Idea Transcript
J. Med. Microbiol. - Vol. 31 (1990), 137-149
OO22-2615/90/OO3 1-0 137/%10.00
01990 The Pathological Society of Great Britain and Ireland
A pyrolysis-massspectrometry study of Corynebacterium spp.
J. M. HINDMARCH, J. T. MAGEE", M. A. HADFIELD and B. I. DUERDENt Department of Bacteriology, Royal Hallamshire Hospital, Shefield S 10 2JF, *Department of Microbiology, Children's Hospital, Shefield S 10 2TH and ?Department of Medical Microbiology, University of Shefield Medical School, Shefield S 10 2RX
Summary. Clinical (66) and collection (38) strains of Corynebacterium spp., including C . jeikeium and CDC group D2, and of Listeria monocytogenes were examined. Conventional characters used in species identification were assessed by a microbiochemical method, and pyrolysis-mass spectrometry (Py-MS) was performed with a Horizon Instruments PYMS 200X. Classification based on Py-MS data yielded clusters that corresponded with species identification and classification groups from conventional data. One small group of clinical strains, homogeneous in conventional tests and Py-MS, comprised isolates from sputum samples from patients undergoing ventilation; they were similar to collection strains of C. renale and C . striatum; the latter species has been implicated in chest infection. Another group, similar to C . minutissimum in both systems, comprised clinical strains isolated from urogenital specimens. L. monocytogenes strains were clearly distinct from Corynebacterium spp. Groups comprising CDC D2 strains and C .jeikeium were resolved, and were similar to other Corynebacterium spp. Two collection strains of C . xerosis were distinct in conventional tests and Py-MS. Introduction The coryneform bacteria include the classical pathogen, Corynebacteriumdiphtheriae,the recently proposed species C .jeikeium (Jackman et al., 1987), that has been implicated in prosthetic valve endocarditis and infections of immunocompromised patients (Riley et al., 1979; Stamm et al., 1979; Wirsing von Koenig et al., 1982;Finger et al., 1983) and the CDC group D2 organisms, that are occasionally found in urinary tract infections (Sorianoet al., 1985).Clinical isolates of C .jeikeium and D2 organisms are often resistant to many antimicrobial agents (Gill et al., 1981; Soriano et al., 1987; van Bosterhaut et al., 1987). Severalother species are recognised: C . ulcerans is a toxinproducing pathogen (Gilbert and Stewart, 192627; Maximescu et al., 1974); C . striatum (synonymous with C . Javidum; Collins and Cummins, 1986), C . xerosis, C . bovis, C . renale, C . minutissimum and C . pseudotuberculosis have only rarely been documented as human pathogens (Lopez et al., 1966; Geraci et al., 1967; Blackwell et al., 1974; Goldbergeret al., 1981). The genus Listeria includes L. monocytogenes which is a primary pathogen, causing meningitis, septicaemia and other infecReceived 1 June 1989; revised version accepted 11 Aug. 1989.
tions (Gray and Killinger, 1966). These are also gram-positive, non-sporing, facultative bacilli, but are clearly distinct from coryneforms in many other properties, and were examined here as a marker group to validate taxonomic findings. There is a growing need in diagnostic microbiology for clarification of the taxonomic structure of the coryneforms to define those species that are human pathogens, and to develop a biochemical scheme that can clearly identify and discriminate between them. Pyrolysis-mass spectrometry (Py-MS) originated as a technique for the analysisof insoluble polymeric materials, but has more recently been applied to characterisation of biological materials. Microbiological applications of pyrolysis include typing within species, classification, identification, and chemical analysis (see reviews by Gutteridge and Norris, 1979; Drucker, 1981; Gutteridge et al., 1985; Shute et al., 1985). Studieson the identification of coryneform bacilli were initiated because of their increasing incidence in serious infections. Py-MS was applied to the coryneforms in parallel with conventional tests. This report describes the results of the two approaches and compares the classificationsresulting from these distinct techniques.
137 Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
138
J. M. HINDMARCH ET AL.
Materials and methods Bacterial strains The test strains comprised 38 from culture collections
and 66 recent clinical isolates. Designations and sources are listed in table I. All strains were preserved at - 70°C in 15% glycerol broth and cultured on horse-blood agar (Columbia Agar Base, Oxoid, with horse blood, Gibco 5%) for 48 h at 37°C before testing.
Table I. Biochemical reactions and classification results of the test strains Clusters
Putative Number of speciest strains
UC PyMSCTRP
++--++++-----+++--+ +- -+- -- -- -+-+- -+-++----------+++- ++ - - +
F1 F2 B5 E
5b 5b 9 9
L .mono L.mono C.jk CDC D2
iii
A4
8b
C.jk
7
iii iii iii iii
B1 B2 B3 B4
8b 8b 8b 8b
C.jk C.jk C.jk C.jk
2 1 6 15
iv v vi
D K A2
Id 7 5a
C .xer C.pstb C.ulc
1 1 3
vi vii vii vii
A2 H1 I H2 I J G1 G2 G1 C A3 Ala Alb Alb Alb Alc Alc Alc Alc A5 A5 A5 Ala Alc A3
6b 4c 4c 4c 4g 4b 4b 4b 4f llb 4e la 3 8a 10 2 4a 4d 6a lb lc 4b lb lla 4a
C.pstb C.min
1 2 1 1 1 1 1 3 1 1 2 3
v11
... ...
Vlll
... ...
Vlll Vlll
ix X
xi xi xi x1
xi xi xi xi xii xii 7 3
? 7
Collection strains
GMLSFORTYCUNWHAZEI D
i i ii ii
Vlll
Biochemical reactions*
-
Cstr CDC G2" C .ren -
C.dip C.xer C.bov -
CDC 1" C.dip
-
CDC 1" CDC 1"
1
2 1 1 1 1 1 13 10 1 1 1 1
++--++----+--++--+-
++--+++++++--+++--+ ++--++----+---+---+
++-+++------++++--+ ++-+++------++++--+ ++--++++-----+++--+ ++--++-----+-+++--+ +-++++-------+++--+ ++++++-----+-+++--+ ++++++-------+++--+ +- -+ -- --+-+-- -- --- -+ -- --+-+-+-- - + ++--+ +---++-------+++--+ ++v-++--++-+--++-+-
++ -++ -++ +- - +- - +- - -v -- ----- -+ -- ++ - - + - + +---+------++-----+
c222 ATCC4504, ATCC43044, ATCC43043, B693 1, B9629 NCTCll914, NCTCAll3/74, C220, C499, B66, B80, API B8937, B6136 C488 C225 B6536, C28, C226, C483, B6650, A2375, C454 B5944 NCTC3450 NCD01896, NCDOl570, NCTC7907 A2354 API
NCTC764
NCTC7188 NCTC3984, NCTC10648 NCD01572 NCTC3224, NCD01930
++-+++-----+--++--+ ++--++--++-+-+++--+ ++-+++----+-+-+---+
+++-++--------++--+ ++--++-v---v--++-+-
+v-+++-v---
v--++-+-
++-+++-------+++--+ +- -+ -+ -- +-+---- ----- +- - - + + - + ++--++--+ +++-++-------+++--+
Type strains are italicised ; tentative identifications suggested by API are indicated by a superscript a. Strains donated by Dr R. Bayston (Institute of Child Health, University of London), Professor W. C. Noble (Institute of Dermatology, St John's Hospital, London), and API-Bio Merieux (Cranbourne Lane, Basingstoke, Hants) are indicated by the prefices B, C and A respectively. Two strains donated by API had no accession number. * Key GMLSFORTYC-acid from glucose, maltose, galactose, sucrose, fructose, mannose, rhamnose, trehalose, glycogen and starch ; U-urease activity; N-nitrate reduction ; W-Tween 80 hydrolysis; H-haemolysis; A-anaerobic growth; Z-large colony (> 2 mm diam.); E-colony sheen; I-irregular colony edge; D-domed colony. t Key L.mono-L. monocytogenes; C.jk-C. jeikeium; C.xer-C. xerosis; C.pstb-C. pseudotuberculosis; C.ulc-C. ulcerans; C.min-4. minutissimum;Cstr-C. striatum; C.ren-C. renale; C.dip-C. diphtheriae; C.bov-C. bovis. Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
PYROLYSIS-MASS SPECTROMETRY OF CORYNEFORMS
Biochemical tests Strains were examined for : reaction with Gram’s stain, cell morphology, catalase production, spore formation, acidification of sugars and nitrate reduction (Thompson et al., 1983), urease activity (Lautrop, 1960), Tween 80 hydrolysis (Sierra, 1957),a-haemolysis,anaerobic growth and colonial morphology. Tween 80 medium was incubated aerobicallyat 37°C for 48 h. Haemolysis was sought on horse-blood agar incubated aerobically for 24 h at 37°C. Anaerobic growth was sought on horse-blood agar incubated for 72 h in an atmosphere of H2 lo%, C 0 2 lo%, N2 80%. The morphology of well separated colonies was examined on horse-blood agar after aerobic incubation at 37°C for 24 h, noting surface sheen, colony edge, profile and diameter; colonies > 2 mm diameter were recorded as large. All strains included were gram-positive, catalase-positive,non-sporing coryneform bacilli. In addition, six strains from sputum samples were sent to API-Bio Merieux (Basingstoke, Hants) where they were identified in a system based on the API20 Strep microtest strip (Tillotson et al., 1988). Three of these were isolated from sputum from neurosurgical patients who were being ventilated and had clinical evidence of chest infection. Gram-stained smears of these sputum samples showed large numbers of coryneform organisms in pus cells and cultures yielded a heavy growth of coryneforms; the other three were isolated from non-ventilated patients with little evidence for a role in infection.
Pyrolysis-mass spectrometry Methods for Py-MS were as described previously (Magee et al., 1988). Samples were pyrolysed at 530°C for 4 s. Collection strains were blind coded and subjected to four replicate analyses. Clinical strains were blind coded as a separate series, and subjected to single analyses. The code was not broken until the classification (fig. 1) was finalised. Strains were cultured on blood agar poured from a single batch; they were incubated concurrently for 48 h at 37”C, and smeared on pyrolysis foils for analysis. Foils were stored in a vacuum over P 2 0 5and analysed over 2 days.
Mathematical analyses The recorded spectra comprised ion intensity measurements for masses 11-200. Spectra were normalised by the iterative technique of Huff et al. (1981) to eliminate variation due to differences in the amount of sample pyrolysed. After normalisation, masses 50-1 29 were selected for further analysis. These showed within-strain coefficientsof variation < 15%. Between 1% and 10% of replicate spectra yielded ion intensities outside the range of the strain mean + 2 x the within-strain standard deviation for these masses, indicating that replicate ion intensities for this mass set showed approximately Gaussian (normal) distributions. Other recorded masses showed poor within-strain reproducibility of normalised intensity. Replicate spectra of the strains were labelled
139
as distinct groups and normalised data for the selected peaks were subjected to stepwise discriminant analysis (SPSSX User’s Guide, 1983). This optimised discrimination between the replicated strains, correcting the data for systematic covariance and for differences between masses in within-strain reproducibility of measurement. Mass 99 was eliminated as non-discriminatory in the Mahalanobis stepwise variate selection employed. The spectra were now represented by co-ordinates on 14 independent (orthogonal) discriminant function axes, accounting for c. 99% of inter-strain variance, and describing a statistical transformation that optimised inter-strain discrimination and corrected for withinstrain statistical effects (Magee et al., 1989). Spectrum co-ordinates on these axes were processed in classification analyses. Normalisation and peak selection were performed in BASIC VM on a Prime 750 computer, and discriminant and classification analyses on an IBM 3083X computer at the University of Sheffield Computer Centre. ClassiJicationanalyses were programmed in Clustan 3.2 (Wishart, 1987). Squared euclidean distance between spectral points in the space described by the 14 discriminant functions was the index of dissimilarity for Py-MS data. This index, multiplied by the number of variables analysed (14), is approximately equal to x2 on 13 degrees of freedom, with the null hypothesis that the pairs of spectra compared were derived from the same strain. Four clustering strategies were explored : hierarchical clustering was performed by UPGMA (Sokal and Michener, 1958; Rohlf, 1963) and Ward’s method (Ward, I963), and non-hierarchical clustering by Relocate (Crawford and Wishart, 1968) and Normix (Wolfe, 1970), with classifications from each hierarchical method as starting points. Co-phenetic correlations with the similarity matrix were calculated for hierarchical methods, and probabilities of N groups (p of null hypothesis) for Normix (Wolfe, 1970). Mean-strain, rather than individual spectrum, co-ordinates were submitted for replicated strains, but full data were also submitted for the most promising strategy-UPGMA clustering. For biochemical data, simple matching (Sokal and Michener, 1958) was the index of similarity, and UPGMA, Ward’s and Relocate analyses were performed. Ident$cation analyses. For biochemical identification, the members of each of the unified classification (UC, see below) groups were randomly divided into two equal sets. The frequency of positive reactions in the conventional tests was calculated for one set, the teaching set, from each group, yielding an identification matrix. The conventional test reaction patterns (CTRPs) of the remaining strains, the blind identification set, were compared with this matrix, identifying each by the method of Lapage et al. (1973) using the normalised identification score. The teaching and blind identification sets were then exchanged, and the procedure was repeated. Py-MS data were similarly divided, and blind identification performed with discriminant analysis using the UC groups. Groups iv, v, ix and x comprised less than three strains and were excluded from these analyses.
Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
140
J. M. HINDMARCH ET AL.
Results Py-MS classijication results :numerical considerations In the various strategies, Py-MS data produced similar cluster memberships at levels from 40 to 20 clusters, suggesting that these clusters approached Wishart’s “global optimum” solution (Wishart, 1978). UPGMA yielded a higher co-phenetic correlation (0.709) than did Ward’s method (0.486), indicating that the former was a better representation of the Py-MS similarity matrix. Normix null hypothesis probabilities showed clear optima at 11 and 22 clusters, suggesting that the clusters at these levels represented homogeneous groups that could be readily discriminated in Py-MS; cluster membership at these two levels was almost identical for all clustering methods, with one exception-a subcluster of C .jeikeium strains (A4, fig. 1) was found in the A cluster in UPGMA, but in the B cluster in Ward’s, and Relocate strategies. The apparently optimal Py-MS UPGMA classification (fig. 1) at 11 clusters (A-K) and 22 clusters (A-K with A, B, F, G and H subdivided into 5, 5, 2, 2 and 2 clusters respectively) were compared with biochemical classifications. The mean within-cluster spectral co-ordinates on the first three discriminant function axes (representing 75.6% of inter-strain discrimination) are shown in the ordination diagrams for the 11 and 22 clusteroptima (figs. 2a and 2b), and the dendrogram obtained with Py-MS data in UPGMA is shown in fig. 1.
as L. monocytogenes; the division into sub-cluster F l and F2 did not correlate with serogroup. The collection strains of C . striatum and C . renale clustered in G1, and were similar to the three strains from sputum of ventilated patients (G2) in Py-MS and CTRPs. These three clinical strains were tentatively identified as members of CDC group G2 (a coincidental designation) by API. Strains of clusters H and I were similar in Py-MS and CTRPs, and comprised the C . minutissimum strain, and four clinical strains from urogenital sites. Py-MS cluster K comprised C .pseudotuberculosis NCTC 3450. In the dendrogram, K fused with C .pseudotuberculosis NCTC 2354 (A2) at D2> 700 on the final stem, yet, in the similarity matrix, NCTC 2354 was its closest neighbour at D2= 192.This portion of the similarity matrix was poorly represented in the dendrogram, although the strains were clearly dissimilar in composition. Several Py-MS clusters contained no collection strains : cluster A3 comprised isolates from sputum (one), vagina (one) and urine (one); cluster A5 comprised isolates from intravenous catheters (seven), urines (six), wound swabs (seven), dialysis fluid (one), ear (one), vagina (one) and umbilicus (one); cluster C comprised one strain, singularly unreactive in biochemistry, originating from hydrocoele fluid; cluster J comprised a single sputum isolate from a ventilated patient, and was associated with cluster H in fig. 2b.
Biochemical homogeneity of the Py-M S clusters Fifteen Py-MS clusters comprised more than one strain. Strains in six of these clusters (B3, E, F1, F2, G2 and H1) showed homogeneous CTRPs and Py-M S classijication results :clusters obtained in another three clusters (A4, Bl and B4) the Six clinical strains clustered in A1 with the CTRPs of member strains differed in only one test. collection strains of C . diphtheriae, C . bovis and one Two further clusters (A2, A3) included more than strain of C . xerosis. The second C . xerosis strain two strains, of which only one showed an atypical (B5944) formed the single member cluster D. CTRP; in both cases the atypical strain was an Cluster A2 comprised the three C . ulcerans strains outlier of the Py-MS cluster. These clusters comand C .pseudotuberculosis NCTC 2354. Five strains prised 57 of the 104 strains studied. Three Py-MS clusters (A5, G1 and I) showed a of C .jeikeium (NCTC 11914, NCTC A113/74, C499, C220, and the API strain) and two clinical lower level of CTRP homogeneity. Cluster A5 strains with similar CTRPs formed cluster A4. The comprised 24 strains: one (D64) was an outlier in remaining 21 collection strains of C .jeikeium, biochemistry and Py-MS ; the remaining strains together with four clinical isolates formed cluster showed variation in only four tests. Cluster G1 B, with one strain as an outlier in cluster B5. The comprised two strains that differed in four reactions latter strain (C222)was received by us as C .jeikeium and cluster I comprised two strains that differed in but resembled the CDC D2 group in CTRP; its three reactions. The remaining cluster (A 1) comprised 13 strains mean spectrum point lay between those for the CDC D2 and C .jeikeium clusters in fig. 2b. Cluster that showed identical reactions in only two of the E comprised all strains received as CDC D2. 19 conventional tests. This diversity was atypical, Cluster F comprisedall strains received or identified and proved illusory at a higher level of similarity Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
141
PYROLYSIS-MASS SPECTROMETRY OF CORYNEFORMS x213df of spectra originating from the same strain 7400
6000
46,OO
132pO
28pO
1
I I I I
I I I I
I I I I I I I I I
I
Alb
I I I
Alc
-4
I
rnI
A1 a
I
I I I I I I I I I I I I I I I I I I I I I I
1-
I
I4
I
I
I
I I I
I
A3 A4
A5
B1 62 B3
B4
E
I
I I I
A2
B5 C D
I I I
1
Group designation
11400 I
1
F1 I I
I
I
I
F2 G1 G2 H1 H2 I J
K
Fig. 1. A dendrogram based on Py-MS data. The dissimilarity measure is squared euclidian distance in inter-strain discriminant space. When multiplied by 14-the number of canonical discriminant functions-this is approximately equal to the probability that the spectra compared were from the same strain, on a x2 distribution with 13 degrees of freedom. The clustering method was UPGMA. The vertical lines delineate optimal cluster levels with 11 and 22 groups respectively.
(D2= 100, p < O.l%),where three sub-clusters were resolved, with more homogeneous CTRPs (table I). One sub-cluster included three collection strains of C. diphtheriae; another, the two collection strains of C. bovis and C.xerosis NCDO 1572; and the third, one C. diphtheriae strain and four clinical strains. Hierarchical cluster analyses with individual spectrum rather than mean-strain data showed enhanced separation of these, particularly for the C. bovislxerosis subgroup.
Biochemical cZassiJication results: numerical considerations Biochemical data also yielded a higher cophenetic correlation with UPGMA (0.926) than with Ward’s method (0.884).The UPGMA dendrogram is shown in fig. 3. An arbitrary cut-off of S = 90%was used for biochemical clusters, to allow comparison of conventional and Py-MS clusters for similar numbers of groups; at this level cluster members differed in less than three test results.
Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
142
J. M. HINDMARCH ET AL.
a
b
l-
LL
n 0
-90
C
d 70
24
3c
12 l -
LL
n
0
T-
k o
0
-12
-24 L
-24
Fig. 2. Ordination diagrams illustrating the similarity between groups. Figs. 2a-c have identical axes and show group positions for
the 11 cluster Py-MS classification (2a),the 22 cluster Py-MS classification (2b)and the groups of the unified Py-MS and biochemical classification (2c). The axes are the first three inter-strain canonical discriminant functions. These axes are corrected for withinstrain statistical effects and account for c. 75.6% of the inter-strain variance. Fig. 2d illustrates the discrimination between groups of the unified classification when used in identification from Py-MS data. The axes in this case account for c. 72.5% of the attainable inter-group discrimination in Py-MS. Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
143
PYROLYSIS-MASS SPECTROMETRY OF CORYNEFORMS
Biochemical classificationresults :clusters obtained
two collection strains of C . pseudotuberculosis differed in four reactions and formed single member Of the 24 clusters resolved, six corresponded to clusters (6b and 7) distinct at SO%S. The remaining recognised groups : cluster 5b to L. monocytogenes, clusters comprised clinical isolates : cluster 1b, with 9 to CDC D2, 8b to C. jeikeiurn, 5a to C. ulcerans, 14 members, was the largest; l c comprised ten 4f to C. renale and 8a to C . bovis. Cluster l a strains, 4a and 4e each comprised two strains, and comprised three collection strains of C. diphtheriae; clusters 2, 4g, 6a, 10, 1l a and 11b each comprised the remaining strain formed cluster 4d and differed single strains. in three reactions. The strain of C. striatum, together with four clinical strains, formed cluster 4b, while cluster 4c comprised the collection strain of C . rninutissirnurn and three clinical strains. The two Py-M S homogeneity of the biochemical clusters Twelve of the biochemical clusters comprised collection strains of C . xerosis differed in eight conventional test results and formed single member more than one strain. Of these, seven comprised clusters (Id and 3) distinct at 70%S. Similarly, the strains homogeneous in Py-MS grouping, accountPercentage similarity (simple matching coefficient) 40
50
60 I
70 I
80
90:
Group designat ion
-
I
I I I
-2 3
4a
-
I I I I I I
I
I
I I I I I I I
8b
10 11
Fig. 3. A dendrogram based on biochemical data. Simple matching coefficient and UPGMA strategy were used in this classification. The vertical line delineates 24 clusters at S,, c. 90%. Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
c
P P
Table 11. The numbers of strains are tabulated showing correspondence of Py-MSand biochemical classification clusters Biochemical cluster
UC group I
1
Py-MS cluster
...
111
iv V
vi
9
8b
Id
7
5a
6b
4c
4g
4f
4b
11
4e
la
2
3
4a
4d
6a
8a
10
lb
Tentative species
lc
F1 F2
B5
11
5b
L. monocytogenes
E
-
B1
2
B2
1
B3
6
B4
15
A4
7
CDC D 2
C .jeikeium
D
K
A2
?
H1
vii
H2
C . minutissimum
I
...
Vlll
G1
C . striatum
G2
and
J
C . renale
ix
C
X
A3
xi
A1
xii
A5
Total
LI 1*
rn
? ?
1*
Several
1*
I T T T T - '1
4
' I
5
"
6
I n n '1
2
11
" 2 3 '
The group designation derived from the unified classificationgroup based on Py-MS and conventional test data is shown in the first column; strains which were not assigned in this classification are marked by *, and do not contribute to totals. Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
PYROLYSIS-MASS SPECTROMETRY OF CORYNEFORMS
145
ing for a total of 44 strains. The remaining five B1-4 and A4. These five Py-MS groups were close biochemical clusters with more than one member neighbours in the ordination diagram, appeared were 4a, 4b, 4c, 8b and 9. Biochemical cluster 4a close neighbours in the similarity matrices, and comprised two strains, one each from Py-MS showed similar CTRPs. There was evidence from clusters A1 and A3; 4b comprised the Py-MS several Py-MS clustering procedures that the cluster G2 (three strains homogeneous in CTRPs) inclusion of A4 in Py-MS cluster A was misleading, and three biochemical outliers, one each from Py- possibly a result of the similarity between A5 MS clusters A5, G1 and J. Biochemical cluster 4c members and C. bovis strains. Groups iv, v and ix comprised the members of Py-MS clusters H1, H2 comprised single members, distinct in Py-MS and and I ; 8b comprised the members of Py-MS clusters CTRP. Group vi comprised Py-MS cluster A2, i.e., A4, Bl, B2, B3 and B4. Biochemical cluster 9 the three strains of C . ulcerans (5a) and one of C . showed an outlier, the single member of Py-MS pseudotuberculosis (6b), a CTRP outlier. Py-MS cluster B5;this strain did not hydrolyse Tween 80, clusters H1, H2 and I formed group vii; strains of unlike the remaining members which comprised all HI and H2 showed similar CTRPs and Py-MS members of Py-MS cluster E and were CDC D2 characteristics, and I was included as an outlier. strains. Group viii comprised strains from Py-MS clusters GI, G2 and J ; these were closest neighbours in the ordination diagram and the Py-MS similarity Comparisonand unijicationof the biochemical and matrix, and showed similar CTRPs. The members Py-MS classijications of A3 formed group x, with the exception of a single Table I1 shows comparisons between the cluster strain, atypical in CTRP, and an outlier in Py-MS, memberships of the strains in the Py-MS and which was not assigned. Group xii comprised the biochemical classifications. There were clear cor- strains of Py-MS cluster AS which showed similar relations in membership for groupings derived by CTRPs, except for a single strain which was not the two methods. As outlined above, however, the assigned, as it was an outlier in Py-MS, and differed structures of the two dendrograms (figs. 1 and 3) at in CTRP from other strains assigned to this group. Group xi, however, was assigned arbitrarily to similarity levels lower than those selected were distinct. Therefore, Py-MS and biochemical clus- comprise the strains recovered in Py-MS group A 1, ters were re-aligned to give maximum correspond- with the exception of two which were Py-MS ence along the diagonal of table 11, thereby outliers and gave CTRPs similar to those of group discarding this lower level structure. This clarified xi; these were not assigned. The group was the comparison of the two classifications, and heterogeneous in CTRPs. Three groups with more suggested that a classification combining the results homogeneous CTRPs, corresponding to the Py-MS of the two approaches could be derived by deline- sub-groups of A1 discussed, could have been ating 12 groups of strains with similar Py-MS and constructed, but the small numbers of members biochemical properties. These groups are shown in would have precluded the use of the confirmatory boxes in table 11. Four strains with anomalous Py- identification procedures discussed below. MS or CTRPs were omitted. This arrangement of the Py-MS and biochemical clusters was not arbitrary; the two similarity Identification of strains in the uniJied classification Each strain was identified from a blind challenge matrices, and the 22-group Py-MS ordination diagram were considered carefully before the set, on the basis of biochemical reactions, and in groups were finalised, to obtain maximum homo- discriminant analysis of Py-MS data, using the geneity within each group for both characterisation unified group classification described above. methods. Thus it seemed reasonable to combine Groups iv, v, ix and x were omitted in these Py-MS group F1 and F2, its closest neighbour, to analyses as they contained too few members to form group i which comprised strains showing allow division into a teaching and challenge set. identical CTRPs. Py-MS cluster B5 was inter- The results are shown in table 111. In identification mediate between Py-MS clusters B and E; it was a from Py-MS data, only five strains were identified gross outlier of cluster B and was similar to strains in disagreement with their group assignment in the of Py-MS cluster E in CTRP, therefore group ii unified classification, and in identification from comprised the five strains of E, with the single biochemical reactions, only six strains showed strain of B5 as a tentative member, atypical in disagreement. The strains which were identified in CTRP and Py-MS. Group iii comprised the strains disagreement were mostly classified in groups x, xi of biochemical cluster 8b, found in Py-MS clusters and xii. Biochemical discrimination between Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
146
J. M. HINDMARCH ET AL.
Table 111. Results of blind identification in the unified Py-MS and biochemical classification Agreement with U C classification in blind identification (%) UC group
Corresponding species
i ii...
L. monocytogenes CDCD2 111 C.jeikeium vi C . ulcerans ? vii C . minutissimum viii C . striatum ? xi ? xii ?
Total
Number of strains
Py-MS data
CTRP data
10 6
100 100
4 5 6 11 23
100
100 100 100
96
100 100
96
95
97
31
97
100
100 73
groups xi and xii was poor, with high probabilities for second choice cross-identificationbetween these groups. By contrast, groups xi and xii were well discriminated in Py-MS. Teaching sets which included all members of the groups were constructed, yielding an identificationmatrix (table IV) and further Py-MS ordination diagrams (figs. 2c and 2d). Fig. 2c is directly comparable to figs. 2a and 2b, and shows the mean spectral points for the 12 groups in inter-strain discriminant space. Fig. 2d shows 72.3% of the maximum attainable intergroup discrimination from normalised Py-MS spectra, distances between the group mean points reflect the ease of discrimination between groups.
75 80 83
and Cummins (1986). In this study, the two collection strains investigated were closely similar in CTRPs and Py-MS, and appeared similar to a group including C. diphtheriae strains. C .jeikeium and the CDC D2 group were not considered in the taxonomic review of Collins and Cummins (1986), despite their growing importance in medical microbiology. Athalye et al. (1984) found similarities in mycolic acid content between the J-K group and Rhodococcus equii. However, Jackman et al. (1987) submitted a formal proposal for the recognition of C. jeikeium sp. nov. on the basis of protein (PAGE) and DNA hybridisation studies. In Py-MS this species was heterogeneous, but appeared similar to other Corynebacterium spp. The CDC D2 strains examined formed a homogeneous group in bioDiscussion chemistry and Py-MS, probably deserving species Initially, the genus Corynebacterium comprised status within the genus Corynebacterium. only human and animal pathogens; however, many The Py-MS classification showed clear correlaill-assorted organisms were assigned to the genus, tions with the species designations of the collection necessitating the introduction of a restrictive strains, and with the biochemical patterns obtained. generic definition based on chemotaxonomic prop- In the ordination diagrams, Py-MS cluster mean erties (see Collins and Cummins, 1986). This re- spectrum points showed an elongated group of definition removed much confusion at genus level, corynebacteria, separated from the mean point but the acceptability and homogeneity of several representing L. monocytogenes. At a more detailed species remains in doubt. C. xerosis strains comprise level, C. jeikeium showed a high level of CTRP two groups, differing markedly in G C mol%, and homogeneity. However, highly significant interthe type strain (ATCC 373) should, strictly, be strain variation in overall chemical composition excluded from the genus (G + C mol% > 63%). The was found in this group, and it may be that this two collection strains of C. xerosis investigated here variation could be exploited as a typing method. By were dissimilar in CTRPs and Py-MS. C. renale contrast, the Py-MS cluster G2 revealed a group of has been shown to be a complex of three species isolates with close similarities in CTRPs, composidiffering immunologically, biochemically and in tion and origin; all three were isolated from sputum G + C content (Collins and Cummins, 1986). of neuro-surgical patients undergoing artificial Collection strains of C. bovis are heterogeneous; ventilation, and judged clinically to have chest this species was excluded from the genus by Collins infections. Laboratory findings suggested that these
+
Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
147
PYROLYSIS-MASS SPECTROMETRY OF CORYNEFORMS
Table IV. Biochemical test results for the groups of the unified Py-MS and biochemical classification Percentage of positive results for strains of U C groups Test Glucose Maltose Galactose Sucrose Fructose Mannose Rhamnose Trehalose Glycogen Starch Urea Nitrate Tween 80 Haemolysis Anaerobic growth Large colony Sheen Irregular edge Domed colony
i
11
100 100 0 0
100 100
100
100 0 0 0 0 0
100 1 00 1 00 0 0 100
0 0 0 0 0 0
0 0 0 0
100 0 83 0 0 0
100 0
100
...
...
111
vi
vii
Vlll
xi
xii
100 56 100
100
100
100 100 0
100
100 73
100
0 0 0 0 0 0 0 0 0 100
0 0 0 100 0 100
isolates were clinically significant. The three strains were identified as members of the CDC G2 group by API and appeared similar to C . striatum. The latter has been documented as a respiratory pathogen in a case report (Barr and Murphy, 1986) that shows many striking similarities to the circumstances outlined above. The group of genital strains, similar to C. rninutissimum,were also closely similar in CTRPs and Py-MS, and may represent another group of previously unrecognised occasional or opportunist pathogens. Turning to theoretical aspects, it is commonly accepted (Sneath and Sokal, 1973) that phenon rank should be delineated at a single similarity level, i.e., as a horizontal straight line across a dendrogram, and, for conventional tests, a simple matching coefficient of 80-85% is usually considered to delineate “species”. The latter criterion is not applicable to euclidean distance measures derived from continuous data, therefore other criteria were sought for interpretation of the PyMS data. Normix allowed tentative group structures derived from hierarchical analysis to be tested for optimality and intra-group homogeneity. This revealed two optima, of which one, the 22-group solution, appeared more relevant at species level. Relocate was used as a further test of intra-group homogeneity. This combination of hierarchical and non-hierarchical methods proved useful in obtain-
0 0
100 100 75 75 75 75 100 0 0 75 100 75 0 0
100
80 100 100 0 0 0 0 0 40 80
100 100
100 0 0
100
83 83 83
100 100
0
0 0 0 17 17 0 100 100 100 0 0 100
45 27
100
13 0 0 36 36 10 55 27 10 64 64 18 36 64
96
0
43
100 100 0 17 0 0 0 52 0 0 100 100 0 1 00 12
ing objective criteria for delineation of the Py-MS groups, and may have other applications in taxonomy. We have commented on diagrammatic representation of classification data previously (Hindmarch and Magee, 1987). Dendrograms preserve similarity structure at high similarity levels, i.e., fusions of stems representing single strains occur at similarities that accurately reflect similarity between these strains. However, at lower similarity levels-fusion of stems representing groups of strains-the fusion point no longer corresponds to the similarity of strain pairs from distinct groups. In effect, the similarity matrix, containing data in N dimensions, where N is the number of attributes considered, is reduced to a two dimensional diagram by discarding successively greater portions of the data at each non-tip stem fusion. The problem of Py-MS cluster K, described in Results is a good example. This reasoning led us to select groups comprising few strains, and to ignore the low-level similarity structure of the dendrograms in forming table 11. Ordination diagrams also represent only a portion of the data; however, when axes of maximum variation that account for a large proportion of the sample variance are plotted, major differences tend to be conserved, while fine detail at high similarity levels on minor axes is not represented. We suggest that ordination diagrams are complementary to
Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
148
J. M. HINDMARCH ET AL.
dendrograms ; the former allow inspection of lowlevel similarity structure, and the latter inspection of high-level structure. Neither, alone, can represent the similarity matrix completely; however, in combination they may be more convenient to publish than the unwieldy, expensive and time consuming shaded similarity matrix. We regard a blind identification trial as an essential portion of any classification study. All classifications are arbitrary, justified only by the predictive aspects of identification to the hypothesised taxa. The ability to identify isolates in a high level of agreement with their classified taxon membership is an essential pre-requisite for prediction of hypothesised group properties (such as pathogenicity) of a new isolate. Position in the unified classification could be determined with accuracy from either of two distinct character sets. Bias towards either characterisation method in this classification would have produced differences in accuracy of identification between the Py-MS and
biochemical methods. In practice, differences were minimal, even for group xi, which was assigned arbitrarily on the basis of Py-MS alone. Furthermore, each hypothesised taxon should delineate a group of organisms with many common properties-not only similarities in their nutritional interactions with artificial environments. Py-MS and biochemical classifications showed a high level of mutual predictivity in this case and both suggested that two previously unrecognised groups of coryneforms may be occasional human pathogens.
REFERENCES
Goldberger A C, Lipsky B A, Plorde J J 1981 Suppurative granulomatous lymphadenitis caused by Corynebacterium ovis (pseudotuberculosis). American Journal of Clinical Pathology 76:486-490. Gray M L, Killinger A H 1966 Listeria monocytogenes and listeric infections. Bacteriological Reviews 30: 309-382. Gutteridge C S, Norris J R 1979 The application of pyrolysis techniques to the identification of micro-organisms. Journal of Applied Bacteriology 47 : 5-43. Gutteridge C S, Vallis L, MacFie H J H 1985 Numerical methods in the classification of micro-organisms by pyrolysis mass-spectrometry. In: Goodfellow M et al. (eds) Computer-assisted bacterial systematics. Academic Press, London, pp 369-401. Hindmarch J M, Magee J T 1987 The staphylococci: a classification and identification study using pyrolysis-gasliquid chromatography. Journal of Analytical and Applied Pyrolysis 11: 527-538. Huff S M, Meuzelaar H L C, Pope D L, Kjeldsberg C R 1981 Characterization of leukemic and normal white blood-cells by Curie-point pyrolysis-mass spectrometry. 1. Numerical evaluation of the results of a pilot-study. Journal ofAnalytica1 and Applied Pyrolysis 3: 95-109. Jackman P J H, Pitcher D G, Pelczynska S, Borman P 1987 Classification of corynebacteria associated with endocarditis (Group JK) as Corynebacterium jeikeium sp. nov. Systematic and Applied Microbiology 9 : 83-90. Lapage S P, Bascomb S, Willcox W R, Curtis M A 1973 Identification of bacteria by computer : general aspects and perspectives. Journal of General Microbiology 77 : 273-290. Lautrop H 1960 Laboratory diagnosis of whooping cough or Bordetella infections. Bulletin of the World Health Organization 23 : 15-3 1. Lopez J F, Wong F M, Quesada J 1966 Corynebacterium pseudotuberculosis :first case of a human infection. American Journal of Clinical Pathology 46: 562-567. Magee J T. Hindmarch J M, Duerden B I, Mackenzie D W R
Athalye M, Noble W C, Mallet A I, Minnikin D E 1984 Gas Chromatography-mass spectrometry of mycolic acids as a tool in the identification of medically important coryneform bacteria. Journal of General Microbiology 130 : 5 13-5 19. Barr J G, Murphy P G 1986 Corynebacterium striatum: an unusual organism isolated in pure culture from sputum. Journal of Infection 13 : 297-298. Blackwell J B, Smith F H, Joyce P R 1974 Granulomatous lymphadenitis caused by Corynebacteriumovis. Pathology 6 : 243-249. van Bosterhaut B, Claeys G, Gigi J, Wauters G 1987 Isolation of Corynebacterium Group D2 from clinical specimens. European Journal of Clinical Microbiology 6 : 418-419. Collins M D, Cummins C S 1986 Genus Corynebacterium. In: Sneath P H A et al. (eds) Bergey’s Manual of systematic bacteriology, vol 2. Williams and Wilkins, Baltimore, pp 1266-1276. Crawford R M M, Wishart D 1968 A rapid classification and ordination method and its application to vegetation mapping. Journal of Ecology 56: 385-404. Drucker D B 1981 Microbiological applications of gas chromatography. Cambridge University Press, Cambridge. Finger H, Wirsing Von Koenig C H, Wichmann S, BeckerBoost E, Drechsler H J 1983 Clinical significance of resistant corynebacteria Group JK. Lancet 1: 538. Geraci J E, Forth R J, Ellis F H 1967 Postoperative prosthetic valve bacterial endocarditis due to Corynebacterium xerosis. Mayo Clinic Proceedings 42: 736-743. Gilbert R, Stewart F C 1926-27 Corynebacterium ulcerans: a pathogenic organism resembling Corynebacterium diphtheriae. Journal of Laboratory and Clinical Medicine 12: 756761. Gill V J, Manning C, Lawson M, Woltering P, Pizzo P A 1981 Antibiotic-resistant group JK bacteria in hospitals. Journal o f Clinical MicrobioloPv 13: 472-477.
We thank Perkin-Elmer for their loan of the PYMS 200X instrument, Trent Region Research Committee for a grant which allowed us to develop the mathematical methods described, Professor W. C. Noble (Institute of Dermatology, St John’s Hospital, London) and Dr R. Bayston (Institute of Child Health, University of London) for the strains which they donated, and API/Bio Merieux (Basingstoke, Hants) for their strains, and permission to publish the API20 Strep identification results.
Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44
PYROLYSIS-MASS SPECTROMETRY OF CORYNEFORMS 1988 Pyrolysis mass spectrometry as a method for interstrain discrimination of Candida albicans. Journalof General Microbiology 134: 2841-2847. Magee J T, Hindmarch J M, Bennett K W, Duerden B I, Aries R E 1989 A pyrolysis mass spectrometry study of fusobacteria. Journal of Medical Microbiology 28 : 227-236. Maximescu P, Oprisan A, Pop A, Potorac E 1974 Further studies on Corynebacterium spp. capable of producing diphtheria toxin (C.diphtheriae, C . ulcerans,C . ovis).Journal of General Microbiology 82 : 49-56. Riley P S, Hollis D G, Utter G B, Weaver R E, Baker C N 1979 Characterisation and identification of 95 diphtheroid (group JK) cultures isolated from clinical specimens. Journal of Clinical Microbiology 3 : 4 18-424. Rohlf F J 1963 Congruence of larval and adult classifications in Aedes (Diptera : Culicidae). Systematic Zoology 12 : 97-1 17. Shute L A, Gutteridge C S, Berkeley R C W 1985 Pyrolysis mass spectrometry in bacterial systematics. In : Goodfellow M, Minnikin D E (eds) Chemical methods in bacterial systematics. Academic Press, London, pp 95-1 14. Sierra G 1957 A simple method for the detection of lipolytic activity of microorganisms and some observations on the influence of the contact between cells and fatty substrates. Antonie van Leeuwenhoek Journal of Microbiology 23 ; 15-22. Sneath P H A, Sokal R R 1973 Numerical taxonomy. W H Freeman and Co., San Francisco. Sokal R R, Michener C D 1958 A statistical method for evaluating systematic relationships. University of Kansas Science Bulletin 38 : 1409- 1438. Soriano F et al. 1985 Corynebacterium group D2 as a cause of alkaline-encrusted cystitis : report of four cases and characterization of the organisms. Journal of Clinical Microbiology 21: 788-792.
149
Soriano F, Ponte C, Santamaria M, Torres A, Fernandez-Roblas R 1987 Susceptibility of urinary isolates of Corynebacterium group D2 to fifteen antimicrobials and acetohydroxamic acid. Journal of Antimicrobial Chemotherapy 20 : 349-355. SPSSX User’s Guide 1983 McGraw-Hill Book Co., New York, pp 623-645. Stamm W E, Tompkins L S, Wagner K F, Counts G W, Thomas E D, Meyers J D 1979 Infection due to Corynebacterium species in marrow transplant patients. Annals of Internal Medicine 91 : 167-173. Tillotson G, Arora M, Robbins M, Holton J 1988 Identification of Corynebacterium jeikeium and Corynebacterium CDC group D2 with the API 20 Strep system. European Journal of Clinical Microbiology and Infectious Disease I ; 675-678. Thompson J S, Gates-Davis D R, Yong D C T 1983 Rapid microbiochemical identification of Corynebacteriumdiphtheriae and other medically important corynebacteria. Journal of Clinical Microbiology 18: 926-929. Ward J H 1963 Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association 58 : 236-244. Wirsing von Koenig C H, Wichmann S, Finger H 1982 Infections caused by corynebacteria Group JK. Zentralblatt fur Bakteriologie, Mikrobiologie und Hygiene I. Abteilung Original A 253: 370-376. Wishart D 1978 Clustan user manual, 3rd edn. Program Library Unit, Edinburgh University, Edinburgh. Wishart D 1987 Clustan user manual, 4th edn. Computing Laboratory, University of St Andrews. Wolfe J H 1970 Pattern clustering by multivariate mixture analysis. Multivariate Behavioral Research 5 : 329-350.
Downloaded from www.microbiologyresearch.org by IP: 72.249.124.149 On: Sun, 25 Feb 2018 14:57:44