Characterization and Identification of Clinically RelevantMicroorganisms Using Rapid Evaporative Ionization MassSpectrometryNicole Strittmatter,† Monica Rebec,‡ Emrys A. Jones,† Ottmar Golf,† Alireza Abdolrasouli,‡ Julia Balog,†

Volker Behrends,† Kirill A. Veselkov,† and Zoltan Takats*,†

†Section of Computational and Systems Medicine, Department of Surgery and Cancer, Imperial College London, London SW7 2AZ,United Kingdom‡Department of Microbiology, Imperial College Healthcare NHS Trust, Charing Cross Hospital, London W6 8RF, United Kingdom

*S Supporting Information

ABSTRACT: Rapid evaporative ionization mass spectrometry (REIMS) wasinvestigated for its suitability as a general identification system for bacteria andfungi. Strains of 28 clinically relevant bacterial species were analyzed in negativeion mode, and corresponding data was subjected to unsupervised and supervisedmultivariate statistical analyses. The created supervised model yielded correctcross-validation results of 95.9%, 97.8%, and 100% on species, genus, and Gram-stain level, respectively. These results were not affected by the resolution of themass spectral data. Blind identification tests were performed for strains culturedon different culture media and analyzed using different instrumental platformswhich led to 97.8−100% correct identification. Seven different Escherichia colistrains were subjected to different culture conditions and were distinguishablewith 88% accuracy. In addition, the technique proved suitable to distinguish fivepathogenic Candida species with 98.8% accuracy without any further modificationto the experimental workflow. These results prove that REIMS is sufficiently specific to serve as a culture condition-independenttool for the identification and characterization of microorganisms.

Development of a fast and reliable identification system formicroorganisms is a rapidly developing field.1 Although a

number of concepts have been proposed, none have proven tobe specific, universal, fast, and at the same time cheap enoughto find widespread application. Until today, in routine clinicalmicrobiology settings, identification of an isolate is mostlyaccomplished by observing phenotypic characteristics such ascolonial morphology, Gram-stain behavior, and differentenzymatic properties or carbon source utilization patterns.However, these techniques are time-consuming, need experi-enced personnel, and often lack specificity.Microbial species are defined by their 16S rRNA sequence;

thus, sequencing of the 16S rRNA encoding gene serves as thegold standard for bacterial identification and classification.Partial or full 16S rRNA sequencing has the advantage of beingculture-independent and thus is especially valuable forfastidious microorganisms. However, the sensitivity andspecificity for direct sample applications varies considerably.2

Despite of its role as gold standard, in some cases bacterialspecies cannot be identified confidently by 16S rRNA whichcan still make the application of additional techniquesnecessary.3,4 In addition, genotypic methods generally needextensive sample preparation, are comparably expensive, andstill need at least several hours for identification. Due to thesereasons, sequencing methods are rarely applied in routine

clinical settings but mostly find application in referencelaboratories.Mass spectrometry (MS) gained attention for microbial

identification more than 4 decades ago due to its intrinsicadvantages of fast data acquisition, high sensitivity, andspecificity.5−7 Early studies were mostly applying pyrolysisfollowed by electron impact or chemical ionization. Due to thedestructive nature of pyrolysis methods, only small moleculesand fragments of larger molecules were detected using theseapproaches.5 A decade later, the introduction of fast atombombardment (FAB-MS) allowed monitoring of largerbiomolecules as intact complex phospholipid species desorbeddirectly from intact bacterial cells.8 Eventually, the advent of thesoft ionization techniques, especially matrix-assisted laserdesorption ionization (MALDI), gave significant momentumto mass spectrometric microbial identification.1,9 Following theinitial reports, MALDI found widespread use for the ionizationof proteins desorbed from intact bacterial cells.1 Nowadaysanalysis is focused on the mass range of 2−20 kDa resulting inthe detection of various protein signals, half of which are ofribosomal origin.10,11 Commercialized systems were demon-

Received: March 25, 2014Accepted: June 4, 2014Published: June 4, 2014

Article

pubs.acs.org/ac

© 2014 American Chemical Society 6555 dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−6562

strated to give comparable or even superior results toconventional identification systems.12−15 Although MALDIcan be used directly on intact cells, it was shown that matrixaddition itself leads to lysis of the bacterial cells and thus releaseof the intracellular proteins. However, short additionalextraction steps significantly increase the identificationaccuracy, especially in case of yeasts and Gram-positivebacteria.12,16 With individual analysis times of 20−90 s perprecultured isolate and reporting times as low as 6 min, thesesystems are able to reduce the average turnover time in clinicalmicrobiology laboratories by about 1 day. In addition, analysiscosts are a fraction of those of conventional techniques.15,17

Besides the generally applied intact protein profilingmethodology for MALDI-based microbial identification sys-tems, peptide mixtures resulting from tryptically digestedproteins can be analyzed to gain further specificity in abottom-up approach termed shotgun mass mapping (SMM).18

The inherent speed of analysis, high sensitivity, andspecificity combined with the good agreement with 16SrRNA sequencing led to the widespread use of MALDI-MSfor the identification of microorganisms, both in research and inclinical microbiology laboratories.1 Commercial MALDI time-of-flight (TOF) microbial identification systems are nowavailable for clinical microbiology routine use in the EuropeUnion, as well as in many countries around the world, and mostrecently gained approval by the U.S. Food and DrugAdministration.A growing interest in lipidomics, and the advent of ambient

ionization techniques as an easy means to generate lipidprofiles, gave new momentum to lipid profile-based identi-fication of microorganisms. Traditionally the only generalidentification system based on bacterial lipid composition hasbeen fatty acid profiling using gas chromatography and flameionization detector (GC-FID). However, bacterial fatty acidprofiles are strongly affected by culturing conditions, whereasthe composition of intact membrane phospholipids provedmore robust. Since the first application of desorption ionizationmethods to obtain lipid spectra from intact bacterial cells in1987,19 many different ionization techniques, including fastatom bombardment (FAB),8 electrospray ionization(ESI),20−22 MALDI,23,24 and desorption electrospray ionization(DESI)25,26 have been used to demonstrate that differentbacteria have species-specific phospholipid profiles. However,none of these methods have been shown to possess thespecificity and robustness required to serve as the basis for ageneral lipid-based identification system. Further interestingdevelopments include mass spectrometric techniques to analyzebacterial metabolites and phospholipids in vivo and directlyfrom the Petri dish using ambient ionization techniques asDESI,27 nanospray desorption electrospray ionization,28 andother liquid microjunction−electrospray setups.29

The recently developed technique rapid evaporativeionization mass spectrometry (REIMS) yields highly specificphospholipid profiles of different biological tissue types andalso offers a new opportunity for the development of a lipid-based, sample-preparation-free microbial identification sys-tem.30,31 In case of REIMS analysis, species-specific massspectral fingerprints are generated by subjecting the cellularbiomass to radiofrequency alternating electric current. Thermaldisintegration of cells produces an aerosol comprising lipid-covered droplets containing intracellular and extracellularmetabolites, which is introduced into the mass spectrometerfor subsequent analysis. The REIMS method shows high

methodological resemblance to pyrolysis approaches; however,unlike pyrolysis systems, REIMS provides soft ionizationyielding predominantly molecular ions.30,31

We have previously described a proof-of-principle studyusing REIMS for microbial identification.31 Our current studyprovides a comprehensive and critical assessment of REIMS-based microbial identification.

■ EXPERIMENTAL SECTIONCulturing of Bacterial Strains. Isolated strains of various

microorganisms were grown on a range of solid agar-basedmedia commonly used in clinical microbiology settings. Mediawere purchased from Oxoid (Basingstoke, U.K.). The bacteriawere incubated under various atmospheric conditions at 37 °Covernight before analysis. For more information on cultureconditions refer to Supporting Information Tables S-4 and S-6.Microorganisms were isolated during routine clinical micro-biological workflow and identified using conventional work-flows and a Microflex LT MALDI TOF instrument (BrukerDaltonics, Bremen, Germany).

REIMS Analysis. For REIMS analysis, two hand-heldelectrodes in form of a forceps were used as the samplingprobe (bipolar forceps, obtained from Erbe Elektromedizin,Tubingen, Germany). A Valleylab Force EZc power-controlledelectrosurgical unit (Covidien, Dublin, Ireland) was used at 60W power setting in bipolar mode as rf alternating currentpower supply (470 kHz, sinusoid). An approximately 1.5 mlong 1/8 in. outer diameter, 1/16 in. inner diameter PTFEtubing (Fluidflon PTFE tubing; LIQUID-scan GmbH Co. KG,Uberlingen, Germany) was applied to connect the embeddedfluid line of the bipolar forceps and the inlet capillary of eitheran LTQ Orbitrap Discovery instrument (Thermo ScientificGmbH, Bremen, Germany), a Thermo Exactive instrument(Thermo Scientific GmbH), or a Xevo G2-S Q-TOFinstrument (Waters Micromass, Manchester, U.K.). In eachcase the inherent vacuum system of the mass spectrometer wasused for aspiration of the aerosol. This setup is shown in Figure1, while instrumental settings are given in SupportingInformation Table S-1.Mass spectrometric analysis of the microorganisms was

performed directly from the solid culture medium (Figure 1).An amount of 0.1−1.5 mg of microbial biomass was scraped offthe agar surface using one of the electrodes of the bipolarforceps. The two electrodes were subsequently brought intoclose proximity (i.e., by pinching the biomass between the tipsof the forceps), and the rf power supply was triggered using afoot switch. The microbial biomass is rapidly heated up due toits nonzero impedance, and an aerosol containing the analytesis produced and transferred directly into the mass spectrometer.Five individual measurements were performed for each strainand averaged as a database entry.

Data Analysis. Raw mass spectrometric files were importedas imzML format32 into MATLAB (Mathworks, Natick, MA;http://www.mathworks.co.uk/) for data preprocessing, patternrecognition analysis, and visualization. All REIMS spectra werelinearly interpolated to a common sampling interval of 0.01 Da.Recursive segmentwise peak alignment was then used toremove small mass shifts in peak positions across spectralprofiles.33 The aligned data were subjected to data normal-ization (median, mean, or TIC normalization) and log-basedtransformation to ensure that the noise structure was consistentwith the downstream application of multivariate statisticaltechniques.34 Principal component analysis (PCA) and

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626556

hierarchical cluster analysis (HCA) were used for unsupervisedanalysis of the data set. A recursive maximum margin criterion(RMMC)35 algorithm and linear discriminant analysis (LDA)were used as supervised classification algorithms. A moredetailed description of the data analysis workflow is given in theSupporting Information.Ionic species in the mass spectra were identified based on

exact mass measurements (mass deviation <3 ppm) and MS/MS fragmentation patterns.Safety Considerations. To avoid any infection caused by

aerosolized pathogenic bacteria, the analysis site was enclosedinto a Class II safety-level glovebox compartment equippedwith UV light source and HEPA filters.

■ RESULTS AND DISCUSSIONREIMS Spectral Content. Although the ionization

mechanism is in theory expected to generate equal numbersof positive and negative ions, the vast majority of bacterialspecies give only poor quality spectra in positive ion mode(Supporting Information Figure S-2). Conversely, 400−1600exclusively singly charged spectral features were detected usingREIMS in negative ion mode in the mass range of m/z = 150−2000. Most REIMS spectral profiles (see Figure 2 and

Supporting Information Figure S-2) are dominated by intactphospholipids in the mass range of m/z = 600−900. Thesesignals are mostly derived from phosphatidylglycerols (PGs),phosphatidylethanolamines (PEs), and phosphatidic acids(PAs). Signals with lower mass-to-charge ratio were associatedwith fatty acids (C12−C20), monorhamno- and dirhamnoli-pids, and a range hydroxyalkylquinolines-derived quorumsensing molecules (including PQS) for Pseudomonas aerugino-sa,36,37 ceramides Cer(34:0), Cer(35:0), Cer(36:0) forBacteroides fragilis,38 and short-chain mycolic acids with C26−C36, including corynomycolic acid (C32H64O3), for Coryne-bacterium species.39 Cardiolipins were identified in the highermass range for all bacterial species analyzed. The signals withhighest mass-to-charge ratio so far were tentatively identified asintact lipid A species in case of Helicobacter pylori (m/z = 1547)and Escherichia coli (m/z = 1796). Besides these lipid speciesand lipid-related species, bacterial polyhydroxybutyrate poly-mers were identified for Bacillus cereus and Burkholderia cepaciacomplex strains.40,41 For more details on the nature ofidentified compounds see the Supporting Information (FiguresS-3 and S-4, Tables S-2 and S-3).In addition, no signals attributed to the growth medium were

observed using REIMS, neither when the agar surface is leftintact during analysis nor when agar is deliberately analyzed.This is tentatively associated with the nature of the

Figure 1. (A) Setup used for analysis of bacteria by REIMS. (B)Scheme of analysis. Microbial biomass is held between the twoelectrodes of the irrigated bipolar forceps, electrical current is applied,sample is evaporated thermally, and the produced aerosol is aspiratedinto the opening of the embedded fluid line.

Figure 2. REIMS spectral profiles obtained for Staphylococcus aureusATCC 25923, P. aeruginosa ATCC 27853, and E. coli ATCC 25922,each grown on five different solid growth media (from front to back:brain-heart infusion agar, columbia horse blood agar, chocolate agar,Mueller−Hinton agar, Trypticase soy agar); acquired using Exactiveinstrument.

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626557

carbohydrate matrix, which undergoes condensation reactionsvia water losses, resulting in extensive cross-linking andeventually charring on heating, efficiently hindering REIMSmechanism.Full scan negative ion mode REIMS profiles for reference

strains of E. coli, P. aeruginosa, and S. aureus are depicted inFigure 2. Each strain was cultured on five different solid growthmedia. Only a minor influence of culturing conditions can beobserved that is mostly resulting in small changes of signalintensity, whereas the overall spectral appearance remainsunchanged and contains a large proportion of conserved peaks.

Comparison of Experimental Setups. Different exper-imental setups following the scheme of monopolar and bipolardiathermy were tested for their suitability as ion sources. Bothsetups (shown in Supporting Information Figure S-6) couldgenerate mass spectral fingerprints of intact bacterial cells. Forthe monopolar setup, bacterial biomass is placed on a largesurface area electrode and the electrical circuit is closed by asharp, hand-held counter electrode. Alternatively, the agar canbe removed from the Petri dish and placed on the counterelectrode. The monopolar setup produces a very high electriccurrent density in the proximity of the hand-held electrodewhich results in high effective evaporation temperatures and

Figure 3. (A and B) Pseudo-two-dimensional PCA and RMMC plots of a model comprising 28 different clinically relevant bacterial species. Gram-positives are indicated by circles (○); Gram-negatives are represented by squares (□). Panel C shows results of HCA of the same data set.

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626558

excessive solid aerosol (soot) formation. In agreement withthese observations, the spectral quality is poor in this case andspectra do not feature intensive signals above m/z = 1000 (seeSupporting Information Figure S-7). This observation wastentatively associated with a high degree of charring andeffective combustion of the sample which cannot be circum-vented by decreasing the electrical current. In addition, thelarge amount of smoke produced (and aspirated by theinstrument) results in significant contamination of the ionoptics. In contrast, the bipolar setup features a pair of hand-heldelectrodes in the form of forceps as the sampling probe andoffers a more evenly distributed electrical current between thetwo electrodes (see Supporting Information Figure S-6B). Thissetup proved to be superior in comparison to the monopolarsetup for the analysis of minor amounts of biomass. Theforceps-based setup provided easier handling, elimination ofmemory effects, and less frequent blocking of transfer devices.In addition, less fragmentation and better sensitivity for highermasses were observed using the bipolar tool. The most effectiveand reproducible aspiration of the produced aerosols wasachieved using an electrode setup with an embedded aspirationline (see Figure 1). The comparatively lower amount of aerosolproduced by the bipolar tool is due to smaller amounts ofbiomass needed for analysis (higher sensitivity) and lowercurrent density due to the parallel geometry of the electrodes.Secondary electrospray ionization was tested using theexperimental setup which was described earlier for thepostionization of surgical aerosols. In agreement with earlierstudies, further increase in sensitivity was obtained only inpositive ion mode.42

Identification of Bacteria. Suitability of the REIMSmethod for bacterial identification requires that interspeciesspectral variance needs to be larger than the intraspeciesvariance for different strains and phenotypes of the samespecies. In order to investigate whether this basic condition isfulfilled, a data set was created comprising 15 different clinicalisolates for each of 28 different bacterial species (seeSupporting Information Table S-4 for more information onculturing conditions).Results of unsupervised and supervised analyses of the

generated data set are given in Figure 3, parts A and B,respectively. Generally the plots resulting from the supervisedand unsupervised analysis of REIMS data show high similarityto each other. This is due to the fact that REIMS featuresexclusively signals originating from the sample (i.e., not fromany chemical background). This is an advantage compared toDESI or MALDI MS where solvent- and matrix-related signalssignificantly contribute to the overall spectral information.Gram-positive and Gram-negative species are separated along

the first multivariate component in both PCA and RMMCanalysis (Figure 3, parts A and B). Compared to Gram-negativespecies, Gram-positive bacteria generally show a higher amountof saturated phospholipid species and lower relative abundanceof phosphatidylethanolamines. These observations are inagreement with the bacterial cell membrane compositionreported in the literature.43 Hierarchical cluster analysis wasperformed in order to investigate how well the REIMS spectralprofiles follow the bacterial taxonomy as determined by 16SrRNA gene sequences. Figure 3C shows that spectral profiles ofclosely related bacterial species are grouped closely togetherwhile rather unrelated bacterial species group separately. Forthe Gram-positive species this is visible for each of theStaphylococcus spp. (S. aureus, S. capitis, S. epidermidis, S.

hominis, and S. haemolyticus), Streptococcus spp. (S. agalactiae, S.pneumoniae, S. pyogenes), and two Enterococcus spp. (E. faecalisand E. faecium). Streptococcus and Enterococcus spp. which bothbelong to the Lactobacillales order are further situated on thesame cluster in the HCA. Regarding Gram-negative species, allmembers of the Enterobacteriaceae family (members of thegenera Escherichia, Citrobacter, Enterobacter, Proteus, Morganella,Klebsiella, and Serratia) grouped closely together. Furthermore,P. aeruginosa, Moraxella catarrhalis, and B. cepacia complexstrains are all located together in a separate cluster whencompared to the other Gram-negative species. Pseudomonasspp. and Moraxella spp. are both part of the Pseudomonadalesorder. Although B. cepacia complex strains belong to β-Proteobacteria today, they were previously classified into thePseudomonas genus,44 thus indicating a high phenotypicsimilarity between Pseudomonas spp. and Burkholderia spp.which explains their proximity on the HCA dendrogram. Thesame trends were observed in the PCA plots for Gram-positiveand Gram-negative-species only (see Supporting InformationFigure S-8). These results demonstrate that REIMS spectralprofiles largely follow taxonomical trends. The overall agree-ment with the bacterial taxonomy is expected to increase withlarger coverage of bacterial diversity among different phyla,classes, and orders.The presented RMMC model (Figure 3B) was cross-

validated in order to assess the specificity of the REIMSmethod on Gram, genus, and species level (see SupportingInformation Figure S-9 for supervised models on genus andGram level). Cross-validation was performed using leave-one-out cross-validation and three nearest neighbors as classifier. Inleave-one-out cross-validation, each data point is left out fromthe supervised model once and then projected into thegenerated data space and classified according to a givencriterion (here according to its three nearest neighbors).Therefore, this type of cross-validation shows similarity withblind identification tests and is applied here to assess thespecificity of the method. However, for Gram-level cross-validation entire bacterial species were left out to perform cross-validations. Cross-validation results yield 95.9%, 97.8%, and100% correct classification at species, genus, and Gram level,respectively. Misclassifications were only observed for closelyrelated bacterial species and comprised three misclassificationswithin the Staphylococcus genus, five misclassifications withinthe Enterococcus genus, and nine misclassifications within theEnterobacteriaceae family. Misclassifications are defined incomparison to identifications provided by a commercialMALDI TOF MS instrument as obtained during routineclinical work. The correct identification statistics is thereforenot expected to be better than those reported in literature forcommercial MALDI MS systems.

Dependence of the Correct Identification Perform-ance on Spectral Resolution. In order to assess the massresolution required to reliably distinguish between the differentbacterial species as shown in Figure 3, the original data wasbinned using varying bin sizes. Aligned raw data was normalizedto the total ion count (TIC) and log-transformed for each binsize individually before unsupervised and supervised multi-variate analyses were carried out (for respective PCA andRMMC plots refer to Supporting Information Figure S-10).The required minimum resolution was assessed by comparingthe results of cross-validations for each bin size. As for themodel shown in Figure 3, leave-one-out cross-validation was

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626559

used at species and genus level, whereas leave-species-out cross-validation was performed at Gram-stain level.The results visualized in Figure 4 show that no considerable

loss in identification accuracy was observed at species, genus, or

Gram-stain level for different bin sizes between 0.01 and 1 Da.These data strongly suggest that a mass analyzer working atunit resolution would be sufficient for the identification ofunknown bacteria. Even a bin size of 5 Da does not result in asignificant loss of identification accuracy, clearly indicating thepresence of major spectral differences even between closelyrelated species. Nevertheless, performing analysis on a high-resolution mass spectrometer facilitates identification ofunknown spectral features by exact mass measurements. Inaddition, using high-resolution data leads to more compact datagroups in PCA (see Supporting Information Figure S-10) andimproved information recovery.Interinstrument/Robustness Study. An independent set

of clinical isolates for blind identification tests was cultured onvarious different growth media (see Supporting InformationTable S-6 for details) and subsequently analyzed using anOrbitrap Discovery or a Xevo G2-S Q-TOF instrumentequipped with modified atmospheric interface. Acquiredspectra (full mass range) were classified based on a subsetcomprising nine different species of the previously generatedmodel shown in Figure 3. Since a 1 Da sampling intervalproved sufficiently specific, these blind identification tests wereperformed at a common bin size of 1 Da. This furthercompensates for the differences in mass resolution of thedifferent instruments.The outlined workflow leads to 100% correct blind

identification results for strains acquired on the Orbitrapinstrument (same mass analyzer) and 97.8% correct identi-fication for the files acquired using the Xevo Q-TOFinstrument (see confusion matrix in Supporting InformationTable S-7). The two misclassifications are tentatively attributedto the differences in noise structure between the different massanalyzers. In addition, as the RMMC plot shown in Figure 5demonstrates, very little separation originating from thedifferent instruments and or mass analyzers is shown in caseof supervised analysis and the interspecies variance clearlydominates the plot. This demonstrates that the spectral profilesobtained using REIMS are highly reproducible, even consider-ing the conceptually different atmospheric pressure interfaces ofthe two different types of instruments.

Strain-Level Specificity. The specificity of the presentedmethod goes beyond genus- and species-level specificity as it isdemonstrated for a set of seven standard E. coli laboratorystrains (NCM3722, MG1655, MC1000, MC4100, DH5a,C600, and OP50). The strains NCM3722, MG1655,MC1000, MC4100, DH5a, and C600 are all derived from theK-12 parent strain, whereas OP50 is a B-strain derivative. Theeffect of culture conditions was investigated in case of cultureage (1−4 days) and culture medium (lysogenic broth agar,blood agar base, brain−heart infusion agar, Trypticase soy agar,MacConkey agar). Data acquisition was performed using thelinear ion trap of the LTQ Orbitrap Discovery at unitresolution. Data was mean normalized due to better perform-ance compared to other normalization strategies applied in thisstudy. A reduction of data to the mass range of m/z = 600−900further improved identification performance. While nosignificant changes in spectra could be observed for differentculture ages between 1 and 4 days, small changes inphospholipid signal ratios were observed for different culturingmedia. However, performing supervised PCA−LDA classifica-tion algorithm leads to 87.3% accurate identification resultsusing leave-one-out cross-validation (see Figure 6). As thefigure shows, OP50 as only B-strain is clearly separated fromthe K-12 derivatives along the first principal component. TheRMMC algorithm used in all other examples was found to beinferior in the present case, only yielding 85.3% correct cross-validation results. This observation (together with thereduction of mass range) suggests that different identificationalgorithms used at different levels of identification (subspecieslevel vs higher taxonomical levels) might further improve theidentification performance.

Analysis of Pathogenic Yeasts. Identification of yeastsusing MALDI TOF MS requires the pretreatment of the yeastsample prior to mass spectrometric analysis in order to giveacceptable identification performance (score >2.0). While therecommended sample pretreatment for MALDI TOF MScomprises the complete extraction of the fungal material usingformic acid and acetonitrile,16 intact yeast species can directlybe analyzed by REIMS without any modification inexperimental setup or analysis workflow. Supporting Informa-tion Figure S-11 shows the PCA and RMMC plots for 87

Figure 4. Cross-validation results for the data set shown in Figure 3 asa function of the bin size on species, genus, and Gram-stain level.Numerical values can be found in Supporting Information Table S-5.

Figure 5. RMMC plot showing the combined data acquired on threedifferent instruments (Exactive, Orbitrap, and Xevo). Bin size 1 Da, m/z = 200−2000. Confusion matrices for cross-platform blindidentification results are given in Supporting Information Table S-7.

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626560

different bacterial species and 16 different fungal (yeast)species. No overlap between bacteria and fungi was observed inthe PCA data space. This was attributed to the markedlydifferent phospholipid composition between bacteria and fungi.Whereas REIMS spectra obtained from bacteria mostly featurephosphatidylglycerols, phosphatidylethanolamines, and low-abundance phosphatidic acids, the spectra obtained fromfungi mainly consist of high-abundance phosphatidic acids,phosphatidylethanolamines, and phosphatidylinositols, withparticularly the last group being very rare among bacteria.Figure 7 shows the separation of five clinically relevant

pathogenic Candida species both in supervised and unsuper-vised analysis. C. glabrata is clearly separated from the threeother species along the first principal component while C.lusitaniae, C. parapsilosis, C. tropicalis, and C. albicans are largelyseparated along the second principal component. Thisseparation into two groups is tentatively attributed to the factthat C. glabrata was found to belong to a separate phylogeneticclade than the other Candida species based on DNAsequences.45 Two misclassifications were observed whenperforming leave-one-out cross-validation, resulting in 98.8%correct classification.

■ CONCLUSION

The results shown in this study clearly demonstrate thatREIMS is a suitable method for the identification ofmicroorganisms in clinical microbiology settings. Both bacteriaand yeasts can be analyzed using the same experimental setupand workflow, which lacks the sample preparation step and thusallows analysis times as short as 3−5 s, using mass analyzerswith arbitrary resolution. The demonstrated subspeciesspecificity and the detection of various secondary metabolitesincluding rhamnolipids or lipid A suggest that the techniqueprovides information well beyond the taxonomical classificationof identified species pertaining to phenotypic factors includingvirulence, antibiotic resistance, serotype, or ribotype of themicroorganisms. The number of conserved and genus-specificspectral features gives strong implications on the feasibility ofculture-free applications by the direct analysis of bacterial cellswithin human biological fluids; however, these types ofapplications remain to be developed in the future.

■ ASSOCIATED CONTENT*S Supporting InformationDetailed information on sample sets analyzed in this study,further plots of multivariate statistical analysis of the presenteddata, detailed information about identified spectral features, andmass spectra resulting from different experimental setups andsettings. This material is available free of charge via the Internetat http://pubs.acs.org.

■ AUTHOR INFORMATIONCorresponding Author*Phone: +44 0-207 5942760. E-mail: z.takats@imperial.ac.uk.Author ContributionsStudy was planned by Z.T., N.S., and M.R. Experiments wereconducted by N.S., M.R., A.A., and V.B. Data was analyzed andinterpreted by N.S., E.A.J., O.G., Z.T., K.A.V., and J.B. Themanuscript was written through contributions of all authors. Allauthors have given approval to the final version of themanuscript.NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSThe work was funded by the European Research Council underthe Starting Grant scheme (contract no. 210356) and theEuropean Commission FP7 Intelligent Surgical Device project(contract no. 3054940). We acknowledge Medimass Ltd.

Figure 6. PCA-LDA model of seven different E. coli strains. Data wasmean normalized and reduced to m/z = 600−900. CV = 87.3%.

Figure 7. PCA (A) and RMMC (B) plot of C. albicans (n = 20), C.glabrata (n = 19), C. lusitaniae (n = 12), C. parapsilosis (n = 19), and C.tropicalis (n = 16). CV = 98.8%.

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626561

(Budapest, Hungary) for the technical support. K.A.V.acknowledges his Imperial College Junior Research Fellowshipfunding.

■ REFERENCES(1) Havlicek, V.; Lemr, K.; Schug, K. A. Anal. Chem. 2013, 85, 790−797.(2) Klouche, M.; Schroder, U. Clin. Chem. Lab. Med. 2008, 46, 888−908.(3) Janda, J. M.; Abbott, S. L. J. Clin. Microbiol. 2007, 45, 2761−2764.(4) Woo, P. C. Y.; Lau, S. K. P.; Teng, J. L. L.; Tse, H.; Yuen, K. Y.Clin. Microbiol. Infect. 2008, 14, 908−934.(5) Anhalt, J. P.; Fenselau, C. Anal. Chem. 1975, 47, 219−225.(6) Meuzelaar, H. L. C.; Kistemaker, P. G. Anal. Chem. 1973, 45,587−590.(7) Schulten, H. R.; Beckey, H. D.; Meuzelaar, H. L. C.; Boerboom,A. J. Anal. Chem. 1973, 45, 191−195.(8) Heller, D. N.; Cotter, R. J.; Fenselau, C. Anal. Chem. 1987, 59,2806−2809.(9) Claydon, M. A.; Davey, S. N.; Edwards-Jones, V.; Gordon, D. B.Nat. Biotechnol. 1996, 14, 1584−1586.(10) Krause, E.; Wenschuh, H.; Jungblut, P. R. Anal. Chem. 1999, 71,4160−4165.(11) Ryzhov, V.; Fenselau, C. Anal. Chem. 2001, 73, 746−750.(12) Bizzini, A.; Durussel, C.; Bille, J.; Greub, G.; Prod’hom, G. J.Clin. Microbiol. 2010, 48, 1549−1554.(13) Cherkaoui, A.; Hibbs, J.; Emonet, S.; Tangomo, M.; Girard, M.;Francois, P.; Schrenzel, J. J. Clin. Microbiol. 2010, 48, 1169−1175.(14) van Veen, S. Q.; Claas, E. C. J.; Kuijper, E. J. J. Clin. Microbiol.2010, 48, 900−907.(15) Tan, K. E.; Ellis, B. C.; Lee, R.; Stamper, P. D.; Zhang, S. X.;Carroll, K. C. J. Clin. Microbiol. 2012, 50, 3301−3308.(16) Cassagne, C.; Cella, A. L.; Suchon, P.; Normand, A. C.; Ranque,S.; Piarroux, R. Med. Mycol. 2013, 51, 371−377.(17) Seng, P.; Drancourt, M.; Gouriet, F.; La Scola, B.; Fournier, P.-E.; Rolain, J. M.; Raoult, D. Clin. Infect. Dis. 2009, 49, 543−551.(18) Schmidt, F.; Fiege, T.; Hustoft, H. K.; Kneist, S.; Thiede, B.Proteomics 2009, 9, 1994−2003.(19) Heller, D. N.; Fenselau, C.; Cotter, R. J.; Demirev, P.; Olthoff, J.K.; Honovich, J.; Uy, M.; Tanaka, T.; Kishimoto, Y. Biochem. Biophys.Res. Commun. 1987, 142, 194−199.(20) Goodacre, R.; Heald, J. K.; Kell, D. B. FEMS Microbiol. Lett.1999, 176, 17−24.(21) Smith, P. B. W.; Snyder, A. P.; Harden, C. S. Anal. Chem. 1995,67, 1824−1830.(22) Vaidyanathan, S.; Kell, D. B.; Goodacre, R. J. Am. Soc. MassSpectrom. 2002, 13, 118−128.(23) Ishida, Y.; Madonna, A. J.; Rees, J. C.; Meetani, M. A.; Voorhees,K. J. Rapid Commun. Mass Spectrom. 2002, 16, 1877−1882.(24) Shu, X.; Li, Y.; Liang, M.; Yang, B.; Liu, C.; Wang, Y.; Shu, J. Int.J. Mass Spectrom. 2012, 321−322, 71−76.(25) Song, Y.; Talaty, N.; Tao, W. A.; Pan, Z.; Cooks, R. G. Chem.Commun. 2007, 61−63.(26) Zhang, J. I.; Talaty, N.; Costa, A. B.; Xia, Y.; Tao, W. A.; Bell, R.;Callahan, J. H.; Cooks, R. G. Int. J. Mass Spectrom. 2011, 301, 37−44.(27) Song, Y.; Talaty, N.; Datsenko, K.; Wanner, B. L.; Cooks, R. G.Analyst 2009, 134, 838−841.(28) Watrous, J.; Roach, P.; Alexandrov, T.; Heath, B. S.; Yang, J. Y.;Kersten, R. D.; van der Voort, M.; Pogliano, K.; Gross, H.;Raaijmakers, J. M.; Moore, B. S.; Laskin, J.; Bandeira, N.;Dorrestein, P. C. Proc. Natl. Acad. Sci. U. S. A. 2012, 109, E1743−E1752.(29) Hsu, C.-C.; ElNaggar, M. S.; Peng, Y.; Fang, J.; Sanchez, L. M.;Mascuch, S. J.; Møller, K. A.; Alazzeh, E. K.; Pikula, J.; Quinn, R. A.;Zeng, Y.; Wolfe, B. E.; Dutton, R. J.; Gerwick, L.; Zhang, L.; Liu, X.;Mansson, M.; Dorrestein, P. C. Anal. Chem. 2013, 85, 7014−7018.

(30) Schaefer, K.-C.; Denes, J.; Albrecht, K.; Szaniszlo, T.; Balog, J.;Skoumal, R.; Katona, M.; Toth, M.; Balogh, L.; Takats, Z. Angew.Chem., Int. Ed. 2009, 48, 8240−8242.(31) Strittmatter, N.; Jones, E. A.; Veselkov, K. A.; Rebec, M.; Bundy,J. G.; Takats, Z. Chem. Commun. 2013, 49, 6188−6190.(32) Race, A. M.; Styles, I. B.; Bunch, J. J. Proteomics 2012, 75, 5111−5112.(33) Veselkov, K. A.; Lindon, J. C.; Ebbels, T. M. D.; Crockford, D.;Volynkin, V. V.; Holmes, E.; Davies, D. B.; Nicholson, J. K. Anal.Chem. 2008, 81, 56−66.(34) Veselkov, K. A.; Vingara, L. K.; Masson, P.; Robinette, S. L.;Want, E.; Li, J. V.; Barton, R. H.; Boursier-Neyret, C.; Walther, B.;Ebbels, T. M.; Pelczer, I.; Holmes, E.; Lindon, J. C.; Nicholson, J. K.Anal. Chem. 2011, 83, 5864−5872.(35) Veselkov, K. A.; Mirnezami, R.; Strittmatter, N.; Goldin, R. D.;Kinross, J.; Speller, A. V. M.; Abramov, T.; Jones, E. A.; Darzi, A.;Holmes, E.; Nicholson, J. K.; Takats, Z. Proc. Natl. Acad. Sci. U. S. A.2014, 111, 1216−1221.(36) Abdel-Mawgoud, A.; Lepine, F.; Deziel, E. Appl. Microbiol.Biotechnol. 2010, 86, 1323−1336.(37) Lepine, F.; Milot, S.; Deziel, E.; He, J.; Rahme, L. G. J. Am. Soc.Mass Spectrom. 2004, 15, 862−869.(38) Miyagawa, E.; Azuma, R.; Suto, T.; Yano, I. J. Biochem. 1979, 86,311−320.(39) Welby-Gieusse, M.; Laneelle, M. A.; Asselineau, J. Eur. J.Biochem. 1970, 13, 164−167.(40) Williamson, D. H.; Wilkinson, J. F. J. Gen. Microbiol. 1958, 19,198−209.(41) Zhu, C.; Nomura, C. T.; Perrotta, J. A.; Stipanovic, A. J.; Nakas,J. P. Biotechnol. Prog. 2010, 26, 424−430.(42) Guenther, S.; Schafer, K.-C.; Balog, J.; Denes, J.; Majoros, T.;Albrecht, K.; Toth, M.; Spengler, B.; Takats, Z. J. Am. Soc. MassSpectrom. 2011, 22, 2082−2089.(43) Lechevalier, M. P.; Moss, C. W. Crit. Rev. Microbiol. 1977, 5,109−210.(44) Yabuuchi, E.; Kosako, Y.; Oyaizu, H.; Yano, I.; Hotta, H.;Hashimoto, Y.; Ezaki, T.; Arakawa, M. Microbiol. Immunol. 1992, 36,1251−1275.(45) Donoho, D. L.; Johnstone, I. M. C. R. Seances Acad. Sci., Ser. A1994, 319, 1317−1322.

Analytical Chemistry Article

dx.doi.org/10.1021/ac501075f | Anal. Chem. 2014, 86, 6555−65626562