Systematic and Applied Microbiology 29 (2006) 650–660 Raman microspectroscopy as an identification tool within the phylogenetically homogeneous ‘Bacillus subtilis’-group Didier Hutsebaut a , Joachim Vandroemme b , Jeroen Heyrman b , Peter Dawyndt b , Peter Vandenabeele a , Luc Moens a , Paul de vos b, a Department of Analytical Chemistry, Ghent University, Proeftuinstraat 86, B-9000 Ghent, Belgium b Department of Biochemistry, Physiology and Microbiology, Laboratory for Microbiology, Ghent University, K.L. Ledeganckstraat 35,B-9000 Ghent, Belgium Received 24 January 2006 Abstract Vibrational methods have multiple advantages compared to more classic, chemotaxonomic and even molecular microbial tools for the identification of bacteria. Nevertheless, their definite breakthrough in diagnostic microbiology laboratories is determined by their identification potential. This paper reports on the profound evaluation of Raman spectroscopy to identify closely related species by means of 68 Bacillus strains that are assigned or closely related to the phylogenetically homogeneous ‘Bacillus subtilis’-group (sensu stricto). These strains were chosen to represent biological variation within the selected species and to create a realistic view on the possibilities of this technique The evaluation resulted in 49/54 correct identifications at the species level for intern and 15/19 for extern testing. The correct identification of strains, which were not represented in the training set, supports the potential as an identification tool within the ‘B. subtilis group’. Considering the vague borderline between the species studied, Raman spectroscopy can be regarded here as a promising application for identifications at the species level. r 2006 Elsevier GmbH. All rights reserved. Keywords: Raman spectroscopy; ‘Bacillus subtilis’-group; Identification; Species level; External evaluation; Blind study; Carotenoid structure Introduction Vibrational spectroscopic methods are currently studied and developed as powerful new techniques for identification of micro-organisms. The multiple advan- tages of these methods, compared to more classic, chemotaxonomic and even molecular tools, make them a great interest for modern diagnostic laboratories. A wide range of micro-organisms can be discriminated at the species level by Fourier Transform Infrared (FT- IR) spectroscopy [3,26,27] and Raman spectroscopy [14,19]. Both techniques provide chemical information regarding the complete molecular composition of the cells (carbohydrates, fatty acids, proteins, RNA/DNA) [23] and hence combine discriminatory abilities of various phenotypic characteristics (e.g. whole-cell fatty acid and protein composition, pigmentation, etc.). Raman spectroscopy combines high information content, speed of analysis, minimal sample preparation and is, at the same time, less labor intensive than classic ARTICLE IN PRESS www.elsevier.de/syapm 0723-2020/$ - see front matter r 2006 Elsevier GmbH. All rights reserved. doi:10.1016/j.syapm.2006.02.001 Corresponding author. Tel.: +32 09 264 5110; fax: +32 09 264 5092. E-mail address: [email protected] (P. de vos).
11
Embed
Raman microspectroscopy as an identification tool within the phylogenetically homogeneous ‘Bacillus subtilis’-group
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ARTICLE IN PRESS
0723-2020/$ - se
doi:10.1016/j.sy
�Correspondfax: +32 09 264
E-mail addr
Systematic and Applied Microbiology 29 (2006) 650–660
www.elsevier.de/syapm
Raman microspectroscopy as an identification tool within the
Didier Hutsebauta, Joachim Vandroemmeb, Jeroen Heyrmanb, Peter Dawyndtb,Peter Vandenabeelea, Luc Moensa, Paul de vosb,�
aDepartment of Analytical Chemistry, Ghent University, Proeftuinstraat 86, B-9000 Ghent, BelgiumbDepartment of Biochemistry, Physiology and Microbiology, Laboratory for Microbiology, Ghent University,
K.L. Ledeganckstraat 35,B-9000 Ghent, Belgium
Received 24 January 2006
Abstract
Vibrational methods have multiple advantages compared to more classic, chemotaxonomic and even molecularmicrobial tools for the identification of bacteria. Nevertheless, their definite breakthrough in diagnostic microbiologylaboratories is determined by their identification potential. This paper reports on the profound evaluation of Ramanspectroscopy to identify closely related species by means of 68 Bacillus strains that are assigned or closely related to thephylogenetically homogeneous ‘Bacillus subtilis’-group (sensu stricto). These strains were chosen to representbiological variation within the selected species and to create a realistic view on the possibilities of this technique
The evaluation resulted in 49/54 correct identifications at the species level for intern and 15/19 for extern testing. Thecorrect identification of strains, which were not represented in the training set, supports the potential as anidentification tool within the ‘B. subtilis group’. Considering the vague borderline between the species studied, Ramanspectroscopy can be regarded here as a promising application for identifications at the species level.r 2006 Elsevier GmbH. All rights reserved.
Vibrational spectroscopic methods are currentlystudied and developed as powerful new techniques foridentification of micro-organisms. The multiple advan-tages of these methods, compared to more classic,chemotaxonomic and even molecular tools, make thema great interest for modern diagnostic laboratories.
e front matter r 2006 Elsevier GmbH. All rights reserved.
A wide range of micro-organisms can be discriminatedat the species level by Fourier Transform Infrared (FT-IR) spectroscopy [3,26,27] and Raman spectroscopy[14,19]. Both techniques provide chemical informationregarding the complete molecular composition of thecells (carbohydrates, fatty acids, proteins, RNA/DNA)[23] and hence combine discriminatory abilities ofvarious phenotypic characteristics (e.g. whole-cell fattyacid and protein composition, pigmentation, etc.).
Raman spectroscopy combines high informationcontent, speed of analysis, minimal sample preparationand is, at the same time, less labor intensive than classic
ARTICLE IN PRESSD. Hutsebaut et al. / Systematic and Applied Microbiology 29 (2006) 650–660 651
methods. Together, this has set the stage for thedevelopment of Raman spectroscopy into a powerfultool for rapid and inexpensive microbial analysis.
However, before this vibrational technique can beused as a more general routine screening tool, theborders of the taxonomic resolution of the method haveto be evaluated for various well-chosen groups of Gram-positive and Gram-negative bacteria. Indeed, since itconcerns a chemotaxonomic method, the reproducibilityand the taxonomic resolution is affected by themolecular composition of the cell, which in turn dependson variations of growth conditions (medium composi-tion, incubation time and temperature, etc.) [12].
To evaluate the feasibility of Raman spectroscopy asan identification tool, the genus Bacillus [4] was selected.This genus has been split at the generic level[2,8,11,24,36,41,44] while new relatives were discovered[17,25,29,35,37,38,40,43]. Among them, a group ofeight species Bacillus axarquiensis, B. malacitensis,B. mojavensis, B. vallismortis, B. amyloliquefaciens,B. atrophaeus, B. velezensis and B. subtilis (with twosubsp. subtilis and subsp. spizizenii) [21] exists that ishomogenous at the phenotypic and phylogenetic level.These species represent the ‘B. subtilis’-group. Threeadditional species, B. licheniformis, B. sonorensis andB. pumilus are also closely related, yet they arephylogenetically clearly separated from the others. Themajority of phenotypic surveys report that members ofthe ‘B. subtilis’-group are very difficult to discriminate.For example, Roberts et al. [31,32] demonstrated that,for the traits tested, fatty acid composition was the onlyphenotypic character that distinguished B. mojavensis
and B. vallismortis from one another or from B. subtilis.Further, phenotypic discrimination between B. atro-
phaeus and B. subtilis was limited to pigmentation [22],fatty acid composition and a positive oxidase test [32].In another extensive study, B. amyloliquefaciens couldonly be distinguished from B. subtilis by three of the 75phenotypic traits tested [16]. Recently, Ruiz Garcia et al.[42] tested the type strains of all species closely related toB. subtilis on 112 characteristics. These authors foundclear differences between the type strains of all species,e.g. four of the traits enabled the differentiation of B.
amyloliquefaciens and B. subtilis. However, since thisstudy included a single strain per species it did notconsider the variation within the species and moreextensive studies are needed to verify whether thesedifferences are stable identifications for species differ-entiations. Palmisano et al. [28] described B. sonorensis
as a novel species and could phenotypically onlydistinguish it from B. licheniformis by salt toleranceand pigmentation.
In modern bacterial classification, 16S rDNA se-quence analysis is standardly used to allocate strains atvarious taxonomic levels. However, 16S rDNA se-quences often show limited or no variation if closely
related taxa are considered [6,9]. This is also the case forcertain members of the ‘B. subtilis’-group. Indeed, the16S rDNA sequences of the type strains of B. subtilis
(subsp. subtilis and subsp. spizizenii), B. amyloliquefa-
ciens, B. atrophaeus, B. mojavensis, B. vallismortis andthe recently described B. velezensis [33] and B. malaci-
tensis [34], show similarities ranging from 99.2% to99.8% (pairwise similarities based on the UPGMAalgorithm of 16S rDNA sequences obtained from theGenBank/EMBL/DDBJ database). In addition, thetype strains of B. licheniformis and B. sonorensis showa 16S rDNA sequence similarity of 99.6%.
Based on the outlined taxonomic situation, the ‘B.
subtilis’-group is considered as an excellent model in theevaluation of Raman spectroscopy as a technique toidentify closely related species. The current classificationof the strains that were used is well established due topolyphasic characterization. The study aims at testingthe feasibility of identification by Raman spectroscopicanalysis against a (limited) database. In order toevaluate the identification results in a realistic way, ablind approach is used.
Materials and methods
Strains and culture conditions
A total of 68 Bacillus strains assigned to the followingeight species were used:
ARTICLE IN PRESSD. Hutsebaut et al. / Systematic and Applied Microbiology 29 (2006) 650–660652
The strains were selected to represent biological
variation of the species. Pure subcultures were tempora-rily stored at�20 1C inMicrobank vessels under 15% v/vglycerol cryoprotection before use.
Strains were grown on brain heart infusion (BHI) thatwas freshly prepared (29.6 g BHI (Becton, Dickinsonand Company) and 16.0 g BactoTM Agar (Becton,Dickinson and Company) were mixed in 800ml distilledwater, stirred for 15min, autoclaved for 20min at 121 1Cand poured into Petri dishes) to comprise the effect ofpotential variation in medium preparation on theRaman spectra. Before use, Petri dishes were incubatedovernight at 28 1C, checked for contamination andstored in a closed plastic bag at room temperature formaximum 1 week.
Prior to Raman analysis, bacteria were grown over-night on BHI at 37 1C. A single well-isolated colony waspicked and streaked again on BHI agar plate using thethird quadrant method. The Petri dishes were thenincubated at 37 1C (71 1C) for exactly 7 h.
Sample preparation
Confluently grown cells were harvested from the thirdquadrant [45] with a disposable plastic inoculation loopand manually spotted on a CaF2 substrate (polished,UV-grade, Crystran, Poole, United Kingdom). If noconfluent growth was present in the third quadrant,biomass was taken from the second quadrant. Thespotted biomass was allowed to dry at room tempera-ture for 20min in a desiccator containing drying beads.
Planning of the experiments
To accommodate potential sample heterogeneity, fourspectra were collected at randomly chosen locationswithin the spotted biomass with a signal collection timeof 60 s, each. An ‘experiment’ is considered as the meanof four Raman measurements of a strain grown on afreshly prepared culture plate (BHI) for 7 h at 37 1C on aparticular day. At least three replicate experiments wererecorded for each strain on different days, usingdifferent medium batches. To eliminate the potentialinterference of unpredictable day-to-day variations,planning of the experiments was completely rando-mized. Data were collected over a 5-month periodencompassing 19 measuring sessions. Each measuringsession consisted of 10–20 experiments. The complete setcontained 219 experiments.
Raman microspectroscopy
Raman spectroscopic measurements were performedsimilarly to Maquelin et al. [18]. In brief, the CaF2
substrate with the dried smears was placed directly
under the microscope of a Kaiser System Hololab5000R modular Raman microspectrometer (f/1.8)(KOSI, Ecully, France). The microscope was fitted witha 100� objective (PL Fluotar L, 100� /0.75, W.D.4.7mm, Leica). Samples were excited using 45–50mWof 785 nm laser light from a diode laser (TopticaPhotonics AG, Martinsried/Munich, Germany). Thescattered light is guided to the spectrograph by means ofa confocal, 15 mm aperture collection fiber. A back-illuminated deep depletion CCD camera (Andor, Bel-fast, Northern Ireland) was used for the detection of thescattered light. Raman signal was collected in thespectral interval of 150–3100 cm�1 with a spectralresolution of 4 cm�1. Calibration of the absolutewavelength-axis was performed using the known wave-lengths of the atomic lines from neon. The referencespectrum of a tungsten band lamp (Optronic labora-tories, Orlando, USA) operated at known fixed currentwas used to correct for the wavelength dependent signaldetection efficiency of the Raman set-up [30].
Analysis of Raman data
Data pre-processing
Dark signal and constant optical background arisingfrom optical components in the Raman set-up weresubtracted from each spectrum. All four spectracollected from the same strain were standardizedtowards a reference spectrum [13] and the firstderivatives were calculated to minimize the influence ofbackground signal caused by slight sample fluorescence.Consequently, the data were averaged for furtheranalysis and treated as one single spectrum (oneexperiment ¼ one average spectrum). Further datapre-processing consisted of cutting out the fingerprintregion (370–1750 cm�1) and scaling to zero mean andunit variance (SNV). The resulting data for each strainwere used in the multivariate analysis.
Multivariate data analysis
After autoscaling, data reduction was first performedusing principal component analysis (PCA) which isperformed with the PLS toolbox (Eigenvector ResearchInc., Manson, WA, USA) for Matlab software (Math-works Inc., Natick, MA, USA). PCA is a well-knownmethod to reduce the dimensionality of the data [5,7,39].The maximum number of n�1 principal components(PCs) was calculated (n ¼ the number of spectra in theanalysis), and those PC’s accounting for 99.9% of thevariation in the data set were retained.
An F-test was used to individually select the mostsignificant PC’s (ao0:05) allowing discriminationamong species. The selected scores served as input forlinear discriminant analysis (LDA) [39] and j�1
canonical (j ¼ number of different species) variates were
ARTICLE IN PRESSD. Hutsebaut et al. / Systematic and Applied Microbiology 29 (2006) 650–660 653
retained in the subsequent identification. LDA was usedonly to calculate the canonical variates (CV’s) minimiz-ing the within-species variance, while maximizing thevariation between different species. In the following, theCV’s thus obtained will be referred to as ‘pattern’. Sincethe LDA model was trained at the species level, strainspecific information was neglected.
Description of data set and identification
During internal evaluation the complete dataset (219experiments) was randomly splitted in a training set(3/4) and a test set (1/4). The test set, an assemblage of54 experiments, was treated as if it concerned indepen-dent, ‘unknown’ samples. Therefore, the data of the testset were projected on the PCA-space as determined bythe training set and the significant PC’s (ao0:05) wereused to calculate the patterns of the test set. Based onthe resulting patterns, the similarities (Euclidean dis-tance) between the 54 experiments of the test set and allexperiments in the training set were calculated. Theidentification was regarded as reliable at the species levelwhen at least three out of five matches pointed to thesame species. Since only the first five matches wereconsidered, this corresponds to a relative proportion ofidentification X0.6. The training set consisted of 165( ¼ 219–54) experiments.
For external evaluation a set of 19 strains wasassembled independently and in a blind way. Thesetester strains encompass (1) strains that were already
Fig. 1. Representative Raman spectra obtained by averaging all R
atrophaeus LMG 16797T, (c) B. mojavensis LMG 17797T, (d) B. subti
were grown on BHI at 37 1C for 7 h. Raman bands caused by caroten
vertically on the intensity axis for clarity (a.u.: arbitrary units).
included in the training set, (2) strains not represented inthe training set, but known to be members of speciesrepresented in the training set, and (3) one strainbelonging to a species other than those represented inthe training set. Strains of this blind study were analyzedin duplicate, using freshly prepared BHI growth media.These 19 strains of which the identity was unknown forthe experimenter, were identified by comparison withthe training set exhibiting the highest similarities(Euclidean distance).
Results and discussion
Comparison of Raman spectra
Representative Raman spectra of some investigatedtype strains are shown in Fig. 1. Correspondingly theirhigh genetic similarity, the Raman spectra of thesestrains are highly similar under the specified growthconditions, with the exception of striking bands near1156 and 1520 cm�1 present in the Raman spectra ofLMG 18725T. These bands together with the band at1004 cm�1 can be attributed to b-carotene [42] and arenot reliable for species discrimination since the Ramanspectra of B. vallismortis LMG 17800 does not showthese bands (data not shown). Strain-dependent carote-noid levels, as measured by Raman analysis, were alsoreported for Enterococcus casseliflavius and E. hirae [15].
aman spectra of (a) B. amyloliquefaciens LMG 9814T, (b) B.
lis LMG 7135T, and (e) B. vallismortis LMG 18725T. All strains
oid structures are highlighted. The spectra have been displaced
ARTICLE IN PRESSD. Hutsebaut et al. / Systematic and Applied Microbiology 29 (2006) 650–660654
Since Raman spectra provide phenotypical informa-tion, it is noted here that a reliable discrimination (andtaxonomic resolution) will only be achieved whenexperimental factors such as preparation and composi-tion of growth media, incubation conditions and theprotocols for measuring and analyzing the Ramanspectra are optimal and reproducible [12].
Internal evaluation of the reference library
To evaluate the identification ability of Raman spectro-scopy, 68 strains belonging to eight species were used forthe construction of a database covering the biologicalvariation per species. In some cases, the biological speciesvariation is unknown because of the lack of sufficientstrains. However, all strains were selected for their knowncorrect assignment at the species level.
For internal evaluation, the database was randomlydivided in a training set (3/4) and a test set (1/4). Theidentification of test entries is based on the highest fivematches with entries of the training set at the species level.The number of times a given species name appears inthese first five matches is represented by its relativeproportion. This proportion equals one, if five out of fivematches point towards the same species, 0.8 if four out offive matches point towards the same species, etc. Theresults are reported in Table 1. The relative proportionsmay be regarded as an appreciation of correct identifica-tion. Misidentifications are marked with an asterisk (*).
Overall identification accuracy at the species level ofmore than 90% (49/54) was obtained. Only threepatterns were misidentified and two could not beidentified unequivocally. Misidentifications concernedstrains attributed to B. amyloliquefaciens, B. subtilis andB. mojavensis. When grown as specified, these species aredistinguished only by subtle spectral differences. Twopatterns obtained from B. amyloliquefaciens (nr. 6. inTable 1) and B. atrophaeus (nr. 11 in Table 1), could notbe identified because the first five matches correspondedto at least three different species without a clearpreference for one of them.
The internal evaluation did not reveal misidentifica-tions for B. atrophaeus and B. vallismortis strains.Although some of the tested patterns matched withmore than one species, the majority of the matches(3 out of 5) pointed to the same species of the‘B. subtilis’-group and the satellite species B. pumilus,B. licheniformis and B. sonorensis. In this context(Table 1), two situations in which the relative propor-tion of identification equals 0.6 can occur (1) 0.6, 0.2, 0.2(indicating a three species match) and (2) 0.6, 0.4(indicating a two species match). Clearly, the firstidentification profile is more explicit and is probablymore reliable. The results of the internal evaluationunderline the potential of Raman spectroscopy to
discriminate highly related species based even on subtlespectral differences.
External evaluation: blind study
One of the drawbacks of internal evaluations is thatthese are sometimes not realistic in terms of identifica-tion accuracy. As such, internal evaluations are knownto exhibit higher identification accuracies than whenindependent, external evaluations are performed [27].Therefore, a modest blind study was set-up, based on 19tester strains that were analyzed in duplicate (i.e. in total38 experiments) and identified towards the training set.To maximize the eventual effect of medium preparationand/or batches, spectra were collected from growth onindependently prepared medium batches.
Identification results of the blind study (strains A–S)for each replicate experiment are shown in Table 2.From the eleven strains that a posteriori turned out tobe included in the training set, all but one (Unknown F)were correctly identified. The identification accuracy of91% (10/11) supports the internal evaluation andherewith the reproducibility of the method followingthe protocol. The patterns of unknown F pointed to B.
subtilis (relative proportion of identification ¼ 0.6) forthe first replicate experiment, but did not reach thethreshold (relative proportion of identificationX0.6) forthe second replicate experiment. Therefore, the cumula-tion of both experiments did not lead to convincingspecies identification, although strain LMG 12263 wasincluded in the training set.
The blind study also reveals the potentiality of themethod as a general identification tool because of theinclusion of eight strains that were not representedhitherto in the database. Five out of these eight strainswere correctly identified at the species level, which isparticular encouraging considered the small spectraldifferences between the species studied. Hence, it isagain concluded that even these subtle differences aresufficiently reproducible for species discrimination.
The other three strains, not included in the database(unknowns B, G and P), were misidentified formiscellaneous reasons. For example, the colony ofunknown B exhibited orange pigmentation. Intensepeaks at 1134 and 1532 cm�1 in the Raman spectra(not shown) of this strain pointed to the presence ofcarotenoid structures [1]. Since the Raman spectra ofthis strain deviated importantly from the spectraencompassed by the training set, unknown B wascategorized a priori as unknown. Unknown B wasdisclosed as B. atrophaeus LMG 8198 ( ¼ NCIB 8058,DSM 675, ATCC 9372), also named the ‘red strain’ [10].
Unknown P was disclosed as the recently describedtype strain of B. velezensis LMG 22478T [33], a memberof the homogeneous ‘B. subtilis’-group that was not
D. Hutsebaut et al. / Systematic and Applied Microbiology 29 (2006) 650–660 657
ARTICLE IN PRESSD. Hutsebaut et al. / Systematic and Applied Microbiology 29 (2006) 650–660658
represented by the training set used during the blindstudy. Hence, the non-identification was a correctanswer. Together, based on the constructed spectrallibrary used during the blind study, 80% of the strainswere correctly identified.
In conclusion, The Raman method as described hereshould not be considered as a stand-alone technique fortaxonomic purposes. It is more an additional tool bothin the polyphasic characterization process of bacteria, aswell as for screening purposes allowing fast identifica-tion at the species or perhaps even finer levels.Supplementarily, it is noted that in this type of whole-cell approaches, the interpretation of the Raman spectramight be affected by over-expressed and/or straindependent characteristics (e.g. the presence of carote-noid structures). Also, the observation that Ramanspectra of closely related species are less-suited for directvisual evaluation might be considered as a drawback,e.g., in relation to SDS-PAGE or AFLP [6].
Modest internal and external evaluations indicate thatRaman spectroscopy applies well as a stable identificationmethod since at least 90% of the strains that wereincluded in the reference data set. However, includingthese strains that were not included in the data set overallidentification accuracy is estimated at 80%. Difficultiesand contradictions in the identifications were observedwithin taxa that are most probably not clearly separated.
Another condition sine qua non, concerns thecomposition of the data set that is used for identifica-tion. Indeed, only unambiguously classified strains,covering the biological variation of the taxon, can beincluded. These results support previous studies on theevaluation of Raman spectroscopy for the identificationof micro-organisms [14,15,20].
A final advantage concerns the very limited samplepreparation and the short period in which the identificationresults are obtained (less than 8h, starting from incubation).
Acknowledgements
The authors are grateful to the Research Founda-tion—Flanders (FWO-Vlaanderen) for financial support(Grants G.0156.02 and G.0343.03) of this researchproject. D.H. greatly acknowledges the Institute forthe Promotion of Innovation through Science andTechnology in Flanders (IWT-Vlaanderen) for hisDoctoral Grant. J.H. and P.V. wish to thank theFWO-Vlaanderen for their postdoctoral fellowships.
Reference
[1] G.A. Armstrong, Genetics of eubacterial carotenoid bio-
synthesis: a colorful tale, Ann. Rev. Microbiol. 51 (1997)
629–659.
[2] C. Ash, F.G. Priest, M.D. Collins, Molecular identifica-
tion of rRNA group 3 bacilli (Ash, Farrow, Wallbanks
and Collins) using a PCR probe test. Proposal for the
creation of a new genus Paenibacillus, Antonie Van
Leeuwenhoek 64 (1994) 253–260.
[3] U. Behrendt, A. Ulrich, P. Schumann, D. Naumann, K.
Suzuki, Diversity of grass-associated Microbacteriaceae
isolated from the phyllosphere and litter layer after
mulching the sward; polyphasic characterization of