993. The tropics as a source of evolutionary novelty through geological time. Nature 364:142–144. Jackson, J. B. C., P. Jung, A. G. Coates, and L. S. Collins. 1993. Diversity and extinction of tropical American mollusks and emergence of the Isthmus of Panama. Science 260:1624– 1626. Judd, W. S., R. W. Sanders, and M. J. Donoghue. 1994. Angiosperm family pairs: preliminary phylogenetic analyses. Harvard Papers in Botany 5:1–51. Karr, J. R. 1971. Structure of avian communities in selected Panama and Illinois habitats. Ecological Monographs 41: 207–233. Kingsland, S. E. 1985. Modeling nature. Episodes in the history of population ecology. University of Chicago Press, Chicago, Illinois, USA. Kishino, H., J. L. Thorne, and W. J. Bruno. 2001. Performance of a divergence time estimation method under a probabilistic model of rate evolution. Molecular Biology and Evolution 18:352–361. Latham, R. E., and R. E. Ricklefs. 1993a. Continental comparisons of temperate-zone tree species diversity. Pages 294–314 in R. E. Ricklefs and D. Schluter, editors. Species diversity: historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Latham, R. E., and R. E. Ricklefs. 1993b. Global patterns of tree species richness in moist forests: energy-diversity theory does not account for variation in species richness. Oikos 67: 325–333. Lennon, J. J., J. R. G. Turner, and D. Connell. 1997. A metapopulation model of species boundaries. Oikos 78:486– 502. MacArthur, R. H. 1965. Patterns of species diversity. Biological Reviews 40:510–533. MacArthur, R. H. 1968. The theory of the niche. Pages 150–176 in R. Lewontin, editor. Population biology and evolution. Syracuse University Press, Syracuse, New York, USA. MacArthur, R. H. 1970. Species packing and competitive equilibrium for many species. Theoretical Population Biology 1:1–11. MacArthur, R. H. 1972. Geographical ecology. Patterns in the distribution of species. Harper and Row, New York, New York, USA. MacArthur, R. H., and R. Levins. 1967. The limiting similarity, convergence, and divergence of coexisting species. American Naturalist 101:377–385. MacArthur, R. H., and E. O. Wilson. 1963. An equilibrium theory of insular biogeography. Evolution 17:373–387. MacArthur, R. H., and E. O. Wilson. 1967. The theory of island biogeography. Princeton University Press, Princeton, New Jersey, USA. Magallo´n, S., and M. J. Sanderson. 2001. Absolute diversification rates in angiosperm clades. Evolution 55:1762–1780. Matthew, W. D. 1915. Climate and evolution. Annals of the New York Academy of Sciences 24:171–318. May, R. M. 1975. Stability and complexity in model ecosystems. Second edition. Princeton University Press, Princeton, New Jersey, USA. Mayr, E., and J. Diamond. 2001. The birds of northern Melanesia: speciation, ecology and biogeography. Oxford University Press, Oxford, UK, and New York, New York, USA. McIntosh, R. P. 1985. The background of ecology. Concept and theory. Cambridge University Press, Cambridge, UK.

Ecology Special Issue

Mittelbach, G. G., C. F. Steiner, S. M. Scheiner, K. L. Gross, H. L. Reynolds, R. B. Waide, M. R. Willig, S. I. Dodson, and L. Gough. 2001. What is the observed relationship between species richness and productivity? Ecology 82:2381–2396. Morton, R. D., R. Law, S. L. Pimm, and J. A. Drake. 1996. On models for assembling ecological communities. Oikos 75: 493–499. Morton, S. R. 1993. Determinants of diversity in animal communities of arid Australia. Pages 159–169 in R. E. Ricklefs and D. Schluter, editors. Species diversity in ecological communities. Historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Nee, S., A. O. Mooers, and P. H. Harvey. 1992. Tempo and mode of evolution revealed from molecular phylogenies. Proceedings of the National Academy of Sciences (USA) 89: 8322–8326. Nekola, J. C., and P. S. White. 1999. The distance decay of similarity in biogeography and ecology. Journal of Biogeography 26:867–878. O’Brien, E. M. 1998. Water–energy dynamics, climate, and prediction of woody plant species richness—an interim general model. Journal of Biogeography 25:379–398. Orians, G. H., and R. T. Paine. 1983. Convergent evolution at the community level. Pages 431–458 in D. J. Futuyma and M. Slatkin, editors. Coevolution. Sinauer, Sunderland, Massachusetts, USA. Orme, C. D., et al. 2005. Global hotspots of species richness are not congruent with endemism or threat. Nature 436:1016– 1019. Pianka, E. R. 1966. Latitudinal gradients in species diversity: a review of concepts. American Naturalist 100:33–46. Pianka, E. R. 1973. The structure of lizard communities. Annual Review of Ecology and Systematics 4:53–74. Pianka, E. R. 1975. Niche relations of desert lizards. Pages 292– 314 in M. L. Cody and J. M. Diamond, editors. Ecology and evolution of communities. Harvard University Press, Cambridge, Massachusetts, USA. Qian, H., and R. E. Ricklefs. 1999. A comparison of the taxonomic richness of vascular plants in China and the United States. American Naturalist 154:160–181. Qian, H., and R. E. Ricklefs. 2000. Large-scale processes and the Asian bias in species diversity of temperate plants. Nature 407:180–182. Qian, H., R. E. Ricklefs, and P. S. White. 2005. Beta diversity of angiosperms in temperate floras of eastern Asia and eastern North America. Ecology Letters 8:15–22. Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in Neotropical birds. American Naturalist 149:715–730. Rahbek, C., and G. R. Graves. 2001. Multiscale assessment of patterns of avian species richness. Proceedings of the National Academy of Sciences (USA) 98:4534–4539. Renner, S. S., and K. Meyer. 2001. Melastomeae come full circle: biogeographic reconstruction and molecular clock dating. Evolution 55:1315–1324. Ricklefs, R. E. 1977. Environmental heterogeneity and plant species diversity: a hypothesis. American Naturalist 111:376– 381. Ricklefs, R. E. 1989. Speciation and diversity: integration of local and regional processes. Pages 599–622 in D. Otte and J. A. Endler, editors. Speciation and its consequences. Sinauer Associates, Sunderland, Massachusetts, USA. Ricklefs, R. E. 2000. The relationship between local and regional species richness in birds of the Caribbean Basin. Journal of Animal Ecology 69:1111–1116. Ricklefs, R. E. 2004. A comprehensive framework for global patterns in biodiversity. Ecology Letters 7:1–15. Ricklefs, R. E. 2005a. Historical and ecological dimensions of global patterns in plant diversity. Biologiske Skrifter (Royal Danish Academy of Sciences and Letters) 55:583–603.

July 2006

THE DIVERSITY–ENVIRONMENTAL RELATIONSHIP

Ricklefs, R. E. 2005b. Phylogenetic perspectives on patterns of regional and local species richness. Pages 16–40 in E. Bermingham, C. Dick, and C. Moritz, editors. Tropical rainforests. Past, present, and future. University of Chicago Press, Chicago, Illinois, USA. Ricklefs, R. E. 2006a. The unified neutral theory of biodiversity: Do the numbers add up? Ecology 87:1424–1431. Ricklefs, R. E. 2006b. Global variation in the diversification rate of passerine birds. Ecology 87, in press. Ricklefs, R. E., and E. Bermingham. 1999. Taxon cycles in the Lesser Antillean avifauna. Ostrich 70:49–59. Ricklefs, R. E., and E. Bermingham. 2001. Nonequilibrium diversity dynamics of the Lesser Antillean avifauna. Science 294:1522–1524. Ricklefs, R. E., and G. W. Cox. 1972. Taxon cycles in the West Indian avifauna. American Naturalist 106:195–219. Ricklefs, R. E., and R. E. Latham. 1993. Global patterns of diversity in mangrove floras. Pages 215–229 in R. E. Ricklefs and D. Schluter, editors. Species diversity in ecological communities. Historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Ricklefs, R. E., and D. B. Miles. 1994. Ecological and evolutionary inferences from morphology: an ecological perspective. Pages 13–41 in P. C. Wainwright and S. M. Reilly, editors. Ecological morphology: integrative organismal biology. University of Chicago Press, Chicago, Illinois, USA. Ricklefs, R. E., H. Qian, and P. S. White. 2004. The region effect on mesoscale plant species richness between eastern Asia and eastern North America. Ecography 27:129–136. Ricklefs, R. E., and S. S. Renner. 1994. Species richness within families of flowering plants. Evolution 48:1619–1636. Ricklefs, R. E., and D. Schluter. 1993. Species diversity: regional and historical influences. Pages 350–363 in R. E. Ricklefs and D. Schluter, editors. Species diversity in ecological communities. Historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Rohde, K. 1992. Latitudinal gradients in species diversity: the search for the primary cause. Oikos 65:514–527. Ronquist, F. 1997. Dispersal–vicariance analysis: a new approach to the quantification of historical biogeography. Systematic Biology 46:195–203. Rosenzweig, M. L. 1995. Species diversity in space and time. Cambridge University Press, Cambridge, UK. Sanmartin, I., H. Enghoff, and F. Ronquist. 2001. Patterns of animal dispersal, vicariance and diversification in the Holarctic. Biological Journal of the Linnean Society 73: 345–390. Sauer, J. D. 1988. Plant migration. The dynamics of geographic patterning in seed plant species. University of California Press, Berkeley, California, USA. Sax, D. F., S. D. Gaines, and J. H. Brown. 2002. Species invasions exceed extinctions on islands worldwide: a comparative study of plants and birds. American Naturalist 160: 766–783. Schemske, D. 2002. Tropical diversity: patterns and processes. Pages 163–173 in R. Chazdon and T. Whitmore, editors. Ecological and evolutionary perspectives on the origins of tropical diversity: key papers and commentaries. University of Chicago Press, Chicago, Illinois, USA. Schluter, D., T. Price, A. O. Mooers, and D. Ludwig. 1997. Likelihood of ancestor states in adaptive radiation. Evolution 51:1699–1711. Schluter, D., and R. E. Ricklefs. 1993. Convergence and the regional component of species diversity. Pages 230–242 in R. E. Ricklefs and D. Schluter, editors. Species diversity in

S13

ecological communities. Historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Schwarzbach, A. E., and R. E. Ricklefs. 2000. Systematic affinities of Rhizophoraceae and Anisophyllaceae, and intergeneric relationships with the Rhizophoraceae, based on chloroplast DNA, nuclear ribosomal DNA, and morphology. American Journal of Botany 87:547–564. Sibley, C. G., and J. E. Ahlquist. 1990. Phylogeny and classification of the birds of the world. Yale University Press, New Haven, Connecticut, USA. Slowinski, J. B., and C. Guyer. 1993. Testing whether certain traits have caused amplified diversification—an improved method based on a model of random speciation and extinction. American Naturalist 142:1019–1024. Srivastava, D. 1999. Using local–regional richness plots to test for species saturation: pitfalls and potentials. Journal of Animal Ecology 68:1–16. Svenning, J. C. 2003. Deterministic Plio–Pleistocene extinctions in the European cool-temperate tree flora. Ecology Letters 6: 646–653. Terborgh, J. 1973. On the notion of favorableness in plant ecology. American Naturalist 107:481–501. Terborgh, J. W., and J. Faaborg. 1980. Saturation of bird assemblages in the West Indies. American Naturalist 116:178–195. Vermeij, G. J. 1987. Evolution and escalation. An ecological history of life. Princeton University Press, Princeton, New Jersey, USA. Vermeij, G. J. 1991. When biotas meet: understanding biotic interchange. Science 253:1099–1104. Wallace, A. R. 1878. Tropical nature and other essays. Macmillan, New York, New York, USA, and London, UK. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Weiher, E., and P. A. Keddy, editors. 1999. The search for assembly rules in ecological communities. Cambridge University Press, Cambridge, UK. Weir, J. T., and D. Schluter. 2004. Ice sheets promote speciation in boreal birds. Proceedings of the Royal Society of London B 271:1881–1887. Westoby, M. 2006. Phylogenetic ecology at world scale, a new fusion between ecology and evolution. Ecology 87:S163– S165. Whittaker, R. H. 1967. Gradient analysis of vegetation. Biological Reviews of the Cambridge Philosophical Society 42:207–264. Wiens, J. J., and M. J. Donoghue. 2004. Historical biogeography, ecology and species richness. Trends in Ecology and Evolution 19:639–644. Willis, J. C. 1922. Age and area. A study in geographical distribution and origin of species. Cambridge University Press, Cambridge, UK. Willis, K. J., and R. J. Whittaker. 2002. Species diversity—scale matters. Science 295:1245–1248. Xiang, Q. Y. J., W. H. Zhang, R. E. Ricklefs, H. Qian, Z. D. Chen, J. Wen, and J. H. Li. 2004. Regional differences in rates of plant speciation and molecular evolution: a comparison between eastern Asia and eastern North America. Evolution 58:2175–2184. Zobel, M. 1997. The relative role of species pools in determining plant species richness: an alternative explanation of species coexistence. Trends in Ecology and Evolution 12: 266–269.

Ecology, 87(7) Supplement, 2006, pp. S14–S28 Ó 2006 by the Ecological Society of America

SIMULTANEOUS EFFECTS OF PHYLOGENETIC NICHE CONSERVATISM AND COMPETITION ON AVIAN COMMUNITY STRUCTURE IRBY J. LOVETTE1

WESLEY M. HOCHACHKA

AND

Laboratory of Ornithology, Cornell University, 159 Sapsucker Woods Road, Ithaca, New York 14850 USA

Abstract. We currently have only a partial understanding of how phylogenetic relationships relate to patterns of community structure, in part because, for most groups of organisms, we do not know the extent to which ecological similarity results from common ancestry. Associations between phylogenetic relatedness and local community structure are particularly interesting for groups in which many species that span a gradient of phylogenetic divergence occur in potential sympatry. We explored the relationship between evolutionary relatedness and current species co-occurrence among the North American wood-warblers (Aves: Parulidae), a group of songbirds known both for its species diversity and for exhibiting high levels of sympatry at breeding sites. Species co-occurrences were derived from North American Breeding Bird Survey transects comprising 160 000 census points distributed across North America. The nested point-within-transect structure of this survey provides an unusual opportunity to remove larger-scale geographical effects on local community composition and thereby consider patterns of co-occurrence only among regionally sympatric pairs of species. We indexed evolutionary relatedness among all pairs of taxa by genetic distances based on long mitochondrial DNA protein-coding sequences. Most regionally sympatric taxon pairs rarely co-occur at local sites, and the most closely related never exhibit high local cooccurrences, as predicted if past or present competitive effects are strongest for these recently separated lineages. Quantile regression shows that, for a subset of taxa, local co-occurrence does increase with time since common ancestry, and that this apparent relaxation of competitive exclusion is strongest for distantly related species that have differentiated in fundamental ecological and behavioral traits, such as terrestrial vs. arboreal foraging. Comparisons against a null model of species co-occurrence further demonstrate that these patterns occur against a background of phylogenetic niche conservatism: across all phylogenetic distances, sympatric species co-occurred at higher rates than expected by chance, a pattern that might stem from a tendency by these species to show conservatism in their selection of similar general habitat types. Considered in concert, these analyses suggest the simultaneous mediation of local community structure by the ecological similarity of closely related species and by trait divergence among a subset of more distant lineages. Key words: community structure; competition; conservation; niche phylogeny; Parulidae; phylogenetic; wood-warbler.

INTRODUCTION The connection between phylogenetic similarity and community composition is an important but largely unresolved issue in evolutionary ecology. Because closely related species will usually share many ecological traits via common ancestry, evolutionary relationships are likely to be associated with patterns of community assembly (Darwin 1859, Lack 1971, Ricklefs and Schluter 1993), but these phylogenetic effects might influence species coexistence in several ways. The general observation that closely related organisms tend to occupy similar habitats and often use similar environmental resources long predates the founding of ecology as a science, and Manuscript received 28 January 2005; revised 6 September 2005; accepted 7 September 2005; final version received 5 October 2005. Corresponding Editor (ad hoc): J. B. Losos. For reprints of this Special Issue, see footnote 1, p. S1. 1 E-mail: [email protected]

the extent to which this ecological stasis results from the inheritance of niche-related traits from a common ancestor is termed ‘‘phylogenetic niche conservation’’ (e.g., Wiens and Graham 2005). Research on phylogenetic niche conservatism has been recently invigorated by the growing availability of robust phylogenetic hypotheses based on data independent from ecological traits, but few generalizable trends are apparent across studies. The present literature suggests that phylogenetic niche conservatism is detectable in some but not all communities of organisms. Furthermore, its magnitude differs depending on the ecological traits under comparison and additional factors, such as the degree of ecological interaction among related species (Losos 1996, McPeek and Miller 1996, Peterson et al. 1999, Prinzing et al. 2001, Webb et al. 2002, Ackerly 2003, Losos et al. 2003, Anderson et al. 2004, Cavender-Bares et al. 2004, 2006, Wiens and Graham 2005, Kembel and Hubbel 2006, Knouft et al. 2006, Weiblen et al. 2006).

S14

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

Competition among related species is likely one of the principal forces causing the erosion of phylogenetic niche conservation: for species that interact ecologically, the initial ecological similarity that results from common ancestry will likely increase competition, potentially leading over time either to competitive exclusion or ecological differentiation (Webb et al. 2002). This process is most evident in adaptive radiations, in which groups of sympatric species have diversified extensively with concomitant differentiation in traits related to their use of ecological resources (Losos and Miles 2002). In such situations, divergent selection that results from competition could overwhelm any underlying retention of niche conservatism, as found by Losos et al. (2003) for a highly diverse community of Cuban Anolis lizards, within which phylogenetic relationships explained only a small amount of the among-species ecological variation. This is perhaps an extreme example of a general trend: most groups of closely related and ecologically interacting species will share some traits by common ancestry and differ in other traits, owing to divergent selection and other processes of evolutionary differentiation. An interesting intermediate situation involves groups in which the species composition of local communities varies substantially in time and space, as these taxa are individually involved in many potential interactions. In the absence of more stable associations between particular species that result in tight coevolutionary responses, species in highly variable communities might be more likely to assort themselves along relatively few axes of dissimilarity and therefore retain a larger signature of phylogenetic niche conservation. In addition, groups of organisms with high diversity, geographic distributions that have been labile through evolutionary time, and high individual dispersal are particularly appropriate models for exploring how phylogenetic relationships relate to the balance between niche conservation and ecological divergence, because these attributes will diminish the degree of biogeographic determinism in community composition by bringing together many pairs of taxa that would otherwise be allopatric. The diverse North American Parulidae fit these criteria, as nearly all wood-warbler species have persisted as independent lineages for at least several million years (see Plate 1; Lovette and Bermingham 1999, 2002). During their tenures in North America, the continental landscape has been transformed many times by glacial and climate cycles and consequent broad-scale shifts in habitat distributions (Graham 1999, Klicka and Zink 1999, Johnson and Cicero 2004, Weir and Schluter 2004, Lovette 2005). Nearly all of the approximately 6.0 3 108 individual North American wood-warblers (Rich et al. 2004) also migrate annually, and many studies of marked individuals (summarized in Poole and Gill [2004]) have shown that natal philopatry is low, at least at the fine local scale of intensively monitored study plots. Resulting from this natal dispersal and movements of adults among sites in

S15

different years (Holmes et al. 1996, Cilimburg et al. 2002), wood-warblers redistribute themselves across local habitats annually. This combination of large population sizes, repeated historical displacement of geographic ranges, and high individual dispersal means that most wood-warbler individuals breed at locations where other wood-warbler species are present. As documented by Robert MacArthur in his classic study of five sympatric ‘‘spruce woods’ warblers’’ (MacArthur 1958), at the local scale co-occurring songbird species show differentiation in their spatial foraging niches and other traits (e.g., Holmes et al. 1979, Morse 1989, Suhonen et al. 1994). In a MacArthurian framework, behavioral and ecological differentiation reduce interspecific competition and permit related species to co-occur. However, whether local niche differentiation explains broad spatial patterns of species co-occurrence is controversial because the species composition of wood-warbler communities varies across sites and years (Wiens 1989). A principal reason why the effect of evolutionary relatedness on community structure remains largely unexplored is that some fundamental tests of the relationship require information from different spatial scales (Anderson et al. 2004, Cavender-Bares et al. 2006): the competitive ecological interactions that can drive local patterns of community structure occur at a fine local scale best studied at points or in small study plots, yet these interactions must simultaneously be integrated across the much broader geographic distributions of potentially co-occurring taxa. Most previous studies of phylogenetic patterns in community structure have concentrated on interactions at one of these scales (e.g., Webb 2000, Silvertown et al. 2001, Losos et al. 2003, Gillespie 2004, Kembel and Hubbell 2006); however, unique avian censuses organized by the North American Breeding Bird Survey (BBS) program (Sauer et al. 2003) allow us to synthesize complementary information at both levels. Distinguishing between these alternatives requires separating the biogeographic or other historical factors that constrain the pool of candidate species from the ecological factors that cause competitive exclusion or facilitate co-occurrence (Ricklefs 1987, 2006). Underlying historical factors unrelated to direct competition include the fact that lineages that have speciated in allopatry might remain separated by the same physiographic features that caused their initial isolation and hence never come into ecological contact. Even pairs of species that do occur in sympatry will usually have geographic distributions that do not perfectly overlap. Current ecological interactions are thus limited to a part of each species’ overall range. The fact that each species has a unique geographic distribution that varies in overlap with those of other species is a general challenge in comparing ecological cooccurrences across many different pairs of species, as an absence of species’ overlap at a given point could result from being outside one species’ geographic range, or

S16

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

Ecology Special Issue

PLATE 1. The Prairie Warbler (Dendroica discolor) is one of 42 common species of North American wood-warblers that frequently breed in sympatry with closely related species. Photo credit: Marie P. Read.

more interestingly from ecological processes, such as differential habitat selection or competitive exclusion (Stone et al. 1996, Ackerly et al. 2006, Silvertown et al. 2006). The spatial structure of the BBS (i.e., many census points replicated along short transects) is particularly useful for our purposes because it allows us to remove the confounding effects of these larger scale differences in total geographic range from our measures of local co-occurrence. Here, we use BBS census information in combination with DNA-based genetic distances to assess whether phylogenetic relatedness among regionally sympatric wood-warblers is positively or negatively correlated with co-occurrence at the same points in space, the birds’ local breeding sites. A positive relationship (i.e., related species co-occurring at higher than expected rates) would suggest that local community composition is mediated in part by phylogenetic niche conservatism of habitat or habitat-linked aspects of niches, whereas a negative relationship would suggest that past or present

competition among closely related species limits their local ecological co-occurrence. In further analyses, we explore whether these patterns are influenced by a fundamental attribute of warbler behavior that leads to strong spatial niche segregation, the species-specific tendency to forage terrestrially or arboreally. We also compare patterns of co-occurrence at the local spatial scale defined by census points, with the regional-scale patterns defined by entire census transects. Unlike most previous studies of phylogenetic effects on community structure, the structure of our data sets requires that we compare matrices of genetic and ecological distances; our comparisons therefore do not require ancestral state reconstructions and thereby minimize the importance of the underlying assumption that we have identified the correct phylogenetic tree topology. Instead, we use a null-model approach to identify and explore associations between phylogenetic relatedness and overlap in species’ use of local breeding sites.

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

S17

FIG. 1. Locations of Breeding Bird Survey (BBS) transects. Colors indicate the total number of wood-warbler species on each transect. Inset histograms depict the distributions of species richness on the .4000 transects and at the .162 000 census points within transects.

METHODS Breeding bird survey co-occurrences Breeding Bird Survey (BBS) censuses are run during the breeding season by experienced volunteers who conduct standardized 3-minute counts of all birds heard or seen at 50 equally spaced locations along a 39.4-km (25-mile) transect (Sauer et al. 2003). Most transects are recensused annually, and data spanning 1997–2003 are available for the species detected at each census point, as well as the summed observations for each transect. The resulting 162 800 census points (Fig. 1) provide a robust representation of all but the most geographically restricted or peripheral North American wood-warblers. Here we define the wood-warblers detected at a single census point as co-occurring in local sympatry, as these individuals are certainly within auditory detection distance of one another and are likely to occupy at

least partially overlapping territories. To derive a measure of local sympatry, we considered only BBS transects on which both members of each pair of species were detected, as this physical proximity indicates that both species are members of the pool of taxa from which we drew the local community at each census point. For each point observation of species A, we determined the probability that wood-warbler species B was detected at that same census point, generating an index of dissimilarity in co-occurrence that spans 0 (complete cooccurrence) to 1 (no co-occurrence). To account for differences in prevalence among species, the matrix of co-occurrences also included the reciprocal calculation for each species pair (i.e., probability of detecting species A at points where species B was found). These values provide a continent-wide assessment of species cooccurrences measured only at the fine spatial scale most relevant to current ecological interactions among

S18

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

species. Scaling these values such that higher numbers indicate lower co-occurrence facilitates comparisons with the corresponding pairwise genetic distances, in which higher values indicate more ancient common ancestry. We have contrasted these patterns of local cooccurrence with patterns of regional co-occurrence, in order to determine whether patterns found in local cooccurrence are the result of phenomena that were not happening at larger scales. Regional co-occurrence was defined as existing when both members of a species pair were reported at least once on an individual 50-point transect in a given year, regardless of whether the two species were reported at the same point. Taken across all BBS transects, regional co-occurrence was given as a value ranging from 0 (species B is always reported on a transect that also reported species A) to 1 (species B is never reported on a transect that has species A). We base both local and regional co-occurrence indices on reported detections of warbler species, and each detection is a combined result of the presence of a bird species at a site and the human detection of the individual bird(s). Because the detection data result from both a biological process and a methodological filter (detection by observers), it is important to know whether the patterns found reflect the biological process and not a methodological artifact. To our knowledge, there has been no comprehensive attempt to estimate the detection probabilities of all of the warbler species examined in this study. While we cannot calculate these detection probabilities in a fully rigorous manner (e.g., MacKenzie et al. 2002) with the data at hand, we can calculate indices of detection probability that allow us to explore some potential biases. For each species of wood-warbler, we randomly selected a single point on each BBS transect that had reported the species on at least two points in at least one year; the actual median number of points at which a species was reported ranging from 2 (for one species only) to 11 across all species. The transect had to be censused for at least five years, and transects censused for longer had one or two years of data randomly discarded to leave exactly five years of available data. Using a standardized number of years’ data avoided inducing variation in the detection index due to differences among species in the number of years that transects were censused. The randomly selected census point within the transect had to have reported the target species in at least two separate years. Our rationale is that we are purposefully selecting transects in regions in which a given species was likely to present, as well as specific census points at which habitat was suitable and the species potentially present every year. The detection index for each species was the proportion of point-years, summed across all selected points, on which that species was detected. The values of the detection indices used here are the median values from 1000 iterations of the process of randomly selecting a single census point per transect. In calculating these indices of detection, we

Ecology Special Issue

assume that among-year detection probability contains information about within-year detection probability, the latter being the detection probability of interest. Values of the detection index spanned 0.43–0.62 for different species, with a median of 0.52 and a mean value of 0.53. Thus, values of the index were roughly symmetrically distributed, and most species detection rates were at intermediate values: the interquartile range spanned 0.50–0.56. While our index of detection probability is biased high relative to true detection probability, due to the use of data for which each species had a minimum probability of 0.4 of being detected at each individual site, this bias should be roughly equal for all species. Spatial foraging niche characters Co-occurring songbirds usually vary in traits that include foraging height, foraging substrate, and nest location (e.g., MacArthur 1958, Lack 1971, Robinson and Holmes 1982, Morse 1989, Richman and Price 1992, Martin 1998), suggesting that within-site niche partitioning is pervasive in these communities. There is also strong evidence that some closely related woodwarblers compete antagonistically in sympatry (Martin and Martin 2001). MacArthur’s 1958 paper spawned a large number of subsequent studies of wood-warbler niche segregation in traits such as foraging behavior, habitat strata, nest locations, and prey types, but methodological differences among studies in different local communities make it difficult to integrate these measures into a common index of niche differentiation. Instead, we coded a fundamental and unambiguously categorizing attribute of the foraging niche of woodwarblers, designating each species as either terrestrial (mean foraging height ,1 m; 10 species) or arboreal (mean height .1 m; 33 species) based on quantitative foraging studies conducted during the breeding season (Poole and Gill 2004). Each pair of taxa could therefore comprise species that forage primarily in similar (terrestrial–terrestrial and arboreal–arboreal pairs) or separated (terrestrial–arboreal) strata. This categorization allowed us to determine whether species pairs that are predicted to interact more intensively (because they occur in the same stratum; MacArthur 1958, Morse 1989:293, Wiens 1989) show a different association between phylogenetic relationship and co-occurrence than do species pairs with a higher degree of spatial separation. Taxon sampling We included 43 Parulidae species that breed within the BBS region in the continental United States and Canada, excluding only six rare or geographically peripheral species that were recorded on few (
10) BBS transects; all included species were recorded on 41 transects. Excluded species were Colima Warbler (Vermivora crissalis), Tropical Parula (Parula pitiayumi), Goldencheeked Warbler (Dendroica chrysoparia), Kirtland’s

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

Warbler (D. kirtlandii), Red-faced Warbler (Cardellina rubrifrons), and Painted Redstart (Myioborus pictus). We excluded two additional species, the Yellow-breasted Chat (Icteria virens) and the Olive Warbler (Peucedramus taeniatus), because, although these morphologically aberrant lineages have traditionally been placed in the Parulidae, more recent phylogenetic analyses (Sibley and Ahlquist 1990, Klicka et al. 2000, Lovette and Bermingham 2002) have shown unambiguously that they fall outside the Parulidae and that they are more closely related to species in other families. Genetic distances As indices of phylogenetic similarity, we generated two matrices of genetic distances among the 43 woodwarbler species based on the complete DNA sequences of the mitochondrially encoded cytochrome oxidase I, cytochrome oxidase II, NADH dehydrogenase II, ATPase 6, and ATPase 8 genes (4116 nucleotides/taxon; GenBank accession numbers AY650182–AY650224). Noncoding spacer regions and tRNA sequences situated between these coding genes were sequenced, but they were excluded from distance calculations owing to their different underlying pattern of molecular evolution. These sequences are substantially longer than the mitochondrial alignments typically generated for species-level avian phylogenetics; by basing our distance metric on this robust mtDNA data set, we minimize the stochastic error in our estimates of pairwise distances. Laboratory methods used to generate these sequences have been described elsewhere (e.g., Lovette 2004). We calculated the first matrix of pairwise distances using the maximum likelihood (ML) method implemented in PAUP*4.0 (Swofford 2002) under the general time reversal plus gamma plus invariant sites (GTR-gþI) model. Because these ML distances could be biased if mitochondrial divergence is not constant across lineages, as would be the case if a subset of the taxa included here had a higher or lower mutation rate than the remaining taxa, we also calculated pairwise distances using the summed branch lengths connecting each pair of termini in an ultrametric tree. We derived this ultrametric tree from an analysis in which the Bayesian topology described below was imported into PAUP* as a constraint tree, the heuristic search algorithm was set to produce a clocklike (ultrametric) topology, and a maximum-likelihood analysis was conducted using mean GTR-gþI parameters derived from the Bayesian Markov chain Monte Carlo (MCMC) analysis. Hereafter, we use the term ‘‘ML distances’’ to refer to the nonultrametric distance matrix, and ‘‘ultrametric distances’’ to refer to the ML distance matrix derived from the ultrametric branch lengths in the topology with an enforced molecular clock. For all statistical analyses involving genetic distance matrices, we report the results from the ML distances followed by the ultrametric distances (hence these summaries each have two successive P values, etc.). As the results from the two

S19

alternative distance metrics were highly congruent (see Results and Discussion), the figures printed here illustrate only the ML distance-based comparisons; the Appendix depicts the alternative figures based on the ultrametric distance matrix. We reconstructed phylogenetic relationships among the 43 species using the same sequences employed for the genetic distance calculations. We generated these phylogenetic analyses via the Bayesian MCMC approach implemented in MrBayes 3.0b4 (Huelsenbeck and Ronquist 2001) under the GTR-gþI model of sequence evolution. The search was run for 5 3 106 generations and sampled every 2500 generations; the initial 1000 samples were discarded as burn-in. An important assumption in our use of mtDNA-based distances as a measure of species relationships is that the mtDNA gene tree reflects the overall organismal ‘‘species tree’’ for these taxa. This assumption would be violated if past hybridization resulted in interspecific mitochondrial transfer (introgression), as has been documented in a few avian groups in which closely allied species breed in sympatry (e.g., Gala´pagos finches; Sato et al. 1999), or between species with active hybrid zones. In the paruline warblers several lines of evidence—including long, species-specific mtDNA lineages (Lovette and Bermingham 1999, 2002) and the very high congruence between the mtDNA gene tree and independent trees from DNA sequences from six unlinked nuclear loci (I. J. Lovette, unpublished data)—suggest that the mitochondrial tree is not highly biased by past introgression. The relative constancy of mitochondrial substitution rates is also apparent in this phylogram, in which there are no conspicuously long or short terminal branches or clades. Quantile regression Conventional regression analyses describe changes in mean response, which might not be biologically meaningful information under some circumstances, including our examination of the relationships between phylogenetic and ecological distances. While the potential for high ecological overlap could change with greater phylogenetic distance between warbler species, this potential might not be realized for all pairs of species. Thus, on average, ecological overlap could vary little with changes in phylogenetic distance, even if some proportion of species pairs that are more distantly related do have greater ecological overlap; i.e., a regression through the mean could show no relationship between phylogenetic distance and ecological overlap, but a regression through data from the most highly overlapping species (e.g., the 95th percentile of highest overlap at each value of phylogenetic distance) would exhibit a relationship between phylogenetic distance and ecological overlap. Thus, we needed to analyze our data using statistical techniques that could provide us regression lines through percentiles (quantiles) of our choosing, if we wanted to detect the patterns that we expected.

S20

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

We did not simply select the data points from the quantiles of our choosing and run conventional regression lines through these. This approach would have required us to arbitrarily divide the continuous variation in phylogenetic distance into discrete intervals, in which to determine the appropriate value of co-occurrence for each quantile used. Such arbitrariness would have meant that regression lines, and thus biological conclusions, would be dependent on the rules used to divide data into groups with differing ranges of phylogenetic distance. In order to avoid these complications in interpretation of results, we used a statistically justifiable and repeatable analytical technique, linear quantile regression (Koenker and Bassett 1978, Cade and Noon 2003), to quantify the relationship between phylogenetic distance and quantiles of the distribution of co-occurrence. Instead of describing a single line through the mean, as is done in conventional linear regression, a linear quantile regression describes linear changes in the shape of the entire distribution of the response variable at all values of a predictor variable, allowing a user to request production of regression lines through one or several arbitrarily chosen quantiles of the response variable’s distribution. Regression equations for any arbitrary quantiles of this distribution can be calculated. In our analyses, we examined changes in median and 5% quantiles of our co-occurrence measures as a function of changes in phylogenetic distance. Remember that our indices of co-occurrence were defined so that high numbers correspond to lower co-occurrence; thus a 5% quantile denotes the 95th percentile of high cooccurrence. Regressions through the median inform us as to whether pervasive changes in co-occurrence were seen along the gradient of phylogenetic distances. A median-quantile regression line and a conventional linear regression line would be essentially identical for data with normally distributed errors. Slopes from the 5% quantiles, in contrast, indicate whether the extent of co-occurrence varied only for a small proportion of species pairs with greater than median co-occurrence. The 5% quantile reflects variation in the relationship between co-occurrence and phylogenetic distance when only a small proportion of species pairs have actually realized the potential for greater co-occurrence with changing phylogenetic distance. Note that, as mentioned above, changes in the 5% quantile reflect changes in the shape of the entire distribution of co-occurrence, and should not be interpreted to mean that only the 5% of most highly co-occurring species varied in co-occurrence with changes in phylogenetic distance. We do not present results for 95% quantiles, as the 95% quantile of co-occurrence did not vary with changes in phylogenetic distance. In all cases the 5% of species pairs with the lowest co-occurrence (i.e., the 95% quantile) abutted the hard boundary of complete non-co-occurrence at point or transect (i.e., had a co-occurrence of zero and hence an ecological distance of 1.0).

Ecology Special Issue

Randomizations and statistical analyses Two attributes of the data require the use of randomization tests for exploring the association between ecological co-occurrence and phylogenetic distance. First, the underlying comparisons between pairs of species are not independent: owing to differences in abundance, each pair of species was represented twice in calculations of ecological co-occurrence (A with B, and B with A), and a given species was also represented in all pairwise associations with co-occurring species. Each species was similarly represented multiple times in the matrix of pairwise genetic distances. As a result, a nonzero slope for the relationship between phylogenetic distance and co-occurrence could be generated by chance alone. This nonindependence would also cause true Type I error rates to differ from the nominal rates calculated by more conventional statistical tests, another problem that is dealt with by randomization tests. For our examination of local (within-BBS-transect) patterns, we generated null distributions against which to compare the observed relationships between phylogenetic and ecological co-occurrence by randomly shuffling species identities among points within each BBS transect. The algorithm used in the randomizations was to randomly reorder the values within the species name column in the input data table, with the reordering done separately and independently for data from each BBS transect. The input data table for this randomization had one line for each species and census point on a BBS transect on which that species was observed, with each BBS transect being represented by only one year of data. The result of the randomization was to maintain the same number of warbler species being recorded at each point on a transect, maintain the number of points at which each species was reported, and maintain the same species composition on a transect as was present in the actual data. The randomization procedure was designed to break up local ecological associations between species, but still (1) maintain the identities and relative abundances of all species within each transect and (2) maintain information on the proportion of points within a transect that were suitable for any species of wood-warbler, for example, ensuring that points with no reported warblers continued to have no warbler species after the randomization. This randomization procedure is similar to a Mantel’s test in which cells in a matrix of similarity values are randomly rearranged to produce null distributions. Our procedure differed from that in a Mantel’s test by (1) constraining the randomization as explained and (2) employing a different test statistic, in our case slopes from quantile regression. In our analyses of among-transect associations between phylogenetic and regional co-occurrence, we used a similar randomization procedure. Both the number of warbler species per transect and the total number of times each species was found in the data table

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

remained identical to the values in the real data. Only the associations between species pairs were broken up by the randomization procedure. Mechanistically, the randomization was done on an input data table that contained a column identifying each transect uniquely, and another column containing species’ names. Each transect appeared 43 times in the file, one time for each warbler’s name. A third column in the input data was a Boolean variable that indicated whether each species in the real data was reported on that specific transect. Each transect was represented by a single year’s data in the input data file. The randomization was accomplished by randomly reordering the warbler species’ names relative to the rest of the information in the input file; this reordering was done separately for species that were listed as having been seen, and those listed as having not been seen on a specific transect in the real data. These two separate reorderings were needed to keep the number of transects on which a species was present in the randomized data at the same value as in the real data. Although the transect-level results are interesting in comparison to the within-transect results that include only species pairs with demonstrated regional sympatry, it is important to note that the transect-level randomization brings together species that never occur together in nature, because they have nonoverlapping geographic distributions. The second reason for using randomization tests was to incorporate replication across multiple years for each transect, as occurrences may differ among years for biological reasons (e.g., demographic productivity in the previous year) and as sampling artifacts. We accounted for this interannual variation by generating multiple data sets, each of which contained a single, randomly selected year’s data for each transect. We combined the two levels of randomization by (1) randomly selecting one year for each transect and (2) creating a null data set by randomizing species’ distributions for that transectyear as we have described. This two-stage randomization was repeated 1000 times each for the withintransect analyses and the among-transect analyses, and we separately performed the quantile regression analyses for each iteration on the true and randomized data. We investigated whether the patterns found in our analyses were substantially influenced by single species through analyses that tested individually for the influence of each species on the overall results. In these influence analyses, all species pairs containing data from a target species were removed from the data set, and quantile regressions were calculated for each of the 1000 randomly sampled years of true data, as just described. We used the differences between regression slopes with and without the target species as our metric of influence of this species on the overall results. We treated each of the 43 species in turn as the target species and compared their influences by examining median changes in regression slopes, from the 1000 data sets, when that species’ data were excluded.

S21

We calculated probabilities from our regressions of the relationships between phylogeny and ecological overlap as the proportion of randomizations in which the regression slope from the true data was greater than the slope from the randomized data. Counts of steeper positive and negative slopes were tallied separately, and the proportion of steeper slopes in one direction gave a one-tailed probability. The two-tailed probabilities that we report in the text were calculated by doubling these one-tailed probabilities. The confidence intervals depicted in the figures are the range of the central 95% of all regressions. Plotted results from quantile regression analyses are not derived from median regression coefficients, but are lines that connect a series of predicted values, with the median predicted values representing the regression line and the range of the central 95% of predicted values representing the 95% confidence limits at each point. As a result of plotting lines connecting point estimates, it is possible for plotted lines to show deviations from perfect linearity even though linear quantile regressions were used. We used the SAS statistical software to generate random data sets for analysis and the quantreg library (Quantreg package, 2004, version 3.35, by R. Koenker [available online])2 for the R statistical software (R Development Core Team 2004) to conduct all quantile regression analyses. Conclusions did not change qualitatively between analyses based on 500 and 1000 randomization iterations, indicating that 1000 iterations were sufficient to describe variability in null associations among taxa. RESULTS

AND

DISCUSSION

Patterns of association within local areas (BBS transects) Three features are immediately obvious in comparisons of ecological dissimilarity with phylogenetic distance (Fig. 2a). First, although pairwise genetic distances varied within 1–18%, the majority of pairwise distances were clustered at intermediate levels of divergence owing to several periods of rapid cladogenesis during the diversification of this group (Lovette and Bermingham 1999, 2002). This temporal clustering of nodes is visible in the phylogenetic reconstruction for these taxa (Fig. 3). Second, there is a notable lack of taxa that are both phylogenetically similar and have high local co-occurrence. This is seen in the empty lower left quadrant in Fig. 2a. Third, across the full gradient of genetic divergence, most species pairs detected on the same Breeding Bird Survey (BBS) transect were rarely detected at the same transect points, leading to the majority of pairwise comparisons of ecological dissimilarity being clustered near the upper boundary in Fig. 2a. A smaller number of species pairs co-occur at higher frequency, with a median (of 1000 random subsamples of BBS data) of only 0.085% of the 1184 pairwise 2

hhttp://cran.r-project.org/doc/packages/quantreg.pdfi

S22

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

Ecology Special Issue

FIG. 2. Ecological co-occurrence of wood-warbler species pairs across a gradient of phylogenetic divergence (maximumlikelihood distances). In each panel, the background of plotted points depicts the same random sample of real census transect-years that is representative of patterns observed across all 1000 such iterations. (a) Distribution of pairwise associations between phylogeny and ecological co-occurrence among taxa that forage at similar (green) or different (black) habitat strata. (b) Median (thick lines) and lower fifth quantile (thin lines) relationships between phylogenetic distance and ecological co-occurrence for species pairs that forage in different habitat strata. Regression through the true data (red) and randomized null-model data (blue) are both plotted. For each of the four lines, the median from the 1000 iterations is plotted with the inner 95% of predicted values forming the confidence intervals (dashed lines). (c) Median and lower fifth quantile regressions for species pairs foraging in the same habitat stratum. (d) Corresponding quantile regressions for pairs of taxa foraging in different strata, but with data from Ovenbirds (Seiurus aurocapilla) excluded from the analyses. In panels (b) and (d), regression lines are only plotted across the range on the xaxis of actual data points used to derive the regressions. In these panels, the background points represent the relevant subset of data from the full distribution in panel (a).

comparisons involving taxa with .50% detected cooccurrence. On a historical note, the five ‘‘spruce-woods warblers’’ featured in Robert MacArthur’s 1958 study all have median (of 1000 random subsamples of BBS data) pairwise dissimilarity values .0.83, suggesting that these species are not usually detected in local sympatry. Our interpretation of the conspicuous absence of species pairs in the lower left quadrant of Fig. 2a is that closely related taxa effectively exclude each other at a local scale, even when they occur in geographic proximity. Some wood-warbler species pairs separated by greater phylogenetic distances had high likelihoods of

local co-occurrence, although across all phylogenetic distances the majority of species pairs had relatively high local ecological dissimilarity. We contrasted these observed patterns with a null model of randomized species occurrences. When all pairs of taxa are considered, the median co-occurrence does not change significantly with phylogenetic distance relative to the null distribution (P ¼ 0.192 for differences in slopes compared to the null model, based on maximumlikelihood distances; P ¼ 0.164, based on ultrametric distances). However, maximal co-occurrence significantly increases as ML distance increases (P ¼ 0.03 from the fifth-quantile regression slope), although the

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

S23

FIG 3. Phylogenetic relationships among the 43 wood-warbler taxa included in this study based on Bayesian Markov chain Monte Carlo analysis of long mitochondrial DNA sequences (see Methods). Thick branches indicate species (or groups of species) coded as terrestrial foragers; the remaining taxa are arboreal foragers. The tree is rooted to various non-wood-warbler outgroup taxa, as in Lovette and Bermingham (2002).

pattern only approached statistical significance when ultrametric distances were used (P ¼ 0.066). These results suggest that a subset of the pairs of more distantly related species is capable of higher local association than expected by chance. Further analyses suggest that the increased cooccurrence of some more distantly related species pairs requires divergence in foraging niche in order to reduce potential competition. Adding each species pair’s foraging strata information (in the form of a categorical variable with two categories) to the quantile regression analyses revealed a nonrandom relationship at the lower (fifth-quantile) boundary for the mixed-strata

pairs (P ¼ 0.026 for ML distances, P ¼ 0.000, for ultrametric distances), for which fifth-quantile co-occurrence increased with greater phylogenetic distance (Fig. 2b), but not for same-stratum pairs, for which the slopes of the real and randomized regressions do not differ significantly (P ¼ 0.526 for ML distances, P ¼ 0.736 for ultrametric distances) (Fig. 2c). Results from analyses using ultrametric distances even suggested the possibility of higher co-occurrrence with greater phylogenetic distance from the median regression for mixed-substrate pairs (P ¼ 0.070), although this pattern was not reflected in the median regression when ML distances were used (P ¼ 0.236). No hint of nonrandom

S24

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

associations were found from the median regressions from same-substrate pairs (P ¼ 0.556 for ML distances, P ¼ 0.478 for ultrametric distances). Thus, in spite of already having a greater average genetic distance associated with their differing foraging substrates (compare the ranges of genetic distances in Fig. 2b, c), only the mixed-stata pairs showed changes in cooccurrence with phylogenetic distance. This result suggests that differences in or associated with foraging substrate are a necessary but not sufficient precondition for co-occurrence to vary with phylogenetic distance. To explore which taxa were driving the fifth-quantile relationship between phylogeny and ecological overlap (Fig. 2b), we repeated these analyses and serially removed comparisons involving each of the 43 taxa. These influence analyses were only conducted on the fifth-quantile regressions for mixed-strata pairs, as this was the only case in which a clear effect of phylogenetic distance on ecological co-occurrence was detected. Median changes in the fifth-quantile slope, with removal of data from a single species, almost all clustered between 0.5 and 0.5 for the analyses based on ML phylogenetic distances, and between 0.2 and 0.08 for analyses based on ultrametric distances. The only exceptions were for Ovenbirds, in whose absence the median increase in slope was 2.9 (2.7 with ultrametric distances) (Fig. 2d) and Palm Warblers (Dendroica palmarum), for which the median change in slope was 1.6 1.2 for ultrametric distances). Only the increase in regression slope in the absence of Ovenbirds is relevant to explaining the increased ecological association with increasing phylogenetic distance, and for analyses based on ML phylogenetic distances only for Ovenbirds was the inner 95% of changes above zero, with none of the 1000 data sets showing a lower slope when Ovenbirds were excluded from the quantile regressions. With ultrametric phylogenetic distances, the slope of the 5% quantile line was always higher in the absence of data from Ovenbirds and always lower in the absence of data from Palm Warblers. The effect of Ovenbirds is not surprising because of the Ovenbird’s placement as the earliest branch within the entire Parulidae (Lovette and Bermingham 2002; I. J. Lovette, unpublished data) (Fig. 3). If ecological differentiation is associated with time, this relatively ancient lineage is expected to exhibit the highest ecological differentiation (and concomitant spatial association) relative to the more derived species within the radiation. A nonrandom increase in the potential for ecological co-occurrence of distantly related species was one of two patterns evident in our data. We also find evidence of consistent phylogenetic niche conservatism in the ubiquitous vertical offset between the actual and null quantile regression lines (Fig. 2b–d). This phenomenon is present in all quantile regressions and is particularly evident in the lower fifth quantiles. The consistently lower position of the regression lines based on the real data, relative to the lines from the corresponding null

Ecology Special Issue

models, indicates that, regardless of phylogenetic distance, similarly related sympatric species choose a larger number of similar habitats than expected by chance. This pattern is consistent with a growing body of evidence showing that phylogenetic relationships sometimes explain substantial variation in how related species are distributed across habitat types, bioclimatic regions, and other niche-related environmental variables (e.g., Ricklefs and Latham 1992, Peterson et al. 1999). The historical legacy in ecological niches is not surprising, given that phylogenetic effects are usually apparent in organismal traits with a genetic basis, from molecular pathways to complex behaviors (Harvey and Pagel 1991, Ackerly 2003). In a different context, niche conservation is also apparent in the distribution of arboreal vs. terrestrial foraging among these 43 wood-warbler species (Fig. 3): evolutionary switches among these strata are uncommon, and no pairs of very closely related species have diverged across this level of spatial segregation. Are patterns driven by differences in species’ detection probabilities? The data presented in Fig. 2 are based on detections of individual birds, which require both that a given species is actually present and also that this species has been detected by the censusing observers. Actual presence, and hence ecological co-occurrence, is thereby systematically underestimated by the inevitable missed detections. As a result, the percentages of co-occurrence presented here are not direct quantifications of ecological co-occurrence, but instead comparable indices of relative co-occurrence. Nevertheless, examination of patterns in our index of detection probability (see Methods) leads us to believe that the general patterns of ecological co-occurrence that we have summarized would be essentially impossible to create as a result of among-species variation in detection probability alone. We expect that relatively low ecological co-occurrence for the majority of species pairs is a biologically real phenomenon, because creation of this pattern through variation in detection probability would require at least one member of the majority of species pairs to have an anomalously low detection probability. As long as the distribution of species’ detection probabilities has a mode at some intermediate value (as our index of detection probability does; see Methods), the observed distribution of ecological co-occurrence would be extremely unlikely to be produced as a result of interspecific variation in detection rates. The products of detection indices for all observed species pairs are, as expected, distributed with a peak at intermediate values, which indicates that the generally low local co-occurrence of species pairs (Fig. 2a) is a biologically real phenomenon. For reasons similar to those we have discussed, increasing ecological co-occurrence with increasing phy-

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

S25

FIG. 4. Regional co-occurrence of wood-warbler species pairs across a gradient of phylogenetic divergence (maximumlikelihood distances). In each panel the background of plotted points depicts the same random sample of real census transect-years that is representative of patterns observed across all 1000 such iterations. (a) Distribution of pairwise associations between phylogeny and regional co-occurrence among taxa that forage at similar (green) or different (black) habitat strata. (b) Median (thick lines) and lower fifth quantile (thin lines) relationships between phylogenetic distance and regional co-occurrence for all species pairs combined. Regressions through the true data (red) and randomized null-model data (blue) are both plotted. For each of the four lines, the median from the 1000 iterations is plotted with the inner 95% of predicted values forming the confidence intervals (dashed lines).

logenetic distance (Fig. 2b) also appears highly unlikely to be an artifact of systematic variation in detection probabilities among species as a function of their phylogenetic distance. For variation in detection probabilities to have caused this pattern, we would need to find detection probabilities of some proportion of species pairs to increase with increasing phylogenetic distance. However, we found no evidence for this, and in fact a significant tendency in the opposite direction using a fifth-quantile regression of minimum detection index (within each of the 903 species pairs) against phylogenetic distance. In this fifth-quantile regression, confidence intervals around the estimated slope did not come close to overlapping zero (slope ¼ 0.45, 95% confidence limits of 0.50 to 0.41). Thus, we believe that the increasing ecological co-occurrence with greater phylogenetic distance is a biological phenomenon and not an artifact of limitations of data collection. Interspecific variation in detection rates could only have caused the significant influence of Ovenbirds on the results (Fig. 2b, d) if increasing genetic distance from Ovenbirds were correspondingly correlated with increasing detection probability (e.g., by louder and more persistent vocalization) for arboreally foraging woodwarblers. While there was an increase in detection probability of arboreally foraging species paired with Ovenbirds as their phylogenetic distance from Ovenbirds increased, the result was not statistically significant (r2 ¼ 0.05, P ¼ 0.19, N ¼ 33, slope ¼ 0.76). Even were this increase in detection probability real, it still would translate to an average change in the detection index of

only 8% across the full range of phylogenetic distances between Ovenbirds and arboreally foraging warblers found on the same transects. In contrast, the presence of species pairs with Ovenbirds leads to a roughly 135% change in ecological co-occurrence at the fifth quantile (Fig. 2d). This contrast of effects suggests that, while systematic variation in detectability contributed slightly to the closer measured ecological co-occurrence of arboreally foraging warblers more distantly related to Ovenbirds, the effect of this measurement artifact was trivial. Finally, the lower position of the real vs. null regression lines that we interpret as indicating higher than expected niche conservatism could only be an artifact of varying detection probabilities if the actual species pairs were composed of pairings in which both members had higher detection probabilities than was typical of randomly selected pairs of warbler species. The median product of the detection probabilities of the actual species pairs was 0.281, and products ranged from 0.198 to 0.375 in the 903 observed species pairs. From 1000 iterations of an equal number of random species pairings, the median product was 0.281 (95% confidence limits 0.279–0.282). There is no evidence to suggest that phylogenetic niche conservatism is a sampling artifact. Contrasting local and regional relationships between phylogenetic distance and spatial co-occurrence At the broader spatial scale involving comparisons among transects, there was a greater range of variation

S26

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

in species co-occurrence (Fig. 4a) than was found for local, ecological co-occurrence (Fig. 2a). However, the higher values of co-occurrence may largely be accounted for by the higher probability of detection of a given species on a transect as a whole, relative to a single point on that transect. Nevertheless, we found a predominance of species pairs with little or no overlap in occurrence on transects, as we would expect given that at maximum 23 of the 46 North American wood-warblers ever occur on the same transect, and most transects support substantially fewer species (Fig. 1). This real geographical separation among many species pairs drives much of the pattern in the transect-level analyses and demonstrates the value of the point-based comparisons we describe, in which we are able to restrict analyses to only those transects where both members of a given pair of species occur. At the transect level, the fact that many pairs of warbler species have largely to fully nonoverlapping ranges is evidenced in the y-axis offsets between actual median regional overlap and the overlap expected by chance (Fig. 4b). Here, the median regional overlap was lower (rather than higher, as in the point-based analyses) than expected by chance, as indicated by the complete displacement of the median-quantile regression line above the null-model regression line. The opposite pattern was found in the fifth-quantile regressions, for which the observed level of regional co-occurrence was higher, for the 5% of most overlapping species, than that expected by chance alone, regardless of phylogenetic distances. This fifth-quantile result is likely a function of some species pairs having very similar habitat associations, and thus ranges. The final pattern to emerge from analyses of regional association is that variation in regional overlap with phylogenetic distance is different from that expected by chance alone, when analyses were based on ML distances. If warbler species were randomly distributed across North America, the randomization suggests that there would be a slight decrease in regional cooccurrence with increasing phylogenetic distance (as indicated by the positive slope of the blue medianquantile regression line in Fig. 4b). What we actually found for the real data is no relationship between regional co-occurrence and phylogenetic distance (as represented by the flat red median-quantile in Fig. 4b), a statistically significant difference from the null slope (P ¼ 0.000). For analyses based on ultrametric distances, the slope of the line through the real data did not differ from zero, although in this case the result was not a departure from the null regression line (P ¼ 0.110), the slope of which also did not differ from zero. This comparison suggests that there is only a weak signature of phylogeny on the overall geographic ranges of these North American warbler species. Previous studies of other taxa have sometimes reported stronger (and usually positive) correlations between phylogenetic distances and the overlap of species’ geographic ranges

Ecology Special Issue

(e.g., Barraclough and Vogler 2000), but the lack of such an association in our analyses is the result predicted for groups of organisms—such as these warblers—in which species’ ranges must have been shifted repeatedly by large-scale climate cycles, thereby scrambling most geographic patterns associated with their initial speciation events (Losos and Glor 2003). In summary, when comparing point (Fig. 2) and transect (Fig. 4) scales, there are clear differences in both the general magnitude of species’ co-occurrence, and in the relationship between phylogenetic distance and species co-occurrence. Broadly speaking, these differences reinforce the notion that the factors that are determining local ecological co-occurrence are different than those biogeographic influences that have created the varying distributional ranges of North America’s parulid warblers. CONCLUSIONS Phylogenetic niche conservation has most often been observed in allopatric species that occupy similar niches despite a period of evolutionary isolation (e.g., Peterson et al. 1999); in such cases, niche conservation may even enhance the probability of continued isolation if it lowers the likelihood that one or both populations evolve ecologically to occupy the previously unsuitable habitat that separated them (Wiens 2004). Stabilizing selection on niche attributes is one potential cause of phylogenetic niche conservation (Harvey and Pagel 1991, Ackerly 2003), but stabilizing selection leading to phylogenetic niche conservatism may be more likely in allopatry because it is less likely to be opposed by divergent selection among closely related sympatric taxa. The analyses here suggest that niche conservation is evident in a large group of species, even with substantial sympatry. In the wood-warblers, divergent selection stemming from interspecific competition has not been sufficient to completely erase the phylogenetic legacy of habitat specialization. The majority of our comparisons address only the component of ecological differentiation that is associated with species’ spatial co-occurrences, or their lack thereof. For example, these comparisons are much more likely to detect patterns related to fundamental differences in species’ selection of habitats than they are to the more subtle ecological differences that likely allow coexisting species to partition ecological resources within habitats. The contrasting quantile regressions for pairs of warbler species that forage in similar vs. mixed vegetation strata (Fig. 2b, c) demonstrate that behavioral differentiation contributes to patterns of local community composition, but we were unable to further address this level of ecological niche differentiation. Another important limitation of our approach is that both present and past competition might cause species to be spatially segregated, and our analyses do not distinguish between the effects of ongoing species interactions and differentiation in site selection caused

July 2006

WARBLER PHYLOGENY AND COMMUNITY STRUCTURE

by competition in the past. Similarly, at an even broader spatial scale, it is possible that competitive interactions feed back to influence species’ ranges, particularly if species with highly similar ecological niches tend to exclude one another at this broad scale. The influence of competition on the regional species pool has been termed the Narcissus Effect in the ecological literature (Colwell and Winkler 1984), but has received less attention in a phylogenetic context. Because the comparisons presented here are most biologically informative at the site-within-transect level, we are unable to provide a robust test for any patterns at the broader regional level. Considered in concert, our results suggest that woodwarbler community assembly is constrained by ecological similarity among closely related species, with subsequent local co-occurrence of more distantly related lineages mediated in part by divergence in ecology and behavior. Considerable evolutionary time appears to be required for high levels of spatial overlap. This interpretation challenges the classic depiction of the North American wood-warbler adaptive radiation as one in which many species have quickly evolved substantial behavioral differences that allow extensive coexistence (Lack 1971, Morse 1989, Price et al. 2000). Phylogenetic niche conservatism (Ricklefs and Latham 1992, Peterson et al. 1999, Webb et al. 2002, Losos et al. 2003) appears to play a continuing role in structuring even these highly spatially and temporally dynamic communities, and the relatively few pairs of taxa that consistently co-occur with high frequency are both distantly related and ecologically differentiated. At the most general level, this new phylogenetic perspective on the community ecology of the parulid warblers reinforces the idea that for most groups of sympatric species, the relationship between phylogeny and community structure likely involves simultaneous stasis and differentiation along many ecological niche axes. ACKNOWLEDGMENTS We thank D. Rabosky, J. Losos, T. Price, R. Ricklefs, C. Webb, and two anonymous reviewers for insightful suggestions and comments, C. Cooper for bringing quantile regressions to our attention, and I. Fiorentino and L. Stenzler for laboratory assistance. We gratefully acknowledge the contributions of the thousands of volunteer BBS participants and the BBS staff at the USGS Patuxent Wildlife Research Center and Canadian Wildlife Service who have collectively assembled a uniquely comprehensive resource for studies of avian abundance and distribution. This research was supported by the National Science Foundation with logistic support during the initial data analyses to WMH from the Max Planck Institute for Ornithology, Vogelwarte Radolfzell. LITERATURE CITED Ackerly, D. D. 2003. Community assembly, niche conservatism, and adaptive evolution in changing environments. International Journal of Plant Science 164:S165–S184. Ackerly, D. D., D. W. Schwilk, and C. O. Webb. 2006. Niche evolution and adaptive radiation: testing the order of trait divergence. Ecology 87:S50–S61.

S27

Anderson, T. M., M. A. Lachance, and W. T. Starmer. 2004. The relationship of phylogeny to community structure: the cactus yeast community. American Naturalist 164:709–721. Barraclough, T. G., and A. P. Vogler. 2000. Detecting the geographical pattern of speciation from species-level phylogenies. American Naturalist 155:419–434. Cade, B. S., and B. R. Noon. 2003. A gentle introduction to quantile regression for ecologists. Frontiers in Ecology and the Environment 1:412–420. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Cavender-Bares, J., A. Keen, and B. Miles. 2006. Phylogenetic structure of Floridian plant communities depends on taxonomic and spatial scale. Ecology 87:S109–S122. Cilimburg, A. B., M. S. Lindberg, J. J. Tewksbury, and S. J. Hejl. 2002. Effects of dispersal on survival probability of adult Yellow Warblers (Dendroica petechia). Auk 119:778– 789. Colwell, R. K., and D. W. Winkler. 1984. A null model for null models in biogeography. Pages 344–359 in D. R. Strong, D. Simberloff, L. G. Abele, and A. B. Thistle, editors. Ecological communities: conceptual issues and the evidence. Princeton University Press, Princeton, New Jersey, USA. Darwin, C. 1859. On the origin of species. Murray, London, UK. Gillespie, R. 2004. Community assembly through adaptive radiation in Hawaiian spiders. Science 303:356–359. Graham, A. 1999. Late Cretaceous and Cenozoic history of North American vegetation. Oxford University Press, Oxford, UK. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Holmes, R. T., R. E. Bonney, and S. W. Pacala. 1979. Guild structure of the Hubbard Brook bird community: a multivariate approach. Ecology 60:512–520. Holmes, R. T., P. P. Marra, and T. W. Sherry. 1996. Habitatspecific demography of breeding Black-throated Blue Warblers (Dendroica caerulescens): implications for population dynamics. Journal of Animal Ecology 65:183–195. Huelsenbeck, J. P., and F. Ronquist. 2001. MRBAYES: Bayesian inference of phylogenetic trees. Bioinformatics 17: 754–755. Johnson, N. K., and C. Cicero. 2004. New mitochondrial DNA data affirm the importance of Pleistocene speciation in North American birds. Evolution 58:1122–1130. Kembel, S. W., and S. P. Hubbell. 2006. The phylogenetic structure of a Neotropical forest tree community. Ecology 87: S86–S99. Klicka, J., K. P. Johnson, and S. M. Lanyon. 2000. New World nine-primaried oscine relationships: constructing a mitochondrial DNA framework. Auk 117:321–336. Klicka, J., and R. M. Zink. 1999. Pleistocene phylogeographic effects on avian evolution. Proceedings of the Royal Society of London B 266:695–700. Knouft, J. H., J. B. Losos, R. E. Glor, and J. J. Kolbe. 2006. Phylogenetic analysis of the evolution of the niche in lizards of the Anolis sagrei group. Ecology 87:S29–S38. Koenker, R., and G. Bassett. 1978. Regression quantiles. Econometrica 46:33–50. Lack, D. 1971. Ecological isolation in birds. Harvard University Press, Cambridge, Massachusetts, USA. Losos, J. B. 1996. Phylogenetic perspectives on community ecology. Ecology 77:1344–1354. Losos, J. B., and R. E. Glor. 2003. Phylogenetic comparative methods and the geography of speciation. Trends in Ecology and Evolution 18:220–227. Losos, J. B., M. Leal, R. E. Glor, K. de Queiroz, P. E. Hertz, L. Rodriguez Schettino, A. Chamino Lara, T. R. Jackman, and

S28

IRBY J. LOVETTE AND WESLEY M. HOCHACHKA

A. Larson. 2003. Niche lability in the evolution of a Caribbean lizard community. Nature 424:542–545. Losos, J. B., and D. B. Miles. 2002. Testing the hypothesis that a clade has adaptively radiated: iguanid lizard clades as a case study. American Naturalist 160:147–157. Lovette, I. J. 2004. Molecular phylogeny and plumage signal evolution in a trans Andean and circum Amazonian avian species complex. Molecular Phylogenetics and Evolution 32: 512–523. Lovette, I. J. 2005. Glacial cycles and the tempo of avian speciation. Trends in Ecology and Evolution 20:57–59. Lovette, I. J., and E. Bermingham. 1999. Explosive speciation in the New World Dendroica warblers. Proceedings of the Royal Society of London B 266:1629–1636. Lovette, I. J., and E. Bermingham. 2002. What is a woodwarbler? Molecular characterization of a monophyletic Parulidae. Auk 119:695–714. MacArthur, R. H. 1958. Population ecology of some warblers of northeastern coniferous forests. Ecology 39:599–619. MacKenzie, D. I., J. D. Nichols, G. B. Lachman, S. Droege, J. A. Royle, and C. A. Langtimm. 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83:2248–2255. Martin, P. R., and T. E. Martin. 2001. Ecological and fitness consequences of species coexistence: a removal experiment with wood warblers. Ecology 82:189–206. Martin, T. E. 1998. Are microhabitat preferences of coexisting species under selection and adaptive? Ecology 79:656– 670. McPeek, M. A., and T. E. Miller. 1996. Evolutionary biology and community ecology. Ecology 77:1319–1320. Morse, D. H. 1989. American warblers: an ecological and behavioral perspective. Harvard University Press, Cambridge, Massachusetts, USA. Peterson, A. T., J. Sobero´n, and V. Sa´nchez-Cordero. 1999. Conservatism of ecological niches in evolutionary time. Science 285:1265–1267. Poole, A., and F. Gill, editors. 2004. The birds of North America series. American Ornithologists’ Union and Cornell Laboratory of Ornithology, Ithaca, New York, USA. Price, T., I. J. Lovette, E. Bermingham, L. Gibbs, and A. Richman. 2000. The imprint of history on two warbler adaptive radiations. American Naturalist 156:354–367. Prinzing, A., W. Durka, S. Klotz, and R. Brandl. 2001. The niche of higher plants: evidence for phylogenetic conservatism. Proceedings of the Royal Society of London B 268: 2383–2389. R Development Core Team. 2004. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Rich, T. D., et al. 2004. Partners in Flight North American landbird conservation plan. Cornell Laboratory of Ornithology, Ithaca, New York, USA. Richman, A. D., and T. Price. 1992. Evolution of ecological differences in the Old-World leaf warblers. Nature 355:817– 821. Ricklefs, R. E. 1987. Community diversity: relative roles of local and regional processes. Science 235:167–171.

Ecology Special Issue

Ricklefs, R. E. 2006. Evolutionary diversification and the origin of the diversity–environment relationship. Ecology 87:S3– S13. Ricklefs, R. E., and R. E. Latham. 1992. Intercontinental correlation of geographical ranges suggests stasis in ecological traits of relict genera of temperate perennial herbs. American Naturalist 139:1305–1321. Ricklefs, R. E., and D. Schluter. 1993. Species diversity in ecological communities. Chicago University Press, Chicago, Illinois, USA. Robinson, S. K., and R. T. Holmes. 1982. Foraging behavior of forest birds: the relationships among search tactics, diet, and habitat structure. Ecology 63:1918–1931. Sato, A., C. O’hUigin, F. Figueroa, P. R. Grant, B. R. Grant, T. Tichy, and J. Klein. 1999. Phylogeny of Darwin’s finches as revealed by mtDNA sequences. Proceedings of the National Academy of Sciences (USA) 96:5101–5106. Sauer, J. R., J. E. Hines, and J. Fallon. 2003. The North American Breeding Bird Survey, results and analysis 1966– 2002. Version 2003.1. USGS Patuxent Wildlife Research Center, Laurel, Maryland, USA. Sibley, C. G., and J. E. Ahlquist. 1990. Phylogeny and classification of birds. Yale University Press, New Haven, Connecticut, USA. Silvertown, J., M. Dodd, and D. Gowing. 2001. Phylogeny and the niche structure of meadow plant communities. Journal of Ecology 89:428–435. Silvertown, J., M. Dodd, D. Gowing, C. Lawson, and K. McConway. 2006. Phylogeny and the hierarchical organization of plant diversity. Ecology 87:S39–S49. Stone, L., T. Dayan, and D. Simberloff. 1996. Community-wide assembly patterns unmasked: the importance of species’ differing geographic ranges. American Naturalist 148:997– 1015. Suhonen, J., R. V. Alatalo, and L. Gustafsson. 1994. Evolution of foraging ecology in Fennoscandian Tits (Parus spp.) Proceedings of the Royal Society of London B 258:127–131. Swofford, D. L. 2002. PAUP*: phylogenetic analysis using parsimony (*and other methods). Sinauer Associates, Sunderland, Massachusetts, USA. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Reviews in Ecology and Systematics 33:475–505. Weiblen, G. D., C. O. Webb, V. Novotny, Y. Basset, and S. E. Miller. 2006. Phylogenetic dispersion of host use in a tropical insect herbivore community. Ecology 87:S62–S75. Weir, J., and D. Schluter. 2004. Ice sheets promote speciation in boreal birds. Proceedings of the Royal Society of London B 271:1881–1887. Wiens, J. A. 1989. The ecology of bird communities. Cambridge University Press, Cambridge, UK. Wiens, J. J. 2004. Speciation and ecology revisited: phylogenetic niche conservatism and the origin of species. Evolution 58: 193–197. Wiens, J. J., and C. H. Graham. 2005. Niche conservatism: integrating evolution, ecology, and conservation biology. Annual Reviews in Ecology and Systematics 36:519–539.

APPENDIX Alternative versions of Figs. 2 and 4, based on ultrametric genetic distances (Ecological Archives E087-109-A1).

Ecology, 87(7) Supplement, 2006, pp. S29–S38 Ó 2006 by the Ecological Society of America

PHYLOGENETIC ANALYSIS OF THE EVOLUTION OF THE NICHE IN LIZARDS OF THE ANOLIS SAGREI GROUP JASON H. KNOUFT,1,4 JONATHAN B. LOSOS,2 RICHARD E. GLOR,3 1

AND

JASON J. KOLBE2

Department of Ecology and Evolutionary Biology and University of Colorado Museum, UCB 265, University of Colorado, Boulder, Colorado 80309 USA 2 Department of Biology, Campus Box 1137, Washington University in St. Louis, St. Louis, Missouri 63130 USA 3 Center for Population Biology, University of California, Davis, California 95616 USA

Abstract. Recent advances in ecological niche modeling (ENM) algorithms, in conjunction with increasing availability of geographic information system (GIS) data, allow species’ niches to be predicted over broad geographic areas using environmental characteristics associated with point localities for a given species. Consequently, the examination of how niches evolve is now possible using a regionally inclusive multivariate approach to characterize the environmental requirements of a species. Initial work that uses this approach has suggested that niche evolution is characterized by conservatism: the more closely related species are, the more similar are their niches. We applied a phylogenetic approach to examine niche evolution during the radiation of Cuban trunk-ground anoles (Anolis sagrei group), which has produced 15 species in Cuba. We modeled the niche of 11 species within this group using the WhyWhere ENM algorithm and examined the evolution of the niche using a phylogeny based on ;1500 base pairs of mitochondrial DNA. No general relationship exists between phylogenetic similarity and niche similarity. Examination of species pairs indicates some examples in which closely related species display niche conservatism and some in which they exhibit highly divergent niches. In addition, some distantly related species exhibit significant niche similarity. Comparisons also revealed a specialist–generalist sister species pair in which the niche of one species is nested within, and much narrower than, the niche of another closely related species. Key words: anole; Anolis; Cuba; ecological niche modeling; fundamental niche; niche conservatism; niche evolution; phylogenetics.

INTRODUCTION The fundamental niche, which encompasses the theoretical range of conditions a species can occupy (Hutchinson 1957), provides a conceptual framework to predict the potential geographic distribution of a species (MacArthur 1972, Sobero´n and Peterson 2005). Species traits, whether morphological, physiological, or behavioral, are often obviously linked to the niche and generally susceptible to the processes of evolution. Consequently, the shaping of niche characteristics can be viewed as an evolutionary phenomenon. Because the fundamental niche provides details about the potential distribution of species, and the niche is determined by the processes of evolution, understanding evolutionary patterns of niche diversification can reveal valuable insights into factors related to the diversity and distribution of species. Studies of the niche have historically involved detailed analyses of local habitat requirements of an organism (Chase and Leibold 2003). Recently, the availability of Manuscript received 25 January 2005; revised 16 August 2005; accepted 24 August 2005. Corresponding Editor (ad hoc): C. O. Webb. For reprints of this Special Issue, see footnote 1, p. S1. 4 E-mail: [email protected]

global climate and land cover Geographic Information System (GIS) and remote-sensing data has provided environmental information at a regional scale. These data, when integrated into ecological niche modeling (ENM) algorithms, have provided a powerful opportunity to characterize the habitat requirements of a variety of species and assess patterns of niche differentiation in a comparative framework. The majority of recent niche modeling efforts have focused on predicting species’ distributions, species’ response to climate change, and potential distributions of invasive species (Guisan and Zimmerman 2000, Peterson 2001, 2003, Peterson and Vieglais 2001, Oberhauser and Peterson 2003, Peterson and Robins 2003, Illoldi-Rangel et al. 2004, Peterson et al. 2004). While these studies have provided novel insights into aspects of broad-scale ecological niche characteristics, recent research has begun to realize the potential to examine the results of niche modeling efforts in an evolutionary context (Peterson et al. 1999, Rice et al. 2003, Graham et al. 2004). Two approaches have been taken to integrating information on evolutionary relationships into niche modeling studies. On one hand, some studies have focused on recent evolutionary events by comparing pairs of closely related taxa (either subspecies or sister

S29

S30

JASON H. KNOUFT ET AL.

FIG. 1. Ultrametric phylogeny for the Anolis sagrei group derived from Bayesian phylogenetic analysis of ;1500 base pairs of mtDNA. The support values and simplified tree presented here were generated by culling taxa from a larger tree that included 306 unique sequences obtained from 315 individuals representing 11 of 15 A. sagrei group species. Numbers above nodes represent posterior probabilities obtained from Bayesian analysis. Boldface numbers below nodes represent bootstrap values obtained from 200 bootstrapped data sets analyzed via maximum parsimony.

species [Peterson et al. 1999, Peterson and Holt 2003]). Results from these studies suggest that conservatism, where taxa that are more closely related possess characteristics that are more similar (Harvey and Pagel 1991, Lord et al. 1995, Webb et al. 2002), is a frequent occurrence and characterizes evolutionary patterns of niche diversification. Alternatively, other studies have included deeper evolutionary divergence by using phylogenetic methods to reconstruct the niche of ancestral taxa (Rice et al. 2003, Graham et al. 2004). Results from these studies suggest that niche conservatism might not be a consistent characteristic during diversification. Both of these approaches have limitations: the former approach has limited scope and only applies to evolutionary events in the recent past, whereas the accuracy of the latter approach is questionable because a number of studies have demonstrated that phylogenetically derived estimates of ancestral states will in many cases have low accuracy and extremely high uncertainty (Schluter et al. 1997, Losos 1999, Martins 1999, Oakley and Cunningham 2000, Webster and Purvis 2002). We propose an alternative approach, similar to one applied by Rice et al. (2003), that makes full use of phylogenetic information and thus permits inferences beyond comparison of closely related taxa, while

Ecology Special Issue

avoiding the pitfalls of ancestor reconstruction. Specifically, following the logic of Harvey and Pagel’s (1991) ‘‘non-directional approach,’’ we examine the extent to which niche similarity among extant species is a function of phylogenetic similarity. Like the widely used independent-contrasts method (Felsenstein 1985), this approach allows the integration of phylogenetic information into comparative analyses without requiring inference concerning the characteristics of hypothetical ancestral taxa. Caribbean lizards of the genus Anolis are a particularly good group for such studies (see Plate 1). Anoles are abundant and diverse on Caribbean islands, with as many as 57 species on a single island (Cuba) and up to 11 species occurring sympatrically, and their ecology has been extensively studied (reviewed in Losos [1994] and Roughgarden [1995]). Moreover, recent research has established a firm phylogenetic framework for the Caribbean anole radiation (Poe 2004, Nicholson et al. 2005). Our focus in this study is on the A. sagrei species group on Cuba. This clade contains 15 species, all but one of which use similar microhabitats, occurring on tree trunks and other broad surfaces low to the ground, and using the ground extensively for foraging and intraspecific interactions (the ‘‘trunk ground ecomorph’’ niche of Williams [1983]). Several species occur widely throughout Cuba, whereas others have restricted distributions; as many as four occur sympatrically (Losos et al. 2003). Recent phylogenetic work (Glor 2004) has provided the phylogenetic framework for the investigation of the evolution of the niche and community composition. In this study, we examine the evolution of niche characteristics in species within the A. sagrei group on Cuba. We characterize the broad-scale environmental components of the niche using ENM algorithms and GIS data. We then incorporate the results of these analyses with information on the phylogenetic relationships within the A. sagrei group to examine patterns of niche diversification within this island group. METHODS Phylogenetic relationships among members of the Anolis sagrei group We obtained an mtDNA phylogeny for the A. sagrei species group from Glor (2004). This tree was reconstructed from 306 unique sequences obtained from 315 individuals representing 11 of 16 A. sagrei group species and including extensive intraspecific sampling within most widespread species (Glor 2004). We treat A. allogus as two separate species, A. allogus (east) and A. allogus (west), because they are highly divergent genetically and do not form a clade (Glor 2004) (Fig. 1); thus the discrepancy between the 15 recognized species and the 16 species referred to in our analyses. Maximumparsimony analysis implemented by PAUP* 4.0b10 (Swofford 2002) and Bayesian analysis implemented in MrBayes 3.0 (Huelsenbeck and Ronquist 2001) yielded

July 2006

NICHE EVOLUTION IN ANOLIS LIZARDS

S31

TABLE 1. Results of niche modeling predictions for species in the Anolis sagrei group.

Species A. A. A. A. A. A. A. A. A. A. A.

N 

ahli allogus (east) allogus (west) bremeri homolechis jubar mestrei ophiolepis quadriocellifer rubribarbus sagrei

GARP external accuracy mean (range)

16 65 23 25 146 63 35 43 24 21 124

0.748 0.542 0.500 0.545 0.962 0.594 0.597 0.814 0.735 0.547 0.953

(0.500–0.999) (0.466–0.972) (0.500–0.500) (0.500–0.952) (0.951–0.972) (0.500–0.984) (0.500–0.993) (0.500–0.973) (0.500–0.992) (0.500–0.971) (0.937–0.968)

WhyWhere External accuracy 0.490 0.758 0.945 0.845 0.595 0.736 0.958 0.756 0.981 0.971 0.627

Z statistic 31.15*** 23.19*** 29.26*** 30.40*** 18.50*** 22.00*** 30.13*** 19.16*** 30.73*** 30.23*** 17.80***

***P , 0.001 for all of the WhyWhere predictions.   Number of localities for each species.

congruent and well-supported topologies (Fig. 1). Phylogenetic analysis of a nuclear DNA fragment (the third intron of the rhodopsin-encoding gene) for a subset of the taxa included in the mtDNA phylogeny also yields a topology that is concordant with all of the nodes presented in Fig. 1 (Glor 2004). We then converted the Bayesian mtDNA tree into an ultrametric form using Sanderson’s (2002) penalizedlikelihood approach, as implemented by the program r8s (Sanderson 2003). A smoothing value for the penalizedlikelihood analysis was determined via cross-validation. Following conversion of the tree into an ultrametric format, taxa within monophyletic groups were pruned until a single representative of each species remained (Fig. 1). In the case of A. allogus, we retained two individuals representing the genetically distinct eastern and western populations. This pruned tree was then used to derive patristic distances among all pairwise taxonomic comparisons. Niche characteristics of members of the Anolis sagrei group Ecological niche modeling provides the ability to estimate the niche of a species based on known species localities and environmental parameters characterized in GIS data sets. This estimation is then used to predict the potential geographic distribution of a species based on the same GIS data (Peterson 2001). These predictions are based on broad-scale environmental data and do not account for microhabitat characteristics. Consequently, microhabitat partitioning among sympatric anoles, which has been an important component in the community structure and evolutionary diversification of anoles (e.g., Williams 1983, Losos et al. 2003), is not considered in the ENM algorithms. Similar broad-scale ENM analyses have suggested a high predictive ability for geographic distribution with the ENM algorithms (Peterson 2001, 2003, Peterson and Vieglais 2001, Oberhauser and Peterson 2003, Peterson and Robins 2003, Illoldi-Rangel et al. 2004, Peterson et al. 2004). In other words, the broad-scale ENM approach is useful

for predicting whether a species will occur in a particular region, but is not insightful regarding microhabitat differences among co-occurring species. We compiled locality information for 11 species in the A. sagrei group from natural history museum collection records (American Museum of Natural History; California Academy of Sciences; Field Museum; Museum of Comparative Zoology, Harvard University; Museum of Vertebrate Zoology, University of California-Berkeley; Smithsonian National Museum of Natural History; University of Kansas Natural History Museum) and published data (Schwartz and Henderson 1991, Rodrı´ guezSchettino 1999) (Table 1). The five species for which phylogenetic data are not available are known from an extremely limited number of individuals or localities (Rodrı´ guez-Schettino 1999). We predicted the niche of each species in the A. sagrei group using the WhyWhere niche modeling program (Stockwell 2006; software available online).5 The WhyWhere algorithm (newly available in June 2004) affords greater predictive ability and decreased computational time compared to the commonly used GIS-based Genetic Algorithm for Rule-Set Prediction (GARP) ENM application (Stockwell and Peters 1999). To achieve results, WhyWhere converts environmental data layers into multicolor images and applies a data-mining approach to image processing methods to sort through large amounts of data to determine the variables that best predict species occurrences. Testing of the predictive ability of each model is performed by calculating the accuracy of the model based on species presence data and randomly generated pseudo-absence points within a specified geographic region (similar to GARP) (Stockwell and Peters 1999), in this case Cuba. We used the WhyWhere ENM algorithm, 184 terrestrial environmental data layers (see Stockwell [2006] for information on data layers), and georeferenced species occurrence data to predict the niche of each species in the A. sagrei group. These ENM 5

hhttp://biodi.sdsc.edu/ww_home.htmli

S32

JASON H. KNOUFT ET AL.

Ecology Special Issue

FIG. 2. Potential distributions of species in the Anolis sagrei group based on WhyWhere and Genetic Algorithm for Rule-Set Prediction (GARP) ecological niche modeling predictions. Points represent locality data used in model development and testing for each species. Species designations are as follows: A, A. ahli; B, A. allogus (east); C, A. allogus (west); D, A. bremeri; E, A. homolechis; F, A. jubar; G, A. mestrei; H, A. ophiolepis; I, A. quadriocellifer; J, A. rubribarbus; K, A. sagrei.

predictions reflect the potential geographic distribution of each species on Cuba based on the ENM algorithms and the GIS data used to construct each prediction. Each prediction was calculated based on a 0.18 resolution per grid cell, a 0.5 occurrence cut for each prediction, and a 1.28 Z score termination condition. For each species, 50% of the locality data were used for model training (internal accuracy), while 50% were held back for testing of model accuracy (external accuracy). We determined accuracy of each prediction using the same protocol as GARP (Stockwell and Peters 1999). Although WhyWhere has been demonstrated to generate predictions with higher accuracy than the frequently used GARP ENM (Stockwell 2006), we also modeled the niche of each species using GARP for

qualitative assessment of the WhyWhere application. Both GARP and WhyWhere were developed by the same person/research group, and both applications calculate model accuracy in a similar manner (Stockwell and Peters 1999, Stockwell 2006). Thus, the accuracy of predictions generated by the two ENM algorithms are qualitatively comparable. Using the Anolis locality data and the GARP ENM application, we again predicted the niche of each species in the A. sagrei group. GIS data sets were at a 0.18 resolution and included layers describing topography (elevation, slope, aspect, flow accumulation, and flow direction) and climate (annual means of total, minimum, and maximum temperature, precipitation, solar radiation, vapor pressure, and wet days). We developed 100

July 2006

NICHE EVOLUTION IN ANOLIS LIZARDS

S33

FIG. 2. Continued.

niche models for each species based on locality data. The GARP algorithm was run for 1000 iterations or until a convergence limit of 0.1 was achieved for each species. During model development, 50% of the localities were used for model training, while 50% of the localities were held back to test model accuracy. Using the best-subset selection criteria (Anderson et al. 2003), we chose 20 models that had an omission error of ,5% based on the localities used to test each model. From these 20 ‘‘best’’ models, we then selected the 10 models that exhibited an intrinsic commission index closest to the mean intrinsic commission index for the 20 models (Anderson et al. 2003; similar to Oberhauser and Peterson [2003]). We then imported these models into GIS software (DIVAGIS [Hijmans et al. 2001]), identified the areas of predicted occurrence that were present in all 10 models, and used this area as our prediction for each species. The use of areas that only occur in the predictions of all 10 models is a conservative approach; however, this methodological choice still resulted in predictions that encompassed all of the locality data for each species.

We investigated the relationship between niche similarity and phylogenetic relatedness in a geographic context by examining the frequency that known localities from a particular species fall within the ENM predicted geographic distribution of another species (similar to Peterson et al. [1999]). Using a binomial probability distribution, we determined if the number of times that the occurrence data points of one species overlapped the predicted distribution of another species was nonrandom. For the binomial probability calculation, the null expectation was that the percentage of actual occurrence data points that fell within the predicted range of the other species would correspond to the proportion of Cuba lying within the predicted distribution range. For each species pair, we conducted reciprocal tests for each species. In sister species pairs, greater than expected overlap by both species indicates niche conservatism, and less than expected overlap in both species indicates niche divergence. Also, in sister species pairs, greater than expected overlap by one species and less than expected overlap by the other could suggest a case of niche specialization. In more distantly

S34

JASON H. KNOUFT ET AL.

Ecology Special Issue

TABLE 2. Pairwise comparisons of percentage of locality points of one Anolis species that fall within the predicted range of a second Anolis species. Species 2 Species 1 A. ahli A. allogus (east) A. allogus (west) A. bremeri A. homolechis A. jubar A. mestrei A. ophiolepis A. quadriocellifer A. rubribarbus A. sagrei

A. ahli 4.6, 4.3, 0.0, 6.5, 0.0, 0.0, 11.6, 0.0, 25.0, 10.2,

0.0 93.8 93.8 68.8 0.0 93.8 93.8 0.0 0.0 25.0

A. allogus (east)

8.7, 4.5, 59.4, 60.3, 0.0, 41.9, 54.2, 40.0, 39.8,

A. allogus (west)

6.2 10.8 95.4 69.2 0.0 73.8 0.0 56.9 86.2

100, 91.3 47.8, 95.7 6.3, 0.0 97.1, 95.7 58.1, 82.6 91.7, 4.3 0.0, 0.0 51.7, 95.7

A. bremeri

39.9, 14.3, 97.1, 44.2, 91.7, 0.0, 44.9,

91.3 12.0 100 100 26.1 0.0 100

A. homolechis

50.8, 94.3, 90.7, 66.7, 35.0, 70.3,

39.1 35.5 62.3 5.8 18.1 76.1

A. jubar

8.6, 30.2, 37.5, 40.0, 30.5,

0.0 42.9 0.0 12.7 57.1

Notes, Percentages in boldface represent cases in which a smaller than expected number of locality points of one species fall within the predicted range of the second species. Percentages in italics represent cases in which a greater than expected number of locality points of one species fall within the predicted range of the second species. In each cell, the value on the left represents the percentage of locality points from Species 1 that fall in the predicted range of Species 2. The value on the right represents the percentage of locality points from Species 2 that fall in the predicted range of Species 1.

related species pairs, greater than expected overlap could suggest either niche convergence or suggest that the species have both retained the ancestral condition (i.e., stasis). In addition to the geographic approach to niche overlap using species locality data and the predictions generated by the ENM algorithms, we also examined overlap of the ‘‘environmental envelopes’’ of species pairs using GIS-derived environmental data extracted from localities for each species. We generated the environmental envelope for each species based on data extracted from WorldClim Global Climate GIS data sets (30-second resolution; WorldClim interpolated global terrestrial climate surfaces, version 1.3; data and software available online)6 (Hijmans et al. 2004). The WorldClim data sets consisted of 19 bioclimatic variables including annual mean temperature, mean diurnal temperature range, isothermality, temperature seasonality, maximum temperature of warmest month, minimum temperature of coldest month, temperature annual range, mean temperature of wettest quarter, mean temperature of driest quarter, mean temperature of warmest quarter, mean temperature of coldest quarter, annual precipitation, precipitation of wettest month, precipitation of driest month, precipitation seasonality, precipitation of wettest quarter, precipitation of driest quarter, precipitation of warmest quarter, and precipitation of coldest quarter, with all temperatures reported in degrees Celsius and all precipitation amounts reported in millimeters. Environmental data for each species was compiled by importing species locality points into DIVA-GIS (Hijmans et al. 2001) to generate longitude–latitude layers for each species. Environmental data at each locality were then extracted from each GIS data set to provide 19 climatic measures 6

hhttp://biogeo.berkeley.edui

at each species locality. All climatic data were log10transformed to standardize data for statistical analyses. A principal-components analysis (PCA) was performed on the correlation matrix of transformed environmental data to generate an environmental envelope for each species. To generate the environmental envelope, the first two axes of the PCA for each species were plotted in x–y space using ArcGIS (version 9.0). A minimum convex polygon (MCP) was then calculated around the points for each species using the Hawth’s Tools extension in ArcGIS (available online).7 The area for each species MCP was then calculated in ArcGIS. The percentage niche overlap for each species pair was calculated as the ratio of the sum of the overlap areas in each species’ MCP to the total area of each species’ MCP, with the resulting quotient multiplied by 100 to yield a percentage. While the geographic approach applied to the ENM predictions and species locality data is useful for assessing similarities between individual species pairs, the environmental-envelope method allows for the examination of the relationship between niche similarity and phylogenetic similarity among all members of the clade. The environmental envelope could not be used to assess differences between individual species pairs, because we do not know the breadth of the environmental envelope for all of Cuba. Consequently, we could not apply a binomial probability calculation to the overlap between species pairs. We calculated the correlation between niche similarity (i.e., percentage envelope overlap) and phylogenetic similarity (patristic distance between taxa) among all species pairs using a Mantel test. A significant negative correlation indicates niche conservatism between closely related species and niche divergence between distantly related species. 7

hhttp://www.spatialecology.com/htoolsi

July 2006

NICHE EVOLUTION IN ANOLIS LIZARDS

S35

little overlap, and most of the cases of the greatest overlap are among distantly related species pairs.

TABLE 2. Extended.

DISCUSSION A. mestrei

48.8, 91.7, 0.0, 42.4,

82.9 11.4 0.0 97.1

A. ophiolepis

A. quadriocellifer

A. rubribarbus

41.7, 7.0 25.0, 14.0 66.9, 88.4

0.0, 0.0 7.6, 75.0

11.8, 50.0

RESULTS Niche characteristics of members of the Anolis sagrei group The modeled niche of each species served as an accurate predictor of the species’ distribution in all cases (Table 1, Fig. 2). The GARP application provided 10 ‘‘best’’ models. We used the average accuracy of these models for comparisons with WhyWhere model accuracy. Prediction accuracy was higher with the WhyWhere algorithm than with GARP for 7 of 11 species (Table 1), thus we used the WhyWhere predictions to assess individual species pair overlap. When locality data from one species were compared to the predicted distribution of a sister species using the WhyWhere predictions, significant cases of niche conservatism are recovered (Table 2). Additionally, several cases occurred in which distantly related species had niches that are more similar than would be expected by chance, whether resulting either from convergence or stasis (Table 2). There are also cases in which species pairs exhibit less overlap than expected by chance; however, whether this limited overlap is due to selectiondriven divergence or random diversification is unclear. In all of these previous scenarios, both members of the sister species pair exhibited the same reciprocal relationship. In addition, in one case, A. quadriocellifer and A. bremeri, the relationship was not reciprocal: although a significant percentage of locality data from A. bremeri does not fall within the A. quadriocellifer prediction, a significant percentage of locality data from A. quadriocellifer does fall with in the A. bremeri prediction. We regard this as indicating that A. quadriocellifer resides in a specialized component of the A. bremeri niche (Fig. 3). The first and second principal components explained 42.4% and 25.7% of the overall variance in the PCA, respectively (Table 3). The Mantel test examination of the relationship between percentage of environmental envelope overlap and phylogenetic similarity indicates no consistent pattern in the evolution of the species niche among species in the A. sagrei group (r ¼ 0.10, P ¼ 0.26) (Fig. 4). Indeed, the most closely related taxa show

A common finding related to trait evolution is that conservatism is the expected pattern during species diversification (Webb et al. 2002). In terms of niche evolution, this conservatism has been hypothesized to result from active, stabilizing selection (Lord et al. 1995), or from fixation of ancestral traits that limit the potential range of outcomes during niche evolution (Westoby et al. 1995; see review in Webb et al. [2002]). Initial ENM work examining niche overlap in species pairs separated by a geographic barrier supported the prediction that niche conservatism characterizes evolutionary diversification (Peterson 2001). However, more recent ENM work has suggested that patterns of niche evolution beyond sister taxa can be inconsistent and not conserved (Rice et al. 2003). Our results from the environmental-envelope overlap analysis are congruent with these more recent findings. No evidence of generalized niche conservatism exists for the A. sagrei group in Cuba. Overall, no relationship was found between phylogenetic and niche similarity. This results because some closely related species have greatly divergent niches, whereas some distantly related species are quite similar in their niches. A closer look at patterns of niche similarity using the ENM predictions and species locality data allows a more detailed level of inference. All combinations of phylogenetic relatedness and degree of niche overlap are seen in the data (Table 2). Examples of closely related taxa with significantly high niche overlap (e.g., A. bremeri–A. ophiolepis, A. bremeri–A. sagrei, A. mestrei–A. ophiolepis) indicate niche conservatism. Additionally, high levels of niche differentiation are seen in close relatives that exhibit significantly little niche overlap (e.g., A. allogus (east)–A. allogus (west), A. homolechis–A. jubar).

FIG. 3. (A) Predicted distribution of A. quadriocellifer with actual localities (points) of A. bremeri. (B) Predicted distribution of A. bremeri with actual localities (points) of A. quadriocellifer.

S36

JASON H. KNOUFT ET AL.

TABLE 3. PC1 and PC2 loadings from principal-components analysis of environmental variables for species in the Anolis sagrei group.

Environmental variable

PC1 loadings

PC2 loadings

Elevation Mean annual temperature Mean diurnal temperature Isothermality Temperature seasonality Maximum temperature, warmest month Minimum temperature, coldest month Temperature annual range Mean temperature, wettest quarter Mean temperature, driest quarter Mean temperature, warmest quarter Mean temperature, coldest quarter Annual precipitation Precipitation, wettest month Precipitation, driest month Precipitation seasonality Precipitation wettest quarter Precipitation, driest quarter Precipitation, warmest quarter Precipitation, coldest quarter


0.738 0.921
0.006 0.023
0.099 0.867 0.801
0.020 0.803 0.837 0.891 0.892
0.794
0.709
0.501 0.301
0.684
0.565
0.520
0.461


0.061
0.145 0.731
0.061 0.562 0.148
0.537 0.864 0.337
0.427
0.038
0.330
0.027 0.045
0.718 0.835 0.216
0.705 0.644
0.796

Many cases of low niche overlap among distantly related taxa are also apparent (e.g., A. allogus (east)–A. bremeri, A. mestrei–A. rubribarbus). In these cases, examination of species’ niches in a phylogenetic context makes evolutionary interpretation, either conservatism or divergence, obvious. By contrast, the evolutionary explanation for similar niches among distantly related taxa is not so clear-cut: long-term conservatism or convergence are both possibilities. Although the natural history of many members of the A. sagrei group is poorly studied, there appear to be at least four axes along which the members of this group partition habitat: heliothermy (sun vs. shade-loving), forest type (xeric vs. mesic), habitat openness (woodland vs. open habitat), and substrate type (trunks vs. rocks) (Ruibal 1961, Ruibal and Williams 1961, Schwartz and Henderson 1991, Rodrı´ guez-Schettino 1999, Losos et al. 2003). Only one of these axes (i.e., forest type) involves a geographic scale of niche partitioning that is appropriate for the methods discussed here. A previous study (Glor 2004) suggested that divergence along this axis, and with respect to substrate type, has occurred repeatedly, perhaps accounting for the observed lack of conservatism. The two other axes (heliothermy and habitat openness) meanwhile, appear more conserved in the sense that they characterize large deeply divergent clades (Glor 2004). Consequently, further study may reveal a greater degree of niche conservatism, particularly along these axes, than we discuss. Our approach differs from previous niche modeling exercises (Rice et al. 2003, Graham et al. 2004) in which phylogenetic methods have been used to infer the niche of hypothetical ancestral taxa. Because in many cases ancestral reconstructions probably have low accuracy (Schluter et al. 1997, Losos 1999, Martins 1999, Oakley

Ecology Special Issue

and Cunningham 2000, Webster and Purvis 2002), we have avoided this approach by focusing only on the niches of extant species in the context of their phylogenetic similarity. The trade-off of this ‘‘nondirectional’’ (sensu Harvey and Pagel 1991) approach, however, is that we are less able to make statements about the direction in which evolution has occurred. In particular, we have a number of cases in which nonsister taxa, and indeed sometimes taxa that are only distantly related, have significantly similar niches. Two processes could account for such cases. On the one hand, two species might have retained the ancestral niche through the course of time; in other words, their conserved niche similarity would be an example of evolutionary stasis. Alternatively, the two species might have independently derived the same niche through convergent evolution. Without inferring ancestral niches, these two possibilities are difficult to distinguish. One other interesting situation occurred between the sister taxa A. quadriocellifer and A. bremeri, in which a significant proportion of A. quadriocellifer localities fall within the A. bremeri prediction, whereas the reciprocal pattern is not exhibited. We interpret this to indicate that the A. quadriocellifer niche is more specialized than, and nested within, the A. bremeri niche. Again, however, the direction in which this evolutionary change occurred, whether from generalist to specialist or vice versa, is difficult to discern. Nevertheless, the sister taxon to the A. quadriocellifer–A. bremeri species pair (Fig. 1) is A. sagrei, a species that has a broad niche similar to that of A. bremeri (Table 2). Consequently, the parsimonious conclusion is that a broader niche is ancestral and the narrower niche of A. quadriocellifer is derived in this species. Despite methodological differences, our results are broadly congruent with previous studies on dendrobatid frogs (Graham et al. 2004) and Aphelocoma jays (Rice et

FIG. 4. Relationship between patristic distance and ecological-envelope overlap in species pairs of the Anolis sagrei group.

July 2006

NICHE EVOLUTION IN ANOLIS LIZARDS

S37

PLATE 1. Example of a Cuban trunk-ground anole lizard species (Anolis rubribarbus) from Guantanamo Province at Sendero Natural et Recreo de Nibujon, Cuba. Photo credit: R. E. Glor.

al. 2003): in all three cases, many instances of closely related taxa that diverge greatly in niche have been discovered. Thus, the balance of evidence to date provides little consistent evidence that environmental niches are phylogenetically conservative. The main counterexample to date is a study of Mexican bird species that showed that allopatric populations on either side of the Isthmus of Tehuantepec tend to exhibit niche conservatism (Peterson et al. 1999, Peterson and Holt 2003). Although it is tempting to suggest that the focus on allopatric populations explains these discrepant findings, this conclusion is unwarranted, as closely related allopatric sister taxa show divergence in dendrobatid frogs and anoles. The broad-scale environmental data used in ENM algorithms successfully predict anole occurrence on a regional scale. Investigating the role that adaptation to different climatic niches has played in anole evolution will contribute importantly to understanding the genesis of the incredible diversity of this species-rich clade. Nonetheless, this approach has limitations. First, the resolution of the GIS data limits analysis to regions, rather than specific localities. Thus, this approach can only investigate environmental determinants of regional co-occurrence, rather than true sympatry. Second, current species distributions may be a result of recent

allopatric speciation and not a consequence of species distributions actually tracking climatic conditions. This can present an interpretive dilemma for sister species that appear to diverge in niche characteristics. This scenario is displayed by only one species pair in our data set (A. jubar–A. homolechis). Finally, an important component of anole community ecology and evolution is partitioning of habitats (e.g., cool/hot; high/low) within a site (Schoener 1968, Williams 1983, Losos et al. 2003), a scale of habitat far too small to be detected by these sorts of data. Clearly, a next step in niche modeling will be the integration, of broad- and fine-scale niche characteristics to elucidate the determinants of local and regional distributions and habitat use. ACKNOWLEDGMENTS We greatly appreciate the time and efforts of the Curators and Collection Managers at the American Museum of Natural History, the California Academy of Sciences, Field Museum, the Museum of Comparative Zoology, Harvard University, the Museum of Vertebrate Zoology, University of California at Berkeley, the Smithsonian National Museum of Natural History, and the University of Kansas Natural History Museum who assisted in providing locality data for species in this study. We also greatly appreciate the efforts of D. Ullman who georeferenced a large percentage of these data. J. H. Knouft was supported by a grant from the National Science Foundation (DBI-204144).

S38

JASON H. KNOUFT ET AL. LITERATURE CITED

Anderson, R. P., D. Lew, and A. T. Peterson. 2003. Evaluating predictive models of species’ distributions: criteria for selecting optimal models. Ecological Modelling 162:211–232. Chase, J. M., and M. A. Leibold. 2003. Ecological niches: linking classical and contemporary approaches. The University of Chicago Press, Chicago, Illinois, USA. Felsenstein, J. 1985. Phylogenies and the comparative method. American Naturalist 125:1–15. Glor, R. E. 2004. Species diversification in an adaptive radiation: a comparative analysis of Anolis lizards. Dissertation. Washington University, St. Louis, Missouri, USA. Graham, C. H., S. R. Ron, J. C. Santos, C. J. Schneider, and C. Moritz. 2004. Integrating phylogenetics and environmental niche models to explore speciation mechanisms in dendrobatid frogs. Evolution 58:1781–1793. Guisan, A., and N. E. Zimmermann. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135: 147–186. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Hijmans, R. J., L. Guarino, M. Cruz, and E. Rojas. 2001. Computer tools for spatial analysis of plant genetic resources data: 1. DIVA-GIS. Plant Genetic Resources Newsletter 127: 15–19. Huelsenbeck, J. P., and F. Ronquist. 2001. MRBAYES: Bayesian inference of phylogenetic trees. Bioinformatics 17:754–755. Hutchinson, G. E. 1957. Concluding remarks. Cold Spring Harbor Symposia on Quantitative Biology 22:415–442. Illoldi-Rangel, P., V. Sanchez-Cordero, and A. T. Peterson. 2004. Predicting distributions of Mexican mammals using ecological niche modeling. Journal of Mammalogy 85:658–662. Lord, J., M. Westoby, and M. Leishman. 1995. Seed size and phylogeny in six temperate floras: constraints, niche conservatism, and adaptation. American Naturalist 146:349–364. Losos, J. B. 1994. Integrative approaches to evolutionary ecology: Anolis lizards as model systems. Annual Review of Ecology and Systematics 25:467–493. Losos, J. B. 1999. Uncertainty in the reconstruction of ancestral character states and limitations on the use of phylogenetic comparative methods. Animal Behaviour 58:1319–1324. Losos, J. B., M. Leal, R. E. Glor, K. de Queiroz, P. E. Hertz, L. Rodriguez Schettı´ no, A. Chamizo Lara, T. R. Jackman, and A. Larson. 2003. Niche lability in the evolution of a Caribbean lizard community. Nature 424:542–545. MacArthur, R. H. 1972. Geographical ecology. Harper and Row, New York, New York, USA. Martins, E. P. 1999. Estimation of ancestral states of continuous characters: a computer simulation study. Systematic Biology 48:642–650. Nicholson, K. E., R. E. Glor, J. J. Kolbe, A. Larson, S. B. Hedges, and J. B. Losos. 2005. Mainland colonization by island lizards. Journal of Biogeography 32:929–938. Oakely, T. H., and C. W. Cunningham. 2000. Independent contrasts succeed where ancestor reconstruction fails in a known bacteriophage phylogeny. Evolution 54:397–405. Oberhauser, K., and A. T. Peterson. 2003. Modeling current and future potential wintering distributions of eastern North American monarch butterflies. Proceedings of the National Academy of Sciences (USA) 100:14063–14068. Peterson, A. T. 2001. Predicting species’ geographic distributions based on ecological niche modeling. Condor 103:599–605. Peterson, A. T. 2003. Predicting the geography of species’ invasion via ecological niche modeling. Quarterly Review of Biology 78:419–433. Peterson, A. T., and R. D. Holt. 2003. Niche differentiation in Mexican birds: using point occurrences to detect ecological innovation. Ecology Letters 6:774–782. Peterson, A. T., E. Martinez-Meyer, C. Gonzalez-Salazar, and P.W. Hall. 2004. Modeled climate change effects on

Ecology Special Issue

distributions of Canadian butterfly species. Canadian Journal of Zoology 82:851–858. Peterson, A. T., and C. R. Robins. 2003. Using ecological-niche modeling to predict barred owl invasions with implication for spotted owl conservation. Conservation Biology 17:1161–1165. Peterson, A. T., J. Sobero´n, and V. Sanchez-Cordero. 1999. Conservatism of ecological niches in evolutionary time. Science 285:1265–1267. Peterson, A. T., and D. A. Vieglais. 2001. Predicting species invasions using ecological niche modeling: new approaches from bioinformatics attack a pressing problem. Bioscience 51:363–371. Poe, S. 2004. Phylogeny of anoles. Herpetological Monographs 18:37–89. Rice, N. H., E. Martinez-Meyer, and A. T. Peterson. 2003. Ecological niche differentiation in the Aphelocoma jays: a phylogenetic perspective. Biological Journal of the Linnean Society 80:369–383. Rodrı´ guez-Schettino, L. 1999. The iguanid lizards of Cuba. University of Florida Press, Gainesville, Florida, USA. Roughgarden, J. 1995. Anolis lizards of the Caribbean: ecology, evolution, and plate tectonics. Oxford University Press, New York, New York. Ruibal, R. 1961. Thermal relations of five species of tropical lizards. Evolution 15:98–111. Ruibal, R., and E. E. Williams. 1961. The taxonomy of the Anolis homolechis complex of Cuba. Bulletin of the Museum of Comparative Zoology 125:211–246. Sanderson, M. J. 2002. Estimating absolute rates of molecular evolution and divergence times: a penalized likelihood approach. Molecular Biology and Evolution 19:101–109. Sanderson, M. J. 2003. r8s: Inferring absolute rates of molecular evolution and divergence times in the absence of a molecular clock. Bioinformatics 19:301–302. Schluter, D., T. Price, A. O. Mooers, and D. Ludwig. 1997. Likelihood of ancestor states in adaptive radiation. Evolution 51:1699–1711. Schoener, T. W. 1968. The Anolis lizards of Bimini: resource partitioning in a complex fauna. Ecology 49:704–726. Schwartz, A., and R. W. Henderson. 1991. Amphibians and reptiles of the West Indies: descriptions, distributions, and natural history. University of Florida Press, Gainesville, Florida, USA. Sobero´n, J., and A. T. Peterson. 2005. Interpretation of models of fundamental ecological niches and species’ distributional areas. Biodiversity Informatics 2:1–10. Stockwell, D. R. B. 2006. Improving ecological niche models by data mining large environmental datasets for surrogate models. Ecological Modelling 192:188–196. Stockwell, D. R. B., and D. Peters. 1999. The GARP modeling system: problems and solutions to automated spatial prediction. International Journal of Geographical Information Science 13:143–158. Swofford, D. 2002. PAUP*. Phylogenetic analysis using parsimony (*and other methods). Sinauer, Sunderland, Massachusetts, USA. Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Webster, A. J., and A. Purvis. 2002. Testing the accuracy of methods for reconstructing ancestral states of continuous characters. Proceedings of the Royal Society of London B 269:143–149. Westoby, M., M. Leishman, and J. Lord. 1995. On misinterpreting the phylogenetic correction. Journal of Ecology 83:531–534. Williams, E. E. 1983. Ecomorphs, faunas, island size, and diverse end points in island radiations of Anolis. Pages 326– 370 in R. B. Huey, E. R. Pianka, and T. W. Schoener, editors. Lizard ecology: studies of a model organism. Harvard University Press, Cambridge, Massachusetts, USA.

Ecology, 87(7) Supplement, 2006, pp. S39–S49 Ó 2006 by the Ecological Society of America

PHYLOGENY AND THE HIERARCHICAL ORGANIZATION OF PLANT DIVERSITY JONATHAN SILVERTOWN,1,4 MIKE DODD,1 DAVID GOWING,1 CLARE LAWSON,2 2

AND

KEVIN MCCONWAY3

1 Department of Biological Sciences, The Open University, Walton Hall, Milton Keynes, MK7 6AA UK Centre for Agri-Environmental Research, School of Agriculture, Policy and Development, The University of Reading, Earley Gate, P.O. Box 237, Reading, RG6 6AR UK 3 Department of Statistics, The Open University, Walton Hall, Milton Keynes, MK7 6AA UK

Abstract. R. H. Whittaker’s idea that plant diversity can be divided into a hierarchy of spatial components from a at the within-habitat scale through b for the turnover of species between habitats to c along regional gradients implies the underlying existence of a, b, and c niches. We explore the hypothesis that the evolution of a, b, and c niches is also hierarchical, with traits that define the a niche being labile, while those defining b and c niches are conservative. At the a level we find support for the hypothesis in the lack of close significant phylogenetic relationship between meadow species that have similar a niches. In a second test, a niche overlap based on a variety of traits is compared between congeners and noncongeners in several communities; here, too, there is no evidence of a correlation between a niche and phylogeny. To test whether b and c niches evolve conservatively, we reconstructed the evolution of relevant traits on evolutionary trees for 14 different clades. Tests against null models revealed a number of instances, including some in island radiations, in which habitat (b niche) and elevational maximum (an aspect of the c niche) showed evolutionary conservatism. Key words: coexistence; community assembly; diversity; evolutionary lability; geographical range; habitat; hydrology; niche overlap; plant community; plant phylogeny.

INTRODUCTION R. H. Whittaker (1975) proposed that diversity should be analyzed at a hierarchy of spatial scales. At the local scale, a diversity represents the number of species found within a habitat. These species occur in sufficient proximity to interact with one another. At intermediate scales, b diversity quantifies the turnover in species that takes place between habitats or along environmental gradients. At a still wider scale, c diversity is the species diversity of a region. Of a and b diversity, Whittaker (1975:119) wrote that they ‘‘will be recognized as consequences of niche differentiation and habitat diversification of species, respectively.’’ Not long afterwards, Pickett and Bazzaz (1978) explicitly referred to these concepts as the a niche and the b niche. (A glossary of terms is given in Table 1.) Although these terms were not widely adopted when they were first introduced, recent research on the phylogenetic structure of ecological communities suggests that the distinction between a and b niches should receive greater attention, because the hierarchical relationship between them might reflect the hierarchical structure of evolutionary trees (Fig.1). For a given phylogeny, heritable traits that vary freely among the terminals (tips) of the tree are likely to be Manuscript received 18 January 2005; revised 9 September 2005; accepted 13 September 2005. Corresponding Editor (ad hoc): C. O. Webb. For reprints of this Special Issue, see footnote 1, p. S1. 4 E-mail: [email protected]

evolutionarily labile. If these traits determine a species’ niche, community structure will appear free of phylogenetic conservatism. In contrast, traits that vary little among terminals on the same tree indicate that their evolution is likely to be more conservative. Niche-related traits of this kind can potentially produce a phylogenetic signal in the structure of ecological communities. Whether conservatively evolving niche traits actually do produce this signal depends upon the ecological processes of community assembly that determine how many representatives of a conservatively evolving clade are present. In areas of high endemism such the Cape Floristic Region of South Africa or oceanic archipelagos such as Hawaii, some communities might have been assembled, at least in part, by adaptive radiation in situ. This is where we might expect to find recent evolutionary events influencing community structure most strongly. However, this form of community assembly is a rare event, and most plant communities, including some on islands such as those in Macaronesia (Santos 2001), have been assembled from plants with quite disparate phylogenetic histories (Pennington et al. 2004, Pennington and Dick 2004). We ask two central questions. First, do ecological traits evolve in a conservative manner? Second, is there a difference in evolutionary lability between the traits that underlie a and b niches? Recent studies of the phylogenetic distribution of ecological traits have tended to emphasize the conservative nature of plant trait evolution and suggested that this influences community assembly (Tofts and Silvertown 2000, Webb 2000,

S39

S40

JONATHAN SILVERTOWN ET AL.

Ecology Special Issue

TABLE 1. Definitions of terms used in the text. Term a niche b niche c niche Community Habitat Lability Niche Niche trait Realized niche

Definition

Source

The region of a species’ realized niche corresponding to species diversity at the local (a) scale where interactions among species occur The region of a species’ niche that corresponds to the habitat(s) where it is found; equivalent to the ‘‘habitat niche’’ Grubb (1977) The geographical range of a species The collection of species that predictably co-occurs within a particular type of habitat The kind of environment where a species occurs, defined largely by physical conditions; note that conditions will usually be influenced by organisms as well as physical factors, but direct interactions among organisms are not used to define habitats The property of evolutionary changeability in a trait An n-dimensional hypervolume defined by axes of resource use and/or environmental conditions and within which populations of a species are able to maintain a long-term average net reproductive rate
1 A measurable property of a species, by which its niche (a, b, or c ) can be defined The region of its niche that a species is able to occupy in the presence of interspecific competition and natural enemies

2, 5 2, 3 6 2

1, 4

1

Sources: 1, Hutchinson (1957); 2, Whittaker (1975); 3, Pickett and Bazzaz (1978); 4, Chase and Leibold (2003); 5, Silvertown (2004); 6, Silvertown et al. (2006).

Prinzing et al. 2001, Webb et al. 2002, Ackerly 2003, 2004, Chazdon et al. 2003). One example of this pattern is the long-standing observation that, in many communities, there is a higher ratio of species per genus than would be expected if communities were assembled by random draws from the species pool (e.g., Williams 1964). If congeneric species are overrepresented in communities, then it follows that they must share ecological traits that influence community assembly and that these traits evolve more slowly than the rate of appearance of new species. Other studies, however, suggest that some traits that influence community structure do not evolve conservatively. Cavender-Bares et al. (2004) detected labile evolution in the soil moisture tolerances of North American oak species and found that these species segregated along soil moisture gradients. Silvertown et al. (1999) found that plant species in English meadow grasslands also segregated on hydrological gradients and later reported that there is no correlation between the ecological distance between species in hydrological niche space and their phylogenetic distance as measured by the evolution of the rbcL gene (Silvertown et al. 2006). How can these data be reconciled with the many other examples of the conservative evolution of ecological traits? Silvertown et al. (2006) suggested that the apparent contradiction between the lack of phylogenetic signal in their data, which implies evolutionary lability in hydrological niches, and contrary findings by other authors implying conservative evolution in some traits could be explained if the traits have different evolutionary lability. They proposed that habitat-determining traits that influence b diversity, and which may be said to define the b niche (Pickett and Bazzaz 1978), evolve conservatively. By contrast, traits involved in coexistence and that influence a diversity, defining the a niche, are evolutionarily labile. Such a pattern could arise if, as most theories of coexistence demand (Chesson 2000), species must differ from each other in order to coexist. The

corollary of this is that a niches and coexistence will necessarily be determined by labile traits. In short, Silvertown et al. (2006) proposed that competing species must share b niches in order to occur in the same habitat, but they must have different a niches in order to coexist. Silvertown et al. (2006) proposed that, by extension of the relationship between a and b niches and Whittaker’s (1975) a and b diversity, the geographical range of a species can be regarded as its c niche. Thus, there is a hierarchy of three niche levels with c at the top (Fig. 1). The little evidence that is so far available suggests that b niche traits are evolutionarily conservative; data pertaining to the evolution of the c niche are even more sparse. Prinzing et al. (2001) analyzed the niches of European plant species using Ellenberg indicator values (Ellenberg 1979, Ellenberg et al. 1991) and found strong evidence of evolutionary conservatism. These values were devised to quantify on an ordinal scale where different plant species are found in central Europe along major environmental axes, such as soil moisture, pH, light, and soil fertility. Several studies have found that Ellenberg values are stable traits that consistently predict the b niche of species across Europe more generally (Thompson et al. 1993, Hill et al. 2000, Schaffers and Sykora 2000, Prinzing et. al. 2002). Ellenberg values can be regarded as b niche traits, because they refer to large-scale environmental gradients. However, since a niches are nested within b niches, some correlation between traits like soil moisture tolerance is to be expected. Ackerly (2004) examined phylogenetic conservatism in the evolution of leaf traits that are associated with adaptation to Mediterranean climates in California chaparral habitat. These are a good example of traits associated with the b niche. Specific leaf area was significantly conserved in all four families analyzed, and leaf size in three. These results suggest that sclerophylly and other leaf traits associated with Mediterranean habitats evolved before California chaparral was colo-

July 2006

HIERARCHY OF PLANT DIVERSITY

nized, supporting the view that the b niche traits evolve in a conservative manner. Very little evidence is available concerning the evolution of c niches. Qian and Ricklefs (2004) found that the latitudinal ranges, and hence c niches, of 57 plant genera with disjunct distributions in North America and Asia were correlated between continents, suggesting that the genera had highly conserved c niches that dated to before the origin of the disjunctions, perhaps 18 million years ago in the case of woody species. How typical this result will prove to be of c niches in general is not clear at present. Other studies have examined the degree of range overlap between members of the same clade (Barraclough and Vogler 2000, Graham et al. 2004). Barraclough and Vogler (2000) found that in a range of vertebrate and insect phylogenies range overlap was low between recently diverged taxa, but increased with time since divergence. This indicates that speciation occurs more often in allopatry (no range overlap) than sympatry, but does not directly address the issue of c niche evolution, since ranges can be split (i.e., become allopatric) by the appearance of environmental barriers, without any need for evolutionary change in the c niche. In this paper we present four new lines of evidence that have a bearing on the evolutionary conservatism of a, b, and c niche traits. Each piece of evidence applies a different kind of test, as appropriate to the data available. First, we take a closer look at the English wet meadow communities in which Silvertown et al. (2006) found a niche traits to be evolutionarily labile. We perform a new analysis of the data in which we ask whether groups of specialists that are confined to particular subcommunity types are more closely related to one another than would be expected for randomly drawn samples from the same community. A smaller phylogenetic distance between specialists than between randomly drawn nonspecialists would imply evolutionary conservatism in the specialist group. Second, we reanalyze published data on various ecological traits in a number of plant communities to determine whether a niche overlap between congeners is greater or less than between noncongeners in the same community. Although we recognize that there is no consistent phylogenetic definition of a genus, and that some may be very old, congeners can usually be expected to be more closely related than species from other genera drawn from the same community. Trait variation ought therefore to be smaller between congeners than noncongeners for conservatively evolving traits that predate the origin of the genus, but similar for labile traits. Third, we conduct a test of a prediction derived from the hypothesis that b niche evolution is conservative by examining the number of inferred transitions in habitat (reflecting the b niche) within plant phylogenies for a sample of 12 independent clades. Changes of habitat should be fewer than expected by chance if b niche

S41

FIG. 1. A Venn diagram showing the nested hierarchy of a, b, and c niches superimposed upon a hypothetical phylogenetic tree. Note that the rectangles representing each kind of niche intersect the phylogenetic tree at progressively deeper levels from a to b to c niches, indicating earlier origin and greater conservatism.

evolution is conservative. As a subsidiary hypothesis, the expected patterns of conservatism should be stronger in clades that have evolved in continental areas, where available habitats are likely to have been already occupied by competitors, than in clades that have radiated on islands where most habitats were unoccupied. Finally, we apply the same kind of test to c niche evolution by optimizing maximum elevation (reflecting the c niche) onto phylogenies for two clades. METHODS Are habitat specialists phylogenetically clustered? Silvertown et al. (1999, 2001) previously analyzed the niche relationships of species in two mesotrophic grassland (meadow) plant communities classified as MG5 and MG8 by the British National Vegetation Classification (Rodwell 1992). Silvertown et al. (2001) suggested that some of the niche separation observed in these communities arose as deep as the split between monocots and eudicots, indicating that niche specialization occurs within particular clades. For the present study, we identified specialists from within each of the two community types using data from an extensive survey made by Gowing et al. (2002). This survey recorded an estimate of percent cover of all species present in 3904 1 3 1 m quadrats across 18 sites representative of MG5, MG8, and other floodplain hay meadow types in England. Quadrats were classified into 12 communities

S42

JONATHAN SILVERTOWN ET AL.

Ecology Special Issue

TABLE 2. Comparison of a niche overlaps between congeneric pairs of herb species in nine genera and mean overlaps between the congeners and species in other genera.

Genus Drosera Prosopis Dentaria Galium Senna Trillium Aster Helictotrichon Festuca

Total no. Overlap Overlap overlapping with with species congeners Comparison noncongeners 9 16 17 17 16 17 10 10 10

0.00 0.15 0.42 0.45 0.48 0.51 0.76 0.87 0.88

, , . ¼ . , . . .

0.60 0.38 0.37 0.45 0.31 0.52 0.44 0.83 0.83

a niche axes

Data source

water table gradient in a bog community soil moisture and nutrients in a desert community forest understory microtopography and light forest understory microtopography and light soil moisture and nutrients in a desert community forest understory microtopography and light phenology and microtopography in forest understory shoot phenology in grassland shoot phenology in grassland

5 4 1 1 4 1 2 3 3

Notes: There is no significant difference overall between the degree of overlap found between congeners and the overlap between noncongeners (Wilcoxon matched-pairs test, Z ¼ 0.42, P ¼ 0.67). Sources: 1, Mann and Shugart (1983); 2, Beatty (1984); 3, Sydes (1984); 4, Shaukat (1994); 5, Nordbakken (1996).

and subcommunities using the program TWINSPAN (Hill 1979). A group of seven species characteristic of the MG5a subtype of the MG5 community consisted of Trifolium pratense (Fabaceae), Rhinanthus minor (Scrophulariaceae), Dactylis glomerata (Poaceae), Prunella vulgaris (Lamiaceae), Heracleum sphodylium (Apiaceae), and Leucanthemum vulgare and Leontodon saxatilis (both Asteraceae). Within the MG8 community type, 10 species were identified as specialists associated with a Carex disticha subcommunity. These were Carex disticha, C. distans, and Eleocharis uniglumis (Cyperaceae); Senecio aquaticus and Bellis perennis (Asteraceae); Juncus inflexus, J. articulatus, and J. subnodulosus (Juncaceae); Festuca arundinaceae (Poaceae), and Trifolium fragiferum (Fabaceae). Specialists occurred more frequently in the designated subcommunity types (MG5a, MG8 C. disticha) than in any other of the 12 communities identified in the TWINSPAN analysis. Phylogenetic distances between all pairwise combinations of 52 species belonging to MG5 and MG8 communities were calculated by Silvertown et al. (2006). Distances were calculated as the sum of branch lengths connecting species in a tree fitted to rbcL sequences using maximum likelihood in PAUP* (Swofford 1996). For each of the two specialist groups, we computed the mean and variance of pairwise phylogenetic distances among members of the group and compared these with expected (null) distributions produced by randomization. Null distributions were derived by sampling groups of n species at random from the 52 species in the meadow species pool for which rbcL sequences are known, where n was the number of species in the specialist group. To avoid bias in the species pool caused by underrepresentation of sequences for Carex and Juncus, we added extra copies of rbcL sequences for species in these genera when conducting the test on the Carex disticha subcommunity type. Using substitutes in this way does not introduce bias, because rbcL sequence differences among species of Carex and among Juncus species are very small. A total of 104 randomizations were

run for each null model. If specialists are significantly clustered phylogenetically, then the mean and variance of pairwise rbcL distances should fall in the lower 5% of values in the null distribution of each statistic. a niche overlap among congeners vs. other species Through an extensive review of the literature on plant niches, we identified five studies of plant communities from which it was possible to compute a niche overlaps within and between genera. There were nine sets of congeners in total. The validity of generic names was checked against the online versions of Clayton and Williamson (2003) for grasses and Brummitt (1992) for other species. A name change affected one genus (Dentaria to Cardamine), but did not alter the implied evolutionary relationships between this genus and the rest of the community with which it was compared. Niche axes varied between studies (Table 2), but overlap was measured using Pianka’s index in all cases (Pianka 1973). The pairwise overlap, Ojk, between the niche of species j and the niche of species k is P pij pik Ojk ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1Þ P 2 2 pij pik for all resource states, i. In Eq. 1, pij is the proportion of total resources used by j that consist of resource state i, and pik is the proportion of total resources used by k that consist of resource state i. Values of Ojk range from 0 to 1. The difference in mean overlap between congeners, and between congeners and the rest of the community, was tested by a Wilcoxon matched-pairs test (Sokal and Rohlf 1995). b and c niche transitions We conducted a search of articles and citations in American Journal of Botany, Systematic Botany, and TreeBASE (University of Buffalo, New York, USA; available online)5 to identify molecular phylogenetic 5

hwww.treebase.orgi

July 2006

HIERARCHY OF PLANT DIVERSITY

studies of plants in which
50% of the extant species in a clade had been sampled (Tables 3 and 4). Phylogenies with ,20 species were excluded because randomization tests of the kind we used to detect phylogenetic conservatism have low statistical power with sample sizes below this limit (Blomberg et al. 2003). Habitat and elevational data were obtained from the same source as the phylogeny wherever they were given, or from standard floras where they were not (Tables 3 and 4). Habitat is by definition a b niche trait. We treated elevational maximum as a c niche trait, because it delimits the vertical dimension of a species’ range and is clearly related to climate. Evolution of habitat and elevational maximum (EM) were optimized onto trees using MacClade 3.06 (Maddison and Maddison 1992). Habitat was treated as a polymorphic character for species that were present in more than one habitat type. Elevational maximum was scored as a categorical variable with four classes: 0, EM  1000 m; 1, 1000 m , EM  2000 m; 2, 2000 m , EM  3000 m; 3, EM . 3000 m. Tests for phylogenetic conservatism were performed by comparing the number of transitions (steps) between habitat or EM states required to account for the observed distribution of habitats among terminal taxa with a null distribution. We obtained a null distribution for the number of habitat or EM transitions to be expected in any given tree by randomly shuffling the observed states among its terminals (Maddison and Slatkin 1991). Using MacClade 3.06, we performed 103 randomizations for each tree. The probability that an observed number of steps occurred by chance was the frequency of transitions of the same or smaller value found in the null distribution. Frequencies ,0.05 were treated as evidence of significant conservatism in the evolution of habitat preference or EM. The randomization test we used is normally employed on binary characters, but some of our tests involved more than two niche categories (e.g., four EM classes of the c niche). In order to test the robustness of our results against the unconventional use of multistate characters, where variables could be combined on the basis of some ecological variable (e.g., dry vs. mesic), we ran tests on data recoded as a single binary character. RESULTS Are habitat specialists phylogenetically clustered? The mean and variance of pairwise rbcL distances among the seven specialists in the MG5a community were 0.112 (P ¼ 0.144) and 0.0017 (P ¼ 0.226), respectively. For the 10 specialists in the Carex disticha community, the mean and variance were respectively 0.125 (P ¼ 0.428) and 0.0023 (P ¼ 0.362). In neither community was the mean or the variance significantly lower than expected by chance; thus the null hypothesis of no phylogenetic clustering among specialists cannot be rejected.

S43

a niche overlap among congeners vs. other species Table 2 compares a niche overlaps between congeneric pairs of species in nine genera with mean overlaps between the congeners and species in other genera. There is no significant difference overall between the degree of overlap found between congeners and the overlap between noncongeners (Wilcoxon matchedpairs test, Z ¼ 0.42, P ¼ 0.67). b and c niche transitions Table 3 presents b niche transitions in seven island and five continental clades. There was significant conservatism in the evolution of the b niche in five of the seven island clades and in three of the six continental cases. Habitat factors associated with conservatism included the six major altitudinal zones in the Canary Islands (the principal archipelago of Macaronesia) in the case of the Aeonium clade, but not in the Echium or Argyranthemum clades. When a binary coding of b niche into dry vs. mesic was used, Argyranthemum and Sonchus did show conservatism, but Aeonium and Sideritis did not (Table 3). Similar patterns were found in Hawaii, with species in the Schiedea clade showing conservative evolution in respect of eight habitat types; this clade and the silversword alliance showed fewer transitions than expected between wet and dry environments (Table 3). In continental clades, conservatism occurred in both habitat variables (serpentine soils and forest vs. open habitats) analyzed in Calochortus, in one of three variables (occurrence in vernal pools) in Mimulus, and in preference among four habitat types in Narcissus (Table 3). Neither Linanthus nor Primula clades showed evidence of b niche conservatism. Of the two clades analyzed for c niche conservatism, EM evolved conservatively in Pinus, but not in Mimulus (Table 4). Whether the data were coded as four EM classes or two did not affect either outcome. DISCUSSION Collectively, the analyses performed here demonstrate a lack of phylogenetic signal in the ecological structure of communities, but, in contrast, indicate its presence in at least some instances of how speciation populates different habitats and how elevational range evolves. The results support the suggestion that a niche traits are evolutionarily labile, while b and c niche traits might evolve in a more conservative manner. However, there are caveats. The species in the samples used to examine ecological structure in communities on the one hand and adaptive radiation among habitats and elevations on the other were differently constituted. In the first instance, we measured phylogenetic distances between a collection of species that had passed through the various ecological filters involved in community assembly. This resulted in an extremely rarefied sampling of disparate branches of the angiosperm phylogeny, including monocot and eudicot clades. It would be necessary to analyze a less

S44

JONATHAN SILVERTOWN ET AL.

Ecology Special Issue

TABLE 3. Characteristics and numbers of observed b niche (habitat) transitions in island and continental clades estimated by phylogenetic optimization, along with the number expected from a null model.

Clade

Sample/Clade size 

Types of b niche (no. habitats)

Macaronesia

51/63

Argyranthemum

Macaronesia

51 pops/23 spp.

Echium

Macaronesia

21/27

Sideritis Sonchus

Macaronesia Macaronesia

32/32 31/34

Schiedea and Alsinidendron

Hawaii

30/28

Silversword alliance

Hawaii

36/36

cliffs and rocks, xerophytic scrub, thermophile woodland, laurel forest, pine forest, subalpine (6) dry, mesic (2) cliffs and rocks, xerophytic scrub, thermophile woodland, laurel forest, pine forest, subalpine (6) dry, mesic (2) cliffs and rocks, xerophytic scrub, thermophile woodland, laurel forest, pine forest, subalpine (6) dry, mesic (2) dry, mesic (2) dry, coastal, mesic (3) dry and coastal, mesic (2) dry forest, dry shrubland, dry cliffs, dry subalpine, shrubland, diverse mesic forest, wesic forest, wet forest (8) dry, mesic (2) dry, mesic (2)

Western North America

67/67

Narcissus

Europe

23/27

Linanthus, Leptosiphon clade

California

28/28

Mimulus

Northwestern North America

Primula sect. Auricula

Alps

Island clades Aeonium, Greenovia, and Monanthes

Continental clades Calochortus

Region

88/;114 25/25

serpentine, other habitats (2) forest, open habitats (2) dry rocky open Mediterranean hillsides, montane wet meadows, oak woodland, lowland mesic Mediterranean (4) woodland and chaparral, grassy areas, serpentine, desert, drying areas in conifer forest (5) dry, other habitats (2) serpentine, other habitats (2) vernal pool, other habitats (2) serpentine, other habitats (2) limestone, acid substratum (2) cliffs and rocks, turf, woodland, alpine/subalpine/tundra (4)

*P  0.05, **P  0.01, ***P  0.001.   Sample/Clade size is the ratio of the no. species (or, in the case of Argyranthemum, no. populations) in the phylogenetic analysis to the estimated no. extant species that belong to the clade. à The expected number of transitions shown is the mode of the distribution of 1000 runs of the null model. § P values are for the difference between the number of expected transitions and the number of observed transitions (number observed not exceeding number expected). Sources: 1, Hickman (1993); 2, Baldwin and Robichaux (1995); 3, Wagner et al. (1995); 4, Weller et al. (1995); 5, Bohle et al. (1996); 6, Francisco-Ortega et al. (1996); 7, Kim et al. (1996); 8, Barber et al. (2000); 9, Bell and Patterson (2000); 10, Bramwell and Bramwell (2001); 11, Mort et al. (2002); 12, Beardsley et al. (2004); 13, Graham and Barrett (2004); 14, Patterson and Givnish (2004); 15, Zhang et al. (2004); 16, S. C. H. Barrett, personal communication.

rarefied sample if we were asking a solely evolutionary question, but the following question is specific and ecological: ‘‘Are specialist members of MG5a communities phylogenetically clustered?’’ In this case, the method we have used is appropriate and it gives the unequivocal answer, ‘‘no.’’ It is interesting that Kembel and Hubbell (2006) found an absence of phylogenetic structure in the overall tropical forest community in the 50-ha plot at Barro Colorado Island, but that phylogenetic structure did occur within specific habitats. Our null model and those of Kembel and Hubbell (2006) were different, and this

cannot be ruled out as a source of the opposing results (Gotelli and Graves 1996). It may also be that the much larger species pool for tropical forest (n ¼ 312) than for English meadows (n ¼ 52) makes phylogenetic structure more likely to occur or easier to detect among specialists. The approach used to compare niche overlap among congeners with that among other species does not raise phylogenetic sampling issues, but it does assume that a niche dimensions relevant to coexistence have been correctly identified. In each case the dimensions measured do seem likely to fulfil this assumption (Table 2), but as yet very few field studies of putative plant

July 2006

HIERARCHY OF PLANT DIVERSITY

S45

Observed

Expectedà

P§

Data sources

used specific leaf area (SLA) as a proxy measure of the a niche in Ceanothus and found that this diverged earlier than their climatically defined measure of the b niche. This finding is at odds with our hypothesis that the a niche is more labile than the b niche (Fig. 1).

20

25

0.003**

10, 11

b niche transitions

TABLE 3. Extended.

No. b niche transitions

10 21

11 24

0.303 0.071

6 11

10 12

0.001*** 0.485

5 13 8 6 18

6 13 11 9 21

0.272 0.681 0.020* 0.036* 0.019*

3, 4

6 5

9 8

0.020* 0.011*

2

10 13 8

15 17 12

0.011* 0.049* 0.003**

17

18

0.284

6 3 3 5 5 12

8 3 6 6 7 12

0.092 1.000 0.002** 0.156 0.120 0.641

6

5, 10

8 7

14 13, 16

1, 9

1, 12 15

niches have proven their role in coexistence beyond all doubt (Silvertown 2004). The result of the congeners test performed here is consistent with large a niche differences that have been found between sympatric species in, among others, the following genera: Acer (Sipe and Bazzaz 1994, 1995), Adenostoma (Redtfeldt and Davis 1996), Dryobalanops (Itoh et al. 2003), Macaranga (Davies 2001), Piper (Fleming 1985), Psychotria (Valladares et al. 2000), Quercus (CavenderBares et al. 2004), Ranunculus (Harper and Sagar 1953), Salix (Dawson 1990), and Typha (Grace and Wetzel 1981). It is clear that coexisting congeners are often as ecologically different from each other as they are from unrelated members of the same communities. This implies that a niche traits are evolutionarily labile, although proof of this requires evolutionary changes to be analyzed against an explicit, and preferably dated, phylogeny. Ackerly et al. (2006) tested the order in which a and b niche traits evolved in the shrub genus Ceanothus. They

Oceanic islands and island-like habitats, such as vernal pools and serpentine barrens in California, contain multiple radiations that provide replicates for the test of b niche conservatism. There are several examples in the endemic flora of vernal pools in the California Floristic Province (CFP). An extreme case is the monophyletic genus Downingia that contains 13 species (Schultheis 2001), all but one of which occur in vernal pools (Ayers 1993). In section Navarretia of the genus Navarretia, four vernal pool species form a clade that is sister to a species that is facultatively associated with the same habitat (Spencer and Rieseberg 1998). In the much larger genus Mimulus, there are roughly six vernal pool species, and four of them are concentrated in one small clade, indicating significant conservatism in this genus, as well (Thompson 1993, Beardsley et al. 2004) (Table 3). Also in the CFP, significantly conservative evolution of serpentine tolerance is found in the large genus Calochortus, where seven of a total of 18 species occurring on serpentine soils belong to a single clade (Patterson and Givnish 2004) (Table 3). Serpentine species in Mimulus show slight, though non-significant phylogenetic association (Table 3). Phylogenetic relationships among Mimulus species are well resolved, but the weak evidence of phylogenetic conservatism might easily be strengthened by more ecological data. Just three of 28 species in the Leptosiphon clade of the genus Linanthus occur on serpentine, but they represent three independent evolutionary events (Patterson 1993, Bell and Patterson 2000), so there is no evidence of conservatism in this case. Kelch and Baldwin (2003) compared the mean genetic divergence measured at ITS and ETS rDNA loci among terminal taxa in seven clades that have evolved within the CFP, in addition to the cases already mentioned. There was a positive relationship between genetic divergence within a clade and the number of plant communities in which its members are found. A clade of Cirsium species endemic to the CFP was an outlier from the relationship as a whole, inhabiting a greater variety of plant communities than would be expected for the degree of genetic divergence among its members. This deviation could result either from an abnormally high rate of evolutionary shifts between habitats in the CFP Cirsium clade, or an abnormally low rate of molecular evolution. High ecological diversity relative to rDNA variation also occurs in the larger North American Cirsium clade of which the CFP endemics form one part (Kelch and Baldwin 2003). This could indicate that the evolutionary lability of habitat depends on lineage.

S46

JONATHAN SILVERTOWN ET AL.

Ecology Special Issue

TABLE 4. Characteristics and no. c niches, along with no. b niche transitions, based upon upper elevational maximum in two large clades estimated by phylogenetic optimization

Clade

Region

No. b niche transitions

Sample/Clade size

No. c niches

Observed

Expected

P

4 2 4 2

27 19 29 13

27 19 37 19

0.399 0.680 0.003 0.003

Mimulus

Northwestern North America

88/;114

Pinus

Europe, Asia, North and Central America

101/;120

Data sources 2, 3 1, 4

Notes: Data were analyzed for c niches coded into four (0–999 m, 1000–1999 m, 2000–3000 m, .3000 m) and two (,2000 m vs. .2001 m) elevational maximum classes (i.e., c niches). Also see Table 3 footnotes for further explanatory details. Sources: (1) Mirov 1967, (2) Hickman 1993, (3) Beardsley et al. 2004, (4) Gernandt et al. 2005.

Radiations on islands also show a mixed picture, although conservatism here is more evident than might have been expected given the extreme evolutionary lability of plant form that is present in Aeonium (Jorgensen and Olesen 2001), Sonchus (Kim et al. 1996), the silversword alliance (Baldwin and Robichaux 1995), and other island endemics. The Hawaiian mints are another endemic group in which considerable morphological variation among species occurs within a restricted range of climatic conditions (Lindqvist et al. 2003). It should be recognized that the crude distinction between wet and dry habitats used for the silverswords in Table 3 does not do justice to the enormous range of soil moisture conditions present in different habitats in Hawaii. The Hawaiian lobeliods are a group that have radiated across the entire soil moisture gradient (Givnish et al. 2004). Nonetheless, the unexpected presence of conservatism of habitat evolution in several island radiations is remarkable. It suggests that speciation often involves interisland colonization between similar habitats (Francisco-Ortega et al. 1996) and that conservative habitat evolution is not confined to continental radiations. c niche transitions The distinction between b and c niches is not clearcut, but neither should we expect it to be. The three niche types of a, b, and c are segments in Hutchinson’s (1957) n-dimensional hypervolume and are bound to overlap along some dimensions. On some dimensions they may be nested, on others they may not. For example, since elevation and habitat are closely correlated in Macaronesia (Bramwell and Bramwell 2001), conservative evolution of habitat in Aeonium and Sonchus (Table 3) also implies conservative evolution of elevational distribution. We analyzed elevational maximum in Mimulus, where its evolution was not conservative and in Pinus where it was (Table 4). Differences in elevational distribution between the pines of different regions of the world were noted by Mirov (1967). Grotkopp et al. (2004) found that species in the subgenus Pinus occupied significantly lower elevations than those in subgenus Strobus. This implies that elevational distribution has been conserved since the

two subgenera diverged, which dates it to the deepest node in the phylogeny of Pinus (Gernandt et al. 2005). Extant members of the genus comprise a mixture of ancient and quite recently evolved species (Farjon 1996), so conservatism in their elevational distribution cannot simply be attributed to the lack of recent speciation. Why should a, b, and c niches evolve with different degrees of lability? All theories of coexistence based upon nonneutral processes require that species have different a niches in order to coexist (Chesson 2000). Silvertown et al. (2006) argued that, for this reason, community assembly will create structure based upon labile traits. (It will not do so if neutral processes dominate community assembly.) The argument is not that competitive exclusion forces a niches to evolve in a labile manner, but rather that it prevents any traits that might, for whatever reason, not be evolutionarily labile from facilitating coexistence. Nonlabile traits are prevented from defining the a niche by default. Webb et al.’s (2006) study of the effect of interspecific relatedness on seedling mortality implies that apparent competition mediated by disease, as well as direct competition, could cause related species that are insufficiently different to exclude one another at the local scale. A filtering process might also operate upon the traits that define the b niche, but with opposite effect. Coexisting species must by definition occupy the same habitat and must therefore have b niches that overlap. Thus, the b niche might come to be defined by nonlabile traits. The filtering processes that could determine the lability of the a niche and the conservatism of the b niche do not as easily explain the conservatism of c niches, such as the latitudinal ranges of woody plants with disjunct distributions (Qian and Ricklefs 2004). For c niches, we must invoke either phylogenetic constraint, such as a lack of appropriate genetic variation, or phylogenetic niche conservatism (PNC) (Harvey and Pagel 1991). Although the result of stabilizing selection, it is not clear why PNC should operate with particular effect on the c niche; we therefore offer a third

July 2006

HIERARCHY OF PLANT DIVERSITY

explanation. If one thinks of the c niche as being a geographical area with climatically defined boundaries, then the problem of why it evolves conservatively is closely allied to another evolutionary question: what prevents species at range boundaries from evolving the ability to escape beyond those boundaries? Haldane (1956) proposed the following answer to this question: Adaptation at range boundaries, which is necessary for spread to be possible, might be genetically constrained by the swamping effect of gene flow from individuals in the hinterland that are not adapted to condition at the boundary. This process requires that populations at the periphery of a distribution exist as demographic sinks that require an input of migrants for persistence (Kirkpatrick and Barton 1997, Barton 2001). This is a condition that can be tested. In this paper we have developed earlier ideas that the hierarchical organization of plant diversity at the a, b, and c scales proposed by Whittaker (1975) corresponds to a hierarchical set of niches. The traits that define the a niche appear to be evolutionarily labile, whereas the phylogenetic evidence suggests that the b niche evolves in a conservative manner. Perhaps most conservative of all is the c niche, which is related to geographic distribution. The more conservative a trait, the more remote its origin in evolutionary time and the deeper this lies in a phylogenetic tree. Further exploration of the correspondence between the ecological and evolutionary hierarchies should illuminate our knowledge of both. ACKNOWLEDGMENTS J. Silvertown acknowledges the support of a Royal Society travel grant and is grateful to Spencer Barrett and the Botany Department of the University of Toronto for their hospitality during the inception of this paper. We thank Konrad Dolphin, Mike Fay, and Jeffrey Joseph for rbcL sequences. D. Gowing and C. Lawson acknowledge the support of Defra and M. Dodd the support of The Open University. LITERATURE CITED Ackerly, D. D. 2003. Community assembly, niche conservatism, and adaptive evolution in changing environments. International Journal of Plant Sciences 164:S165–S184. Ackerly, D. D. 2004. Adaptation, niche conservatism and convergence: comparative studies of leaf evolution in California chaparral. American Naturalist 163:654–671. Ackerly, D. D., D. W. Schwilk, and C. O. Webb. 2006. Niche evolution and adaptive radiation: testing the order of trait divergence. Ecology 87:S50–S61. Ayers, T. 1993. Downingia. Pages 460–462 in J. C. Hickman, editor. The Jepson manual. Higher plants of California. University of California Press, Berkeley, California, USA. Baldwin, B. G., and R. H. Robichaux. 1995. Historical biogeography and ecology of the Hawiian silversword alliance (Asteraceae). Pages 259–287 in W. L. Wagner and V. A. Funk, editors. Hawaiian biogeography. Smithsonian Institution Press, Washington, D.C., USA. Barber, J. C., J. F. Ortega, A. Santos-Guerra, A. Marrero, and R. K. Jansen. 2000. Evolution of endemic Sideritis (Lamiaceae) in Macaronesia: insights from a chloroplast DNA restriction site analysis. Systematic Botany 25:633–647. Barraclough, T. G., and A. P. Vogler. 2000. Detecting the geographical pattern of speciation from species-level phylogenies. American Naturalist 155:419–434.

S47

Barton, N. 2001 Adaptation at the edge of a species’ range. Pages 3–20 in J. Silvertown and J. Antonovics, editors. Integrating ecology and evolution in a spatial context. Blackwell, Oxford, UK. Beardsley, P. M., S. E. Schoenig, J. B. Whittall, and R. G. Olmstead. 2004. Patterns of evolution in western North American Mimulus (Phrymaceae). American Journal of Botany 91:474–489. Beatty, S. W. 1984. Influence of microtopography and canopy species on spatial patterns of forest understory plants. Ecology 65:1406–1419. Bell, C. D., and R. W. Patterson. 2000. Molecular phylogeny and biogeography of Linanthus (Polemoniaceae). American Journal of Botany 87:1857–1870. Blomberg, S. P., T. Garland, and A. R. Ives. 2003. Testing for phylogenetic signal in comparative data: Behavioral traits are more labile. Evolution 57:717–745. Bohle, U. R., H. H. Hilger, and W. F. Martin. 1996. Island colonization and evolution of the insular woody habit in Echium L (Boraginaceae). Proceedings of the National Academy of Sciences (USA) 93:11740–11745. Bramwell, D., and Z. Bramwell. 2001. Wild flowers of the Canary Islands. Rueda, Madrid, Spain. Brummitt, R. K. 1992. Vascular plant families and genera. Royal Botanic Gardens, Kew, Surrey, UK. hhttp://www. rbgkew.org.uk/data/vascplnt.htmli Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Chazdon, R. L., S. Careaga, C. O. Webb, and O. Vargas. 2003. Community and phylogentic structure of reproductive traits of woody species in wet tropical forests. Ecological Monographs 73:331–348. Chesson, P. 2000. Mechanisms of maintenance of species diversity. Annual Review of Ecology and Systematics 31:343–366. Clayton, W. D., and H. Williamson. 2003. World grass flora. Royal Botanic Gardens, Kew, Surrey, UK. hhttp://www. rbgkew.org.uk/data/grasses-syn.htmli Davies, S. J. 2001. Tree mortality and growth in 11 sympatric Macaranga species in Borneo. Ecology 82:920–932. Dawson, T. E. 1990. Spatial and physiological overlap of three co-occurring alpine willows. Functional Ecology 4:13–25. Ellenberg, H. 1979. Zeigerwerte von Gefa¨sspflanzen Mitteleuropas. Scripta Geobotanica 9:1–122. Ellenberg, H., H. E. Weber, R. Du¨ll, V. Wirth, W. Werner, and D. Paulißen. 1991. Zeigerwerte von Pflantzen in Mitteleuropa. Scripta Geobotanica 18:1–248. Farjon, A. 1996. Biodiversity of Pinus (Pinaceae) in Mexico: Speciation and palaeo-endemism. Botanical Journal of the Linnean Society 121:365–384. Fleming, T. H. 1985. Coexistence of five sympatric Piper (Piperaceae) species in a tropical dry forest. Ecology 66:688– 700. Francisco-Ortega, J., R. K. Jansen, and A. Santos-Guerra. 1996. Chloroplast DNA evidence of colonization, adaptive radiation, and hybridization in the evolution of the Macaronesian flora. Proceedings of the National Academy of Sciences (USA) 93:4085–4090. Gernandt, D. S., G. G. Lopez, S. O. Garcia, and A. Liston. 2005. Phylogeny and classification of Pinus. Taxon 54:29–42. Givnish, T. J., R. A. Montgomery, and G. Goldstein. 2004. Adaptive radiation of photosynthetic physiology in the Hawaiian lobeliads: light regimes, static light responses, and whole-plant compensation points. American Journal of Botany 91:228–246. Gotelli, N. J., and G. R. Graves. 1996. Null models in ecology. Smithsonian Institution Press, Washington, D.C., USA. Gowing, D. J. G., C. S. Lawson, E. G. Youngs, K. R. Barber, M. V. Prosser, H. Wallace, J. S. Rodwell, J. O. Mountford, and G. Spoor. 2002. The water-regime requirements and the response to hydrological change of grassland plant commun-

S48

JONATHAN SILVERTOWN ET AL.

ities. Project BD1310, Final report to DEFRA (Conservation Management Division), London, UK. Grace, J. B., and R. G. Wetzel. 1981. Habitat partitioning and competitive displacement in cattails (Typha): experimental field studies. American Naturalist 118:463–474. Graham, C. H., S. R. Ron, J. C. Santos, C. J. Schneider, and C. Moritz. 2004. Integrating phylogenetics and environmental niche models to explore speciation mechanisms in dendrobatid frogs. Evolution 58:1781–1793. Graham, S. W., and S. C. H. Barrett. 2004. Phylogenetic reconstruction of the evolution of stylar polymorphisms in Narcissus (Amaryllidaceae). American Journal of Botany 91: 1007–1021. Grotkopp, E., M. Rejmanek, M. J. Sanderson, and T. L. Rost. 2004. Evolution of genome size in pines (Pinus) and its lifehistory correlates: supertree analyses. Evolution 58:1705– 1729. Grubb, P. J. 1977. The maintenance of species richness in plant communities: the importance of the regeneration niche. Biological Reviews 52:107–145. Haldane, J. B. S. 1956. The relation between density regulation and natural selection. Proceedings of the Royal Society of London B 145:306–308. Harper, J. L., and G. R. Sagar. 1953. Some aspects of the ecology of buttercups in permanent grassland. Proceedings of the British Weed Control Conference 1:256–265. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Hickman, J. C. 1993. The Jepson manual. Higher plants of California. University of California Press, Berkeley, California, USA. Hill, M. O. 1979. TWINSPAN—a FORTRAN program for arranging multivariate data in an ordered two-way table by classification of the individuals and attributes. Cornell University, Ithaca, New York, USA. Hill, M. O., D. B. Roy, J. O. Mountford, and R. G. H. Bunce. 2000. Extending Ellenberg’s indicator values to a new area: an algorithmic approach. Journal of Applied Ecology 37:3–15. Hutchinson, G. E. 1957. The mutivariate niche. Cold Spring Harbour Symposia in Quantitative Biology 22:415–421. Itoh, A., T. Yamakura, T. Ohkubo, M. Kanzaki, P. A. Palmiotto, J. V. LaFrankie, P. S. Ashton, and H. S. Lee. 2003. Importance of topography and soil texture in the spatial distribution of two sympatric dipterocarp trees in a Bornean rainforest. Ecological Research 18:307–320. Jorgensen, T. H., and J. M. Olesen. 2001. Adaptive radiation of island plants: evidence from Aeonium (Crassulaceae) of the Canary Islands. Perspectives in Plant Ecology Evolution and Systematics 4:29–42. Kelch, D. G., and B. G. Baldwin. 2003. Phylogeny and ecological radiation of New World thistles (Cirsium, Cardueae–Compositae) based on ITS and ETS rDNA. Molecular Ecology 12:141–151. Kembel, S. W., and S. P. Hubbell. 2006. The phylogenetic structure of a Neotropical forest tree community. Ecology 87: S86–S99. Kim, S. C., D. J. Crawford, J. Francisco-Ortega, and A. Santos-Guerra. 1996. A common origin for woody Sonchus and five related genera in the Macaronesian islands— molecular evidence for extensive radiation. Proceedings of the National Academy of Sciences (USA) 93:7743–7748. Kirkpatrick, M., and N. H. Barton. 1997. Evolution of a species’ range. American Naturalist 150:1–23. Lindqvist, C., T. J. Motley, J. J. Jeffrey, and V. A. Albert. 2003. Cladogenesis and reticulation in the Hawaiian endemic mints (Lamiaceae). Cladistics 19:480–495. Maddison, W. P., and D. R. Maddison. 1992. MacClade. Analysis of phylogeny and character evolution. Version 3. Sinauer Associates, Sunderland, Massachusetts, USA.

Ecology Special Issue

Maddison, W. P., and M. Slatkin. 1991. Null models for the number of evolutionary steps in a character on a phylogenetic tree. Evolution 45:1184–1197. Mann, L. K., and H. H. Shugart. 1983. Discriminant analysis of some east Tennessee forest herb niches. Vegetatio 52:77–89. Mirov, N. T. 1967. The genus Pinus. Ronald Press, New York, New York, USA. Mort, M. E., D. E. Soltis, P. S. Soltis, J. Francisco-Ortega, and A. Santos-Guerra. 2002. Phylogenetics and evolution of the Macaronesian clade of Crassulaceae inferred from nuclear and chloroplast sequence data. Systematic Botany 27:271– 288. Nordbakken, J. F. 1996. Plant niches along the water-table gradient on an ombrotrophic mire expanse. Ecography 19: 114–121. Patterson, R. W. 1993. Linanthus. Pages 1037–1046 in J. C. Hickman, editor. The Jepson manual. Higher plants of California. University of California Press, Berkeley, California, USA. Patterson, T. B., and T. J. Givnish. 2004. Geographic cohesion, chromosomal evolution, parallel adaptive radiations, and consequent floral adaptations in Calochortus (Calochortaceae): evidence from a cpDNA phylogeny. New Phytologist 161:253–264. Pennington, R. T., Q. C. B. Cronk, and J. A. Richardson. 2004. Introduction and synthesis: plant phylogeny and the origin of major biomes. Philosophical Transactions of the Royal Society of London B 359:1455–1464. Pennington, R. T., and C. W. Dick. 2004. The role of immigrants in the assembly of the South American rainforest tree flora. Philosophical Transactions of the Royal Society of London B 359:1611–1622. Pianka, E. R. 1973. The structure of lizard communities. Annual Review of Ecology and Systematics 4:53–74. Pickett, S. T. A., and F. A. Bazzaz. 1978. Organization of an assemblage of early successional species on a soil moisture gradient. Ecology 59:1248–1255. Prinzing, A., W. Durka, S. Klotz, and R. Brandl. 2001. The niche of higher plants: evidence for phylogenetic conservatism. Proceedings of the Royal Society of London B 268: 2383–2389. Prinzing, A., W. Durka, S. Klotz, and R. Brandl. 2002. Geographic variability of ecological niches of plant species: Are competition and stress relevant? Ecography 25:721–729. Qian, H., and R. E. Ricklefs. 2004. Geographical distribution and ecological conservatism of disjunct genera of vascular plants in eastern Asia and eastern North America. Journal of Ecology 92:253–265. Redtfeldt, R. A., and S. D. Davis. 1996. Physiological and morphological evidence of niche segregation between two cooccurring species of Adenostoma in California Chaparral. Ecoscience 3:290–296. Rodwell, J. S. 1992. British plant communities. Vol. 3. Grasslands and montane communities. Cambridge University Press, Cambridge, UK. Santos, A. 2001. Flora vascular nativa. Pages 185–192 in J. M. Ferna´ndez-Palacios and J. L. M. Esquivel, editors. Naturaleza de las Islas Canarias. Turquesa, Santa Cruz de Tenerife, Canary Islands. Schaffers, A. P., and K. V. Sykora. 2000. Reliability of Ellenberg indicator values for moisture, nitrogen and soil reaction: a comparison with field measurements. Journal of Vegetation Science 11:225–244. Schultheis, L. M. 2001. Systematics of Downingia (Campanulaceae) based on molecular sequence data: implications for floral and chromosome evolution. Systematic Botany 26:603– 621. Shaukat, S. S. 1994. A multivariate analysis of the niches and guild structure of plant populations in a desert landscape. Pakistan Journal of Botany 26:451–465.

July 2006

HIERARCHY OF PLANT DIVERSITY

Silvertown, J. 2004. Plant coexistence and the niche. Trends in Ecology and Evolution 19:605–611. Silvertown, J., M. Dodd, and D. Gowing. 2001. Phylogeny and the niche structure of meadow plant communities. Journal of Ecology 89:428–435. Silvertown, J., M. E. Dodd, D. Gowing, and O. Mountford. 1999. Hydrologically-defined niches reveal a basis for speciesrichness in plant communities. Nature 400:61–63. Silvertown, J., K. McConway, D. Gowing, M. Dodd, M. F. Fay, J. A. Joseph, and K. Dolphin. 2006. Absence of phylogenetic signal in the niche structure of meadow plant communities. Proceedings of the Royal Society of London B 273:39–44. Sipe, T. W., and F. A. Bazzaz. 1994. Gap partitioning among maples (Acer) in central New England: shoot architecture and photosynthesis. Ecology 75:2318–2332. Sipe, T. W., and F. A. Bazzaz. 1995. Gap partitioning among maples (Acer) in central New England: survival and growth. Ecology 76:1587–1602. Sokal, R. R., and F. J. Rohlf. 1995. Biometry. Fourth edition. W. H. Freeman, San Francisco, California, USA. Spencer, S. C., and L. H. Rieseberg. 1998. Evolution of amphibious vernal pool specialist annuals: putative vernal pool adaptive traits in Navarretia (Polemoniaceae). Pages 76– 85 in C. W. Witham, E. T. Bauder, D. Belk, W. R. Ferren, Jr., and R. Ornduff, editors. Ecology, conservation, and management of vernal pool ecosystems—proceedings from a 1996 Conference. California Native Plant Society, Sacramento, California, USA. Swofford, D. L. 1996. PAUP*: phylogenetic analysis using parsimony (and other methods). Version 4.0. Sinauer, Sunderland, Massachusetts, USA. Sydes, C. L. 1984. A comparative-study of leaf demography in limestone grassland. Journal of Ecology 72:331–345. Thompson, D. M. 1993. Mimulus. Pages 1037–1046 in J. C. Hickman, editor. The Jepson manual. Higher plants of California. University of California Press, Berkeley, California, USA.

S49

Thompson, K., J. G. Hodgson, J. P. Grime, I. H. Rorison, S. R. Band, and R. E. Spencer. 1993. Ellenberg numbers revisited. Phytocoenologia 23:277–289. Tofts, R., and J. Silvertown. 2000. A phylogenetic approach to community assembly from a local species pool. Proceedings of the Royal Society of London B 267:363–370. Valladares, F., S. J. Wright, E. Lasso, K. Kitajima, and R. W. Pearcy. 2000. Plastic phenotypic response to light of 16 congeneric shrubs from a Panamanian rainforest. Ecology 81: 1925–1936. Wagner, W. L., S. G. Weller, and A. Sakai. 1995. Phylogeny and biogeography in Schiedea and Alsinidendron. Pages 221–258 in W. L. Wagner and V. A. Funk, editors. Hawaiian biogeography. Smithsonian Institution Press, Washington, D.C., USA. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Webb, C. O., G. S. Gilbert, and M. J. Donoghue. 2006. Phylodiversity-dependent seedling mortality, size structure, and disease in a Bornean rain forest. Ecology 87:S123–S131. Weller, S. G., W. L. Wagner, and A. K. Sakai. 1995. A phylogenetic analysis of Schiedea and Alsinidendron (Caryophyllaceae, alsinoideae)—implications for the evolution of breeding systems. Systematic Botany 20:315–337. Whittaker, R. H. 1975. Communities and ecosystems. Macmillan, New York, New York, USA. Williams, C. B. 1964. Patterns in the balance of nature and related problems in quantitative ecology. Academic Press, London, UK. Zhang, L. B., H. P. Comes, and J. W. Kadereit. 2004. The temporal course of quaternary diversification in the European high mountain endemic Primula sect. Auricula (Primulaceae). International Journal of Plant Sciences 165:191–207.

Ecology, 87(7) Supplement, 2006, pp. S50–S61 Ó 2006 by the Ecological Society of America

NICHE EVOLUTION AND ADAPTIVE RADIATION: TESTING THE ORDER OF TRAIT DIVERGENCE D. D. ACKERLY,1,4 D. W. SCHWILK,2 2

AND

C. O. WEBB3,5

1 Department of Integrative Biology, University of California, Berkeley, California 94720 USA U.S. Geological Survey/WERC, Sequoia and Kings Canyon Field Station, Three Rivers, California 93271 USA 3 Department of Ecology and Evolution, Yale University, New Haven, Connecticut 06511 USA

Abstract. In the course of an adaptive radiation, the evolution of niche parameters is of particular interest for understanding modes of speciation and the consequences for coexistence of related species within communities. We pose a general question: In the course of an evolutionary radiation, do traits related to within-community niche differences (a niche) evolve before or after differentiation of macrohabitat affinity or climatic tolerances (b niche)? Here we introduce a new test to address this question, based on a modification of the method of independent contrasts. The divergence order test (DOT) is based on the average age of the nodes on a tree, weighted by the absolute magnitude of the contrast at each node for a particular trait. The comparison of these weighted averages reveals whether large divergences for one trait have occurred earlier or later in the course of diversification, relative to a second trait; significance is determined by bootstrapping from maximum-likelihood ancestral state reconstructions. The method is applied to the evolution of Ceanothus, a woody plant group in California, in which co-occurring species exhibit significant differences in a key leaf trait (specific leaf area) associated with contrasting physiological and life history strategies. Cooccurring species differ more for this trait than expected under a null model of community assembly. This a niche difference evolved early in the divergence of two major subclades within Ceanothus, whereas climatic distributions (b niche traits) diversified later within each of the subclades. However, rapid evolution of climate parameters makes inferences of early divergence events highly uncertain, and differentiation of the b niche might have taken place throughout the evolution of the group, without leaving a clear phylogenetic signal. Similar patterns observed in several plant and animal groups suggest that early divergence of a niche traits might be a common feature of niche evolution in many adaptive radiations. Key words: adaptive radiation; Ceanothus; Cerastes; Coast Range; community assembly; Euceanothus; habitat, niche conservatism; phylogenetic comparative methods; specific leaf area; Sierra Nevada; trait divergence; Transverse Range.

INTRODUCTION Ecologists have long considered niche differences among species to be essential for species coexistence (Chesson 2000, Chase and Leibold 2003; but see Hubbell [2001]). The evolution of niche differences among closely related species has received particular attention. Because close relatives tend to be ecologically similar in many respects (Darwin 1859, Felsenstein 1985, Harvey and Pagel 1991, Webb et al. 2002), those features that do diverge during speciation will provide important insights into ecological differentiation and consequences for coexistence of closely related species. It is useful in this context to distinguish two scales of niche differentiation, corresponding to different scales Manuscript received 21 January 2005; revised 9 August 2005; accepted 11 August 2005. Corresponding Editor: A. A. Agrawal. For reprints of this Special Issue, see footnote 1, p. S1. 4 E-mail: [email protected] 5 Present affiliation: Arnold Arboretum of Harvard University.

of species distributions. At large spatial scales, species can occupy different macrohabitats or climatic envelopes; the resulting distributions will be largely allopatric, or, if they do overlap geographically, individuals of the two species would rarely encounter one another due to habitat differentiation. At smaller scales, related species that co-occur in local communities usually exhibit spatial or temporal differentiation in microhabitat, resource use, diet, or other factors. It is at this local scale, where the balance of intra- and interspecific interactions influences coexistence and community structure, that the niche concept has played the most important role. Following Pickett and Bazzaz (1978) and Silvertown et al. (2006), we employ the term a niche to describe these small-scale components of the niche that differ among co-occurring species, corresponding to Whittaker’s (1975) use of a diversity for diversity of local communities. In contrast, the b niche is defined as macrohabitat and climate factors related to larger scale distributions, corresponding to the b component of diversity among habitats in a landscape. In this paper, we do not distinguish the proposed b and c niche, which

S50

July 2006

NICHE EVOLUTION AND ADAPTIVE RADIATION

refers to distributions at the scale of habitat vs. geographic range (Silvertown et al. 2006), as these have equivalent implications in terms of species interactions at the community level. It has been argued that species within local communities tend to be phylogenetically overdispersed; i.e., closely related species co-occur less than expected, relative to an appropriate null model (Elton 1946, Williams 1964, Gotelli and Graves 1996, CavenderBares et al. 2004). This pattern would suggest that a niche parameters are evolutionarily conserved and/or b niche parameters are highly divergent, such that close relatives tend to occupy different macrohabitats and hence different communities (Elton 1946, Williams 1964, Gotelli and Graves 1996, Cavender-Bares et al. 2004). The alternative, if a niche traits were more labile, would facilitate coexistence and divergent resource use in sibling species within local communities. Divergence of macrohabitat parameters among sibling taxa is compatible with the allopatric model of speciation, as disjunct populations in a heterogeneous landscape are likely to encounter distinct habitats (Graham et al. 2004; but see Wiens [2004]). Differentiation of the a niche would then represent a later stage of evolutionary divergence, either resulting from or directly promoting species coexistence as the ranges of the now distinct species expand into each other’s territory. In an important paper, Diamond (1986) argued for this ‘‘habitat-first’’ model of speciation, based on his observations that closely related bird species in New Guinea tend to be allopatric and occupy distinct macrohabitats along elevational or climatic gradients (see review by Schluter [2000]). In addition, Diamond and Schluter both argued that the habitat-first speciation model can be extended to the analysis of adaptive radiations, based on a parsimonious assumption that rates of evolution do not dramatically change. In other words, if habitat divergence represents the first stage of speciation among close relatives, then it would also be characteristic of early speciation events at the base of an adaptive radiation. As a corollary, if differentiation of a niche occurs late in speciation, or is observed among distant relatives, then it would be characteristic of the later stage of adaptive radiation. Streelman and Danley (2003) presented a related model, arguing that vertebrate adaptive radiations follow a trajectory of divergence along three axes: habitat, trophic morphology, and communication, usually in that order. However, their use of ‘‘habitat’’ refers more to microhabitats within a community (e.g., benthic vs. limnetic sticklebacks), rather than Diamond’s larger scale differentiation among elevational bands and different forest types occupied by New Guinea birds. This ambiguity over the use of the word habitat is unfortunate, as there is a substantive difference between these models in their emphasis on large-scale habitat differences, implying allopatric populations, vs. microhabitat differentiation within local communities.

S51

Recent developments of phylogenetic methodology offer outstanding opportunities to reevaluate these classic questions. Here, we present a comparative approach to the problem of niche evolution by introducing a new comparative method designed to test the relative timing of divergence in two ecological traits (e.g., a vs. b niche axes); we then use the test to evaluate the sequence of trait divergence in the radiation of the woody plant group Ceanothus in California. Our results indicate that a niche traits related to local scale coexistence diverge first in the radiation of this group, a pattern that is shared with several other plant and animal radiations. DIVERGENCE ORDER TEST The divergence order test (DOT) was designed to address these questions of niche evolution, as well as other questions regarding the relative sequence of diversification events in a clade. The test examines the relative timing of evolutionary divergence for two continuous characters, and it is based on a modification of the method of phylogenetic independent contrasts (Felsenstein 1985). Independent contrasts transform the trait data for N species into a set of N – 1 contrasts, each based on the difference between trait values across a phylogenetic divergence. Under the assumption that the trait evolved independently at each divergence, the contrasts provide a robust basis on which to test hypotheses of correlated evolution, addressing the underlying historical pattern of trait evolution as well as better meeting the assumptions of standard parametric statistics (Garland et al. 1992). For the DOT, we modify this method and use the absolute differences between related nodes derived from maximum-likelihood estimates of ancestral trait reconstructions, obtained using ANCML (Schluter et al. 1997). This approach allows us to incorporate the uncertainty of reconstructions in deeper nodes. The divergence order test is based on two sets of numbers: (1) the absolute value of the unstandardized contrasts for trait i across the nodes (k ¼ 1, 2, . . . , N) of a phylogeny (Cik), which measures the magnitude of the divergence that occurred at each node regardless of the direction of change; and (2) the age of each node (Ak). We then calculate a weighted mean age of divergence for each trait as follows: N X

Ak Cik

Wi ¼ k ¼N1 X

:

ð1Þ

Cik

k¼1

The result is an average age, in units of time, that indicates whether the large divergences in a trait tended to occur early or late in the diversification of a group (Fig. 1). Note that this age will not generally correspond to the age of any single divergence; it is simply a statistical measure of the tendency toward early or late

S52

D. D. ACKERLY ET AL.

FIG. 1. Example of the divergence order test (DOT). Two patterns of trait divergence are illustrated on a simple phylogeny. Numbers at the tips of the phylogeny indicate two possible trait states, 0 or 1. Numbers at interior nodes (in italics) show the contrast for that node. Trait A (open circles) exhibits a pattern of late divergence, whereas Trait B (shaded circles) exhibits a pattern of early divergence. The lower panel plots contrast magnitude vs. age and shows the calculated weighted divergence age for Trait A (WA ¼ 1 Ma [i.e., 1 million years ago]) and Trait B (WB ¼ 2 Ma).

divergence for the trait in question. The DOT is then based on the comparison of the weighted divergence age for two traits (D ¼ Wi
Wj) to determine whether one trait diverged significantly earlier than the other, on average. The DOT statistic is derived from contrasts between ancestral states and does depend on the accuracy of the ancestral estimates themselves (Oakley and Cunningham 2000). We do not consider the traits of outgroups in calculating ancestral states, as these are only necessary to identify trends in trait evolution within a group, and not the magnitude of divergences. The maximumlikelihood algorithm of ANCML assumes that the pattern of trait evolution fits a model of Brownian motion. Global squared-change parsimony, which is also based on Brownian motion, provides the same ancestral estimates, but no confidence limits. The fit to Brownian motion can be tested in several ways. First, using Felsenstein’s (1985) algorithm, the correlation of

Ecology Special Issue

the absolute value of standardized independent contrasts and the standard deviation of those contrasts (the square root of subtending branch lengths) should be nonsignificant (Garland et al. 1992). A negative correlation would indicate larger contrasts than expected on rapid bifurcations, and vice versa for a positive correlation. The absolute values of standardized contrasts should also fit a half-normal distribution, and this can be checked visually using truncated normal probability plots. In addition, if several distinct clades are present (as in the Ceanothus case), homogeneity of evolutionary rates can be tested using a nonparametric comparison of standardized contrasts between groups (Garland 1992), or a recently introduced maximum-likelihood approach (O’Meara et al. 2005). In general, methods based on independent contrasts are fairly robust to violations of Brownian motion (Diaz-Uriarte and Garland 1996, Ackerly 2000), but this has not been evaluated for the calculation of standard errors by ANCML or the DOT analysis. It can be useful to examine correlations between the two traits under consideration, though DOT does not require that the traits exhibit any particular pattern of correlated or independent evolutionary change. If changes in the two traits are tightly linked, then DOT will certainly not be significant, as the contrasts will be similar in magnitude at each node. However, differences in the magnitude of a few basal or distal nodes could result in a significant DOT outcome, and trait evolution could still be correlated overall on the tree. We have explored several approaches to significance testing of the D statistic (see Appendix A). We present here our preferred method, based on a bootstrapping approach, to obtain confidence intervals for the two estimates of W and their difference, D. The rationale for this approach is that comparative methods, particularly independent contrasts, tend to underestimate the magnitude of older divergences for rapidly evolving traits. As a simple example, consider a bifurcating tree with four species at the tips. If each pair of sister taxa has divergent trait values, reflecting rapid trait evolution, then the averages for their respective common ancestors could be virtually identical and the basal contrast will be nearly or exactly zero (e.g., Fig. 1, Trait A). However, given the rapid rate of evolution for this trait, it is also possible that a large divergence occurred at the first node, followed by reversals at the subsequent nodes resulting in convergence among extant taxa. Maximumlikelihood estimates of ancestral states allow for this possibility by placing confidence limits on the ancestors (Schluter et al. 1997). If a trait evolves rapidly, then the confidence limits at deeper nodes will be large (see Appendix A, Fig. A1). We use ANCML (Schluter et al. 1997) to generate maximum-likelihood estimates and confidence limits of ancestral states at each node; we then create bootstrap distributions of the potential magnitude of each divergence event (see Appendix A, Fig. A1). Hypo-

July 2006

NICHE EVOLUTION AND ADAPTIVE RADIATION

thetical ancestral values are sampled from the distribution for each trait at each node, and from each sample we calculate the corresponding values of Cik, Cjk, Wi, Wj, and D. We then examine the distribution of D values to determine whether D ¼ 0 falls outside the 95% confidence limits on the mean, indicating significance of the observed values at P  0.05. The calculation and significance of the DOT method were implemented on Mac OS X, using awk scripts, R (R-project 2004), and ANCML (see Supplement). Branch lengths and node ages As the objective of this test is to calculate relative timing of divergence events, the analysis should ideally be conducted with ultrametric, calibrated phylogenies based on a molecular clock, or rate-smoothed branches if the branch lengths violate a molecular clock (Sanderson 2002). Methods for obtaining relative ages are improving, although there can still be considerable uncertainty due to heterogeneous rates of molecular evolution and difficulty in establishing calibration points across different clades (Sanderson 2002, 2003). The DOT method will be robust to much of this uncertainty, because the same ages are used in the calculation of weighted divergence times for both traits. For the nodes along any contiguous path from the root to the tips, ages will always be correctly ordered, even if the actual values are uncertain. Problems would most likely arise if incorrect calibrations were applied to two or more independent segments of the tree (i.e., along different root-to-tip paths) in which different traits exhibited large evolutionary divergences. Attention to this problem is warranted in applications of the test.

S53

Despite these consistent differences in drought tolerance and fire response, both clades are widespread in the California Floristic Province, inhabiting chaparral, semiarid forests, and oak woodland. Often, species pairs representing one species from each of the subgroups cooccur, and it has been suggested that differences in fire response and/or tolerance of water stress facilitate this coexistence (Keeley 1977, Keeley and Zedler 1978, Davis et al. 1999). These patterns suggest that the basal divergence between the two clades involved a niche traits related to drought tolerance or postfire regeneration strategies. In contrast, both clades are represented by species throughout California, suggesting more recent differentiation of the climatic niche envelope, representing the b niche (Knight and Ackerly 2001). Here we undertake four new analyses to quantify and test these observations. (1) We have assembled a co-occurrence data set from the literature, and we use null models to test for nonrandom patterns of co-occurrence between species from the two subgroups. (2) We combine the cooccurrence data with a trait data set to test whether cooccurring species differ significantly for traits related to plant growth strategies (a niche) and are more similar than expected for traits related to climatic envelopes (b niche). (3) We reanalyze the Ceanothus phylogeny, based on internal transcribed spacer (ITS) sequence data, to obtain an ultrametric tree with branch lengths fit to a molecular clock. (4) We use this phylogeny and the trait data set to apply the DOT analysis, testing the prediction that a niche traits diverged earlier than b niche traits in the evolution of Ceanothus. METHODS Occurrence data

CASE STUDY: COMMUNITY ASSEMBLY AND NICHE EVOLUTION IN CEANOTHUS The woody plant group Ceanothus comprises ;55 minimum-rank taxa (species or subspecies) that have primarily diversified within the California Floristic Province (McMinn 1942, Hardig et al. 2000). The group is divided into two well-supported clades (Jeong et al. 1997, Hardig et al. 2000) that differ consistently in several morphological and physiological traits related to drought tolerance. The two groups are considered subgenera, and are currently designated Cerastes and Ceanothus; for greater clarity (and consistency with phylogenetic naming conventions), we prefer the older name Euceanothus for the latter group. Species in Cerastes have thick leaves with stomatal crypts and have shallower roots and more embolism-resistant xylem than do members of Euceanothus (McMinn 1942, Hellmers et al. 1955, Davis et al. 1999). Additionally, species of Cerastes establish only from seed following fire, while Euceanothus species generally resprout as well (Wells 1969, Schwilk and Ackerly 2005). In a Mediterraneantype climate, seedlings of nonsprouting species must survive an intense summer drought period after winter or spring germination.

A matrix of co-occurrence data for Ceanothus in California was obtained from a search of the literature and consultation with colleagues. A total of 51 sites were obtained that had two or more co-occurring Ceanothus, with a total of 16 different taxa (plots in the same location with the same species composition were recorded as one site; Nicholson 1993). Of these sites, 35, with 13 taxa, were located in chaparral of the Coast or Transverse ranges, while 16 sites with 7 taxa were from the Sierra Nevada region (CT and SN, respectively). Of the 51 sites, 48 had just two Ceanothus species, while the remainder had three. (See Appendix B for details of the occurrence data matrix.) Trait selection Although coexistence of Ceanothus species, in a matrix of other taxa in the community, can involve contrasting physiological or regeneration strategies, no direct studies of coexistence mechanisms in chaparral have been conducted. To reflect differences in drought tolerance, the ideal traits would be either a direct measure of xylem tolerance to embolism under water deficit or wood density, which is a close correlate of xylem tolerance (Hacke et al. 2001). These traits are known to vary

S54

D. D. ACKERLY ET AL.

between the two subgroups (Davis et al. 1999), but they are not available for large numbers of species. As a proxy for contrasting physiological strategies, we used specific leaf area (SLA), the ratio of fresh leaf area to dry mass. This well-studied leaf functional trait plays an important role in the ‘‘leaf economic spectrum’’ of variation in plant metabolic rates (Wright et al. 2004). In general, species with higher SLA have shorter leaf life span and higher photosynthetic rates. In chaparral shrubs such as Ceanothus, high SLA is associated with less droughttolerant leaves (Ackerly 2004b), and we have a large data set available for roughly two-thirds of the Ceanothus taxa (Ackerly 2004a). We do not claim that differences in SLA per se promote coexistence, but rather that they could be associated with suites of traits that are related to niche partitioning and differentiation in ecological strategies of co-occurring species. Data for SLA are species means collected previously from field, herbarium, and botanic garden specimens (Ackerly 2004a); three new taxa that appeared in the occurrence data set were added to this trait data set (C. greggii, C. parvifolius, and C. sanguineus) based on measurements of specimens in the University and Jepson Herbaria (University of California–Berkeley, California, USA). Values were log-transformed for all analyses, as relative differences are a better measure of physiological differentiation (Reich et al. 1997). (See Appendix C for details of the trait data set.) At larger spatial scales (b niche), Ceanothus are differentially distributed with respect to edaphic conditions (e.g., serpentine specialists), habitat (chaparral vs. conifer forests), elevational range, and macroclimate (precipitation and temperature) (Hickman 1993, Nicholson 1993, Davis et al. 1999). We quantified the realized climatic niche, to characterize these large-scale distributions, by overlaying species distributions on climate maps of California and calculating mean climatic parameters for each species (Knight and Ackerly 2001). We have selected the mean precipitation and the mean January temperatures within the geographic range of each species, reflecting distributions along geographic and elevational gradients in California. The climate niche analysis serves as an indirect surrogate for unmeasured physiological traits related to species tolerances and distributions along climate gradients. This interpretation assumes that distributions do not simply reflect historical factors and limited dispersal potential. The contraction and expansion of chaparral during the interglacial periods (Graham 1999) argues against a strong role for dispersal limitation as a longterm constraint on species distributions. Community assembly We used five null models for community assembly to test several predictions, relative to patterns that would be expected by chance: (1) Species from the two clades (Cerastes and Euceanothus) co-occur more often than expected. (2) Values of SLA show greater variation among co-occurring species than expected. (3) Climate

Ecology Special Issue

niche parameters show greater similarity among cooccurring species than expected. The measure of trait dissimilarity within communities was simply the difference between species values for two-species samples, and the mean of the successive differences among ranked values in three-species samples. We also calculated the mean trait value for each community, which allowed us to test our additional null models. (4) The standard deviation of site means should be higher for climate niche parameters than expected by chance, due to turnover of species along climatic gradients. Finally, (5) among-site standard deviation in mean SLA should be lower than expected under the null, as the combinations of species from the two clades result in high trait disparity within sites and low disparity across sites. One-tailed tests were conducted for all hypotheses, based on these predictions, comparing the observed data to 999 randomizations. As a null model, we use the ‘‘independent-swap’’ algorithm (Gotelli and Entsminger 2001), preserving both site diversity and species frequency of occurrence while randomizing assignments of species to sites. This is critical to ensure that patterns of trait assembly do not simply reflect differential abundance of particular species. All calculations were carried out in R (R-project 2004); the swap algorithm was implemented by S. Kembel in C as part of the PHYLOCOM package (Webb et al. 2004). Phylogeny The Ceanothus phylogeny of Hardig et al. (2000) was reanalyzed to obtain an ultrametric tree fit to a molecular clock for the taxon sample in our trait data set. Conflicting phylogenies for Ceanothus based on ITS vs. matK sequence data could reflect lineage sorting during rapid radiations or hybridization (Hardig et al. 2000), and ITS (which is a nuclear marker) was selected as a more reliable estimator of the ‘‘true’’ species tree. Limited sampling of ndhF (10 taxa; Jeong et al. 1997) is insufficient to incorporate in the broader analysis considered here. ITS sequence data for 76 accessions (73 Ceanothus and 3 outgroups [Adolphia californica, Zizyphus obtusifolia, and Spyridium parvifolium]) were obtained from GenBank (accessions GBAN-AF048901 through GBAN-AF048975; Hardig et al. 2000). Sequences were aligned with ClustalX using default parameters, and alignments were checked by eye (no manual adjustments were made); total aligned sequence length for ITS1, ITS2, and the intervening 15S region was 627 nucleotides. Taxa with identical sequences were kept in the analysis, for use later in comparative analyses. Multiple sequences for individual taxa were pruned to one representative sequence, based on preliminary analyses (Hardig et al. 2000). The resulting analysis included 56 Ceanothus sequences and 96 informative characters. Phylogenetic analysis was conducted with PAUP*, using parsimony criteria and a heuristic search (random addition sequence with 10 replicates, TBR, MULTREES in effect, collapse zero-length branches in

July 2006

NICHE EVOLUTION AND ADAPTIVE RADIATION

effect, and steepest descent not in effect) (Swofford 2002). This analysis resulted in 174 equally parsimonious trees of length 322 on one island, which was hit in all 10 replicate searches. The strict consensus of the equally parsimonious trees was very similar to the results reported by Hardig et al. (2000). The most significant difference was that we obtained more resolution within Euceanothus, with the Western group (Hardig et al. 2000: Fig. 1) monophyletic and nested within a paraphyletic Eastern group. For comparative analysis, ANCML (Schluter et al. 1997) requires fully bifurcating phylogenies. It is possible to generate these by randomly resolving the trees obtained in the analysis just described (in which zero-length branches were collapsed), but we are not aware of software that will provide alternative resolutions while maintaining branch lengths. As an alternative, we conducted a second parsimony search in PAUP* with the same parameters, but with zerolength branches not collapsed and MAXTREES ¼ 10 000. Ceanothus oliganthus ssp. oliganthus, which was present in our ecological data but missing from the molecular data, was added to the matrix with the same sequence as C. oliganthus ssp. sorediatus. Given taxa with identical sequences, this search quickly reached the maximum number of trees, with length ¼ 322, as in the prior case. We then pruned the trees to include only the 39 taxa for which we had phenotypic data, resulting in 3254 unique topologies (outgroups were not considered in the analysis of ancestral states). For further analysis, 100 trees were selected from this set by sampling every 20th tree (because immediately adjacent trees tend to be similar to each other). This set of 100 alternative, fully bifurcating trees was used for all subsequent analyses. Branch lengths Maximum-likelihood methods were used to fit branch lengths to the 100 topologies, based on the HKY85 model with transition/transversion rates fit empirically from the data (Swofford 2002). A molecular clock was not rejected using this model (P . 0.05 for all trees), so the branch lengths fit with a molecular clock were used for comparative analyses. Based on an independent analysis of rates of rbcL evolution in Ceanothus (Jeong et al. 1997), the split between Cerastes and Euceanothus was calibrated at 18–39 Ma. Fossil evidence provides independent confirmation of the minimum age, as taxa assignable to both clades appear in the fossil record by 18 Ma (Chaney 1927, Axelrod 1956). Using this calibration for the basal node, the clock-calibrated tree suggests that radiation of each of the two (Cerastes and Euceanothus) began no more recently than 4–5 Ma, at about the same time as the onset of the Mediterraneantype climate in California. The inclusion of taxa with identical sequences (a common occurrence for rapidly speciating groups) results in zero-length branches in the phylogeny (i.e., a

S55

polytomy). Zero-length branches create problems for both ANCML and independent contrasts, as they imply instantaneous evolutionary divergence (a hard polytomy), whereas they might reflect uncertainty of the sequence of speciation events (soft polytomy) as much as rapid radiation. Zero-length branches represent truncated estimates of elapsed time since speciation, since each branch will remain at zero length until the first fixed base change occurs. We addressed this problem by adjusting zero-length branches to a small nonzero value (1.0 3 10
4) slightly lower than the shortest branches resulting from the maximum-likelihood fit to the molecular clock, which ranged from 1.03 3 10
4 to 4.94 3 10
4 across the 100 trees. Based on our calibration, this adjustment represents an absolute time of ;5 3 104 yr. Note that these small divergences (especially if they occur on both sides of a bifurcation) will lead to an inflated estimate of the Brownian motion rate parameter and, hence, broader confidence intervals on ancestral states. This will increase the standard errors of weighted divergence age (W ) from our bootstrap procedure, leading to a more conservative test of significance for the DOT statistic. We recognize that this is an imperfect solution and hope that the problem of zero-length branches will be addressed in future research on branch length calibration for comparative analysis. Divergence order test analysis We conducted the DOT analysis based on the average age of internal nodes, weighted by the absolute value of the unstandardized independent contrasts for each trait. Significance of the difference in ages was assessed by bootstrapping trait histories from the mean and standard deviations obtained from ANCML. Given the strong preliminary data that we have introduced, we conducted one-tailed tests of the hypothesis that SLA divergence occurred earlier than divergences in climate niche parameters. RESULTS Trait variation Across the entire trait data set (39 taxa), specific leaf area (SLA) was significantly higher in Euceanothus than Cerastes. Precipitation and SLA were weakly correlated overall (R ¼ 0.25), and they were essentially independent based on independent contrasts (R ¼ 0.04) (Fig. 2A). January temperature and SLA were negatively correlated across species (R ¼
0.35) and based on independent contrasts (R ¼
0.31). This negative relationship was also apparent within Euceanothus, but not in Cerastes (Fig. 2B). The decline in Euceanothus reflects the transition from deciduous taxa occupying the coldest ranges (e.g., C. parvifolius) to evergreens at lower latitudes or elevations. The most striking aspect of both relationships is the marked difference in SLA between the clades that is maintained across the climatic gradients, consistent with the prediction of local cooccurrence between species that differ in SLA.

S56

D. D. ACKERLY ET AL.

Ecology Special Issue

FIG. 2. Specific leaf area (SLA; originally measured in mm2/mg) vs. (A) mean precipitation, and (B) January temperatures within species ranges for species of Cerastes (solid circles) and Euceanothus (open circles).

Community assembly Taxonomic co-occurrence.—Of the 48 sites with two species, 38 had one from each of the two major clades (Cerastes and Euceanothus); this pattern was particularly striking in the Coast/Transverse (CT) chaparral sites (31 of 35 sites), but was not significant in the Sierra Nevada (SN) (Table 1). For both the full data set and the coastal partition, the elevated co-occurrence of species from different clades was highly significant, relative to the null model that maintained both site diversity and species occurrence (P  0.001). The frequent co-occurrence of taxa from the two clades has long been noted in the literature, but not previously quantified and tested relative to a null model. Trait disparity.—Across all sites, the mean difference in log10(SLA) between co-occurring species was 0.32 units, and as predicted this value was significantly greater than the null expectation (P , 0.04; Table 2). In contrast, differences between climate niche parameters of co-occurring species were much smaller than expected (P  0.001 in both cases; Table 2). When the data are partitioned geographically, the within-site disparity in SLA was still significantly greater in CT, but disparity in SN was less than expected. Disparity in climate niche parameters is significantly lower than the null, as predicted, in both areas (Table 2). The among-site standard deviation in trait means was significantly greater than expected for climate niche traits, as predicted, in the entire data set and both geographic partitions. For SLA, it was significantly lower than expected in CT, but the pattern was reversed in SN (Table 2). Collectively, these analyses indicate that in Coast and Transverse range chaparral, co-occurring Ceanothus species exhibit greater disparity in SLA than expected under a null model of community assembly. In contrast, in the Sierra Nevada it appears that SLA varies among sites, while within-site disparity is low,

perhaps due to distribution of evergreen vs. deciduous (low- vs. high-SLA) species along elevational gradients. As expected, climate niche parameters are always similar among co-occurring species and significantly different among sites. Divergence order test analysis These results support the selection of SLA and realized climate niche parameters as traits reflecting local (a) vs. regional (b) differentiation, respectively. The patterns are stronger in the Coast and Transverse ranges, and if there were a monophyletic group in Ceanothus restricted to these communities we would limit our analyses to this group. However, the evolutionary radiation into the two major clades, and subsequently within clades, encompasses both geographic areas (and beyond). We feel it is important to TABLE 1. Relative frequency of local co-occurrence patterns for Ceanothus, in terms of the number of species from each major subgroup (CT ¼ Coast and Transverse ranges; SN ¼ Sierra Nevada). No. communities Community pattern Cerastes

Euceanothus

Total (P  0.001)

0 1 2 0 1 2 3

2 1 0 3 2 1 0

7 38 3 2 1 0 0

CT (P  0.001)

SN (NS)

2 31 2 0 0 0 0

5 7 1 2 1 0 0

Notes: Example of how to read Table 1: The first row indicates that a total of seven communities were recorded (two in CT and five in SN) with two co-occurring Euceanothus species and no Cerastes. P values indicate the probability of obtaining the observed number of sites occupied by species from the two subgroups, relative to a null model of community assembly (see text).

July 2006

NICHE EVOLUTION AND ADAPTIVE RADIATION

S57

TABLE 2. Mean trait disparity among co-occurring species, and standard deviation of trait means among sites. Mean disparity Trait

SD

of site means

Prediction All (N ¼ 51) CT (N ¼ 35) SN (N ¼ 16) Prediction All (N ¼ 51) CT (N ¼ 35) SN (N ¼ 16)

Specific leaf area (mm2/mg, log) Observed mean Expected mean (SD) P

.

Annual precipitation (mm) Observed mean Expected mean (SD) P January minimum temperature (8C) Observed mean Expected mean (SD) P

0.32 0.29 (0.018) ,0.04

0.35 0.28 (0.021) 0.001

0.25 0.28 (0.016)

,

117 280 (25.2) 0.001

123 187 (17.3) 0.002

103 121 (8.7) ,0.03

,

1.25 2.53 (0.19) 0.001

0.92 1.24 (0.093) 0.002

1.95 2.49 (0.21) , 0.02

0.170 0.14 (0.014)

NS

0.093 0.15 (0.013) 0.001

.

217 155 (12.1) 0.001

159 124 (10.5) 0.002

83 67 (8.22) ,0.02

.

2.001 1.50 (0.11) 0.001

0.985 0.81 (0.065) , 0.01

1.571 1.23 (0.14) ,0.02

,

NS

0.158 0.17 (0.012)

NS

Notes: Observed values are compared to expectations based on a null model of community assembly. Analyses were conducted for all sites, and for Coast and Transverse ranges (CT) and Sierra Nevada (SN) sites separately. Prior predictions regarding the direction of the difference between observed and expected values are listed, and all significance tests are one-tailed.

conduct evolutionary analyses on all available data for the entire group. The three traits considered here generally fit the Brownian motion model underlying the use of ANCML for ancestral states. For all three traits, inspection of normal probability plots for the absolute value of standardized contrasts indicated a close fit to a truncated normal distribution. For SLA and January temperatures, correlations of standardized contrasts and their standard deviation (i.e., the square root of the subtending branch lengths) were not significant. For precipitation, the correlation was significantly negative across all trees, reflecting larger than expected divergences across rapid speciation events, and a small divergence across the long branches between Cerastes and Euceanothus. This pattern will lead to a relatively high Brownian motion rate parameter, to accommodate these rapid divergences, and will thus inflate the confidence intervals around the ancestral estimates for precipitation, making the DOT more conservative. Across most of the 100 trees, rates of trait evolution were homogeneous between Cerastes and Euceanothus; in 3 and 24 trees, rates of SLA and January temperature evolution, respectively, were significantly higher in Euceanothus (0.04 , P , 0.05). Fig. 3 illustrates trait divergences on a randomly selected tree from the analysis. The largest contrast for SLA was located at the basal node between Cerastes and Euceanothus (Fig. 3). The weighted divergence age for SLA, averaged across the 100 alternative phylogenies, was 6.5 3 10
3 branch length units (Table 3). For the climate parameters, the basal contrasts were much smaller, and larger divergences were noted within each of the two major clades (Fig. 3). The weighted divergence ages were 4.30 3 10
3 and 4.24 3 10
3 for January temperatures and precipitation, respectively.

For all 100 alternative phylogenies, weighted divergence ages were older for SLA than for both climate parameters, and the DOT was significant in 94 of 100 trees for the SLA vs. January temperature difference, and for 92 out of 100 trees for SLA vs. precipitation. DISCUSSION This analysis of the radiation of Ceanothus demonstrates an initial shift in a niche traits that subsequently promote (or at least facilitate) coexistence among related species. These traits were then conserved as the radiation progressed, and later speciation events were characterized primarily by divergence in climate envelopes, corresponding to geographic differentiation along latitudinal and elevational gradients. The result is that cooccurring Ceanothus species within local communities are more distantly related than expected by chance, relative to the group as a whole. Similar patterns are evident in other clades that have been analyzed in a phylogenetic context. In the oaks (Quercus) of Florida, local communities tend to be phylogenetically overdispersed, often with one or two members each drawn from three distinct clades (red, white, and live oaks). Habitat preferences diverge repeatedly within each of these clades; deeper divergences between the clades involve traits, such as seed maturation time and wood density, which may promote coexistence through differential regeneration or pathogen tolerance, respectively (Cavender-Bares et al. 2004). In studies of Phylloscopus warblers in Kashmir, Richman and Price (1992, Richman 1996) argued that deep divergences in the group involved differentiation in feeding strategies, while more recent speciation events were related to macrohabitat distributions (coniferous vs. deciduous forests) (but see Forstmeier et al. [2001]). In the radiation of Anolis lizards, distinctive ecomorphs, which coexist by feeding in different parts of the canopy,

S58

D. D. ACKERLY ET AL.

FIG. 3. Divergence order test (DOT) for Ceanothus, illustrated for one randomly chosen tree out of 100 equally parsimonious trees used for analysis. For all panels, ages along the x-axis are clock-calibrated molecular branch length units; the breaks in the axis indicate the long branches connecting the two recently diversifying clades. (A) Ceanothus phylogeny, based on reanalysis of internal transcribed spacer (ITS) sequence data from Hardig et al. (2000). Species names are omitted for clarity. (B–D) Open circles indicate the magnitude of unstandardized contrasts vs. node age for specific leaf area (SLA), January temperatures, and mean precipitation, respectively (see Table 3 for weighted mean ages). Standard errors of the contrasts (derived from the bootstrap procedure) are illustrated for the basal contrast only. The solid diamonds indicate weighted mean divergence age for each trait (6SE) based on 200 bootstrap randomizations of the contrasts (vertical position of this point is arbitrary).

have evolved independently on multiple islands (Losos et al. 1998, Knouft et al. 2006). On the larger islands, however, there has been continued speciation involving diversification of macrohabitats within ecomorph

Ecology Special Issue

groups (Glor et al. 2003). In our terminology, the earlier divergence events in each of these cases involve shifts in the a niche, whereas b niche traits continue to diverge later in the radiation. Streelman and Danley (2003) include the anoles as one of the case studies in their review, considering the diversification of ecomorphs as an example of the first stage of radiation, involving microhabitat divergence. This is consistent with our interpretation of ecomorphs as diversification of the a niche, although our terminology is different. These case studies provide a striking contrast to Diamond’s (1986) habitat-first model, which proposed that the first stage of speciation and adaptive radiation involved allopatric divergence along habitat gradients. These contrasting interpretations reflect differences in the interpretation of fast- vs. slow-evolving traits, and they highlight underlying philosophical differences in the interpretation of comparative data. It is well known that phylogenetic inference works best for slowly evolving traits (Felsenstein 1988). Rapid evolution erodes the signal of early events on a phylogeny, due to reversals on the same or adjacent branches and convergence among the terminal states (strictly speaking, this is only true for traits with a finite number or range of states; Donoghue and Ree 2000, Ackerly and Nyffeler 2004). For this reason, independent contrasts will generally provide an extremely poor estimate of the timing of divergence for rapidly evolving traits, as it will appear that there was little divergence in deeper nodes (e.g., Fig. 1). The high level of uncertainty in ML ancestral state reconstructions reflects this problem for rapidly evolving traits (Schluter et al. 1997, Cunningham et al. 1998) and led us to adopt the bootstrap method proposed here. In this situation, if the most recent speciation events in an adaptive radiation involve divergence in macrohabitat or habitat-related traits, it is parsimonious to assume that deeper events also involved such divergences (T. Price, personal communication), though evidence of this will not be available from phylogenetic analysis. In contrast, for slowly evolving traits phylogenetic methods are quite powerful, leading to a different interpretation of the comparative data. For this case, as in the contrasting SLA values in Ceanothus, or divergent ecomorphs in Anolis, the greatest trait differences will generally be observed among distantly related species. The conservatism of these traits in diverse clades within each group strongly argues that the underlying divergences occurred during the initial speciation events early in the overall radiation. The view that a niche divergence occurs late in the course of species differentiation, because these traits tend to differ between distantly related species, is not compatible with comparative phylogenetic analysis. Taken together, these views lead to a possible synthesis of contrasting interpretations. In the cases discussed here, adaptive radiation may have proceeded by ‘‘a niche early’’ and ‘‘b niche throughout.’’ In other words, habitat divergence can occur frequently at all

July 2006

NICHE EVOLUTION AND ADAPTIVE RADIATION

S59

TABLE 3. Results of the divergence order test (DOT) for relative divergence times of specific leaf area (SLA) and climate niche parameters in Ceanothus. Trait

Mean age (SE)

D (SE)

No. nonzero

SLA January temperature Precipitation

0.00654 (0.00083) 0.00430 (0.00071) 0.00424 (0.00083)

 0.00225 (0.00108) 0.00230 (0.00117)

 94 92

Notes: Mean age is the average, over 100 alternative trees, of the weighted divergence times (W) for each trait. For each tree, W is derived by bootstrapping the ancestral states from maximum likelihood distributions for ancestral states, with 200 replicates; D is the mean difference in age of each climate parameter vs. SLA, over the 100 alternative trees; no. nonzero is the number of trees (out of 100) in which D was significantly different from zero, based on a one-tailed test of the hypothesis that mean age for SLA was older.

depths in the tree, although it can only be reconstructed with confidence in recent speciation events. Habitat differences clearly can play an important role in allopatric speciation, though it is important to note that allopatric populations might also occupy similar environments in different geographic areas (Peterson et al. 1999, Wiens 2004). In some cases, speciation also involves a shift in a niche traits related to resource partitioning in local communities, but these events are apparently less frequent. Divergence in a niche traits might be due to incidental divergence under conditions that favor different traits (e.g., island vs. mainland populations), or divergence by character displacement in sympatry due to direct competitive interactions between the incipient species (Schluter 2000, Levin 2004). If correct, the ‘‘a early, b throughout’’ model presents two unresolved questions about the evolution of a niche traits. First, why should these traits exhibit evolutionary transitions early on in adaptive radiations? (At the time of this initial divergence, the adaptive radiation is still an unrealized future of the clade.) And second, why do a niche traits often exhibit phylogenetic conservatism during subsequent diversification. With regard to the first question, it is tempting to identify divergence in a niche traits as key innovations, or invasions of a novel adaptive zone (sensu Simpson 1953), contributing to the subsequent radiation. For example, in the case of Ceanothus we have strong evidence that California chaparral communities are capable of supporting at least one sprouting and one nonsprouting Ceanothus species. Thus, when this trait diverged early in the evolution of Ceanothus (the direction of evolution between ancestral and derived states is unknown), the two descendent subclades were both able to diversify in parallel across a broad gradient of climatic conditions. While such scenarios are plausible, and may be correct, it is difficult to conduct rigorous analyses of diversification hypotheses in terms of the timing of phenotypic innovation and hypothesized shifts in speciation rates (Sanderson and Donoghue 1994, Hodges 1995). The second question addresses the important topic of niche conservatism: What are the mechanisms promoting evolutionary stasis in ecological traits through speciation and diversification of a clade? Recent theoretical analyses support the view that contrasting selection pressures in heterogeneous environments, combined with gene flow, interspecific competition, and/or habitat selection, can

generate stabilizing selection effects that lead to evolutionary stasis in niche parameters (Holt 1987, 2003, Kirkpatrick and Barton 1997, Case and Taper 2000). Empirical studies of these predictions are needed. The roles of niche conservatism in speciation, the evolution of regional biota and the assembly of communities has recently received increased attention (Webb et al. 2002, Ackerly 2003, Wiens 2004), and each of these offers a counterpoint to the emphasis on ecological divergence as a key component of evolutionary radiations. The conclusion that a niche traits evolve relatively slowly during an adaptive radiation contrasts with the views of Silvertown et al. (2006a, b) on niche evolution. They argue, based on several lines of evidence, that the a niche evolves rapidly, and as a corollary local communities usually show little phylogenetic signal in their species composition. This conclusion is supported by their analysis of the phylogenetic structure of English meadow communities (Silvertown et al. 2001, 2006a, b) and by comparison of niche overlap between congeneric and noncongeneric species within local communities. The apparent conflict between their analyses and our conclusions is most likely due to the different scales of analysis and sampling of taxa. The species of the English meadow communities are widely dispersed across the angiosperm phylogeny. When the phylogeny is pruned for analysis of niche distributions, the closest relatives remaining on the tree are rarely if ever immediate sibling species. As a result, even the most recent ‘‘events’’ represented on such a phylogeny are relatively old compared to our analysis of adaptive radiations. Thus, Silvertown et al.’s (2006a) conclusions and our analyses may be entirely compatible, but focused on different scales of analysis. CONCLUSIONS The divergence order test (DOT) introduced here provides a quantitative approach to test hypotheses about the relative sequence of divergence for continuous traits. Considering the potential pitfalls in comparative analysis of rapidly evolving traits, the bootstrap method incorporating the uncertainty of ancestral state reconstructions provides a conservative approach for significance testing. Our application of the DOT to Ceanothus, and interpretation of other cases in the literature, leads to the conclusion that a niche traits often diverge early in the course of adaptive radiations. The b niche traits,

S60

D. D. ACKERLY ET AL.

which are related to macrohabitat distributions, might evolve rapidly throughout the radiation, although the signal of early divergences is erased by the high rates of evolution. Further application of the DOT, or improved tests addressing these questions, will provide an important step towards synthesis of niche evolution and adaptive radiation. ACKNOWLEDGMENTS D. D. Ackerly thanks C. O. Webb and J. B. Losos for the invitation to contribute this paper to the ESA symposium and this special issue on Phylogenetics and Community Ecology. The authors thank D. Schluter and T. Price for valuable discussions and comments that improved the manuscript. This research was supported by National Science Foundation grants 0212873 to C. O. Webb, M. J. Donoghue, and D. D. Ackerly and 0078301 to D. D. Ackerly. LITERATURE CITED Ackerly, D. D. 2000. Taxon sampling, correlated evolution and independent contrasts. Evolution 54:1480–1492. Ackerly, D. D. 2003. Community assembly, niche conservatism and adaptive evolution in changing environments. International Journal of Plant Sciences 164:S165–S184. Ackerly, D. D. 2004a. Adaptation, niche conservatism and convergence: comparative studies of leaf evolution in the California chaparral. American Naturalist 163:654–671. Ackerly, D. D. 2004b. Functional strategies of chaparral shrubs in relation to seasonal water deficit and disturbance. Ecological Monographs 75:25–44. Ackerly, D. D., and R. Nyffeler. 2004. Evolutionary diversification of continuous traits: phylogenetic tests and application to seed size in the California flora. Evolutionary Ecology 18: 249–272. Axelrod, D. I. 1956. Mio-Pliocene floras from west-central Nevada. University of California Publications in Geological Sciences 33:1–316. Case, T. J., and M. L. Taper. 2000. Interspecific competition, environmental gradients, gene flow, and the coevolution of species’ borders. American Naturalist 155:583–605. Cavender-Bares, J., D. D. Ackerly, D. Baum, and F. A. Bazzaz. 2004. Phylogenetic repulsion in the assembly of Floridean oak communities. American Naturalist 163:823–843. Chaney, R. W. 1927. Geology and palaeontology of the Crooked River Basin with special reference to the Bridge Creek Flora. Carnegie Institution of Washington Publication 346:45–138. Chase, J., and M. Leibold. 2003. Ecological niches: linking classical and contemporary approaches. University of Chicago Press, Chicago, Illinois, USA. Chesson, P. L. 2000. Mechanisms of maintenance of species diversity. Annual Review of Ecology and Systematics 31: 343–367. Cunningham, C. W., K. Omland, and T. H. Oakley. 1998. Reconstructing ancestral character states: a critical reappraisal. Trends in Ecology and Evolution 13:361–366. Darwin, C. 1859. On the origin of species. Murray, London, UK. Davis, S. D., F. W. Ewers, J. Wood, J. J. Reeves, and K. J. Kolb. 1999. Differential susceptibility to xylem cavitation among three pairs of Ceanothus species in the Transverse mountain ranges of Southern California. Ecoscience 6:180–186. Diamond, J. M. 1986. Evolution of ecological segregation in the New Guinea montane avifauna. Pages 98–125 in J. M. Diamond and T. J. Case, editors. Community ecology. Harper and Row, Cambridge, Massachusetts, USA. Diaz-Uriarte, R., and T. Garland, Jr. 1996. Testing hypotheses of correlated evolution using phylogenetically independent contrasts: sensitivity to deviations from Brownian motion. Systematic Biology 45:27–47. Donoghue, M. J., and R. Ree. 2000. Homoplasy and developmental constraint: a model and example from plants. American Zoologist 40:759–769.

Ecology Special Issue

Elton, C. 1946. Competition and the structure of ecological communities. Journal of Animal Ecology 15:54–68. Felsenstein, J. 1985. Phylogenies and the comparative method. American Naturalist 125:1–15. Felsenstein, J. 1988. Phylogenies and quantitative characters. Annual Review of Ecology and Systematics. 19:445–471. Forstmeier, W., O. Bourski, and B. Leisler. 2001. Habitat choice in Phylloscopus warblers: the role of morphology, phylogeny and competition. Oecologia 128:566–576. Garland, T., Jr 1992. Rate tests for phenotypic evolution using phylogenetically independent contrasts. American Naturalist 140:509–519. Garland, T., Jr., P. H. Harvey, and A. R. Ives. 1992. Procedures for the analysis of comparative data using phylogenetically independent contrasts. Systematic Biology 41:18–32. Glor, R. E., J. J. Kolbe, R. Powell, A. Larson, and J. B. Losos. 2003. Phylogenetic analysis of ecological and morphological diversification in Hispaniolan trunk-ground anoles (Anolis cybotes group). Evolution 57:2383–2397. Gotelli, N. J., and G. Entsminger. 2001. Swap and fill algorithms in null model analysis: rethinking the knight’s tour. Oecologia 129:281–291. Gotelli, N., and G. Graves. 1996. Null models in ecology. Smithsonian Institution Press, Washington D.C., USA. Graham, A. 1999. Late Cretaceous and Cenozoic history of North American vegetation. Oxford University Press, New York, New York, USA. Graham, C. H., S. R. Ron, J. C. Santos, C. J. Schneider, and C. Moritz. 2004. Integrating phylogenetics and environmental niche models to explore speciation mechanisms in dendrobatid frogs. Evolution 58:1781–1793. Hacke, U., J. Sperry, W. Pockman, S. Davis, and K. McCulloch. 2001. Trends in wood density and structure are linked to prevention of xylem implosion by negative pressure. Oecologia 126:457–461. Hardig, T. M., P. S. Soltis, and D. E. Soltis. 2000. Diversification of the North American shrub genus Ceanothus (Rhamnaceae): conflicting phylogenies from nuclear ribosomal DNA and chloroplast DNA. American Journal of Botany 87:108–123. Harvey, P. H., and M. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Hellmers, H., J. S. Horton, G. Juhren, and J. O’Keefe. 1955. Root systems of some chaparral shrubs in Southern California. Ecology 36:667–678. Hickman, J. C. 1993. The Jepson manual: higher plants of California. University of California Press, Berkeley, California, USA. Hodges, S. 1995. Spurring plant diversification: Are floral nectar spurs a key innovation? Proceedings of the Royal Society (London) B 262:343–348. Holt, R. D. 1987. Population dynamics and evolutionary processes: the manifold roles of habitat selection. Evolutionary Ecology 1:331–347. Holt, R. D. 2003. On the evolutionary ecology of species’ ranges. Evolutionary Ecology Research 5:159–178. Hubbell, S. P. 2001. The unified neutral theory of biodiversity and biogeography. Princeton University Press, Princeton, New Jersey, USA. Jeong, S.-C., A. Liston, and D. D. Myrold. 1997. Molecular phylogeny of the genus Ceanothus (Rhamnaceae) using rbcL and ndhF sequences. Theoretical and Applied Genetics 94: 852–857. Keeley, J. E. 1977. Fire-dependent reproductive strategies in Arctostaphylos and Ceanothus. Pages 391–396 in H. A. Mooney and C. E. Conrad, editors. Proceedings of the Symposium on Environmental Consequences of Fire and Fuel Management in Mediterranean Ecosystems. USDA Forest Service General Technical Report WO-3, Washington, D.C., USA. Keeley, J. E., and P. H. Zedler. 1978. Reproduction of chaparral shrubs after fire: a comparison of sprouting and seeding strategies. American Midland Naturalist 99:142–161. Kirkpatrick, M., and N. H. Barton. 1997. Evolution of a species’ range. American Naturalist 150:1–23.

July 2006

NICHE EVOLUTION AND ADAPTIVE RADIATION

Knight, C., and D. D. Ackerly. 2001. Correlated evolution of chloroplast heat shock protein expression in closely related plant species. American Journal of Botany 88:411–418. Knouft, J. H., J. B. Losos, R. E. Glor, and J. J. Kolbe. 2006. Phylogenetic analysis of the evolution of the niche in lizards of the Anolis sagrei group. Ecology 87:S29–S38. Levin, D. A. 2004. Ecological speciation: crossing the divide. Systematic Botany 29:807–816. Losos, J., T. Jackman, A. Larson, K. de Queiroz, and L. Rodriguez-Schettino. 1998. Contingency and determinism in replicated adaptive radiations of island lizards. Science 279: 2115–2118. McMinn, H. 1942. Ceanothus. Volume II. A systematic study of the genus Ceanothus. Santa Barbara Botanical Gardens, Santa Barbara, California, USA. Nicholson, P. 1993. Ecological and historical biogeography of Ceanothus (Rhamnaceae) in the Transverse Ranges of Southern California. Dissertation. University of California, Los Angeles, California, USA. Oakley, T. H., and C. W. Cunningham. 2000. Independent contrasts succeed where ancestral reconstruction fails in a known bacteriophage phylogeny. Evolution 54:397–405. O’Meara, B. C, C. Ane´, M. J. Sanderson, and P. C. Wainwright. In press. Testing for different rates of evolution using likelihood. Evolution. Peterson, A. T., J. Soberon, and V. Sanchez-Cordero. 1999. Conservatism of ecological niches in evolutionary time. Science 285:1265–1267. Pickett, S., and F. Bazzaz. 1978. Organization of an assemblage of early successional species on a soil moisture gradient. Ecology 59:1248–1255. R Development Core Team. 2004. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Reich, P. B., M. B. Walters, and D. S. Ellsworth. 1997. From tropics to tundra: global convergence in plant functioning. Proceedings of the National Academy of Sciences (USA) 94: 13730–13734. Richman, A. D. 1996. Ecological diversification and community structure in the Old World leaf warblers (genus Phylloscopus): a phylogenetic perspective. Evolution 50:2461–2470. Richman, A., and T. Price. 1992. Evolution of ecological differences in Old World warblers. Nature 355:817–821. Sanderson, M. J. 2002. Estimating absolute rates of molecular evolution and divergence times: a penalized likelihood approach. Molecular Biology and Evolution. 19:101–109. Sanderson, M. J. 2003. r8s: Inferring absolute rates of molecular evolution and divergence times in the absence of a molecular clock. Bioinformatics 19:301–302. Sanderson, M., and M. Donoghue. 1994. Shifts in diversification rate with the origin of angiosperms. Science 264:1590– 1593.

Schluter, D. 2000. The ecology of adaptive radiations. Oxford University Press, Oxford, UK. Schluter, D., T. Price, A. Mooers, and D. Ludwig. 1997. Likelihood of ancestor states in adaptive radiation. Evolution 51:1699–1711. Schwilk, D. W., and D. D. Ackerly. 2005. Is there a cost to resprouting? Seedling growth rate and drought tolerance in sprouting and nonsprouting Ceanothus (Rhamnaceae). American Journal of Botany 92:404–410. Silvertown, J., M. Dodd, and D. Gowing. 2001. Phylogeny and the niche structure of meadow plant communities. Journal of Ecology 89:428–435. Silvertown, J., M. Dodd, D. Gowing, C. Lawson, and K. McConway. 2006a. Phylogeny and the hierarchical organization of plant diversity. Ecology 87:S39–S49. Silvertown, J., K. McConway, D. Gowing, M. Dodd, M. F. Fay, J. A. Joseph, and K. Dolphin. 2006b. Absence of phylogenetic signal in the niche structure of meadow plant communities. Proceedings of the Royal Society (London) B 273:39–44. Simpson, G. G. 1953. The major features of evolution. Columbia University Press, New York, New York, USA. Streelman, J. T., and P. D. Danley. 2003. The stages of vertebrate evolutionary radiation. Trends in Ecology and Evolution 18:126–131. Swofford, D. L. 2002. PAUP*: phylogenetic analysis using parsimony. Version 4.10b. Sinauer, Sunderland, Massachusetts, USA. Webb, C. O., D. D. Ackerly, and S. Kembel. 2004. Phylocom: software for the analysis of community phylogenetic structure and character evolution. hhttp://www.phylodiversity.net/ phylocomi Webb, C. O., D. D. Ackerly, M. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Wells, P. V. 1969. The relation between mode of reproduction and extent of speciation in woody genera of the California chaparral. Evolution 23:264–267. Whittaker, R. 1975. Communities and ecosystems. Macmillan, New York, New York, USA. Wiens, J. J. 2004. Speciation and ecology revisited: phylogenetic niche conservatism and the origin of species. Evolution 58: 193–197. Williams, C. B. 1964. Patterns in the balance of nature. Academic Press, New York, New York, USA. Wright, I. J., et al. 2004. The leaf economic spectrum worldwide. Nature 428:821–827.

APPENDIX A Discussion of alternative null models (Ecological Archives E087-110-A1).

APPENDIX B Ceanothus occurrence matrix (Ecological Archives E087-110-A2).

APPENDIX C Ceanothus trait matrix (Ecological Archives E087-110-A3).

SUPPLEMENT Source code for DOT (Ecological Archives E087-110-S1).

S61

Ecology, 87(7) Supplement, 2006, pp. S62–S75 Ó 2006 by the Ecological Society of America

PHYLOGENETIC DISPERSION OF HOST USE IN A TROPICAL INSECT HERBIVORE COMMUNITY GEORGE D. WEIBLEN,1,6 CAMPBELL O. WEBB,2,7 VOJTECH NOVOTNY,3 YVES BASSET,4

AND

SCOTT E. MILLER5

1

Department of Plant Biology, University of Minnesota, Saint Paul, Minnesota 55108 USA Section of Evolution and Ecology, University of California, Davis, California 95616 USA Institute of Entomology, Czech Academy of Sciences and Biological Faculty, University of South Bohemia, 370 05 Ceske Budejovice, Czech Republic 4 Smithsonian Tropical Research Institute, Balboa, Ancon, Panama 5 National Museum of Natural History, Smithsonian Institution, Washington, D.C. 20013-7012 USA 2

3

Abstract. Theory has long predicted that insect community structure should be related to host plant phylogeny. We examined the distribution of insect herbivore associations with respect to host plant phylogeny for caterpillars (Lepidoptera), beetles (Coleoptera), and grasshoppers and relatives (orthopteroids) in a New Guinea rain forest. We collected herbivores from three lineages of closely related woody plants and from more distantly related plant lineages in the same locality to examine the phylogenetic scale at which host specificity can be detected in a community sample. By grafting molecular phylogenies inferred from three different genes into a supertree, we developed a phylogenetic hypothesis for the host community. Feeding experiments were performed on more than 100 000 live insects collected from the 62 host species. We examined patterns of host use with respect to the host plant phylogeny. As predicted, we found a negative relationship between faunal similarity, defined as the proportion of all herbivores feeding on two hosts that are shared between the hosts, and the phylogenetic distance between hosts based on DNA sequence divergence. Host phylogenetic distance explained a significant fraction of the variance (25%) in herbivore community similarity, in spite of the many ecological factors that probably influence feeding patterns. Herbivore community similarity among congeneric hosts was high (50% on average) compared to overlap among host families (20–30% on average). We confirmed this pattern using the nearest taxon index (NTI) and net relatedness index (NRI) to quantify the extent of phylogenetic clustering in particular herbivore associations and to test whether patterns are significantly different from chance expectations. We found that 40% of caterpillar species showed significant phylogenetic clustering with respect to host plant associations, somewhat more so than for beetles or orthopteroids. We interpret this as evidence that a substantial fraction of tropical forest insect herbivores are clade specialists. Key words: community ecology; community phylogenetics; herbivory; host specialization; host specificity; plant–insect interactions; phylogenetic dispersion; phylogeny; tropical rain forest.

INTRODUCTION In the era before automated DNA sequencing and molecular phylogenetics, Daniel H. Janzen stated that ‘‘the systematics and taxonomy of interactions is hopeless’’ (Janzen 1977). As robust phylogeny estimates for plants and insects become available, investigating the evolutionary history of their interactions is no longer a fruitless endeavor. It is now possible to examine the historical associations of plants and insects by comparing molecular phylogenies for the interacting lineages (Becerra 1997, Weiblen and Bush 2002, Percy et al. 2004). Phylogenetic studies of host use by phytophagous Manuscript received 24 January 2005; revised 20 June 2005; accepted 23 June 2005. Corresponding Editor: A. A. Agrawal. For reprints of this Special Issue, see footnote 1, p. S1. 6 E-mail: [email protected] 7 Present affiliation: Arnold Arboretum of Harvard University.

insects have tended to focus on the reconstruction of ancestral associations for particular groups (Kelley and Farrell 1998) or whether particular insect groups and their host plants have diversified in parallel (Farrell and Mitter 1990, 1998). Other macroevolutionary studies have examined patterns of phylogenetic conservatism in the host plant associations of phytophagous insects (Farrell 1998, Janz and Nylin 1998, Ward et al. 2003). Ecologists interested in patterns of herbivore community structure are faced with a different set of questions. For example, to what extent do insects feed on closely related host plants in a particular community? How likely are host shifts to occur between divergent host lineages? Few studies have attempted to integrate the knowledge of phylogeny in the study of community structure (Connor et al. 1980, Strong et al. 1984, Marquis 1991, Losos 1996, Ødegaard 2003, Ødegaard et al. 2005). Concern over the lack of statistical independence among species led Kelly and Southwood

S62

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

(1999) to control for phylogenetic effects in demonstrating that host plant abundance can predict herbivore species richness in the temperate forest of Britain. But still more can be learned from phylogeny. The members of any biotic community are related in some fashion, and insights can be gained by examining ecological patterns with respect to patterns of descent from common ancestors. The incorporation of phylogenetic knowledge in ecological studies can inform our understanding of community structure (Webb et al. 2002) and of evolutionary constraints on the distribution of traits in ecological communities (Chazdon et al. 2003). A useful approach is to apply clustering indices to the phylogenetic distribution of species that belong to a particular community sample drawn from a larger species pool (Futuyma and Gould 1979, Webb 2000). Such indices were first applied to the distribution of phytophagous insects across a host plant phylogeny in order to quantify diet breadth (Symons and Beccaloni 1999, Beccaloni and Symons 2000). Early studies of diet breadth failed to consider the phylogenetic nonequivalence of taxonomic ranks (e.g. families and orders), and the phylogenetic diversity index and the clade dispersion index, in particular, were proposed to address this problem (Symons and Beccaloni 1999). However, these indices measured relatedness in terms of the branching order, not branch lengths, of phylogenies. Branch lengths are especially critical for studies of phylogenetic dispersion in ecological communities with an uneven distribution of closely related and distantly related species (CavenderBares et al. 2004). Consider lowland tropical rain forest tree communities, for example, which are often dominated by a relatively small number of highly species-rich genera and families (Novotny et al. 2002). In such cases, narrow host specificity of herbivores has been invoked to explain the maintenance of high insect species richness, but this conclusion was reached with little regard for host plant relatedness (Basset 1992). The analysis presented here builds on an earlier study (Novotny et al. 2002), expanding a New Guinea host plant assemblage from 51 to 62 species and applying new indices of phylogenetic dispersion to herbivore associations. The island of New Guinea is the third largest remaining area of tropical forest wilderness in the world and includes ;5% of global plant and insect diversity while occupying only 0.5% of the land area (Miller 1993). Our study site near Madang, on the north coast of Papua New Guinea, includes ;150 tree species/ha that measure .5 cm dbh, and species richness is dominated by approximately a dozen genera. We quantified the relationship between the herbivore community similarity of host trees and the phylogenetic distance between hosts. We defined similarity as the ratio of the number of herbivore species sharing two hosts to the total number of herbivore species feeding on the pair of hosts. Phylogenetic distance between host species was based on DNA sequence divergence

S63

integrated across three genes and rate-smoothed across the community phylogeny using penalized likelihood. If herbivores tend to feed on closely related plants more than on distantly related plants, as we expect, then faunal similarity should decline with increasing phylogenetic distance between host species. Indices of phylogenetic dispersion that incorporate null models can be especially useful as quantitative tests of host specificity in community samples. We used the nearest taxon index (NTI) and net relatedness index (NRI) to quantify the extent of phylogenetic clustering in particular herbivore associations and to test whether patterns are significantly different from chance expectations (Webb 2000, Webb et al. 2002). These indices measure the mean phylogenetic distance between plants that share a particular herbivore, relative to the mean and standard deviation of herbivore associations randomly distributed on the phylogeny, as obtained by multiple iteration. The NRI measures the average distance between all plants that share an herbivore species (i.e., the extent of overall clustering), while the NTI measures the average minimal distance between plants that share an herbivore species (i.e., the extent of terminal clustering). METHODS Community ecology Leaf-chewing insects were collected from 62 plant species representing 41 genera and 18 families (Table 1). Sampling effort was equalized across all host plants to provide quantitative estimates of herbivore relative abundance. Parataxonomists and village collectors surveyed 1500 m2 of foliage over nearly 1600 field-days and .6 3 104 tree inspections. Live insects were subjected to feeding trials with fresh foliage of the plant species from which they were collected in the field. These procedures are detailed in Novotny et al. (2002). We recorded 961 species and 62 193 individuals feeding on the 62 host plant species. Additionally, 40 000 insects that failed to feed on the plant from which they were collected were discarded. Local parataxonomists assigned feeding specimens to morphospecies (Basset et al. 2000), and taxonomic specialists later identified known taxa. Details on plant and insect identification are reported in Miller et al. (2003). One-quarter of all species were identified to named species, and 44% were identified to genus, but taxonomic knowledge varied from group to group. For example, 90% of the Lepidoptera species were assigned to a genus, and 72% were associated with a known species, while only 39% of beetles were assigned to genus and 19% to species. The locality, collection date, and host plant species for 37 972 mounted specimens are also available in our database. Digital photographs of many species are archived and available online.8 Sampling included 388 species and 8

hhttp://www.entu.cas.cz/png/index.htmli

S64 TABLE 1.

GEORGE D. WEIBLEN ET AL.

Ecology Special Issue

Plant species and gene sequences included in a phylogenetic study of host use in a tropical insect herbivore community. Species

Code

Family

Order

Clade

GenBank

Amaracarpus nymanii Valeton Artocarpus camansi Blanco Breynia cernua (Poir.) Muell. Arg Casearia erythrocarpa Sleum. Celtis philippensis Blanco Codiaeum ludovicianum Airy Shaw Dolicholobium oxylobum K. Schum. Dracaena angustifolia Roxb. Endospermum labios Schodde Eupomatia laurina R. Br. Excoecaria agallocha L. Ficus bernaysii King Ficus botryocarpa Miq. Ficus conocephalifolia Ridley Ficus copiosa Steud. Ficus dammaropsis Diels Ficus hispidioides S. Moore Ficus microcarpa L. Ficus nodosa Teysm. & Binn. Ficus phaeosyce Laut. & K. Schum. Ficus pungens Reinw. ex Bl. Ficus septica Burm. f. Ficus tinctoria Forst. Ficus trachypison K. Schum. Ficus variegata Bl. Ficus wassa Roxb. Gardenia hansemannii K. Schum. Gnetum gnemon L. Homalanthus novoguineensis (Warb.) K. Schum. Hydriastele microspadix (Becc.) Burret. Kibara cf. coriacea (Bl.) Tul. Leucosyke capitellata (Poir.) Wedd. Macaranga aleuritoides F. Muell. Macaranga bifoveata J. J. Smith Macaranga brachytricha A. Shaw Macaranga densiflora Warb. Macaranga novoguineensis J. J. Smith Macaranga quadriglandulosa Warb. Mallotus mollissimus (Geisel.) Airy Shaw Melanolepis multiglandulosa (Reinw. ex Bl.) Reichb. f. Morinda bracteata Roxb. Mussaenda scratchleyi Wernh. Nauclea orientalis (L.) L. Neonauclea clemensii Merr. & Perry Neuburgia corynocarpa (A.Gray) Leenh. Osmoxylon sessiliflorum (Lauterb.) W.R.Philipson Pavetta platyclada Lauterb. & K. Schum. Phyllanthus lamprophyllus Muell. Arg. Pimelodendron amboinicum Hassk. Pometia pinnata Forster Premna obtusifolia R.Br. Psychotria leptothyrsa Miquel Psychotria micralabastra (Laut. & Schum.) Val. Psychotria micrococca (Laut. & Schum.) Val. Psychotria ramuensis Sohmer Pterocarpus indicus Willd. Randia schumanniana Merrill & Perry Sterculia schumanniana (Lauterb.) Mildbr. Tabernaemontana aurantica Gaud. Tarenna buruensis (Miq.) Val. Timonius timon (Spreng.) Merr. Versteegia cauliflora (K. Schum. & Laut.)

AMA ART BRE CAS CEL COD DOL DRA END EUP EXC BER BOT CON COP DAM HIS MIC NOD PHA PUN SEP TIN TRA VAR WAS GAR GNE HON ARE STG LEU MAA MAP MAF MAD MAU MAQ MAL MEL MOR MUS SAR NEO NEU OSM PAV PHY PIM POM PRE PSF PSM PSS PSL PTE MEN STR TAB TAR TIT VER

Rubiaceae Moraceae Phyllanthaceae Flacourtiaceae Ulmaceae Euphorbiaceae Rubiaceae Agavaceae Euphorbiaceae Eupomatiaceae Euphorbiaceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Moraceae Rubiaceae Gnetaceae Euphorbiaceae Arecaceae Monimiaceae Urticaceae Euphorbiaceae Euphorbiaceae Euphorbiaceae Euphorbiaceae Euphorbiaceae Euphorbiaceae Euphorbiaceae Euphorbiaceae Rubiaceae Rubiaceae Rubiaceae Rubiaceae Loganiaceae Araliaceae Rubiaceae Phyllanthaceae Euphorbiaceae Sapindaceae Verbenaceae Rubiaceae Rubiaceae Rubiaceae Rubiaceae Fabaceae Rubiaceae Malvaceae Apocynaceae Rubiaceae Rubiaceae Rubiaceae

Gentianales Rosales Malphigiales Malphigiales Rosales Malphigiales Gentianales Asparagales Malphigiales Magnoliales Malphigiales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Rosales Gentianales Gnetales Malphigiales Arecales Laurales Rosales Malphigiales Malphigiales Malphigiales Malphigiales Malphigiales Malphigiales Malphigiales Malphigiales Gentianales Gentianales Gentianales Gentianales Gentianales Apiales Gentianales Malphigiales Malphigiales Sapindales Lamiales Gentianales Gentianales Gentianales Gentianales Fabales Gentianales Malvales Gentianales Gentianales Gentianales Gentianales

euasterids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 euasterids 1 monocots eurosids 1 basals eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 euasterids 1 outgroup eurosids 1 monocots basals eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 eurosids 1 euasterids 1 euasterids 1 euasterids 1 euasterids 1 euasterids 1 euasterids 2 euasterids 1 eurosids 1 eurosids 1 eurosids 2 euasterids 1 euasterids 1 euasterids 1 euasterids 1 euasterids 1 eurosids 1 euasterids 1 eurosids 2 euasterids 1 euasterids 1 euasterids 1 euasterids 1

AJ002176  AY289288 AY374311 AF206746  D86309  AY374312 AJ318445 AF206729  AY374313 L12644  AY374314 AF165378 AF165379 AF165381 AF165382 AF165383 AF165388 AF165393 AF165395 AF165401 AF165404 AF165409 AF165413 AF165414 AF165415 AF165418 AJ318446 AY056577 AY374315  AY012504  AF050221  AY208707  AY374319 AY374321 AY374316 AY374317 AY374320 AY374318 AY374322 AY374323 AJ318448 AJ318447 AJ318449 AJ318450 AJ001755 U50257  AJ318451 AY374325 AY374324 AJ403008  U28883  AJ318452 AJ318453 AJ318454 AJ318455 AF308721  AJ318456 AJ233140  X91772  AJ318457 AJ318458 AJ318459

Notes: When sequences were not available for particular species, substitutions of near relatives from GenBank were made (hhttp://www.ncbi.nlm.nih.govi). For example, rbcL sequences from Artocarpus altilis (AF500345) and Ficus heterophylla (AF500351) were substituted for ART and VAR, respectively. Additional substitutions are footnoted.   Substituted rbcL sequences Amaracarpus sp. (AMA), Casearia sylvestris (CAS), Celtis sinensis (CEL), Agave ghiesbreghtii (DRA), Eupomatia bennetti (EUP), Gnetum parvifolium (GNE), Kibara rigidifolia (STG), Hydriastele wendlandiana (ARE), Urtica dioica (LEU), Teraplasandra hawaiensis (OSM), Talisia nervosa (POM), Premna microphylla (PRE), Willardia mexicana (PTE), Sterculia apetala (STR), and Tabernaemontana divaricata (TAB).

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

24 481 individuals for beetles (Coleoptera), 464 species and 31 108 individuals for moths and butterflies (Lepidoptera; see Plate 1), and 109 species and 6605 individuals of orthopteroids (Orthoptera and Phasmatodea). Among the caterpillars, ;14 000 were matched with adults, amounting to 298 species of Lepidoptera with known larval and adult stages. Molecular phylogenetics Phylogenetic relationships for the 62 host plant species were drawn from multiple molecular data sets including a three-gene phylogeny for all angiosperms (Soltis et al. 1998). We used additional molecular markers for species of Moraceae, Rubiaceae, and Euphorbiaceae, including the internal transcribed spacer (ITS) region of nuclear ribosomal DNA for Ficus (Weiblen 2000), rbcL, encoding the large subunit of ribulose-1,5-bisphosphate carboxylase, and the 30S ribosomal protein S16 gene (rps16) for Rubiaceae (Novotny et al. 2002), and ndhF, encoding a subunit of NADH-plastoquinone oxidoreductase, for the Euphorbiaceae. Phylogenetic analyses of Euphorbiaceae based on ndhF are presented in the Appendix. Community phylogenetics A phylogeny estimate for the community sample was obtained by grafting less inclusive single-gene phylogenies for Ficus, Euphorbiaceae, and Rubiaceae into a more inclusive phylogeny of angiosperms based on three genes (Soltis et al. 1998). The assembly of a community phylogeny can follow supertree methods (Sanderson et al. 1998) or other approaches (Lapointe and Cucumel 1997), but one crucial difference is that only members of the community are retained in the supertree, while all other lineages are pruned away. It is important to consider the impact of branch length considerations on indices of phylogenetic clustering drawn from community samples. When branch lengths are assumed equal, using the number of intervening nodes as a proxy for phylogenetic distance (Novotny et al. 2002), relationships between intensively sampled congeneric species are given the same weight as relationships among representatives of major clades. Branch length information can distinguish between these two very different cases, short distances between congenerics and long distances between members of major lineages. Therefore, to incorporate information from all three molecular data sets, we scaled branch lengths in the supertree to the relative rate of change in two genes compared between pairs of taxa. For example, the relative rate of ITS to ndhF was calculated by counting the absolute number of character differences in each gene between Ficus microcarpa and F. variegata. Including all characters, there were 15 ndhF differences between these species and 58 ITS differences, yielding a relative rate of 0.259 for ndhF to ITS (Weiblen 2000, Datwyler and Weiblen 2004). Fifty-eight pairwise differences between Artocarpus camansi and Ficus variegata

S65

for rbcL and 111 for ndhF yielded a rate of 1.914 for ndhF relative to rbcL. We rescaled the branch lengths by these rates to approximate the phylogenetic distance between taxa sampled for genes showing radically different rates of molecular divergence. The assumption of this method is that rates of divergence for each gene are homogeneous among the lineages comprising the community sample. In the case of plant families other than Moraceae, Rubiaceae, and Euphorbiaceae, rbcL sequences were not necessarily available from the particular species, and in these instances sequences from related species or genera were obtained from GenBank as indicated in Table 1. The next challenge is to obtain a phylogeny for which all distances from the root of the tree to the tips are equal, also known as an ultrametric tree. Ultrametricity is necessary to make direct comparisons of phylogenetic distance (as measured by rescaled molecular branch lengths) among pairs of host species distributed across the phylogeny. Each individual data set rejected a molecular clock assumption, so we applied nonparametric rate smoothing and penalized likelihood as implemented in the program r8s (Sanderson 2002) to the rescaled branch lengths of the supertree to obtain an ultrametric tree accommodating rate heterogeneity across lineages. Penalized likelihood is a semiparametric method that allows substitution rates to vary among lineages according to a smoothing parameter (Sanderson 2002). The optimal smoothing parameter was chosen on the basis of the data by cross-validation involving the sequential pruning of taxa from the tree and parameter estimation to best predict the branch length of the pruned taxon (Sanderson 2003). We compared 20 cross-validation parameters beginning with zero and increasing by increments of log10(0.05) and chose the optimal smoothing parameter to minimize v2 error. Cross-validation was performed with the age of the root node fixed at one. Penalized-likelihood search parameters included 2000 maximum iterations, 10 multiple starts, and 30 optimization runs. Phylogenetic dispersion of herbivore associations Herbivore associations with each of the 62 host species were coded as either present or absent under two different assumptions, including or excluding solitary observations. Where r denotes the number of feeding records for a particular herbivore species on a particular host species, associations were coded as present when r . 1 or when r . 0 to exclude or include singletons, respectively. Varying this threshold allowed us to examine the sensitivity of findings based on presence/absence to extreme variation in herbivore abundance. We examined the distribution of herbivore associations across the host phylogeny, with indices of phylogenetic clustering as implemented in the program Phylocom (Webb et al. 2004).

S66

GEORGE D. WEIBLEN ET AL.

Ecology Special Issue

FIG. 1. Phylogenetic relationships of host plant species included in the study (see Table 1 for abbreviations). Brackets indicate the three major angiosperm clades that were sampled intensively. A supertree was assembled from separate analyses of DNA sequences for Rubiaceae (Novotny et al. 2002), Ficus (Weiblen 2000), Euphorbiaceae (see the Appendix), and angiosperms as a whole (Soltis et al. 1998, Angiosperm Phylogeny Group 2003). Branch lengths based on ITS, rbcL, and ndhF sequences for partially overlapping sets of taxa were rescaled in proportion to pairwise differences between selected species with published ITS and ndhF, or ndhF and rbcL, sequences (see Methods). Branch lengths as shown are proportional to absolute numbers of nucleotide changes under parsimony. The scale bar indicates 10 changes.

The net relatedness index measured the mean phylogenetic distance between all plants sharing a particular herbivore: NRI ¼ –(Xnet – X(n))/SD(n) where Xnet is the mean phylogenetic distance between all pairs of n host plants sharing an herbivore, and X(n) and SD(n) are the mean and standard deviation of phylogenetic

distance for n host plants randomly distributed on the phylogeny, obtained by multiple iteration. The nearest taxon index measured the distance between the two nearest hosts sharing a particular herbivore. This index is calculated in the same manner as NRI, except that Xnear is substituted for Xnet, where Xnear is the shortest

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

S67

FIG. 2. Molecular divergence among 62 selected, woody host plant species in lowland tropical rain forest on the island of New Guinea (see Table 1 for species abbreviations). The ultrametric tree was derived from penalized-likelihood analysis. (a) Shallowest split between families, Loganiaceae and Apocynaceae, (b) deepest crown radiation of a genus, Psychotria, and (c) shallowest crown radiation of a genus, Ficus. Brackets mark the angiosperm families and genera that were the focus of herbivore sampling. Branch lengths as shown are proportional to the number of nucleotide changes per site under maximum likelihood. The scale bar indicates 0.05 substitutions per site.

distance between all pairs of n host plants sharing an herbivore. High values of these indices suggest clustering, whereas low values point to evenness (i.e., overdispersion). We tested whether these measures of phylogenetic dispersion of herbivore associations across the community phylogeny were significantly different from chance expectations. Under a null model of

random association, we performed 1000 permutations of host associations to simulate a distribution of NRI and NTI for each herbivore species. A two-tailed test of significance evaluated the rank of observed values at P ¼ 0.05. For example, a rank of ,25 or .975 of 1000 permutations constituted significant overdispersion or clustering, respectively.

S68

GEORGE D. WEIBLEN ET AL.

Ecology Special Issue

TABLE 2. Numbers and percentages of insect herbivore species with significantly clustered (and overdispersed) patterns of host association across a community sample of 62 woody plant species from New Guinea lowland rain forest. Excluding singletons NRI Taxon Lepidoptera Lepidoptera (%) Coleoptera Coleoptera (%) Orthopteroids Orthopteroids (%) Total herbivores Herbivores (%)

NB 76 55 30 30 12 30 118 43

(0) (0) (1) (1) (0) (0) (1) (1)

BL 65 47 43 43 14 35 122 44

(1) (1) (0) (0) (0) (0) (1) (1)

NTI PL 61 45 46 46 9 22 116 42

(5) (4) (1) (1) (0) (0) (6) (2)

LF 61 45 46 46 9 22 116 42

(5) (4) (1) (1) (0) (0) (6) (2)

NB 68 50 26 26 8 20 102 37

(0) (0) (0) (0) (0) (0) (0) (0)

BL 58 42 27 27 4 10 89 32

(3) (2) (0) (0) (1) (2) (4) (1)

PL 60 44 24 24 4 10 88 32

(9) (6) (3) (3) (2) (5) (14) (5)

LF 60 44 24 24 4 10 88 32

(9) (6) (1) (1) (6) (15) (16) (6)

Notes: Two-tailed tests of phylogenetic dispersion assessed significance at P ¼ 0.05 with ranks .975 (or ,25) out of 1000 randomizations. Abbreviations: NRI ¼ net relatedness index; NTI ¼ nearest taxon index; NB ¼ no. branch lengths (no. intervening nodes); BL ¼ rescaled molecular branch lengths (nonultrametric); PL ¼ rescaled ultrametric branch lengths (penalized likelihood); LF ¼ rescaled ultrametric branch lengths (Langely-Fitch nonparametric rate smoothing).

We further examined the relationship of herbivore community similarity to the phylogenetic distance between hosts. We calculated community similarity as the percentage of the total number of herbivore species feeding on any pair of host species that were shared between the hosts (Novotny et al. 2002). We estimated phylogenetic distance from branch lengths based on DNA sequence divergence under penalized likelihood as implemented in r8s (Sanderson 2002). We used linear regression to analyze the direction and linear regression and Mantel tests to assess the significance of this relationship. RESULTS Community ecology Among the 62 193 insects, including 464 caterpillar species reared to adults, 388 beetle species, and 109 orthopteroids, there were 281 species collected as single individuals (singletons). Singleton species were excluded from subsequent analyses, because it is impossible to assess host range when a species is known from only one feeding record (Novotny and Basset 2000). Apart from singletons, our sample also included 156 herbivore species that fed on a single plant species. Our analysis did not examine whether these species are truly monophagous or were simply sampled in insufficient numbers. Rather, we focused on the host phylogenetic distribution of associations for the remaining 524 herbivore species (55% of the total) that were found to feed on more than one plant species. Community phylogeny A phylogeny was obtained for the host plant community sample by grafting hypotheses of relationship for selected Euphorbiaceae (see Appendix). Rubiaceae (Novotny et al. 2002), and Ficus (Weiblen 2000) to an ordinal phylogeny based on multiple data sets (Angiosperm Phylogeny Group 2003). The phylogeny is shown in Fig. 1 with rbcL and ITS branch lengths

rescaled in terms of ndhF substitutions. Nonparametric rate smoothing (Langely-Fitch) and penalized likelihood yielded highly similar ultrametric trees (Fig. 2). As expected, phylogenetic distances between congeneric species were lower than between confamilial genera and extaordinal families. Phylogenetic dispersion Each of 226 Lepidoptera, 212 Coleoptera, and 87 orthopteroid species observed on multiple hosts was tested for nonrandom patterns of association with respect to host plant phylogeny. Under a more stringent coding of host association that excluded all solitary feeding records, the 137 Lepidoptera, 99 Coleoptera, and 40 orthopteroid species encountered on multiple hosts (multiple times each) were also analyzed with respect to host phylogenetic dispersion. Results under four different branch length assumptions, two different indices of phylogenetic dispersion, and two feeding thresholds indicated that herbivores with nonrandom dispersion of associations feed on closely related hosts more often than on distantly related hosts (Table 2). In particular, 25–43% of the herbivore species we analyzed were significantly clustered on the host plant phylogeny compared to 0–6% that were overdispersed. The incorporation of sequence divergence in branch length estimation had a dramatic impact on the detection of phylogenetic dispersion. In the case of nearest taxon index, for example, results under the assumption of equal branch lengths only agreed with those under molecular branch length assumptions in 65% of cases, three variations on the latter agreed in 93% of cases, and the two assumptions based on ultrametric trees agreed in all cases. Exclusion of feeding records represented by single observations also enhanced the detection of nonrandom associations with respect to host plant phylogeny. Without singletons, 32–37% of herbivore species rejected the null model of association

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

S69

TABLE 2. Extended.

Including singletons NRI NB 124 55 41 19 20 23 185 35

(0) (0) (6) (3) (1) (1) (7) (1)

BL 93 41 70 33 27 31 190 36

(0) (0) (0) (0) (0) (0) (0) (0)

NTI PL 90 40 75 35 23 26 188 36

(4) (2) (3) (1) (0) (0) (7) (1)

LF 91 40 76 36 24 27 191 36

(4) (2) (3) (1) (1) (1) (8) (2)

compared to 25–31% including singletons in the analysis, a trend that was upheld by each of three insect groups. Community similarity and phylogenetic distance Herbivore community similarity, defined as the fraction of the total herbivore species on two host species that are shared between the hosts (Novotny et al. 2002), was negatively associated with phylogenetic distance as estimated by rate-smoothed molecular divergence under penalized likelihood (Fig. 3). The regression of community similarity against phylogenetic distance was highly significant (ANOVA, F1, 3842 ¼ 1243.5, P , 0.0001), and the correlation between these variables was also significant according to a Mantel test (Pearson’s productmoment correlation, r ¼ 0.423, P , 0.01). Declining community similarity with increasing phylogenetic distance between hosts indicates that herbivores tend to feed on closely related plants more often than on distantly related plants.

NB 103 46 45 21 13 15 161 31

BL

(0) (0) (0) (0) (0) (0) (0) (0)

78 34 44 21 8 9 130 25

(7) (3) (0) (0) (3) (3) (10) (2)

PL 75 33 49 23 9 10 133 25

(13) (6) (1) (0) (2) (2) (16) (3)

LF 76 34 50 23 10 11 136 26

(13) (6) (1) (0) (2) (2) (16) (3)

extant taxa. This is why we applied indices of phylogenetic dispersion to examine the relationship between herbivore associations and host plant phylogeny. Phylogenetic dispersion of host associations Community phylogenies, null models, and measures of phylogenetic dispersion taken together increase the precision with which herbivore associations can be studied. Previous attempts to quantify host specificity, for example, have either relied on taxonomic ranks that are not commensurate with plant lineages or ignored the branch length information contained in molecular phylogenies (Symons and Beccaloni 1999, Novotny et al. 2002). Branch length assumptions influence our

DISCUSSION While it is tempting to trace ecological character evolution on community phylogenies, ancestral reconstructions of host associations in community samples often yield implausible inferences. Equally weighted parsimony for highly polyphagous species implies that these herbivores colonized the common ancestors of major angiosperm clades and were subsequently lost from some host lineages. Consider for example the ancestral association of Rhinoscapha tricolor under equally weighted parsimony (Fig. 4). It is highly unlikely that this particular polyphagous species was associated with the common ancestor of the angiosperms and the gymnosperm Gnetum. Ancestral state reconstructions are sensitive to taxon sampling (Cunningham et al. 1998, Cunningham 1999), and colonization or extinction patterns cannot necessarily be inferred from local assemblages because community phylogenies are incomplete by definition. This problem is not unique to the evolution of host associations, but also occurs whenever the included taxa might be a subset of an entire clade of

FIG. 3. Herbivore community similarity as a function of the phylogenetic distance between host plants. Similarity is the fraction of the total fauna on two hosts that is shared between the hosts. Phylogenetic distance was derived from the penalized-likelihood ultrametric phylogram shown in Fig. 2. Means, standard deviations, and ranges of community similarity are shown for selected distance intervals. The outgroup is excluded from the regression.

S70

GEORGE D. WEIBLEN ET AL.

FIG. 4. Erroneous inference of ancestral host use in community samples under equally weighted parsimony. Rhinoscapha tricolor is a polyphagous generalist that parsimony suggests had an implausible, ancient association with the common ancestor of Gnetum and flowering plants. See Table 1 for species abbreviations.

power to detect patterns of phylogenetic dispersion in at least one important way. Failure to consider the extent of molecular divergence between hosts will underestimate the extent of herbivore clustering (or overdispersion) given that closely related hosts and extremely divergent hosts with the same number of intervening

Ecology Special Issue

nodes in the community phylogeny are assumed to be equidistant when they are not. Branch lengths scaled to molecular divergence distinguish between these cases and enhance the power to detect significant patterns in host use (Table 2). Ultrametric molecular branch lengths approximate relative ages of lineages, and thus the length of time for ecological associations or adaptations to arise. We found that a large proportion of herbivores feed on closely related plants, including congeneric species and confamilial genera, and that a small number of herbivores feed on more divergent hosts than expected by chance. The former pattern is expected in cases of herbivore specialization (Jaenike 1990, Futuyma et al. 1993) and the latter pattern when herbivores are tracking convergent chemical, morphological, or ecological host traits (Becerra 1997). The incorporation of molecular branch lengths in a community phylogeny assembled from multiple genes poses interesting methodological challenges that invite further exploration. Communities are usually composed of heterogeneous taxa, some very closely related and others distantly so. Grafting of multiple phylogenies based on different genes could be necessary when no single gene resolves phylogenetic relationships at all taxonomic levels in the community sample. This was the case in our sample, where ITS sequences were employed to resolve relationships among Ficus, but this region could not be aligned across plant families. Rescaling of branch lengths from different gene regions based on the ratio of absolute character differences between taxon pairs represents one possible solution among many. An improvement on our method would be to correct for multiple substitutions in a model-based maximum-likelihood framework when rescaling branch lengths across grafted phylogenies. We do not know the extent to which the phylogenetic dispersion of herbivores in our samples is representative of herbivore community structure on the complete local plant community or tropical rain forests in general. The scope of our sampling universe is incomplete for even the local community. Fifteen figs, 13 Rubiaceae, 13 Euphorbiaceae, and 21 other angiosperms hardly encompass the woody vegetation of a study area that contains hundreds of flowering plant species. The selection of study plants was made to replicate the taxonomic ranks of family and genus, and is at best a highly skewed sample in terms of local abundance and distribution. At least one way to avoid artifacts due to taxonomic unevenness is to restrict analyses to single representatives of given taxonomic ranks, such as families, but this is not satisfactory owing to the phylogenetic nonequivalence of taxa at any single rank. Age estimates of family clades in a recent study of angiosperms range from ,25 Ma to .150 Ma (Davies et al. 2004). The problem of taxonomic unevenness could be addressed by including all members of a local community, provided that the boundaries of the community can be defined. We intend to explore these

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

S71

FIG. 5. Phylogenetic dispersion of host range in 30 herbivore species arbitrarily selected from the community sample to illustrate the extremes of variation. Herbivores are grouped into nonsignificantly clustered species including polyphagous generalists, and significantly clustered species including oligophagous specialists feeding on Macaranga, Ficus, or Psychotria. Branch lengths of the host community phylogeny are proportional to molecular divergence as in Fig. 4, except for the truncated root indicated by a slash. Host species codes are defined in Table 1, and herbivore species codes are defined in Table 3. As in Fig. 4, solid boxes indicate herbivore presence and open boxes indicate herbivore absence.

issues in the future through the complete enumeration of vegetation in specific areas of forest (Novotny et al. 2004a). At the very least, it is encouraging that the relationship between herbivore community similarity and host phylogenetic distance was strengthened by the expansion of our sample from 51 host species in

Novotny et al. (2002) to 62 in the present study, and through the incorporation of branch length information. It is remarkable that a full quarter of the variance of herbivore community similarity can be explained by the phylogenetic relationships among hosts (r2 ¼ 0.244) when we consider the variability that environmental and

S72

GEORGE D. WEIBLEN ET AL.

Ecology Special Issue

TABLE 3. Herbivore species from Fig. 5 arranged alphabetically by morphospecies code. Code

Order

Family

Species

N

H

NRI

NTI

ACRI001 ACRI014 ACRI044 ARCT002 BUPR002 CHOR008 CHRY004 CHRY076 CHRY124 CRAM003 CRAM005 CRAM006 CRAM044 CURC002 CURC005 GEOM021 LYCA006 NOCT002 NYMP001 NYMP002 PHAS002 PHAS004 PHAS016 PSYC001 PYRA002 SPHI004 TORT006 TORT008 TORT040 TORT075 XXXX021 XXXX048 XXXX076

Orthopteroid Orthopteroid Orthopteroid Lepidoptera Coleoptera Lepidoptera Coleoptera Coleoptera Coleoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Coleoptera Coleoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Orthopteroid Orthopteroid Orthopteroid Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera Lepidoptera

Pyrgomorphidae Acrididae Eumastacidae Arctiidae Buprestidae Choreutidae Chrysomelidae Chrysomelidae Chrysomelidae Crambidae Crambidae Crambidae Crambidae Curculionidae Curculionidae Geometridae Lycaenidae Noctuidae Nymphalidae Nymphalidae Heteronemiidae Phasmatidae Phasmatidae Psychidae Pyralidae Sphingidae Choreutidae Tortricidae Tortricidae Thyrididae Pyralidae Gelechiidae Gelechiidae

Desmopterella biroi (Bolivar, 1905) Valanga papuasica (Finot, 1907) Paramnesicles buergersi Bolivar, 1930 Darantasia caerulescens Druce, 1899 Habroloma sp. Brenthia salaconia Meyrick, 1910 genus indeterminate Deretrichia sp. Deretrichia sp. Glyphodes margaritaria (Clerck) 1794 Talanga deliciosa (Butler) 1887 Talanga sexpunctalis (Moore) 1877 ‘‘Coelorhycidia’’ nr. purpurea Hampson, 1907 Apirocalus ebrius Faust, 1892 Alcidodes elegans (Guerin) 1838 Cleora repetita Butler, 1882 Philiris helena Snellen, 1887 Asota heliconia Linnaeus, 1758 Euploea leucosticos Gmelin, 1788 Cyrestis acilia Godart, 1819 Neopromachus vepres (Brunner von Wattenwyl) 1907 Dimorphodes prostasis Redtenbacher, 1908 Eurycantha insularis Lucas, 1869 Eumeta variegata Snellen, 1879 Orthaga melanoperalis Hampson, 1906 Macroglossum melas Rothschild & Jordan, 1930 Choreutis lutescens (Felder and Rogenhofer) 1875 Adoxophyes templana complex Homona mermerodes Meyrick, 1910 Mellea ramifera Warren, 1897 Unadophanes trissomita Turner, 1913 Dichomeris ochreoviridella (Pagenstecher) 1900 Dichomeris sp. nr. resignata Meyrick, 1929

2215 273 111 3 20 389 125 16 16 318 856 329 234 2349 141 12 121 257 108 156 211 98 38 33 246 167 332 482 815 56 301 394 324

58 49 29 3 3 7 6 5 6 11 14 13 5 54 22 9 8 9 11 12 41 35 25 19 7 5 13 29 25 7 5 6 10


0.36 1.09 1.28
0.54 3.34 5.60 6.60 5.01 0.55 9.04 10.02 9.50 2.65
1.31 1.94 0.11 5.54 7.70 8.12 9.52
0.15 0.06 1.65 0.72 5.99 4.32 9.50
1.54
0.1 5.99 4.97 5.79 6.01


0.74 0.55 1.94
0.89 2.33 2.38 2.60 2.45
0.26 3.22 3.37 3.21 1.51
1.38 2.51 0.21 2.00 2.94 3.04 3.32
1.43
0.74
0.68 0.12 2.55 2.11 3.21
1.65
1.53 2.55 2.35 2.48 1.88

Notes: The total number of individuals (N) and the total number of host species (H), including solitary feeding records, are indicated. Net relatedness (NRI) and nearest taxon (NTI) indices are reported as calculated under the penalized-likelihood tree (Fig. 2).

population demographical heterogeneity must inevitably contribute to samples of herbivore associations from any site over any period of time. A recent community phylogenetic analysis of host use by beetles in Panamanian rain forest revealed the same pattern (Ødegaard et al. 2005). These findings provide quantitative support for long-standing theory (Ehrlich and Raven 1964, Strong et al. 1984, Schoonhoven et al. 1998). There are at least two explanations for the decline in herbivore similarity with increasing phylogenetic distance between hosts. The first has to do with the phylogenetic conservatism of host choice as manifest in the tendency for herbivore offspring to feed on the same host lineages as their parents. Second, it is possible that host choice is influenced by the conservatism of chemical, morphological, ecological, and physiological plant traits affecting herbivore performance. Power to detect phylogenetic conservatism in community samples could be improved by considering species abundance and increasing the universe of sampled hosts. Nonetheless, species presence/absence and a limited sample of the local plant community indicated a relatively strong influence of host relatedness on herbivore community composition.

Tests of host specificity Community phylogenies and null models provide a quantitative test of significance for host specificity at the clade level. Examples of specialists on Macaranga, Ficus, and Psychotria that rejected null models of host association are shown in Fig. 5 and Table 3, along with nonspecialists that failed to reject null models. These examples were chosen to illustrate extreme cases and to reinforce the point that a quantitative definition of host specificity based on phylogenetic dispersion is more practical and powerful than definitions based on arbitrary taxonomic ranks. When singleton records were excluded from analyses, more host clade specialists were detected in all herbivore assemblages (Table 2). This result is not surprising given that herbivores tend to have a highly skewed distribution of abundance across the host range. The average herbivore species in New Guinea secondary forest, for example, has .90% of individuals aggregated on a single host species and feeds on one to three host species (Novotny et al. 2004b). Singletons representing rare or anomalous feeding events are likely to increase error rates in analyses such as ours that ignore abundance distributions and simply treat the associations as present or absent.

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

S73

PLATE 1. Darantasia caerulescens Druce (Lepidoptera: Arctiidae): (A) Adult, (B) larva, and (C) male genitalia, with aedoeagus separated and vesica inflated. Genitalia, illustrated here for the first time, allow differentiation from similar-looking species. The caterpillars of this moth fed on three distantly related plant species (Fig. 5). Photo and dissection credit: Karolyn Darrow.

Excluding singletons and considering molecular branch lengths, the NRI and NTI differed as expected, considering that NTI quantifies dispersion near the tips of the phylogeny whereas NRI measures overall dispersion. According to NRI, Lepidoptera was proportionally the most specialized fauna, with 42–44% of species significantly clustered with respect to host plant phylogeny, compared with 24–27% of coleopterans and 10% of orthopteroids. By contrast grasshoppers and relatives showed the highest proportion of overdispersed species (2–15%), compared with Lepidoptera (2–6%) and Coleoptera (0–3%). We attribute these differences to variation among feeding guilds. We expected Lepidoptera to show the greatest overall degree of host specificity, due to the fact that caterpillars feeding on foliage were reared from larvae to adults and host specificity is manifest at the larval stage. Coleoptera, on the other hand, were tested for feeding only as adults and potentially feed on a more restricted set of hosts as larvae. Root-, stem-, and wood-boring beetle larvae are expected to exhibit greater host specialization than adult stages, because the impact of plant chemistry on insect development is strongest in the early life stages (Mattson et al. 1988). The nonholometabolous assemblage of grasshoppers and relatives, feeding as nymphs and adults, are widely regarded as polyphagous (Chapman and Sword 1997) and therefore expected to show less phylogenetic clustering and greater overdispersion than the other assemblages. Orthopteroid nymphs are more mobile than caterpillars, enhancing opportunities to graze on multiple hosts and presumably selection for greater breadth of diet (Chapman and Sword 1997).

Clustering of similar plant traits in close relatives due to phylogenetic conservatism (Cavender-Bares et al. 2004) provides a simple explanation for the extreme patterns of clade specialization observed in many herbivore species. We believe that herbivore tracking of phylogenetically conserved plant traits is a more plausible explanation than co-cladogensis for patterns of association in many plant-herbivore interactions. Predation and parasitism might also indirectly promote specialization in phytophagous insect communities (Bernays and Graham 1988). Attack rates of caterpillar parasitoids in temperate forests, for example, vary among host plant species, and this variation has the potential to influence the evolution of herbivore host range (Lill et al. 2002). Apart from patterns of clade specialization, we also detected a small number of cases of phylogenetic overdispersion (3–6% of all herbivores) that could have a biological explanation. Significant overdispersion is expected for herbivores feeding on distantly related hosts when hosts share a set of convergent traits that are palatable to particular herbivores (Cavender-Bares and Wilczek 2003). Convergence in plant traits can result from adaptive evolution (Agrawal and Fishbein 2006) or habitat specialization (Fine et al. 2006). For example, Ficus tinctoria (Moraceae) and Excoecaria agallocha (Euphorbiaceae) share an extreme environment and a unique set of herbivores along the seacoast. The identification of convergent ecophysiological, morphological, and chemical traits in distantly related hosts sharing similar herbivores might point to factors that limit the evolution of host range. Where trait convergence enables similar insects to feed on highly diverged plant lineages, we expect significant

S74

GEORGE D. WEIBLEN ET AL.

herbivore clustering in more than one place on the plant phylogeny, causing the nearest taxon index to be significantly high when the net relatedness index is not. Conclusions This study of herbivore associations illustrates how the integration of community ecology and phylogeny can detect patterns of host specialization. A community phylogeny with molecular branch lengths and null models identified patterns of phylogenetic clustering in the associations of insect herbivores feeding on a sample of tropical rain forest vegetation in New Guinea. Quantitative, community phylogenetic studies such as ours show a general tendency for insects to feed on closely related hosts (Ødegaard et al. 2005). As predicted, we found greater phylogenetic structure in caterpillar associations than in herbivorous beetles or orthopteroids. Quantifying the phylogenetic dispersion of host associations has advantages over approaches that ignore phylogenetic distance or assume the equivalence of taxa of the same rank. Indices of phylogenetic dispersion provide a quantitative definition of host specificity that can be compared among studies, solving a problem that has plagued herbivore community ecology from the very beginning. The approach provides not only a standard for the identification of specialists, but also holds promise for the study of host shifts and the identification of alternative hosts. ACKNOWLEDGMENTS We thank D. Althoff, K. Seagraves, and two anonymous reviewers for helpful comments; S. I. Silvieus for Euphorbiaceae sequencing; S. L. Datwlyer and S. Swenson for analytical assistance; C. Bellamy, J. Brown, K. Darrow, D. R. Davis, J. D. Holloway, M. Horak, S. James, E. G. Munroe, P. Nasrecki, D. Perez, G. Robinson, G. A. Samuelson, K. Sattler, M. Shaffer, M. A. Solis, G. Setliff, W. Takeuchi, K. Tuck, H. C. M. Van Herwaarden, and P. van Welzen for taxonomic assistance; and the staff of the New Guinea Binatang Research Center for field assistance. This material is based upon work supported by the National Science Foundation under Grants 9407297, 9628840, 9707928, 0212873, and 0211591, the Czech Ministry of Education (ME646, 6007665801), the Czech Grant Agency (206/04/0725), the Czech Academy of Sciences (Z50070508), and a David and Lucille Packard Fellowship in Science and Engineering. LITERATURE CITED Agrawal, A. A., and M. Fishbein. 2006. Plant defense syndromes. Ecology 87:S132–S149. Angiosperm Phylogeny Group. 2003. An update of the Angiosperm Phylogeny Group classification for the orders and families of flowering plants. Botanical Journal of the Linnean Society 141:399–436. Basset, Y. 1992. Host specificity of arboreal and free-living insect herbivores in rain forests. Biological Journal of the Linnean Society 47:115–133. Basset, Y., V. Novotny, S. E. Miller, and R. Pyle. 2000. Quantifying biodiversity: experience with parataxonomists and digital photography in Papua New Guinea and Guyana. BioScience 50:899–908. Beccaloni, G. W., and F. B. Symons. 2000. Variation of butterfly diet breadth in relation to host-plant predictability: results from two faunas. Oikos 90:50–66.

Ecology Special Issue

Becerra, J. X. 1997. Insects on plants: macroevolutionary chemical trends in host use. Science 276:253–256. Bernays, E., and M. Graham. 1988. On the evolution of phytophagous arthropods. Ecology 69:886–892. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Cavender-Bares, J., and A. Wilczek. 2003. Integrating microand macroevolutionary processes in community ecology. Ecology 84:592–597. Chapman, R. F., and G. A. Sword. 1997. Polyphagy in the Acridomorpha. Pages 183–195 in S. K. Gangwere, M. C. Muralirangan, and M. Muralirangan, editors. The bionomics of grasshoppers, katydids and their kin. CAB International, Wallingford, UK. Chazdon, R. L., C. Careaga, C. Webb, and O. Vargas. 2003. Community and phylogenetic structure of reproductive traits of woody species in wet tropical forests. Ecological Monographs 73:331–348. Connor, E. F., S. H. Faeth, D. Simberloff, and P. A. Opler. 1980. Taxonomic isolation and the accumulation of herbivorous insects: a comparison of introduced and native trees. Ecological Entomology 5:205–211. Cunningham, C. W. 1999. Some limitations of ancestral character-state reconstruction when testing evolutionary hypotheses. Systematic Biology 48:665–674. Cunningham, C. W., K. E. Omland, and T. H. Oakley. 1998. Reconstructing ancestral character states: a critical reappraisal. Trends in Ecology and Evolution 13:361–366. Datwyler, S. L., and G. D. Weiblen. 2004. On the origin of the fig: phylogenetic relationships of Moraceae from ndhF sequences. American Journal of Botany 91:767–777. Davies, T. J., T. G. Barraclough, M. W. Chase, P. S. Soltis, D. E. Soltis, and V. Savolainen. 2004. Darwin’s abominable mystery: insights from a supertree of the angiosperms. Proceedings of the National Academy of Sciences (USA) 101:1904–1909. Ehrlich, P. R., and P. H. Raven. 1964. Butterflies and plants: a study in coevolution. Evolution 18:586–608. Farrell, B. D. 1998. ‘‘Inordinate fondness’’ explained: Why are there so many beetles? Science 281:555–559. Farrell, B. D., and C. Mitter. 1990. Phylogenesis of insect/plant interactions: Have Phyllobrotica leaf beetles (Chrysomelidae) and the Lamiales diversified in parallel? Evolution 44:1389– 1403. Farrell, B. D., and C. Mitter. 1998. The timing of insect/plant diversification: Might Tetraopes (Coleoptera: Cerambycidae) and Asclepias (Asclepiadaceae) have co-evolved? Biological Journal of the Linnean Society 63:553–577. Fine, P. V. A., Z. J. Miller, I. Mesones, S. Irazuzta, H. M. Appel, M. H. H. Stevens, I. Sa¨a¨ksja¨rvi, J. C. Schultz, and P. D. Coley. 2006. The growth–defense trade-off and habitat specialization by plants in Amazonian forests. Ecology 87:S150–S162. Futuyma, D. J., and F. Gould. 1979. Associations of plants and insects in a deciduous forest. Ecological Monographs 49:33– 50. Futuyma, D. J., M. C. Keese, and S. J. Scheffer. 1993. Genetic constraints and the phylogeny of insect–plant associations: responses of Ophraella communa (Coleoptera: Chrysomelidae) to host plants of its congeners. Evolution 47:888–905. Jaenike, J. 1990. Host specialization in phytophagous insects. Annual Review of Ecology and Systematics 21:243–273. Janz, N., and S. Nylin. 1998. Butterflies and plants: a phylogenetic study. Evolution 52:486–502. Janzen, D. H. 1977. Promising directions of study in tropical animal–plant interactions. Annals of the Missouri Botanical Garden 64:706–736. Kelley, S. T., and B. D. Farrell. 1998. Is specialization a dead end? The phylogeny of host use in Dendroctonus bark beetles (Scolytidae). Evolution 52:1731–1743. Kelly, C. K., and T. R. E. Southwood. 1999. Species richness and resource availability: a phylogenetic analysis of insects

July 2006

PHYLOGENY OF HOST ASSOCIATIONS

associated with trees. Proceedings of the National Academy of Sciences (USA) 96:8013–8016. Lapointe, F.-J., and G. Cucumel. 1997. The average consensus procedure: combination of weighted trees containing identical or overlapping sets of taxa. Systematic Biology 46:306– 312. Lill, J. T., R. J. Marquis, and R. E. Ricklefs. 2002. Host plants influence parasitism of forest caterpillars. Nature 417:170– 173. Losos, J. B. 1996. Phylogenetic perspectives on community ecology. Ecology 77:1344–1354. Marquis, R. J. 1991. Herbivore fauna of Piper (Piperaceae) in a Costa Rican wet forest: diversity, specificity and impact. Pages 179–208 in P. W. Price, T. M. Lewinsohn, G. W. Fernandes, and W. W. Benson, editors. Plant–animal interactions: evolutionary ecology in tropical and temperate regions. Wiley, New York, New York, USA. Mattson, W. J., R. K. Lawrence, R. A. Haack, D. A. Herms, and P. J. Charles. 1988. Defensive strategies of woody plants against different insect-feeding guilds in relation to plant ecological strategies and intimacy of association with insects. Pages 1–38 in W. J. Mattson, J. Levieux, and C. BernardDagan, editors. Mechanisms of woody plant defenses against insects: search for pattern. Springer-Verlag, New York, New York, USA. Miller, S. E. 1993. Biodiversity and conservation of the nonmarine invertebrate fauna of Papua New Guinea. Pages 227–325 in B. Beehler, editor. Papua New Guinea conservation needs assessment. Volume 2. Department of Environment and Conservation, Boroko, Papua New Guinea. Miller, S. E., V. Novotny, and Y. Basset. 2003. Studies of New Guinea moths. 1. Introduction (Lepidoptera). Proceedings of the Entomological Society of Washington 105:1034–1042. Novotny, V., and Y. Basset. 2000. Rare species in communities of tropical insect herbivores: pondering the mystery of singletons. Oikos 89:564–572. Novotny, V., Y. Basset, S. E. Miller, R. L. Kitching, M. Laidlaw, P. Drozd, and L. Cizek. 2004a. Local species richness of leaf-chewing insects feeding on woody plants from one hectare of a lowland rainforest. Conservation Biology 18: 227–237. Novotny, V., Y. Basset, S. E. Miller, G. D. Weiblen, B. Bremer, L. Cizek, and P. Drozd. 2002. Low host specificity of herbivorous insects in a tropical forest. Nature 416:841–844. Novotny, V., S. E. Miller, J. Leps, Y. Basset, D. Bito, M. Janda, J. Hulcr, K. Damas, and G. D. Weiblen. 2004b. No tree an island: the plant–caterpillar food web of a secondary rainforest in New Guinea. Ecology Letters 7:1090–1100. Ødegaard, F. 2003. Taxonomic composition and host specificity of phytophagous beetles in a dry forest in Panama. Pages 220–236 in Y. Basset, V. Novotny, S. E. Miller, and R. L. Kitching, editors. Arthropods of tropical forests: spatio-

S75

temporal dynamics and resource use in the canopy. Cambridge University Press, Cambridge, UK. Ødegaard, F., O. H. Diserud, and K. Østbye. 2005. The importance of plant relatedness for host utilization among phytophagous insects. Ecology Letters 8:612–617. Percy, D. M., R. D. M. Page, and Q. C. B. Cronk. 2004. Plant– insect interactions: double-dating associated insect and plant lineages reveals asynchronous radiations. Systematic Biology 53:120–127. Sanderson, M. J. 2002. Estimating absolute rates of molecular evolution and divergence times: a penalized likelihood approach. Molecular Biology and Evolution 19:101–109. Sanderson, M. J. 2003. r8s: inferring absolute rates of molecular evolution and divergence times in the absence of a molecular clock. Bioinformatics 19:301–302. Sanderson, M. J., A. Purvis, and C. Henze. 1998. Phylogenetic supertrees: assembling the trees of life. Trends in Ecology and Evolution 13:105–109. Schoonhoven, L. M., T. Jermy, , and J. J. A. van Loon, editors. 1998. Insect–plant biology: from physiology to evolution. Chapman and Hall, London, UK. Soltis, D. E., P. S. Soltis, M. E. Mort, M. W. Chase, V. Savolainen, S. B. Hoot, and C. M. Morton. 1998. Inferring complex phylogenies using parsimony: an empirical approach using three large DNA data sets for angiosperms. Systematic Biology 47:32–42. Strong, D. R., J. H. Lawton, and S. R. Southwood. 1984. Insects on plants. Harvard University Press, Cambridge, Massachusetts, USA. Symons, F. B., and G. W. Beccaloni. 1999. Phylogenetic indices for measuring the diet breadths of phytophagous insects. Oecologia 119:427–434. Ward, L. K., A. Hackshaw, and R. T. Clark. 2003. Do foodplant preferences of modern families of phytophagous insects and mites reflect past evolution with plants? Biological Journal of the Linnean Society 78:51–83. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, and S. Kembel. 2004. Phylocom: software for the analysis of community phylogenetic structure and character evolution. hhttp://www.phylodiversity.net/ phylocomi Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Weiblen, G. D. 2000. Phylogenetic relationships of functionally dioecious Ficus (Moraceae) based on ribosomal DNA sequences and morphology. American Journal of Botany 87:1342–1357. Weiblen, G. D., and G. L. Bush. 2002. Speciation in fig pollinators and parasites. Molecular Ecology 11:1573–1578.

APPENDIX A description of Euphorbiaceae phylogeny (Ecological Archives E087-111-A1).

Ecology, 87(7) Supplement, 2006, pp. S76–S85 Ó 2006 by the Ecological Society of America

ECOLOGICAL FITTING AS A DETERMINANT OF THE COMMUNITY STRUCTURE OF PLATYHELMINTH PARASITES OF ANURANS DANIEL R. BROOKS,1,5 VIRGINIA LEO´N-RE`GAGNON,2 DEBORAH A. MCLENNAN,3 2

AND

DEREK ZELMER4

1 Department of Zoology, University of Toronto, Ontario M5S 3G5 Canada Laboratorio de Helmintologı´a, Instituto de Biologı´a, Universidad Nacional Auto´noma de Me´xico, C. P. 04510, D. F. Me´xico, Me´xico 3 Department of Zoology, University of Toronto, Ontario M5S 3G5 Canada 4 Department of Biological Sciences, Emporia State University, Emporia, Kansas 66801 USA

Abstract. Host–parasite associations are assumed to be ecologically specialized, tightly coevolved systems driven by mutual modification in which host switching is a rare phenomenon. Ecological fitting, however, increases the probability of host switching, creating incongruences between host and parasite phylogenies, when (1) specialization on a particular host resource is a shared characteristic of distantly related parasites, and (2) the resource being tracked by the parasite is widespread among many host species. We investigated the effect of ecological fitting on structuring the platyhelminth communities of anurans from a temperate forest and grassland in the United States and tropical dry and wet forests in Mexico and Costa Rica. The six communities all exhibit similar structure in terms of the genera and families inhabiting the frogs. Parasite species richness is highly correlated with the amount of time a host spends in association with aquatic habitats, a conservative aspect of both parasite and host natural history, and determined in a proximal sense by host mobility and diet breadth. The pattern of parasite genera and families within host genera across the regions examined is consistent with the prediction that ecological fitting by phylogenetically conservative species, coupled with historical accidents of speciation and dispersal, should be evidenced as a nestedsubset structure; the shared requirement for aquatic habitats of tadpoles provides a baseline assemblage to which other parasite taxa are added as a function of adult host association with aquatic habitats. We conclude that parasite communities are structured by both ecological fitting and coevolution (mutual modification), the relative influences of which are expected to vary among different communities and associations. Key words: anurans; coevolution; community structure; Costa Rica; ecological fitting; frogs; Mexico; nested subset; parasitic platyhelminths; phylogenetic conservatism; toads; United States of America.

INTRODUCTION There are two approaches to studying the evolution of host–parasite associations. The first and newer research program, maximum co-speciation, assumes that hosts and their parasites share such a specialized and exclusive evolutionary association (Page 2003, Clayton et al. 2004, Johnson and Clayton 2004) that speciation in one lineage causes speciation in the other (synchronous cospeciation; Hafner and Nadler 1988, 1990). Host– parasite phylogenies are thus expected to be completely congruent, with departures from congruence explained by invoking extinction in one lineage or the other. The second and original research program (Brooks 1979) is also based upon comparing host–parasite phylogenies and identifying points of congruence as instances of cospeciation (the term coined by Brooks in [1979]). There are, however, no assumptions about underlying proManuscript received 20 January 2005; revised 12 September 2005; accepted 21 September 2005. Corresponding Editor (ad hoc): C. O. Webb. For reprints of this Special Issue, see footnote 1, p. S1. 5 E-mail: [email protected]

cesses, nor is there an expectation of complete congruence. Brooks proposed that the incongruent portions of host–parasite phylogenies falsified the hypothesis of co-speciation at those nodes and thus required investigations into the influence of other factors (e.g., dispersal and host switching) on the evolution of the association. For example, parasites might diverge more rapidly than their hosts via sympatric speciation, producing sister species inhabiting the same host (Brooks and McLennan 1993; or ‘‘lineage duplication’’ sensu Page [2003]), or ecological or immunological evolution in the host lineage could cause parasite extinction (lineage sorting or ‘‘missing the boat’’ sensu Page [2003]). Although the maximum co-speciation program has been moving closer to Brooks’ propositions about the way incongruences should be treated, there is still one area of dispute between the two perspectives, the importance of host switching during the evolution of host–parasite associations. This debate is a logical extension of the assumption that hosts and parasites share a specialized exclusive evolutionary association, making it extremely unlikely that a parasite could

S76

July 2006

ECOLOGICAL FITTING OF ANURAN PARASITES

S77

PLATE 1. Major habitat types in Area de Conservacion Guanacaste, Costa Rica (clockwise from upper left); Pitilla approach, tropical cloud forest habitat at entrance to Pitilla Field Station; Cuajiniquil approach, Santa Elena Peninsula, seen from road to Cuajiniquil, tropical dry forest habitat; Rio Pizote, permanent swamp near Rio Pizote, between Dos Rios and Brasilia, tropical rain forest habitat. Pitilla Forest, tropical cloud forest habitat near Pitilla Field Station. All photos were taken in June 2005. Photo credits: D. R. Brooks

change host species. This assumption, however, arises from believing that it is the host species, not a biological characteristic or combination of characteristics of the host, that is important to the parasite (Brooks and McLennan 1993). Once researchers began thinking in terms of traits rather than taxonomy, it became evident that parasites might be able to switch hosts if the trait they were tracking was shared among two or more hosts. The fact that present-day associations might be shaped in part by the distribution of phylogenetically conservative traits is called ecological fitting (Janzen 1985). There are many macroevolutionary manifestations of ecological fitting. For example, any given parasite species might be a resource specialist, but also might share that specialist trait with one or more close relatives. That is, specialization on a particular resource can be plesiomorphic within a group (for an extensive discussion and examples see Brooks and McLennan [2002]). On the other hand, the resource itself might be at once very specific and taxonomically and geographically widespread if it is a persistent plesiomorphic trait in the hosts. The evolutionary basis for ecological fitting is thus deceptively simple, yet powerful. If specific cues/ resources are widespread, or if traits can have multiple

functions (or both), then the stage is set for the appearance of ecological specialization and close (co)evolutionary tracking as well as host switching. Ecological fitting thus explains how a parasite can be ecologically specialized and still switch hosts: if the resource is widespread across many host species, then the parasite can take advantage of an opportunity to establish a ‘‘new’’ specialized association without the cost of evolving novel abilities (Brooks and McLennan 2002). Just because a resource is widespread does mean that it is automatically available. The geographic distribution of the parasite might not coincide with the geographic distribution of all hosts having the resource (Pellmyr 1992a, b), or some other aspect of host biology might make the resource inaccessible to the parasite. For example, if host species A bearing resource x is highly abundant in a community, then less-abundant host species B and C, which also bear x might not be ‘‘apparent’’ to a parasite specializing on that resource (Feeny 1976, Wiklund 1984, Courtney 1985). Such density-dependent factors provide the appearance of close ecological tracking between the parasite and species A at time T0. If some environmental stressor later decreases the abundance of species A, and C

S78

DANIEL R. BROOKS ET AL.

Ecology Special Issue

FIG. 1. Ecological fitting in frog lung flukes. A clade of lung flukes (Haematoloechus spp.) arose in conjunction with the evolution of leopard frogs (rpg; Rana pipiens group). Haematoloechus floedae arose through speciation by host switching to bullfrogs (bg; Rana catesbeiana). Haematoloechus floedae was introduced into Costa Rica with bullfrogs, where it expanded its host range to include local species of leopard frogs, members of the ancestral host group. Bullfrogs subsequently died out in Costa Rica, but H. floedae survives today due to ecological fitting (from Brooks et al., in press). Thick lines indicate episodes of cospeciation; thin lines indicate episodes of speciation by host switching.

becomes relatively more apparent, then the parasite will become associated with C at time T1. This manifestation of ecological fitting could explain seemingly rapid and virtually unconstrained evolution of novel specialized host associations. Finally, a parasite might have a hierarchy of host preferences, even though it is tracking the same resource (host rank order; Singer et al. 1971; Janz and Nylin 1998 and references therein). The hierarchy arises because the costs of accessing the resource might not be identical across all host species or even across individuals in the same species (Singer et al. 1992). Such costs will depend on many different factors, including concentration of the resource, host density, and difficulty in extracting the resource. Overall, parasites accessing a plesiomorphically (or, less often, homoplasiously) distributed resource are ‘‘faux generalists’’ (Brooks and McLennan 2002): specialists whose host range appears large, but who are in reality using the same resource. If a parasite species evolves the ability to utilize a novel resource, a second and more complicated type of host preference hierarchy can arise if the parasite also retains sufficient information to use the plesiomorphic resource (Wiklund 1981, Courtney et al. 1989). For example, Haematoloechus floedae is a fluke native to the southeastern United States where it lives in the lungs of the bullfrog, Rana catesbeiana. When bullfrogs were introduced to the southwestern United States, the

Yucata´n, and Costa Rica, the parasite went with them, and is now found in bullfrogs in those areas, as well as in leopard frogs in the Yucata´n and Costa Rica. Leopard frogs (Rana pipiens clade) are the plesiomorphic hosts for Haematoloechus (Fig. 1). Although the ancestor of H. floedae switched to bullfrogs, the presence of the fluke in leopard frogs indicates that the parasite has retained its clade’s plesiomorphic ability to infect leopard frogs (Brooks et al., in press). Interestingly, bullfrogs have disappeared from Costa Rica, but the parasite persists, having survived the ‘‘extinction’’ of its preferred host. This is the first demonstration that parasites, like phytophagous insects (Janz et al. 2001 and references therein) might display ancestral host preferences under certain circumstances. Ecological fitting is generally investigated in insect– plant systems, because researchers can reconstruct phylogenetic patterns of association between the two clades, then examine the processes underlying those patterns by (1) identifying the resource being tracked by the insect, (2) determining the distribution of that resource among host plants, and (3) delineating the host preference hierarchy of the insects (Brooks and McLennan 2002). Currently, we do not have this degree of detailed information for any host–parasite system. It is possible, however, to take advantage of ‘‘natural experiments’’ (e.g., the case of H. floedae), or even to make inferences based on contemporary patterns of

July 2006

ECOLOGICAL FITTING OF ANURAN PARASITES

host–parasite association, if hosts vary in their use of a habitat to which parasite species are constrained. The associations between anurans and their platyhelminth parasites provide a model system for such an investigation, because the majority of helminths require water for the development and transmission of infective stages, while most, but not all, major groups of anurans have a sexual and developmental tie to aquatic habitats. Brandt (1936) suggested that species richness in anuran parasite communities was directly related to the amount of time the host spent in or near water, an observation confirmed by subsequent studies (Prokopic and Krivanec 1975, Brooks 1976). A shared plesiomorphic requirement for an aquatic habitat, coupled with a gradient of adult anuran preferences ranging from aquatic to arboreal, suggests that ecological fitting as a determinant of the parasites associated with a given anuran taxon should be evidenced as a nested-subset structure (Patterson and Atmar 1986) of host–parasite associations across anuran taxa (Zelmer et al. 2004). At one extreme, if all the host–parasite associations are the result of ecological fitting, then all host taxa are interchangeable from the point of suitability for the parasites, and associations will be determined solely by the habitats the host utilizes and its feeding preferences. The shared requirement of tadpoles for aquatic habitats should thus provide a baseline assemblage of parasites that infect the tadpole stage, while the parasites of adult anurans should accumulate in anuran host species as a function of the time they spend in aquatic habitats as adults. If specialized coevolutionary processes dominate, sympatry between anurans and the infective parasitic stages will result in parasitism of only appropriate hosts, producing idiosyncratic (i.e., ‘‘unexpected’’) presences and absences in the matrix of host–parasite associations. MATERIALS

AND

METHODS

Compound parasite communities are defined as the array of parasite species inhabiting an array of host species in a given area (Holmes and Price 1986). We have data for six compound communities of platyhelminths that parasitize frogs as definitive hosts in North and Central America: the temperate hardwood forests of North Carolina (Brandt 1936), the temperate grasslands of Nebraska (Brooks 1976), and the tropical wet and dry forests of Costa Rica (see Plate 1) and Mexico, derived from biodiversity inventories currently being coordinated by D. R. Brooks (Costa Rica) and V. Leo´nRe`gagnon (Mexico) (see the Appendix). We sampled 75 anuran species in the six areas; 59 were sampled in one area, 14 species were sampled in two areas, and two species were sampled in three areas (see the Appendix, Table A1). Of the 57 platyhelminth species collected, 38 were found in one area, 13 species were found in two areas, four species were found in three areas, and two species (Langeronia macrocirra and Haematoloechus complexus) were found in four areas (see the Appendix, Table A2). The parasites inhabit 34

S79

of the 75 sampled anurans, only six of which (Rana catesbeiana, Rana vaillanti, Smilisca baudenii, Smilisca phaeota, Leptodactylus melanonotus, and Bufo marinus) have been sampled in two areas, and one of which (Bufo marinus) was sampled in three areas. From this we conclude that comparisons of compound community structure among the six sites will not be confounded by multiple samples of the same anuran community, and therefore the same anuran parasite (i.e., pseudoreplication). Moreover, given the geographical and taxonomic breadth of the surveys, it is assumed that the resultant presence/absence matrices of host–parasite associations, at the taxonomic levels examined (i.e., host genera and parasite genera and families) are representative of the possible associations, and not strongly biased by ecological factors, such as host and parasite ranges and relative abundances. Anuran species were ranked based on their association with aquatic habitats as follows: 7, riparian, prolonged breeding (several months); 6, semiaquatic, prolonged breeding; 5, terrestrial, prolonged breeding; 4, terrestrial, explosive breeding (1–2 wk); 3, arboreal, prolonged breeding; 2, arboreal, explosive breeding; 1, fossorial. The relationship between the ranked association and trematode species richness was evaluated using Spearman’s rank correlation analysis. Without data from experimental infections, ecological fitting and co-speciation cannot be distinguished as explanations for extant, and apparently specific, host– parasite associations. Thus, parasite species and host species were grouped by genera for the purpose of nested-subset analysis, increasing the likelihood that the host and parasite clades had at one time been sympatric. Given the degree of local adaptation for both the host and parasite species, pooling hosts by genera and parasites by genera and families should not increase the likelihood of a nested-subset pattern occurring, given a mechanism of co-speciation. Thus it is necessary to view such a pattern as having been produced by ecological fitting. Examination of the nested-subset structure of parasite genera within the pooled anuran genera across all six localities was conducted using the nestedness temperature calculator (Atmar and Patterson 1995), which calculates the temperature of the matrix (a measure of order, with lower temperatures indicating a higher degree of order) and idiosyncratic host and parasite temperatures, which indicate host species and parasite species contributing disproportionately to the lack of order in the matrix (Atmar and Patterson 1993). Nested-subset patterns can arise as artifacts of random draws of individual items from categories that vary in their representation (Connor and McCoy 1979). In a proximal sense, within a given locality, this would involve host individuals acquiring parasites from a species pool where the probabilities of infection varied among the parasite species because of an uneven distribution of infective stages within the environment. Considering the patterns of association between host

S80

DANIEL R. BROOKS ET AL.

and parasite taxa, assuming that the various host taxa are sympatric in a regional sense, nestedness could be expected to arise by a similar passive mechanism if the parasite taxa vary in the degree of sympatry between their respective geographic ranges and those of the hosts. For tests of passive sampling involving community data, the relative abundances of the species sampled is not known for the source pool, requiring estimation of these relative abundances from the available data. Models that test for passive sampling typically base this estimate on the occurrence of species in the sample (RANDOM1; Patterson and Atmar 1986, Fischer and Lindenmayer 2002), which will result in overestimation of the colonization probabilities of rare species unless none of the populations present in the sample were further supplemented by dispersal from the source pool following the initial colonization (Andre´n 1994). Constructing an appropriate null model for passive sampling would require knowledge of the contribution of immigration from the source pool to the observed relative abundances. In the absence of such information, a null model (RELABUND) defining the opposite extreme, i.e., each individual present in a population is assumed to be an immigrant from the source pool, can be used in concert with RANDOM1, with the appropriate, but unavailable, null model falling between these extremes (Zelmer et al. 2004). Given that the distributions of temperatures of matrices produced by these models represent extremes in terms of the effect of immigration on the observed population sizes within a community, overlap with the tails of these distributions cannot be evaluated with a simple decision rule and must be interpreted in light of ecological evidence for the expected effects of immigration. By analogy, the evaluation of passive mechanisms that produce nested-subset patterns of associations between host taxa and parasite taxa would require an understanding of the contribution of host capture to the observed associations. Species-level host and parasite phylogenies do not yet exist for the taxa in question (an exception is Haematoloechus; Leo´n-Re`gagnon and Brooks 2003), so the number of times a particular host genus acquired any particular parasite genus or family cannot be directly inferred, and must be estimated from the available presence/absence data. Analogous models to RANDOM1 and RELABUND were employed, using the occurrence of parasite taxa within host taxa to parameterize the Monte Carlo simulations for RANDOM1, and using the number of independent host– parasite associations to parameterize RELABUND. (For example, there are two species of Langeronia, one infecting four host species, the other infecting one. Thus, for the RELABUND model considering parasite genera, five ‘‘individuals’’ of Langeronia are distributed randomly among the host taxa. Within the Lecithodendriid family, in addition to the associations mentioned for Langeronia, there are two other parasite species, one infecting two host species, and one infecting a single

Ecology Special Issue

species. For the RELABUND model considering parasite families, eight ‘‘individual’’ lecithodendriids are distributed randomly among the host taxa.) As with the interpretation of these models for nestedness in communities, some of the associations observed will be the result of host capture, and some by inheritance, placing the appropriate, but unknown, null model between the extremes. Both the RANDOM1 and RELABUND null models were applied to presence/ absence matrices of platyhelminth genera within anuran genera, trematode genera within anuran genera, and matrices where parasite genera not represented in all three regions were pooled by family and evaluated within host genera and host ecotype ranking. To evaluate whether the patterns of presence and absence revealed across regions by the nested-subset analysis were reflected at smaller scales, we employed Spearman’s rank correlation analysis to assess covariance between the total number of host genera occupied by a parasite genus or family (pooled across all six localities), and the number of host species, genera, and families occupied by each species within that taxon within each of the six localities. We also used Pearson’s analysis to determine covariance between the total number of host genera occupied by a parasite genus or family (pooled across all six localities) and the number of host species, genera, and families occupied by those taxa within each region (United States, Mexico, and Costa Rica). RESULTS The ranked association with aquatic habitats of the anuran species with nine or more individuals necropsied per locality positively covaried (r ¼ 0.785; P , 0.0001) with the trematode richness of the frog host species (Fig. 2), with no clear differences in the pattern of increase among the six localities. The temperature (the measure of matrix order derived by Atmar and Patterson [1993]) of the presence/absence matrix of platyhelminth genera within anuran genera (Fig. 3) was significantly more ordered than the matrices produced by the RANDOM1 (P ¼ 0.00063) or RELABUND (P ¼ 0.00002) null models. Nested-subset analysis designated four of the 21 parasite taxa as idiosyncratic; two monogenean genera and two cestode genera. Such idiosyncrasies are an indication of different colonization histories for these genera (Atmar and Patterson 1993) relative to the other parasites considered, suggesting the importance of phylogenetic congruence as a determinant of the anuran associations with monogenean and cestode species. Consequently, the remaining analyses focused on the trematodes. The nested-subset structure of the trematode genera within the pooled anuran genera (Fig. 4) also was significantly colder than the matrices generated from both null models (RANDOM1, P ¼ 0.000001; RELABUND, P ¼ 0.0000004), and also revealed idiosyncratic parasite genera. These idiosyncrasies all occurred in

July 2006

ECOLOGICAL FITTING OF ANURAN PARASITES

S81

FIG. 2. Increasing trematode species richness in anuran host species from each of six localities, with increased anuran association (ranked) with aquatic habitats. Multiple observations at a single coordinate are indicated parenthetically above the coordinate. ACG denotes Area de Conservacio´n Guanacaste.

genera that were missing from at least one of the three regions sampled (North America, Mexico, or Costa Rica). Thus, in order to evaluate the potential for the interaction of the anuran host genera with specific landscapes to produce nested subset patterns, we arbitrarily pooled parasite genera not represented in all three areas into their respective families, thereby ensuring that all anuran genera included in the analysis have the potential to draw from the same parasite pool. The resulting matrix is depicted in Fig. 5. The temperature of the presence/absence matrices of trematode genera/families in pooled anuran genera overlaps the cold tail of the distribution produced by the RANDOM1 model (P ¼ 0.0025) and the cold tail of the distribution produced by the RELABUND model (P ¼ 0.0885). Thus, the observed nested-subset pattern could only be attributed to passive mechanisms by adhering to the RELABUND model’s assumption that all host–parasite associations occur independently. Given that extreme assumption, however, interpretations of the observed matrix as nonsignificant with regards to passive sampling must be made with caution. Examination of the presence/absence matrix of trematode genera and families within ranked anuran ecotypes (Fig. 6) supports the contention that the semiaquatic anuran habitat creates overlap with infective stages of a greater number of trematodes than a purely aquatic habit. The temperature of this matrix falls within the cold tail of the distribution of the temperatures produced by both models (RANDOM1, P ¼ 0.0018; RELABUND, P ¼ 0.165). As with the parasite associations with anuran genera, one must conclude that passive mechanisms could produce this pattern only under the independent-association assumption of the RELABUND model, again with

the caveat that the distribution of the appropriate null model presumably has a warmer central tendency than that produced by the RELABUND model. The total number of host genera occupied by a parasite genus or family (pooled across all six localities) positively covaried with the number of host species (r ¼ 0.321; P ¼ 0.0104), genera (r ¼ 0.425; P ¼ 0.0005), and families (r ¼ 0.426; P ¼ 0.0005) occupied by each species within that taxon within each of the six localities. The total number of host genera occupied by a parasite genus or family also positively covaried with the number of host species (r ¼ 0.492; P ¼ 0.0147), genera (r ¼ 0.668; P ¼ 0.0004), and families (r ¼ 0.645; P ¼ 0.0007) occupied by each species within that taxon within each of the three regions (United States, Mexico, and Costa Rica). DISCUSSION Parasite habitat preference and transmission patterns Fifty-four of the 57 parasite species (see the Appendix, Table A3) exhibit the plesiomorphic pattern of requiring water for transmission, either by utilizing aquatic intermediate hosts (digeneans and cestodes), or by swimming from one host to another (monogeneans). In other words, transmission patterns are phylogenetically conservative in this phylum (Brooks and McLennan 1993, Adamson and Caira 1994). This explains why 45 of the 57 platyhelminth species were found only in aquatic and semiaquatic frogs. Of the remaining 12 species, 10 occur in terrestrial, arboreal, and fossorial frogs, but infect the tadpole stage of their hosts. Digeneans in the genus Glypthelmins and the family Paramphistomidae cluster with the brachycoelids (Brachycoelium and Mesocoelium) in the maximally

S82

DANIEL R. BROOKS ET AL.

Ecology Special Issue

FIG. 3. Maximally packed presence/absence matrix for pooled parasite genera (columns) within pooled host genera (rows) from all six localities. Stars indicate idiosyncratic hosts and parasite species. Letters within the matrix denote geographic region where associations were observed: A, United States only; B, Mexico only; C, Costa Rica only; D, United States and Mexico; E, United States and Costa Rica; F, Mexico and Costa Rica; G, United States, Mexico, and Costa Rica.

packed matrix (Fig. 5) as parasite taxa common to a number of anuran genera within localities and regions, as well as across regions. All that is required for infection is that the frog species comes to water to breed in a population density high enough to ensure infection. This behavior is plesiomorphic for, and phylogenetically conservative among, frogs. The last two species, members of the sister groups Brachycoelium and Mesocoelium, have terrestrial life cycles, which explains why they occur so frequently in terrestrial anurans and in frogs that occasionally forage away from water.

Choledocystus intermedius (one of the idiosyncratic taxa in the presence/absence matrix depicted in Fig. 4) inhabits only Bufo marinus, and it is the only adult platyhelminth restricted to that host. Razo-Mendivil et al. (in press) recently have shown C. intermedius to be closely related to members of the families Ochetosomatidae and Telorchiidae. Life cycles for members of those families involve aquatic molluscs as first intermediate hosts, and tadpoles as second intermediate hosts, which are ingested by the final host. The absence of C. intermedius from other anuran hosts that ingest tadpoles might indicate that the

FIG. 4. Maximally packed presence/absence matrix for pooled trematode genera (columns) within pooled host genera (rows) from all six localities. Stars indicate idiosyncratic host and parasite species. Letters within the matrix are as in Fig. 3.

July 2006

ECOLOGICAL FITTING OF ANURAN PARASITES

FIG. 5. Maximally packed presence/absence matrix for trematodes (columns) pooled by genera, or by family for those genera that did not occur in all three regions (United States, Mexico, and Costa Rica), within pooled host genera (rows) from all six localities. Stars indicate idiosyncratic host and parasite species. Letters within the matrix are as in Fig. 3.

association between C. intermedius and Bufo marinus involves more specialization than ecological fitting. The remaining parasite taxa, whose associations with their hosts cannot easily be interpreted as manifestations of ecological fitting (based on the idiosyncratic patterns revealed in the nested subset analysis), also infect tadpoles at some stage in their lives. The monogeneans Polystoma naevius, and the probable sister species Neodiplorchis scaphiopi and Pseudiplorchis americanus, infect tadpoles and develop into adults when the tadpoles metamorphose. Anecdotal reports exist of tadpoles eating proglottids of nematotaeniid cestodes, suggesting that four additional species (Cylindrotaenia americana and C. sp., Distoichometra bufonis and D. kozloffi) gain infection in a manner similar to the first three species. Perhaps infection of a tadpole requires greater specificity on the part of the parasite than infections of adult anurans, making parasites with such life cycle patterns less amenable to ecological fitting.

S83

in plethodontid salamanders and has a terrestrial life cycle. Not surprisingly it is more common in a terrestrial nonranid (Pseudacris brimleyi) than in a ranid host. It is clear that a role for coevolutionary processes exists for variations in associations of parasite genera within families, and species within those genera in terms of their specific host association. However, no explanatory power is gained from considering anurans by their genera, as opposed to their ecotype, in terms of associations with the genera and families of the trematodes infecting them. This equivalence occurs, in part, because ecotype preference for anuran hosts (e.g., all aquatic and semiaquatic host species in all six sites are members of the same genus, Rana), and transmission dynamics for the parasites (e.g., all species of Haematoleochus utilize odonate naiads as second intermediate hosts) are phylogenetically conservative (Snyder and Janovy 1994, Wetzel and Esch 1996). The host landscape As evidence for phylogenetic conservatism in host and parasite biology, 80% of the parasite species discovered in these six communities inhabit only 13% of the frog species sampled. How do these species coexist? Part of the answer lies in perhaps the most fundamental element of ecological fitting: allopatry. Only 38% of the 48 parasite species inhabiting aquatic and semiaquatic anurans occur in more than one community. Another aspect of the process of co-ocurrence lies in parasite microhabitat diversification, or, as commonly

Habitat use by hosts Forty-two of the 58 species of adult platyhelminths (72%) occur in ranids. Of those 42 species, 27 are found only in ranids, indicating that some character or suite of characters associated with being a ranid is the resource being tracked by the parasites. Of the remaining 15 species, 11 always occur in higher prevalences in ranids (one measure of the host preference hierarchy), two occur at lower prevalences, and two are equivocal (possibly an artifact of small sample size). The lowprevalence occurrences in nonranid hosts might be an additional example of ecological fitting if the nonranids are suitable hosts, but their lack of exposure to aquatic habitats renders them ‘‘not apparent’’ to the parasites. Brachycoelium hospitale, for example, is generally found

FIG. 6. Maximally packed presence/absence matrix for trematodes (columns) pooled by genera, or by family for those genera that did not occur in all three regions (United States, Mexico, and Costa Rica), within host species from all six localities, pooled by ecotype (ranked association with aquatic habitats). Stars indicate idiosyncratic host and parasite species. Variation in rank within genera was (no. species at rank) as follows: ranids, two at rank 7, six at rank 6, and two at rank 5; bufonids, two at rank 4, three at rank 5; leptodactylids, two at rank 4, one at rank 5; hylids, two at rank 3; see Materials and methods for rank designations. No trematodes were found to infect the fossorial species.

S84

DANIEL R. BROOKS ET AL.

phrased, diversity in preferred site of infection within the host (Brooks and McLennan 1993, 2002, Adamson and Caira 1994, Radtke et al. 2002). Platyhelminth species occur in the buccal cavity, lungs, gall bladder and hepatic ducts, small intestine, rectum, and urinary bladder of anuran hosts, such that multiple species from different clades can infect the same host and form complicated communities without interacting physically, i.e., they are microallopatric. Many species having similar transmission modes occur in the same hosts, but live in different parts of that host. On the other hand, because the diversification of infection site is phylogenetically conservative, multiple, distantly related parasite species living within a given geographic area can exhibit the same kind of site specificity, which should amplify competition. In some cases, the parasite species occur in different host species; for example, polystome monogeneans and gorgoderid digeneans live in the urinary bladder, but do not infect the same species of frogs. In other cases, the parasites have markedly different biological requirements. Cestodes living in the host gut absorb nutrients from the host intestinal contents, whereas digeneans living in the host gut forage for host epithelial cells, cell and tissue exudates, and blood. CONCLUSIONS These six communities of frog parasites are both complex and similar to each other. Their complexity rules out simple phylogenetic replication, namely, that these communities are products of a simple history of vicariance and/or co-speciation. The taxonomic similarity of the communities, coupled with their occurrence in such markedly different environments, rules out the possibility that they are the result of convergent adaptation. Brooks (1980, 1985), Futuyma and Slatkin (1983), and Janzen (1985) suggested that relatively weak phenomena (in this case, phylogenetic conservatism in host and parasite natural history) have the potential to produce marked ecological structure. That a great deal of the stable and predictable structure of contemporaneous anuran parasite communities appears to be a result of phylogenetic conservatism in the evolution of both parasite and host biology, coupled with the historical biogeographic contingencies of speciation and dispersal, is consistent with those views. These observations, of course, do not rule out the possibility of ongoing strong evolutionary interactions between any of these parasites and their hosts or each other, particularly at the small spatial and short temporal scales associated with Thompson’s (1994) coevolutionary mosaic. Nor do the observations imply that ecological fitting explains everything; only that assumptions about the low probability of host switching must be viewed with far more caution in the future. Tracking a plesiomorphic resource in parasites is the equivalent of free-living organisms dispersing into new habitat, but retaining their ecological niche; both are aspects of ecological fitting. Given this, we expect that

Ecology Special Issue

parasite communities will be macroevolutionary mosaics of ecological fitting and co-speciation, just as are communities of free-living organisms (e.g., colubrid snakes; Cadle and Greene 1993). Additionally, because communities and associations are subject to evolutionary forces that will vary across space and time, we also expect that the importance of ecological fitting and co-speciation will vary among communities and among associations. Finally, our analysis implies that many parasites currently restricted to particular hosts in particular localities could survive in other hosts and other localities if they could get there. Episodes of major climate change, for example, result in range contractions and expansions bringing together species that have been allopatric during their previous evolutionary histories. In such cases, we would expect an increase in host switching, not as a result of evolution of novel host utilization capabilities, but as a manifestation of ecological fitting. As discussed above (see the Introduction), some parasites might well survive extinction of their original hosts. Discovering the importance of ecological fitting as a determinant of the structure of anuran parasite communities thus underscores the need for more comprehensive ecological and evolutionary understanding of host specificity in assessing the risk of parasite transmission into native hosts resulting from the introduction of exotic host species along with their parasites. ACKNOWLEDGMENTS We express our appreciation to Cam Webb for inviting us to contribute to this issue. D. R. Brooks thanks the scientific and technical staff of the Area de Conservacio´n Guanacaste for support of this study, in particular: Elda Araya, Roger Blanco, Carolina Cano, Maria Marta Chavarrı´ a, Felipe Chavarrı´ a, Roberto Espinoza, Dunia Garcia, Guillermo Jimenez, Elba Lopez, Sigifredo Marin, Alejandro Masis, Calixto Moraga, Fredy Quesada, and Petrona Rios. Thanks to Dan Janzen and Winnie Hallwachs, scientific advisers to the ACG, for their support and insights. D. R. Brooks and D. A. McLennan acknowledge support from the Natural Sciences and Engineering Research Council (NSERC) of Canada. V. Leo´n-Re`gagnon thanks Ma. Antonieta Arizmendi, Luis Garcı´ a, Rosario Mata, Laura Paredes, Elizabeth Martı´ nez, Agustı´ n Jime´nez, Rogelio Rosas, Ulises Razo (Instituto de Biologı´ a, Universidad Nacional Auto´noma de Me´xico [UNAM]), Edmundo Pe´rez, Adria´n Nieto, (Facultad de Ciencias, UNAM), David Lazcano, and Javier Banda (Universidad Auto´noma de Nuevo Leo´n) for their help in the field and lab, and acknowledges support from NSF grant DEB-0102383. LITERATURE CITED Adamson, M. L., and J. N. Caira. 1994. Evolutionary factors influencing the nature of parasite specificity. Parasitology 109:S85–S95. Andre´n, H. 1994. Can one use nested subset pattern to reject the random sample hypothesis? Examples from boreal bird communities. Oikos 70:489–491. Atmar, W., and B. D. Patterson. 1993. The measure of order and disorder in the distribution of species in fragmented habitat. Oecologia 96:373–382. Atmar, W., and B. D. Patterson. 1995. The nestedness temperature calculator: a Visual Basic program, including 294 presence–absence matrices. AICS Research, University

July 2006

ECOLOGICAL FITTING OF ANURAN PARASITES

Park, New Mexico, USA, and the Field Museum, Chicago, Illinois, USA. Brandt, B. B. 1936. Parasites of certain North Carolina Salientia. Ecological Monographs 6:491–532. Brooks, D. R. 1976. Parasites of amphibians of the Great Plains. II. Platyhelminths of amphibians in Nebraska. Bulletin of the University of Nebraska State Museum 10:65–92. Brooks, D. R. 1979. Testing the context and extent of host– parasite coevolution. Systematic Zoology 28:299–307. Brooks, D. R. 1980. Allopatric speciation and non-interactive parasite community structure. Systematic Zoology 29:192–203. Brooks, D. R. 1985. Historical ecology: a new approach to studying the evolution of ecological associations. Annals of the Missouri Botanical Garden 72:660–680. Brooks, D. R., and D. A. McLennan. 1993. Parascript: parasites and the language of evolution. Smithsonian Institution Press, Washington, D.C., USA. Brooks, D. R., and D. A. McLennan. 2002. The nature of diversity: an evolutionary voyage of discovery. University of Chicago Press, Chicago, Illinois, USA. Brooks, D. R., D. A. McLennan, V. Leo´n-Re`gagnon, and E. Hoberg. In press. Phylogeny, ecological fitting and lung flukes: helping solve the problem of emerging infectious diseases. [In English.] Revista Mexicana de Biodiverisidad 77. Cadle, J. E., and H. W. Greene. 1993. Phylogenetic patterns, biogeography, and the ecological structure of neotropical snake assemblages. Pages 281–293 in R. E. Ricklefs and D. Schluter, editors. Species diversity in ecological communities. University of Chicago Press, Chicago, Illinois, USA. Clayton, D. H., S. E. Bush, and K. P. Johnson. 2004. Ecology of congruence: Past meets present. Systematic Biology 53: 165–173. Connor, E. F., and E. D. McCoy. 1979. The statistics and biology of the species–area relationship. American Naturalist 113:791–833. Courtney, S. P. 1985. Apparency in coevolving relationships. Oikos 44:91–98. Courtney, S. P., G. K. Chen, and A. Gardner. 1989. A general model for individual host selection. Oikos 55:55–65. Darlington, P. J., Jr. 1957. Zoogeography: the geographical distribution of animals. John Wiley and Sons, New York, New York, USA. Feeny, P. P. 1976. Plant apparency and chemical defense. Pages 1–40 in J. Wallace and R. Mansell, editors. Recent advances in phytochemistry. Volume 10. Biochemical interactions between plants and insects. Plenum, New York, New York, USA. Fischer, J., and D. B. Lindenmighter. 2002. Treating the nestedness temperature calculator as a ‘‘black box’’ can lead to false conclusions. Oikos 99:193–199. Futuyma, D. J., and M. Slatkin, editors. 1983. Coevolution. Sinauer Associates, Sunderland, Massachusetts, USA. Hafner, M. S., and S. A. Nadler. 1988. Phylogenetic trees support the coevolution of parasites and their hosts. Nature 332:258–260. Hafner, M. S., and S. A. Nadler. 1990. Cospeciation in host– parasite assemblages: comparative analysis of rates of evolution and timing of cospeciation events. Systematic Zoology 39:192–204. Holmes, J. C., and P. W. Price. 1986. Communities of parasites. Pages 187–213 in J. Kikkawa and C. J. Anderson, editors. Community ecology: pattern and process. Blackwell Scientific, Melbourne, Australia. Janz, N., and S. Nylin. 1998. Butterflies and plants: a phylogenetic study. Evolution 52:486–502.

S85

Janz, N., S. Nylin, and K. Bybom. 2001. Evolutionary dynamics of host plant specialization: a case study. Evolution 55:783–796. Janzen, D. H. 1985. On ecological fitting. Oikos 45:308–310. Johnson, K. P., and D. H. Clayton. 2004. Untangling coevolutionary history. Systematic Biology 53:92–94. Leo´n-Re`gagnon, V., and D. R. Brooks. 2003. Molecular phylogeny of Haematoloechus Looss, 1899 (Digenea: Plagiorchiidae), with emphasis on North American species. Journal of Parasitology 89:1206–1211. Page, R. D. M., editor. 2003. Tangled trees. University of Chicago Press, Chicago, Illinois, USA. Patterson, B. D., and W. Atmar. 1986. Nested subsets and the structure of insular mammalian faunas and archipelagos. Biological Journal of the Linnean Society 28:65–82. Pellmyr, O. 1992a. Evolution of insect pollination and angiosperm diversification. Trends in Ecology and Evolution 7:46–49. Pellmyr, O. 1992b. The phylogeny of a mutualism: evolution and coadaptation between Trollius and its seed-parasitic pollinators. Biological Journal of the Linnean Society 47: 337–365. Prokopic, J., and K. Krivanec. 1975. Helminths of amphibians, their interaction and host–parasite relationships. Acta Scientiarum Naturalium Brno 9:1–48. Radtke, A., D. A. McLennan, and D. R. Brooks. 2002. Evolution of host specificity in Telorchis spp. (Digenea: Plagiorchiformes: Telorchiidae). Journal of Parasitology 88: 874–879. Razo-Mendivil, U. J., and V. Leo´n-Re`gagnon. and G. Pe´rezPonce de Leo´n. In press. Systematic position of Glypthelmins (Digenea) within the Plagiorchiida, based on molecular phylogenetic analysis using partial lsrDNA sequences. Organisms, Diversity and Evolution 6. Singer, M. C., P. R. Ehrlich, and L. E. Gilbert. 1971. Butterfly feeding on lycopsid. Science 172:1341–1342. Singer, M. C., D. Ng, and C. D. Thomas. 1992. Rapidly evolving associations among oviposition preferences fail to constrain evolution of insect diet. American Naturalist 139:9–20. Snyder, S. D., and J. J. Janovy, Jr. 1994. Second intermediate host-specificity of Haematoloechus complexus and Haematoloechus medioplexus (Digenea: Haematoloechidae). Journal of Parasitology 80:1052–1055. Thompson, J. N. 1994. The coevolutionary process. University of Chicago Press, Chicago, Illinois, USA. Wetzel, E. J., and G. W. Esch. 1996. Influence of odonate intermediate host ecology on the infection dynamics of Halipegus spp., Haematoloechus longiplexus, and Haematoloechus complexus (Trematoda: Digenea). Journal of the Helminthological Society of Washington 63:1–7. Wiklund, C. 1981. Generalist vs. specialist oviposition behaviour in Papilio machaon (Lepidoptera) and functional aspects on the hierarchy of oviposition preferences. Oikos 36:163–170. Wiklund, C. 1984. Egg-laying patterns in butterflies in relation to their phenology and the visual apparency and abundance of their host plants. Oecologia 63:23–29. Zelmer, D. A., and H. P. Arai. 2004. Development of nestedness: host biology as a community process in parasite infracommunities of yellow perch (Perca flavescens (Mitchill)) from Garner Lake, Alberta. Journal of Parasitology 90:435–436. Zelmer, D. A., L. Paredes-Caldero´n, V. Leo´n-Re`gagnon, and L. Garcı´ a-Prieto. 2004. Nestedness in colonization-dominated systems: helminth infracommunities of Rana vaillanti Brocchi (Anura:Ranidae) in Los Tuxtlas, Veracruz, Mexico. Journal of Parasitology 90:705–710.

APPENDIX Geographic distributions of anuran hosts and their platyhelminth parasites in six areas in North and Central America (Ecological Archives E087-112-A1).

Ecology, 87(7) Supplement, 2006, pp. S86–S99 Ó 2006 by the Ecological Society of America

THE PHYLOGENETIC STRUCTURE OF A NEOTROPICAL FOREST TREE COMMUNITY STEVEN W. KEMBEL1,3 1

AND

STEPHEN P. HUBBELL2

Department of Biological Sciences, University of Alberta, Edmonton, Alberta T6G2E9 Canada 2 Department of Plant Biology, University of Georgia, Athens, Georgia 30602 USA

Abstract. Numerous ecological and evolutionary processes are thought to play a role in maintaining the high plant species diversity of tropical forests. An understanding of the phylogenetic structure of an ecological community can provide insights into the relative importance of different processes structuring that community. The objectives of this study were to measure the phylogenetic structure of Neotropical forest tree communities in the Forest Dynamics Plot (FDP) on Barro Colorado Island, Panama, to determine how the phylogenetic structure of tree communities varied among spatial scales and habitats within the FDP, and to study the effects of null-model choice on estimates of community phylogenetic structure. We measured community phylogenetic structure for tree species occurring together in quadrats ranging in size from 10 3 10 m to 100 3 100 m in the FDP. We estimated phylogenetic structure by comparing observed phylogenetic distances among species to the distribution of phylogenetic distances for null communities generated using two different null models. A null model that did not maintain observed species occurrence frequencies tended to find nonrandom community phylogenetic structure, even for random data. Using a null model that maintained observed species frequencies in null communities, the average phylogenetic structure of tree communities in the FDP was close to random at all spatial scales examined, but more quadrats than expected contained species that were phylogenetically clustered or overdispersed, and phylogenetic structure varied among habitats. In young forests and plateau habitats, communities were phylogenetically clustered, meaning that trees were more closely related to their neighbors than expected, while communities in swamp and slope habitats were phylogenetically overdispersed, meaning that trees were more distantly related to their neighbors than expected. Phylogenetic clustering suggests the importance of environmental filtering of phylogenetically conserved traits in young forests and plateau habitats, but the phylogenetic overdispersion observed in other habitats has several possible explanations, including variation in the strength of ecological processes among habitats or the phylogenetic history of niches, traits, and habitat associations. Future studies will need to include information on species traits in order to explain the variation in phylogenetic structure among habitats in tropical forests. Key words: community phylogenetic structure; environmental filtering; Forest Dynamics Plot; null models; Panama; phylogenetic clustering; phylogenetic overdispersion.

INTRODUCTION Why are there so many species of trees in the tropics? Tropical forests are incredibly biologically diverse, and numerous ecological and evolutionary processes, such as niche differentiation, herbivory, dispersal, competition, parasitism, and disease, appear to interact to play a role in maintaining the high species diversity of tropical tree communities at a range of spatial scales (Wright 2002). Although all of these processes have been demonstrated to occur, their relative importance in structuring ecological communities is not well understood. Numerous studies have demonstrated niche differentiation and Manuscript received 21 January 2005; revised 6 September 2005; accepted 8 September 2005. Corresponding Editor (ad hoc): J. B. Losos. For reprints of this Special Issue, see footnote 1, p. S1. 3 E-mail: [email protected]

habitat specialization in tropical tree species (Hubbell and Foster 1983, Condit et al. 1996, Clark et al. 1998, Webb and Peart 2000, Harms et al. 2001), but the strength of habitat specialization is often not sufficient to explain observed levels of species richness in tropical forests (Webb and Peart 2000, Harms et al. 2001). Similarly, finding niche differentiation among species does not mean that niche differences are more important than species-neutral processes in structuring a community (Hubbell 2001). The phylogenetic structure of ecological communities should provide insights into the relative importance of different ecological processes, as these processes interact with the evolutionary history of plant traits and leave their signature on the phylogenetic structure of a community (Webb et al. 2002). Previous studies have identified two general types of processes that can interact with phylogenetic patterns of niche and trait evolution to give rise to nonrandom

S86

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE

phylogenetic community structure, namely, competitive exclusion and environmental filtering processes (Webb 2000, Webb et al. 2002, Cavender-Bares et al. 2004). Competitive exclusion and other negative densitydependent interactions among ecologically similar species can create nonrandom community phylogenetic structures. Many processes, such as herbivory (Novotny et al. 2002) and competition (Uriarte et al. 2004), can have negative density-dependent effects, not only on conspecific individuals, but on close phylogenetic relatives as well. Density-dependent processes that negatively affect close phylogenetic relatives could give rise to phylogenetic overdispersion of co-occurring species, meaning that co-occurring species are more distantly phylogenetically related than expected by chance. Direct competitive exclusion, as well as indirect interactions among relatives mediated by herbivores, parasites, or pathogens, could all give rise to the same pattern of community phylogenetic overdispersion (Webb et al. 2002). Environmental filters (Weiher and Keddy 1995, Webb et al. 2002, Cavender-Bares et al. 2004) or assembly rules (Weiher and Keddy 1999) restricting community membership to individuals possessing certain traits might also affect community phylogenetic structure. If environmental conditions in a habitat act as a filter that select for species in possession of certain traits, we would expect to find either phylogenetic clustering or overdispersion, depending on the evolutionary history of ecologically important traits and species niches (Cavender-Bares et al. 2004, Ackerly et al. 2006, CavenderBares et al. 2006, Silvertown et al. 2006). The phylogenetic structure of a community might also be random. If species-neutral interactions (Hubbell 2001) structure a community, if the strength of density-dependent and environmental filtering processes are balanced or weak, or if species niches or traits are phylogenetically random, local communities could exhibit phylogenetic structures indistinguishable from random. Given the variation of measures of community structure, such as taxonomic and functional diversity with spatial scale and extent (Levin 1992), and changes in the phylogenetic signal of niches and traits with spatial scale (Cavender-Bares et al. 2006, Silvertown et al. 2006), we would predict that the phylogenetic structure of a community will be highly dependent on the spatial scale and extent used to define the community. The choice of an appropriate null model to use when measuring the structure of ecological communities has been very contentious (Gotelli and Graves 1996, Gotelli 2004), since analyses of the same data set with different null models can lead to very different conclusions. Simulation studies have been used to directly assess the performance of different null models when measuring species co-occurrence patterns (e.g., Gotelli 2000), and a number of different null models have been used to measure community phylogenetic structure, but the effects of different null models on estimates of community phylogenetic structure have not been evaluated quantitatively.

S87

In this study, we used large data sets on the phylogenetic relationships and co-occurrence patterns of Neotropical forest tree species in communities at a range of local spatial scales (10 3 10 m to 100 3 100 m) to address several questions. First, we compared the performance of different null models used to study the phylogenetic structure of ecological communities. Second, to test the relative importance of different ecological processes in structuring these communities and to test whether community phylogenetic structure changes with spatial scale, we asked whether trees occurring together in communities at a range of local spatial scales are phylogenetically clustered or overdispersed. Third, we asked whether the phylogenetic structure of tree communities differs among habitats characterized by different environmental conditions. METHODS Ecological data We estimated the phylogenetic structure of a Neotropical forest tree community using data from the 50-ha Forest Dynamics Plot (FDP) on Barro Colorado Island in the Republic of Panama. The moist lowland forests in the 1000 3 500 m FDP receive ;2600 mm of rain each year, with a dry season during January–April, and mean annual temperatures of 278C (Dietrich et al. 1982). Within the FDP, a variety of habitats have been identified (Harms et al. 2001), including (in approximate order of decreasing water availability during the dry season) swamp, stream, slope, and upland plateaus. The majority of the FDP contains old-growth primary forests, although some relatively young secondary forest habitat is found within the plot (Harms et al. 2001). Within the 1000 3 500 m FDP, all tree and shrub stems
1 cm dbh have been mapped and identified to species in repeated censuses conducted since 1982 (Condit 1998). We measured the phylogenetic structure of the forests on Barro Colorado Island using data on occurrence of tree and shrub species in the FDP. Mapped tree locations from the 1982 census of the FDP were divided into square nonoverlapping quadrats of four different spatial scales (10 3 10 m, 20 3 20 m, 50 3 50 m, and 100 3 100 m). We defined communities at a given spatial scale to include species of all tree and shrub stems
1 cm dbh present together in individual quadrats at that scale. Although we refer to the communities in the FDP as tree communities throughout this paper for the sake of convenience, shrubs
1 cm dbh were also included in all analyses of community structure. Phylogenetic data We constructed a hypothesized phylogenetic tree for the 312 tree species occurring in the FDP using Phylomatic version R20031202 software (Webb and Donoghue 2005), a phylogenetic database and toolkit for the assembly of phylogenetic trees. The tree created by Phylomatic used information from numerous published molecular phylogenies to create a tree containing all of

S88

STEVEN W. KEMBEL AND STEPHEN P. HUBBELL

Ecology Special Issue

FIG. 1. Hypothesized phylogenetic relationships among woody plant species occurring in the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. Circles indicate nodes dated based on divergence dates reported by Wikstro¨m et al (2001). Undated nodes were spaced evenly between dated nodes to minimize variation in branch length.

the species in the FDP, based on Phylomatic reference tree R20031202 with the APG II (Angiosperm Phylogeny Group 2003) phylogenetic classification of flowering plant families (P. F. Stevens, available online)4 forming the backbone of the tree. In the absence of detailed information on phylogenetic relationships within many of the 55 families and 192 genera found in the FDP, we assumed most families and genera were monophyletic and polytomous when placing them on the tree. We assigned branch lengths to the phylogenetic tree using the BLADJ module of the Phylocom version 3.19 software package (Webb et al. 2004), creating a pseudochronogram with branch lengths based on clade ages reported by Wikstro¨m et al. (2001). Nodes in the phylogenetic tree for which age estimates were available were fixed at their estimated ages (Wikstro¨m et al. 2001), and all remaining branch lengths were set by spacing undated nodes in the tree evenly between dated nodes to minimize variance in branch lengths. Age estimates were available for 34 of the 122 internal nodes in the phylogeny. Although 57 of the 122 internal nodes were polytomous, the majority of these polytomies were within families and genera, with the backbone of the tree relatively well resolved and dated. The resulting phylogenetic tree (Fig. 1; also, see the Supplement) was used for all subsequent analyses of community phylogenetic structure. Community phylogenetic structure We calculated several measures of community phylogenetic structure for all quadrats at each spatial scale. 4

hhttp://www.mobot.org/MOBOT/research/APwebi

All analyses of community phylogenetic structure were conducted using the Phylocom version 3.19 software package (Webb et al. 2004). Three steps were taken to measure community phylogenetic structure in each quadrat. First, we calculated raw phylogenetic distances among species occurring together in each quadrat. Second, we created numerous randomly generated null communities corresponding to each quadrat and estimated raw phylogenetic distances among species occurring together in the null communities. Finally, we calculated measures of standardized effect size (Gotelli and Rohde 2002) of community phylogenetic structure for our raw measures of phylogenetic distance in each quadrat, by comparing observed phylogenetic distances to the distribution of phylogenetic distances for the randomly generated null communities. We calculated raw phylogenetic distances among species in quadrats in two ways, each of which captures a different aspect of the phylogenetic relatedness of cooccurring species (Webb 2000). The mean pairwise distance (MPD) was calculated as the mean phylogenetic distance among all pairwise combinations of species occurring together in each quadrat, and the mean nearest neighbor distance (MNND) was calculated as the mean phylogenetic distance to the nearest relative for all species occurring together in each quadrat (Webb 2000, Webb et al. 2002). Null models To determine whether the phylogenetic structure of local tree communities differed from the phylogenetic community structure expected by chance, we compared observed phylogenetic distances among species in each

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE

quadrat to the distribution of phylogenetic distances for randomly generated null communities (Gotelli and Graves 1996). To assess the effect of null-model choice on ability to measure community phylogenetic structure, we generated null communities in two ways. The first method maintained the total species richness of each quadrat, with species in each quadrat chosen equiprobably at random, without replacement from the pool of species present in the FDP. We refer to this null model as the ‘‘unconstrained’’ model, since the species richness of each quadrat remained the same in the null communities, but species occurrence frequencies in the null communities were not constrained to be equal to their actual occurrence frequency among quadrats in the FDP data set. This null model assumes that all species present in the FDP are equally able to colonize any quadrat. The second method maintained both the total species richness of each quadrat, as well as the occurrence frequency of each species, by randomly swapping species occurrences among all quadrats at a scale subject to the constraint that the species richness of each quadrat remain constant and that the relative frequency of all species occurrences in quadrats remain constant. We refer to this null model as the ‘‘constrained’’ model, since the occurrence frequencies of species in the null community were constrained to be equal to their actual frequency in quadrats at that spatial scale. This null model assumes that a species’ ability to colonize a quadrat is proportional to its frequency in the FDP. We generated the constrained null communities using the independent swap algorithm (Gotelli 2000, Gotelli and Entsminger 2003) by holding the row and column sums of the quadrat/species occurrence matrix constant while swapping species among quadrats using a checkerboard swap. The checkerboard swap searches the quadrat/species matrix for submatrices of the form (0,1)(1,0) or (1,0)(0,1) (where 1 and 0 represent species presence and absence, respectively, in two quadrats) and swaps species presences between quadrats when these checkerboard submatrices are found. This maintains species frequencies and quadrat species richnesses while randomizing patterns of species co-occurrence. Each null community was created by swapping subsequent matrices many times relative to the number of species presences in the quadrat/species matrix, creating serially independent randomized matrices. The first null community for each spatial scale was created by checkerboard swapping the original quadrat/species matrix for that scale 30 000 times, with each subsequent null community generated by checkerboard swapping the previous matrix 10 000 times. We recorded the mean and standard deviation of MPD and MNND among species in each quadrat for 1000 null communities generated using the constrained and unconstrained null models. We then calculated measures of standardized effect size (Gotelli and Rohde 2002) of the observed phylogenetic distances among

S89

species occurring in each quadrat, relative to the distribution of distances calculated for null communities in each quadrat. These effect size measures compared the observed phylogenetic distance in each quadrat to the distribution of phylogenetic distances in null communities corresponding to that quadrat, and can be used to test for phylogenetic clustering or overdispersion. We used two metrics of community phylogenetic structure similar to those first proposed by Webb (2000), but based on comparisons with different null models. The net relatedness index (NRI) of each quadrat (Webb 2000, Webb et al. 2002) was defined as [–(MPD – MPDnull)/SD(MPDnull)], where MPD is the mean pairwise phylogenetic distance among species in the quadrat, MPDnull is the mean MPD for that quadrat in 1000 null communities, and SD(MPDnull) is the standard deviation of MPD for that quadrat in 1000 null communities. The net relatedness index has been proposed as a measure of tree-wide phylogenetic clustering and overdispersion of species (Webb 2000). Positive NRI scores indicate that species occurring together in a quadrat are more closely phylogenetically related than expected by chance, generally due to tree-wide phylogenetic clustering of co-occurring species. Negative NRI scores indicate that co-occurring species are less phylogenetically related than expected by chance, generally due to tree-wide phylogenetic overdispersion of co-occurring species. The nearest taxon index (NTI) of each quadrat (Webb et al. 2002) was defined as [–(MNND – MNNDnull)/ SD(MNNDnull)], where MNND is the mean nearest neighbor phylogenetic distance among species in the quadrat, MNNDnull is the mean MNND for that quadrat in 1000 null communities, and SD(MNNDnull) is the standard deviation of MNND for that quadrat in 1000 null communities. The nearest taxon index has been proposed as a measure of terminal (branch tip) phylogenetic clustering of species on a phylogeny (Webb 2000). If species tend to occur together with other closely related species (e.g., with congeners or confamilials), NTI scores will generally be positive due to this terminal phylogenetic clustering of species toward the tips of the phylogenetic tree. If species tend not to occur together with other closely related species, NTI scores will be negative due to terminal phylogenetic overdispersion. Estimating plot-wide phylogenetic structure To test whether the average phylogenetic structure of local tree communities at a given spatial scale differed from random, we calculated the mean phylogenetic structure of all quadrats at each scale as the mean NRI and NTI of all quadrats at that scale. If the mean NRI or NTI for all quadrats at a given spatial scale differed from zero according to a one-sample t test, we could conclude that the tree communities at that scale were significantly phylogenetically clustered or overdispersed on average, since both NRI and NTI are standardized effect sizes whose expected values are zero for phylo-

S90

STEVEN W. KEMBEL AND STEPHEN P. HUBBELL

Ecology Special Issue

FIG. 2. Spatial patterns of (a) habitats (from Harms et al. [2001]), (b) net relatedness index (NRI), and (c) nearest taxon index (NTI) in 20 3 20 m quadrats within the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. The indices NRI and NTI are measures of community phylogenetic structure based on a constrained null model that shuffled species co-occurrence patterns in null communities while maintaining observed species occurrence frequencies and quadrat species richnesses (see Methods for description). Positive NRI and NTI values indicate phylogenetic clustering, and negative values indicate phylogenetic overdispersion of species occurring together in a quadrat.

genetically random communities, positive for phylogenetically clustered communities, and negative for phylogenetically overdispersed communities, with approximately 95% of NRI and NTI values expected to fall in the range of 2 to þ2 for random communities (Gotelli and Rohde 2002). We estimated phylogenetic

structure for each quadrat using the constrained and unconstrained null models. We also assessed the phylogenetic structure of each quadrat by comparing the observed MPD and MNND values in each quadrat to the distribution of these values in the null communities. Quadrats were considered to be

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE

S91

TABLE 1. Results of a randomization study to assess the ability of different null models to measure community phylogenetic structure for 50 randomly selected 20 3 20 m quadrats in the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. A) Mean pairwise phylogenetic distance (MPD) and net relatedness index (NRI) MPD (Ma) Randomization null model Constrained Unconstrained

Analysis null model constrained unconstrained constrained unconstrained

Randomization

Analysis

NRI

Mean

Mean

SD

Mean

Type I error rate

214.3 214.3 217.9 217.9

214.4 217.9 217.9 217.9

5.9 7.3 7.2 7.2

0.00 0.49 0.00 0.00

0.06 0.04 0.06 0.06

B) Mean nearest neighbor phylogenetic distance (MNND) and nearest taxon index (NTI) MNND (Ma) Randomization method Constrained Unconstrained

Analysis method constrained unconstrained constrained unconstrained

Randomization

Analysis

NTI

Mean

Mean

SD

Mean

Type I error rate

78.0 78.0 79.2 79.2

78.1 79.2 79.2 79.2

6.6 7.2 7.1 7.2

0.02 0.18 0.00 0.00

0.05 0.05 0.06 0.05

Notes: Species co-occurrences in these quadrats were first randomized using either the constrained or unconstrained null model (see Methods for description). Phylogenetic distances among co-occurring species (mean pairwise phylogenetic distance [MPD] and mean nearest neighbor phylogenetic distance [MNND]) were first calculated for the randomized data. Both MPD and MNND are reported in terms of millions of years (Ma). The distributions of phylogenetic distances in null communities were then calculated for 999 subsequent randomizations of the data, using the analysis null model, and used to calculate measure of community phylogenetic structure for each randomization (net relatedness index [NRI] and nearest taxon index [NTI]; see Methods for description). This process was repeated 500 times for each combination of randomization and analysis null models. The average community phylogenetic structure (mean NRI and NTI) and Type I error rate (proportion of randomizations indicating a significant phylogenetic structure [absolute value of NRI or NTI . 2, P , 0.05]) were calculated for each combination of randomization and analysis null models. Boldface type indicates mean NRI or NTI values that were significantly different from zero according to a one-sample t test (N ¼ 50 quadrats 3 500 runs, P , 0.05).

significantly phylogenetically overdispersed or clustered if they occurred in the lowest or highest 2.5% of the distribution of distances from the null communities, respectively (a ¼ 0.05). A one-tailed binomial test was then used to assess whether the numbers of quadrats with significantly overdispersed or clustered phylogenetic distances were greater than expected at each spatial scale. Differences in sample sizes among spatial scales were accounted for using bootstrap estimation (Manly 1997) of the mean and standard error of NRI and NTI at each spatial scale. At the largest spatial scale examined (100 3 100 m), the sample size was 50 quadrats. At smaller spatial scales (50 3 50 m, 20 3 20 m, 10 3 10 m), we estimated the mean and standard error of phylogenetic structure (NRI and NTI) for 4999 random draws without replacement of 50 quadrats. The bootstrap estimates of mean and standard errors of NRI and NTI were then used to calculate a bootstrap t statistic and P value for NRI and NTI at each spatial scale. This method allowed direct comparisons of the mean and standard error of NRI and NTI values among spatial scales, taking into account the original differences in sample size at each spatial scale. Because phylogenetic similarities among co-occurring species were positively spatially autocorrelated (Fig. 2), we also tested whether NRI and NTI values at each spatial scale differed from zero using generalized leastsquares models with simultaneous spatial autoregression (SAR) covariance structures in SþSpatialStats (Kaluzny

et al. 1998). These models calculate an estimate of the mean and standard error of the coefficients in the model (NRI or NTI values), taking into account the nonindependence of spatially adjacent samples (Cressie 1993). In all cases, adding a first-order spatial-neighbor autoregressive term to the model removed autocorrelation from the residuals and improved the fit of the model, relative to a nonspatial model, and so we report only the results of the spatial models. Null model comparisons To compare null models, we employed a method similar to that used by Gotelli (2000), whereby random communities are created by shuffling ‘‘real’’ data using null models in order to randomize patterns of species cooccurrence, and then the randomized data are analyzed using the same set of null models. We randomly chose 50 of the FDP’s 20 3 20 m quadrats to conduct our randomization study. We first randomized the original data using the unconstrained or constrained null models. This created a new set of samples with the same phylogenetic relationships among species and sample species richnesses as the original data, but for which species cooccurrences were completely randomized. We then calculated phylogenetic distances (MPD and MNND) and standardized effect sizes of community phylogenetic structure (NRI and NTI) for the randomized samples using 1000 runs of the unconstrained and constrained null models. This process was repeated 500 times for

S92

STEVEN W. KEMBEL AND STEPHEN P. HUBBELL

Ecology Special Issue

FIG. 3. Tree community phylogenetic structure in 20 3 20 m quadrats within the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. Observed phylogenetic distances among co-occurring species are presented for 1250 quadrats, along with phylogenetic distances (in millions of years, Ma) among species in corresponding null communities generated using a constrained and an unconstrained null model (see Methods for description).

each combination of randomization and analysis null models. We estimated the proportion of randomized samples for which each null model found a significantly nonrandom phylogenetic distance among species (number of times observed distances were in the top or bottom 2.5% of randomized distances, an estimate of the Type I error rate), as well as calculating the average degree of phylogenetic clustering or overdispersion (mean NRI and NTI) estimated by each null model. Null models that maintain species frequencies have been criticized for potentially including the effects of any

process that acts to determine the occurrence frequency of a species (Colwell and Winkler 1984). If this were occurring, we would expect some phylogenetic signal in the distribution of species occurrence frequencies. To determine whether the frequencies of species in the FDP exhibited a phylogenetic signal, we calculated phylogenetic distances among all pairs of species occurring in the entire FDP at each spatial scale examined. We then calculated the dissimilarity of species frequencies as the square root of the absolute difference in occurrence frequency rank for all pairs of species, and we tested the significance of the correlation among the resulting

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE

S93

FIG. 4. Relationship between measures of phylogenetic community structure calculated using two null models in 1250 20 3 20 m quadrats within the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. Net relatedness index (NRI) and nearest taxon index (NTI) are measures of community phylogenetic structure based on constrained and unconstrained null models (see Methods for description).

phylogenetic and frequency distance matrices using a Mantel test (Legendre and Legendre 1998). Habitat phylogenetic structure Based on Harms et al.’s (2001) classification of 20 3 20 m quadrats within the FDP into seven habitat types (high plateau, low plateau, mixed, slope, stream, swamp, and

young), we asked whether the phylogenetic structure of tree communities at this spatial scale differed among habitats within the FDP. We tested for overall differences in phylogenetic structure among habitats using generalized least-squares models with simultaneous spatial autoregression (SAR) covariance structures, as described for the plot-wide tests. We estimated the

FIG. 5. Relationship between phylogenetic distances among species and similarity of species occurrence frequency ranks in 20 3 20 m quadrats within the 50-ha Forest Dynamics Plot (FDP) on Barro Colorado Island, Panama. Data points represent all pairwise combinations of species present in the FDP. Frequency rank differences were calculated as the absolute difference in the frequency occurrence ranks of each species pair. Phylogenetic distances were calculated as the branch length (in millions of years, Ma) connecting each species pair. Mantel tests indicated that relationships between square-root transformed frequency rank difference and phylogenetic distance were not statistically significant at any spatial scale examined.

S94

STEVEN W. KEMBEL AND STEPHEN P. HUBBELL

Ecology Special Issue

TABLE 2. Tree community phylogenetic structure in quadrats at four spatial scales within the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. Net relatedness index

Nearest taxon index

Estimated mean

SE

P

Estimated mean

SE

P

Unconstrained null model 10 3 10 m 5000 20 3 20 m 1250 50 3 50 m 200 100 3 100 m 50

0.260 0.389 0.771 0.074

0.008 0.016 0.047 0.099

0.0001 0.0001 0.0001 0.4605

0.142 0.154 0.453 0.019

0.012 0.021 0.045 0.099

0.0001 0.0001 0.0001 0.8459

Constrained null model 10 3 10 m 5000 20 3 20 m 1250 50 3 50 m 200 100 3 100 m 50

0.061 0.056 0.051 0.043

0.016 0.033 0.093 0.180

0.0001 0.0952 0.5825 0.8110

0.070 0.020 0.046 0.003

0.014 0.028 0.076 0.161

0.0001 0.4854 0.5427 0.9858

Spatial scale

N

Notes: Net relatedness index (NRI) and nearest taxon index (NTI) are measures of community phylogenetic structure based on constrained and unconstrained null models (see Methods for description). Positive NRI and NTI values indicate phylogenetic clustering; negative values indicate phylogenetic overdispersion of species occurring together in a quadrat. Parameter estimates at each scale are based on a spatial generalized least-squares model with a first-order spatial-neighbor simultaneous spatial autoregression term (SAR). Significant P values indicate that the phylogenetic structure at a given spatial scale differed from zero according to a two-tailed t test.

overall significance of differences in phylogenetic structure among habitats using NRI and NTI scores based on the constrained null model, as well as estimating the mean and standard errors of NRI and NTI scores within each habitat type to test for significant phylogenetic clustering or overdispersion. In all cases, adding a firstorder spatial-neighbor autoregressive term to the model removed autocorrelation from the residuals and improved the fit of the model relative to a nonspatial model, and so we report only the results of the spatial models. RESULTS Null model comparisons When analyzing community data generated by randomizing 50 randomly selected 20 3 20 m quadrats from the Forest Dynamics Plot with the unconstrained null model, both null models performed well (Table 1), with appropriate Type I error rates and mean NRI and NTI values of approximately zero. When analyzing data generated by randomizing these quadrats with the constrained null model, the unconstrained null model concluded that the randomized data were significantly phylogenetically clustered or overdispersed (mean NRI and NTI different from zero), although the Type I error rates of both null models remained correct (Table 1). A comparison of the phylogenetic distances among co-occurring species in all 1250 20 3 20 m quadrats in the FDP with the corresponding mean pairwise phylogenetic distances in null communities (Fig. 3) showed that phylogenetic distances in the null communities were much less variable than the observed distances, especially the mean pairwise distances. Mean phylogenetic distances among co-occurring species in the unconstrained null communities were higher than the mean distances in the observed and constrained null communities. As a result of the higher mean phylogenetic distances in the unconstrained null communities, NRI

and NTI values calculated for each quadrat using the different null models were tightly correlated (Fig. 4), but NRI and NTI values calculated using the unconstrained null model tended to be higher than those calculated using the unconstrained null model. We found no statistically significant relationships between phylogenetic distances among species and the square root of differences in species frequency ranks at any spatial scale (Mantel tests, P . 0.5 at all scales), although there was a slight but nonsignificant trend of the most closely related species pairs having similar frequency ranks (Fig. 5). Community phylogenetic structure The phylogenetic structure of tree communities in the FDP was highly variable among quadrats and dependent on the choice of null model. Using the unconstrained null model, tree communities in the FDP were phylogenetically clustered on average (mean NRI and NTI . 0; Table 2, Fig. 6) at spatial scales from 10 3 10 m to 50 3 50 m. Although the mean NRI and NTI were greater than zero at most scales examined, relatively few quadrats were significantly phylogenetically clustered or overdispersed according to the unconstrained null model (Table 3, Fig. 6). According to the constrained null model, mean community phylogenetic structure (NRI and NTI) across the entire plot did not differ from zero, except at the smallest spatial scale examined (10 3 10 m), where the mean NRI was greater than zero, indicating a slight overall trend of phylogenetic clustering at this scale. The magnitude of this effect was very small, and, after accounting for differences in sample size among spatial scales using bootstrap resampling (Table 4), the average phylogenetic structure (NRI and NTI) of tree communities did not differ from zero at all spatial scales examined according to the constrained null model. However, there was substantial variation in phylogenetic

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE

S95

FIG. 6. Tree community phylogenetic structure in quadrats at four spatial scales within the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. See Fig. 4 legend for definition and description of (a) NRI and (b) NTI. Dashed lines indicate expected 95% CI.

structure around the mean at all spatial scales examined (Fig. 6), and more quadrats than expected contained communities that were significantly phylogenetically overdispersed or clustered (Table 1). In the 20 3 20 m quadrats (Fig. 2), the phylogenetic structure of a quadrat was positively correlated with the phylogenetic structures of that quadrat’s first-order spatial neighbors (NRI, Moran’s I ¼ 0.34, P , 0.01; NTI, Moran’s I ¼ 0.14, P , 0.01), whereas the residuals of the spatial autoregression models for these values were not significantly spatially correlated (NRI residuals, Moran’s I ¼ 0.05, P ¼ 0.93; NTI residuals, Moran’s I ¼ 0.02, P ¼ 0.80). Similar patterns of signi-

ficant positive spatial autocorrelation of NRI and NTI values and nonsignificant spatial autocorrelation of residuals from the spatial models were observed at all spatial scales examined. Habitat influences on community phylogenetic structure Based on the constrained null model, the phylogenetic structure of the tree communities in the FDP differed among habitats at the spatial scale (20 3 20 m) for which habitat data were available (Fig. 1, Table 5). Species occurring together in the high plateau, low plateau, and young habitats tended to be significantly phylogenetically clustered (NRI . 0 or NTI . 0, P , 0.05), while

S96

STEVEN W. KEMBEL AND STEPHEN P. HUBBELL

Ecology Special Issue

TABLE 3. Number of quadrats with statistically significant phylogenetic clustering or overdispersion at four spatial scales in the 50ha Forest Dynamics Plot on Barro Colorado Island, Panama. MPD

Spatial scale

Total quadrats N

Unconstrained null model 10 3 10 m 5000 20 3 20 m 1250 50 3 50 m 200 100 3 100 m 50 Constrained null model 10 3 10 m 5000 20 3 20 m 1250 50 3 50 m 200 100 3 100 m 50

Overdispersed quadrats

MNND Clustered quadrats

N

Binomial P

69 36 17 0

1.0000 0.2171 ,0.001 1.0000

4 1 1 3

188 76 18 3

,0.001 ,0.001 ,0.001 0.1294

204 63 6 1

N

Binomial P

Overdispersed quadrats

Clustered quadrats

N

Binomial P

N

Binomial P

1.0000 1.0000 0.9937 0.1294

92 10 1 0

0.9992 1.0000 0.9937 1.0000

17 0 0 1

1.0000 1.0000 1.0000 0.7180

,0.001 ,0.001 0.3840 0.7180

150 30 4 4

0.0151 0.6144 0.7385 0.0362

98 40 6 2

0.9950 0.0716 0.3840 0.3565

Notes: Phylogenetic distances among co-occurring species in each quadrat (mean pairwise phylogenetic distance [MPD] and mean nearest neighbor phylogenetic distance [MNND]) were compared with phylogenetic distances in 1000 null communities generated using the constrained and unconstrained null models (see Methods for description). Quadrats were considered to be significantly phylogenetically overdispersed or clustered if they occurred in the lowest or highest 2.5% of the distribution of distances from the null communities, respectively (a ¼ 0.05). A one-tailed binomial test was then used to assess whether the numbers of quadrats with significantly overdispersed or clustered phylogenetic distances were greater than expected at each spatial scale.

communities in the swamp habitat tended to contain species that were phylogenetically overdispersed (NRI ¼ 1.096, P , 0.0001). Communities in the slope habitat exhibited both tree-wide phylogenetic overdispersion (NRI ¼ 0.286, P ¼ 0.0005) and marginally significant terminal phylogenetic clustering (NTI ¼ 0.130, P ¼ 0.0589). DISCUSSION Null model choice and community phylogenetic structure Measures of phylogenetic distances among species in a community must be compared with the phylogenetic distances generated by a null model in order to determine whether a community is more phylogenetically clustered or overdispersed than expected by chance. A potential problem with analyses of the phylogenetic structure of a community is that null models can simultaneously affect not only the co-occurrence patterns of

species, but also their occurrence frequencies across samples and the distribution of frequencies on the phylogeny. Randomization tests showed that the unconstrained null model can indicate nonrandom community phylogenetic structure (mean NRI and NTI different from zero) when used with data containing nonuniform species frequencies, even when patterns of species cooccurrence in samples are completely random (Table 1). A similar pattern was found in the tree communities within the FDP (Fig. 3). In the unconstrained null communities, mean pairwise phylogenetic distance converged on the average pairwise phylogenetic distance among all species occurring in the FDP, with every species and its associated phylogenetic distance to other species given equal weighting. In the constrained null communities, species and associated phylogenetic branches were effectively weighted by their frequency

TABLE 4. Tree community phylogenetic structure in quadrats at four spatial scales within the 50-ha Forest Dynamics Plot on Barro Colorado Island, Panama. Net relatedness index

Nearest taxon index

Estimated mean

SE

P

Estimated mean

SE

P

Unconstrained null model 10 3 10 m 50 20 3 20 m 50 50 3 50 m 50 100 3 100 m 50

0.261 0.388 0.771 0.074

0.077 0.079 0.094 0.099

0.0013 ,0.0001 ,0.0001 0.4583

0.143 0.156 0.452 0.019

0.118 0.103 0.090 0.099

0.2327 0.1347 ,0.0001 0.8455

Constrained null model 10 3 10 m 50 20 3 20 m 50 50 3 50 m 50 100 3 100 m 50

0.058 0.053 0.047 0.043

0.156 0.157 0.186 0.180

0.7128 0.7389 0.8028 0.8122

0.070 0.021 0.047 0.003

0.140 0.139 0.151 0.161

0.6201 0.8786 0.7546 0.9852

Spatial scale

N

Notes: Parameter estimates at each scale are based on 4999 bootstrap resamples of 50 quadrats from that spatial scale, except at the 100 3 100 m scale where parameters represent the actual parameter estimates for the 50 quadrats at that scale. Significant P values indicate that the phylogenetic structure at a given spatial scale differed from zero according to a two-tailed t test. See Table 2 notes for additional definitions and explanations.

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE

S97

TABLE 5. Tree community phylogenetic structure in 20 3 20 m quadrats in different habitats within the 50-h Forest Dynamics Plot on Barro Colorado Island, Panama. Net relatedness index

Nearest taxon index

Habitat

N

Mean

SE

P

Mean

SE

P

High Plateau Low Plateau Mixed Slope Stream Swamp Young

170 620 66 284 32 30 48

0.338 0.117 0.140 0.286 0.198 1.096 0.570

0.117 0.060 0.146 0.827 0.219 0.244 0.206

0.0039 0.0533 0.3393 0.0005 0.3666 0.0001 0.0057

0.021 0.054 0.141 0.130 0.315 0.116 0.440

0.094 0.049 0.129 0.069 0.191 0.206 0.169

0.8216 0.2619 0.2723 0.0589 0.0990 0.5720 0.0095

Notes: Net relatedness index (NRI) and nearest taxon index (NTI) are measures of community phylogenetic structure based on a constrained null model that shuffled species co-occurrence patterns in null communities while maintaining observed species occurrence frequencies and quadrat species richnesses (see Methods for description). Parameter estimates in each habitat are based on a spatial generalized least-squares (GLS) model with a first-order spatial neighbor simultaneous spatial autoregression term (SAR). Overall differences among habitats in NRI and NTI were statistically significant according to the spatial GLS tests (P , 0.0001). Significant P values in the table indicate that the phylogenetic structure in a habitat differed from zero (significant phylogenetic clustering or overdispersion) according to a two-tailed t test. See Table 2 notes for additional definitions and explanations.

within the plot, leading to similar mean phylogenetic distances in the observed and constrained null communities. Our results highlight the sensitivity of measures of phylogenetic community structure, not only to patterns of species co-occurrence, but also to phylogenetic tree topology, branch lengths, and species frequencies. By giving species such as the tree fern Cnemidaria petiolaria (which is both rare and distantly related to all other species in the FDP) an equal chance of occurring in every null community, phylogenetic distances among species in the unconstrained null communities were inflated relative to the observed communities, causing mean NRI and NTI values to be greater than zero on average. Conversely, by maintaining species frequencies in null communities, the constrained null communities tended to have the same mean phylogenetic structure as the observed communities, leading to a distribution of NRI and NTI values whose mean was close to zero. When using the unconstrained null model with the FDP data, it is difficult to attribute any differences in phylogenetic structure between the observed and null communities to the impact of ecological processes on the phylogenetic structure of the community, since they might simply be due to differences in species frequencies between the observed and null communities. Additionally, given the widespread spatial aggregation of trees (Condit et al. 2000) and strong dispersal limitation (Hubbell et al. 1999) demonstrated to occur in the FDP and other ecological communities, the assumption that all species are equally able to colonize any sample in the null communities is not realistic. Much of the previous debate surrounding the use of null models in ecology has focused on the relative merits of null models that maintain or do not maintain species frequencies (Gotelli and Graves 1996). Null models that do not maintain species frequencies have been criticized as being overly statistically liberal; this shortcoming has been described previously as the ‘‘Jack Horner effect’’

(Wilson 1995). Conversely, null models that maintain species frequencies have been criticized for potentially ‘‘smuggling in’’ the effects of processes such as competition or environmental filtering on species frequencies and community phylogenetic structure (the ‘‘Narcissus effect’’ [Colwell and Winkler 1984]), and for potentially being too statistically conservative. If this were the case, we might expect some relationship between species’ frequencies and their phylogenetic relatedness, but relationships between frequency similarity and phylogenetic distance were not statistically significant at any spatial scale examined according to Mantel tests (Fig. 2). We might also expect the constrained null model to find nonrandom community structure in fewer quadrats within the FDP if it were too statistically conservative, but in fact the constrained model found many more quadrats to be phylogenetically nonrandom compared to the unconstrained null model (Table 2), although mean NRI and NTI values from the constrained null model were always close to zero. Recent attempts to resolve the null model debate have focused on the use of simulation studies to quantify the Type I and II error rates of different null models when confronted with randomized or simulated community co-occurrence data (Gotelli 2000). Although we quantified the Type I error rate and bias of null models in this study, we did not assess the Type II error rate, which will need to be measured using simulation studies with data generated using different models of the interaction among phylogenetic relationships, trait evolution, and community structure. Several additional issues related to the use of null models to measure community phylogenetic structure remain unresolved. The constrained null model used in this study maintains observed species frequencies in null communities, but swaps species presences among samples, and thus can only be used with species presence/ absence data matrices. Species abundances within samples contain useful information that is discarded

S98

STEVEN W. KEMBEL AND STEPHEN P. HUBBELL

when using this approach, but it is probably not possible to simultaneously constrain species frequencies, abundances, and sample species richnesses. The constrained null model also ignores species in the regional species pool when generating null communities, since only species present in local communities are used to generate the null communities. To determine how community phylogenetic structure varies across a larger range of spatial scales (Webb et al. 2002, CavenderBares et al. 2006), it will be necessary to compare the phylogenetic relatedness of species present in communities at one scale to those present in some regional species pool. It is not clear how to separate the effects of variation in species frequency from the effects of ecological and evolutionary processes on community phylogenetic structure for these types of data, but clearly more research on the effects of null-model and species pool choice is needed. Community phylogenetic structure in the FDP Although the average phylogenetic structure of the tree communities in the FDP was close to random on average across the entire plot at spatial scales from 10 3 10 m to 100 3 100 m according to the constrained null model (Tables 2 and 4), the phylogenetic relatedness of tree species occurring together in individual quadrats varied greatly from phylogenetic overdispersion to phylogenetic clustering (Table 2, Fig. 6), and more quadrats than expected exhibited significant phylogenetic clustering and overdispersion at most spatial scales (Table 1, Fig. 6). Habitats within the plot differed strongly in their phylogenetic structure (Table 5, Fig. 2). At the spatial scale for which habitat data were available, phylogenetic clustering in some habitats combined with phylogenetic overdispersion in other habitats appeared to result in a pattern of phylogenetic structure indistinguishable from random on average across the entire FDP, obscuring the strong differences in phylogenetic structure among habitats within the FDP. In the seasonally dry plateau habitats and in the young secondary growth forests within the plot, cooccurring species were phylogenetically clustered (Table 5). The tree-wide phylogenetic clustering in the plateau habitats was largely due to the co-occurrence of numerous species from large, highly speciose clades such as the eurosids. Numerous species from these clades tended to occur together in plateau quadrats, leading to a trend of tree-wide phylogenetic clustering of co-occurring species in the plateau. Similarly, in the young habitats, species from families and genera concentrated in a few orders commonly occurred together in most quadrats, leading to a pattern of both tree-wide and terminal phylogenetic clustering. Given the broad phylogenetic niche and trait conservatism documented across plants in general (Prinzing et al. 2001, Cavender-Bares et al. 2006, Silvertown et al. 2006) and tropical trees in particular (Chazdon et al. 2003), the phylogenetic clustering in the plateau and

Ecology Special Issue

young forests in the FDP suggests that some sort of environmental filter is structuring tree communities in these habitats. Habitats such as the dry plateaus within the FDP would be predicted to contain phylogenetically clustered species, since the relatively stressful dry-season soil moisture conditions in these environments would be more likely to act as an environmental filter on broadly conserved ecological traits than in moister habitats (Harms et al. 2001, Webb et al. 2002). Communities in the relatively moist slope and swamp habitats were phylogenetically overdispersed (Table 5). Tree-wide phylogenetic overdispersion in the swamp habitats was caused by the presence of species from families such as the Moraceae and Arecaceae (Harms et al. 2001) that are widely scattered across the angiosperm phylogeny. In the slope habitats, a pattern of tree-wide overdispersion, but terminal phylogenetic clustering, was found. Several processes could give rise to these patterns of phylogenetic overdispersion, depending on the interaction between the phylogenetic history of trait evolution and contemporary ecological interactions in these habitats. Even within a single habitat type, the hierarchical nature of trait and niche evolution (Ackerly et al. 2006, Silvertown et al. 2006), interactions among multiple ecological processes (Webb et al. 2002), and the phylogenetic history of species habitat associations (Brooks and McLennan 1991) could lead to complicated patterns of community phylogenetic structure, making it difficult to attribute these patterns to any one process. More data on the ecological traits of species in the FDP and the evolutionary history of habitat associations in tropical trees will be necessary to determine the relative importance of processes such as environmental filtering or competitive exclusion in different habitats within the FDP. However, it is clear that either the relative importance of nonneutral ecological processes or the evolutionary history of niches, traits, or habitat associations must vary along environmental gradients within the FDP to explain the observed variation in phylogenetic structure among habitats. ACKNOWLEDGMENTS Thanks to Cam Webb for inviting this contribution and for suggesting the phylogenetic analysis of the Forest Dynamics Plot (FDP) data. Thanks to Kyle Harms for providing the habitat data, and to Rick Condit for making the FDP data available. Thanks to J. F. Cahill and M. R. T. Dale for statistical advice. This manuscript benefited greatly from comments by C. Webb and three anonymous reviewers. S. W. Kembel acknowledges support from the Natural Sciences and Engineering Research Council of Canada. The Forest Dynamics Plot of Barro Colorado Island has been made possible through the generous support of the U.S. National Science Foundation, The John D. and Catherine T. MacArthur Foundation, and the Smithsonian Tropical Research Institute and through the hard work of over 100 people from 10 countries over the past two decades. The BCI Forest Dynamics Plot is part of the Center for Tropical Forest Science, a global network of large-scale demographic tree plots.

July 2006

FOREST COMMUNITY PHYLOGENETIC STRUCTURE LITERATURE CITED

Ackerly, D. D., D. W. Schwilk, and C. O. Webb. 2006. Niche evolution and adaptive radiation: testing the order of trait divergence. Ecology 87:S50–S61. Angiosperm Phylogeny Group. 2003. An update of the Angiosperm Phylogeny Group classification for the orders and families of flowering plants: APG II. Botanical Journal of the Linnean Society 141:399–436. Brooks, D. R., and D. A. McLennan. 1991. Phylogeny, ecology, and behavior: a research program in comparative biology. University of Chicago Press, Chicago, Illinois, USA. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Cavender-Bares, J., A. Keen, and B. Miles. 2006. Phylogenetic structure of Floridian plant communities depends on taxonomic and spatial scale. Ecology 87:S109–S122. Chazdon, R. L., C. Careaga, C. Webb, and O. Vargas. 2003. Community and phylogenetic structure of reproductive traits of woody species in wet tropical forests. Ecological Monographs 73:331–348. Clark, D. B., D. A. Clark, and J. M. Read. 1998. Edaphic variation and the mesoscale distribution of tree species in a neotropical rain forest. Journal of Ecology 86:101–112. Colwell, R. K., and D. W. Winkler. 1984. A null model for null models in biogeography. Pages 344–359 in D. R. Strong, D. Simberloff, L. G. Abele, and A. B. Thistle, editors. Ecological communities: conceptual issues and the evidence. Princeton University Press, Princeton, New Jersey, USA. Condit, R. 1998. Tropical forest census plots. Springer-Verlag, Berlin, Germany. Condit, R., et al. 2000. Spatial patterns in the distribution of tropical tree species. Science 288:1414–1418. Condit, R., S. P. Hubbell, and R. B. Foster. 1996. Changes in tree species abundance in a Neotropical forest: impact of climate change. Journal of Tropical Ecology 12:231–256. Cressie, N. A. 1993. Statistics for spatial data. Revised edition. John Wiley and Sons, New York, New York, USA. Dietrich, W. E., D. M. Windsor, and T. Dunne. 1982. Geology, climate, and hydrology of Barro Colorado Island. Pages 21– 46 in E. G. Leigh, A. S. Rand, and D. M. Windsor, editors. The ecology of a tropical forest. Smithsonian Institution Press, Washington, D.C., USA. Gotelli, N. J. 2000. Null model analysis of species co-occurrence patterns. Ecology 81:2606–2621. Gotelli, N. J. 2004. Research frontiers in null model analysis. Global Ecology and Biogeography 10:337–343. Gotelli, N. J., and G. L. Entsminger. 2003. Swap algorithms in null model analysis. Ecology 84:532–535. Gotelli, N. J., and G. R. Graves. 1996. Null models in ecology. Smithsonian Institution Press, Washington, D.C., USA. Gotelli, N. J., and K. Rohde. 2002. Co-occurrence of ectoparasites of marine fishes: a null model analysis. Ecology Letters 5:86–94. Harms, K. E., R. Condit, S. P. Hubbell, and R. B. Foster. 2001. Habitat associations of tree and shrubs in a 50-ha Neotropical forest plot. Journal of Ecology 89:947–959. Hubbell, S. P. 2001. The unified neutral theory of biodiversity and biogeography. Princeton University Press, Princeton, New Jersey, USA. Hubbell, S. P., and R. B. Foster. 1983. Diversity of canopy trees in a Neotropical forest and implications for conservation.

S99

Pages 25–41 in S. L. Sutton, T. C. Whitmore, and A. C. Chadwick, editors. Tropical rain forest: ecology and management. Blackwell Scientific, Oxford, UK. Hubbell, S. P., R. B. Foster, S. T. O’Brien, K. E. Harms, R. Condit, B. Wechsler, S. J. Wright, and S. Loo de Lao. 1999. Light-gap disturbances, recruitment limitation, and tree diversity in a Neotropical forest. Science 283:554–557. Kaluzny, S. P., S. C. Vega, T. P. Cardoso, and A. A. Shelly. 1998. SþSpatialStats user’s manual for Windows and Unix. Springer-Verlag, New York, New York, USA. Legendre, P., and L. Legendre. 1998. Numerical ecology. Second English edition. Elsevier, New York, New York, USA. Levin, S. A. 1992. The problem of pattern and scale in ecology. Ecology 73:1943–1967. Manly, B. F. J. 1997. Randomization, bootstrap and Monte Carlo methods in biology. Second edition. Chapman and Hall, London, UK. Novotny, V., Y. Basset, S. E. Miller, G. D. Weiblen, B. Bremer, L. Cizek, and P. Drozd. 2002. Low host specificity of herbivorous insects in a tropical forest. Nature 416:841–844. Prinzing, A., W. Durka, S. Klotz, and R. Brandl. 2001. The niche of higher plants: evidence for phylogenetic conservatism. Proceedings of the Royal Society of London B 268: 2383–2389. Silvertown, J., M. Dodd, D. Gowing, C. Lawson, and K. McConway. 2006. Phylogeny and the hierarchical organization of plant diversity. Ecology 87:S39–S49. Uriarte, M., R. Condit, C. D. Canham, and S. P. Hubbell. 2004. A spatially explicit model of sapling growth in a tropical forest: Does the identity of neighbours matter? Journal of Ecology 92:348–360. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, and S. W. Kembel. 2004. Phylocom: software for the analysis of community phylogenetic structure and character evolution. Version 3.19. hhttp:// www.phylodiversity.net/phylocomi Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Webb, C. O., and M. J. Donoghue. 2005. Phylomatic: tree assembly for applied phylogenetics. Molecular Ecology Notes 5:181–183. Webb, C. O., and D. R. Peart. 2000. Habitat associations of trees and seedlings in a Bornean rain forest. Journal of Ecology 88:464–478. Weiher, E., and P. A. Keddy. 1995. Assembly rules, null models, and trait dispersion: new questions from old patterns. Oikos 74:159–164. Weiher, E., and P. A. Keddy, editors. 1999. Ecological assembly rules: perspectives, advances, retreats. Cambridge University Press, Cambridge, UK. Wikstro¨m, N., V. Savolainen, and M. W. Chase. 2001. Evolution of the angiosperms: calibrating the family tree. Proceedings of the Royal Society of London B 268:2211–2219. Wilson, J. B. 1995. Null models for assembly rules: the Jack Horner effect is more insidious than the Narcissus effect. Oikos 72:139–144. Wright, S. J. 2002. Plant diversity in tropical forests: a review of mechanisms of species coexistence. Oecologia 130:1–14.

SUPPLEMENT A phylogenetic tree and key to species codes for tree species occurring within the 50-ha Forest Dynamics Plot (FDP) on Barro Colorado Island, Panama (Ecological Archives E087-113-S1).

Ecology, 87(7) Supplement, 2006, pp. S100–S108 Ó 2006 by the Ecological Society of America

PHYLOGENETIC CLUSTERING AND OVERDISPERSION IN BACTERIAL COMMUNITIES M. CLAIRE HORNER-DEVINE1,3 1

AND

BRENDAN J. M. BOHANNAN2

School of Aquatic and Fishery Science, University of Washington, Seattle, Washington 98195 USA 2 Department of Biological Science, Stanford University, Stanford, California 94305 USA

Abstract. Very little is known about the structure of microbial communities, despite their abundance and importance to ecosystem processes. Recent work suggests that bacterial biodiversity might exhibit patterns similar to those of plants and animals. However, relative to our knowledge about the diversity of macro-organisms, we know little about patterns of relatedness in free-living bacterial communities, and relatively few studies have quantitatively examined community structure in a phylogenetic framework. Here we apply phylogenetic tools to bacterial diversity data to determine whether bacterial communities are phylogenetically structured. We find that bacterial communities tend to contain lower taxonomic diversity and are more likely to be phylogenetically clustered than expected by chance. Such phylogenetic clustering may indicate the importance of habitat filtering (where a group of closely related species shares a trait, or suite of traits, that allow them to persist in a given habitat) in the assembly of bacterial communities. Microbial communities are especially accessible for phylogenetic analysis and thus have the potential to figure prominently in the integration of evolutionary biology and community ecology. Key words: bacteria; community structure; microbial ecology; phylogenetic clustering and overdispersion; phylogenetic diversity; phylogenetic structure; relatedness.

INTRODUCTION Although there may be millions of species of bacteria on Earth, we are only beginning to investigate patterns in their diversity (Horner-Devine et al. 2004a). Understanding patterns of bacterial diversity is of particular importance, because bacteria likely comprise the majority of the planet’s biodiversity, they mediate many environmental processes that sustain life, and their diversity is of great importance in medicine, agriculture, and industry. Recent evidence suggests that bacteria can exhibit patterns in taxonomic diversity and community composition similar to those of plants and animals (e.g., Horner-Devine et al. 2003, 2004b). However, most of these studies have relied on measures of diversity that do not consider phylogenetic relatedness (Bohannan and Hughes 2003), and few studies have quantitatively examined bacterial communities within a phylogenetic context (Martin 2002, Davelos et al. 2004, Martin et al. 2004). Phylogenetic measures can reveal differences in the richness or composition of two communities that would be identical using standard measures of species richness and composition (Martin 2002). Phylogenetic analyses of diversity have proven valuable in studies of plant and animal diversity, because such an approach can lend Manuscript received 9 February 2005; revised 13 July 2005; accepted 12 August 2005; final version received 7 September 2005. Corresponding Editor (ad hoc): J. B. Losos. For reprints of this Special Issue, see footnote 1, p. S1. 3 E-mail: [email protected]

insight into the relative importance of evolutionary and ecological forces in shaping communities (Elton 1946, Webb et al. 2002, Cavender-Bares and Wilczek 2003). The idea that closely related taxa are more likely to interact intensely with each other than with more distantly related taxa is an old one (Darwin 1859); more recently, this idea has been expanded to suggest that interspecific interactions are influenced by the net ecological similarity of taxa, and closely related taxa tend to be more similar ecologically than distantly related taxa (Harvey and Pagel 1991). For example, cooccurring rainforest tree species have been observed to be more closely related than expected by chance (Webb 2000); such a pattern of phylogenetic attraction or clustering can indicate that these closely related taxa share traits important for their persistence in a particular environment (Webb et al. 2002). Such habitat filtering is important and might be more important than competition, in maintaining rain forest tree species diversity (see also Tofts and Silvertown 2000, Webb 2000, Kembel and Hubbell 2006). In contrast, a community could be composed of distantly related taxa as a result of current or past competitive exclusion between similar (and thus closely related) taxa and/or as a result of convergent evolution in traits important for persistence in a given environment (Cavender-Bares et al. 2004, Kembel and Hubbell 2006). However, even for macro-organisms, relatively few studies have quantitatively examined community structure in a phylogenetic framework (but see other articles this issue), and even fewer have done so for

S100

July 2006

RELATEDNESS IN BACTERIAL COMMUNITIES

microbes (but see Martin 2002, Francis et al. 2003, Martin et al. 2004). We know of no other microbial studies that employ the approach used here to quantify community structure. Bacteria offer a unique opportunity to examine the phylogenetic structure of multiple communities, because most recently published bacterial diversity data are molecular in nature and thus can be more easily interpreted within a phylogenetic context than data from many community studies of macro-organisms (Bohannan and Hughes 2003). A large proportion of microbes cannot be cultured with current laboratory techniques (Brock 1987), and thus bacterial taxa are often identified from the sequences of indicator genes extracted from environmental samples (Stackebrandt and Goebel 1994). Here we take a quantitative approach to examining the phylogenetic structure of bacterial communities from a number of different environments. We ask whether bacterial communities exhibit phylogenetic structure (e.g., significant degrees of clustering or overdispersion of taxa across a phylogenic tree) and whether such patterns vary along environmental gradients. METHODS Data We used existing bacterial sequence data from four different environments and for three different genes. We selected data sets that were of high resolution (e.g., cloned and sequenced DNA sequences, rather than gradient gel or restriction fragment length data), that were from extensively sampled communities (relative to many studies of bacteria diversity), and that were replicated or spanned ecologically interesting environmental gradients in aquatic, soil, and sediment habitats. The first data set consists of partial 16S rDNA sequences (the most common indicator gene used for bacterial diversity studies) sampled from freshwater mesocosms that span a productivity gradient (HornerDevine et al. 2003). Each mesocosm consisted of a 2 m diameter polyethylene cattle tank with a screen cover, filled with well water. Each mesocosm was inoculated from the same-pooled sample collected from six ponds in southern Michigan that spanned a natural gradient in primary productivity. A gradient of primary productivity was established across the mesocosms by maintaining otherwise identical mesocosms with different input concentrations of nitrogen and phosphorus. At the end of a 4-mo growing season, one composite water column sample was used to generate a clone library from each of the five mesocosms. We selected ;100 clones from each of these libraries and sequenced 500 nucleotides from the 5 0 terminal of each clone (GenBank accession numbers DQ064816–DQ065575). The second data set consists of 16S rDNA sequences sampled from soil communities at different depths that differed in water saturation and total organic carbon (Zhou et al. 2002). Soil cores were collected from

S101

previously described sites in northern Virginia (Abbot’s Pitt) and central Delaware (Zhou et al. 2002) at both the soil surface (depth ¼ 0.05 m; GenBank accession numbers AY280351–AY289492) and subsurface (approximately at the depth of the water table, depth . 4.0 m; GenBank accession numbers AY456755–AY456883 and AY456885–AY456903) for a total of five samples (note that two samples were collected from the subsurface of Dover Air Force Base; hereafter, subsurface-D1 and subsurface-D2). The third data set consists of 16S rDNA sequences sampled specifically from ammonia-oxidizing bacteria in Costa Rican soils (Carney et al. 2004). Samples were collected from three land use types on sandy loam soils: forest, pasture, and tree plantations. The tree plantation sites included three different site types that differed in plant community composition and richness. The onespecies sites contained only Cordia alliodora, the threespecies sites contained C. alliodora, an herb, and a palm, and the five-species sites contained C. alliodora, two palm species, and two other hardwoods. Each of the five site types (forest, pasture, and the three plantation treatments) was replicated three times, with the exception of the five-species site, which had two replicates. For each site type, a composite soil sample was collected from each of the replicate plots for a total of 14 samples. Partial 16S rDNA sequences from these samples were deposited in GenBank under accession numbers AY631475–AY631851. The fourth data set consists of sequences of functional genes amplified from five sediments samples collected along a salinity and nitrogen gradient in the Chesapeake Bay (Francis et al. 2003; C. A. Francis, J. C. Cornwell, and B. B. Ward, unpublished manuscript). One of these functional genes (amoA) codes for a subunit of ammonia monooxygenase, an enzyme found only in ammoniaoxidizing bacteria (bacteria that mediate the transformation of ammonia into nitrite). A 450 base pair (bp) region was chosen for phylogenetic analysis, representing 150 amino acids (GenBank accession numbers AY352899–AY353054; Francis et al. 2003). The second functional gene (nirS) codes for a subunit of nitrite reductase, an enzyme found in denitrifying bacteria (bacteria that mediate the transformation of nitrite into nitrogen gas). A 233-bp region was used for analyses (C. A. Francis, J. C. Cornwell, and B. B. Ward, unpublished manuscript). Both gene fragments span the active site of their respective proteins (Berks et al. 1995, Rotthauwe et al. 1997, Braker et al. 2000). For each data set, we screened sequences for chimeras and aligned them using the 2002 version of the ARB software package (for 16S genes; available online)4 or Sequencher software (for functional genes; Gene Codes Corporation, Ann Arbor, Michigan, USA). We used only unambiguously aligned positions to construct the 4

hhttp://www.arb-home.de/i

S102

M. C. HORNER-DEVINE AND B. J. M. BOHANNAN

phylogenetic hypotheses, and duplicate sequences were not used when generating phylogenetic trees. Thus sample sizes represent the number of unique sequences observed, rather than the total number of sequences analyzed. Analysis We used indices of community phylogenetic structure to compare these communities (Webb 2000). The net relatedness index (NRI) and nearest taxa index (NTI) measure the degree of phylogenetic clustering of taxa across a phylogenetic tree in a given sample relative to the regional pool of taxa. Positive values indicate that a community is clustered, whereas negative values indicate that community members are evenly spread or overdispersed across a phylogenetic tree. In other words, a positive NRI value indicates a community where members are on average more closely related to one another than they are to members of the regional taxon pool. Such a community thus appears to be clustered on a phylogenetic tree of the regional taxa. The NRI measures overall clustering across the phylogeny as the average distance between all pairs of taxa in a community. Specifically, NRI ¼
(Xnet – X(n))/SD(n), where Xnet is the mean phylogenetic distance, measured as the mean pairwise branch lengths and thus a measure of pairwise sequence divergence, between all pairs of n taxa in a particular community; X(n) and SD(n) are the mean and standard deviation of phylogenetic distance for n taxa randomly distributed on the phylogeny. We obtain these latter values by 1000 random draws from the entire pool of taxa in the phylogeny. Alternatively, NTI measures the extent of terminal clustering on the phylogeny by determining the minimal distance or branch length between taxa in a particular community. The two indices are calculated similarly, except NTI substitutes Xnear for Xnet, where Xnear is the shortest mean distance between all pairs of n taxa in a community sample. We calculated NRI and NTI using Phylocom 3.22 (Webb et al. 2005). High and positive values of these indices indicate clustering of taxa across the overall phylogeny, whereas low or negative values indicate overdispersion of taxa across the phylogeny. We tested whether these values (and thus whether the extent of clustering) significantly differed from that of a randomly assembled community with a null model (1000 permutations of randomly drawn communities). We used a two-tailed significance test to evaluate the rank of observed values at P ¼ 0.05, such that an observed rank of ,25 or .975 was assumed to be significant overdispersion or clustering, respectively. Calculation of NRI and NTI relies on a community phylogeny. We used ModelTest 3.06 to determine the best models of sequence evolution for the unique amoA and nirS sequences from the Chesapeake Bay sediment samples (Posada and Crandall 1998). Using the Akaike Information Criterion (Akaike 1973), we selected

Ecology Special Issue

K81ufþIþG as the best model of sequence evolution for the amoA sequences and TVMþIþG for the nirS sequences. The 16S rDNA trees were constructed using neighbor-joining distance clustering with a HKY þ gamma substitution model (Hasegawa et al. 1985), where gamma was estimated from the data. We used PAUP* to construct trees for all data sets (Swofford 2002). Maximum likelihood methods were used to estimate branch lengths based on the above HKY and gamma DNA substitution models. Trees were bootstrapped to examine phylogenetic robustness. We also examined how phylogenetic clustering varies along environmental gradients. NRI and NTI values were standardized by the mean expected value for the number of taxa found in each community (Webb et al. 2002). We then used regression and ANOVA implemented in JMP, version 4.0, to examine the relationship between clustering values and environmental parameters (Sokal and Rohlf 1995). Where data were not normally distributed (as determined by the Shapiro-Wilk W test), we used the Kruskal-Wallis one-way analysis of variance by ranks (Sokal and Rohlf 1995). The following environmental parameters were considered: chlorophyll a (for the mesocosm data from Horner-Devine et al. [2003]), carbon content (for the soil community data from Zhou et al. [2002]), plant diversity and ammonia (for the ammonia oxidizer data from Carney et al. [2004]), ammonia and salinity (for the amoA data), and nitrate and salinity (for the nirS data). These environmental parameters were identified in the respective prior studies as important to taxonomic richness and/or community composition. RESULTS Phylogenetic structure We observed that most of the bacterial communities we examined exhibited significant phylogenetic structure (i.e., bacteria tended to co-occur with other bacteria that were more closely related than expected by chance). For example, bacterial communities from freshwater mesocosms exhibited significant and positive net relatedness index (NRI) and nearest taxa index (NTI) values (Table 1). This was true when all bacteria were considered, as well when the three most common groups of bacteria sampled from each of the mesocosms (Alphaproteobacteria, Betaproteobacteria, and CytophagaBacteroides-Flavobacteria, or CFB) were considered separately. While there was some variation in relatedness among the different communities and groups of taxa, bacteria in the three most common taxonomic groups in these communities tended to be clustered. Soil communities sampled at different depths showed different patterns of phylogenetic structure (Table 2). Subsurface soil communities showed significant clustering for both NRI and NTI. In contrast, one surface sample was randomly structured phylogenetically, and the other exhibited significant overdispersion for both indices.

July 2006

RELATEDNESS IN BACTERIAL COMMUNITIES

TABLE 1. Net relatedness index (NRI) and nearest taxa index (NTI) results for the 16S rDNA sequences from the five freshwater mesocosm communities.

TABLE 3. The NRI and NTI results for the Costa Rican soil nitrifiers. Group

Community

N

NRI

NRI_gt

NTI

NTI_gt

All bacteria 1 2 3 4 5

108 114 87 117 104

1.47* 7.28* 2.47*
2.18* 3.8*

934 999 988 13 998

5.54* 4.95* 2.81* 3.18* 2.51*

999 999 998 998 998

Betaproteobacteria 1 18 2 35 3 2 4 19 5 7

2.34* 4.62* 1.02 0.47 0.31

991 999 76 690 627

2.71* 0.87
1.36 1.89  0.98

999 814 77 972 825

Alphaproteobacteria 1 18
1.48 2 28 2.96* 3 14 1.53 4 33
2  5 42 4.03*

74 999 935 25 999

2.44* 1.97* 1.35 1.94* 2.35*

998 994 913 994 998

Cytophaga-Bacteroides-Flavobacteria (CFB) 1 26 3.44* 999 4.57* 2 20 2.52* 996 3.29* 3 13 1.21 889 0.48 4 24 2.38* 996 2.68* 5 7 0.08 504
0.78

999 999 673 996 234

Notes: N ¼ no. taxa in a community. NRI_gt and NTI_gt represent the number of times the observed NRI and NTI values for a community, respectively, were greater than the value for randomly permuted communities. * Communities that are significantly structured at the P ¼ 0.05 level.   Communities that are significantly structured at the P ¼ 0.10 level.

Ammonia-oxidizer communities from Costa Rican soils exhibited the most variation in phylogenetic structure of all the data sets considered (Table 3). While communities from forest soils showed no significant phylogenetic structure, pasture communities tended to be overdispersed. Communities from the experimental plant treatments with one, three, or five plant species TABLE 2. The NRI and NTI results for the 16S rDNA sequences from soil communities at different depths. Community

N

NRI

NRI_gt

NTI

NTI_gt

Subsurface A Subsurface D1 Subsurface D2 Surface D Surface A

43 27 20 66 65

3.15* 2.59* 3.81*
3.4*
0.98

999 994 999 0 159

3.72* 2.97* 3.81*
1.88 
0.29

999 999 999 27 382

Notes: Labels D and A refer to samples collected at Dover Air Force Base (Delaware, USA) and Abbot’s Pitt (Virginia, USA), respectively. As two subsurface samples were collected at Dover Air Force Base, they are denoted D1 and D2. Other abbreviations and symbols are as in Table 1. * Communities that are significantly structured at the P ¼ 0.05 level.   Communities that are significantly structured at the P ¼ 0.10 level.

S103

Forest Forest Forest Pasture Pasture Pasture One One One Three Three Three Five Five

Community N F1 F2 F3 PF PR PS One1 One2 One3 Three1 Three2 Three3 Five1 Five2

18 18 18 20 20 19 29 33 34 36 35 31 30 27

NRI

NRI_gt

NTI

NTI_gt


0.58 0.05
0.46
2.06*
2.42*
4.71*
1.99* 2.73* 2.32* 2.97* 1.19 2.43* 0.91 2.5*

264 508 314 26 9 0 23 997 996 998 887 995 802 998

0.77 0.77 1.4
0.16
2.41*
3.76* 1.53 1.84* 1.26 1.93*
0.45 1.8* 0.3 2.25*

751 755 937 394 10 0 959 986 882 987 332 983 588 999

Notes: ‘‘One,’’ ‘‘Three,’’ and ‘‘Five’’ indicate the three plantation treatments containing the respective number of different species. Community labels describe the group and sample number. Other abbreviations and symbols are as in Table 1. * Communities that are significantly structured at the P ¼ 0.05 level.

tended to be phylogenetically clustered overall with no clear pattern in the NTI. Finally, sediment bacterial communities sampled at five sites in the Chesapeake Bay were phylogenetically structured (i.e., clustered) and contained less genetic diversity than a randomly assembled community (Table 4). This was true for both the ammonia-oxidizing bacteria and denitrifying bacteria sampled. All but one sample showed significant overall phylogenetic structure (as estimated by NRI). Interestingly, one community of denitrifying bacteria exhibited significant overdispersion as measured by NRI. Phylogenetic clustering measured by NTI was more common for denitrifying bacteria than for ammonia-oxidizing bacteria. Only one of the five TABLE 4. The NRI and NTI results for the five Chesapeake sediment communities. Community

N

NRI

amoA CB1 CB2 CB3 CT1 CT2

NRI_gt

NTI

NTI_gt

24 18 22 26 14

3.58* 5.66* 6.17* 5.04* 1.95*

999 999 999 999 965


0.64 2.79* 0.47 0.88 1.56

275 999 688 789 940

nirS CB1 CB2 CB3 CT1 CT2

79 46 53 75 87


4.66* 5.13* 4.27* 0.39 3.09*

0 999 999 631 999


0.32 3.56* 3.45* 2.45* 5.54*

363 998 999 990 999

Note: Sampling stations were located in the Choptank River (CT) as well as in the main channel of the Chesapeake Bay (CB). CT1 was located in the upper Choptank, while CT2 was located in the lower Choptank. Main channel stations were located in the north bay (CB1), mid-bay (CB2), and south bay (CB3). Other abbreviations and symbols are as in Table 1. * Communities that are significantly structured at the P ¼ 0.05 level.

S104

M. C. HORNER-DEVINE AND B. J. M. BOHANNAN

Ecology Special Issue

was a negative trend between NRI estimated for Betaproteobacteria and primary productivity in aquatic mesocosms (Fig. 1A). Relatedness did not vary with productivity for any of the other bacterial groups examined in these mesocosms (results not shown). Phylogenetic structure varied significantly with depth for soil communities (NRI, t ¼ 5.319, df ¼ 3, P ¼ 0.013; NTI, t ¼ 6.693, df ¼ 3, P ¼ 0.0068). In addition, communities sampled from soils with high total organic carbon had lower relatedness values than those sampled from low total organic carbon soils (Fig. 1B). In the Chesapeake Bay, relatedness of denitrifying bacteria exhibited a nonsignificant trend with nitrate and salinity (Fig. 1C for nitrate results; salinity: NRI, NS, results not shown; NTI, R2 ¼ 0.5121, P ¼ 0.107). In contrast, relatedness measures of ammonia-oxidizing bacteria did not vary with any environmental parameters measured (ammonium and salinity, results not shown). Phylogenetic structure of communities of ammoniaoxidizing bacteria varied with plant community composition in Costa Rican soils for NRI (ANOVA, F4,9 ¼ 5.507, MSE ¼ 2.33, P ¼ 0.0160), but not for NTI (Kruskal-Wallis: v2 ¼ 6.5095, df ¼ 4, P ¼ 0.164). Pairwise post-hoc comparisons of the different treatments revealed that the bacterial communities from sites with one, three, or five focal plant species were more clustered than pasture communities as measured by NRI. There was a weak positive relationship between ammonia and NTI, but not NRI, for these bacteria (R2 ¼ 0.218, P ¼ 0.052). DISCUSSION

FIG. 1. Variation of relatedness along environmental gradients. (A) Relatedness decreased with increasing productivity in freshwater mesocosms for Betaproteobacteria (NRI, R2 ¼ 0.829, P ¼ 0.0588; mesocosm three was excluded due to small community size). (B) Relatedness decreased with total organic carbon in bacterial soil communities (NRI, R2 ¼ 0.925, P ¼ 0.0058; NTI, R2 ¼ 0.966, P ¼ 0.0017). (C) There was a significant negative relationship between the relatedness of nirS genes and nitrate in Chesapeake Bay sediment communities (NRI, R2 ¼ 0.909, P ¼ 0.0077; NTI, R2 ¼ 0.601, P ¼ 0.0768).

ammonia-oxidizing communities sampled showed significant phylogenetic clustering as estimated by NTI. Phylogenetic structure and the environment We also observed that measures of phylogenetic structure can vary along environmental gradients. There

Our results suggest that bacteria tend to co-occur with other closely related bacteria more often than expected by chance, as has been observed for some plant species (Webb 2000; also see Cavender-Bares et al. 2006, Kembel and Hubbell 2006, Lovette and Hochachka 2006, Weiblen et al. 2006). In addition, we observed that phylogenetic structure can vary along environmental gradients. We observed significant net relatedness index (NRI) and nearest taxa index (NTI) values for freshwater bacterial communities from experimental mesocosms. Relatedness information provides a different window into bacterial communities than does information concerning richness or taxonomic composition. Accordingly, our observations that the relatedness of cooccurring Betaproteobacteria decreases with productivity, is in contrast to previous observations that their taxonomic richness does not (Horner-Devine et al. 2003). We observed that each of the five communities contained approximately the same number of taxa regardless of productivity, but these taxa tended to be more distantly related at higher productivities. Decreasing relatedness with increasing productivity might indicate that low productivity environments are more ‘‘stressful’’ (e.g., impose a stronger ‘‘filter’’ on a

July 2006

RELATEDNESS IN BACTERIAL COMMUNITIES

community) for Betaproteobacteria than do more productive environments. In contrast, we did not observe a relationship between relatedness of Alphaproteobacteria or Cytophaga-Bacteroides-Flavobacteria (CFB) and productivity in the mesocosm study, despite our previous observations of changes in taxonomic richness of these groups with productivity (HornerDevine et al. 2003). Clustered distributions have been interpreted as evidence of habitat filtering, where a group of closely related species shares a trait, or suite of traits, that allow them to persist in a given habitat (Webb et al. 2002). Alternatively, significant phylogenetic clustering could be the result of differential dispersal and/or colonization abilities, or an adaptive radiation event. While it is beyond the scope of this work to determine the process responsible for the clustering we observed, the results from the freshwater mesocosms suggest that habitat filtering, rather than an adaptive radiation event or colonization effects, was important in the assembly of these bacterial communities. In this mesocosm study, bacterial communities were established in each mesocosm from the same inoculum of bacteria from natural pond communities (Horner-Devine et al. 2003). Thus, the history of colonization could not play a role in the patterns we observed. Although dispersal among mesocosms was not prevented (and likely occurred), the productivity gradient was randomized in space (i.e., mesocosms with similar productivities were not clustered in space), and thus clinal dispersal is unlikely to underlie the patterns we observed. Finally, since the mesocosm communities were sampled four months after they were initiated, it is unlikely that the ribosomal gene evolved during the course of the experiment, and thus it is unlikely that the clustering is due to adaptive radiation during the experiment. Without such information for the other data sets we analyzed, it is difficult to distinguish among habitat filtering, adaptive radiation, or colonization processes in the systems that were not manipulated. However, the results from the mesocosm study suggest that habitat filtering could be an important force in the assembly of at least some bacterial communities. Overdispersion of taxa across a phylogeny has been observed in natural communities (Ackerly et al. 2006, Cavender-Bares et al. 2006, Silvertown et al. 2006) and could indicate that negative interactions (e.g., competition) are important in community assembly (Graves and Gotelli 1993, Webb et al. 2002). Although we did observe significant phylogenetic overdispersion in one of the freshwater bacterial communities, our observations do not suggest that competition played an overwhelming role in structuring the communities we studied at the scales examined here. However, it is important to note that habitat filtering and competition likely act in concert to produce the communities we observe. Thus even where the phylogenetic signature suggests the importance of habitat filtering, local competition can also be occurring.

S105

Phylogenetic scale (the taxonomic group or rank under consideration) has been shown to influence the observation of phylogenetic patterns (Silvertown et al. 2001, 2006, Cavender-Bares et al. 2004, 2006). However, the effect of phylogenetic scale was not evident for bacteria from the freshwater mesocosms. We tended to observe phylogenetic clustering both when we examined all bacteria together and when we examined taxonomic subsets of the bacteria (Alphaproteobacteria, Betaproteobacteria, and CBF). It is possible that Alphaproteobacteria, Betaproteobacteria, and CBF encompass such a broad range of bacterial ecotypes that, even with the NTI (which focuses on terminal clustering and thus is particularly sensitive), it is less likely that one will observe overdispersion. Recent work by Zhou et al. (2002) and Treves et al. (2003) suggests that spatial isolation plays an important role in structuring soil bacterial communities. They observed that unsaturated, surface communities had more uniform rank abundance patterns than did communities from saturated, subsurface communities, which exhibited high-dominance distributions (Zhou et al. 2002). They interpreted the uniform distribution from the surface samples as evidence that local competition does not play a significant role in structuring the soil communities they studied. Their observations (and subsequent mathematical modeling and laboratory experimentation) suggested that spatial isolation might limit competition in the surface soils (Treves et al. 2003). Our results do not support this hypothesis. We observed phylogenetic clustering in the subsurface samples, where spatial isolation was predicted to be minimized due to high water content and thus where there was an expectation for strong competition and phylogenetic overdispersion. In contrast we observed phylogenetic overdispersion in one of the surface samples, where isolation was predicted to be high and competition weak. We did observe that phylogenetic clustering decreased with increasing total organic carbon (which covaried with depth) in the Zhou et al. (2002) soil data set. Thus as carbon availability increased, the strength of clustering and perhaps habitat filtering decreased. The potential decrease in the strength of filtering is unlikely to be related to an increase in the role of competition for carbon, since competition would likely decrease with increasing carbon. More information on the types of carbon present, as well as C:N ratios, might lend more insight into the underlying processes. We observed both phylogenetic clustering and overdispersion for ammonia-oxidizing bacteria from Costa Rican soils. It is possible that, for ammonia-oxidizing bacteria, a more restricted (and potentially more ecologically similar) group of taxa, it might be easier to detect interactions among taxa, because ammoniaoxidizing bacteria likely compete for similar resources. Carney et al. (2004) found that neither bacterial richness nor composition changed across plant diversity treat-

S106

M. C. HORNER-DEVINE AND B. J. M. BOHANNAN

ments (one, three, or five focal plant species). Similarly, we did not observe pairwise differences in phylogenetic structure among the plant treatments using post-hoc comparisons. Carney et al. (2004) also observed differences in the ammonia-oxidizer community among landuse types in some measures of diversity and in composition. We observed that, while pasture and forest did not differ in phylogenetic clustering, pasture communities were less phylogenetically diverse (i.e., less clustered) than each of the plant treatments. In fact, pasture communities tended to be overdispersed. We also found a weak positive relationship between terminal clustering and ammonia, such that clustering (and perhaps the importance of habitat filtering rather than competition) increased with ammonia. This is consistent with an increase in the importance of habitat filtering (and conversely, perhaps a decrease in the relative strength of competition) for ammonia as ammonia concentrations increased. Analysis of amoA and nirS genes sampled from the Chesapeake Bay offered us an opportunity to examine community patterns deduced from potentially ‘‘ecologically relevant’’ genes. We assumed that phylogenetic overdispersion would be more prevalent in such data, where the taxa sampled are essentially from a single guild (i.e., a group performing the same function and requiring the same resources). However, we did not observe overdispersion for amoA, and we observed overdispersion in only one sample of nirS. In hindsight this is not overly surprising for nirS, given that denitrifying bacteria can be ecologically very different despite sharing the nirS gene (Shapleigh 2000). However, ammonia-oxidizing bacteria are believed to be a physiologically constrained group, existing solely on the oxidation of ammonia. The lack of overdispersion suggests that the amoA gene may be too conserved to reveal ecological differences or that sequence variation at this scale does not reflect ecological differences. Previous analysis of the amoA samples demonstrated a nonsignificant trend toward decreasing richness with increasing salinity (Francis et al. 2003). We did not observe a relationship between phylogenetic structure as inferred from this gene and salinity or ammonia. In contrast, we observed a significant decrease in phylogenetic clustering as inferred from the nirS gene with increasing nitrate and a weak relationship between clustering and salinity. It is difficult to interpret the relationship with nitrate; one might expect that increasing nitrate availability would lead to decreased competition for this substrate by denitrifiers or stronger habitat filtering for denitrifying bacteria with genes that confer an advantage at high nitrate concentrations; we observed the opposite trend. The interpretations of our results are based on the assumption that closely related taxa are more ecologically similar than distantly related taxa. This assumption has been shown to be true for some plants, animals, and microbes (Kuittinen et al. 1997, Nubel et al. 1999,

Ecology Special Issue

Morgan et al. 2001, Prinzing et al. 2001), but not all (Losos et al. 2003, Rice et al. 2003, Knouft et al. 2006). How universally the assumption about similarity of closely related organisms applies to microorganisms is currently unknown. If this assumption does not hold for most bacteria, other explanations might be necessary for the patterns we observe. For example, some bacteria are capable of lateral gene transfer (LGT; Ochman et al. 2000, Lerat et al. 2003). Lateral gene transfer among cooccurring bacteria could weaken or uncouple the relationship between ecological similarity and evolutionary relatedness, if ecologically relevant genes are exchanged more often than phylogenetically informative housekeeping genes (e.g., ribosomal genes) as has been suggested (Lerat et al. 2003). Rampant LGT would reduce the prevalence of phylogenetic clustering or overdispersion due to ecological processes. However, we observed a significant level of phylogenetic clustering in the communities that we examined, suggesting that LGT does not substantially overwhelm phylogenetic patterns in these communities. In addition, recent work in environmental genomics suggests that on recent evolutionary time scales horizontal gene transfer is not rampant in natural microbial communities (Lerat et al. 2003). Martin et al. (2004) used a different approach (lineage-per-time analysis) to look for phylogenetic patterns in microbial diversity data. They failed to show significant phylogenetic structure (i.e., an overabundance of closely related or distantly related sequences) across several different data sets. However their study differed from ours in that they assumed a ‘‘universal’’ null model (an exponential increase in lineages) for all data sets, rather than creating a null expectation for each data set by resampling of a regional phylogenetic tree. In the approach used here, we are interested in whether observed communities differ in phylogenetic diversity from communities created by a random draw from the available taxa in the regional pool. The Martin et al. (2004) approach suffers from a lack of power if the fraction of diversity sampled is small, making the task of detecting phylogenetic structure very difficult. While the communities examined here are also undersampled, the use of a relative measure of community relatedness decreases the influence of undersampling, provided communities in a given analysis are sampled with equal effort. Furthermore, if the questions of interest concern community assembly (as they do in our study), assuming a null model based on regional tree resampling is appropriate because it should model the assembly process. We have observed that bacterial communities exhibit phylogenetic structure, in some cases similar to that observed for plants, and that this structure can vary along environmental gradients. Our results suggest that habitat filtering might be relatively more important to the assembly of bacterial communities than competition. Why might this be the case? Recent work suggests that

July 2006

RELATEDNESS IN BACTERIAL COMMUNITIES

the greater the degree of environmental heterogeneity over which one samples a community, the more likely that phylogenetic clustering, rather than overdispersion, will be present (Cavender-Bares et al. 2006, Silvertown et al. 2006). The data sets used in our study consisted of samples that were extremely large relative to the size of the target organisms and the scales over which individuals interact (as is the case in most studies of microbial diversity) and thus likely included substantial environmental heterogeneity. As such, the spatial scale of sampling might bias the results towards clustering rather than overdispersion. Recent studies have also suggested that phylogenetic scale should affect the prevalence of clustering; increasing phylogenetic scale (i.e., an increase in taxonomic lumping) might result in an increased prevalence of clustering (Silvertown et al. 2006). Substantial ecological diversity has been shown to be present within microbial taxa defined using molecular markers, especially ribosomal markers (e.g., Ward et al. 1998, Rocap et al. 2002). The molecular approach used to characterize microbial diversity might bias microbial diversity data sets toward detection of phylogenetic clustering when relatively large groups of organisms are targeted. Finer scale markers (e.g., internal transcribed spacer (ITS) or multi-gene approaches) might reveal the presence of increased phylogenetic overdispersion. The growing possibility of using environmental genomics to examine full genomes of different lineages from environmental samples will provide even more power to this approach (R. Whitaker and J. F. Banfield, unpublished manuscript). The search for patterns in microbial biodiversity is in its infancy, and it is premature to make strong conclusions regarding the exact mechanisms responsible for the patterns we have described. To make such conclusions with confidence, a better understanding of the relationship between community assembly mechanisms and phylogenetic patterns is necessary. Such an understanding could be developed through studies of controlled experimental systems, where, for example, one can manipulate environmental parameters such as resource availability and tease apart the effects of mechanisms that occur on large scales of time and space (such as evolution and differential colonization) from those that occur on smaller scales (such as habitat filtering and competition). Microbial model systems could serve as excellent experimental systems in which to explore these ideas (Jessup et al. 2004). We also suggest that future work in natural systems examine gradients that span a greater range of environmental characteristics to test the hypothesis that phylogenetic clustering and thus habitat filtering increases with environmental extremes. Such studies coupled with a better understanding of the extent of LGT, how traits map onto phylogenies, and the evolutionary history of these traits (Ackerly et al. 2006, Silvertown et al. 2006) will help to explain the patterns of phylogenetic structure we observed. Where we do observe similar patterns of

S107

phylogenetic structure for microbial and macrobial communities, it is possible that similar mechanisms could be responsible for the structure of communities from these very different forms of life. ACKNOWLEDGMENTS We are grateful to David Ackerly, Will Cornwell, Clara Davis, and Cam Webb for many informative discussions about phylogenetic structure and community assembly. We would also like to thank Karen Carney, Chris Francis, and Jizhong Zhou for allowing us to use their data for these analyses. Finally, we thank Jonathan Losos and two anonymous reviewers for their comments and suggestions. M. C. HornerDevine was supported by funding from the American Association of University Women and the Center for Evolutionary Studies at Stanford. This work was also supported by a grant from the NSF (DEB0108556) to B. Bohannan. LITERATURE CITED Ackerly, D. D., D. W. Schwilk, and C. O. Webb. 2006. Niche evolution and adaptive radiation: testing the order of trait divergence. Ecology 87:S50–S61. Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267–281 in B. N. Petrov and F. Csaki, editors. Second International Symposium on Information Theory. Academiai Kiado, Budapest, Hungary. Berks, B. C., J. W. B. Moir, and D. J. Richardson. 1995. Enzymes and associated electron transport systems that catalyse the respiratory reduction of nitrogen oxides and oxyanions. Biochimica et Biophysica Acta 1232:97–173. Bohannan, B. J. M., and J. B. Hughes. 2003. New approaches to analyzing microbial biodiversity data. Current Opinion in Microbiology 6:282–287. Braker, G., J. Zhou, L. Wu, A. H. Devol, and J. M. Tiedje. 2000. Nitrite reductase genes (nirK and nirS) as functional markers to investigate diversity of denitrifying bacteria in Pacific northwest marine sediment communities. Applied and Environmental Microbiology 66:2096–2104. Brock, T. D. 1987. The study of microorganisms in situ: progress and problems. Symposium of the Society for General Microbiology 41:1–17. Carney, K. M., P. A. Matson, and B. J. M. Bohannan. 2004. Diversity and composition of tropical soil nitrifiers across a plant diversity gradient and among land-use types. Ecology Letters 7:684–694. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Cavender-Bares, J., A. Keen, and B. Miles. 2006. Phylogenetic structure of Floridian plant communities depends on taxonomic and spatial scale. Ecology 87:S109–S122. Cavender-Bares, J., and A. Wilczek. 2003. Integrating microand macroevolutionary processes in community ecology. Ecology 84:592–597. Darwin, C. 1859. On the origin of species. John Murray, London, UK. Davelos, A. L., K. Xiao, D. A. Samac, A. P. Martin, and L. L. Kinkel. 2004. Spatial variation in Streptomyces genetic composition and diversity in a prairie soil. Microbial Ecology 48:601–612. Elton, C. S. 1946. Competition and the structure of ecological communities. Journal of Animal Ecology 15:54–68. Francis, C. A., G. D. O-Mullan, and B. B. Ward. 2003. Diversity of ammonia monooxygenase (amoA) genes across environmental gradients in Chesapeake Bay sediments. Geobiology 1:129–140. Graves, G. R., and N. J. Gotelli. 1993. Assembly of avian mixed-species flocks in Amazonia. Proceedings of the National Academy of Sciences (USA) 90:1388–1391.

S108

M. C. HORNER-DEVINE AND B. J. M. BOHANNAN

Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Hasegawa, M., H. Kishino, and T. Yano. 1985. Dating of the human–ape splitting by a molecular clock of mitochondrial DNA. Journal of Molecular Evolution 22:160–174. Horner-Devine, M. C., K. M. Carney, and B. J. M. Bohannan. 2004a. An ecological perspective on bacterial biodiversity. Proceedings of the Royal Society of London B 271:113–122. Horner-Devine, M. C., M. Lage, J. B. Hughes, and B. J. M. Bohannan. 2004b. A taxa–area relationship for bacteria. Nature 432:750–753. Horner-Devine, M. C., M. A. Leibold, V. H. Smith, and B. J. M. Bohannan. 2003. Bacterial diversity patterns along a gradient of primary productivity. Ecology Letters 6:613–622. Jessup, C. M., R. Kassen, S. E. Forde, B. Kerr, A. Buckling, P. B. Rainey, and B. J. M. Bohannan. 2004. Big questions, small worlds: microbial model systems in ecology. Trends in Ecology and Evolution 19:189–197. Kembel, S. W., and S. P. Hubbell. 2006. The phylogenetic structure of a Neotropical forest tree community. Ecology 87: S86–S99. Knouft, J. H., J. B. Losos, R. E. Glor, and J. J. Kolbe. 2006. Phylogenetic analysis of the evolution of the niche in lizards of the Anolis sagrei group. Ecology 87:S29–S38. Kuittinen, H., A. Mattila, and O. Savolainen. 1997. Genetic variation at marker loci and in quantitative traits in natural populations of Arabidopsis thaliana. Heredity 79:144–152. Lerat, E., V. Daubin, and N. A. Moran. 2003. From gene trees to organismal phylogeny in prokaryotes: the case of the gamma-Proteobacteria. Public Library of Science 1:101–108. Losos, J. B., M. Leal, R. E. Glor, K. de Queiroz, P. E. Hertz, L. R. Schettino, A. C. Lara, T. R. Jackman, and A. Larson. 2003. Niche lability in the evolution of a Caribbean lizard community. Nature 424:542–545. Lovette, I. J., and W. M. Hochachka. 2006. Simultaneous effects of phylogenetic niche conservatism and competition on avian community structure. Ecology 87:S14–S28. Martin, A. P. 2002. Phylogenetic approaches for describing and comparing diversity of microbial communities. Applied and Environmental Microbiology 68:3673–3682. Martin, A. P., E. K. Costello, A. F. Meyer, D. R. Nemergut, and S. K. Schmidt. 2004. The rate and pattern of cladogenesis in microbes. Evolution 58:946–955. Morgan, K. K., J. Hicks, K. Spitze, L. Latta, M. E. Pfrender, C. S. Weaver, M. Ottone, and M. Lynch. 2001. Patterns of genetic architecture for life-history traits and molecular markers in a subdivided species. Evolution 55:1753–1761. Nubel, U., F. Garcia-Pichel, M. Kuhl, and G. Muyzer. 1999. Quantifying microbial diversity: morphotypes, 16S rRNA genes, and carotenoids of oxygenic phototrophs in microbial mats. Applied and Environmental Microbiology 65:422–430. Ochman, H., J. G. Lawrence, and E. A. Groisman. 2000. Lateral gene transfer and the nature of bacterial innovation. Nature 405:299–304. Posada, D., and K. A. Crandall. 1998. ModelTest: testing the model of DNA substitution. Bioinformatics 14:817–818. Prinzing, A., W. Durka, S. Klotz, and R. Brandl. 2001. The niche of higher plants: evidence for phylogenetic conservatism. Proceedings of the Royal Society of London B 268: 2383–2389.

Ecology Special Issue

Rice, N. H., E. Martinez-Meyer, and A. T. Peterson. 2003. Ecological niche differentiation in the Aphelocoma jays: a phylogenetic perspective. Biological Journal of the Linnean Society 80:369–383. Rocap, G., D. L. Distel, J. B. Waterbury, and S. W. Chisholm. 2002. Resolution of Prochlorococcus and Synechococcus ecotypes by using 16S–23S ribosomal DNA internal transcribed spacer sequences. Applied and Environmental Microbiology 68:1180–1191. Rotthauwe, J. H., K. P. Witzel, and W. Liesack. 1997. The ammonia monooxygenase structural gene amoA as a functional marker: molecular fine-scale analysis of natural ammonia-oxidizing populations. Applied and Environmental Microbiology 63:4704–4712. Shapleigh, J. P. 2000. The denitrifiers. In M. Dworkin, editor. The prokaryotes: an evolving electronic resource for the microbiological community. Springer-Verlag, New York, New York, USA. Silvertown, J., M. Dodd, and D. Gowing. 2001. Phylogeny and the niche structure of meadow plant communities. Journal of Ecology 89:428–435. Silvertown, J., M. Dodd, D. Gowing, C. Lawson, and K. McConway. 2006. Phylogeny and the hierarchical organization of plant diversity. Ecology 87:S39–S49. Sokal, R. R., and F. J. Rohlf. 1995. Biometry: the principles and practice of statistics in biological research. Third edition. W. H. Freeman and Company, New York, New York, USA. Stackebrandt, E., and B. M. Goebel. 1994. Taxonomic note: a place for DNA–DNA reassociation and 16S rRNA sequence analysis in the present species definition in bacteriology. International Journal of Systematic Bacteriology 44:846–849. Swofford, D. 2002. PAUP*: phylogenetic analysis using parsimony (*and other methods). Sinauer, Sunderland, Massachusetts, USA. Tofts, R., and J. Silvertown. 2000. A phylogenetic approach to community assembly from a local species pool. Proceedings of the Royal Society of London B 267:363–369. Treves, D. S., B. Xia, J. Zhou, and J. M. Tiedje. 2003. A twospecies test of the hypothesis that spatial isolation influences microbial diversity in soil. Microbial Ecology 45:20–28. Ward, D. M., M. J. Ferris, S. C. Nold, and M. M. Bateson. 1998. A natural view of microbial biodiversity within hot spring cyanobacterial mat communities. Microbiology and Molecular Biology Reviews 62:1353–1370. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, and S. W. Kembel. 2004. Phylocom: software for the analysis of community phylogenetic structure and character evolution. Version 3.22. hhttp:// www.phylodiversity.net/phylocomi Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Weiblen, G. D., C. O. Webb, V. Novotny, Y. Basset, and S. E. Miller. 2006. Phylogenetic dispersion of host use in a tropical insect herbivore community. Ecology 87:S62–S75. Zhou, J., B. Xia, D. S. Treves, L. Y. Wu, T. L. Marsh, R. V. O’Neill, A. V. Palumbo, and J. M. Tiedje. 2002. Spatial and resource factors influencing high microbial diversity in soil. Applied and Environmental Microbiology 68:326–334.

Ecology, 87(7) Supplement, 2006, pp. S109–S122 Ó 2006 by the Ecological Society of America

PHYLOGENETIC STRUCTURE OF FLORIDIAN PLANT COMMUNITIES DEPENDS ON TAXONOMIC AND SPATIAL SCALE JEANNINE CAVENDER-BARES,1 ADRIENNE KEEN,

AND

BRIANNA MILES

Department of Ecology, Evolution, and Behavior, University of Minnesota, St. Paul, Minnesota 55108 USA

Abstract. Consideration of the scale at which communities are defined both taxonomically and spatially can reconcile apparently contradictory results on the extent to which plants show phylogenetic niche conservatism. In plant communities in north central Florida, we collected species abundances in 55 0.1-ha plots in several state parks. When communities were defined narrowly to include a single phylogenetic lineage, such as Quercus, Pinus, or Ilex, neighbors tended to be less related than expected (phylogenetic overdispersion) or there was no pattern. If the same communities were defined more broadly, such as when all seed plants were included, neighbors tended to be more related than expected (phylogenetic clustering). These results provide evidence that species interactions among close relatives influence community structure, but they also show that niche conservatism is increasingly evident as communities are defined to include greater phylogenetic diversity. We also found that, as the spatial scale is increased to encompass greater environmental heterogeneity, niche conservatism emerges as the dominant pattern. We then examined patterns of trait evolution in relation to trait similarity within communities for 11 functional traits for a single phylogenetic lineage (Quercus) and for all woody plants. Among the oaks, convergent evolution of traits important for environmental filtering contributes to the observed pattern of phylogenetic overdispersion. At the broader taxonomic scale, traits tend to be conserved, giving rise to phylogenetic clustering. The shift from overdispersion to clustering can be explained by the increasing conservatism of traits at broader phylogenetic scales. Key words: environmental heterogeneity; Florida; Ilex; niche conservatism; overdispersion; phylogenetic structure of communities; Pinus; Quercus; taxonomic scale; trait convergence.

INTRODUCTION There is growing recognition that species evolve within communities and that community interactions influence the evolutionary process (Antonovics 1992, Neuhauser et al. 2003, Whitham et al. 2003). At the same time, we are becoming increasingly aware that evolutionary processes, particularly the way that traits evolve within lineages, influence species distributions and assembly in communities (McPeek 1996, Webb et al. 2002, Ackerly 2003, Chazdon et al. 2003, CavenderBares et al. 2004a). This study seeks to understand the role that trait evolution plays in determining the phylogenetic structure of communities and the extent to which phylogenetic structure depends on how communities are defined. Two processes are often considered as central to the assembly of communities: (1) filtering of species that can persist within a community on the basis of their tolerance of the abiotic environment (e.g., Weiher and Keddy 1995), and (2) competitive interactions among species that limit their long-term coexistence (Elton Manuscript received 26 January 2005; revised 17 August 2005; accepted 18 August 2005. Corresponding Editor: A. A. Agrawal. For reprints of this Special Issue, see footnote 1, p. S1. 1 E-mail: [email protected]

1946, MacArthur and Levins 1967, Chesson 1991, Leibold 1998). The two processes lead to opposite predictions about the phenotypic similarity and phylogenetic relatedness of co-occurring species (Tofts and Silvertown 2000, Webb et al. 2002). If closely related species share similar physiological limitations and exhibit evolutionary niche conservatism, environmental filtering will tend to cause closely related species to cooccur (phylogenetic clustering). In contrast, competitive exclusion should limit the coexistence of closely related species if species compete for the same limiting resources, leading to the opposite pattern of phylogenetic overdispersion. Both processes can operate simultaneously in real communities, but have greater influence at different scales. Keddy and Weiher (1999) hypothesized that limiting similarity should have greater importance at smaller spatial scales, whereas environmental filtering should predominate at larger spatial scales. Evidence supporting this view has been found in meadow communities in Great Britain (Silvertown et al. 2005). At the same time, the ecological process that appears to predominate might also depend on how broadly or narrowly communities are defined. For example, phylogenetic clustering was found among tree species in rainforest communities in Borneo (Webb 2000), as well as in herbaceous communities in Great Britain (Tofts

S109

S110

JEANNINE CAVENDER-BARES ET AL.

Ecology Special Issue

FIG. 1. (A) Phylogenetic structure of communities. An observed correlation (solid line) that is more negative than expected (dashed line) indicates phylogenetic clustering (left panel), because closely related species occur together more often than expected by chance; an observed correlation that is more positive than expected indicates phylogenetic overdispersion (right panel), because closely related species do not occur together. (B) Trait similarity in communities. An observed correlation that is more negative than expected indicates that co-occurring species show similar trait values (phenotypic clustering, left panel); an observed correlation that is more positive than expected indicates that trait values can be highly variable within communities (phenotypic overdispersion, right panel).

and Silvertown 2000). Both studies considered a large number of angiosperm lineages, providing support for the generalization that evolutionary stasis, or phylogenetic niche conservatism, is widespread among plant communities around the globe (Harvey and Pagel 1991, Eldridge 1995, Wen 1999, Webb et al. 2002, Ackerly 2003, Qian and Ricklefs 2004). In contrast, phylogenetic overdispersion has been found in narrowly defined communities that include a single phylogenetic lineage, including within Caribbean lizard communities (Losos et al. 2003) and among co-occurring oaks in north central Florida (Cavender-Bares et al. 2004a). In the latter study, evidence of convergence, rather than conservatism, in the evolution of species niches highlighted the prevalence of species interactions over niche conservatism in community assembly. These studies suggest that patterns of phylogenetic dispersion might vary systematically with taxonomic scale. The present study seeks to evaluate this possibility by examining the vegetation of Florida at various taxonomic and spatial scales. In addition, the study addresses the role that trait evolution plays in the phylogenetic structure of communities. Specifically, we make the following predictions: 1) The pattern of phylogenetic structure among communities depends on how a community is defined in terms of the taxa included. The more broadly a community is defined, the more likely it will be to show phylogenetic clustering as a result of trait conservatism

and environmental filtering. Narrowly defined communities, in contrast, are more likely to show phylogenetic overdispersion, either as a result of trait convergence, trait overdispersion, or both. 2) The phylogenetic structure of communities depends on the spatial scale of the analysis. As the spatial scale is increased to encompass greater environmental heterogeneity, species interactions should become less important, and phylogenetic clustering should emerge as the dominant pattern. 3) The relationship between trait evolution (conservative vs. convergent) and trait similarity within communities (clustered vs. overdispersed) should predict the phylogenetic structure of communities at different scales of analysis. MATERIALS

AND

METHODS

Tests of phylogenetic structure of communities We examined the phylogenetic structure of communities by comparing the degree of co-occurrence of species pairs in relation to the phylogenetic distance between them (Fig. 1A). Due to nonindependence and nonnormality of the data points, these correlations were compared to null models. Tests for phylogenetic clustering and overdispersion were conducted for a suite of different data sets in north central Florida and in the entire state of Florida that vary in spatial extent and

July 2006

SCALE AND FLORIDIAN PLANT COMMUNITIES

resolution of data. We compared the correlation coefficient of the relationship between co-occurrence and phylogenetic distance of species pairs to a null model in which either species distributions were permuted or phylogenetic relationships among species were randomized (Fig. 1A; Cavender-Bares and Wilczek 2003). The pairwise values of co-occurrence (C) were calculated based on proportional similarity (Schoener 1970) as follows: Cih ¼ 1 – 0.5 Rjpij – phjj, where Cih is the co-occurrence of species i and h, and pij is the proportion of total basal area or the proportion of occurrences of the ith species in the jth plot. We calculated phylogenetic distances from the estimated intervening branch length distances (measured in millions of years) between species pairs based on community phylogenies created for each test (see Phylogenetic analyses). Three null models were used. The first two generated a null distribution of expected species occurrence patterns against which to compare the observed data. These allow us to ask the following question: Given the evolutionary history of the taxa in the regional species pool, does phylogenetic relatedness influence the way that species have assembled in communities? In null model 1, basal area of species within plots was randomized by reshuffling raw data values 999 times across plots, but constraining the total basal area per species. Null model 2 used presence/absence data, instead of basal area, and constrained both total occurrences per species and total number of species per plot, using the sequential swap algorithm (Gotelli and Entsminger 2001b). The null model for the presence/ absence data may be the most biologically realistic, because it constrains both species abundances and plot diversity levels (Gotelli and Graves 1996). However, presence/absence data provide a lower degree of resolution for the distributions of species than basal area, because differences in relative abundance within plots are not considered. A null model that constrains total basal area per species and plot would probably be impossible to construct and was not attempted. Randomizations were carried out using Ecosim (Gotelli and Entsminger 2001a), and the co-occurrence vs. phylogenetic distance correlations were carried out in a self-written Visual Basic program, modified to accommodate large data sets from Cavender-Bares et al. (2004a). A third null model kept constant the distribution of species within communities, but randomized the phylogenetic tree topology. Null model 3 allows us to ask, given the distribution of species in communities, does the phylogenetic relatedness of species within those communities differ from random expectation? To randomize, we used the ‘‘random branch moves’’ algorithm in Mesquite (Maddison and Maddison 2000) and assigned the number of branch moves to equal the number of taxa in the tree. The total branch length distance from the basal node to the tips was kept constant. A distance matrix was then automatically

S111

computed from each of the 999 randomized trees, as well as from the original phylogeny, using a program in Visual Basic (J. Cavender-Bares, unpublished software). We used the result of this computation to calculate a null distribution of correlation coefficients between pairwise species co-occurrence and phylogenetic distance. In all analyses, a P value was calculated based on a two-tailed test. Since the null models use 999 randomizations plus the observed value, the minimum P value is 0.002 (Manly 1991). Trait similarity within communities We used a similar test to examine trait similarity within communities (Fig. 1B) by comparing the absolute value of pairwise differences in trait values to the degree of co-occurrence between species pairs. The observed correlation coefficients were tested against null model 2, in which the presence/absence of species within plots was randomized keeping row and column totals constant. The ranking of the observed correlation coefficient relative to that of the null model provides a way to order the measured traits in terms of their similarity within communities. The term ‘‘phenotypic clustering’’ refers to high trait similarity within communities, while the term ‘‘phenotypic overdispersion’’ refers to low trait similarity within communities (Fig. 1B) The spatial extent of communities In order to test the importance of (1) the spatial resolution of data, and (2) how communities are defined, we analyzed a series of community data sets that differ in the resolution of species abundances and in the community definition. First, we examined the influence of using basal area or presence/absence to evaluate species abundance in random 0.1-ha plots of north central Florida. Second, we examined the influence of the way community boundaries are defined by using these random 0.1-ha plots, where community boundaries are fixed by a standard area (hereafter referred to as the plot survey), or by using previously established community classifications and vegetation maps within state parks (hereafter referred to as community classifications). In the latter case, community boundaries in north central Florida were defined by environmental variation and the vegetation itself according to the Florida Natural Areas Inventory and the Florida Department of Natural Resources (FNAI and FDNR 1990). Previously determined community classifications for the entire state of Florida were also used, so that environmental variation among communities encompasses topographic and edaphic variation, as well as climatic variation. To examine the influence of the inclusion or exclusion of communities on phylogenetic structure, we performed tests of co-occurrence vs. phylogenetic distance for three individual genera (Quercus, Pinus, and Ilex), using community classification data sets for six state parks in north central Florida relative to two null models (both

S112

JEANNINE CAVENDER-BARES ET AL.

corresponding to null model 2). We created the first null model by permuting the presence of species across only those communities that contained a member of the genus (no community had zero as a matrix element). We created the second null model by permuting species’ presences across all communities that existed in the six parks examined, regardless of whether any members of the genus were found in them or not (a number of communities were represented by zeros in the matrix). This allowed us to test whether increasing the number of communities (effectively increasing the spatial scale and the degree of environmental heterogeneity) would alter the results. Inclusiveness of taxa within communities To examine the influence of phylogenetic scale, or the inclusiveness of taxa, on phylogenetic structure of communities, we carried out the same tests for the plot survey data to examine the phylogenetic structure of plant species within communities for (1) species within a monophyletic lineage (Quercus), (2) all angiosperms in the same guild (tree and shrub strata), (3) all plants in the tree and shrub strata, and (4) all recorded plants in all strata, including trees, shrubs, vines, epiphytes, and herbaceous species. These were tested against null models 1, 2, and 3, and against null models 2 and 3 only in the latter case, as only presence/absence data were available. We ran the same tests for the community classification data. The three most diverse woody plant genera within north central Florida, including the oaks (18 species), the hollies (seven species), and the pines (six species), were included to examine lineage-specific patterns in the context of their biogeographic and evolutionary history. Trait similarity within communities in relation to trait evolution For suites of life history and functional traits, we used the relationship between trait conservatism and trait similarity within communities to understand the phylogenetic structures of communities for the oaks only, as well as for all trees and shrubs in the plot survey. Trait similarity within communities was determined from the correlation of trait differences between species pairs and their degree of co-occurrence, relative to null model 1 (Fig. 1B). Trait conservatism (or phylogenetic signal) was calculated using methods developed by Ackerly in the ‘‘Analysis of Traits’’ module in Phylocom (Webb et al. 2004), based on trait differences between nodes, normalized by the standard deviation of the trait within a given lineage relative to a null model in which species are randomized across the phylogeny (Moles et al. 2005). We used the ranking of the observed phylogenetic signal relative to the simulations of the null model to order the traits by their degree of conservatism. We recognize that measures of trait conservatism are influenced by taxon sampling and might be biased in highly pruned phylogenies, such as

Ecology Special Issue

those used in the present study (Ackerly 2000). Nevertheless, examination of patterns of trait evolution with increasing taxonomic diversity is useful for understanding how trait conservatism shifts with phylogenetic scale in a single study system. Trait data collection We measured a suite of leaf traits on mature trees across the range of their local environmental distributions, collected from five state and city parks in north central Florida, including San Felasco Hammock State Preserve, Ichetucknee Springs State Park, Morning Side Nature Center, Payne’s Prairie State Preserve, and O’Leno State Park. For five individuals of each species, three sun and three shade leaves from each individual were measured for leaf area, scanned, and dried for leaf mass. We then used scanned leaf images to calculate perimeter (P), the perimeter-to-area ratio (P/A), which has been shown to be correlated with leaf hydraulic conductance (Sack et al. 2003), and lobedness (PL/A), determined as the perimeter-to-area ratio, scaled by leaf length (L). In addition, we took plant height, seed mass, and cotyledon type (photosynthetic or storage cotyledons), which has been shown to influence growth dynamics (Kitajima and Fenner 2000), from published (Kurz and Godfrey 1962, Mirov 1967, Schopmeyer 1974, Godfrey and Wooten 1981, Wenger 1983, Godfrey 1988) or online sources. We were able to collect leaf trait data for 90 of the 122 species examined in the plot survey; maximum height and leaf habit were obtained for 115 species, seed mass for 63 species, and cotyledon type for 87 species. It is useful to include as many different kinds of traits as possible that might contribute to ecological filtering and interactions among species. The present study, however, examined a limited number of easily measured traits. A suite of additional traits, including maximum hydraulic conductivity, vessel diameter, wood density, leaf longevity, and others were available from previous studies for the Quercus species (Cavender-Bares and Holbrook 2001, Cavender-Bares et al. 2004b), and these are presented in the oak analysis for comparison. Community data collection Plot survey data in north central Florida.—Quantitative vegetation data were collected from randomly located 20 3 50 m (0.1 ha) plots established in a previous study (Cavender-Bares et al. 2004a, b) in several state parks in north central Florida. Seventy four plots were originally established in 1998 for a study on oaks, and 55 of these original plots were resampled to determine the basal area of all woody species as well as the presence/absence of all plant species common enough within a plot to be conspicuous. Insufficient time and inability to relocate some of the plots prevented a complete resampling of all of the original plots. Within each plot, the diameter at breast

July 2006

SCALE AND FLORIDIAN PLANT COMMUNITIES

height of each tree over 1 m height was measured for calculation of basal area for a total of 122 tree and shrub species (see Appendix A). The presence/absence of all seed plants was also recorded for each of the plots (141 species; see Appendix A). The majority of the plots are located in three state parks, including San Felasco Hammock State Preserve (28 plots), a 2803-ha park in Alachua County; Ichetucknee Spring State Park (13 plots), a 921-ha park bridging Columbia and Suwanee counties and bisected by the Ichetucknee River; and Manatee Springs State Park (12 plots), a 960-ha park abutting the Suwanee River in Levy County. Two more plots were resampled at other sites in the region, including one at Morning Side Nature Center and one at Paynes Prairie Preserve State Park. An effort was made to sample across a range of the major woody plant community types in north central Florida. Community classification data in north central Florida.—We obtained species lists for each of six state parks in north central Florida and corresponding maps of natural community types from the District 2 Park Service in Gainesville, Florida, USA. Parks included San Felasco Hammock, Ichetucknee Springs, O’Leno State Park, Big Shoals State Park, Stephen Foster State Park, and Goldhead Branch State Park. In Florida, 63 terrestrial natural communities were identified and defined by the Florida Natural Areas Inventory and Florida Department of Natural Resources (FNAI and FDNR 1990), 29 of which occur in terrestrial areas of these state parks in north central Florida. Aquatic communities, which comprised nearly half of the defined communities, were excluded from the analyses because plant species were either poorly represented or poorly documented. A natural community is defined by the FNAI and FDNR (1990) as a ‘‘distinct and reoccurring assemblage of populations of plants, animals, fungi, and microbes naturally associated with each other and their physical environment.’’ In each park, we assigned species to designated communities, either by data directly from the Park Service or from a master list from the FNAI and FDNR (1990). If community assignments were not available from these sources, we determined them from two other sources (Kurz and Godfrey 1962, Godfrey 1988). Data were sufficient to include only the tree and shrub species (216 species). There were 29 possible community types, replicated according to whether they occurred in multiple state parks, giving a total of 75 sample communities used for the community classification analysis. Community classification data for the state of Florida.—A master species list with the community affiliations of these species was developed for the entire state of Florida from the FNAI and FDNR report. We ran two analyses: one for all listed plant species with identified community types (383 species), and a second for plants in the tree and shrub guild only (255 species) (see Appendix A). We included 50 terrestrial commun-

S113

ities in the analyses, including wetlands and coastal communities. Unlike the analyses for the parks in north central Florida, each community was represented only once, regardless of how commonly it occurs. Phylogenetic analyses Phylogenetic reconstruction.—Community phylogenies were created for each test using Phylomatic (Webb and Donoghue 2005). Phylomatic is an online application for creating a backbone phylogeny based on family and genera.2 The maximally resolved seed plant tree used for our trees relies on the online resource continually updated by P. F. Stevens, i.e., Angiosperm Phylogeny.3 Sources for tree construction down to the family level are extensively documented on this web site, although phylogenies created through the Phylomatic maximally resolved seed plant tree are not entirely resolved to the family level. We used a supertree of the angiosperms (Davies et al. 2004) to manually resolve all trees to the family level. Higher taxa were resolved according to published phylogenies (see Appendix B). If no published information was available to resolve polytomies, they remained unresolved. (See Appendix C for all reconstructed phylogenies.) It is important to recognize that the community phylogenies generated using this approach represent only approximations of true species relationships and should be refined as more data and other methods for constructing supertrees become available. The Quercus phylogeny was based on Cavender-Bares et al. (2004a) and was consistent with Manos et al. (1999). The four most parsimonious trees from that analysis were tested, as well as several less resolved phylogenies that collapsed nodes with low bootstrap support. We converted branch lengths to millions of years. A less resolved topology was used in the analyses presented here (see Appendix C), but the results were indistinguishable from those using the more resolved phylogenies (data not shown). We based the Ilex phylogeny on Cenoud (2000), and the Pinus phylogeny on Millar (1993), Schwilk and Ackerly (2001), Grotkopp et al. (2004), and Gernandt et al. (2005). Branch length estimation.—Branch lengths were based on minimum ages of nodes determined for genera, families, and higher orders from fossil data, and we extrapolated to higher order branches by spacing undated nodes in the tree evenly between dated nodes. This was done using an averaging algorithm in Phylocom (Webb et al. 2004) called ‘‘BLADJ’’ (Branch Length ADJustment, available online).4 Most of the node ages at the family level were taken from Wikstro¨m et al. (2001). At the genus level, we used additional sources, including Daghlian and Crepet (1983) for Quercus, based on the first appearance of the genus in 2 3 4

hhttp://www.phylodiversity.net/phylomatici hhttp://www.mobot.org/MOBOT/research/APwebi hhttp://www.phylodiversity.net/bladji

S114

JEANNINE CAVENDER-BARES ET AL.

Ecology Special Issue

TABLE 1. Tests for patterns of phylogenetic structure for community data sets that vary in how communities are defined. For each data set, the number of taxa, the number of communities, the data type (basal area or presence/absence), and the null model used in the analysis are given.

Data type and taxa included North central Florida Plot survey data (three parks, 55 plots) Quercus species Quercus species Quercus species Quercus species Angiosperm trees and shrubs Angiosperm trees and shrubs/no oaks All tree and shrub species All tree and shrub secies All tree and shrub species All tree and shrub species All angiosperms All angiosperms/no oaks All plant species All plant species

r Communities No. taxa No. communities Data type Null model (observed)

all all all all all all all all all all all all all all

17 17 17 17 113 96 122 122 122 122 130 113 141 141

55 55 55 55 55 55 55 55 55 55 55 55 55 55

basal area pres/abs basal area pres/abs basal area basal area basal area pres/abs basal area pres/abs pres/abs pres/abs pres/abs pres/abs

1 2 3 3 1 1 1 2 3 3 2 2 2 3

0.184 0.157 0.184 0.157
0.030
0.019
0.036
0.043
0.036
0.043
0.008
0.014
0.027
0.027

North central Florida Community classification data (six parks, 29 terrestrial community types) Quercus species all 18 Quercus species oak only 18 Quercus species all 18 Pinus species all 6 Pinus species pine only 6 Pinus species all 6 Ilex species all 7 Ilex species holly only 7 Ilex species all 7 Angiosperm trees and shrubs all 204 Angiosperm trees and shrubs all 204 All tree and shrub species all 216 All tree and shrub species all 216

75 38 75 75 29 75 75 35 75 75 75 75 75

pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs pres/abs

2 2 3 2 2 3 2 2 3 2 3 2 3

0.153 0.153 0.153
0.284
0.284
0.284 0.243 0.243 0.243
0.046
0.046
0.015
0.015

State of Florida Florida community classifications (50 terrestrial community types) All tree and shrub species all 255 All species all 383

50 50

pres/abs pres/abs

2 2


0.023
0.057

Notes: Observed and expected correlation coefficients (r) for the relationship between co-occurrence and phylogenetic distance of species pairs are shown. P values (two-tailed test) are determined from a null distribution of 999 randomizations plus the observed value.

the Americas, and Cenoud et al. (2000) for Ilex, based on the Eocene radiation of the genus, not the earliest appearance of extinct basal lineages. Higher level branch lengths, if available, were converted to millions of years, based on the fossil age for the deepest node. This method of branch length calculation provides only a first approximation of relative evolutionary distances between species. However, given that molecular data could not be used for branch length estimation, this method is an improvement over using only the tree topology itself. RESULTS Plot survey data Oak species: basal area vs. presence/absence.—Quercus species showed significant phylogenetic overdispersion when we compared pairwise co-occurrence values to phylogenetic distances and used null models 1 or 3 with basal area (Table 1, Fig. 2A). We obtained similar

results when using presence/absence data, rather than basal area, and applying null models 2 or 3, although the results were somewhat less statistically significant (Table 1). These results are very similar to previous analyses (Cavender-Bares et al. 2004a), despite reduced sampling and a less resolved phylogeny. Increasing taxonomic inclusiveness.—The effect of increasing the number of species in the community analysis was examined with data sets that included angiosperm trees and shrubs, all trees and shrubs, all recorded angiosperm species, and all recorded plants (Table 1). When angiosperm tree and shrub species were included, we found no pattern to the data. This result did not change significantly if we excluded oaks from the analysis. However, the observed correlation coefficient became more negative than 818 of the simulated r values, up from only 650 (Table 1), indicating a shift toward greater clustering. When examining all tree and shrub species (122 species), again using basal area and

July 2006

SCALE AND FLORIDIAN PLANT COMMUNITIES

TABLE 1. Extended.

r (expected)

Obs.Sim

0.014 0.004
0.024 0.019
0.025
0.006
0.009
0.010 0.005 0.009
0.009 0.008
0.014 0.005

P

Phylogenetic pattern

8 31 1 33 650 818 979 977 992 997 578 990 999 960

0.016 0.064 0.002 0.066 0.700 0.364 0.042 0.046 0.016 0.006 0.844 0.020 0.002 0.080

overdispersion weak overdispersion overdispersion weak overdispersion no pattern no pattern clustering clustering clustering clustering no pattern clustering clustering weak clustering

0.013 0.028
0.014
0.424
0.135 0.005
0.029
0.062 0.005 0.005 0.038 0.005 0.026

45 1 9 254 465 943 109 34 175 999 999 998 979

0.090 0.002 0.018 0.508 0.930 0.114 0.218 0.068 0.350 0.002 0.002 0.004 0.042

weak overdispersion overdispersion overdispersion no pattern no pattern no pattern no pattern weak overdispersion no pattern clustering clustering clustering clustering

0.021 0.005

999 999

0.002 0.002

clustering clustering

either null model 1 or 3, a clear pattern of phylogenetic clustering emerged. A nearly identical result was obtained when presence/absence and null models 2 or 3 were used. When all recorded plant species were examined (141 species), the pattern of phylogenetic clustering became even more significant using null model 1. All recorded angiosperm species (130 species) showed no pattern. However, upon removal of the 17 oak species from the analysis, the angiosperms were also significantly clustered. We obtained very comparable results for the two types of null models (community randomizations vs. phylogeny randomizations) (Table 1), although we did not attempt an exhaustive comparison. Community classification data In general, similar results were obtained when community classifications were used, rather than actual plot survey data. Based on the community classification, angiosperm trees and shrubs, as well as all trees and shrubs, showed significant phylogenetic clustering. Oaks still showed a pattern of phylogenetic overdispersion

S115

(Table 1, Fig. 2B). This pattern was marginally significant if all communities were used in the analysis and became highly significant if communities where oaks did not occur were excluded from the analysis. One difference that is apparent between the plot survey and the community classification is that closely related oak species show higher degrees of co-occurrence when the community classification is used (Fig. 2A, B). This is not surprising, given that defined community types necessarily include all species that regularly occur in them, and they do not take into account spatial distances among individuals or smaller scale environmental variation that might occur within community types. Despite this difference, the overall pattern of phylogenetic overdispersion in the oaks was clear in both analyses. The hollies, although much less diverse, showed a similar but only marginally significant pattern of overdispersion, based on null model 2 only (Fig. 2C), due in large part to the non-co-occurrence of pairs of close relatives, i.e., Ilex cassine and Ilex opaca, as well as Ilex ambigua and Ilex decidua. Note, however, that the overdispersion pattern was only apparent if community types that did not include any Ilex species were excluded from the analysis. The pattern was not significant using null model 3. The pines, a much older lineage, showed a somewhat different pattern (Fig. 2D). The two closest relatives, Pinus palustris and Pinus taeda (subsection Australes), do not co-occur, similar to the pattern in the other two genera. However, at the other extreme, distantly related pines also do not occur. Pinus clausa, which is in a different subsection of the genus (Contortae) from all of the other pines in the region (section Australes), does not co-occur with any other Pinus species. Extending the analysis to communities for the entire state of Florida, a clear pattern of phylogenetic clustering emerged (Table 1). At this scale, which includes coastal and subtropical communities, there is an increase in the number and variety of community types and the degree of environmental heterogeneity that is encompassed. Correspondingly, the number of taxa also increases. Acknowledging the incompleteness of the species list, the number of plant species for the entire state more than doubles that of the north central Florida communities. The large spatial scale and high environmental heterogeneity, along with the greater inclusiveness of taxa, leads to a highly significant pattern of phylogenetic clustering. Trait evolution and trait similarity within communities Analysis of trait evolution for the leaf data and life history attributes for the 120-species data set from the plot survey showed that all traits examined were significantly conserved (Fig. 3A). A subset of these traits also showed high similarity within communities (clustering), including specific leaf area for both sun and shade leaves, leaf habit, maximum height, leaf perimeter-to-area ratio, and cotyledon type.

S116

JEANNINE CAVENDER-BARES ET AL.

Ecology Special Issue

FIG. 2. Results of tests for phylogenetic overdispersion or clustering within three lineages in north central Florida. (A) There is a higher than expected correlation between pairwise co-occurrence (based on measurements of basal area of species within 55 0.10-ha plots) and phylogenetic distance (measured in millions of years [Ma], based on fossil records) for oak species. Closely related species tend not to co-occur, whereas species from different lineages do tend to co-occur. (B) A similar pattern emerges when the analysis is based on presence/absence of oaks in defined community types (based on community classification data) within six state parks in the same region. The significance of the depicted relationship is dependent on whether the null model only includes communities in which oaks occur, or whether all defined terrestrial communities found in the parks are included. (C) A similar, but less significant pattern is found for the less speciose Ilex genus. (D) The Pinus genus shows a different pattern. Note that in all graphs multiple data points may be superimposed.

In contrast, when the same traits were examined only for the oaks, many fewer traits showed conservatism (Fig. 3B). Conserved traits included leaf mass, leaf area, leaf habit, seed mass, and cotyledon type. None of these traits showed high similarity within communities. Other traits showed a lack of conservatism, which can also be interpreted as various degrees of convergence. Maximum height was the most convergent of the traits examined in this study. The high convergence results from the fact that in each of the major oak lineages, including the red oaks, the white oaks, and the live oaks, there exist both short and tall species. This trait also showed very high similarity within communities. A number of other traits were examined in a previous study (Cavender-Bares et al. 2004a) and are indicated with small, open circles in Fig. 3B. Relatively convergent traits from this analysis included maximum hydraulic conductivity, whole-shoot transpiration rate, vulnerability to cavitation during drought, absolute and relative growth rates, ability to resprout from rhizomes, and bark thickness. At the same time, traits that were fairly conserved tended to show overdispersion (or low

similarity within communities), such as seed mass and specific leaf area (particularly for shade leaves), or wood density, acorn maturation time, and leaf lifespan from the previous analysis. In both the present study and the previous analysis, there was a remarkable absence of conserved traits that showed clustering (or high similarity) within communities. This contrasts sharply with the pattern seen for all woody plants (Fig. 3A). For the oaks, trait similarity was highest for convergent traits and lowest for conserved traits. These results indicate that the pattern of phylogenetic overdispersion in oak communities is generated by the combination of environmental filtering of convergent traits and overdispersion of conserved traits. DISCUSSION It has long been recognized that plants can show a high degree of evolutionary stasis (e.g., Li 1952, Wen 1999, Qian and Ricklefs 2004) and niche conservatism (Webb et al. 2002, Ackerly 2003, Reich et al. 2003). At the same time, evolutionary processes that cause differentiation of sister taxa, such as character displacement

July 2006

SCALE AND FLORIDIAN PLANT COMMUNITIES

S117

FIG. 3. Conservatism of traits in relation to trait similarity within communities for (A) all woody species and (B) oak species only. Traits are ordered on the x- and y-axes by the ranking of the observed similarity or phylogenetic signal of each trait relative to 999 simulations in the null model (the ranking of the r values of observed data for trait similarity relative to simulated data increases with overdispersion [vs. clustering], and the ranking of the observed phylogenetic signal relative to simulated data decreases with conservatism [vs. convergence]). Traits include the following (labeled solid circles): leaf area, specific leaf area (SLA)sun, SLA-shade (SLA-sh), perimeter-to-area ratio (P/A), leaf habit, leaf mass, lobing, petiole length, cotyledon type, seed mass, and height. Open circles represent a suite of other morphological, physiological, and life history traits, measured from trees in the field as well as from seedlings in a common garden from a previous study (Cavender-Bares et al. 2004). The contrasting patterns of trait evolution and community similarity for the whole woody plant community and the oak lineage help explain the contrasting phylogenetic structures of communities observed at these different taxonomic scales.

(e.g., Schluter 2000) and adaptive radiation (e.g., Givnish et al. 2000) are well documented. To the extent that evolutionary stasis predominates, close relatives are likely to occur in similar habitats, given that plants will track environments for which they are adapted (Ackerly 2003). The principle of competitive exclusion presents a paradox, however, because close relatives that are too

similar cannot coexist (e.g., Elton 1946, MacArthur and Levins 1967, Chesson 2000). In this study, we suggest a solution for this paradox by showing that plant communities can simultaneously exhibit niche conservatism and overdispersion of close relatives. The signal that dominates depends on the scale at which communities and the taxa in them are examined.

S118

JEANNINE CAVENDER-BARES ET AL.

FIG. 4. The phylogenetic structure of communities is dependent on the scale at which they are defined. Apparent overdispersion within forest communities (A), for example, can appear as clustering when grassland and wetland communities are included (B). At the same time, phylogenetic structure also depends on the phylogenetic scale or taxonomic inclusiveness. Apparent overdispersion within a lineage is more likely to appear as a dominant pattern of clustering when the analysis includes many lineages.

Phylogenetic structure of communities at different spatial and taxonomic scales Both the spatial scale and resolution of species abundance data influence the observed phylogenetic structure in community assemblages. Greater overdispersion is detected at smaller spatial scales and with higher resolution data. This is demonstrated by more significant overdispersion when using basal area within plots rather than presence/absence (Table 1). Basal area provides higher resolution for species distributions, and the extent to which they overlap within communities, and hence could provide more precision for detecting overdispersion. The difference in null model is apparently not a factor in this outcome, as the same results were found using null model 3 (randomization of the phylogeny instead of the communities; Table 1). Phylogenetic overdispersion is more easily detected among close relatives when the communities included in the analysis are limited to only those in which the focal lineage occurs (Table 1). This effectively reduces the spatial area and environmental heterogeneity examined. These results make clear that phylogenetic overdispersion can occur at small spatial scales, even while

Ecology Special Issue

clustering occurs at larger spatial scales when a greater range of community types is included (Fig. 4). The phylogenetic structure of communities depends even more strikingly on the phylogenetic scale, or inclusiveness of taxa. Communities within a region may appear overdispersed at one phylogenetic scale (e.g., within one lineage), but clustered at a broader phylogenetic scale (e.g., when including all seed plants in the community). This allows a reconciliation of apparently contradictory results, as both niche conservatism and species interactions are important forces in community assembly, but are dominant at different scales. Species interactions that give rise to phylogenetic overdispersion show a stronger signal when communities are delimited to include a single monophyletic lineage. Niche conservatism shows a stronger signal the more broadly a community is defined taxonomically. Speciose lineages, such as the oaks, may be more likely to show overdispersion than less speciose lineages if, for example, increased diversity leads to increased competition among closely related species. Phylogenetic patterns are also likely to be lineagespecific, and it is not known the extent to which the oaks represent an exceptional scenario. Hence, the specific results found here might not be generalizable to other systems (see, for example, Kembel and Hubbell 2006). Nevertheless, the study does demonstrate the potential of phylogenetic patterns to vary predictably with scale and demonstrates that both phylogenetic overdispersion and clustering can occur at different spatial and taxonomic scales in the same study system. Phylogenetic structure within single lineage communities Phylogenetic dispersion patterns of species in communities that are narrowly defined taxonomically appear to be lineage specific, and may depend on intrinsic properties of lineages, biogeographic history, and lineage diversity. One possibility is that phylogenetic overdispersion might be more likely in speciose lineages due to intensification of species interactions. In addition, overdispersion might be more likely in lineages that have undergone adaptive radiations more or less in situ, as the evolutionary process should result in close relatives occupying different habitats as well as high local diversity. The American oaks are a likely example of such a radiation. Oaks are believed to have reached the Americas from Eurasia ;40 million years ago and are subsequently thought to have undergone a rapid radiation with all of the major subgenera appearing in the fossil record by 35 million years ago (Crepet and Nixon 1989). Many of the southeastern oaks are endemic to the region, and biogeographic evidence suggests that diversification of the oaks in the Americas occurred in their current localities (Manos et al. 1999, Manos and Stanford 2001). The Florida peninsula was not connected with North America and was submerged under the ocean until 20 million years ago (Webb 1990). The colonization of Florida, therefore, occurred subsequent to this time period. Fossil records indicate that

July 2006

SCALE AND FLORIDIAN PLANT COMMUNITIES

the flora was originally tropical, but increasingly invaded by deciduous hammock communities, similar to those of today, starting about 17 million years ago as the climate cooled. The availability of new habitats for colonization might have facilitated rapid adaptation, possibly aided by promiscuous exchange of genetic material among species. The present-day overdispersion of oaks (Table 1; Mohler 1990, Cavender-Bares et al. 2004a) could thus be the result of their evolutionary history of adaptive radiation into novel habitats. The hollies show some degree of overdispersion (Fig. 3C) under null model 2, although this pattern is not significant under null model 3, and hence the signal is somewhat ambiguous. Ilex diversified in the Eocene, ;50 million years ago and was already dispersed over all of the major continents by that time (Cuenoud et al. 2000). These small-seeded, bird-dispersed species arrived in Florida over the course of the past 20 million years, but for the most part evolved before then on other continents in other contexts. With the exception of two pairs, Ilex cassine and Ilex myrtifolia, and Ilex decidua and Ilex ambigua, speciation did not occur in the context of the other hollies that currently inhabit Florida, and currently sympatric congeners could have large intervening branch length distances (Cuenoud et al. 2000). The lack of cooccurrence between the closely related members in each pair gives rise to the apparent trend of overdispersion under null model 2 (Fig. 3C). The relatively weak signal, however, might result from a very different biogeographic history of the hollies or, alternatively, from the fact that the hollies are much less speciose than the oaks, particularly in this study region. The pines show a hump-shaped phylogenetic pattern, in which only pines of intermediate relatedness co-occur (Fig. 3D). It has previously been noted for the pines in Florida that sister species do not co-occur and that cooccurring species are not closely related (Adams and Jackson 1997). Hence, some degree of phylogenetic overdispersion has been observed within the genus, even though we did not detect it in north central Florida. The lack of clear phylogenetic signal in pine communities might result from the lower diversity of the pines relative to the oaks, causing congeneric competition to be less important among the pines. The strong signal of overdispersion in the oaks might result from their history of adaptive radiation and their high local diversity. Additional studies are needed to determine (1) the extent to which overdispersion among close relatives depends on historical and lineage-specific factors and (2) the extent to which local diversity within lineages might influence the intensity of species interactions that give rise to phylogenetic overdispersion in communities. Trait evolution and community assembly Nonneutral ecological processes that influence community assembly act on the phenotypes of species. Therefore, the phylogenetic structure of communities

S119

ultimately depends on (1) the evolutionary history of species’ traits and (2) the extent to which traits influence species distributions across environmental gradients or prevent coexistence due to species interactions. Closely related species are likely to co-occur if traits important for environmental filtering are conserved. On the other hand, if traits important for environmental filtering are convergent, or many different strategies for existing in a given habitat are possible, then closely related species might not co-occur. Likewise, if similarity in particular traits prevent coexistence, and such traits are conserved, then closely related species are again unlikely to co-occur. Of course, if species distributions are not related to their phenotypes, as predicted by neutral models (Hubbell 2001), then phylogenetic patterns in community structure are unlikely to emerge (Kembel and Hubbell 2006). However, lack of clear phylogenetic patterns might also result from species interactions and environmental filtering operating in opposing directions. In this study, for example, phylogenetic overdispersion of oaks apparently masks the pattern of phylogenetic clustering that predominates among angiosperm lineages. When oaks are removed from the analysis, the clustering pattern becomes apparent (Table 1). Among species in the broadly defined Floridian communities, all traits examined showed a fairly high degree of conservatism (Fig. 3A). A subset of these traits showed high similarity within communities (clustering), including specific leaf area for both sun and shade leaves, leaf habit, maximum height, leaf perimeter-toarea ratio, and cotyledon type. These traits have generally been shown to be important in the ability of species to respond to abiotic stress factors and to influence species distributions across environmental gradients, critical evidence for their potential role in environmental filtering. For example, specific leaf area and leaf habit, a proxy for leaf lifespan, are both important in the carbon economy of plants (e.g., Kikuzawa 1991, Damesin et al. 1998, Kikuzawa and Ackerly 1999) and have been well documented to vary with soil nutrient availability at various spatial scales (Monk 1966, Reich et al. 1999, Wright et al. 2002, Reich et al. 2003). Specific leaf area is also likely to be associated with canopy openness and soil moisture availability (Wright et al. 2002). Maximum height has been linked to growth rate (Thomas 1996), and taller height is a competitive strategy for accessing light in productive environments (Tilman 1988). In lower productivity environments or in communities prone to burning, less investment in aboveground biomass is expected (Schwilk and Ackerly 2001, Cavender-Bares et al. 2004b). Leaf perimeter-to-area ratio has been correlated to hydraulic conductance (Sack et al. 2003) and is likely to be important in distribution patterns across soil moisture gradients (Brodribb and Holbrook 2003). The primary role of cotyledons as either storage organs or photosynthetic organs represents a trade-off in regeneration strategies that depend on resource

S120

JEANNINE CAVENDER-BARES ET AL.

availability (Kitajima and Fenner 2000). The previously established linkages to resource capture and use for all of the traits that show high similarity (clustering) within communities support the interpretation that these traits are likely to be important for habitat tracking and environmental filtering. Clustering of conserved traits is critical in explaining the emergent pattern of phylogenetic clustering in Floridian plant communities. Other traits, including leaf mass, leaf area, lobedness, petiole length, and seed mass showed low similarity within communities. These traits might be less adaptive to environmental factors, or at least to those that vary at the same scale as the plot survey, and might be associated with architectural constraints. Alternatively, there could be multiple trait strategies that are successful in a given community. For example, leaf size and leaf lobing have been linked to boundary layer conductance and heat load (Givnish 1976), which could be important in environmental filtering. However, various combinations of leaf size, shape, leaf angle, leaf display, and pubescence can achieve a similar energy balance (Lambers et al. 1998, Ackerly 1999). Leaf size, seed size, and petiole length might also be more related to plant architecture than to external environmental gradients (e.g., Mazer 1989, Lord et al. 1995, Moles et al. 2005), and these traits have not generally shown strong trait–environment correlations. While all the traits measured in this study were fairly conserved in the broad analysis, many of the same traits tended to be convergent, or not different from random expectation, within the oak genus (Fig. 3B). For example, maximum height, the perimeter-to-area relationship, and specific leaf area were more convergent within the oaks than among the larger species pool. This result is not surprising if trait variance among lineages is higher than the variance within a lineage (Felsenstein 1985, Harvey and Pagel 1991). In other words, even if sister taxa show divergent morphology, recent common ancestry of closely related species is likely to limit the amount of divergence within a lineage, relative to that found among distantly related lineages. This could explain the shift to conservatism at broader taxonomic scales. Specific leaf area, a fairly labile trait within the oaks, for example, varied roughly threefold within the genus, but exceeded fivefold among all species sampled. However, in a study of almost 13 000 seed plants, Moles et al. (2005) showed that higher divergences sometimes occurred within a family than between families. A number of genera, particularly speciose genera in which many members co-occur, have also been shown to have divergences within them equal to or greater than those between co-occurring species that are distantly related (Silvertown et al. 2006). Therefore, the shift toward trait conservatism at the broader phylogenetic scale can also be explained as a result of swamping out the signal of high trait lability within the oaks by the addition of many more taxa that have conserved traits. Maximum height, for example, shows as much variation within the

Ecology Special Issue

oaks (;60-fold difference) as it does across all species examined at the broader scale, but this level of variation within a clade is unique to the oaks in our study. The shift in patterns of trait evolution toward increasing conservatism at broader taxonomic scales is likely to explain the concomitant shift in the phylogenetic structure of communities toward clustering. It also lends support to the view that niche conservatism is widespread among plants (Wen 1999, Webb et al. 2002, Ackerly 2003, Qian and Ricklefs 2004). The oaks, which dominate many woody communities in north central Florida, appear to represent a special case in this study system. They show an unusual amount of lability in certain functional traits, such as maximum height, water transport capacity, growth rate, and the ability to resprout from rhizomes (Cavender-Bares et al. 2004a), traits important for habitat specialization (Cavender-Bares et al. 2004b). The observed phylogenetic overdispersion among the Floridian oaks could be a result of their history of adaptive radiation and might be maintained by reduced competition among cooccurring species of different subgenera or lower density-dependent mortality (Cavender-Bares et al. 2004a). The latter possibility has been hypothesized as a result of pathogen specificity at taxonomic levels above the species (Janzen 1970, Webb and Gilbert 2006). At larger taxonomic and spatial scales, however, Floridian plant communities show phylogenetic clustering. This can be accounted for by conservatism of functional traits that influence species distributions. These contrasting patterns that emerge within the same study system illustrate the importance of scale in detecting opposing ecological and evolutionary forces. ACKNOWLEDGMENTS J. Cavender-Bares would like to thank Cam Webb and Jonothan Losos for the invitation to participate in this special issue. We thank Kelly McPhearson and Anne Barkdoll at the District 2 Park Service for providing species lists and community data maps for the six state parks in north central Florida. Ginger Morgan is acknowledged for facilitating permits. George Otto, Tony Davanzo, and Michael Stevens are gratefully acknowledged for resampling 32 of the permanent plots. Elizabeth Salisbury collected and sent to us all of the leaves for the leaf trait analysis, and Marjorie May Haight helped scan the leaves. We thank Doug Hornbeck, Sue Mauk, Kent Cavender-Bares, David Ackerly, Jason Teisinger, Hannah Nendick-Mason, Erik Smith, and Park Manager Randy Brown for logistical and other support or assistance with various aspects of the work in Florida. Anurag Agrawal and two anonymous reviewers are gratefully acknowledged for valuable comments on the manuscript. LITERATURE CITED Ackerly, D. 1999. Self-shading, carbon gain and leaf dynamics: a test of alternative optimality models. Oecologia 119:300– 310. Ackerly, D. D. 2000. Taxon sampling, correlated evolution, and independent contrasts. Evolution 54:1480–1492. Ackerly, D. D. 2003. Community assembly, niche conservatism and adaptive evolution in changing environments. International Journal of Plant Sciences 164:S165–S184.

July 2006

SCALE AND FLORIDIAN PLANT COMMUNITIES

Adams, D. C., and J. F. Jackson. 1997. A phylogenetic analysis of the southern pines (Pinus subsect. Australes Loudon): biogeographical and ecological implications. Proceedings of the Biological Society of Washington 110:681–692. Antonovics, J. 1992. Toward community genetics. Pages 426– 449 in R. S. Fritz and E. L. Simms, editors. Plant resistance to herbivores and pathogens: ecology, evolution, and genetics. University of Chicago Press, Chicago, Illinois, USA. Brodribb, T. J., and N. M. Holbrook. 2003. Changes in leaf hydraulic conductance during leaf shedding in seasonally dry tropical forest. New Phytologist 158:295–303. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004a. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Cavender-Bares, J., and N. M. Holbrook. 2001. Hydraulic properties and freezing-induced cavitation in sympatric evergreen and deciduous oaks with, contrasting habitats. Plant Cell and Environment 24:1243–1256. Cavender-Bares, J., K. Kitajima, and F. A. Bazzaz. 2004b. Multiple trait associations in relation to habitat differentiation among 17 Florida oak species. Ecological Monographs 74:635–662. Cavender-Bares, J., and A. Wilczek. 2003. Integrating microand macroevolutionary processes in community ecology. Ecology 84:592–597. Chazdon, R. L., S. Careaga, C. Webb, and O. Vargas. 2003. Community and phylogenetic structure of reproductive traits of woody species in wet tropical forests. Ecological Monographs 73:331–348. Chesson, P. L. 1991. A need for niches? Trends in Ecology and Evolution 6:26–28. Chesson, P. L. 2000. Mechanisms of maintenance of species diversity. Annual Review of Ecology and Systematics 31: 343–366. Crepet, W. L., and K. C. Nixon. 1989. Earliest megafossil evidence of Fagaceae: phylogenetic and biogeographic implications. American Journal of Botany 76:842–855. Cuenoud, P., M. A. DelPeroMartinez, P.-A. Loizeau, R. Spichiger, S. Andrews, and J.-F. Manen. 2000. Molecular phylogeny and biogeography of the genus Ilex L (Aquifoliaceae). Annals of Botany 85:111–222. Daghlian, C. P., and W. L. Crepet. 1983. Oak catkins, leaves and fruits from the Oligocene Catahoula formation and their evolutionary significance. American Journal of Botany 70: 639–649. Damesin, C., S. Rambal, and R. Joffre. 1998. Co-occurrence of trees with different leaf habit: a functional approach on Mediterranean oaks. Acta Oecologica-International Journal of Ecology 19:195–204. Davies, T. J., T. G. Barraclough, M. W. Chase, P. S. Soltis, D. E. Soltis, and V. Savolainen. 2004. Darwin’s abominable mystery: insights from a supertree of the angiosperms. Proceedings of the National Academy of Sciences (USA) 101:1904–1909. Eldridge, N. 1995. Reinventing Darwin: the great debate at the high table of evolutionary theory. Wiley, New York, New York, USA. Elton, C. 1946. Competition and the structure of ecological communities. Journal of Animal Ecology 15:54–68. Felsenstein, J. 1985. Phylogenies and the comparative method. American Naturalist 125:1–15. FNAI and FDNR [Florida Natural Areas Inventory and Florida Department of Natural Resources]. 1990. Guide to the natural communities of Florida. Florida Natural Areas Inventory and the Florida Department of Natural Resources, Tallahassee, Florida, USA Gernandt, D. S., G. G. Lopez, S. O. Garcia, and A. Liston. 2005. Phylogeny and classification of Pinus. Taxon 54:29–42. Givnish, T. 1976. Sizes and shapes of liane leaves. American Naturalist 110:743–778.

S121

Givnish, T. J., T. M. Evans, M. L. Zjhra, T. B. Patterson, P. E. Berry, and K. J. Sytsma. 2000. Molecular evolution, adaptive radiation, and geographic diversification in the amphiatlantic family Rapateaceae: evidence from ndhF sequences and morphology. Evolution 54:1915–1937. Godfrey, R. K. 1988. Trees, shrubs, and woody vines of northern Florida and adjacent Georgia and Alabama. University of Georgia Press, Athens, Georgia, USA. Godfrey, R. K., and J. W. Wooten. 1981. Aquatic and wetland plants of southeastern United States: dicotyledons. University of Georgia Press, Athens, Georgia, USA. Gotelli, N. J., and G. L. Entsminger. 2001a. EcoSim: null models software for ecology. Version 6.0. Acquired Intelligence and Kesey-Bear, Jericho, Vermont, USA. hhttp:// homepages.together.net/;gentsmin/ecosim.htmi Gotelli, N. J., and G. L. Entsminger. 2001b. Swap and fill algorithms in null model analysis: rethinking the knight’s tour. Oecologia 129:281–291. Gotelli, N. J., and G. R. Graves. 1996. Null models in ecology. Smithsonian Institution Press, Washington, D.C., USA. Grotkopp, E., M. Rejmanek, M. J. Sanderson, and T. L. Rost. 2004. Evolution of genome size in pines (Pinus) and its lifehistory correlates: supertree analyses. Evolution 58:1705– 1729. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Hubbell, S. 2001. The unified neutral theory of biodiversity and biogeography. Princeton University Press, Princeton, New Jersey, USA. Janzen, D. 1970. Herbivores and the numbers of tree species in tropical forests. American Naturalist 104:501–528. Keddy, P., and E. Weiher. 1999. Introduction: the scope and goals of research on assembly rules. Pages 1–22 in E. Weiher and P. Keddy, editor. Ecological assembly rules: perspectives, advances, retreats. Cambridge University Press, Cambridge, UK. Kembel, S. W., and S. P. Hubbell. 2006. The phylogenetic structure of a Neotropical forest tree community. Ecology 87: S86–S99. Kikuzawa, K. 1991. A cost–benefit analysis of leaf habit and leaf longevity of trees and their geographical patterns. American Naturalist 138:1250–1263. Kikuzawa, K., and D. D. Ackerly. 1999. Significance of leaf longevity in plants. Plant Species Biology 14:39–45. Kitajima, K., and M. Fenner. 2000. Ecology of seedling regeneration. Pages 331–360 in M. Fenner, editor. Seeds: the ecology of regeneration in plant communities. CAB International, Wallingford, UK. Kurz, H., and R. K. Godfrey. 1962. Trees of northern Florida. University of Florida, Gainesville, Florida, USA. Lambers, H., F. S. Chapin, III, and T. L. Pons. 1998. Plant physiological ecology. Springer, Berlin, Germany. Leibold, M. 1998. Similarity and local coexistence of species in regional biotas. Evolutionary Ecology 12:95–100. Li, H.-L. 1952. Floristic relationships between eastern Asia and eastern North America. Transactions of the American Philosophical Society 42:371–429. Lord, J., M. Westoby, and M. Leishman. 1995. Seed size and phylogeny in six temperate floras: constraints, niche conservatism, and adaptation. American Naturalist 146:349–364. Losos, J. B., M. Leal, R. E. Glor, K. de Queiroz, P. E. Hertz, L. R. Schettino, A. C. Lara, T. R. Jackman, and A. Larson. 2003. Niche lability in the evolution of a Caribbean lizard community. Nature 424:542–545. MacArthur, R., and R. Levins. 1967. The limiting similarity, convergence and divergence of coexisting species. American Naturalist 101:377–385. Maddison, W., and D. Maddison. 2000. Mesquite: a modular programming system for evolutionary analysis. University of Arizona, Tucson, Arizona, USA.

S122

JEANNINE CAVENDER-BARES ET AL.

Manly, B., editor. 1991. Randomization, bootstrap and Monte Carlo methods in biology. Second edition. Chapman and Hall, London, UK. Manos, P. S., J. J. Doyle, and K. C. Nixon. 1999. Phylogeny, biogeography, and processes of molecular differentiation in Quercus subgenus Quercus (Fagaceae). Molecular Phylogenetics and Evolution 12:333–349. Manos, P. S., and A. M. Stanford. 2001. The historical biogeography of Fagaceae: tracking the tertiary history of temperate and subtropical forests of the northern hemisphere. International Journal of Plant Sciences 162:S77–S93. Mazer, S. J. 1989. Ecological, taxonomic, and life history correlates of seed mass among Indiana Dune angiosperms. Ecological Monographs 59:153–175. McPeek, M. A. 1996. Linking local species interactions to rates of speciation in communities. Ecology 77:1355–1366. Millar, C. I. 1993. Impact of the Eocene on the evolution of Pinus L. Annals of the Missouri Botanical Garden 80:471– 498. Mirov, N. T. 1967. The genus Pinus. Ronald Press, New York, New York, USA. Mohler, C. L. 1990. Co-occurrence of oak subgenera: implications for niche differentiation. Bulletin of the Torrey Botanical Club 117:247–255. Moles, A. T., D. D. Ackerly, C. O. Webb, J. C. Tweddle, J. B. Dickie, and M. Westoby. 2005. A brief history of seed size. Science 305:576–580. Monk, C. D. 1966. An ecological significance of evergreenness. Ecology 47:504–505. Neuhauser, C., D. Andow, G. Heimpel, G. May, R. Shaw, and S. Wagenius. 2003. Community genetics: expanding the synthesis of ecology and genetics. Ecology 84:545–558. Qian, H., and R. E. Ricklefs. 2004. Geographical distribution and ecological conservatism of disjunct genera of vascular plants in eastern Asia and eastern North America. Journal of Ecology 92:253–265. Reich, P. B., D. S. Ellsworth, M. B. Walters, J. M. Vose, C. Gresham, J. C. Volin, and W. D. Bowman. 1999. Generality of leaf trait relationships: a test across six biomes. Ecology 80:1955–1969. Reich, P., I. Wright, J. Cavender-Bares, J. Craine, J. Oleksyn, M. Westoby, and M. Walters. 2003. The evolution of plant functional variation: traits, spectra and strategies. International Journal of Plant Sciences 164:S143–S164. Sack, L., P. D. Cowan, N. Jaikumar, and N. M. Holbrook. 2003. The ‘‘hydrology’’ of leaves: co-ordination of structure and function in temperate woody species. Plant Cell and Environment 26:1343–1356. Schluter, D. 2000. Ecological character displacement in adaptive radiation. American Naturalist. 156:S4–S16. Schoener, T. W. 1970. Nonsynchronous spatial overlap of lizards in patchy habitats. Ecology 51:408–418. Schopmeyer, C. S. 1974. Seeds of woody plants in the United States. USDA Agricultural Handbook 450. USDA, Washington, D.C., USA.

Ecology Special Issue

Schwilk, D. W., and D. D. Ackerly. 2001. Flammability and serotiny as strategies: correlated evolution in pines. Oikos 94: 326–336. Silvertown, J., M. Dodd, D. Gowing, C. Lawson, and K. McConway. 2006. Phylogeny and the hierarchical organization of plant diversity. Ecology 87:S39–S49. Thomas, S. C. 1996. Asymptotic height as a predictor of growth and allometric characteristics in Malaysian rain forest trees. American Journal of Botany 83:556–566. Tilman, D. 1988. Plant strategies and the dynamics and structure of plant communities. Princeton University Press, Princeton, New Jersey, USA. Tofts, R., and J. Silvertown. 2000. A phylogenetic approach to community assembly from a local species pool. Proceedings of the Royal Society of London B 267:363–369. Webb, C. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, and S. W. Kembel. 2004. Phylocom: software for the analysis of community phylogenetic structure and character evolution. Versions 3.19–3.22. hhttp://www.phylodiversity.net/phylocomi Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Webb, C. O., and M. J. Donoghue. 2005. Phylomatic: tree assembly for applied phylogenetics. Molecular Ecology Notes 5:181. Webb, C. O., G. S. Gilbert, and M. J. Donoghue. 2006. Phylodiversity-dependent seedling mortality, size structure, and disease in a Bornean rain forest. Ecology 87:S123–S131. Webb, S. D. 1990. Historical biogeography. Pages 70–100 in R. L. Myers and J. J. Ewel, editors. Ecosystems of Florida. University of Central Florida Press, Orlando, Florida, USA. Weiher, E., and P. Keddy. 1995. The assembly of experimental wetland plant communities. Oikos 73:323–335. Wen, J. 1999. Evolution of eastern Asian and eastern North American disjunct distributions in flowering plants. Annual Review of Ecology and Systematics 30:421–455. Wenger, K. F. 1983. Forestry handbook. Second edition. John Wiley and Sons, New York, New York, USA. Whitham, T. G., W. Young, G. D. Martinsen, C. A. Gehring, J. A. Schweitzer, S. M. Shuster, G. M. Wimp, D. G. Fischer, J. K. Bailey, R. L. Lindroth, S. Woolbright, and C. R. Kuske. 2003. Community genetics: a consequence of the extended phenotype. Ecology 84:559–573. Wikstro¨m, N., V. Savolainen, and M. W. Chase. 2001. Evolution of the angiosperms: calibrating the family tree. Proceedings of the Royal Society of London B 268:2211– 2220. Wright, I. J., M. Westoby, and P. B. Reich. 2002. Convergence towards higher leaf mass per area in dry and nutrient-poor habitats has different consequences for leaf life span. Journal of Ecology 90:534–543.

APPENDIX A Species used in each analysis (Ecological Archives E087-114-A1).

APPENDIX B References for phylogenetic data (Ecological Archives E087-114-A2).

APPENDIX C Phylogenetic trees used in each analysis (Ecological Archives E087-114-A3).

Ecology, 87(7) Supplement, 2006, pp. S123–S131 Ó 2006 by the Ecological Society of America

PHYLODIVERSITY-DEPENDENT SEEDLING MORTALITY, SIZE STRUCTURE, AND DISEASE IN A BORNEAN RAIN FOREST CAMPBELL O. WEBB,1,3 GREGORY S. GILBERT,2

AND

MICHAEL J. DONOGHUE1

1

Department of Ecology and Evolutionary Biology, Yale University, New Haven, Connecticut 06520 USA 2 Environmental Studies Department, University of California-Santa Cruz, California 95064 USA

Abstract. Density-dependent models that partition neighbors into conspecifics and heterospecifics ignore the great variation in effect of heterospecifics on focal plants. Both evolutionary theory and empirical results suggest that the negative effect of other plants on a focal plant should be higher for closely related neighbors than for less related neighbors. Using community-wide seedling mortality data from a forest where density dependence has previously been found, we searched for significant phylogenetic neighborhood effects (the ‘‘phylodiversity’’ neighborhood) on seedling (,50 cm tall) survival at various spatial scales. Logistic regression models were used, with 19-mo survival of individual seedlings as the response. We found a significant positive effect of nearest taxon phylodiversity on seedling survival at the 36-m2 scale and the 4-m2 scale, indicating that seedling survival is enhanced by being in a neighborhood where heterospecifics are not closely related. At all scales there was a strong negative effect of conspecific seedling density on focal survival, and at small scales there was also an effect of heterospecific density, indicating generalized competition. We place these results (for seedling dynamics over a relatively short period of time) in the context of changes in phylodiversity between different size classes of plants in the same forest, which integrate the effects of dynamics of all size classes over long time periods. At the 36-m2 scale, there was an increase in nearest taxon phylodiversity (i.e., a decrease in phylogenetic clustering) from the seedlings (,50 cm tall) to the poles (1–5 cm diameter), consistent with the positive effect of local phylodiversity on seedling survival. In contrast, there was a marked decrease in average phylodiversity from seedlings to saplings at the same scale. The trees in the 1600 m2 surrounding the seedling plots had much lower phylodiversity than either the seedlings or saplings. Taken together, these results suggest that (1) over short time and spatial scales, local seedling phylodiversity has a positive effect on seedling survival, possibly via interaction with pathogens (which we discuss in detail), but (2) over longer time periods and larger spatial scales the effect of abiotic-related mortality results in habitat filtering for phylogenetically conserved traits. Key words: Borneo; community phylogenetic structure; density dependence; phylodiversity; plant pathogen infection; seedlings.

INTRODUCTION In both theoretical and empirical studies, intraspecific negative density dependence has been shown to be important for slowing competitive exclusion and maintaining diversity in rain forest trees (Wright 2002). Most analyses focus on the statistical effects of conspecific density on either focal-plant performance or species demographic parameters (Martinez-Ramos et al. 1988, Condit et al. 1994, Harms et al. 2000). In some cases, the effect of distance from focal plant to conspecifics has been used as a correlate of plant density (Hubbell 1980, Connell et al. 1984, Hubbell et al. 1990, Gilbert et al. 1994). Manuscript received 26 January 2005; revised 18 July 2005; accepted 21 July 2005; final version received 9 September 2005. Corresponding Editor: A. A. Agrawal. For reprints of this Special Issue, see footnote 1, p. S1. 3 Present affiliation: Arnold Arboretum of Harvard University. E-mail: [email protected]

The strength of conspecific density (or interplant distance), relative to the effect of overall competitive pressure, can be measured by including the density (or distance) of heterospecifics as a separate factor in models (Connell et al. 1984, Uriarte et al. 2004). However, simply dividing species into conspecifics and heterospecifics obscures the great variation in effects of different species on any focal species (Pacala et al. 1996). One solution is to give separate ‘‘competition’’ parameters to each pairwise interaction (Canham et al. 2004, 2006). Unfortunately, a great deal of data is required to fit a model with so many free parameters. If the effect of one species on another was random with respect to the plants’ phenotypes, there would be no solution other than this multiparameter modeling. However, interspecific interactions are influenced by the anatomical and physiological similarity of the species involved, and gross similarity is not distributed at random among

S123

S124

CAMPBELL O. WEBB ET AL.

Ecology Special Issue

PLATE 1. The Air Putih River as it passes through granite hill forest at the Cabang Panti study area, Gunung Palung National Park, Indonesia. Photo Credit: C. Webb.

organisms; it is generally highest among taxa that share a recent ancestor (Felsenstein 1985, Harvey and Pagel 1991). In general, we expect greater negative interactions among individuals that are more phenotypically similar, be they competing for a similar vector of resources or negatively affecting each other via shared seed predators. Thus, we expect that closely related taxa should have a greater negative influence on each other than taxa that are more distantly related. Indeed, Uriarte and colleagues (2004) recently analyzed neighbor-dependent sapling growth at Barro Colorado Island and found that the negative effect of a neighbor on a target was greater when both plants were in the same taxonomic family. Adding a relatedness parameter to models of density dependence might therefore increase the predictive power of these models significantly, at the expense of just a few degrees of freedom. The development of phylogenetic methods over the past few decades has exposed the nonequivalence of taxa assigned to the same traditional rank. For example, in a recent supertree study of the relationships among angiosperms (Wikstro¨m et al. 2001), the estimated minimum age of family clades varied over an order of magnitude (e.g., 108 Ma for the Aristolochiaceae, 9 Ma

for the Rhizophoraceae). Additionally, higher taxonomic groups defined using morphological (e.g., floral) characters might have little ecological coherence; either more- or less-inclusive clades than named-rank clades might be more ecologically meaningful classes. With the great progress made recently in resolving the phylogenetic relationships among organisms, we can now move beyond ranks (such as family and genus) to use phylogenetic distances (e.g., age) among taxa. In this paper, we include such a phylogenetic distance factor in the analysis of community seedling mortality in lowland rain forest at Gunung Palung in Indonesian Borneo (see Plate 1; Webb and Peart 2000). At this site, densitydependent seedling mortality has previously been found both in single species (Webb and Peart 1999, Blundell and Peart 2004) and as a community-wide compensatory trend (Webb and Peart 1999). We take two approaches toward testing the contribution of the phylogenetic relatedness structure of neighbors (hereafter the neighborhood ‘‘phylodiversity’’; cf. Faith 1992) to seedling dynamics. First, we examine its effect on the survival of seedlings in small quadrats in multifactorial models that also include the effect of conspecific density and heterospecific density. The

July 2006

PHILODIVERSITY AND SEEDLING DYNAMICS

period chosen for the analysis is the same as that for which community-wide density dependence was found (Webb and Peart 1999). Second, we analyze changes in local phylogenetic structure with plant size class. Any influence of phylodiversity on seedling survival (a dynamic measure) should eventually leave an imprint on the phylogenetic structure of surviving plants (a static pattern). If, for example, mortality is higher in species that are sharing a neighborhood with closely related species than in species that are phylogenetically isolated from their neighbors, then this should be observed as an increase in phylodiversity (and a decrease in net relatedness) of plants of increasing size. Alternatively, if habitat filtering leads to strong phenotypic and phylogenetic attraction (Webb et al. 2002, CavenderBares et al. 2004), the plants in increasingly taller size classes should show decreased phylodiversity. We thus ask the following questions: (1) Does adding a phylogenetic neighborhood (phylodiversity) term increase the fit of density-driven models of seedling mortality? (2) Does the influence of neighborhood phylodiversity vary with neighborhood scale? (3) Are changes in the (static) phylogenetic structure with increasing size classes consistent with phylodiversitydependent seedling mortality? Finally, we develop a mechanistic hypothesis for the mediation of these phylodiversity effects by pathogens, the most likely causal agents of density dependence in these forests. METHODS In 1993, Webb set up 28 36-m2 seedling plots in lowland rain forest in the Gunung Palung National Park, West Kalimantan (Indonesian Borneo; site location: 1.21258 S, 110.10788 E; see Plate 1). The height of all woody plants was measured for plants taller than 5 cm and ,1.0 cm dbh (diameter at breast height [1.3 m high]), and dbh was measured for plants 1.0–5.0 cm dbh. Each seedling plot was nested inside a 40 3 40 m tree plot in which all trees .10.0 cm dbh were measured and identified (see Webb and Peart [1999, 2000] for further details). We constructed a single phylogeny that included all plant species occurring in any of the plots. We first ran the species list through the online tool Phylomatic (Chazdon et al. 2003, Webb and Donoghue 2005) to produce a tree topology based on the Angiosperm Phylogeny Group (APG) II backbone (APG 2003; P. F. Stevens, Angiosperm Phylogency Website, version 6, (available online),4 using the Phylomatic reference megatree R20040402. We then used the BLADJ algorithm of Phylocom (Webb et al. 2004, Moles et al. 2005) to constrain the internal nodes of the tree to the age estimates of Wikstro¨m et al. (2001). The algorithm then interpolates the other nodes of the tree for which 4

hhttp://www.mobot.org/MOBOT/research/APwebi

S125

direct age estimates are not available. Because many of the genera in the plot have yet to be sampled in a phylogenetic study, many of the families were modeled in the final supertree as a polytomy, and all the genera were polytomies. This means that the significant phylogenetic results reported here are not dependent on, for example, phylogenetic conservatism in pathogen host use only among sibling species. Instead, they must reflect patterns of conservatism that extend at least to the root of genera nodes, and possibly beyond. It is likely that some ecologically relevant characters are even conserved at family nodes and above. For the analysis of seedling dynamics, we used measures of survival of all individuals of all seedling species (5–50 cm tall) over a 27-month period from March 1994, in a subset of 12 of the plots (n ¼ 2296 seedlings total). This subset included only plots with completed seedling identification as of November 1994, and was a sample of all 28 plots stratified by elevation and subhabitat. Individual plants were mapped to a 0.25-m2 quadrat within the 36-m2 seedling plot, allowing analysis in nested square neighborhoods of 0.25, 1, 4, and 36 m2 (the design of the seedling plots included two 1-m walkways, which precluded using neighborhoods between 4 m2 and 36 m2; see plot diagram in Webb and Peart [1999]). We first counted the total seedling number per quadrat (for each sized quadrat: 0.25, 1, 4, and 36 m 2) at the beginning of the census period, and determined whether each individual was alive or dead at the end of 27 months. We also calculated the number of conspecific individuals and the relative phylodiversity of each species in the quadrat. Phylogenetic diversity (Faith 1992), or phylodiversity, is negatively related to the extent of phylogenetic clustering in a sample (Webb 2000, Webb et al. 2002, Cavender-Bares et al. 2004), and positively related to measures of phylogenetic distance among taxa in a sample. We measured relative phylodiversity in two ways, (1) using the mean phylogenetic distance (in units of millions of years) from the species of the focal individual to all other (n – 1) species in the quadrat, and (2) using the minimum phylogenetic distance to any heterospecific species in the quadrat (i.e., to the nearest taxon, or taxa, where several were equidistant). Both distances were then standardized by the mean expected phylogenetic distance, given the number of species in the quadrat, in order to correct for the effect of sample species richness; details of the same standardization for the indices of phylogenetic clustering, net relatedness index (NRI) and nearest taxon index (NTI), are given in Webb et al. (2002). We refer to the resulting metrics as relative average phylodiversity, APd 0 , and relative nearest taxon phylodiversity, NTPd 0 (the prime indicates that the metric is relative to the identity of focal species; absolute APd and NTPd are used in analysis of size class; and ‘‘Pd’’ is used to differentiate from Faith’s [1992] ‘‘PD,’’ which is measured differently). Phylodiversity is oriented here

S126

CAMPBELL O. WEBB ET AL.

so that larger positive values of phylodiversity indicate communities whose species are less closely related. We modeled survival of each seedling (categorical data: lived vs. died) using four separate logistic regressions: (1) survival as a function of the total number of seedlings in the same quadrat, (2) survival as a function of the number of conspecifics in the same quadrat and the number of heterospecifics in the same quadrat, (3) survival as a function of the number of conspecifics in the same quadrat, and (4) survival as a function of the number of conspecifics in the same quadrat, number of heterospecifics in the same quadrat, and relative phylodiversity of the community in the same quadrat. The total number of species in the quadrat was included in preliminary analyses, but was never a significant effect, and was dropped from final models. For the static size class analysis, we used identity and height for all species alive in June 1996 in the full set of 28 seedling and tree plots. Based on plant size at that time, we assigned plants to four size classes: seedlings (0– 50 cm tall), saplings (50 cm tall to 1 cm dbh), poles (1–5 cm dbh), and trees (.10 cm dbh). A total of 389 species were recorded as seedlings, 277 as saplings, 161 as poles, and 325 species as trees. We drew up a species list for each size class in each of the 28 plot locations, and this was passed through Phylocom (Webb et al. 2004) to calculate the net relatedness index and the nearest taxon index (Webb et al. 2002) of these species lists on the phylogeny of the whole pool of 548 species (see the Supplement). We recognize that not all taxa in the pool can occur in every size class (e.g., shrub and liana taxa will not be found in the tree plot samples), but we use a common phylogeny because measures of phylogenetic structure (NRI, NTI) are directly comparable only within a common pool phylogeny. We do not attempt to test whether the indices are different from zero (indicating either phylogenetic clustering or evenness; Webb 2000), because the null model of random sampling from the pool would be inappropriate. Because we are using estimates of absolute age between taxa, the effect of taxon sampling on phylogenetic distance is largely removed. The indices NRI and NTI were multiplied by negative one to obtain measures of phylodiversity (average, APd, and nearest taxon, NTPd, respectively), directly comparable to those used in the analysis we have described of seedling mortality. Median values of phylodiversity for the 28 plots were compared across size classes. We conducted all analyses for both subprojects using the statistical language R (R Project 2004) and Phylocom (Webb et al. 2004). RESULTS Phylodiversity and seedling survival A summary of the variation in numbers of individuals and species follows: for quadrats of 36, 4, 1, and 0.25 m2, respectively, the number of individuals per quadrat

Ecology Special Issue

(mean 6 SE) was 209 6 22, 48.4 6 4.4, 12.3 6 0.63, and 3.66 6 0.11; the number of species per quadrat was 48.0 6 3.5, 18.8 6 1.1, 7.46 6 0.29, and 2.88 6 0.072; the mean number of APG family clades per quadrat was 23.1 6 1.1, 12.5 6 0.52, 5.97 6 0.19, and 2.70 6 0.061; the mean number of individuals per species per quadrat was 4.36 6 0.41, 2.56 6 0.13, 1.65 6 0.045, and 1.27 6 0.018. A total of 20 models were fit (Table 1), and, in all models in which it was present as a factor, the (logarithm of the) number of conspecific individuals was the most predictive factor for seedling survival. At quadrat sizes of
4 m2, seedling survival was strongly negatively related to (the logarithm of) total seedling density. Heterospecific (log of) seedling density was negatively related to survival only at the smallest quadrat size (0.25 m2), at which scale the effect of total density was also strongest. Relative nearest taxon phylodiversity (NTPd 0 ) was positively associated with survival at the 36-m2 and 4-m2 scales. At both scales, the addition of the NTPd 0 phylodiversity term provided the best overall model fit, as measured by Akaike’s Information Criterion (AIC). Relative average phylodiversity (APd 0 ) was negatively correlated with seedling survival at the largest (36-m2) scale. Change in phylodiversity with increasing size class From seedlings to poles, average phylodiversity (APd) decreased and nearest taxon phylodiversity (NTPd) increased, while phylodiversity declined from seedlings to trees for both measures (Fig. 1). The influence of the four size classes on phylodiversity was significant overall (Kruskall-Wallis rank-sum test; APd, v2 ¼ 24.1, df ¼ 3, P ¼ 2.3 3 105; NTPd, v2 ¼ 34.7, df ¼ 3, P ¼ 1.4 3 107), and the following pairwise comparisons were significant using Wilcoxon rank-sum tests: APd, seedling vs. sapling (W ¼ 209, P ¼ 0.0023), seedling vs. pole (W ¼ 257, P ¼ 0.026), seedling vs. tree (W ¼ 130, P ¼ 6.5 3 106), pole vs. tree (W ¼ 195, P ¼ 0.0010); NTPd, seedling vs. tree (W ¼ 207, P ¼ 0.002), sapling vs. tree (W ¼ 131, P ¼ 7.2 3 106), pole vs. tree (W ¼ 31, P ¼ 9.2 31012). In a separate analysis, lists of seedling and sapling species were made for the 252 quadrats of 4 m2: both APd and NTPd decreased from seedlings to saplings (APd, v2 ¼ 7.17, df ¼ 1, P ¼ 0.0073; NTPd, v2 ¼ 10.3, df ¼ 1, P ¼ 0.0012). DISCUSSION Our analysis of community-dependent seedling survival suggests that at larger scales (4 m2 and 36 m2) there is a significant beneficial effect of local phylodiversity (NTPd): that is, the chance that a seedling survives increases if surrounding plants are not closely related to it, even after the effects of conspecific and heterospecific density have been removed. There was no such effect at smaller scales. At the same scale of 36 m2, we also observed an increase in phylodiversity from the seedling to the sapling to the pole size classes, when measured as

July 2006

PHILODIVERSITY AND SEEDLING DYNAMICS

S127

TABLE 1. Effect of local seedling density and phylodiversity on seedling survival, for four quadrat sizes. Five logistic models were fit for each of the four quadrat sizes: 0.25, 1, 4, and 36 m2. Significance of factor Log (total N)

Quadrat size and model

AIC

df

36 m2 Total density Partitioned density Conspecifics only Complete (APd 0 ) Complete (NTPd 0 )

3138 3114 3118 3103 3101

2295 2294 2295 2293 2293

NS

4 m2 Total density Partitioned density Conspecifics only Complete (APd 0 ) Complete (NTPd 0 )

3129 3114 3112 3113 3111

2295 2294 2295 2293 2293

0.21**

1 m2 Total density Partitioned density Conspecifics only Complete (APd 0 ) Complete (NTPd 0 )

3123 3109 3107 3110 3110

2291 2290 2291 2289 2291

0.20**

0.25 m2 Total density Partitioned density Conspecifics only Complete (APd 0 ) Complete (NTPd 0 )

2838 2834 2838 2835 2837

2093 2092 2093 2827 2091

0.34**

Log (conspecific N)

Log (heterospecific N)

0.11*** 0.13*** 0.15*** 0.085**

þ0.28* þ0.25* þ0.25*

0.17*** 0.18*** 0.19*** 0.14**

NS

0.23*** 0.24*** 0.25*** 0.21**

NS

0.31*** 0.33*** 0.33*** 0.33**

0.18*

NS

Phylodiversity (APd 0 )

0.25* þ0.22**

NS

þ0.13**

NS

NS

NS

NS

0.18* 0.17*

Phylodiversity (NTPd 0 )

NS

NS NS

Notes: Survival is measured as a positive response, so a negative parameter estimate indicates a negative relationship of the factor with seedling survival. Phylodiversity increases with decreasing relatedness of the focal taxon to the other taxa in the sample. Akaike’s Information Criterion (AIC) measures the complexity of the fitted model; a lower value indicates a better fit of the model to the data. Key to abbreviations: APd 0 , relative average phylodiversity; NTPd 0 , relative nearest taxon phylodiversity. *P ¼ 0.05; **P ¼ 0.01; ***P ¼ 0.001.

nearest taxon phylodiversity, NTPd. These two results are consistent with each other: if survival is higher for seedlings that are more distantly related, then as the cohort ages the overall net relatedness should decrease, as closely related taxa are ‘‘weeded out.’’ However, a contrasting pattern is observed in the measure of average phylodiversity: at the 36-m2 scale, there is an apparent detrimental effect of APd on seedling survival, and a consistent decrease in APd phylodiversity is observed in the static data, from seedlings to saplings, and from seedlings to poles. We also observed that for both NTPd and APd measures, there was a large decrease in phylodiversity from seedlings to trees, i.e., tree species were far more clustered phylogenetically than were seedling species. Previous work showed trees in a plot to be more phylogenetically clumped than expected by chance (Webb 2000). How can we reconcile these different results? When comparing different size classes for the same area, we must be aware that the different number of individuals in the different size classes, and thus the different expected number of species, could be leading to an artifactual change in metrics. While this remains a possibility, we feel it is not the cause of the patterns we observe, because the standardization of the indices uses the expected value of relatedness for a given n species (Webb et al. 2002),

which removes the main effect of sample species richness on phylogenetic relatedness. Additionally, there was no significant relationship of plot species richness with APd or NTPd within any size class (8 tests, n ¼ 28 plots). However, thorough simulation studies to explore the behavior of these metrics are desirable and are underway (S. Kembel, personal communication; N. Kraft, personal communication). Instead, we believe that the observed patterns result from the different aspects of phylogenetic structure captured by the two measures of phylodiversity. Imagine that the seeds arriving at a particular site are a random sample of all the plants in a forest, plants that occupy a number of habitats. Imagine also that some large clades of many taxa possess characters that will increase survival in this particular habitat, while other clades do not have these characters. Over time, there should be a net increase in average relatedness of the survivors at a site (i.e., a decrease in APd). However, if negative interactions are highest among taxa that are very closely related (sister species, or consectionals), there may simultaneously be a reduction over time in the phylogenetic distance to the most closely related surviving species (i.e., an increase in NTPd). All that is required for these two apparently opposite changes to occur simultaneously is that the characters for habitat filtering

S128

CAMPBELL O. WEBB ET AL.

Ecology Special Issue

become trees, nearest taxon phylodiversity (NTPd) must also eventually decrease, as the only survivors will belong to numerous ‘‘clumps’’ of related taxa. While this explanation surely oversimplifies the phylogenetic distribution of characters necessary for survival, we believe it helps explain the observed patterns. The large decrease of average and nearest taxon phylodiversity in trees is expected because of both the larger spatial scale of the tree plots and the length of time over which processes have operated as seedlings grow to trees. During this time, the effects of habitat filtering have dominated change in both measures of phylodiversity. We note that it is the combination of short-term dynamic data and static data reflecting long-term outcomes that provides these insights into community dynamics. A framework for the influence of pathogens on phylodiversity

FIG. 1. Change in phylodiversity with size class (seedlings, 0–50 cm tall; saplings, 50 cm tall to 1 cm dbh; poles, 1–5 cm dbh; trees, .10 cm dbh) for 28 plots (36 m2 for seedlings to poles; 0.16 ha for trees, hence separation by dashed vertical line); box indicates median (heavy line) and quartiles, whiskers reach to points
1.5 times the interquartile range. Units of both APd and NTPd are one standard deviation of a random distribution of phylogenetic distance (Pd), zeroed to the distribution’s mean. (a) Average phylodiversity (APd ¼ NRI), which becomes more positive with decreasing mean relatedness among all pairs of taxa in the sample. (b) Nearest taxon phylodiversity (NTPd ¼ –NTI, which is the mean of the relatedness to the most closely related taxon to each taxon). Abbreviations are as follows: NRI, net relatedness index; NTI, nearest taxon index.

(Webb et al. 2002, Cavender-Bares et al. 2004) must be plesiomorphic to a clade, whereas negative interactions must occur among the most closely-related taxa within the clade. For instance, if disease cross-species susceptibility is highest in sister taxa, while drought tolerance important for long-term survival on ridges is shared among many taxa in a phylogenetic family, we should see a simultaneous increase in nearest taxon phylodiversity (NTPd) and a decrease in ‘‘deeper,’’ average phylodiversity (APd). If only taxa with a particular synapomorphy can survive in a particular place to

As at other tropical rain forest sites (e.g., Augspurger 1984), available evidence points to pathogens being the primary cause of density-dependent mortality at Gunung Palung (Webb and Peart 1999). Pathogens have been shown to strongly influence plant community structure and diversity in a number of natural and agricultural plant systems (see review by Gilbert [2002]). To the extent that pathogens are more likely to crossinfect closely related hosts (Parker and Gilbert 2004), negative interactions between plant species should be strongest among very close relatives (Mack 1996). While the data presented in this paper are phenomenological, and do not point directly to pathogen influences, we use pathogens as an example in developing a framework that integrates the expected phylogenetic component of density dependence that we have outlined with a mechanistic, causal hypothesis. Invertebrate herbivores, with a species richness comparable to pathogens, might interact with plant phylodiversity in a similar way (Weiblen et al. 2006). Within a plant community, a plant pathogen species can have a single species of host plant (‘‘monophagy’’) or several (‘‘polyphagy’’). Monophagous pathogen–host dynamics have been well explored and tend to result in classic negative density-dependent effects in the plant (Burdon and Chilvers 1982). This will be the case either in a monoculture of the plant species or in a mixture with nonhost plant species. However, the population dynamics that result when pathogens can attack several plant species in a community are more complicated, and have been addressed mainly in the context of apparent competition (Price et al. 1988, Alexander and Holt 1998). Here, an increase in numbers of one plant species leads to an increase in pathogen inoculum, which could then affect other susceptible plant species in the community. In a tropical rain forest, the hundreds of plant species present are exposed to thousands of potential fungal pathogens, mostly unidentified (Arnold et al. 2000, Gilbert et al. 2002). Some of the pathogens are monophagous and some have many hosts (Lindblad

July 2006

PHILODIVERSITY AND SEEDLING DYNAMICS

S129

FIG. 2. Theoretical framework for the phylogenetic effects of pathogens. (a) Model of spread of single pathogen to other host plants (i þ 1 to n) in the community. The output vector of probabilities of cross-infection is a function of the phylogenetic relatedness of other taxa to the host plant (i), the degree of polyphagy in the pathogen, and the ‘‘ease’’ of host switching to unrelated host plants. (b) Model of community-wide cross-infection by all pathogens. The output is a matrix of cross-infection probabilities, declining from a high rate (H) in closely related taxa to a low rate (L) in unrelated taxa. The mean baseline rate (l) depends on the mean ‘‘ease’’ of pathogen host switching to unrelated taxa, both historically, and with contemporary rapid evolution in ecotypes. Note that the lower half of the matrix can differ from the upper half if the probability of infection from plant i to j is not equal to the probability from j to i.

2000, Woolhouse et al. 2001, Gilbert et al. 2002). Individual plant species might host dozens of species of fungal symbionts (Frohlich and Hyde 1999, Arnold et al. 2001), many with negative effects on the host (reviews in Gilbert 2002, 2005). Host susceptibility to pathogens will also be influenced in plastic ways by plant abundance, phenological state, season, nutrient availability, and other factors which stress a plant. Analyzing and understanding this complicated network of interactions and effects might be greatly simplified by modeling pathogen host breadth as a phylogenetic function (Webb et al. 2002; cf. Weiblen et al. 2006). The barriers for a pathogen to infect a host plant are morphological and biochemical. Moreover, because both morphology and plant secondary chemistry are often conserved phylogenetically (Farrell and

Mitter 1993, Farrell 2001), a pathogen that can infect one species can most easily become adapted to phylogenetically closely related species. At the same time, while phylogenetic conservatism of morphology and biochemistry is pervasive in nature, so is parallel evolution, or homoplasy, of similar characters in unrelated lineages. Hence, pathogen adaptations to infection barriers might also be effective in phylogenetically unrelated plants, which partially explains why some pathogens can infect many unrelated plants in a community (Weste and Marks 1987, Eckenwalder and Heath 2001). In general, the probability that a single pathogen can infect a particular set of hosts in a plant community should decline with increasing phylogenetic separation among the hosts, but with some host-sharing across some

S130

CAMPBELL O. WEBB ET AL.

apparently random plant species as well (Fig. 2a). Ecological association, through consistent community cooccurrence of unrelated but common species, might facilitate more distant host jumps. Considering the many pathogen species likely to be present, we expect variation in (1) the number of hosts that pathogens can attack, (2) the slope of their phylogenetically determined drop-off from closely related cohosts to less related nonhosts, and (3) the extent of nonphylogenetic host ‘‘jumps.’’ Combining the variation in all these pathogens, we expect that the net interspecific cross-infection probability will decline with increasing phylogenetic distance, down to a low baseline that represents the frequency of phylogenetically unrelated host ‘‘jumping’’ (Fig. 2b, matrix element L). In this way, pathogens, which are already known to exert strong influences on plant survival in natural systems, might operate in a phylodiversity-dependent manner and might underlay the phylodiversity-dependent seedling survival observed at Gunung Palung. The most powerful way to confirm the phylogenetic-distance-dependent signal in pathogen interactions will be to experimentally cross-inoculate pathogens in replicated samples of both closely and distantly related taxa and to record the success in transmitting disease symptoms. The overall influence of pathogens might be confirmed with pathogen exclusion (i.e., fungicide) experiments. Irrespective of their underlying mechanisms, phylodiversity-dependent dynamics have the potential to contribute to high overall species diversity, in a different way than other density-dependent phenomena. Singlespecies density dependence has the potential to cap the population size of species (Hubbell 1980) and, if present as a community compensatory trend across all species (Connell et al. 1984, Webb and Peart 1999, Wright 2002), will stabilize community composition and promote diversity by allowing rare immigrants to increase in number. Diversity-dependent dynamics, if found (Wills et al. 1997; but see critique in Wright [2002]), should increase overall survival in diverse sites and thus provide source or refuge populations that influence the metapopulation dynamics of species that would otherwise be poor competitors. None of these density and diversity dynamics reacts to or influences the specific identity of taxa: in theory, all taxa are considered to have equivalent influence or response. Phylodiversity-dependent dynamics, however, should operate differently, influencing the taxonomic (or rather phylogenetic) structure of the species composition of communities. The survival of species will be higher if they occur with unrelated taxa rather than with close relatives, and thus phylodiverse communities should be promoted, independently of their absolute species number, as found in this study. There has been much concern in the literature for the need to actively conserve sites and communities that have high phylogenetic diversity (e.g., Vane-Wright et al. 1991, Faith 1992, Moritz and Faith 1998). If phylodiversity-dependent dynamics operate in natural

Ecology Special Issue

systems, then the persistence and recovery of phylogenetic diversity might be assisted by nature itself. ACKNOWLEDGMENTS C. O. Webb thanks David Ackerly for informative discussions about phylogenetic structure and David Peart for many discussions on the nature of density dependence. Karen Garrett provided feedback in the development of the conceptual model. C. O. Webb was funded during analysis and writing by a grant from the National Science Foundation (DEB-0212873) and during the field work by a graduate fellowship (GER-9253849) and a Small Grants for Exploratory Research (SGER) grant to David Peart (DEB-9520889). We thank the various branches of the Indonesian government that permitted the fieldwork, and Darmawan and Suryadi, C. O. Webb’s field assistants at Gunung Palung. The review comments of Mark Westoby and an anonymous reviewer are much appreciated. LITERATURE CITED Alexander, H. M., and R. D. Holt. 1998. The interaction between plant competition and disease. Perspectives in Plant Ecology Evolution and Systematics 1:206–220. APG [Angiosperm Phylogeny Group]. 2003. An update of the Angiosperm Phylogeny Group classification for the orders and families of flowering plants: APG II. Botanical Journal of the Linnean Society 141:399–436. Arnold, A. E., Z. Maynard, and G. S. Gilbert. 2001. Fungal endophytes in dicotyledonous neotropical trees: patterns of abundance and diversity. Mycological Research 105:1502– 1507. Arnold, A. E., Z. Maynard, G. S. Gilbert, P. D. Coley, and T. A. Kursar. 2000. Are tropical fungal endophytes hyperdiverse? Ecology Letters 3:267–274. Augspurger, C. K. 1984. Seedling survival of tropical tree species: interactions of dispersal distance, light gaps, and pathogens. Ecology 65:1705–1712. Blundell, A. G., and D. R. Peart. 2004. Density-dependent population dynamics of a dominant rain forest canopy tree. Ecology 85:704–715. Burdon, J. J., and G. A. Chilvers. 1982. Host density as a factor in plant disease ecology. Annual Review of Ecology and Systematics 20:143–166. Canham, C. D., P. T. LePage, and K. D. Coates. 2004. A neighborhood analysis of canopy tree competition: effects of shading versus crowding. Canadian Journal of Forest Research 34:778–787. Canham, C. D., M. J. Papaik, M. Uriarte, W. H. McWilliams, J. C. Jenkins, and M. J. Twery. 2006. Neighborhood analyses of canopy tree competition along environmental gradients in New England forests. Ecological Applications 16:540–554. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Chazdon, R. L., S. Careaga, C. O. Webb, and O. Vargas. 2003. Community and phylogenetic structure of reproductive traits of woody specie in wet tropical forests. Ecological Monographs 73:331–348. Condit, R., S. P. Hubbell, and R. B. Foster. 1994. Density dependence in two understory tree species in a Neotropical forest. Ecology 75:671–680. Connell, J. H., J. G. Tracey, and L. J. Webb. 1984. Compensatory recruitment, growth, and mortality as factors maintaining rain forest tree diversity. Ecological Monographs 54:141–164. Eckenwalder, J. E., and M. C. Heath. 2001. The evolutionary significance of variation in infection behavior in two species of rust fungi on their hosts and related nonhost plant species. Canadian Journal of Botany 79:570–577.

July 2006

PHILODIVERSITY AND SEEDLING DYNAMICS

Faith, D. P. 1992. Conservation evaluation and phylogenetic diversity. Biological Conservation 61:1–10. Farrell, B. D. 2001. Evolutionary assembly of the milkweed fauna: cytochrome oxidase I and the age of Tetraopes beetles. Molecular Phylogenetics and Evolution 18:467–478. Farrell, B. D., and C. Mitter. 1993. Phylogenetic determinants of insect/plant community diversity. Pages 253–266 in R. Ricklefs and D. Schluter, editors. Species diversity in ecological communities: historical and geographical perspectives. University of Chicago Press, Chicago, Illinois, USA. Felsenstein, J. 1985. Phylogenies and the comparative method. American Naturalist 125:1–15. Frohlich, J., and K. D. Hyde. 1999. Biodiversity of palm fungi in the tropics: are global fungal diversity estimates realistic? Biodiversity and Conservation 8:977–1004. Gilbert, G. S. 2002. Evolutionary ecology of plant diseases in natural ecosystems. Annual Review of Phytopathology 40: 13–43. Gilbert, G. S. 2005. Dimensions of plant disease in tropical forests. Pages 141–164 in D. Burslem, M. Pinard, and S. Hartley, editors. Biotic interactions in the tropics: their role in the maintenance of species diversity. Cambridge University Press, Cambridge, UK. Gilbert, G. S., A. Ferrer, and J. Carranza. 2002. Polypore fungal diversity and host density in a moist tropical forest. Biodiversity and Conservation 11:947–957. Gilbert, G. S., S. P. Hubbell, and R. B. Foster. 1994. Density and distance-to-adult effects of a canker disease of trees in a moist tropical forest. Oecologia 98:100–108. Harms, K. E., S. J. Wright, O. Calderon, A. Hernandez, and E. A. Herre. 2000. Pervasive density-dependent recruitment enhances seedling diversity in a tropical forest. Nature 404: 493–495. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Hubbell, S. P. 1980. Seed predation and the coexistence of tree species in tropical forests. Oikos 35:214–229. Hubbell, S. P., R. Condit, and R. B. Foster. 1990. Presence and absence of density dependence in a Neotropical tree community. Philosophical Transactions of the Royal Society of London B 330:269–281. Lindblad, I. 2000. Host specificity of some wood-inhabiting fungi in a tropical forest. Mycologia 92:399–405. Mack, R. N. 1996. Biotic barriers to plant naturalization. Pages 39–46 in V. Moran and J. H. Hoffman, editors. Proceedings of the ninth international symposium on the biological control of weeds. University of Capetown, Stellenbosch, South Africa. Martinez-Ramos, M., J. K. Sarukhan, and D. D. Pinero. 1988. The demography of tropical trees in the context of forest gap dynamics: the case of Astrocaryum mexicanum at Los Tuxtlas tropical rain forest. Pages 293–313 in D. Davy, M. J. Hutchings, and A. R. Watkinson, editors. Plant population ecology. Blackwell, Oxford, UK. Moles, A. T., D. D. Ackerly, C. O. Webb, J. C. Tweddle, J. B. Dickie, and M. Westoby. 2005. A brief history of seed size. Science 307:576–580.

S131

Moritz, C., and D. P. Faith. 1998. Comparative phylogeography and the identification of genetically divergent areas for conservation. Molecular Ecology 7:419–429. Pacala, S. W., C. D. Canham, J. Saponara, and J. A. Silander. and Kobe. 1996. Forest models defined by field measurements: estimation, error analysis and dynamics. Ecological Monographs 66:1–43. Parker, I. M., and G. S. Gilbert. 2004. The evolutionary ecology of novel plant–pathogen interactions. Annual Review of Ecology, Evolution, and Systematics 35:675–700. Price, P. W., M. Westoby, and B. Rice. 1988. Parasite-mediated competition: some predictions and tests. American Naturalist 131:544–555. R Development Core Team. 2004. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Uriarte, M., R. Condit, C. D. Canham, and S. P. Hubbell. 2004. A spatially explicit model of sapling growth in a tropical forest: Does the identity of neighbours matter? Journal of Ecology 92:348–360. Vane-Wright, R. I., C. J. Humphries, and P. H. Williams. 1991. What to protect? Systematics and the agony of choice. Biological Conservation 55:235–254. Webb, C. O. 2000. Exploring the phylogenetic structure of ecological communities: an example for rain forest trees. American Naturalist 156:145–155. Webb, C. O., D. D. Ackerly, and S. W. Kembel. 2004. Phylocom: software for the analysis of community phylogenetic structure and character evolution. Version 3.22. hhttp:// www.phylodiversity.net/phylocomi Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475–505. Webb, C. O., and M. J. Donoghue. 2005. Phylomatic: tree assembly for applied phylogenetics. Molecular Ecology Notes 5:181–183. Webb, C. O., and D. R. Peart. 1999. Seedling density dependence promotes coexistence of Bornean rain forest trees. Ecology 80:2006–2017. Webb, C. O., and D. R. Peart. 2000. Habitat associations of trees and seedlings in a Bornean rain forest. Journal of Ecology 88:464–478. Weiblen, G. D., C. O. Webb, V. Novotny, Y. Basset, and S. E. Miller. 2006. Phylogenetic dispersion of host use in a tropical insect herbivore community. Ecology 87:S62–S75. Weste, G., and G. C. Marks. 1987. The biology of Phytophthora cinnamomi in Australasian forests. Annual Review of Phytopathology 25:207–229. Wikstro¨m, N., V. Savolainen, and M. W. Chase. 2001. Evolution of angiosperms: calibrating the family tree. Proceedings of the Royal Society of London B 268:2211–2220. Wills, C., R. Condit, R. B. Foster, and S. P. Hubbell. 1997. Strong density- and diversity-related effects help to maintain tree species diversity in a Neotropical forest. Proceedings of the National Academy of Sciences (USA) 94:1252–1257. Woolhouse, M. E. J., L. H. Taylor, and D. T. Haydon. 2001. Population biology of multihost pathogens. Science 292: 1109–1112. Wright, S. J. 2002. Plant diversity in tropical forests: a review of mechanisms of species coexistence. Oecologia 130:1–14.

SUPPLEMENT Supertree phylogeny files and taxonomic list (Ecological Archives E087-115-S1).

Ecology, 87(7) Supplement, 2006, pp. S132–S149 Ó 2006 by the Ecological Society of America

PLANT DEFENSE SYNDROMES ANURAG A. AGRAWAL1,3 1

AND

MARK FISHBEIN2,4

Department of Ecology and Evolutionary Biology and Department of Entomology, Cornell University, Ithaca, New York 14853 USA 2 Department of Biological Sciences, Mississippi State University, Mississippi State, Mississippi 39762 USA

Abstract. Given that a plant’s defensive strategy against herbivory is never likely to be a single trait, we develop the concept of plant defense syndromes, where association with specific ecological interactions can result in convergence on suites of covarying defensive traits. Defense syndromes can be studied within communities of diverse plant species as well as within clades of closely related species. In either case, theory predicts that plant defense traits can consistently covary across species, due to shared evolutionary ancestry or due to adaptive convergence. We examined potential defense syndromes in 24 species of milkweeds (Asclepias spp.) in a field experiment. Employing phylogenetically independent contrasts, we found few correlations between seven defensive traits, no bivariate trade-offs, and notable positive correlations between trichome density and latex production, and between C:N ratio and leaf toughness. We then used a hierarchical cluster analysis to produce a phenogram of defense trait similarity among the 24 species. This analysis revealed three distinct clusters of species. The defense syndromes of these species clusters are associated with either low nutritional quality or a balance of higher nutritional quality coupled with physical or chemical defenses. The phenogram based on defense traits was not congruent, however, with a molecular phylogeny of the group, suggesting convergence on defense syndromes. Finally, we examined the performance of monarch butterfly caterpillars on the 24 milkweed species in the field; monarch growth and survival did not differ on plants in the three syndromes, although multiple regression revealed that leaf trichomes and toughness significantly reduced caterpillar growth. The discovery of convergent plant defense syndromes can be used as a framework to ask questions about how abiotic environments, communities of herbivores, and biogeography are associated with particular defense strategies of plants. Key words: Asclepias; cardenolides; chemical ecology; cluster analysis; coevolution; Danaus plexippus; herbivory; latex; milkweed; monarch butterfly; phylogenetically independent contrasts; phytochemistry; plant–insect interactions.

INTRODUCTION Understanding the macroevolution of adaptive traits has inspired biologists for decades, yet has been challenging to study (Schluter 2000). The difficulty lies in (1) identifying the trait, or more commonly suites of traits, responsible for the adaptation of interest, (2) having adequate phylogenies to examine macroevolutionary patterns, (3) distinguishing apparent adaptation from ‘‘random’’ evolution along a diversifying phylogeny, and (4) matching the origins of adaptations to contemporaneous biotic and abiotic environmental factors that have likely driven adaptive changes. Despite these difficulties, the study of plant–herbivore interactions has contributed substantially to understanding the macroevolution of adaptive traits. Research on the macroevolution of plant defense has played a prominent Manuscript received 21 January 2005; revised 9 May 2005; accepted 21 June 2005; final version received 1 August 2005. Corresponding Editor (ad hoc): C. O. Webb. For reprints of this Special Issue, see footnote 1, p. S1. 3 E-mail: [email protected] 4 Present address: Department of Biology, Portland State University, Portland, Oregon 97207 USA.

role in the development of ideas on both the ecology and evolution of plants and insect herbivores, two of the most diverse lineages of eukaryotes (Ehrlich and Raven 1964, Coley 1983, Farrell et al. 1991, Farrell and Mitter 1993, 1998, Becerra 1997, Berenbaum 2001). In this paper, we present a new synthesis of ideas on the macroevolution of plant defense traits, with an attempt to identify the relative roles of phylogenetic history and ecological variables in shaping the expression of suites of defense traits within species. We then test some of our proposed ideas and assumptions utilizing new data on the phylogeny and defense of milkweeds (Asclepias spp.). Although it is convenient to consider plant defense as a single trait, plants typically utilize a broad arsenal of defensive traits against herbivores (Duffey and Stout 1996, Romeo et al. 1996). Even when a plant species is apparently defended by a single type or class of defense chemical, there are typically many specific forms of those compounds (Berenbaum et al. 1986, Malcolm 1991, Bennett and Wallsgrove 1994, Becerra 1997). Thus, it is more useful to think about plant defense as a suite of traits, which might include aspects of a plant’s

S132

July 2006

PLANT DEFENSE SYNDROMES

nutritional quality (e.g., proteins and antiproteins), physical characteristics (e.g., spines, trichomes, and leaf toughness), toxicity (e.g., cyanides and alkaloids), phenology, regrowth capacity (i.e., tolerance), and indirect defenses (e.g., volatiles and branching architecture). Synergistic interactions between multiple traits is particularly important in potentially providing a greater level of defense than would be possible if the traits were present independently (Broadway and Duffey 1988, Gunasena et al. 1988, Berenbaum et al. 1991, Stapley 1998). Nonetheless, most attention historically has considered defenses as singleton strategies, with the typical prediction that there should be trade-offs among different antiherbivore strategies (because they could be costly and/or redundant) (Steward and Keeler 1988, Herms and Mattson 1992). We argue that this reasoning, in its simplest form, is inaccurate, because plants do simultaneously employ multiple defense traits. Of course, particular plant defenses might trade off against each other, but this should not be the a priori expectation for any two defense-related traits. If plant defenses, like most adaptations, are composed of multiple traits, they might be organized into coadapted complexes (Dobzhansky 1970). The categorization of an organism’s phenotype into suites of potentially covarying traits has a long tradition in biology, including the concepts of guilds and syndromes (Root 1967, Grime 1977, Fægri and van der Pijl 1979, Simberloff and Dayan 1991, Chapin et al. 1993, Cunningham et al. 1999, Fenster et al. 2004, Wilson et al. 2004). The syndrome concept has been rightly criticized as overly simplistic when applied uncritically (e.g., Waser et al. 1996); however, suites of covarying traits can be usefully employed in some cases to infer adaptation to specific selective contexts. For example, suites of traits associated with hummingbirdpollinated flowers (i.e., red color, tubular morphology, extended anther position, dilute nectar concentration, etc.) have repeatedly evolved from those of beepollinated ancestors in diverse lineages (McDade 1992, Beardsley et al. 2003, Castellanos et al. 2003, 2004, Fenster et al. 2004). Patterns of natural selection imposed by birds and bees implicate these pollinators as the agents favoring floral syndromes (e.g., Schemske and Bradshaw 1999). Although any two traits in a syndrome can be positively or negatively (or not at all) correlated across taxa, the plant defense syndrome hypothesis rejects the prediction that any two plant defenses are redundant, and thus should be negatively associated across species (Steward and Keeler 1988, Twigg and Socha 1996, Rudgers et al. 2004). Defense syndromes as whole, however, can trade off if they truly represent alternative adaptive strategies. If a syndrome has independently evolved multiple times, it suggests that the selective forces driving the evolution of these convergent adaptations are common and widespread. If, however, closely related species

S133

share a syndrome due to common ancestry, then common ancestry is sufficient to explain the association, and adaptive advantage need not necessarily be invoked. Like trade-offs among supposedly redundant traits, the correlated evolution of traits composing syndromes needs to be tested explicitly rather than assumed. Although plant defense syndromes have received little attention in the past (but see Feeny 1976, Kursar and Coley 2003), we suggest that where distantly related plant species share a common assemblage of herbivores, they are likely to defend themselves with similar strategies. For example, plant species attacked primarily by vertebrate grazers should employ quite different strategies (e.g., spines, leaf toughness, and out-of-reach morphologies) than plants primarily attacked by caterpillars (e.g., trichomes, toxins, and parasitoid-attracting volatiles). Perhaps the original ‘‘plant defense syndrome’’ was the set of traits proposed to be associated with highly apparent plants (Feeny 1976). Feeny argued that apparent plants like oak trees were bound to be found, and thus defended with suites of traits that make them nutritionally poor: high concentrations of tannins, low water and nitrogen content, and tough leaves. We agree that such patterns are likely to have evolved repeatedly to converge on distinct syndromes of defensive traits, even in distantly related plant taxa. So why have most ecologists not explicitly considered plant defense syndromes? We believe that the traits of importance in defense against herbivores are too often unobservable by the naked eye of naturalists, and therefore have escaped attention. Plant defense syndromes in communities versus taxonomic groups The concept that plant species within a particular community may coexist due to trade-offs in fitnessenhancing traits is an old and central paradigm in plant ecology (reviewed by Tilman and Pacala [1993]). This logic was later applied to plant defense traits for coexisting species. Van der Meijden et al. (1988) proposed that suites of traits associated with resistance vs. regrowth (tolerance) might represent alternative strategies for coexisting temperate herbaceous plants. Spurred by Janzen’s (1974) study of tropical blackwater rivers and emerging physiological plant defense theories (Bryant et al. 1983), Coley and colleagues hypothesized that plant species colonizing light gaps would have divergent (trading-off) suites of defense traits, compared to species colonizing the more resource-poor understory in tropical forests (Coley 1983, 1987, Coley et al. 1985). Such trade-offs explained divergent patterns of herbivory in different microenvironments and also likely contributed to the maintenance of diversity at larger spatial scales. Recent experimental work on phylogenetically paired species from white sand and clay soil habitats of Peru support the hypothesis (Fine et al. 2004, 2006).

S134

ANURAG A. AGRAWAL AND MARK FISHBEIN

Kursar and Coley (2003) have also expanded these ideas to explicitly consider the evolution of convergent defense syndromes in tropical trees. They argue that trees fall along an escape–defense continuum: extreme ‘‘escape’’ species are predicted to have few chemical defenses, but rapid synchronous leaf expansion, and low leaf nutritional quality during expansion; extreme ‘‘defense’’ species have high chemical defense, low nutritional quality, and asynchronous leaf expansion (Kursar and Coley 2003). Using phylogenetically independent contrasts, Silvertown and Dodd (1996) showed that herbaceous vs. woody plants had distinct types of chemical defenses (tannins vs. alkaloids), consistent with classic apparency theory (Feeny 1976). In each of these examples, the general finding has been that unrelated plant species within a particular community have converged on a suite of similar strategies that maximize fitness, given a particular set of ecological interactions. Thus within a regional community, plant species can converge on a few defense syndromes, yet the divergent strategies (across the syndromes) also can promote the coexistence of species. The approach of examining syndromes within communities is an ecological question, first and foremost, and begins with identifying the traits of species that occur within a plant community irrespective of their evolutionary relationships. If patterns of defensive traits consistently occur among coexisting taxa, then the patterns of defense among species could be explained by either shared phylogenetic history or convergence due to similar selective agents causing repeated evolution of a defensive syndrome. We propose that phylogenetic history should provide the null hypothesis to explain the distribution of defense traits among species (Silvertown and Dodd 1996). The alternative hypothesis is that species’ defensive traits are evolutionarily labile, and species have repeatedly evolved homoplasious (i.e., independent, but similar) solutions to biotic and abiotic environmental challenges. For instance, in a temperate grassland community, patterns of plant defense could consistently fall into a few categories (e.g., toxic and antdefended vs. tolerant and poor nutritional quality), but each of these categories might be phylogenetically homologous (e.g., evolving once each in legumes and grasses). Alternatively, unrelated taxa could converge on homoplasious suites of traits in response to the same or similar selective forces (Fine et al. 2004). In community studies of syndromes, phylogenetic information need not be extremely detailed, and taxonomic information might be sufficient. If two species within a genus have different suites of defense traits and are ecologically divergent (i.e., have different types of species interactions, live in different microhabitats, etc), then selection is a reasonable hypothesis for the differences between the close relatives; if multiple, phylogenetically independent congeneric pairs show similar suites of defensive traits and divergent ecologies, convergence can be invoked (Conway-Morris 2003, Fine

Ecology Special Issue

et al. 2004). Of course, such an association does not prove adaptation, but, at minimum, implies an evolutionary association between traits and ecology. The macroevolution of syndromes has also been studied within clades. Within the plant genus Penstemon, species have repeatedly evolved suites of traits associated with hummingbird pollination from Hymenopterapollinated ancestors (Thomson et al. 2000, Wilson et al. 2004). The repeated evolution of particular plant defense trait combinations within a taxonomic group (e.g., genus) is suggestive of ‘‘syndromes in clades.’’ Traits associated with induced resistance to herbivores evolved multiple times in the cotton genus, Gosspium (Thaler and Karban 1997). Work with ant-associated and nonant-associated Acacia spp. may reveal patterns of divergent defensive strategies in plants with altered species interactions (Rehr et al. 1973, Seigler and Ebinger 1987, Heil et al. 2000, 2002, 2004). Recent work by Becerra and colleagues (Becerra 1997, Becerra et al. 2001) demonstrates that there has been convergent evolution of terpene chemical defense and ‘‘squirt-gun’’ defense combinations in tropical Bursera spp. Species of specialist Blepharida beetles typically consume plants with similar defense traits, which are not necessarily the most closely related species or those with geographically similar ranges (Becerra 1997, Becerra and Venable 1999). Thus, traits can covary within a clade and are predicted to be associated with particular interactions (although convergent taxa do not necessarily coexist). The approach of examining syndromes within a clade requires an accurate plant phylogeny and the existence of two or more syndromes (suites of traits). Thus, if suites of traits converge in different parts of the phylogeny, the question becomes this: Are there consistent ecological correlates that match these different syndromes? A key difference from the community approach is that syndromes within a closely related set of species in a clade can surface in geographically separated species that do not naturally coexist. The community approach specifically starts with species that coexist under relatively similar ecological conditions. Although similar in trying to disentangle the effects of phylogenetic history and ecology, the ‘‘community’’ and ‘‘clade’’ approaches start at the opposing ends of the phylogeny–ecology continuum. Defense syndromes in milkweeds (Asclepias) There are ;130 species of Asclepias native to North America, including Mesoamerica and the Caribbean (Woodson 1954; M. Fishbein et al., unpublished manuscript). Approximately six additional species are native to temperate South America (Bollwinkel 1969). The situation is more complicated in Africa, where up to 200 species have been or could potentially be included in the genus Asclepias, depending on the breadth of circumscription and phylogenetic relationships among African and American species (Fishbein 1996). However, preliminary analyses suggest that American and African

July 2006

PLANT DEFENSE SYNDROMES

S135

TABLE 1. Species sampled for defense-related traits, effects on herbivore performance, and phylogenetic relationships. Species

Native Range

Asclepias amplexicaulis Sm. Asclepias asperula (Decne.) Woodson ssp. asperula Asclepias californica Greene Asclepias cordifolia (Benth.) Jeps. Asclepias curassavica L. Asclepias eriocarpa Benth. Asclepias exaltata L. Asclepias fascicularis Decne. Asclepias hallii A. Gray Asclepias hirtella (Pennell) Woodson Asclepias incarnata L. ssp. incarnata Asclepias incarnata L. ssp. pulchra (Ehrh. ex Willd.) Woodson Asclepias oenotheroides Schltdl. & Cham. Asclepias perennis Walter Asclepias purpurascens L. Asclepias speciosa Torr. Asclepias sullivantii Engelm. ex A. Gray Asclepias syriaca L. Asclepias tuberosa L. Asclepias variegata L. Asclepias verticillata L. Asclepias viridis Walter Gomphocarpus cancellatus (Burm. f.) Bruyns Gomphocarpus fruticosus (L.) W.T. Aiton

Eastern USA Southwestern USA and Mexico Western USA Western USA Neotropics Western USA Eastern USA Western USA and Mexico Western USA Central USA Eastern USA Eastern USA Southwestern USA, Mexico, and Central America Southeastern USA Eastern USA Western USA Central USA Eastern USA and Canada Eastern USA, southwestern USA, and Mexico Eastern USA Eastern USA and Canada Eastern USA Southern Africa Africa

species belong to distinct sister clades (Rapini et al. 2003; M. Fishbein et al., unpublished manuscript), and thus Asclepias may be considered an exclusively American genus of ;150 species. In this study, we employ the results of the most intensively sampled study of Asclepias to date (M. Fishbein et al., unpublished manuscript), focusing on the American species as a framework for investigating macroevolution of defense traits. Of the species we studied, all are perennial, some are clonal, and genets are probably very long-lived (Table 1). In addition, most of the species are relatively rare in the landscape; those that are common now, such as A. syriaca in eastern North America, were probably much less abundant and widespread prior to the clearing of the eastern deciduous forest by colonists. Plants in the genus Asclepias have been of major importance in the development of theories about plant– herbivore and tri-trophic interactions, and historically the focus has been primarily on cardenolides as a plant defense and the Monarch butterfly as the herbivore (Fig. 1) (Brower et al. 1967, 1972, Malcolm et al. 1989). However, the herbivore community of Asclepias spp. consists of tens of species that are principally hostspecific insects (Weiss and Dickerson 1921, Rothschild et al. 1970, Wilbur 1976, Blakley and Dingle 1978, Price and Willson 1979, Morse 1985, Malcolm 1991, Fordyce and Malcolm 2000, Agrawal and Malcolm 2002). These herbivores fill almost every conceivable feeding guild: aphids feed on the phloem, beetles and caterpillars chew the leaves, flies mine the leaves, hemipterans eat the seeds, weevils bore through the stem and eat the pith, and beetle larvae bore through the roots. For example, the common milkweed, A. syriaca, has 12 reported insect herbivores, including members of all

of the guilds. Some of the herbivores, such as Monarchs (Danaus plexippus) and Milkweed aphids (Aphis nerii), are broadly distributed and feed on many milkweeds. Other groups, such as the cerambycid beetles in the genus Tetraopes have radiated with the milkweed genus, and each of the 24 Tetraopes species is primarily associated with a single milkweed species (Farrell and Mitter 1998). Thus, each milkweed species is likely attacked by several herbivores, although the species composition and guilds attacking particular species vary widely. An important future goal in the study of plant defense macroevolution is to detail the particular insect communities that correspondingly attack plants with different defense syndromes. The community of insects on milkweed thrives by either actively avoiding defenses in the plant or by consuming, sequestering, and advertising these same defenses. Probably the two most potent aspects of herbivore resistance in milkweeds are the production of cardenolides and latex. Cardenolides are bitter tasting steroids that occur in all milkweed tissues, including the latex, that act by disrupting the sodium and potassium flux in cells, and have toxic effects on most animals (Malcolm 1991). The sticky white latex is delivered via specialized canals (laticifers) to most plant parts, and can be copiously exuded upon tissue damage (Fig. 1). Latex of milkweed has been strongly implicated as a physical impediment to herbivory (Dussourd and Eisner 1987, Zalucki et al. 2001). Other potentially defensive and nutritional constituents that could influence herbivores include leaf toughness, trichome density, water content, nitrogen content, and specific leaf area (Mattson 1980, Coley 1983, Haddad and Hicks 2000, Lavoie and Oberhauser 2004). Most of these traits show plastic

S136

ANURAG A. AGRAWAL AND MARK FISHBEIN

Ecology Special Issue

FIG. 1. (A) Common milkweed (Asclepias syriaca) viewed from above early in the growing season. (B) A newly hatched monarch caterpillar (Danaus plexippus). Before it can get its first meal, the caterpillar must graze the trichomes and avoid the droplets of latex. Exposure to latex is reduced by clipping laticifers in a circle and feeding on the tissue in the middle.

variation within species, are genetically variable in natural populations of single species, and are highly variable across species (Table 2). Using 24 species of Asclepias, we begin to address the role of convergent evolution in giving rise to antiherbivore defense syndromes. We specifically addressed the following questions: (1) What are the pairwise associations of defense traits across taxa? Are defense traits typically negatively correlated as predicted by redundancy or trade-off theory, are they positively correlated as predicted by syndrome theory, or are they not correlated at all? (2) Do species cluster into syndromes of common patterns of defense trait expression? (3) Are phylogenetic relationships among Asclepias species inferred from DNA sequences congruent with patterns of defense trait similarity? If so, we can conclude that phylogenetic history is sufficient to explain plant defense trait associations; if not, there is a suggestion for convergence in defense syndromes. And (4) Are particular defense traits or syndromes associated with the performance of a common oligophagous herbivore of milkweeds (monarch caterpillars, Danaus plexippus)? MATERIALS

AND

METHODS

Measuring plant traits Seeds for 24 species of milkweeds (Table 1) were obtained from field collections, nurseries, and native plant seed suppliers. We germinated seeds by nicking the tips and placing the seeds on moist filter paper. Seedlings were grown in potting soil (;10-cm pots) in growth chambers for one month and out-planted in a completely randomized common garden in a plowed field at the Koffler Scientific Reserve at Jokers Hill, Southern Ontario (44803 0 N, 79829 0 W; information available online).5 After 5

hhttp://www.zoo.utoronto.ca/jokershilli

accounting for plant mortality, our common garden had approximately 12 replicate plants per species (mean 6 SE, 11.8 6 0.7). All measurements were taken from newly expanded, undamaged leaves of plants in the common garden. We measured cardenolide concentrations as digitoxin equivalents (grams per gram dry tissue) extracted from 50 mg dry leaf tissue (n ¼ 5 replicates/species); we employed a spectrophotometric assay modified from Nelson (1993). We adapted the assay for sampling using a microplate reader (PowerWave X, Bio-Tek Instruments, Winooski, Vermont, USA). Field-collected leaf tissue was kept on ice, then frozen, freeze-dried, ground with a mortar and pestle, and weighed in 2-mL boilproof tubes. To each tube, we added 1.9 mL of 95% ethanol, tubes were then vortexed, and floated in a sonicating water bath (658C) for 10 min. We then centrifuged the tubes at 5000 rpm for five minutes at room temperature. Two 45-lL aliquots of the supernatant from each tube were then pipetted into the wells of a 96-well plate, one above the other (active sample and blank, respectively). Each plate also contained six samples of digitoxin for the standard curve used to determine concentrations of cardenolides (0.125–3 mg/ mL; Sigma Chemical, St. Louis, Missouri, USA). We then added 90 lL of ethanol to the blanks and 90 lL of 0.15% 2,2 0 4,4 0 -tetranitrodiphenyl (TNDP) in ethanol to the active samples. Finally, 70 lL of 0.1 mol/L aqueous NaOH was added to all wells to make the solutions basic and to catalyze the colorimetric reaction. After 15 minutes, all wells in the plate were read at 620 nm using the microplate reader. We measured latex from 5–10 replicates from each species by cutting the tip off (0.5 cm) an intact leaf in the field and collecting the exuding latex onto a 1 cm disc of filter paper. Latex stopped flowing after ;10 s, all latex was absorbed on the filter paper, and this disc was placed

July 2006

PLANT DEFENSE SYNDROMES

S137

TABLE 2. Seven putative defensive traits of milkweeds. Ranges for the actual trait values for treatment and for family and species means are presented, followed by percentage variation (in parentheses). Putative defense trait

Plastic variation

Full-sib variation

Species level variation

Latex (mg) Cardenolides (% dry mass) Trichomes (no. hairs/cm2) Toughness (g) C:N Water content (%) Specific leaf area (mm2/mg)

1.05–1.35 (29%) 2.7–4.0 (48%) 7.8–9.0 (15%)

0.9–4.5 (400%) 1.7–3.9 (129%) 580–1090 (88%) 96–128 (33%) 11–14 (27%)

NA

NA

NA

NA

0–4.8 (infinite) 2–16 (800%) 0–2019 (infinite) 58–177 (205%) 10–15 (50%) 78–90 (15%) 8–23 (188%)

NA NA

Notes: In experiments, we have characterized the level of variation in these traits among control Asclepias syriaca plants and those experimentally damaged by herbivores (plastic variation) (Van Zandt and Agrawal 2004a; A. A. Agrawal, unpublished data), 23 full-sibling families of A. syriaca (Agrawal 2004b, 2005), and 24 different species of Asclepias (Agrawal 2004a, this study). All data were collected from at least five individuals (per treatment, family, or species) typically growing in growth chambers (plastic variation) or independent field common gardens (full-sibling and species variation). There were significant differences between treatments, families, and species for all traits measured (ANOVA for all traits, P
0.05); families and species vary continuously between the extremes presented.

on top of another dry filter paper disc in a 24-well plate. The discs were dried at 608C and then weighed to the microgram. This method has proven to be highly repeatable (Van Zandt and Agrawal 2004a, Agrawal 2005). We assessed the trichome density of 5–10 replicate plants from each species by counting the tops and bottoms of leaf discs (28 mm2) under a dissection microscope. Leaf discs were taken from the tips of leaves. Water content was assessed by first weighing leaf discs wet and again after drying in an oven (608C). Specific leaf area (SLA) was calculated as the area of the leaf disc divided by the dry mass. This measure can be thought of as an indicator of thickness or density; leaves with a higher SLA are typically thin and have greater levels of herbivory (Scha¨dler et al. 2003). We measured leaf toughness on 10 replicate plants from each species with a force gauge penetrometer (Type 516; Chatillon, Largo, Florida, USA) that measures the mass (in grams) needed to penetrate a surface. We sandwiched the leaf between two pieces of Plexiglas, each with a 0.5-cm hole, pushed the probe of the penetrometer through the leaf, and recorded the maximum force required for penetration. For each leaf, we measured toughness on each side of the mid-rib; these two measures were averaged and used as a single data point per plant. Total leaf carbon (C) and nitrogen (N) concentration was measured from five replicates from each species by microcombustion, using 5 mg of dried ground leaf material in an Elemental Combustion System 4010, CHNS-O analyzer (Costech Analytical Technologies, Valencia, California, USA). We used the C:N ratio as indicator of plant nutritional quality. Means and standard errors of the seven defenserelated traits for each of the 24 milkweed species are reported in Appendix A. Statistical analyses of phylogenetic and nonphylogenetic trait associations Pairwise correlations among traits were calculated using raw trait values and phylogenetically independent

contrasts (PICs) (Felsenstein 1985). We calculated PICs using the maximum-likelihood (ML) estimate of the phylogeny of the 24 species of Asclepias (Fig. 2) for which defense traits were measured (Table 2), based on noncoding plastid DNA (cpDNA) sequences (rpl16 intron and trnC–rpoB intergenic spacer; M. Fishbein et al., unpublished manuscript). We obtained sequences directly, using standard protocols, and manually aligned them (M. Fishbein et al., unpublished manuscript). The best fitting nucleotide substitution model under the ML criterion (general time reversible plus invariant sites plus gamma [GTRþIþC]) was selected using hierarchical likelihood ratio tests and ML trees were sought using a standard he uristic s ea rch strateg y ( us ing PAUP*4.0b10; Swofford 2001) (see Fishbein et al. 2001). Clade support was assessed with Bayesian probabilities estimated under the GTRþIþC model, assuming uninformative priors for model parameters and tree topologies (using MrBayes 3.0b4, Ronquist and Huelsenbeck 2003) (see Fishbein and Soltis 2004). Markov chains were iterated for 5 3 105 generations with trees sampled during the first 105 generations discarded as transitory samples. The ML tree for these 24 species did not differ significantly from relationships found in a comprehensive analysis of 105 species of Asclepias (M. Fishbein et al., unpublished manuscript; results not shown). We calculated PICs using Felsenstein’s (1985) method, as implemented in COMPARE (available online).6 We have no a priori evidence concerning rates of evolution of defense traits. Thus, we conducted PIC analyses under the assumptions of (1) equal branch lengths, corresponding to a speciational model of evolution (Mooers et al. 1999), and (2) branch lengths estimated from cpDNA sequences. However, the branch lengths estimated from molecular data caused pathological behavior in the PIC analysis: several exceedingly short internodes (Fig. 2) resulted in widely inflated 6

hhttp://compare.bio.indiana.edui

S138

ANURAG A. AGRAWAL AND MARK FISHBEIN

Ecology Special Issue

FIG. 2. Maximum-likelihood phylogram of the 24 species of Asclepias employed in this study. Numbers indicate Bayesian posterior probabilities. Letters to the right of taxon names represent defense syndrome clusters (see Fig. 4 and Results: Assessing defense syndromes).

estimates of contrasts involving these branches. Thus, the few data points involving these nodes drove the pattern of correlation among contrasts (results not shown), which often differed dramatically from those found in equal branch length analyses. Despite the drawbacks of assuming equal probabilities of change across all branches of the phylogeny, this seems preferable to alternative assumptions about unknown rates of evolution. Thus, we only present the results based on assuming equal branch lengths. The nodes of the phylogeny subtended by extremely short branches correspond to areas of uncertainty in the phylogeny of Asclepias (see posterior clade probabilities in Fig. 2). A consequence of this uncertainty is possible error in the PIC analysis, due to using an incorrect phylogenetic estimate. To examine this source of error, we conducted PIC analyses on alternative topologies and assessed the robustness of PIC-based correlations

(cf. Housworth and Martins 2001). Starting with a tree in which all clades with posterior probabilities ,0.70 in Fig. 2 were collapsed, we generated 100 trees that randomly resolved all polytomies using MacClade 4.06 (Maddison and Maddison 2003). We present 95% confidence limits for our correlation coefficients by excluding the lowest and highest 2.5% of the r values. We conducted PIC analyses, as described, with COMPARE on each of these 100 trees. We used a hierarchical cluster analysis to produce a defense phenogram (cf. Becerra’s ‘‘chemograms’’) (Becerra 1997, Legendre and Legendre 1998) to group species by expression of defense traits. We used the mean values for each of the seven plant traits measured for each species to generate a phenogram. Mean trait values were transformed to Z scores (mean ¼ 0, SD ¼ 1) so that they were measured on a comparable scale. Following Becerra’s (1997) model, we employed Ward’s method for

July 2006

PLANT DEFENSE SYNDROMES

S139

TABLE 3. Pairwise correlations of defense-related traits among Asclepias species.

Latex (mg) Latex (mg) Trichomes (no. hairs/cm2) Toughness (g) C:N Water content (%) Cardenolides (% dry mass) SLA (cm2/mg)

Trichomes (no. hairs/cm2)

Toughness (g)

C:N

Water content (%)

Cardenolides SLA (% dry mass) (cm2/mg)

0.60** 0.10 0.04 0.21 0.13 0.62** 0.10 0.17 0.04 0.05 (0.47–0.70) 0.06 0.05 0.32 0.19 0.06 (0.25–0.10) (0.27–0.18) 0.03 0.18 0.43* 0.25 0.25 (0.25–0.10) (0.43 to 0.115) (0.18–0.455) 0.29 0.01 0.34  0.17 0.27 (0.32 to 0.045) (0.08–0.165) (0.35 to 0.13) (0.30 to 0.01) 0.08 0.1 0.01 0.24 0.16 (0.16–0.02) (0.015–0.195) (0.095–0.105) (0.32 to 0.14) (0.065–0.265) 0.57** 0.45* 0.15 0.07 0.57** 0.21 (0.58 to 0.42) (0.495 to 0.295) (0.24–0.055) (0.215–0.01) (0.46–0.61) (0.12–0.27)

0.59** 0.43* 0.26 0.26 0.39  0.15

Notes: Correlation coefficients (r) above the diagonal are uncorrected for bias due to phylogenetic history; those below the diagonal are phylogenetically independent contrasts (Felsenstein 1985) assuming a speciational model of evolution (with the 95% confidence limits, based on random resolution of poorly supported nodes, shown in parentheses). SLA (last column) is specific leaf area. None of the significant correlations represents a trade-off in defenses (see Results). * P , 0.05, **P , 0.01; these values are also highlighted in boldface type; df ¼ 22.   P  0.1; df ¼ 22.

linkage and Euclidean distances, which combines subgroups (initially building from one species) at each iteration so as to minimize the within-cluster ANOVA sum of squares (Wilkinson 1997). Alternate joining algorithms provided qualitatively similar relationships. Clusters were designated by distances from nodes; nodes separated by a distance of ,0.5 were included in the same cluster, whereas nodes separated by .0.5 were placed in separate clusters (Wilkinson 1997). We further examined the strength to which particular traits contributed to differences among clusters using discriminant function analysis (Wilkinson 1997). To investigate whether factors related to geographical distribution were associated with the repeated emergence of trait combinations, we conducted a preliminary test for an association between geography and membership in the hierarchical clusters (Thaler and Karban 1997, Becerra and Venable 1999). Biogeographical associations of defense clusters were investigated by contingency table analysis. The bulk of species we studied have ranges in eastern North American (United States and Canada, east of the Tall Grass Prairie–Great Plains ecotone) or western North American (west of the ecotone) (Table 1). Congruence between the ML phylogeny of Asclepias and hierarchical clustering of species based on defense traits was assessed using the test of Shimodaira and Hasegawa (SH test) (Shimodaira and Hasegawa 1999, Goldman et al. 2000), as implemented in PAUP* 4.0b10 (Swofford 2001), using RELL resampling to estimate site likelihoods. Statistical incongruence between the phylogeny of Asclepias and clustering of species by defense traits would indicate that the evolution of defensive traits does not passively track phylogeny, but instead is evolutionarily labile, and suggests phylogenetically independent derivations of associations among traits. This approach evaluates whether associations among traits generate hierarchical similarity among taxa

that is statistically independent of phylogenetic relationships. Stochastically evolving traits will generate clusters of taxa statistically indistinguishable from the estimated phylogeny. Significant deviations in the phenotypic clustering could be caused by convergent adaptation of traits or other processes that generate trait correlations that are independent of association due to phylogenetic history. This test is similar in aim to tests of trait conservatism (e.g., Ackerly and Donoghue 1998, Cavender-Bares et al. 2004). An advantage to our approach is the ability to consider the associations among all traits simultaneously, with the concomitant disadvantage that the conservatism of a single trait cannot be assessed. Effects of defense traits on caterpillars To assess how individual plant traits and defense syndromes affect the performance of one natural feeder of milkweeds, we conducted a bioassay with monarch caterpillars (Danaus plexippus) (Fig. 1). We placed single eggs of freshly hatched caterpillars on all plants (n ’ 12 for each of the 24 plant species) in the field. Caterpillars were from a laboratory colony established from local individuals collected the previous summer and maintained on frozen milkweed foliage. Each caterpillar was caged in a spun polyester bag (Rockingham Opportunities, Reidsville, North Carolina, USA) on the apical meristem with four fully expanded leaves. After five days, we recorded mortality and collected and weighed each living caterpillar. To assess whether plant species that were in the three different defense clusters (see Results) showed differential resistance to monarch caterpillars, we conducted a one-way ANOVA. The species cluster was used as the factor (each species was assigned to one cluster) and mean monarch mass and percentage survival on each species were the response variables (total n ¼ 24). In addition, we employed multiple regression to assess the

S140

ANURAG A. AGRAWAL AND MARK FISHBEIN

Ecology Special Issue

FIG. 3. Significant pairwise correlation between raw values of latex production and trichome density among 24 Asclepias species (see Table 2). Raw correlations and those of phylogenetically independent contrasts (PIC) (Felsenstein 1985) are significant and of similar magnitude.

effects of plant traits on monarch performance across the 24 species. For this analysis, we used species means for each of the seven traits (cardenolides, latex, trichomes, water content, specific leaf area, toughness, and C:N ratio) and mean caterpillar mass and percentage mortality as the response variables. We employed a stepwise multiple regression with backwards removal (P ¼ 0.15 to enter or remove) (Wilkinson 1997). RESULTS Pairwise correlations between traits Three of the 21 raw pairwise correlations (and five based on phylogenetically independent contrasts, PICs) among seven defense-related traits were statistically significant (Table 3). This observed frequency of phylogenetically independent correlations is unlikely to have occurred by chance (binomial expansion test, P ¼ 0.0028) (Zar 1996). Each of the significant raw correlations involved just three traits showing complementary patterns of variation. Species with lower specific leaf areas (SLA) exhibited higher leaf trichome densities and latex production (Fig. 3). Species with lower SLA also exhibited lower water content (P ¼ 0.06), as expected (SLA is derived from leaf dry mass). Recall that SLA is a physiological plant trait, often indicative of rapid growth and high palatability to herbivores, and that low SLA might defend against herbivory. Leaves of species with high C:N ratios were also tougher, although this relationship was not significant (P ¼ 0.13). Thus, we found no indications of trade-offs among defense traits. Indeed, all correlations were ‘‘positive’’; the significant negative correlations in Table 2 represent negative associations between nutritional quality and defense. The phylogenetically independent correlations did not differ substantially from raw correlations (Table 3, Fig. 3). The phylogenetically independent correlation

between SLA and water content and between leaf C:N ratio and toughness were stronger and statistically significant compared to the raw correlations. In addition, a weak and nonsignificant negative correlation between leaf water content and toughness had a nearly significant phylogenetically independent correlation (P ¼ 0.10; Table 3). Correlations based on PICs were generally robust to phylogenetic uncertainty (Table 3). For the latex– trichome, latex–SLA and SLA–percentage water correlations, every one of the randomly resolved topologies generated significant correlations. The trichome–SLA correlation was significant in just 41% of the random resolutions, although the upper bound of the 95% confidence limit (r ¼ 0.295) was still strongly negative. The C:N–toughness correlation was less robust, with only 12% of the random resolutions remaining significant. The only instance of a nonsignificant PIC correlation that was found to be significant in randomly resolved topologies was the C:N–trichome correlation, but this occurred in only 3% of the resolutions. Overall, little bias can be attributed to using the ML phylogeny in the face of weak support for a number of clades. Assessing defense syndromes Our hierarchical cluster analysis of the seven defenserelated traits revealed three distinct clusters (Fig. 4). For illustrative purposes only, in post hoc analyses, we determined that clusters differed strongly from one another (MANOVA results from discriminant function analysis: Wilks’ lambda ¼ 0.033, F14,30 ¼ 9.688, P , 0.001; see Appendix B). The analysis also indicates how each trait correlated with each discriminant function (Table 4); although most traits contributed substantially to at least one function, cardenolides and leaf toughness contributed the least to both. We also identified which traits were significantly heterogeneous among the three

July 2006

PLANT DEFENSE SYNDROMES

S141

FIG. 4. A defense phenogram that depicts similarity among 24 species of Asclepias generated by hierarchical cluster analysis of seven defense-related traits. Tightly clustered species are defensively similar and can be considered to form defense syndromes, A–C (see Results: Assessing defense syndromes).

clusters (Table 4). These analyses revealed that most traits were expressed differently in the three defense clusters, and again cardenolides and leaf toughness were the least distinct. The three clusters (Fig. 4, Table 4) represent species with combinations of: (A) high nitrogen (i.e., low C:N) coupled with high physical defense traits (trichomes, latex), (B) very high C:N ratio, coupled with tough leaves and low water content (hard to eat, little reward), and (C) low C:N and SLA coupled with high cardenolides. Although differences in cardenolides were not significantly different between clusters (Table 3), we found 17% and 50% higher levels in cluster C than in A

or B, respectively. Clusters A and C are strategies that both represent the coupling of a trait increasing the reward to herbivores (nitrogen or SLA) with high defense allocation. The three defense clusters were not significantly associated with distribution in eastern vs. western North America (v2 ¼ 8.77, df ¼ 5, P ¼ 0.12). Phylogenetic congruence test The phylogenetic relationships suggested by the defense trait cluster analysis (Fig. 4) are significantly incongruent with relationships estimated from noncoding cpDNA sequences (Fig. 2). The difference in log likelihood [ln(L)] between the maximum likelihood

TABLE 4. The relationship between seven defensive traits and defensive clustering of milkweed species. Coefficients of standardized canonical discriminant functions (CDFs) indicate how each trait contributes to the two factors generated by discriminant analysis (Wilkinson 1997). Trait values (mean 6 SE) appear for the three defense syndromes (clusters) identified in this study. Significance of differences among traits tested by ANOVA.

Latex (mg) Trichomes (no. hairs/cm2) Toughness (g) C:N Water content (%) Cardenolides (% dry mass) Specific leaf area (mm2/mg)

CDF1

CDF2

Cluster A (n ¼ 4)

Cluster B (n ¼ 6)

–0.495 –0.907 –0.089 0.637 0.102 –0.011 –0.456

0.404 0.316 0.285 0.703 –0.388 –0.008 –0.284

3.13 1630.08 111.08 11.48 0.83 4.97 9.98

1.28 355.40 131.00 14.09 0.81 3.89 10.41

6 6 6 6 6 6 6

0.72 146.57 17.93 0.08 0.01 0.67 0.69

6 6 6 6 6 6 6

0.56 185.22 12.01 0.32 0.01 0.76 0.99

Cluster C (n ¼ 14) 0.31 337.47 104.86 11.66 0.84 5.85 15.80

6 6 6 6 6 6 6

0.09 67.11 5.53 0.21 0.01 1.11 1.02

P ,0.001 ,0.001 0.136 ,0.001 0.088 0.508 0.002

S142

ANURAG A. AGRAWAL AND MARK FISHBEIN

Ecology Special Issue

FIG. 5. Schematic depiction of the lack of congruence between the molecular phylogeny of Asclepias and the defense trait phenogram (see Fig. 4).

tree for the sequence data (–ln(L) ¼ 4225.61) and the highest likelihood tree congruent with the phenogram (–ln(L) ¼ 4532.10) was 306.49, which was highly significant (SH test: P , 0.05). Thus, multivariate defense clusters did not come about only as a consequence of tracking the speciational history of Asclepias (Fig. 5). Effects of defense traits on caterpillars When plant species were classified by defense cluster (Fig. 4), we found no difference in the performance of monarch caterpillars among groups (mass, F2,20 ¼ 2.460, P ¼ 0.111; survival, F2,21 ¼ 2.080, P ¼ 0.150). Nonetheless, in multiple regression analyses, plant traits significantly explained monarch performance (Fig. 6) (overall model for mass, R2 ¼ 0.33, F2,20 ¼ 4.906, P ¼ 0.018, trichome coefficient ¼ 0.130, P ¼ 0.022, toughness coefficient ¼ 0.698, P ¼ 0.041; overall model for survival, R2 ¼ 0.26, F1,22 ¼ 7.566, P ¼ 0.012, latex coefficient ¼ 0.060, P ¼ 0.012). The latter positive correlation between latex and percentage of caterpillar

survival is difficult to explain and is discussed below (see the Discussion). DISCUSSION The concept of convergent expression of suites of traits acting as syndromes has been a useful framework for conceptualizing adaptations in ecological communities. For example, plant ecologists have proposed that specific suites of traits might be associated with particular types of environmental stress (Grime 1977, Chapin et al. 1993). In animal ecology, the guild concept is used to characterize species that exploit the same type of resources in a similar manner (Root 1967). Species forming a guild can share a common set of traits, but are explicitly not necessarily unified by phylogenetic relatedness. Nonetheless, in the past there was limited ability to disentangle the role of phylogenetic history and convergence in generating such species groups. For example, in the plant pollination literature, phylogenetic analyses of convergence on plant traits associated with bee and hummingbird pollination have only recently been employed (McDade 1992, Fenster et al. 2004,

July 2006

PLANT DEFENSE SYNDROMES

S143

Wilson et al. 2004). Other assessments of syndromes have also recently adopted a phylogenetically explicit approach (Cunningham et al. 1999, Fine et al. 2004). Thus, we see our concept of plant defense syndromes as an extension and phylogenetic modernization of classic plant defense theory (e.g., Feeny 1976). Correlations between traits In our study of Asclepias, none of the measured defense-related traits showed the bivariate trade-off predicted by traditional redundancy theory. Indeed, all of the pairwise correlations between defensive traits were positive; the significant correlations with a negative sign were between nutritional quality and defense, thus, not indicating a trade-off among traits providing resistance. Using phylogenetically independent contrasts among cotton (Gossypium) species, Rudgers et al. (2004) also found no trade-off between the production of extrafloral nectaries (EFNs) and trichomes, or between EFNs and chemical defense glands; however, they did find a negative association between trichomes and glands. This negative correlation could reflect allocation costs of the traits or redundancy in their ecological functions. Given that the costs and ecological functions of most defenserelated traits are unknown, however, in this paper we have argued that any two traits should not be a priori predicted to negatively covary. This same conclusion has been reached in a recent meta-analysis of intraspecific genetic correlations of defense traits (Koricheva et al. 2004). We found a positive (phylogenetically independent) correlation between plant production of latex and trichomes (Fig. 3). In an analysis of the correlation between the same two traits across 23 genetic families of A. syriaca, we found no correlation (Agrawal 2005). Although other intraspecific correlations are needed, Armbruster et al. (2004) and others have argued that such correlations across species, but not within species, are suggestive of adaptation over constraint. We speculate that this association could be due to the fact that latex is a water-intensive defense, and the protection from water loss by trichomes facilitates use of latex. Additionally, latex has widely been reported to function as a defense against monarch caterpillars and other milkweed herbivores (Dussourd and Eisner 1987, Zalucki et al. 2001, Agrawal and Van Zandt 2003, Agrawal 2004b, 2005), suggesting that the two strategies might be employed in concert to produce a synergistic defense. Although we did not detect a negative effect of latex on monarchs in the current study, it is perhaps in this situation where plants derive benefits from employing dual strategies. We speculate that the positive correlation we observed between latex production and percentage caterpillar survival is an artifact of a correlation between latex and some unmeasured trait. It is also possible that plant species with higher levels of latex prevented caterpillars from feeding, which protected them from death due to other plant defenses; in

FIG. 6. Effects of milkweed defensive traits on monarch caterpillar growth. Seven plant traits were assayed, and only the significant predictors are shown: (A) trichome density and (B) leaf toughness. Residual growth refers to the residuals from the statistical model with the other factor in the analysis.

our short-term assays this protective effect is, thus, perhaps an artifact. A notable outlier in the positive correlation between trichomes and latex (Fig. 3) was A. sullivantii, which has high latex production but no trichomes. Two other possible independent origins of species with high latex production coupled with the absence of trichomes are A. humistrata (southeastern USA) and a small clade of Mexican species (A. glaucescens, A. elata, A. mirifica, and A. lynchiana), neither of which were sampled in the current study. These species (including A. sullivantii) all have pronounced depositions of epicuticular wax (Fishbein 1996), which we speculate may be a substitute for trichomes in a physical defense syndrome. Thus, in summary, the admittedly adaptationist hypothesis under the defense syndrome concept is not of specific a priori trade-offs, but a prediction that the multivariate trait complex is grouped such that costs are minimized and defense is maximized, given the biotic and abiotic environment of the species. Defense syndromes

S144

ANURAG A. AGRAWAL AND MARK FISHBEIN

Ecology Special Issue

FIG. 7. The plant defense syndrome triangle. Low nutritional defense syndrome is consistent with that outlined for apparent plants by Feeny (1976); a similar group was found in this study. Tolerance follows the fast-growth and high-edibility pattern outlined by Coley (Coley et al. 1985, Kursar and Coley 2003). Nutrition and defense is a strategy that couples a toxic defense or barrier to feeding with relatively high edibility and digestibility. SLA denotes specific leaf area.

themselves might trade-off if they represent multivariate strategies targeted at alternative ecological interactions. Defense syndromes in milkweeds and beyond We distinguished three syndromes (clusters) in our multivariate analysis of seven defense-related traits, which grouped as plant species employing (A) high physical defenses, (B) low nutritional quality, and (C) putatively high chemical defense (Figs. 1 and 3). We found that, overall, our clustering of plant species by defense traits was not congruent with the molecular phylogeny of the group. At a minimum, this means that the defense characteristics of Ascelpias are evolutionarily labile, and do not simply track phylogenetic history. For example, in the well-resolved group of three species that co-occur in eastern North America (A. syriaca, A. tuberosa, and A. exaltata) all three clusters are represented. In addition, it is apparent that some taxa, although distantly related, have converged on similar defense phenotypes. Although we have not reconstructed the relative timing of changes in plant nutritional quality and expression of defense traits, we assume that when such traits are associated, changes in defense traits follow after divergences in nutrition traits rather than the reverse. For two of the three defense clusters (A and C in Fig. 4) there is an association between high resource

quality (from the herbivore’s perspective) and high expression of defense traits. In cluster A, low C:N ratio (i.e., relatively high nitrogen levels) is coupled with high latex production and trichome density. In cluster C, both a low C:N ratio and high specific leaf area (SLA; i.e., relatively concentrated plant tissue) is coupled with higher levels of cardenolides than the other clusters. In the case of the cluster-A species, repeated evolution of a latex–trichome associated syndrome is statistically supported (Table 3, Fig. 3) Although this analysis of Asclepias focused primarily on resistance, plant growth traits and tolerance are likely to be an important component of particular defense syndromes (van der Meijden et al. 1988, Fornoni et al. 2004). For example, Kursar and Coley (2003) made the argument that a trade-off between defense and growth promoted divergent strategies among tropical tree species. We offer the ‘‘defense syndrome triangle’’ hypothesis to include all defense categories (Fig. 7). Presumably other traits, such as plant traits that affect herbivore preference and attraction or facilitation of enemies of herbivores will also be important to integrate into the concept of plant defense syndromes. Nearest to the origin of the two axes of edibility/ digestibility and toxicity/barriers is the low nutritional quality syndrome (Fig. 7). This syndrome, empirically demonstrated in Asclepias (Syndrome B in Fig. 4, Table

July 2006

PLANT DEFENSE SYNDROMES

4), very closely mirrors Feeny’s (1976) grouping of defensive traits for apparent plants, and Coley et al.’s (1985, Kursar and Coley 2003) ‘‘defense syndrome’’ for plants that have evolved in low-resource environments. Note that the defense syndrome concept does not implicate a particular selective agent in generating the suite of adaptations. Feeny hypothesized that the selective agents were herbivores under the constraint of the probability of being found; Coley also hypothesized that herbivores were the selective agent; however, this was presumed to be constrained by the abiotic (resource) environment. The tolerance or escape syndrome (proposed by Kursar and Coley [2003], and not investigated in the current study) predicts a lack of resistance traits in very fast-growing and nutritive plants. Finally, our study revealed two syndromes with intermediate levels of edibility coupled with barriers to consumption (Syndromes A and C, Fig. 4, Table 4). In some ways, this mixed strategy is similar to Feeny’s prediction for unapparent plants (i.e., qualitative defenses that cannot be overcome by generalist herbivores, but specialists feed on these plants with impunity). The important distinction is the implicit recognition that these toxins or barriers to consumption (latex, trichomes, cardenolides, etc), typically have a dose-dependent (or quantitative) effect on specialist herbivores (Berenbaum et al. 1989, Adler et al. 1995, Agrawal and Kurashige 2003, Agrawal and Van Zandt 2003, Agrawal 2004a, b, 2005). Plants with high levels of toxicity or barriers to feeding, however, are not predicted to have particularly high or low levels of nutritional quality (Fig. 7). We reach this conclusion based on the reasoning that plants with extreme investments in toxins should not have the ability to produce very nutritive tissues; likewise plants with very low nutritive quality should not need to invest in toxins. Caveats The performance of monarch butterfly caterpillars on the 24 milkweed species did not differ on plants from the three syndromes. This suggests that plant species might have alternate mechanisms to achieve similar levels of defense, and these might be driven by environmental differences in the habitats in which these species live naturally. Alternatively, our analysis should be viewed as preliminary and is limited by (1) our somewhat arbitrary distinction of clusters (see Materials and methods: Effects of defense traits on caterpillars), (2) unmeasured defense traits (epicuticular waxes, proteases, etc.), (3) a limited sample of species, and (4) limited information on the ecology of each species that would be informative for why particular defense syndromes were expressed across these species. Thus, further analyses are needed. In addition, our approach was to sample broadly across Asclepias, but more detailed analyses of well-resolved clades would also be valuable. Such fine-scale sampling would better control for the confounding effects of unmeasured traits by

S145

limiting the time of divergence among species and permitting denser taxon sampling. It is important to bear in mind that our working hypothesis for the phylogeny of Asclepias (M. Fishbein et al., unpublished manuscript) is not highly resolved with strong statistical support (see Fig. 2). The lack of resolution reflects the reality of the apparently rapid initial radiation of Asclepias. It is also important to consider the effect of our assumption of speciational evolution of morphology (i.e., equal morphological branch lengths across the tree) in calculating phylogenetically independent contrasts (PICs). Both of these factors could bias our estimates of evolutionary correlations among traits. We found that estimated correlations among PICs were very robust to uncertainty regarding the correct relationships in poorly supported regions of the tree. Conversely, the assumption concerning the mode of evolution of defense traits had a dramatic effect on the sign and magnitude of correlations. Ideally, an individual model of evolution should be developed for each trait in order to accurately estimate correlations (e.g., Pagel 1999, Lewis 2001). Because of the relatively small number of taxa (i.e., low sample size) and uncertainty regarding the correct phylogenetic relationships, we did not pursue individualized fitting of evolutionary models for the defense traits. We did attempt to analyze correlations assuming that rates of morphological evolution were accurately estimated by branch lengths inferred from the rates of nucleotide substitution in the cpDNA sequences used to estimate the tree. Correlations estimated under this assumption differed substantially from those reported here. Although we are not comfortable assuming equal branch lengths across the tree (which is almost certainly incorrect), we prefer this assumption to the use of molecular branch lengths. Generally, there is little evidence for correspondence between rates of molecular and morphological evolution (Bromham et al. 2002). Specifically, the extreme branch length heterogeneity for the molecular data is likely unreasonable for the morphological traits under study here. Species of Asclepias that exhibit very low levels of divergence in cpDNA sequences can be dramatically different in morphology, whereas species that differ only subtly (A. fascicularis, A. verticillata, A. perennis, and A. incarnata) are separated by branches that are orders of magnitude longer than those separating morphologically diverse species (see Fig. 2). Given the apparently poor correspondence between rates of molecular and morphological evolution, we prefer the assumption of speciational changes in morphology over any assumption about specific rates across lineages. Directions We see the identification of syndromes as a starting point to test alternative hypotheses for why plant defenses have converged. In this regard, the biotic and

S146

ANURAG A. AGRAWAL AND MARK FISHBEIN

abiotic environment could conspire to drive particular syndromes. For example, plant trichomes are known to have many functions, including defense against herbivory as well as a barrier against evapotranspirative water loss (e.g., Woodman and Fernandes 1991, Agrawal and Spiller 2004). Thus, trichomes might be particular to plant species in environments with low water availability (i.e., deserts) and subject to particular types of herbivores. We have shown here that trichomes are a significant barrier to growth of monarch caterpillars, but benefit some aphids (Agrawal 2004a). Thus, trichomes might only be favored in particular combinations of biotic and abiotic conditions. We suggest working toward formalizing hypotheses about the particular types of herbivores that likely drive the evolution of particular syndromes. It might be that plants that share guilds of herbivores (i.e., those that attack the same plants in a similar way) defend in a similar manner. The example described initially (see Introduction), that of plants defending against vertebrate megafaunal grazers vs. insect herbivores, serves as a starting point. For North American Asclepias, the herbivore fauna of some species is .12 insect herbivore species that eat every plant part, from the flowers and seed pods, to the stems and leaves, to the phloem and roots. But all milkweeds are not subject to herbivory by all insect species, and levels of attack presumably vary considerably. For example, the very damaging stem weevil (Rhyssomatus lineaticollis) is primarily known from A. syriaca, and there apparently are not analogous herbivores on other milkweeds (Agrawal and Van Zandt 2003). Twenty-four species of root-boring Tetraopes beetles attack a single species of milkweed each; apparently the ;100 other North American milkweeds are not attacked in this way. Monarch butterflies only attack a subset of the milkweed flora, probably because their flight paths do not overlap with all species, and some species inhabit forest understories or occur as widely scattered individuals. Given that herbivore guilds vary on plant species and there is a high level of specificity in the effects of particular plant defenses on particular herbivores (e.g., Da Costa and Jones 1971, Giamoustaris and Mithen 1995, Van Dam and Hare 1998, Agrawal and Karban 2000, Van Zandt and Agrawal 2004a, b), we predict that future studies will find that certain herbivore species or guilds have been critical in generating patterns of defense. In identifying such convergent plant defense syndromes, we may finally understand the evolutionary association between herbivore communities and adaptive variation in plant species. ACKNOWLEDGMENTS We thank Rowan Barrett, Erin Douthit, Dana Farmer, Karen Hooper, Deborah Hopp, Shelley McMahon, Michael Moody, Lisa Plane, Ana Lilia Reina G., Karin Rotem, Victor Steinmann, Deborah Tam, Jennifer Thaler, Tom Van Devender, Pete Van Zandt, and Sergio Zamudio for help in the lab

Ecology Special Issue

and field; Robie Mason-Gamer and Steve Lynch for contributions to the phylogenetic study of Asclepias; David Goyder, Mark Chase, and Marlin Bowles for providing DNA samples; Bobby Gendron for seeds; Steve Malcolm for discussions on cardenolide analysis; and Paul Fine, Marc Johnson, Rick Karban, Marc Lajeunesse, Peter Price, Jennifer and Jon Thaler, and Cam Webb for comments and discussion. This research and our laboratories are supported by Natural Science and Engineering Research Council of Canada, the U.S. National Science Foundation, and the U.S. Department of Agriculture. LITERATURE CITED Ackerly, D. D., and M. J. Donoghue. 1998. Leaf size, sapling allometry, and Corner’s rules: phylogeny and correlated evolution in maples (Acer). American Naturalist 152:767– 791. Adler, L. S., J. Schmitt, and M. D. Bowers. 1995. Genetic variation in defensive chemistry in Plantago lanceolata (Plantaginaceae) and its effect on the specialist herbivore Junonia coenia (Nymphalidae). Oecologia 101:75–85. Agrawal, A. A. 2004a. Plant defense and density dependence in the population growth of herbivores. American Naturalist 164:113–120. Agrawal, A. A. 2004b. Resistance and susceptibility of milkweed: competition, root herbivory, and plant genetic variation. Ecology 85:2118–2133. Agrawal, A. A. 2005. Natural selection on common milkweed (Asclepias syriaca) by a community of specialized insect herbivores. Evolutionary Ecology Research 7:651–667. Agrawal, A., and R. Karban. 2000. Specificity of constitutive and induced resistance: pigment glands influence mites and caterpillars on cotton plants. Entomologia Experimentalis et Applicata 96:39–49. Agrawal, A. A., and N. S. Kurashige. 2003. A role for isothiocyanates in plant resistance against the specialist herbivore Pieris rapae. Journal of Chemical Ecology 29: 1403–1415. Agrawal, A. A., and S. B. Malcolm. 2002. Once upon a milkweed. Natural History 111(7):48–53. Agrawal, A. A., and D. A. Spiller. 2004. Polymorphic buttonwood: effects of disturbance on resistance to herbivores in green and silver morphs of a Bahamian shrub. American Journal of Botany 91:1990–1997. Agrawal, A. A., and P. A. Van Zandt. 2003. Ecological play in the coevolutionary theatre: genetic and environmental determinants of attack by a specialist weevil on milkweed. Journal of Ecology 91:1049–1059. Armbruster, W. S., C. Pe´labon, T. F. Hansen, and C. P. H. Mulder. 2004. Floral integration, modularity, and accuracy. Pages 23–49 in M. Pigliucci and K. Preston, editors. Phenotypic integration. Oxford University Press, Oxford, UK. Beardsley, P. M., A. Yen, and R. G. Olmstead. 2003. AFLP phylogeny of Mimulus section Erythranthe and the evolution of hummingbird pollination. Evolution 57:1397–1410. Becerra, J. X. 1997. Insects on plants: macroevolutionary chemical trends in host use. Science 276:253–256. Becerra, J. X., and D. L. Venable. 1999. Macroevolution of insect–plant associations: the relevance of host biogeography to host affiliation. Proceedings of the National Academy of Sciences (USA) 96:12626–12631. Becerra, J. X., D. L. Venable, P. H. Evans, and W. S. Bowers. 2001. Interactions between chemical and mechanical defenses in the plant genus Bursera and their implications for herbivores. American Zoologist 41:865–876. Bennett, R. N., and R. M. Wallsgrove. 1994. Tansley review no. 72. Secondary metabolites in plant defence mechanisms. New Phytologist 127:617–633. Berenbaum, M. R. 2001. Chemical mediation of coevolution: phylogenetic evidence for Apiaceae and associates. Annals of the Missouri Botanical Garden 88:45–59.

July 2006

PLANT DEFENSE SYNDROMES

Berenbaum, M. R., J. K. Nitao, and A. R. Zangerl. 1991. Adaptive significance of furanocoumarin diversity in Pastinaca sativa (Apiaceae). Journal of Chemical Ecology 17:207– 215. Berenbaum, M. R., A. R. Zangerl, and K. Lee. 1989. Chemical barriers to adaptation by a specialist herbivore. Oecologia 80: 501–506. Berenbaum, M. R., A. R. Zangerl, and J. K. Nitao. 1986. Constraints on chemical coevolution: wild parsnips and the parsnip webworm. Evolution 40:1215–1228. Blakley, N. R., and H. Dingle. 1978. Competition: butterflies eliminate milkweed bugs from a Caribbean island. Oecologia 37:133–136. Bollwinkel, C. W. 1969. A revision of the South American species of Asclepias L. Dissertation. Southern Illinois University, Carbondale, Illinois, USA. Broadway, R. M., and S. S. Duffey. 1988. The effects of plant protein quality on insect digestive physiology and the toxicity of plant proteinase inhibitors. Journal of Insect Physiology 34:1111–1118. Bromham, L., M. Woolfit, M. S. Y. Lee, and A. Rambaut. 2002. Testing the relationships between morphological and molecular rates of change along phylogenies. Evolution 56: 1921–1930. Brower, L. P., P. B. McEvoy, K. L. Williamson, and M. A. Flannery. 1972. Variation in cardiac glycoside content of monarch butterflies from natural populations in eastern North America. Science 177:426–429. Brower, L. P., J. Van Zandt, and J. M. Corvino. 1967. Plant poisons in a terrestrial food chain. Proceedings of the National Academy of Science (USA) 57:893–898. Bryant, J. P., F. S. Chapin, and D. R. Klein. 1983. Carbon/ nutrient balance of boreal plants in relation to vertebrate herbivory. Oikos 40:357–368. Castellanos, M. C., P. Wilson, and J. D. Thomson. 2003. Pollen transfer by hummingbirds and bumblebees, and the divergence of pollination modes in Penstemon. Evolution 57:2742– 2752. Castellanos, M. C., P. Wilson, and J. D. Thomson. 2004. ‘‘Antibee’’ and ‘‘pro-bird’’ changes during the evolution of hummingbird pollination in Penstemon flowers. Journal of Evolutionary Biology 17:876–885. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Chapin, F. S., K. Autumn, and F. Pugnaire. 1993. Evolution of suites of traits in response to environmental stress. American Naturalist 142:S78–S92. Coley, P. D. 1983. Herbivory and defensive characteristics of tree species in a lowland tropical forest. Ecological Monographs 53:209–233. Coley, P. D. 1987. Interspecific variation in plant anti-herbivore properties: the role of habitat quality and rate of disturbance. New Phytologist 106:251–264. Coley, P. D., J. P. Bryant, and F. S. Chapin. 1985. Resource availability and plant antiherbivore defense. Science 230:895– 899. Conway-Morris, S. 2003. Life’s solution: inevitable humans in a lonely universe. Cambridge University Press, Cambridge, UK. Cunningham, S. A., B. Summerhayes, and M. Westoby. 1999. Evolutionary divergences in leaf structure and chemistry, comparing rainfall and soil nutrient gradients. Ecological Monographs 69:569–588. Da Costa, C. P., and C. M. Jones. 1971. Cucumber beetle resistance and mite susceptibility controlled by the bitter gene in Cucumis sativus L. Science 172:1145–1146. Dobzhansky, T. G. 1970. Genetics of the evolutionary process. Columbia University Press, New York, New York, USA. Duffey, S. S., and M. J. Stout. 1996. Antinutritive and toxic components of plant defense against insects. Archives of Insect Biochemistry and Physiology 32:3–37.

S147

Dussourd, D. E., and T. Eisner. 1987. Vein-cutting behavior: insect counterploy to the latex defense of plants. Science 237: 898–900. Ehrlich, P. R., and P. H. Raven. 1964. Butterflies and plants: a study in coevolution. Evolution 18:586–608. Fægri, K., and L. van der Pijl. 1979. The principles of pollination ecology. Third edition. Pergamon Press, New York, New York, USA. Farrell, B. D., D. E. Dussourd, and C. Mitter. 1991. Escalation of plant defense: Do latex and resin canals spur plant diversification? American Naturalist 138:881–900. Farrell, B. D., and C. Mitter. 1993. Phylogenetic determinants of insect/plant community diversity. Pages 253–266 in R. Ricklefs and D. Schluter, editors. Historical and geographical determinants of community diversity. University of Chicago Press, Chicago, Illinois, USA. Farrell, B. D., and C. Mitter. 1998. The timing of insect–plant diversification: Might Tetraopes (Coleoptera: Cerambycidae) and Asclepias (Asclepiadaceae) have co-evolved? Biological Journal of the Linnean Society 63:553–577. Feeny, P. P. 1976. Plant apparency and chemical defense. Pages 1–40 in J. W. Wallace and R. L. Mansell, editors. Biochemical interaction between plants and insects. Plenum, New York, New York, USA. Felsenstein, J. 1985. Phylogenies and the comparative method. American Naturalist 125:1–15. Fenster, C. B., W. S. Armbruster, M. R. Dudash, P. Wilson, and J. D. Thomson. 2004. Pollination syndromes and floral specialization. Annual Review of Ecology and Systematics 35:375–403. Fine, P. V. A., I. Mesones, and P. D. Coley. 2004. Herbivores promote habitat specialization by trees in Amazonian forests. Science 305:663–665. Fine, P. V. A., Z. J. Miller, I. Mesones, S. Irazuzta, H. M. Appel, M. H. H. Stevens, I. Sa¨a¨ksja¨rvi, J. C. Schultz, and P. D. Coley. 2006. The growth–defense trade-off and habitat specialization by plants in Amazonian forests. Ecology 87: S150–S162. Fishbein, M. 1996. Phylogenetic relationships of North American Asclepias L. and the role of pollinators in the evolution of the milkweed inflorescence. Dissertation, University of Arizona, Tucson, Arizona, USA. Fishbein, M., C. Hibsch-Jetter, D. E. Soltis, and L. Hufford. 2001. Phylogeny of Saxifragales (angiosperms, eudicots): analysis of a rapid, ancient radiation. Systematic Biology 50: 817–847. Fishbein, M., and D. E. Soltis. 2004. Further resolution of the rapid radiation of Saxifragales (angiosperms, eudicots) supported by mixed-model Bayesian analysis. Systematic Botany 29:883–891. Fordyce, J. A., and S. B. Malcolm. 2000. Specialist weevil, Rhyssomatus lineaticollis, does not spatially avoid cardenolide defenses of common milkweed by ovipositing into pith tissue. Journal of Chemical Ecology 26:2857–2874. Fornoni, J., P. L. Valverde, and J. Nu´n˜ez-Fa´rfan. 2004. Population variation in the cost and benefit of tolerance and resistance against herbivory in Datura stramonium. Evolution 58:1696–1704. Giamoustaris, A., and R. Mithen. 1995. The effect of modifying the glucosinolate content of leaves of oilseed rape (Brassica napus ssp. oleifera) on its interaction with specialist and generalist pests. Annals of Applied Biology 126:347–363. Goldman, N., J. P. Anderson, and A. G. Rodrigo. 2000. Likelihood-based tests of topologies in phylogenetics. Systematic Biology 49:652–670. Grime, J. P. 1977. Evidence for existence of three primary strategies in plants and its relevance to ecological and evolutionary theory. American Naturalist 111:1169–1194. Gunasena, G. H., S. B. Vinson, H. J. Williams, and R. D. Stipanovic. 1988. Effects of caryophyllene, caryophyllene oxide, and their interaction with gossypol on the growth and

S148

ANURAG A. AGRAWAL AND MARK FISHBEIN

development of Heliothis virescens (F) (Lepidoptera, Noctuidae). Journal of Economic Entomology 81:93–97. Haddad, N. M., and W. M. Hicks. 2000. Host pubescence and the behavior and performance of the butterfly Papilio troilus (Lepidoptera: Papilionidae). Environmental Entomology 29: 299–303. Heil, M., T. Delsinne, A. Hilpert, S. Schurkens, C. Andary, K. E. Linsenmair, M. S. Sousa, and D. McKey. 2002. Reduced chemical defence in ant-plants? A critical re-evaluation of a widely accepted hypothesis. Oikos 99:457–468. Heil, M., S. Greiner, H. Meimberg, R. Kruger, J.-L. Noyer, G. Heubl, K. E. Linsenmair, and W. Boland. 2004. Evolutionary change from induced to constitutive expression of an indirect plant resistance. Nature 430:205–208. Heil, M., C. Staehelin, and D. McKey. 2000. Low chitinase activity in Acacia myrmecophytes: A potential trade-off between biotic and chemical defences? Naturwissenschaften 87:555–558. Herms, D. A., and W. J. Mattson. 1992. The dilemma of plants: to grow or defend. Quarterly Review of Biology 67:283–335. Housworth, E. A., and E. P. Martins. 2001. Random sampling of constrained phylogenies: conducting phylogenetic analyses when the phylogeny is partially known. Systematic Biology 50:628–639. Janzen, D. H. 1974. Tropical blackwater rivers, animals, and mast fruiting by the Dipterocarpaceae. Biotropica 6:69–103. Koricheva, J., H. Nyka¨nen, and E. Gianoli. 2004. Metaanalysis of trade-offs among plant antiherbivore defenses: Are plants jacks-of-all-trades, masters of all? American Naturalist 163:E64–E75. Kursar, T. A., and P. D. Coley. 2003. Convergence in defense syndromes of young leaves in tropical rainforests. Biochemical Systematics and Ecology 31:929–949. Lavoie, B., and K. S. Oberhauser. 2004. Compensatory feeding in Danaus plexippus (Lepidoptera: Nymphalidae) in response to variation in host plant quality. Environmental Entomology 33:1062–1069. Legendre, P., and L. Legendre. 1998. Numerical ecology. Second English edition. Elsevier Science BV, Amsterdam, The Netherlands. Lewis, P. L. 2001. Maximum likelihood phylogenetic inference: modeling discrete morphological characters. Systematic Biology 50:913–925. Maddison, D. R., and W. P. Maddison. 2003. MacClade. Version 4.06. Sinauer Associates, Sunderland, Massachusetts, USA. Malcolm, S. B. 1991. Cardenolide-mediated interactions between plants and herbivores. Pages 251–296 in G. A. Rosenthal and M. R. Berenbaum, editors. Herbivores: their interactions with secondary plant metabolites. Volume 1. Academic Press, San Diego, California, USA. Malcolm, S. B., B. J. Cockrell, and L. P. Brower. 1989. Cardenolide fingerprint of monarch butterflies reared on common milkweed, Asclepias syriaca L. Journal of Chemical Ecology 15:819–853. Mattson, W. J., Jr. 1980. Herbivory in relation to plant nitrogen content. Annual Review of Ecology and Systematics 11:119– 161. McDade, L. A. 1992. Pollinator relationships, biogeography, and phylogenetics. BioScience 42:21–26. Mooers, A. Ø., S. M. Vamosi, and D. Schluter. 1999. Using phylogenies to test macroevolutionary hypotheses of trait evolution in cranes (Gruinae). American Naturalist 154:249– 259 Morse, D. H. 1985. Milkweeds and their visitors. Scientific American 253:112–119. Nelson, C. J. 1993. Sequestration and storage of cardenolides and cardenolide glycosides by Danaus plexippus and D. chrysippus petila when reared on Asclepias fruticosa: with a review of some factors that influence sequestration. Pages 91– 105 in S. B. Malcolm and M. P. Zalucki, editors. Biology and

Ecology Special Issue

conservation of the monarch butterfly. Natural History Museum of Los Angeles County, Los Angeles, California, USA. Pagel, M. 1999. The maximum likelihood approach to reconstructing ancestral character states of discrete characters on phylogenies. Systematic Biology. 48:612–622. Price, P., and M. Willson. 1979. Abundance of herbivores on six milkweed species in Illinois. American Midland Naturalist 101:76–86. Rapini, A., M. W. Chase, D. J. Goyder, and J. Griffiths. 2003. Asclepiadeae classification: evaluating the phylogenetic relationships of New World Asclepiadoideae (Apocynaceae). Taxon 52:33–50. Rehr, S. S., P. P. Feeny, and D. H. Janzen. 1973. Chemical defence in Central American non-ant-Acacias. Journal of Animal Ecology 42:405–416. Romeo, J. T., J. A. Saunders, and P. Barbosa, editors. 1996. Phytochemical diversity and redundancy in ecological interactions. Plenum Press, New York, New York, USA. Ronquist, F., and J. P. Huelsenbeck. 2003. MrBayes 3: Bayesian phylogenetic inference under mixed models. Bioinformatics 19:1572–1574. Root, R. B. 1967. The niche exploitation pattern of the Bluegray Gnatcatcher. Ecological Monographs 37:317–350. Rothschild, M., J. V. Euw, and T. Reichste. 1970. Cardiac glycosides in Oleander aphid, Aphis nerii. Journal of Insect Physiology 16:1141–1145. Rudgers, J. A., S. Y. Strauss, and J. F. Wendel. 2004. Tradeoffs among anti-herbivore resistance traits: insights from Gossypieae (Malvaceae). American Journal of Botany 91: 871–880. Scha¨dler, M., G. Jung, H. Auge, and R. Brandl. 2003. Palatability, decomposition and insect herbivory: patterns in a successional old-field plant community. Oikos 103:121– 132. Schemske, D. W., and H. D. Bradshaw. 1999. Pollinator preference and the evolution of floral traits in monkeyflowers (Mimulus). Proceedings of the National Academy of Sciences (USA) 96:11910–11915. Schluter, D. 2000. The ecology of adaptive radiation. Oxford University Press, Oxford, UK. Seigler, D. S., and J. E. Ebinger. 1987. Cyanogenic glycosides in ant-Acacias of Mexico and Central America. Southwestern Naturalist 32:499–503. Shimodaira, H., and M. Hasegawa. 1999. Multiple comparisons of log-likelihoods with applications to phylogenetic inference. Molecular Biology and Evolution 16:1114–1116. Silvertown, J., and M. Dodd. 1996. Comparing plants and connecting traits. Philosophical Transactions of the Royal Society of London B 351:1233–1239. Simberloff, D., and T. Dayan. 1991. The guild concept and the structure of ecological communities. Annual Review of Ecology and Systematics 22:115–143. Stapley, L. 1998. The interaction of thorns and symbiotic ants as an effective defence mechanism of swollen-thorn acacias. Oecologia 115:401–405. Steward, J. L., and K. H. Keeler. 1988. Are there trade-offs among antiherbivore defenses in Ipomoea (Convolvulaceae)? Oikos 53:79–86. Swofford, D. L. 2001. PAUP*. Phylogenetic analysis using parsimony (*and other methods). Version 4. Sinauer Associates, Sunderland, Massachusetts, USA. Thaler, J. S., and R. Karban. 1997. A phylogenetic reconstruction of constitutive and induced resistance in Gossypium. American Naturalist 149:1139–1146. Thomson, J. D., P. Wilson, M. Valenzuela, and M. Malzone. 2000. Pollen presentation and pollination syndromes, with special reference to Penstemon. Plant Species Biology 15:11– 29. Tilman, D., and S. Pacala. 1993. The maintenance of species richness in plant communities. Pages 13–25 in R. Ricklefs and

July 2006

PLANT DEFENSE SYNDROMES

D. Schulter, editors. Species diversity in ecological communities. University of Chicago Press, Chicago, Illinois, USA. Twigg, L. E., and L. V. Socha. 1996. Physical versus chemical defence mechanisms in toxic Gastrolobium. Oecologia 108: 21–28. Van Dam, N. M., and J. D. Hare. 1998. Differences in distribution and performance of two sap-sucking herbivores on glandular and non-glandular Datura wrightii. Ecological Entomology 23:22–32. van der Meijden, E., M. Wijn, and H. J. Verkaar. 1988. Defense and regrowth, alternative plant strategies in the struggle against herbivores. Oikos 51:355–363. Van Zandt, P. A., and A. A. Agrawal. 2004a. Specificity of induced plant responses to specialist herbivores of the common milkweed, Asclepias syriaca. Oikos 104:401–409. Van Zandt, P. A., and A. A. Agrawal. 2004b. Community-wide impacts of herbivore-induced plant responses in milkweed (Asclepias syriaca). Ecology 85:2616–2629. Waser, N. M., L. Chittka, M. V. Price, N. M. Williams, and J. Ollerton. 1996. Generalization in pollination systems, and why it matters. Ecology 77:1043–1060.

S149

Weiss, H. B., and E. L. Dickerson. 1921. Notes on milkweed insects in New Jersey. Journal of the New York Entomological Society 20:123–145. Wilbur, H. 1976. Life history evolution in seven milkweeds of the genus Asclepias. Journal of Ecology 64:223–240. Wilkinson, L. 1997. SYSTAT 7.0: the system for statistics. SPSS, Chicago, Illinois, USA. Wilson, P., M. C. Castellanos, J. N. Hogue, J. D. Thomson, and W. S. Armbruster. 2004. A multivariate search for pollination syndromes among penstemons. Oikos 104:345–361. Woodman, R. L., and G. W. Fernandes. 1991. Differential mechanical defense: herbivory, evapotranspiration, and leaf hairs. Oikos 60:11–19. Woodson, R. E., Jr. 1954. The North American species of Asclepias L. Annals of the Missouri Botanical Garden 41:1– 211. Zalucki, M. P., L. P. Brower, and A. Alonso. 2001. Detrimental effects of latex and cardiac glycosides on survival and growth of first-instar monarch butterfly larvae Danaus plexippus feeding on the sandhill milkweed Asclepias humistrata. Ecological Entomology 26:212–224. Zar, J. H. 1996. Biostatistical analysis. Prentice Hall, Englewood Cliffs, New Jersey, USA.

APPENDIX A Means and standard errors for seven defense-related traits of the 24 species of milkweed employed in this study (Ecological Archives E087-116-A1).

APPENDIX B The three defense clusters separated in multivariate space (Ecological Archives E087-116-A2).

Ecology, 87(7) Supplement, 2006, pp. S150–S162 Ó 2006 by the Ecological Society of America

THE GROWTH–DEFENSE TRADE-OFF AND HABITAT SPECIALIZATION BY PLANTS IN AMAZONIAN FORESTS PAUL V. A. FINE,1,2,3,9 ZACHARIAH J. MILLER,3 ITALO MESONES,4 SEBASTIAN IRAZUZTA,5 HEIDI M. APPEL,6 M. HENRY H. STEVENS,7 ILARI SA¨A¨KSJA¨RVI,8 JACK C. SCHULTZ,6 AND PHYLLIS D. COLEY1 1

2

Department of Biology, University of Utah, Salt Lake City, Utah 84112 USA Environmental and Conservation Programs and Department of Botany, Field Museum of Natural History, Chicago, Illinois 60605 USA 3 Department of Ecology and Evolutionary Biology, University of Michigan, Ann Arbor, Michigan 48109-1048 USA 4 Department of Forestry, Universidad Nacional de la Amazonı´a Peruana, Plaza Serafı´n Filomeno 246, Iquitos, Peru 5 Department of Biology, McMaster University, Hamilton, Ontario L8S4K1 Canada 6 Pesticide Research Laboratory, Pennsylvania State University, University Park, Pennsylvania 16802 USA 7 Department of Botany, Miami University, Oxford, Ohio 45056 USA 8 Zoological Museum, Centre for Biodiversity, FIN-20014, University of Turku, Finland

Abstract. Tropical forests include a diversity of habitats, which has led to specialization in plants. Near Iquitos, in the Peruvian Amazon, nutrient-rich clay forests surround nutrient-poor white-sand forests, each harboring a unique composition of habitat specialist trees. We tested the hypothesis that the combination of impoverished soils and herbivory creates strong natural selection for plant defenses in white-sand forest, while rapid growth is favored in clay forests. Recently, we reported evidence from a reciprocal-transplant experiment that manipulated the presence of herbivores and involved 20 species from six genera, including phylogenetically independent pairs of closely related white-sand and clay specialists. When protected from herbivores, clay specialists exhibited faster growth rates than white-sand specialists in both habitats. But, when unprotected, white-sand specialists outperformed clay specialists in whitesand habitat, and clay specialists outperformed white-sand specialists in clay habitat. Here we test further the hypothesis that the growth–defense trade-off contributes to habitat specialization by comparing patterns of growth, herbivory, and defensive traits in these same six genera of white-sand and clay specialists. While the probability of herbivore attack did not differ between the two habitats, an artificial defoliation experiment showed that the impact of herbivory on plant mortality was significantly greater in white-sand forests. We quantified the amount of terpenes, phenolics, leaf toughness, and available foliar protein for the plants in the experiment. Different genera invested in different defensive strategies, and we found strong evidence for phylogenetic constraint in defense type. Overall, however, we found significantly higher total defense investment for white-sand specialists, relative to their clay specialist congeners. Furthermore, herbivore resistance consistently exhibited a significant trade-off against growth rate in each of the six phylogenetically independent species-pairs. These results confirm theoretical predictions that a trade-off exists between growth rate and defense investment, causing white-sand and clay specialists to evolve divergent strategies. We propose that the growth–defense trade-off is universal and provides an important mechanism by which herbivores govern plant distribution patterns across resource gradients. Key words: Amazon; ecological gradient; growth–defense trade-off; habitat specialization; herbivory; phenolics; phylogenetic control; rainforest; reciprocal-transplant experiment; terpenes; tropical trees.

INTRODUCTION The regional diversity of plant species arises, in part, because a given species is restricted to a subset of environmental conditions. But how and why does this habitat specialization occur? The most common explanation is that habitat specialists are physiologically adapted to growing in their particular abiotic environManuscript received 8 February 2005; revised 29 April 2005; accepted 3 May 2005; final version received 11 July 2005. Corresponding Editor: A. A. Agrawal. For reprints of this Special Issue, see footnote 1, p. S1. 9 Present address: Department of Ecology and Evolutionary Biology, University of Michigan, Ann Arbor, Michigan 48109-1048 USA. E-mail: paulfi[email protected]

ment and out-compete other plants that are not so closely suited to the local conditions (Ashton 1969, Cody 1978, Bunce et al. 1979). However, herbivore– plant interactions can also contribute to the evolution of habitat specialization. Theoretical work has demonstrated that herbivores can alter competitive relationships among plants, especially when there is spatial heterogeneity of resources (Louda et al. 1990, Grover and Holt 1998). Empirical studies at the population and community levels have documented that herbivores can reduce plants’ potential distributions, restricting them to a subset of the habitats that they might physiologically tolerate (Parker and Root 1981, Louda 1982, 1983, Louda and Rodman 1996, Olff and Ritchie 1998,

S150

July 2006

PLANT TRADE-OFFS AND SPECIALIZATION

Carson and Root 2000, Harley 2003). Thus, herbivores can play a major role in determining which species of plants dominate in a community, as well as in which habitats a species will be successful. The lowland Amazonian rainforest near Iquitos, Peru provides an ideal system to study habitat specialization and the role of herbivores. Forests in the Iquitos area grow on a mosaic of soil types; including red clay soils and extremely infertile white-sand soils (Kauffmann et al. 1998). The two soil types lie immediately adjacent to each other, the boundaries are well defined, and each soil type is associated with a distinctive flora (Gentry 1986, Va´squez 1997, Fine 2004). White-sand forests are much more resource limited than clay soil forests (Medina and Cuevas 1989, Coomes and Grubb 1998, Moran et al. 2000). Resource availability theory proposes that resource-limited species will have slower growth rates and higher optimal levels of defense, reflecting the decreased ability of a resource-limited plant to compensate for tissues lost due to herbivory (Janzen 1974, Coley et al. 1985, Coley 1987b). Thus we predict that species growing in white-sand forests should evolve to allocate relatively more resources to defense than species growing in clay forests (Fine et al. 2004). Recently, we reported the results of a reciprocaltransplant experiment of 20 species of seedlings from six genera of phylogenetically independent pairs of whitesand and clay specialist plants (Fine et al. 2004). We manipulated the presence of herbivores and found that clay specialists grew significantly faster than did whitesand specialists in both habitats when protected from herbivores. But when herbivores were not excluded, white-sand specialists out-performed clay specialists in white-sand forests, and clay specialists grew faster than white-sand specialists in clay forests. These results strongly supported the existence of a growth–defense trade-off, with habitat specialization being enforced by herbivores (Fine et al. 2004). Here, we test further the predictions of the growth– defense trade-off by comparing species-level patterns of growth, herbivory, and defense in this same phylogenetically diverse group of tree species. We predicted that closely related species specialized to contrasting soil types should diverge in traits that confer defense vs. those that confer growth. We investigated the evidence for such differential investment while controlling for phylogeny. Therefore, any differences in defense allocation found between closely related white-sand and clay specialists can be inferred to be traits derived for habitat specialization. This phylogenetically controlled approach enabled us to investigate the degree of constraint involved in the type and amount of defense, and to separate this from the repeated and independent evolution of defensive traits due to selection from similar ecological conditions. Second, examining defense investment with a reciprocal-transplant experiment allowed us to identify which traits (if any) are phenotypi-

S151

cally plastic as opposed to genetically controlled adaptations to a particular habitat. Thus, to test whether the growth–defense trade-off contributes to habitat specialization in white-sand and clay forests, we combined field observations and a reciprocal-transplant experiment to ask the following questions: (1) Are there differences in herbivore abundance in the two habitats? (2) Is there a difference in the impact of herbivory in the two habitats, suggesting selection for greater defense investment in white-sand habitats? (3) Do white-sand and clay specialists differ in their type of defensive strategy or in their amount of defense investment? Are these differences phylogenetically constrained or repeatedly and independently evolved? (4) Are defensive traits in white-sand and clay specialists affected by resource-driven phenotypic plasticity? (5) Do white-sand and clay specialists follow the predictions of the growth–defense trade-off? MATERIALS

AND

METHODS

Study site and study species We conducted this research in the AllpahuayoMishana National Reserve near Iquitos, Peru (3857 0 S, 73824 0 W). This 57 600-ha reserve is at ;130 m elevation and receives more than 3000 mm of precipitation during the year, with no distinct dry season (Marengo 1998). Many white-sand specialist trees belong to the same genera as neighboring clay forest specialists, allowing for a phylogenetically controlled experiment using edaphic specialist species. For a reciprocal-transplant experiment, we chose 20 common white-sand and clay specialists from six genera from five families (see Fine et al. [2004] for a phylogeny). The genera were Mabea (Euphorbiaceae), Oxandra (Annonaceae), Pachira (Malvaceae sensu lato), Parkia (Fabaceae), Protium (Burseraceae), and Swartzia (Fabaceae). Each genus was represented by one white-sand specialist and one clay specialist, except for Protium, which was represented by six clay specialists and four white-sand specialists. Designation of habitat for each species was accomplished by extensive inventories (Fine 2004, Fine et al. 2005) as well as consultation of local floras and other published species lists from the western Amazon (Va´squez 1997, Ruokolainen and Tuomisto 1998, Jørgensen and Le´on-Ya´nez 1999, Garcı´ a et al. 2003). Nitrogen availability To test for differences in nitrogen availability between white-sand and clay habitats, we filled 27 nylon stocking bags filled with 8 g of Rexyn 300 (H-OH) analytical grade resin beads. In May 2002, we placed the ionexchange resin bags beneath the litter layer and root mat at the organic material–mineral soil interface at our white-sand and clay sites (Binkley and Matson 1983). The bags were collected after five weeks, extracted with KCl, and measured by standard techniques with an autoanalyzer (University of Wisconsin Soils Laboratory). Nitrate, ammonium, and root mat depth differ-

S152

PAUL V. A. FINE ET AL.

ences were tested for significance between soil types with a Wilcoxon signed-ranks test. The reciprocal-transplant experiment We used a reciprocal-transplant experiment to test whether white-sand and clay specialists had different growth rates and defense investments as predicted by the growth–defense trade-off hypothesis. In addition, the reciprocal-transplant experiment allowed us to test for phenotypic plasticity of defense investments under different edaphic and herbivore treatments. In May 2001, we built 22 control and 22 herbivore exclosures (3 3 3 3 2 m); half were located in clay forest and half in white-sand forest. We transplanted 880 seedlings from the six genera into the controls and exclosures (see Fine et al. 2004). Using the results of the reciprocal-transplant experiment reported in Fine et al. (2004), we compared the amount of leaf and height growth of the plants grown in herbivore exclosures to the unprotected controls, and estimated the effect herbivory had on growth rates for each white-sand and clay specialist. This measure is referred to throughout as ‘‘protection effect.’’ The experiment lasted until February 2003 (21 mo after transplanting, 18 mo after first data collection), at which point leaves were collected to measure defensive traits. Insect abundance and species richness To evaluate differences in insect abundance and composition across habitats, we used a portable black light attached to a battery to attract insects in five whitesand and five clay sites. During 8–20 December 2002, on rain-free evenings between 1900–2000, the black-light was illuminated and suspended above white sheets. We collected all insects from the families/orders Blattoideae, Coleoptera, Hemiptera, Homoptera, and Orthoptera. We excluded all obvious predators and collected all herbivorous insects from these five groups and counted and identified them to order and family and then separated them into morphospecies. Parasitoid wasps were collected with malaise traps over a two-year period in 15 white-sand and nonwhite-sand forest sites in the Allpahuayo-Mishana National Reserve (from 15 of the same sites described in Fine [2004]) as a part of a much larger study on ichneumonid wasps (for detailed methods see Sa¨a¨ksja¨rvi [2003]). Since these parasitoid wasps attack either herbivorous insects (or predators of herbivorous insects), we would expect parasitoid diversity and abundance to track herbivorous insect diversity and abundance in white-sand and clay forests. To test for differences between white-sand and clay habitats (both the black light trap data and the wasp data), Wilcoxon signed-ranks tests were conducted on the ranked abundances and numbers of species. Field herbivory For herbivory comparisons in addition to those from the transplant experiment, we chose six genera that were

Ecology Special Issue

common in both white sand and clay forests: Protium (Burseraceae), Hevea and Mabea (Euphorbiaceae), Pachira (Malvaceae s.l.), and Macrolobium (Fabaceae). In September 2000, in the same white-sand and clay sites where the wasps were collected, we sampled 355 individuals in the field from .20 species of Protium, two species of Hevea, two species of Mabea, two species of Pachira, and three species of Macrolobium. Most of these species were found in only one of the two habitats. Plants were 1–3 m tall (juvenile trees). We marked newly expanding leaves (or leaves that had already expanded but were not toughened) with small colored wires, from 1–10 leaves or leaflets per plant. After five to seven weeks we estimated the amount of leaf area missing from the marked leaflet (0–100%). Average amount of leaf area missing was divided by number of days between marking and the census (damage per day). These data were arcsine square-root transformed to improve normality, and a mixed-model ANOVA (including the random factor genus and the fixed factor habitat) was performed on the data to test for differences in herbivory rate between white-sand and clay habitats. Impact of herbivory (defoliation experiment) In February 2003, after collecting leaf material for chemical analyses from all of the seedlings in the reciprocal-transplant experiment, we removed 100% of the remaining leaves to test the effect of defoliation on white-sand and clay specialists in the two habitats. After three months, we counted the number of seedlings that survived defoliation. To compare mortality rates, we averaged mortality for white-sand specialists and clay specialists in each of the 44 controls and exclosures (Protium species in each control and exclosure were weighted to give each genus equal importance in the analyses). A fixed-factor ANOVA including the terms habitat (white-sand or clay), origin (white-sand or clay specialist), and the origin 3 habitat interaction was used to assess the effects of origin and habitat on mortality due to defoliation. Post hoc tests on the individual group means were performed using the studentized t distribution (appropriate for equal sample sizes; Sokal and Rohlf 1995). Defensive characteristics of white-sand and clay specialists Comparing differences in herbivory and growth is the best method of comparing defense investment in whitesand and clay specialists, since this approach takes into account the entire arsenal of plant defenses as experienced by the actual herbivores (cf. Simms and Rausher 1987). However, to investigate which particular defensive traits are deterring herbivores, we measured two classes of chemical defenses, a physical defense, and the nutritional quality of white-sand and clay specialists. After the transplant experiment was completed, we collected leaves from all surviving plants to compare defense investment in white-sand and clay specialists,

July 2006

PLANT TRADE-OFFS AND SPECIALIZATION

and to assess the effect of habitat and treatment on the plasticity of defense investment for each species. We collected marked mature leaves that were produced after plants were transplanted. We measured terpenes, total phenolics, toughness, and available protein for all seedlings in the reciprocal-transplant experiment. Terpenes and phenolics are carbon-based secondary compounds common to many families of plants, including those in our research (Mabry and Gill 1979, Bernays et al. 1989, Schultes and Raffauf 1990). Although phenolics and terpenes have alternative functions, they commonly function in herbivore deterrence (Mabry and Gill 1979, Bernays et al. 1989, Herms and Mattson 1992, Langenheim 1994; but see Harborne 1991, Close and McArthur 2002). Increased toughness of leaves (sclerophylly) is a mechanical antiherbivore defense that is commonly found worldwide in plants that live in resource-limited environments (Coley 1983, 1987a, Grubb 1986, Turner 1994). Finally, available foliar protein is a good measure of a plant’s nutritional quality. Moran and Hamilton (1980) hypothesized that plant nutrition can be considered a defensive trait if it can be selected for by herbivore attack. This can result if herbivores detect nutritional differences and prefer plants with higher nutrition (cf. Scheirs et al. 2003). A second mechanism is if slow growth by herbivores due to low nutrition results in higher predation rates (cf. Denno et al. 2002). Chemical defenses To compare terpene investment among the species, ;500 mg (fresh mass) leaves from the experimental seedlings were collected at the experimental sites in 2-mL glass vials and filled with dichloromethane (DCM). This leaf–DCM mixture was used for qualitative and quantitative analyses with gas chromatography (GC) and gas chromatography–mass spectrometry (GCMS). (See Appendix A for detailed methods of terpene extraction and analysis.) For comparisons of total phenolics, ;2 g fresh mass of mature leaves of 16 individuals (8 protected and 8 unprotected) from each species in the reciprocal-transplant experiment were collected and immediately placed in plastic tubes containing silica gel desiccant. Leaves were later analyzed for phenolic compounds in the Appel/Schultz laboratory at Penn State University. Whenever possible, bulk tannins were prepared to provide standards for the phenolic assays of individual samples. This is a crude purification, and although nonphenolic materials are unlikely to be present (Hagerman and Klucher 1986; H. M. Appel and J. C. Schultz, unpublished data), the product is merely a more representative sample of extractable polyphenols found in the actual plant than is a commercial standard from some other source. (See Appendix A for detailed information on all methodology of phenolic extraction, purification, and analysis.) Because total phenolics likely function as an antiherbivore defense by precipitating available protein (Herms and Mattson 1992), we divided

S153

our total phenolics obtained as described with available foliar protein data to create a phenolic : protein ratio (Nichols-Orians 1991). Leaf toughness A ‘‘penetrometer’’ (Chatillon Universal Tension and Compression Tester, Largo, Florida, USA) was used to puncture holes through the leaf (or leaflet) lamina to give a measure of toughness. It was impossible to test the pair of species from the genus Parkia, since both have bipinnately compound leaves, with leaflets not much larger than the 3 mm diameter of the testing machine’s rod. We standardized the punch position to midway between leaf tip and base, between the midrib and the leaf margin, avoiding the main veins where possible. The punch test measures a combination of shear and compressive strength and resistance to crack propagation. For these reasons, it has been criticized as not specifically measuring leaf toughness (Choong et al. 1992). Nevertheless, it is easy to perform in the field and highly correlated with more specific tests to measure the physical properties of leaf toughness (shearing and tearing parameters) (Edwards et al. 2000). Soluble protein assays The amount of available foliar protein was measured at the Appel/Schultz laboratory using the same driedleaf samples collected for the phenolics analyses. (See Appendix A for detailed methods.) Statistical analyses of growth and defensive traits Clay and white-sand specialists in each of the six genera were the experimental unit. Because there were four white-sand specialists and six clay specialists from the genus Protium, the responses for all Protium individuals were weighted to give each genus equal importance in the analysis. The four white-sand specialist Protium species were weighted at 0.25, the six clay specialist Protium species were weighted at 0.167, and species from all other genera were weighted at 1. We used fixed factor ANOVAs to test for genus, origin (the difference between white-sand specialists and clay specialists), habitat (whether species responded differently depending on where they were planted), and treatment (whether defense investment differed depending on whether the plants were exposed to herbivores). Since we had a priori knowledge that different genera would have different defensive strategies (i.e., some species have terpene investment, others do not), we used fixed-factor ANOVAs for defensive traits (genus was treated as a fixed factor), since our ability to generalize our results in these analyses to unsampled genera is limited. Subsequent to the overall test, individual group means were compared with Tukey hsd post hoc tests. Defense index Because different species of plants are likely to employ different defensive strategies, we therefore devised a

S154

PAUL V. A. FINE ET AL.

simple method to combine all measures of chemical defense, leaf toughness, and available protein to investigate whether, for each genus, white-sand specialists were more defended than clay specialists. Values for phenolics, terpenes, and leaf toughness were averaged across both habitats and Z-transformed to give the defense traits among the six pairs of white-sand and clay specialists a mean of zero and a standard deviation of one. Missing data was scored as zero. For available protein, we standardized the inverse of the species averages, because a larger amount of available protein corresponds to lower defense. All four standardized defense variables were then summed to create a defense index (DI). For each genus, the DI for the clay specialist was subtracted from the DI of the white-sand specialist. This method has the assumption that each of these four measures has equal weight, which is undoubtedly incorrect, but preferable than subjectively assigning different weights to defense types. These difference scores (DIWS – DICL) were used to test the prediction that white-sand specialists are more defended than clay specialists with a one-tailed Wilcoxon paired signedranks test (Zar 1999). Phylogenetic independence of growth, herbivory, and defense traits In order to evaluate whether growth, herbivory, and defense traits were more similar in closely related genera, we mapped each of the indices listed above, as well as each individual defensive trait onto a phylogeny representing the relationships among the six genera and 20 species (see Fine et al. 2004). Using the program Phylogenetic Independence 2.0, we tested whether traits exhibited significant phylogenetic independence by comparing the average contrast values (C-stat) among the actual trait values for the plant species to the distribution of contrast values created by randomly placing the trait values at the tips within the topology 2000 times and testing for serial independence (TFSI) (Abouheif 1999). If a trait is significantly phylogenetically constrained, then the average C-stat for the actual value will be greater than 95% of the average contrast values generated by the randomization. Correlations of growth, defense, and herbivory data for the six genera Species averages for growth (leaf area and height, averaged across habitats), the effect of herbivore protection on growth (arithmetic difference between the average leaf area and height with and without protection, for each white-sand and clay genus averaged across habitats) and defenses, as described, were Ztransformed and analyzed by a method analogous to phylogenetically independent contrasts (Harvey and Pagel 1991). To test for trade-offs, we plotted the values for the six species pairs and analyzed the six slopes, to see if the relationship between traits was consistent when controlling for phylogenetic relationship. We used these

Ecology Special Issue

plots to test our predictions that (1) growth and herbivory would be positively correlated, (2) growth and defense would be negatively correlated, and (3) herbivory and defense investment would be negatively correlated. Hypotheses about the correlations of traits were tested by the difference scores of the slopes and were evaluated for significance with one-tailed Wilcoxon paired sign-rank tests (Zar 1999). RESULTS Differences in nutrient availabilities Clay forest sites contained significantly more available nitrogen (Z ¼ 3.53, P , 0.0004) than white-sand forests, more than twice as much available ammonium (Z ¼ 2.71, P , 0.0061), more than an order of magnitude more available nitrate (Z ¼ 3.59, P , 0.0003), and a much thinner root mat (Z ¼ 4.89, P , 0.0001; Table 1). Habitat differences in herbivore abundance We found no significant differences in herbivore abundance or species richness between habitats for all herbivores or any of the six orders of herbivorous insects that we collected (P . 0.05, Wilcoxon signed-ranks tests; Table 1). Of the 311 morphospecies collected, 208 were collected only once (67%). Of the morphospecies collected more than once, 41 were collected only in white-sand forest, 28 were collected only in clay forest, and 34 were collected in both forests (33% of the morphospecies collected more than once). For parasitoid wasps, no statistical differences in abundance or morphospecies diversity were found between white-sand and nonwhite-sand forest sites (Table 1). Moreover, in the reciprocal-transplant experiment, mean effect of protection for white-sand and clay specialists did not change between habitats (Fig. 1a, b). Differences in the magnitude of herbivore attack Clay specialists showed an average increase in growth of 0.25 cm2/d in leaf area (paired t test, df ¼ 5, t ¼
2.91, P , 0.05) and 0.0018 cm/d in height (paired t test, df ¼ 5, t¼
2.59, P , 0.05) when protected from herbivores, while white-sand specialists grew just as well or better in the unprotected vs. protected treatments. When the effect of herbivore protection on leaf area and height growth are Z-transformed and summed, all genera show the same pattern that clay specialists received a greater benefit from herbivore protection than did white-sand specialists. During our study of field herbivory rates in the two habitats, plants in clay forest sites suffered more than twice the herbivory rates on their new leaves than did plants in white-sand sites (mixed model ANOVA, F1, 349 ¼ 6.69, P , 0.01). Clay plants lost almost 23% of their new leaves per month, while white-sand plants lost slightly .10% (Table 1). Habitat differences in the impact of herbivory As predicted, seedlings overall suffered higher mortality due to total defoliation in white-sand habitat than

July 2006

PLANT TRADE-OFFS AND SPECIALIZATION

S155

TABLE 1. Comparisons of white-sand and clay forests for leaf litter depth, nitrogen availability, young-leaf herbivory, and insect abundance and morphospecies richness (means 6 SE reported). Variable

Clay forest sites a

Root mat (cm) (N ¼ 44 plots)

0.91 6 1.0

Nitrogen availability (ppm) from ion-exchange resin bags (N ¼ 27 resin bags) NO3– NH4þ Total nitrogen

349.2 6 25.7 135.2 6 32.7 484.4 6 43.0

8.48 6 0.6 b b b

b

Herbivory (% leaf eaten/mo) (N ¼ 355 individuals) Insect herbivore abundance ((no. individuals)(light trap)
1h
1) Total insect herbivore abundance Blattoid abundance Coleopteran abundance Hemipteran abundance Homopteran abundance Orthopteran abundance

87.2 3.0 20.0 7.6 20.0 36.6

6 6 6 6 6 6

12.6 a 0.7 a 9.0 a 4.9 a 4.5 a 8.2 a

Insect herbivore species richness ((no. morphospecies)(light trap)
1h
1) Total insect herbivore morphospecies Blattoid morphospecies Coleopteran morphospecies Hemipteran morphospecies Homopteran morphospecies Orthopteran morphospecies

45.0 2.6 7.6 3.2 15.8 15.8

6 6 6 6 6 6

4.3 0.5 2.1 1.0 2.8 1.7

Parasitoid wasp ((no. individuals)site
1(malaise trap)
1 for 2 yr) Total parasitoid wasp abundance Total parasitoid species and morphospecies

67.7 6 28.5 a 25.5 6 6.4 a

22.8 6 4.3

White-sand forest sites

a a a a a a

b

25.6 6 13.8 62.1 6 17.5 87.7 6 23.0 10.3 6 3.3

a a a

a

74.8 2.6 22.4 13.4 17.0 19.2

6 6 6 6 6 6

18.1 a 0.9 a 7.8 a 11.7 a 1.9 a 3.6 a

34.8 2.4 8.0 2.0 11.4 10.8

6 6 6 6 6 6

3.9 0.8 2.0 0.4 2.0 2.1

a a a a a a

59.9 6 10.8a 22.0 6 3.3 a

Note: Significant differences between forests are indicated by different superscript letters within a row (mixed-model ANOVA, effect of habitat for herbivory, Wilcoxon signed-ranks tests between habitats for litter depth, nitrogen availability, insect abundance, and species richness).

they did in clay habitat (effect of habitat, F1,84 ¼4.96, P , 0.05). In addition, white-sand specialists suffered significantly more mortality than did clay specialists in both habitats (effect of origin, F1,84 ¼ 22.8, P , 0.0001; Fig. 2). Differences in defense investment Type of defense.—We found strong evidence for phylogenetic constraint for type of defense. The main effect of genus was always significant for differences in terpenes, phenolics, leaf toughness, and available foliar protein. Moreover, it is clear that different genera are relying on different defense strategies, as each of the six genera had a distinct defense investment pattern (see Appendix C). For example, only two genera, Oxandra and Protium, contained measurable terpenes identified by GCMS (Appendix C). Similarly, only two genera, Pachira and Parkia had white-sand specialists with obviously tougher leaves than clay specialists. The pattern of high phenolic investment and low available foliar protein in white-sand specialists was more consistent across the six genera, but still there were exceptions (Oxandra and Protium for phenolics, Mabea for available protein; Fig. 3). Whereas different genera invest in different defensive strategies, we found no consistent relationship between any particular defensive traits that would suggest either a negative trade-off or a synergistic relationship between defense types (Fig. 3). Amount of defense investment.—We found that fivesixths of the genera have a higher defense index (DI) in

the white-sand congener than in the clay congener, and that our prediction of higher defense in the white-sand is supported (one-tailed Wilcoxon paired signed-ranks test, T0.05(1), 6 ¼ 1, P , 0.05, Fig. 3). For phenolic compounds, white-sand specialists overall had significantly higher values for both total phenolics (effect of origin, F1, 292 ¼ 50.3, P , 0.0001) and phenolic : protein ratios (F1, 292 ¼ 128.2, P , 0.0001) with, respectively, three-sixths and four-sixths of the genera exhibiting significant relationships in the predicted direction (Fig. 3, see Appendix D). The two genera that invested in terpenes, Protium and Oxandra, exhibited very different patterns of terpene investment in their white-sand and clay specialists (see Appendix D). Oxandra xylopiodies, the clay specialist, had significantly higher sesquiterpene and total terpene concentrations than O. euneura, the white-sand specialist (P , 0.05, Tukey tests; see Appendix D). In contrast, Protium white-sand specialists had higher monoterpene and total terpene concentrations than did Protium clay specialists (P , 0.05, Tukey tests; see Appendix D). Both Oxandra and Protium white-sand species had significantly higher concentrations of diterpenes and other resins than did their respective clay specialists (see Appendix D). There was no overall effect of origin on leaf toughness (see Appendix D). In contrast, white-sand species had lower available protein in their leaves than clay specialists (significant effect of origin, F1, 292 ¼ 393.5, P , 0.0001; see Appendix D).

S156

PAUL V. A. FINE ET AL.

Ecology Special Issue

FIG. 1. The effect of origin and habitat in the reciprocal-transplant experiment for (a) the effect of herbivore protection on leaf growth (cm2/d), (b) the effect of herbivore protection on height growth (cm/d), (c) total terpenes and resins (mL terpenes/mg of dry leaf material), (d) total phenolics (g phenolics/g dry leaf material), (e) phenolic : protein ratio (phenolics divided by available protein, a unitless ratio), (f) leaf toughness (grams of mass to punch a 3-mm rod through a leaf; 1.0 g ¼ 1.38 kPa), and (g) available protein (g soluble protein/g dry leaf material). Histograms show means 6 SE. Different letters above bars indicate significant differences among the different groups (Tukey tests).

Defensive traits and phenotypic plasticity There was no significant overall effect of habitat for terpenes (Fig. 1c). Aside from the outlier behavior by one species, there was no evidence of phenotypic plasticity in phenolic investment (Fig. 1d). Swartzia cardiosperma is the only species of 20 in the experiment that showed a significant effect of habitat for phenolic : protein ratios (see Appendix C).

The effect of habitat on leaf toughness was highly significant (F1, 388 ¼ 51.6, P , 0.0001; Fig. 1f). Sixteen of 17 species measured had greater leaf toughness in whitesand than clay habitat; three of those were significant (see Appendix C). In contrast, even though nitrogen availabilities differed by more than five times in the two habitats, there was no significant effect of habitat on available protein for either white-sand or clay specialists (Fig. 1g).

July 2006

PLANT TRADE-OFFS AND SPECIALIZATION

S157

Phylogenetic independence of growth and defensive traits

FIG. 2. Mortality results of the 100% defoliation experiment. Bars show average mortality (6SE) for each origin and habitat combination. Different letters above bars indicate significant differences (post hoc tests, studentized t distribution).

Evaluating the trade-off: growth vs. defense vs. herbivory The growth index (GI) and the herbivory index (HI) showed a significant positive relationship (all six of the genera with positive slope, T0.05(1), 6 ¼ 0, P , 0.025; Fig. 4a). There was a significant negative trade-off between GI and total DI, with five-sixths of the genera showing a negative slope (T0.05(1),6 ¼ 1, P , 0.05, Wilcoxon paired signed-ranks test; Fig. 4b). Finally, DI showed a significant negative relationship with HI (T0.05(1), 6 ¼ 1, P , 0.05; Fig. 4c).

There was evidence for significant phylogenetic dependence for total phenolics (C-stat ¼ 0.34, P , 0.002), terpenes (C-stat ¼ 0.34, P , 0.002), and leaf toughness (C-stat ¼ 0.32, P , 0.012). The defense index (C-stat ¼ 0.23, P , 0.11) and available protein (C-stat ¼ 0.11, P , 0.148) exhibited a trend toward phylogenetic constraint. We found no evidence for phylogenetic constraint in GI (C-stat ¼ 0.07, P , 0.35) and the protection effect index (C-stat ¼ 0.01, P , 0.399), results that in part might reflect an artifact of our design because our sampling within each genus was limited to paired white-sand and clay specialists, which maximized the variation between closely related species. DISCUSSION Habitat differences in herbivore populations Two separate measures of herbivorous insect communities found statistically similar diversity and abundance in the two forest types. In addition, a full third of the morphospecies that were collected more than once occurred in both habitats. These results are likely explained by the large home range and dispersal capabilities of many herbivorous insects (Stork 1988), coupled with the fact that most white-sand forest

FIG. 3. The defense index (DI) scores for each genus are plotted, showing the difference between clay (CL) and white-sand (WS) specialists. The three-letter labels of the lines correspond to the genus table below the plot. Black boxes in the table indicate a significantly higher defensive trait for that genus in the white-sand specialist, and shaded boxes indicate a significantly higher defensive trait for the clay specialist (contrary to predictions). The final column shows the DI scores for each genus, with a negative number signifying a score in the direction contrary to predictions.

S158

PAUL V. A. FINE ET AL.

Ecology Special Issue

insect herbivores indeed range into white-sand forests. Moreover, these patterns are consistent with our herbivory data from the reciprocal-transplant experiment showing that clay specialist seedlings were attacked at similar frequencies whether they were transplanted into clay or white-sand forests (Fig. 1a, b). Habitat differences in the impact of herbivory We predicted that the impact of herbivory would be greater in a white-sand forest, because it is more difficult for plants to replace the nitrogen lost to herbivores (Coley 1987b, Craine et al. 2003). This prediction was supported by the fact that all plants transplanted into white-sand forest had significantly higher mortality when defoliated than those transplanted into clay forest (Fig. 2). In the defoliation experiment, white-sand specialists suffered a significantly higher mortality rate than did clay specialists (Fig. 2), confirming a key prediction of the growth–defense trade-off that white-sand specialists ought to have more difficulty replacing tissue lost to herbivores (Coley et al. 1985). This differential response to defoliation by species adapted to low-fertility soils vs. species adapted to high-fertility soils was also found in a study in Singapore (Lim and Turner 1996). Thus, when heavily defended white-sand species are defoliated, they lose costly leaves that represent a high percentage of their energy budget. Due to their slow growth rate, they are then unable to compensate, and this in turn increases their mortality rate (Coley et al. 1987b). For this reason the impact of herbivory appears to be substantially greater for plants adapted to low-resource conditions. Differences in defense investment

FIG. 4. Plots of the six species pairs for (a) growth rate index vs. protection effect index, (b) growth rate index vs. defense index, and (c) herbivory vs. defense index. The consistency and magnitude of these slopes were used to test the predictions of the growth–defense trade-off hypotheses. The three-letter labels correspond to the six genera listed in Fig. 3.

habitats in the Iquitos area are only a few square kilometers. It is important to recognize that our herbivore sampling was extremely limited and precludes us from drawing definitive conclusions concerning the relative abundance of herbivore populations in whitesand and clay habitats. Nevertheless, our herbivore estimates represent two independent corroborations that

Type of defense.—Different genera adopted dramatically different defensive strategies. There was a consistent signal of phylogenetic constraint in our analyses of plant defenses, as genus was a significant factor in each defense variable (see Appendices B and D), and tests for phylogenetic independence confirmed this. In terms of terpenes, phenolics, toughness, and low nutrition, there was no consistent ‘‘syndrome’’ of defensive investment in the six genera; instead, each genus allocated to different combinations of these (and presumably other unmeasured) traits. Indeed, there is little theoretical or empirical support for the idea of a general negative trade-off between types of defensive strategies (Koricheva et al. 2004, Agrawal and Fishbein 2006). Amount of defensive investment.—For Protium, we found higher concentrations of terpenes in white-sand specialists as predicted, but for Oxandra the reverse pattern was found (Fig. 3). The terpene profile of Oxandra is driven by sesquiterpenes, which could possibly be serving a function other than defense, or do not function in a dosage-dependent fashion (Gershenzon and Croteau 1991, Langenheim 1994). In contrast to sesquiterpenes, both Protium and Oxandra

July 2006

PLANT TRADE-OFFS AND SPECIALIZATION

white-sand specialists were found to have higher diterpenes and other resins compared to clay specialists (see Appendix B). Diterpenes are not volatile and are thought to be either toxic (Lerdau and Penuelas 1993) or a type of physical defense against herbivores or pathogens (Langenheim 1994). Total phenolics and phenolic : protein ratios were significantly higher overall for white-sand specialists than for clay specialists (see Appendix B). In our study, percentage dry mass of total phenolics ranged 3–37%, a large range that is certainly an overestimate and highlights the difficulty of precise phenolic quantification in the laboratory (Appel et al. 2001). Finally, we found significantly less available protein in white-sand specialists. This was the most consistent trait, with fivesixths of the species pairs showing the same pattern (see Appendix D). Defensive traits and phenotypic plasticity We did not find many cases of phenotypic plasticity in the seedlings’ allocation to chemical defenses. Very few species had significant increases or decreases in terpenes or phenolics due to habitat (Fig. 1c, d). Similarly, available foliar protein did not change depending on where the seedlings were planted (Fig. 1g), even though nutrient levels were significantly different between the two habitats. We conclude that, for the genera in our study, herbivore resistance due to chemical defenses and available protein content is due to genetically based, fixed traits (but see Boege and Dirzo 2004). Thus, defense differences result from natural selection by herbivores and are not just passive responses to differences of available nutrients in the soils. In contrast to our results with chemical and nutritional defenses, we found a strong overall effect of habitat on leaf toughness, which was significant for three species (Fig. 1f; see Appendix C). Overall, we found that leaf toughness was significantly higher for white-sand species in only two genera, Parkia (which we were not able to measure with our penetrometer, but for which the pattern was obvious) and Pachira. In contrast, two previous studies found that white-sand plants had significantly tougher leaves than clay plants (Coley 1987a, Choong et al. 1992). These studies did not take phylogeny into account, but their results for white-sand and clay species averages were much more divergent than ours. One possibility for the discrepancy is that toughness in these two studies was only measured in the plants’ home habitats. While our results in no way negate the potentially strong selective effect of herbivores on sclerophylly, they do suggest that future comparisons of white-sand and clay species should not only be controlled for phylogenetic relationships, but also for resource availability. Evaluating the growth–defense trade-off The evolutionary trade-off between growth and defense is illustrated by the data graphed in Fig. 4.

S159

When the protection effect of each species is plotted against the overall growth rate (Fig. 4a), all six genera exhibited a positive relationship. In each genus, herbivores selectively attacked the faster growing species more than the slower growing species. This is evidence that faster growing plants have lower resistance to herbivores, consistent with the predictions of the growth–defense trade-off. Coley (1983, 1987b) found the same relationship in Panama where the growth rates of 40 species of trees were positively correlated with their rates of herbivory. In the graphs of Fig. 4a, the lengths of the lines correspond to the amount of variation in growth rate and antiherbivore traits within the species (white-sand and clay) of each genus. For example, some genera like Parkia are represented by longer lines in the horizontal direction (Fig. 4a), because this genus includes both shade-tolerant species and those that thrive in high-light conditions. Therefore, the clay specialist in Parkia is a very fast grower relative to the Protium and Swartzia species, all of which are shade-tolerant species and never found in tree-fall gaps (P. V. A. Fine, personal observation). Yet the fact that the slopes of the six lines in Fig. 4a are so similar suggests the existence of a universal trade-off, even among species with such disparate growth rates and defensive strategies. When the defense index (DI) scores for the six genera are plotted against their growth rate index (GI) (Fig. 4b), we found a significant negative relationship, with five of the six genera having higher DI scores in the slower vs. faster growing species. The slopes in this graph exhibit much more variation than the growth vs. herbivory graph (Fig. 4a), likely due to the coarse method by which we attempted to quantify defense investment in these species. The one outlier genus, Mabea, shows the opposite relationship than the other five genera, with a higher DI score in its faster growing, clay specialist. Because the slower growing (white-sand specialist) Mabea received the least amount of attack from herbivores in the experiment (see Appendix B), it seems likely that it actually is very well defended and we failed to accurately quantify its defensive investment. One reason for this may be that Mabea is the only genus of the six that produces copious white latex, and we did not quantify this trait in our comparisons. The herbivory vs. defense graph (Fig. 4c) echoes this point, with Mabea the only genus whose DI score does not match its herbivory index score. Phylogenetic approach to studying the growth–defense trade-off Our approach using multiple pairs of congeners from ecologically divergent habitats differs from some other more quantitative approaches that have used data on branch lengths from a phylogenetic tree to test for correlations between particular traits and habitat association (Cavender-Bares et al. 2004). In our approach, we ignore branch lengths by design, since

S160

PAUL V. A. FINE ET AL.

each of our genus pairs includes just one representative from each habitat type. But in terms of comparisons of growth rate, herbivory, and overall defense as it relates to white-sand and clay specialization, our results indicate that variation in branch lengths among our pairs matters very little: All six pairs exhibit similar trade-offs (Fig. 4a). Moreover, if this trade-off has a bearing on a plant’s distribution onto white-sand and clay soils, then evidence for it must be present both in the most recently derived specialist pairs as well as in pairs of species that have persisted for millennia in their particular habitats. By contrast, if we were interested in the evolution of particular traits (like phenolics per se), then inclusion of some estimate of divergence time (and denser sampling within genera) would certainly be warranted. One limitation of the congeneric pair approach is that one’s sample is limited to genera that include species that occur in both of the habitats of interest. It would be interesting to compare genera that were restricted to only white-sand or only clay habitats to see if the growth–defense trade-off was evident in comparisons with their sister taxa (that were specialists to the contrasting habitat). Our way of calculating a defense index (DI) works well precisely because defensive traits are phylogenetically conserved between close relatives, allowing for quantitative comparisons of the same qualitative trait. If we used pairs of taxa that were not closely related, it would become much more difficult to capture the defense allocation of each contrasting species within a DI, although protection effect would still be an appropriate measure for comparison. Including a phylogenetic context is vital for studies of the growth–defense trade-off for at least two reasons. First, controlling for phylogeny is critical because it reduces the noise of interspecific variation that can easily obscure the true patterns in the data (Agrawal and Kotanen 2003). For example, there is substantial variation in both growth and herbivory rates among these six species pairs (Fig. 4a). Indeed, if phylogenetic relationships are ignored and one plots all 12 species averages for growth and herbivory together, the correlation between growth and herbivory disappears. Such an analysis treats each species’ average for growth rate and defense as an independent data point, an assumption that is clearly not valid (Harvey and Pagel 1991). Second, it allows one to make direct inferences about the phylogenetic patterns of plant defensive traits and how they relate to habitat specialization. For example, terpenes, phenolics, and leaf toughness in our genera exhibit strong signals of phylogenetic constraint. But, since species within those genera have a diverse group of defensive options, this apparent lack of evolutionary lability to completely turn on or off investment into a particular class of defense does not result in lineages becoming ecologically constrained to one particular soil type. For this reason, we observed no signal of

Ecology Special Issue

phylogenetic constraint in protection effect (i.e., amount of herbivory) or growth in the genera. This is almost certainly due to the fact that the relevant traits that confer resistance to herbivores in low-resource habitats and faster growth in high-resource habitats are evolutionarily labile and involve quantitative increases and decreases of already-present qualitative traits related to growth and defense. CONCLUSIONS By manipulating the presence of herbivores, we discovered that defense differences interact with edaphic factors to restrict species to their specialized habitats. Although the potential for herbivore attack was similar in the two habitats, the impact of herbivory on growth and survivorship was much stronger in white-sand forest, giving solid evidence of strong selection for effective defense in white-sand forests. Measurements of defenses confirmed that white-sand specialists have a higher overall defense investment, although each genus expressed a different combination of defensive traits. These results confirmed theoretical predictions that species in low resource habitats evolve a higher optimal defense investment. By testing for defense and growth differences in white-sand and clay specialists within an explicit phylogenetic framework, our results strengthen the case that the trade-off between growth and defense is universal and governs patterns of plant distribution. This fundamental trade-off, mediated by herbivores, represents an important mechanism of plant coexistence that has been largely overlooked in studies of plant habitat specialization and niche assembly. Furthermore, this interaction between herbivory and resource heterogeneity should promote divergent selection in plant growth and defense strategies that increase the potential for ecological speciation. ACKNOWLEDGMENTS We thank the Peruvian Ministry of Natural Resources (INRENA) for permission to conduct this study; D. Del Castillo, L. Campos, E. Rengifo, and S. Tello of the Instituto de Investigaciones de la Amazonı´ a Peruana (IIAP) for logistical support and permission to work in and around the Estacio´n Allpahuayo; E. Aquituari, M. Ahuite, J. Guevara, M. Jackson, M. Olo´rtegui, J. Reed, and F. Vacalla for field assistance; P. Evans, J. Becerra, and M. Lott, for advice on terpene analyses; J. Heath, K. Pickering, and C. Cohen for assistance in the Appel/Schultz lab; J. A´lvarez, L. Bohs, D. Dearing, D. Feener, R. Foster, T. Kursar, and S. Schnitzer, for advice during the entire project; and A. Agrawal, M. Ayres, S. DeWalt, G. Paoli, and an anonymous reviewer for improving the manuscript. Support was provided by an NSF Predoctoral Fellowship to P. V. A. Fine, an NSF Doctoral Dissertation Improvement Grant to P. V. A. Fine and P. D. Coley, the Michigan Society of Fellows to P. V. A. Fine, an NSF Long-term Research in Environmental Biology Grant to J. C. Schultz, and NSF grant DEB 0234936 to P. D. Coley. LITERATURE CITED Abouheif, E. 1999. A method for testing the assumption of phylogenetic independence in comparative data. Evolutionary Ecology Research 1:895–909.

July 2006

PLANT TRADE-OFFS AND SPECIALIZATION

Agrawal, A. A., and M. Fishbein. 2006. Plant defense syndromes. Ecology 87:S132–S149. Agrawal, A. A., and P. M. Kotanen. 2003. Herbivores and the success of exotic plants: a phylogenetically controlled experiment. Ecology Letters 6:712–715. Appel, H. M., H. Govenor, M. D’Ascenzo, E. Siska, and J. C. Schultz. 2001. Limitations of Folin assays of foliar phenolics in ecological studies. Journal of Chemical Ecology 27:761– 778. Ashton, P. S. 1969. Speciation among tropical forest trees: some deductions in the light of recent evidence. Biological Journal of the Linnean Society 1:155–196. Bernays, E. A., G. Cooper Driver, and M. Bilgener. 1989. Herbivores and plant tannins. Advances in Ecological Research 19:263–302. Binkley, D., and P. Matson. 1983. Ion-exchange resin bag method for assessing forest soil-nitrogen availability. Soil Science Society of America Journal 47:1050–1052. Boege, K., and R. Dirzo. 2004. Intraspecific variation in growth, defense and herbivory in Dialium guianense (Caesalpiniaceae) mediated by edaphic heterogeneity. Plant Ecology 175:59–69. Bunce, J. A., B. F. Chabot, and L. N. Miller. 1979. Role of annual leaf carbon balance in the distribution of plant species along an elevational gradient. Botanical Gazette 140:288–294. Carson, W., and R. Root. 2000. Herbivory and plant species coexistence: community regulation by an outbreaking phytophagous insect. Ecological Monographs 70:73–99. Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic overdispersion in Floridian oak communities. American Naturalist 163:823–843. Choong, M, P. Lucas, J. Ong, B. Pereira, H. Tan, and I. Turner. 1992. Leaf fracture toughness and sclerophylly: their correlations and ecological implications. New Phytologist 121:597–610. Close, D. C., and C. McArthur. 2002. Rethinking the role of many plant phenolics—Protection from photodamage not herbivores? Oikos 99:166–172. Cody, M. L. 1978. Distribution ecology of Happlopappus and Chrysothamnus in the Mohave desert. I. Niche position and nice shifts on north-facing granitic slopes. American Journal of Botany 65:1107–1116. Coley, P. D. 1983. Herbivory and defensive characteristics of tree species in a lowland tropical forest. Ecological Monographs 53:209–233. Coley, P. D. 1987a. Patrones en las defensas de las plantas: ¿porque´ los herbı´ voros prefieren ciertas especies? Revista Biologica Tropical 35(Supplement 1):151–164. Coley, P. D. 1987b. Interspecific variation in plant antiherbivore properties: the role of habitat quality and rate of disturbance. New Phytologist 106(Supplement):251–263. Coley, P. D., J. P. Bryant, and F. S. Chapin, III. 1985. Resource availability and plant anti-herbivore defense. Science 230: 895–899. Coomes, D. A., and P. J. Grubb. 1998. Responses of juvenile trees to above- and belowground competition in nutrientstarved Amazonian rain forest. Ecology 79:768–782. Craine, J., W. Bond, W. Lee, P. Reich, and S. Ollinger. 2003. The resource economics of chemical and structural defenses across nitrogen supply gradients. Oecologia 442:547–556. Denno, R. F., C. Gratton, M. A. Peterson, G. A. Langellotto, D. L. Finke, and A. F. Huberty. 2002. Bottom-up forces mediate natural-enemy impact in a phytophagous insect community. Ecology 83:1443–1458. Edwards, C., J. Read, and G. Sanson. 2000. Characterising sclerophylly: some mechanical properties of leaves from heath and forest. Oecologia 123:158–167. Fine, P. V. A. 2004. Herbivory and the evolution of habitat specialization by trees in Amazonian forests. Dissertation. University of Utah, Salt Lake City, Utah, USA.

S161

Fine, P. V. A., D. Daly, G. Villa, I. Mesones, and K. Cameron. 2005. The contribution of edaphic heterogeneity to the evolution and diversity of Burseraceae trees in the western Amazon. Evolution 59:1464–1478. Fine, P. V. A., I. Mesones, and P. D. Coley. 2004. Herbivores promote habitat specialization by trees in Amazonian forests. Science 305:663–665. Garcı´ a, R., M. Ahuite, and M. Olo´rtegui. 2003. Clasificacio´n de bosques sobre arena blanca de la zona reservada AllpahuayoMishana. Folia Amazo´nica 14:17–26. Gentry, A. H. 1986. Endemism in tropical versus temperate plant communities. Pages 153–181 in M. E. Soule´, editor. Conservation biology: the science of scarcity and diversity. Sinauer, Sunderland, Massachusetts, USA. Gershenzon, J., and R. Croteau. 1991. Terpenoids. Pages 165– 219 in G. A. Rosenthal and M. R. Berenbaum, editors. Herbivores, their interactions with secondary metabolites. Volume 1. The chemical participants. Academic Press, New York, New York, USA. Grover, J. P., and R. D. Holt. 1998. Disentangling resource and apparent competition: realistic models for plant– herbivore communities. Journal of Theoretical Biology 19: 353–376. Grubb, P. J. 1986. Sclerophylls, pachyphylls, and pycnophylls: the nature and significance of hard leaf surfaces. Pages 137– 150 in B. Juniper and T. Southwood, editors. Insects and the plant surface. Edward Arnold, London, UK. Hagerman, A. E., and K. M. Klucher. 1986. Tannin–protein interactions. Pages 67–76 in V. Cody, E. Middleton, and J. Harborne, editors. Plant flavonoids in biology and medicine: biochemical, pharmacological and structure activity relationships. Alan R. Liss, New York, New York, USA. Harborne, J. B. 1991. Recent advances in the ecological chemistry of plant terpenoids. Pages 399–426 in J. B. Harborne and F. A. Tomes-Barberan, editors. Ecological chemistry and biochemistry of plant terpenoids. Clarendon Press, Oxford, UK. Harley, C. D. G. 2003. Abiotic stress and herbivory interact to set range limits across a two-dimensional stress gradient. Ecology 84:1477–1488. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in evolutionary biology. Oxford University Press, Oxford, UK. Herms, D. A., and W. J. Mattson. 1992. The dilemma of plants: to grow or defend. Quarterly Review of Biology 67: 283–335. Janzen, D. H. 1974. Tropical blackwater rivers, animals, and mast fruiting by the Dipterocarpaceae. Biotropica 6:69–103. Jørgensen, P., and S. Le´on-Ya´nez. 1999. Catalogue of the vascular plants of Ecuador. Missouri Botanical Garden Press, St. Louis, Missouri, USA. Kauffman, S., G. Paredes Arce, and R. Marquina. 1998. Suelos de la zona de Iquitos. Pages 139–230 in R. Kalliola and S. Flores Paita´n, editors. Geoecologia y desarrollo Amazonico: estudio integrado en la zona de Iquitos, Peru. University of Turku Press, Turku, Finland. Koricheva, J., H. Nykanen, and E. Gianoli. 2004. Meta analysis of trade-offs among plant antiherbivore defenses: Are plants jacks-of-all-trades, masters of all? American Naturalist 163: E64–E75. Langenheim, J. 1994. Higher plant terpenoids: a phytocentric overview of their ecological roles. Journal of Chemical Ecology 20:1223–1280. Lerdau, M. T., and J. Penuelas. 1993. Terpenes in plants: links between the biosphere and the atmosphere. Mundo Cientifico 13:60–64. Lim, W., and I. Turner. 1996. Resource availability and growth responses to defoliation in seedlings of three early-successional, tropical, woody species. Ecological Research 11:321– 324.

S162

PAUL V. A. FINE ET AL.

Louda, S. M. 1982. Distribution ecology: variation in plant recruitment over a gradient in relation to insect seed predation. Ecological Monographs 52:25–41. Louda, S. M. 1983. Seed predation and seedling mortality in the recruitment of a shrub Haplopappus venetus (Asteraceae), along a climatic gradient. Ecology 64:511–521. Louda, S. M., K. H. Keeler, and R. D. Holt. 1990. Herbivore influences on plant performance and competitive interactions. Pages 413–444 in J. Grace and D. Tilman, editors. Perspectives in plant competition. Academic Press, New York, New York, USA. Louda, S. M., and J. Rodman. 1996. Insect herbivory as a major factor in the shade distribution of a native crucifer (Cardamine cordifolia A. Gray, bittercress). Journal of Ecology 84:229–237. Mabry, T. J., and J. E. Gill. 1979. Sesquiterpene lactones and other terpenoids. Pages 502–538 in G. Rosenthal and D. H. Janzen, editors. Herbivores: their interaction with secondary plant metabolites. Academic Press, New York, New York, USA. Marengo, J. 1998. Climatologia de la zona de Iquitos, Peru. Pages 35–57 in R. Kalliola and S. Flores Paita´n, editors. Geoecologia y desarrollo Amazonico: estudio integrado en la zona de Iquitos, Peru. University of Turku Press, Turku, Finland. Medina, E., and E. Cuevas. 1989. Patterns of nutrient accumulation and release in Amazonian forests of the upper Rio Negro basin. Pages 217–240 in J. Proctor, editor. Mineral nutrients in tropical forest and savanna ecosystems. Blackwell Scientific, Oxford, UK. Moran, J. A., M. G. Barker, A. J. Moran, P. Becker, and S. M. Ross. 2000. A comparison of the soil water, nutrient status, and litterfall characteristics of tropical heath and mixed dipterocarp forest sites in Brunei. Biotropica 32:2–13. Moran, N., and W. D. Hamilton. 1980. Low nutritive quality as defense against herbivores. Journal of Theoretical Biology 86:247–254.

Ecology Special Issue

Nichols-Orians, C. 1991. The effects of light on foliar chemistry growth and susceptibility of seedlings of a canopy tree to an attine ant. Oecologia 86:552–560. Olff, H., and M. E. Ritchie. 1998. Effects of herbivores on grassland plant diversity. Trends in Ecology and Evolution 13:261–265. Parker, M. A., and R. B. Root. 1981. Insect herbivores limit habitat distribution of a native composite, Machaeranthera canescens. Ecology 62:1390–1392. Ruokolainen, K., and H. Tuomisto. 1998. Vegetacio´n de la zona de Iquitos. Pages 253–368 in R. Kalliola and S. Flores Paita´n, editors. Geoecologia y desarrollo Amazonico: estudio integrado en la zona de Iquitos, Peru. University of Turku Press, Turku, Finland. Sa¨a¨ksja¨rvi, I. E. 2003. Diversity of parasitic wasps (Hymenoptera: Ichneumonidae) in a lowland rain forest mosaic in Peruvian Amazonia. Dissertation. University of Turku, Turku, Finland. Schiers, J., L. de Bruyn, and R. Verhagen. 2003. Host nutritive quality and host plant choice in two grass miners: Primary roles for primary compounds? Journal of Chemical Ecology 29:1373–1389. Schultes, R. E., and R. F. Raffauf. 1990. The healing forest: medicinal and toxic plants of the northwest Amazonia. Dioscorides, Portland, Oregon, USA. Simms, E., and M. Rausher. 1987. Cost and benefits of plant resistance to herbivory. American Naturalist 130:570–581. Sokal, R. R., and F. J. Rohlf. 1995. Biometry. Third edition. W. H. Freeman, New York, New York, USA. Stork, N. E. 1988. Insect diversity: facts, fiction and speculation. Biological Journal of the Linnean Society 35:321– 337. Turner, I. M. 1994. Sclerophylly: Primarily protective? Functional Ecology 8:669–675. Va´squez Martı´ nez, R. 1997. Florula de las reservas biologicas de Iquitos, Peru. Missouri Botanical Garden Press, St. Louis, Missouri, USA. Zar, J. H. 1999. Biostatistical analysis. Fourth edition. Prentice Hall, Upper Saddle River, New Jersey, USA.

APPENDIX A Detailed methods for the chemical analysis of terpenes, phenolics, and soluble protein (Ecological Archives E087-117-A1).

APPENDIX B A table presenting all fixed-factor ANOVAs conducted on the growth and defense variables (Ecological Archives E087-117-A2).

APPENDIX C A table presenting growth, herbivory, and defensive traits measured in the experiment for each species in the two soil types (Ecological Archives E087-117-A3).

APPENDIX D Figures showing the effect of origin (white-sand vs. clay specialists) and the genus 3 origin interaction for (a) leaf growth, (b) height growth, (c) the effect of herbivory on leaf growth (protection effect), (d) protection effect on height growth, (e–j) chemical defenses, (k) leaf toughness, and (l) available foliar protein (Ecological Archives E087-117-A4).

ENDPLATES. (Top) A British mesotrophic grassland (meadow) plant community. Plant diversity can be divided into a hierarchy of special components, which correspond to a hierarchical set of niches. For an exploration of the correspondence between ecological and evolutionary hierarchies, see the article by Silvertown et al. (pp. S39–S49). (Bottom) Larval monarch butterfly (Danaus plexippus) on Asclepias californica. This plant species has high levels of latex and trichomes, which reduce feeding by monarchs and form the basis for one of the milkweed defense syndromes (see the article by Agrawal and Fishbein, pp. S132–S149).