Applications of MALDI-TOF mass spectrometry in clinical diagnostic microbiology

Authors


Correspondence: Gilbert Greub, Institute of Microbiology, University Hospital Center and University of Lausanne, 1011 Lausanne, Switzerland. Tel.: +41 21 314 49 79; fax: +41 21 314 40 60; e-mail: gilbert.greub@chuv.ch

Abstract

Until recently, microbial identification in clinical diagnostic laboratories has mainly relied on conventional phenotypic and gene sequencing identification techniques. The development of matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOFMS) devices has revolutionized the routine identification of microorganisms in clinical microbiology laboratories by introducing an easy, rapid, high throughput, low-cost, and efficient identification technique. This technology has been adapted to the constraint of clinical diagnostic laboratories and has the potential to replace and/or complement conventional identification techniques for both bacterial and fungal strains. Using standardized procedures, the resolution of MALDI-TOFMS allows accurate identification at the species level of most Gram-positive and Gram-negative bacterial strains with the exception of a few difficult strains that require more attention and further development of the method. Similarly, the routine identification by MALDI-TOFMS of yeast isolates is reliable and much quicker than conventional techniques. Recent studies have shown that MALDI-TOFMS has also the potential to accurately identify filamentous fungi and dermatophytes, providing that specific standardized procedures are established for these microorganisms. Moreover, MALDI-TOFMS has been used successfully for microbial typing and identification at the subspecies level, demonstrating that this technology is a potential efficient tool for epidemiological studies and for taxonomical classification.

Introduction

The matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) is a rapid, accurate, and cost-effective method of microbial characterization and identification. This technology generates characteristic mass spectral fingerprints, that are unique signatures for each microorganism and are thus ideal for an accurate microbial identification at the genus and species levels and has a potential to be used for strain typing and identification.

MS has been used for several decades in chemistry but it was in 1975 that Anhalt and Fenselau (Anhalt & Fenselau, 1975) proposed for the first time that bacterial characterization could be achieved using this technique. They noticed that unique mass spectra were produced from bacterial extracts of different genera and of different species. In the 1980s, the development of desorption/ionization techniques such as plasma desorption (PD), laser desorption (LD) and fast atom bombardment (FAB) allowed the generation of molecular biomarker ions from microorganisms leading to bacterial profiling (Heller et al., 1987; Platt et al., 1988). In early experiments, only biomarker molecules of low molecular masses such as bacterial lipids were analysed (Shah & Collins, 1980; Heller et al., 1988), as the processes used for the ionization of biomolecules were too energy rich to avoid unpredictable analyte decomposition. The evolution of soft ionization techniques such as MALDI and electrospray ionization (ESI) developed in the late 1980s (Tanaka & Fenn, 2002) made possible analysis by MS of large biomolecules such as intact proteins. Several groups (Cain et al., 1994; Girault et al., 1996; Liang et al., 1996) demonstrated that MALDI-TOF could be used to produce protein profiles following cellular extraction and purification. However, Holland et al. (1996) reported for the first time in 1996 that MALDI-TOF spectral fingerprints could be obtained from whole bacterial cells without pretreatment before the MS analysis. This approach was then used to identify bacteria at the genus and species levels by multiple research teams (Claydon et al., 1996; Krishnamurthy & Ross, 1996; Haag et al., 1998; Pribil & Fenselau, 2005; Pignone et al., 2006; Vargha et al., 2006; Nagy et al., 2009). Since the late 1990s, the success of this technique for the rapid identification of bacteria but also fungi and yeast was demonstrated by an impressive exponential increase in the number of publications concerning MALDI-TOF identification of microorganisms.

MALDI-TOF MS has been used to characterize a wide variety of microorganisms including bacteria, fungi, and viruses (Giebel et al., 2010). The capability of MALDI-TOF to rapidly characterize microorganisms favors its potential applications in multiple areas including medical diagnostics, biodefense, environmental monitoring, and food quality control. MALDI-TOF MS is suitable for high-throughput and rapid microbial identification at low costs and is an alternative for conventional laboratory biochemical and molecular identification systems.

In this review, we will first present the technical background of the MALDI-TOF MS method including sample preparation, MS, bioinformatics analysis of fingerprint spectra, and their comparison with databases for microorganism identification. We will then present the application of MALDI-TOF MS microbial identification from bacteria and fungi isolates and directly from clinical samples. We will then address the important issues of quality control, maintenance, time-to results, and cost effectiveness. Finally, we will discuss the use of MALDI-TOF MS for additional applications such as microbial taxonomy, typing and identification of virulence factors (Bizzini & Greub, 2010; Murray, 2010).

Technique

A mass spectrometer is composed of three functional units, (1) an ion source to ionize and transfer sample molecules ions into a gas phase, (2) a mass analyser that separate ions according to their mass-to-charge ratio (m/z), and (3) a detection device to monitor separated ions. Several ionization methods have been developed including PD, FAB, chemical ionization (CI), atmospheric pressure CI, electrospray (ESI), LD, and MALDI. The method of ionization is determined according to the nature of the sample and the goal of the MS analysis, but ESI and MALDI are soft ionization techniques that allow ionization and vaporization of large nonvolatile biomolecules such as intact proteins (Emonet et al., 2010). In contrast to ESI, MALDI generates mostly singly charged ions and thus MALDI derived spectra may include larger numbers of proteins. LD has been successfully coupled to several kinds of mass analysers to characterize microorganisms such as TOF (Lay, 2001), Fourier transform ion cyclotron resonance (Ho & Fenselau, 1998), quadrupole-TOF (She et al., 2001), and quadrupole ion trap (Meetani et al., 2007). The various operative modes of the different mass analysers confer strengths and weaknesses in their performance characterized by mass accuracy, resolution, mass range, sensitivity, and scan speed and cost (Table S1) (Jonsson, 2001; Aebersold & Mann, 2003; Domon & Aebersold, 2006; Graham et al., 2007). The performance of mass analysers can be improved by combining and/or summing the advantages of one sort of analyser (tandem MS) or of different analysers (hybrid MS) through the development of multistage instruments such as hybrid quadrupole TOF (Q-Q-TOF), tandem TOF (TOF-TOF), and triple quadrupole (Domon & Aebersold, 2006). Overall, the required performance expected form a mass analyser depends on the type of sample to be analysed (complex/simple mixtures, proteins, peptides, lipids, and polysaccharides) and the ultimate goal of the analysis (quantification, protein identification, microorganism identification, and biotyping).

TOF mass analysers (Cotter, 1997) have been used for intact microorganisms detection for many years (Heller et al., 1987), because they are suited for interfacing with pulse laser ionization and offer the possibility of rapid analysis and miniaturization.

In MALDI analysis, samples are prepared by mixing the samples with a matrix which results in the crystallization of the sample within the matrix. The matrix is composed of small acid molecules that have a strong optical absorption in the range of the laser wavelength used. The matrix composition varies according to the biomolecule to be analysed and the type of laser used (Fenselau & Demirev, 2001). The most frequently used matrices are 2,5-dihydroxybenzoic acid (DHB), α-cyano-4-hydroxycinnamic acid (CHCA), sinapinic acid (SA), ferulic acid (FA), and 2,4-hydroxy-phenyl benzoic acid. FA, SA, and CHCA have been shown to be effective for the detection of proteins biomarkers (Fenselau & Demirev, 2001; Vaidyanathan et al., 2002; Williams et al., 2003), whereas DHB appears to be the best choice for the detection of glycopeptides and glycoproteins (Giebel et al., 2010). The size and the intensities of the peaks of the detected molecules are dependent on the matrix selected for the experiment. DHB and CHCA are usually optimal for the detection of lower mass ions (Hathout et al., 2000; Williams et al., 2003; Ruelle et al., 2004) with a detection up to 10 kDa when the proper solvent is used. Both SA and FA have been shown to be better for the detection of higher mass ions (above 15 kDa) (Madonna et al., 2000; Conway et al., 2001; Ruelle et al., 2004; Vargha et al., 2006) but provide a lower sensitivity than CHCA (Wang et al., 1998; Ruelle et al., 2004).

Intact microorganisms can be directly processed to MALDI-TOF without pretreatment because most vegetative bacteria are lysed following exposure to water, organic solvent and/or strong acid in the MALDI matrix. When resistant microrganims such as some viruses, bacterial spores, and yeast cells have to be analysed by MALDI, strong organic acids and/or alcohols are usually added in pretreatment steps. Similarly, for some bacterial species (such as the Actinomyces), specific pretreatment or protein extraction procedures may be useful (Bizzini et al., 2011).

The marked differences in bioanalytes fingerprints observed with the different matrix demonstrate that similar standardized preanalytical and analytical procedures than those performed to establish a database must be followed to ensure accurate identification. This implies that any modification of the procedure (1) should conformed to the manufacturer's recommendations, (2) should be analysed to demonstrate that protein profiles remained consistent with database fingerprints or (3) should use a new database created with the modified protocol.

In practice, a microbial sample is mixed with a matrix on a conductive metal plate. The mixture can be deposited on the metal support or alternatively the microbial sample is deposited and dried out on the support before the addition of the matrix. After the crystallization of the matrix and compound, the target on the metal plate is introduced in the mass spectrometer where it is bombarded with brief laser pulses from usually a nitrogen laser (Fig. 1). The matrix absorbs energy from the laser leading to the desorption of the analytes that are then vaporized and ionized in the gas phase. This matrix-assisted desorption and ionization of the analytes leads to the formation of predominantly singly charged sample ions. The desorbed and ionized molecules are first accelerated through an electrostatic field and are then ejected through a metal flight tube that is subjected to a vacuum until they reach a detector, with smaller ions traveling faster than larger ions. The TOF required to reach the detector is dependent on the mass (m) and charge (z) of the bioanalyte and is proportional to the square root of m/z. Thus, bioanalytes with different m/z that composed a complex sample are separated according to their TOF and create a mass spectrum that is characterized by both the m/z and the intensity of the ions, which is the number of ions of a particular m/z that struck the detector. The results of a spectral signature is composed of spikes ranging usually from 1000 to 20 000 m/z. Usually, MALDI produces single charged (z = 1) ions and thus the m/z of an analyte corresponds to the value of its mass.

Figure 1.

Technical description of MALDI-TOFMS. The sample is mixed with a matrix on a conductive metal plate. After crystallization of the matrix and microbial material, the metal plate is introduced in the mass spectrometer and is bombarded with brief laser pulses. The desorbed and ionized molecules are accelerated through an electrostatic field and ejected through a metal flight tube subjected to vacuum until they reach a detector, with smaller ions traveling faster than larger ions. Thus, bioanalytes separated according to their TOF create a mass spectrum that is composed by mass to charge ratio (m/z) peaks with varying intensities. A spectrum is thus a microbial signature that is compared with a database for the identification at the species or genus level.

Data analysis

Biomarkers

Several research groups have demonstrated that biomolecules desorbed from whole unfractionated cells and detected above 4 kDa are intact proteins (Arnold et al., 1999; Dai et al., 1999; Holland et al., 1999; Ryzhov & Fenselau, 2001). Most of the biomarkers detected in MALDI-TOF spectra of intact bacterial cells have a molecular mass below 15 kDa. A thorough characterization of MALDI biomarkers performed on intact Escherichia coli cells have demonstrated that the MALDI detected biomolecules corresponds to proteins from the inside of bacterial cell that are abundant, basic, and of medium hydrophobicity (Ryzhov & Fenselau, 2001). Among these MALDI-detected proteins, about half matched ribosomal proteins that are abundant and very basic (Arnold et al., 1999; Ryzhov & Fenselau, 2001), which is a biochemical trait favorable for efficient ionization during the MALDI process (Krause et al., 1999). Thus, the abundance (more than 20% of total cell proteins) and the basic nature of ribosomal proteins explain why the majority of the peaks detected in a MALDI-TOF spectrum correspond to ribosomal proteins. In addition, the lysis of bacterial cells in organic solvents and in acidic conditions favoring the extraction of ribosomal proteins combined with the utilization of a specific matrix allowed the development of a method leading to the ionization of mainly ribosomal proteins (Suh & Limbach, 2004).

In a study performed by Ryzhov & Fenselau (2001) on E. coli to characterize the nature of the proteins favored by MALDI, several additional groups of proteins than ribosomal proteins were identified. These also included abundant nucleic acid-binding proteins, such as E. coli DNA-binding protein HU α- and β-subunits (DbhA and DbhB, respectively) and cold-shock proteins, such as cold-shock proteins A, C, and E (CspA, CspC, and CspE). Similar to ribosomal proteins, these protein families are highly abundant, basic, and of medium hydrophobicity. Holland et al. (1999) potentially identified the acid-resistant precursor proteins HdeA and HdeB observed in the MALDI analysis of both intact E. coli and Shigella flexneri. The ion at m/z 7643 in the spectra from Pseudomonas aeruginosa was mapped to the cold-shock protein CspA and similarly the ion at m/z 7684 observed in Pseudomonas putida was identified as the cold-acclimation protein CapB (Fenselau & Demirev, 2001). Sun et al. 2006 have selected in the MALDI-TOF spectra of Lactobacillus plantarum 34 reliable biomarkers including 31 ribosomal subunit proteins and three ribosome-associated proteins identified as a small heat shock protein, a methylase and the DNA-binding protein II. A comprehensive study was published by Dieckmann et al. 2008 where protein identities were assigned to biomarker peaks obtained by whole-cell MALDI-TOF MS of salmonellae. Most of the proteins identified in this study were abundant cytosolic proteins that were highly basic including in particular ribosomal proteins, proteins involved in DNA or RNA binding, and other abundant proteins, most of which have a high isoelectric point > 9. Thus, examples of biomarker peaks assigned to proteins characterized by a high isoelectric point included not only ribosomal proteins but also many other proteins such as cold shock-like protein CspH, translation initiation factor IF-1 (pI 9.23), DNA binding protein HU α- and β- (pI 9.69 and 9.57), the ribosome modulation factor (pI 10.56), and integration host factors A and B (both pI 9.34). Lower pI value proteins detected in this study were in general very abundant proteins including the nucleoid-associated protein H-NS (pI 5.32), the RNA chaperone CspE (pI 8.08), glutaredoxin-1 (pI 5.63), and the phosphocarrier protein HPr (pI 5.6).

Thus, most of MALDI-TOF spectra are composed of very conserved proteins with house-keeping functions affected to a minimal extent by environmental conditions and thus considered to be optimal for routine identification of bacteria.

Biomarkers reproducibility

The reproducibility of MALDI-TOF spectrum of whole bacterial sample is problematic as large variations can be seen in spectra of the same bacterial species obtained in different conditions. Many experimental parameters can have an important effect on the observed mass spectra (Wang et al., 1998). The reproducibility is dependent on the MALDI-TOF instrument, the matrix used, the age of the microorganism, the sample : matrix ratio, the sample concentration, the culture medium and growth conditions (Valentine et al., 2005). However, several studies have shown that a subset of peaks from genetically identical bacteria was conserved in spectra obtained in different experimental conditions (Wang et al., 1998; Welham et al., 1998). These conserved peaks, among which ribosomal proteins are well represented, explain the feasibility to use MALDI-TOF for bacterial identification even without the standardization of experimental conditions. These results also suggest that selected specific conserved biomarker proteins could be used for bacterial identification irrespective of changes in other biomarkers. However, to optimize the reproducibility, a standardization of sample preparation (e.g. choice of matrix, concentrations, solvent and crystallization conditions) has to be established by diagnostic laboratories.

Intra-laboratory reproducibility

Only a few studies have reported investigations on intra-laboratory reproducibility. In two studies that were focusing only on the presence or absence of particular peaks but not on peak intensities, the level or reproducibility obtained was equal to 75% or higher using the same mass spectrometer and similar sample preparation techniques (Saenz et al., 1999; Walker et al., 2002). The reproducibility of MALDI-TOF MS fingerprints was high even when mass spectra were obtained from bacterial cultures propagated in the same standard culture conditions during weeks and even months (Bernardo et al., 2002).

In our laboratory, we observed that the reproducibility is mainly dependent on the quality of the deposit and that the extraction step is not associated with significant supplementary variability. Thus, when considering the higher peaks (≥ 200 intensity units) obtained with E. coli strain ATCC 25922 in 10 independent analyses, 66 peaks were present in all 10 replicates performed with proteins obtained from the same extraction whereas 78 peaks were present in the 10 spectra obtained from 10 independent extractions (Fig. 2a). Similar results were obtained with Staphylococcus aureus strain ATCC 25923 (Fig. 2b).

Figure 2.

Intra-laboratory reproducibility tested by measuring the number of conserved peaks in two experimental settings. In the first setting, ethanol/formic acid extraction was applied on Escherichia coliATCC 25922 (a) or Staphylococcus aureusATCC 25923 (b) and 10 replicates of one extraction were spotted onto the MALDI-TOF microplate (1 μL, about 106 bacteria μL−1). In the second setting, 10 independent extractions were performed and one replicate of each extraction was spotted onto the MALDI-TOF microplate. After smoothing and baseline subtraction, the 100 highest peaks from each spot were selected and the number of time a given peak is present out of 10 times was determined for each peak in the two different settings for Escherichia coli and Staphylococcus aureus.

Inter-laboratory reproducibility

The variation of inter-laboratory reproducibility is not surprisingly much higher as small variation in sample preparation and analysis may significantly affect mass fingerprints. Unfortunately, only a few attempts have been made to compare results from different laboratories on the same organism (Mellmann et al., 2009). In addition, inter-laboratory studies based on different comparative settings have been performed and gave very different results. Studies comparing mass spectra of identical microorganisms obtained in different locations but with the same experimental protocols and instruments hardware and software have shown promising results in terms of inter-laboratory reproducibility. For instance, two studies using different bacterial species have demonstrated that more than 60% of the peaks observed in mass spectra were similar in separate laboratories (Wang et al., 1998; Walker et al., 2002). However, a poor inter-laboratory reproducibility of MALDI-TOF MS of intact microorganisms was demonstrated in a study where three independent laboratories using three different commercial instruments performed a MALDI-TOF MS analysis of identical aliquots of E. coli culture prepared and analysed in the same experimental conditions (Wunschel et al., 2005). In this study, only 25% of the biomarkers were found in common by all three laboratories and more than 50% of the peaks were detected in spectra from only one of the three laboratory. Of importance, when the mass spectra collected from the instrument of one of the laboratory was analysed in the other two laboratories, 70% of the mass fingerprints could be identified correctly as E. coli. This finding underlines the importance of the instrument in the establishment of bacterial fingerprint databases.

Application for microorganism identifications

Two general MALDI-TOF MS methods have been proposed to characterize microorganisms: (1) mass spectra comparison with fingerprints database and (2) matching of biomarker masses to a proteome database (see Box S1). In the first approach, generated unique spectra of intact cells are compared with previously collected fingerprint libraries that are commercially available. This solution is rapid, simple, and is easily adaptable for routine use in diagnostic laboratories. This approach is convenient to develop specific databases constituted of unique and conserved peaks that can be used for species and subspecies identification, independently of the culture conditions used to grow the microorganism (Carbonnelle et al., 2007). In the second approach, the biomarker masses associated with an unknown microorganism are identified by matching protein molecular masses in the spectrum with protein molecular masses predicted from sequenced genomes (Demirev et al., 1999). This method is based on the observation that the majority of observed biomolecules above 4000 m/z in MALDI-TOF spectra of whole cell extracts are proteins. An available algorithm predicts protein masses in silico from the genomes and seeks matches with experimentally derived masses (Pineda et al., 2000). However, this application is limited to microorganisms whose genomes are sequenced and further development in strategies for organization of the proteome database is required. The advantage of such a bioinformatics-based approach compared with bacterial fingerprinting is that the identification tolerates variations in the protein profiles and thus differences in culture growth and sample treatment conditions.

Bacterial identification

The identification of microorganisms in clinical diagnostic microbiology laboratories is nowadays mainly performed by analysis of biochemical reactions and phenotypic characteristics, such as growth on different media, colony morphology and Gram staining. When combined, these routine laboratory techniques ensure an accurate identification of most microorganisms but are costly, require time, and need in some cases well-trained technician for correct interpretation. One of the major advantages of using MALDI-TOF technology for bacterial identification is the time-to-result, which is reduced from 24 to 48 h to less than an hour. In addition, MALDI-TOF allowed accurate bacterial identification of a large variety of bacteria, that only exhibit few phenotypic traits and that were identified by 16S rRNA gene sequencing prior to the MALDI-TOF era (Bizzini et al., 2011).

Routine identification

Several approaches are commonly used in routine diagnostic laboratories. One typical approach is to pick bacterial isolates colonies freshly grown on defined agar medium with a sterile tip and to smear a thin film onto a ground steel MALDI target plate. The microbial film is then overlaid with a MALDI matrix selected as recommended by the MALDI-TOF manufacturers, typically 1.5 μL CHCA in 50% acetonitrile/2.5% trifluoroacetic acid for the Bruker instrument and 0.5 μL of 20 mg DHB dissolved in 1 mL water-ethanol-acetonitrile [1 : 1 : 1] mix for the Shimadzu instrument. In our laboratory, to increase identification yield, we routinely add formic acid on smeared microorganisms before adding the matrix. The sample-matrix mixture is dried at room temperature and then introduced in the MALDI-TOF instrument for data acquisition. The data are processed by the associated softwares (see above) and the spectra are compared with reference libraries for bacterial identification. A protein extraction step using ethanol-acetonitrile is performed when direct application procedure failed.

Several studies have analysed the bacterial identification efficiency of these two instruments (Bruker and Shimadzu) with their respective software and databases (Table 1). We will present here the results of (1) two intra-laboratory studies using the Bruker system (Seng et al., 2009; van Veen et al., 2010), (2) an intra-laboratory study evaluating the two instruments using the same microbial samples (Cherkaoui et al., 2010), and (3) a large international interlaboratory study performed with the Bruker instrument (Mellmann et al., 2009).

Table 1. Performance of routine identification by MALDI-TOFMS
ReferencesNDescriptionGenus level (%)Species level (%)Mis-identification (%)No ID (%)Comments
  1. Genus and species level (%), percent of identification at the genus and species level respectively. Misidentification (%), percent of misidentified samples. No ID (%), percent of samples not identified. *Bruker system, Shimatzu.

  2. N, sample number; GP, Gram-positive; GN, Gram-negative.

van Veen et al. (2010)327Total95.185.633.9Retrospective intra-laboratory study
Misidentification: associated with absence or mislabeling of reference spectra in the database
89Enterobacteriaceae10096.63.40
55Non-fermentative GN81.874.55.414.5
87GP cocci97.780.51.12.3
77Miscellaneous bacteria (HACCEK)94.884.43.91.3
19Yeasts10094.700
van Veen et al. (2010)980Total98.8921.71.1Prospective intra-laboratory study
Misidentification: associated with absence or mislabeling of reference spectra in the database
Viridans streptococci and pneumococci characterized by many misidentifications
Weakness of the study: lack of sufficient anaerobic bacteria and GP aerobic rods included in the tested samples
311Enterobacteriaceae10097.70.30
88Non-fermentative GN94.3921.14.6
261GP cocci in cluster10094.30.40
165GP cocci in chains98.884.87.31.2
 Enterococci10098.400
 Hemolytic streptococci10010000
 Streptococcus milleri group10077.800
 Pneumococci10086.400
 Viridans streptococci90.59.557.19.5
94Miscellaneous bacteria (HACCEK)96.88403.2
61Yeasts96.785.23.33.2
Seng et al. (2009)1660Total95.484.11.72.8Prospective intra-laboratory study
Lack of identification mainly for non-Clostridium anaerobes due to an absence of reference spectra in the database
NDGP  0.82
NDGN  0.90.7
Cherkaoui et al. (2010)720Total10094*/890.9*/0.5 Comparative intra-laboratory study of two commercial MALDI-TOF MS devices (Bruker and Shimatzu)

Poor yield for streptococcal species and for GN anaerobes (< 50% correct identification)
416Enterobacteriaceae 99.8*/95.90*/0 
80Aerobic GN 97.5*/96.31.25*/0 
111Staphylococci 98.2*/96.40.9*/0.9 
87Aerobic GP 73.6*/55.23.4*/2.3 
6Anaerobic GN 17*/00*/0 
7Anaerobic GP 57*/4325*/0 
Mellmann et al. (2009)480Total 98.71.040.2Interlaboratory study
60 non-fermenting bacteria were shipped to 8 different laboratories with access to Bruker platforms
No significant difference between direct application and preprocessed samples
Misidentification: sample interchange (4) and skin contamination (1)
60Laboratory A 10000
60Laboratory B 96.671.71.7
60Laboratory C 10000
60Laboratory D 93.336.660
60Laboratory E 10000
60Laboratory F 10000
60Laboratory G 10000
60Laboratory H 10000
Bizzini et al. (2010)1371Total direct application 70.3  Intra-laboratory study
Protein extraction increases the total yield of valid results by 25% compared with direct application
Misidentification: inaccurate taxonomic assignment, change in the taxonomy, limit of resolution of the method
1371Total protein extraction95.492.64.24.6
525GP direct application 73.7  
525GP protein extraction99.698.850.950.4
729GN direct application 71.6  
729GN protein extraction9792.273
24Yeasts direct application 4.1  
24Yeasts protein extraction10010000
Marklein et al. (2009)267Total before complementation of database 92.5 7.5Intra-laboratory study of clinical yeast isolates [Candida (n = 250), Cryptococcus, Saccharomyces, Trichosporon, Geotrichum, Pichia and Blastoschizomyces spp.]
All isolates identified upon complementation of the database with appropriate reference strains
All samples were preprocessed with a protein extraction step before deposition
267Total after complementation of the database 100 0
Bader et al. (2010)1192Total 97.6*/96.10.7*/0.21.7*/3.7Comparative intra-laboratory study of two commercial MALDI-TOF MS devices (Bruker and Shimatzu) on clinical yeast isolates

Better yield observed when the performance is only tested on species present in respective databases
All sample were preprocessed with a protein extraction step before deposition
1175*/1152Challenged against respective database 99*/99.40.5*/00.5*/0.6

Using the Bruker system, 327 clinical isolates previously identified by conventional techniques were analysed using MALDI-TOF MS by van Veen et al. 2010. The authors observed a 95.1% correct identification at the genus level and a 85.6% at the species level. The same group performed a prospective validation study on 980 clinical isolates of bacteria and yeast that showed a 92.2% correct identification by MALDI-TOF, a performance significantly better than the 83.1% identification obtained with conventional biochemical systems. Correct species identification by MADLI-TOF MS was observed in 97.7% of Enterobacteriaceae, 92% of nonfermentative Gram-negative bacteria, 94.3% of staphylococci, 84.8% of streptococci, 84% of bacteria of the HACCEK group (Haemophilus, Actinobacillus, Cardiobacterium, Capnocytophaga, Eikenella, and Kingella) and 85.2% of yeasts. In this work, misidentification was clearly associated with insufficient spectra from suitable reference strains in the reference spectra database.

Seng et al. (2009) have conducted a prospective routine MALDI-TOF MS identification analysis with the Bruker system on 1660 bacterial isolates in parallel with conventional phenotypic bacterial identification. Discrepancies were resolved by 16S rRNA and rpoB gene sequencing. They have shown that 95.4% of the isolates were correctly identified using MALDI-TOF MS among which 84.1% at the species level and 11.3% at the genus level only. Absence of identification (2.8% of isolates) and incorrect identification (1.7% of isolates) were mainly due to improper database entries. They have estimated that the MALDI-TOF identification required an average time of 6 min for an estimated 70–80% reduced cost compared with conventional methods of identification.

Cherkaoui et al. 2010) have evaluated the two main MALDI-TOF MS systems, Bruker and Shimatzu, in a comparative study with 720 bacterial isolates under routine clinical laboratory conditions. The isolates were analysed in parallel on both devices according to the manufacturers’ default recommendations. The MALDI-TOF MS results were compared with conventional biochemical identification tests and discordant results were resolved with 16S rRNA gene sequencing. The Bruker MS system gave high-confidence identification for 680 of 720 isolates (94.4%) whereas the Shimadzu MS showed a high-confidence identification for 639 isolates (88.8%). These results showed also that only 6/680 (0.9%) of the Bruker and 3/639 (0.5%) of the Shimadzu identifications gave an incorrect high-confidence identification at the species level. All the high-confidence MS identifications were accurate at the genus level. In addition, the Bruker MS system has identified 9 (69%) and the Shimadzu system 5 (38%) of 13 isolates that were not identified by conventional phenotyping methods.

Interlaboratory species identification was assessed in a large international multicenter study (Mellmann et al., 2009) using the Bruker system. In this study, eight participating laboratories received 60 blind-coded samples for MALDI-TOF MS species identification and 98.75% were correctly identified at the species level. Six of the eight laboratories identified all 60 samples correctly. Out of a total of 480 samples, six samples were misidentified and one sample did not give any valid result due to low signal intensity. Thus, in contrast to other studies (Wang et al., 1998), the utilization of a commercial system developed for routine use provided high reliability for bacterial identification.

Although excellent results have been shown for bacterial identification by MALDI-TOF MS, both the diagnostic yield (identification at the genus, species or strain level) and the accuracy of identification depends on the taxonomy and in the quality of the databases. As shown by many studies, most of the bacterial groups including Enterobacteriaceae, nonfermentative Gram-negative bacteria, staphylococci, and hemolytic streptococci were correctly identified as the species level (Seng et al., 2009; Cherkaoui et al., 2010; van Veen et al., 2010). Regarding staphylococci, the MALDI-TOF MS has brought up an important advantage by allowing rapid and simplified identification of both S. aureus and of some species belonging to coagulase-negative staphylococci (CoNS). Unlike most commercial identification systems that allow a rapid identification of only S. aureus, the MALDI-TOF MS allows a correct identification of various CoNS species (Speers et al., 1998; Bernardo et al., 2002; Dupont et al., 2010). In a comparative study between MALDI-TOF MS and two rapid identification automated systems, BD Phoenix (BD Diagnostic Systems, France) and Vitek-2 (bioMérieux, Marcy l'Etoile, France), the identification of 234 CoNS belonging to 20 different species showed that the MALDI-TOF performance was significantly better (93.2%) than Phoenix (75.6%) and Vitek-2 performance (75.2%) (Dupont et al., 2010). Overall, MALDI-TOF appears to be excellent at identifying various staphylococci species as demonstrated by the congruence of 99.3% (444/447) between MALDI-TOF and rpoB sequence-based identifications (Spanu et al., 2011). As CoNS can cause serious infections and are frequently associated with hospital-acquired infections, the rapid identification at the species level by MALDI-TOF is very useful in distinguishing clinically significant CoNS from contaminant strains (von Eiff et al., 2002).

Problematic identifications

Most of the problematic identifications encountered in most of the recent studies concerned the viridans streptococci group, the pneumococci and anaerobic bacteria (Table 2). Viridans streptococci and pneumococci were commonly misidentified mainly due to an incomplete database reference library. In the study by Seng et al. (2009), nearly 50% of Streptococcus pneumoniae isolates were misidentified as Streptococcus parasanguinis [a close related species within the mitis group of Streptococcus species (Kawamura et al., 1995)] because the database contained only three S. pneumoniae and two S. parasanguinis reference spectra. The problem was solved by adding additional S. pneumoniae isolates reference spectra to the database which clearly indicates that the database need to be updated with multiple spectra of well-characterized streptococcal species. Until now, the identification of Streptococcus spp. remains a problem for MALDI-TOF MS identification especially for closely related species such as S. pneumoniae, Streptococcus mitis, and S. parasanguinis. In the case of pneumococci, the use of MALDI-TOF MS for identification is further impaired by the weak extraction yield caused by the presence of a capsule. Thus, identification of S. pneumoniae should not solely rely on MALDI-TOF as false identification can results in important clinical outcomes. In the study by Cherkaoui et al. 2010), the diagnostic yield for streptococcal species and for Gram-negative anaerobes was < 50% with an accuracy of high-confidence species identification of only 57.1% for streptococci with the Bruker system and of 71.4% with the Shimadzu MS system.

Table 2. Problems commonly found in routine identification by MALDI-TOFMS
ProblemsExamples
  1. a

    A higher number of reference spectra in the database are usually required to accurately identify closely related microorganisms that display a high degree of spectrum similarities. Thus, these two parameters are interdependent.

Limit of resolution of the MALDI-TOF MS methodShigella spp. identified as E. coli
Database discordancesPropionibacterium acnes wrongly identified as Eubacterium brachy
Errors in the reference spectradue to incorrect reference spectra in the database
Similarities of spectra present in the databaseaIncomplete reference librairies for viridans streptococci and pneumococci
Absence or insufficient reference spectra in the databaseaNo reference of non-Clostridium anaerobes in the database
 Insufficient number of reference spectra of Streptococcus pneumoniae and Streptococcus parasanguinis in the database to differentiate accurately these two closely related species
 Only one spectrum of Propionibacterium acnes or Bacillus cereus present in the database is not enough to be representative of the true diversity of P. acnes and Bcereus profiles
Taxonomical discordancesStenotrophomonas maltophila misidentified as Pseudomonas hibiscicola, which is an invalid name for S. maltophila
 Agrobacterium tumefaciens is synonymous of Rhizobium rhizogenes
Insufficient protein signalYeasts require a protein extraction procedure to be correctly identified
Difficult to lyse cell wall structuresPneumococci as well as most strains of Haemophilus influenzae and
Small amount of material sampleKlebsiella pneumoniae possess a capsule which prevents efficient lysis and results to poor spectral quality
 Actinomyces, Gemella, Nocardia, and Streptomyces species usually display weak protein signals.
 Better signal for Enterobacteriaceae grown on blood agar vs. MacConkey agar

To date, only a few studies have analysed the usefulness of MALDI-TOF for routine identification of anaerobic bacteria. Seng et al. 2009) showed that an improved database is required as 50% of the total isolates (46) showing no identification by MALDI-TOF MS were anaerobic bacteria including Fusobacterium nucleatum and non-Clostridium anaerobes that had no reference in the Bruker database. In contrast, for anaerobic species (e.g. several Bacteroides species) with sufficient spectra in the database, the identification by MALDI-TOF MS was better compared with that by conventional biochemical methods. These fastidious organisms are poorly identified using phenotypic methods with a lack of specificity and ambiguous or false identification. There is thus an important need to improve the database entries with additional anaerobes isolates (Bessede et al., 2011).

Similar to other studies, Blondiaux et al. (2010) have demonstrated the difficulties to identify viridans streptococci, pneumococci as well as HACCEK bacteria but also Shigella and several strictly aerobic bacteria (Aeromonas spp., Achromobacter spp., Alcaligenes spp.). In this study, the mass spectra of six Shigella isolates were similar to several E. coli strains present in the Biotyper database. In another study, the misidentification of all Shigella sonnei isolates with E. coli was also documented (Seng et al., 2009).

Using 43 selected Mycobacteria strains, a mycobacterial database could be engineered comprising species-specific spectral profiles allowing identification of 44 species at the species level and of nine strains of the Mycobacterium abscessus complex and the Mycobacterium tuberculosis complex at the mycobacterial clade level (Lotz et al., 2010). Under the preanalytical and analytical conditions used in this study, subspecies of the Mycobacterium abscessus complex (Mycobacterium abscessus, Mycobacterium massiliense and Mycobacterium bolletti) and the Mycobacterium tuberculosis complex (Mycobacterium tuberculosis, Mycobacterium bovis, Mycobacterium microti and Mycobacterium africanum) produced indistinguishable mass profiles due to their high degree of genetic similarity. Using this microbial database, 311 strains grown on solid medium were analysed using MALDI-TOF MS allowing a 97% correct identification, 67% at the species level and 30% at the complex level, respectively. No misidentification was observed. When bacteria were grown on liquid media, correct identification was reduced to 77%, likely due to a reduced number of bacterial load or to potential interference with components of the liquid media. Interestingly, the authors observed that an increase in number of replicates (up to 5) did correlate with an increase in the probability of good identification, especially for slow-growing mycobacteria. Overall, several studies demonstrate that MALDI-TOF MS provides high reproducibility and specificity for mycobacterial identification and represents an alternative to other time consuming and fastidious conventional mycobacterial identification methods (Hettick et al., 2004; Lefmann et al., 2004; Pignone et al., 2006; Lotz et al., 2010).

In our study analysing the performance of MALDI-TOF MS for the identification of 1371 bacterial isolates routinely isolated in clinical microbiology laboratories and characterized by conventional methods, 1278 (93.2%) bacterial isolates were identified at the species level, 73 (5.3%) were only identified at the genus level and 20 (1.5%) gave no reliable identification (Bizzini et al., 2010). Among the 1278 isolates identified at the species level, 63 (4.9%) were misidentified. The majority of discordant results (42/63) were explained by discordances due to the MALDI-TOF database, 14 were due to poor discrimination of the spectra of closely related species such as Shigella spp. and E. coli, and seven were caused by errors in the initial conventional phenotypic and biochemical identifications.

Many of the bacterial identifications that can be peformed only at the genus level are due to incomplete reference spectra covering many different isolates or species from a given genus. For instance, only one reference spectrum of Propionibacterium acnes (strain DSM 1897) or Bacillus cereus are included in the Bruker database which is totally insufficient to cover the true diversity of these bacteria and thus to identify accurately these microorganisms (Bizzini et al., 2010). In addition, mislabelling of bacterial species in the database can cause misidentification by MALDI-TOF MS. In the study by Seng et al. (2009), seven Stenotrophomonas maltophilia isolates were incorrectly identified as Pseudomonas hibiscicola which is an invalid name for a nonfermenting Gram-negative rod that was demonstrated to be S. maltophilia.

These studies show that a complete and representative database is an essential requirement for accurate identification of isolates by MALDI-TOF MS. A frequent update of the reference library database with spectra of appropriate poorly represented reference strains by the manufacturers but also by routine diagnostic laboratories can significantly impact the MALDI-TOF identification performance.

Impact of sample preparation on MS identification

Sample preparation has not been shown to be a major issue for problematic identifications. Samples and matrix prepared according to manufacturer's instructions usually guarantee high quality spectral fingerprints required for efficient MALDI-TOF identifications. In most cases, the composition of an appropriate matrix containing strong solvent ensures efficient bacterial lysis required for MALDI-TOF analysis. When invalid results are initially obtained by MALDI-TOF MS following analysis of intact bacteria directly deposited on MALDI target plate, a step of bacterial protein extraction with acid-containing sample solvents improving cell lysis solves the problem in most cases. Bizzini et al. (2010) have shown that a formic acid-acetonitrile extraction step was required to get a valid MALDI-TOF MS identification for 25.6% of the 1278 valid isolates. The yield of valid score from direct application was almost the same, about 75%, for both Gram-positive and Gram-negative bacteria. For instance, the yields of valid scores without extraction step were equal to 79% for S. aureus, 82% for Enterococcus faecalis, 92% for P. aeruginosa, 74% for E. coli, 58% for Klebsiella pneumoniae and 58% for Staphylococcus epidermidis. However, the authors concluded that protein extraction prior to MALDI-TOF analysis should be performed only in particular cases known to be problematic such as colonies isolated from urine culture devices and/or MacConkey agar which contains crystal violet, a possible interfering substance affecting mass peak signals. Indeed, the higher identification rate obtained after an extraction step largely compensates the longer hand-on-time associated with an extraction procedure.

This was also demonstrated in a study performed by Liu et al. (2007) who have developed a universal sample preparation for characterization of bacteria by MALDI-TOF MS. The protocol, consisting of a pretreatment of bacteria with acidic sample solvents and mixing with CHCA or 5-chloro-2-mercaptobenzothiazole (CMBT) matrix, could be used to analyse both Gram-positive bacteria, including spore-producing B. anthracis and non-spore-producing S. aureus, and Gram-negative bacteria such as Yersinia pestis, E. coli, and Burkholderia cepacia that are characterized by high extracellular-polysaccharide contents.

However, for some microorganisms, insufficient cell lysis and/or low quantity of sample material can be problematic for efficient MALDI-TOF identification and require the use of alternative identification approaches such as 16S rRNA gene sequencing. A study by Bizzini et al. (2011) focusing on the identification of 410 clinical isolates that could not be identified with conventional laboratory methods showed that 133/410 (32.4%) isolates could not be either reliably identified with MALDI-TOF. The failure to obtain a reliable identification was due to the absence of reference spectra in the BioTyper database for 58% (78/133) of the isolates and to poor protein spectral signals for 41.4% (55/133) of the isolates. The authors supposed that the poor spectrum quality observed could be due to either the difficulty to lyse the cell wall of some bacteria such as Gram-positive bacilli and/or to fastidious growth of some isolate, which yielded only small amount of available sample material. Difficult to identify microorganisms belonging to the genera Actinomyces, Gemella, Nocardia, and Streptomyces could be observed in this study.

The rate of successful identification is directly linked to the amount of microorganism available. We have observed that the rate of correct identification is especially poor when the subculture has been incubated on a plate for < 4 h (Fig. 3). Thus, when we tried to identify various bacterial species isolated from positive blood cultures and sub-cultured on agar, we observed that only 30% of bacteria could be successfully identified after 2 h of incubation (Fig. 3c). This identification rate was significantly lower for Gram-positive cocci, remaining below 60% after 6 h. This is likely due to the amount of bacteria available after short incubation period (Fig. 3b). Indeed, using 10-fold dilutions of E. coli strain ATCC 25922 and S. aureus strain ATCC 25923, we could show that as many as 106 bacteria per well were necessary to consistently obtain a spectrum (Fig. 3a) ensuring a score above 2 and that when the amount of bacteria was ranging between 102 to 104 bacteria per well, the score was always below 1.7 (data not shown). A lower amount of bacteria may, however, be identified using a different identification algorithm. Hsieh et al. (2008) have demonstrated that as few as 5 × 103 cells from a pure strain can be identified by MALDI-TOF MS using particular analysis approaches based on selected markers (Hierarchical Clustering Analysis and direct classification model construction). Interestingly, using the classification model analysis, the authors showed that successful MALDI-TOF MS identification can also be achieved from a bacterial species mixture consisting of as few as 3 × 104 cells. This study opens new perspectives for the direct identification of low abundant bacteria located in mixed flora without the prerequirement of bacterial isolation and culturing.

Figure 3.

The amount of material is a critical factor for accurate microbial identification by MALDI-TOFMS. (a) Spectral fingerprints obtained from various quantities (106–103) of Escherichia coliATCC 25922 grown on blood agar plates. The inoculums were prepared from a sample with a turbidity of 4.0 McFarland that was diluted to obtain the approximate quantity used for direct microplate deposition (106–103 bacteria per spot). Protein extraction was performed by directly mixing the samples with formic acid on microplate. The results show that the quality of the spectrum and thus the performance of identification are largely dependent on the sample amount spotted on the microplate. (b) Growth of Staphylococcus epidermidis on blood agar plates at 35 °C in 5% CO2 atmosphere after an incubation of 2, 4, 6, and 8 h, respectively. Usually, at least 6 h of incubation are required to get sufficient amount of material to obtain an efficient MALDI-TOFMS identification (c) Cumulative percentage of MALDI-TOF identification obtained from Gram-negative [Escherichia coli (13), Pseudomonas putida (2), Klebsiella pneumoniae (2), Enterobacter cloacae (1)] and Gram-positive [Staphylococcus epidermidis (13), Staphylococcus aureus (9), Streptococcus pyogenes (4), Streptococcus pneumoniae (1), Staphylococcus hominis (1)] bacteria after short-time plating on agar during 2, 4, 6, and 8 h, respectively. The number of samples for each bacterial species analysed is indicated in brackets.

Fungal identification

Fungal identification still largely relies on phenotypic traits. However, a few days are necessary to obtain mature fungi with phyalids (phyalids are conidiogenous cells observed in a type of fungal asexual reproduction leading to the production of conidia). This time delay may be important given the morbidity and mortality of fungal infections, especially common and life threatening among neutropenic patients. Thus, like PCR and sequencing, MALDI-TOF has the potential to provide accurate and objective identification at species level, with the additional advantage of rapidity and reduced costs compared with PCR and sequencing.

MALDI-TOF MS systems for the identification of microorganisms were successfully adapted for the identification of fungi in the past 10 years. In 2000, Welham et al. were among the first to perform fungal identification using the MALDI-TOF MS approach (Welham et al., 2000). Three fungal species, Penicillium spp., Scytalidium dimidiatum, and Trychophyton rubrum, showed distinct spectral fingerprints allowing accurate species distinction. Since then, many studies have demonstrated the usefulness of the MALDI-TOF application for the identification of various fungal groups such as penicillia, aspergilla, Fusarium, Trichoderma, and dermatophytes. However, until now, MALDI-TOF MS is mainly used for the routine identification of yeasts whereas further development has to be accomplished in database libraries and sample preparation protocols to implement this identification approach to other group a of fungi such as filamentous fungi and dermatophytes.

Yeasts

The databases of the associated softwares (Biotyper and Saramis of the two main MALDI-TOF instruments contains reference spectra of multiple clinical yeast isolates, including several Candida spp. and Cryptococcus neoformans, which allows the use of MALDI-TOF in routine yeast laboratory identification (Table 1).

Marklein et al. (2009) evaluated MALDI-TOF MS for the rapid routine identification of clinical yeast isolates and showed that 92.5% (247/267) of clinical isolates of Candida, Cryptococcus, Saccharomyces, Trichosporon, Geotrichum, Pichia, and Blastoschizomyces spp. were accurately identified. In a study performed by van Veen et al. (2010), 85% of 61 yeast isolates comprising 12 different species were correctly identified without occurrence of major errors. Bizzini et al. (2010) achieved a 100% correct identification of 24 yeast isolates belonging to 12 different species. The suitability of the two commercially available MALDI-TOF MS systems, Bruker and Shimadzu, and their respective associated softwares and databases, Biotyper and Saramis, was tested for rapid species identification of yeasts in a clinical diagnostic approach (Bader et al., 2010). Both MALDI-TOF MS systems have showed a similar species identification rate of 97.6% for Bruker/Biotyper and 96.1% for Shimadzu/Saramis that were comparable with the biochemical tests rate (96.9%). Based on isolates that were contained in the respective database, no misclassifications were seen with Saramis and fewer misidentifications were reported by the Biotyper compared with classical approaches. Using the Bruker MALDI-TOF system, Stevenson et al. (2010b) have created a spectral database library for 109 reference strains of yeast representing 44 species and eight genera to evaluate the use of MALDI-TOF MS for the rapid identification of yeast species. This library was challenged with 197 clinical isolates. Three isolates gave no spectral score as no reference spectrum were included in the database library. Of the remaining 194 clinical isolates, 192 (99.0%) were correctly identified at the species level and two organisms gave consistently low spectral scores that could not be identified. In summary, the use of MALDI-TOF MS for the identification of clinically relevant yeasts is rapid and accurate providing that the database is constructed with a comprehensive collection of accurately identified reference strains.

Filamentous fungi

Chen & Chen (2005) successfully identified several Penicillium species directly from intact fungal spores mixed with the MALDI matrix. Hettick et al. (2008b) have achieved 100% correct identifications of 12 Penicillium species by bead beating fungal samples resuspended in a acetonitrile/trifluoroacetic acid solvent prior to MALDI-TOF analysis. The observed mass spectra contained abundant peaks in the range 5000–20 000 m/z allowing unambiguous discrimination between species. In addition, a biomarker common to all Penicillium mass fingerprints was observed at m/z of 13.9 kDa.

Using an extraction method similar to that used for Penicillium species, Hettick et al. (2008ab) obtained also highly reproducible mass spectral fingerprints for 12 species of Aspergillus and five strains of Aspergillus flavus. The 12 species were correctly identified but only a 95% accurate identification was obtained at the strain level. It was also pointed out that Aspergillus niger could not be distinguished from Aspergillus chevalieri. The authors concluded that the identification of Aspergillus spp. with MALDI-TOF MS would require a comprehensive database of at least 180 species of Aspergillus. A study showed that different species of Aspergillus, including aflatoxigenic and nonaflatoxigenic spp., could be characterized directly from intact spores (Li et al., 2000). However, the authors reported certain discrepancies due to the difficulties encountered to discriminate the spectra obtained with some of the analysed species. A database including the reference spectra of 28 clinically relevant Aspergillus species was engineered in a recent study by including species-specific fingerprints of both young and mature colonies of reference strains (Alanio et al., 2010). The performance of the database was tested on 124 clinical and 16 environmental Aspergillus isolates resulting in a 98.6% (138/140) correct identification with 100% specificity (0% misidentification). This study has demonstrated that a complete fingerprint database including spectra from both young and mature fungal colonies makes MALDI-TOF a robust method for Aspergillus species identification regardless of the maturity of the tested isolates.

The identification of multiple Fusarium spp. has also been demonstrated by various studies. In the study by Marinach-Patrice et al. (2009), 62 strains or isolates belonging to nine Fusarium spp. were subjected to both molecular identification and MALDI-TOF MS analysis. Following updating of the BioTyper database with 13 strains of five Fusarium spp., 57 (92%) strains were correctly identified using MALDI-TOF MS analysis. Only one Fusarium pseudonygamai isolate was misidentified and four Fusarium isolates were not identified due to absence of reference spectra in the database. MALDI-TOF MS was also used successfully to identify five mycotoxin-producing Fusarium spp. by direct analysis of spores which yielded highly reproducible MS profiles (Kemptner et al., 2009).

Dermatophytes

The most important clinical fungal dermatophytes species, T. rubrum, Trychophyton interdigitale, Trychophyton tonsurans, and Arthroderma benhamiae, originating from skin and nail were recently identified using the Saramis database (Erhard et al., 2008). Except for one T. rubrum strain, a high level of confidence (99.9%) was obtained in this study where sufficient MS spectra were used to produce a super-spectrum for each species.

Problematic identifications

Similar to bacteria, absence of identification or misidentification of fungal species by MALDI-TOF MS analysis are essentially due to absence, mistakes or incomplete reference spectra in the database (Table 2). High quality MS spectra are usually easily obtained with both fungal hyphae and spores following the manufacturer's instruction or based on the recommendation of reference studies. For instance, because of their cell wall structure, yeasts need an extraction step to yield a valid score of identification by MALDI-TOF MS. In the study of Bizzini et al. (2010) only 4% (1/24) of the valid results were obtained by direct application of the colony on the MALDI plate and an extraction method prior to analysis by MALDI-TOF was also shown to be mandatory to obtain appropriate spectra in the study performed by van Veen et al. (2010). The spectra of several Penicillium spp. obtained from bead beating fungal samples resuspended in an acidic solvent prior to MALDI-TOF analysis was more discriminative (abundant peaks in the range 5000–20 000 m/z) than spectra obtained directly from intact spores (range of 2600–7378 m/z) (Chen & Chen, 2005; Hettick et al., 2008ab). However, Valentine et al. (2002) identified A. niger, Rhizopus oryae, Trichoderma reesei, and Phanerochaete chrysosporium using either intact spores, hyphae or extracts showing that intact cells, sonicated cells, and acid-treated cells yielded similar spectra.

The main problem is that, unfortunately, very few reference spectra are currently included in the database of commercially available MALDI-TOF MS systems. Most of the studies showing that the MALDI-TOF MS identification is a powerful system for the characterization and identification of fungi have built and used their own reference spectra database and have developed their own sample preparation techniques. There is thus still a lack of standardized extraction protocols regarding filamentous fungi.

In addition, the spectral signal of filamentous fungi may be strongly influenced by the phenotype of the fungus including basidiospore, monokaryon, dikaryon, fruiting body, surface mycelium, strands and substrate mycelium. Moreover, vegetative mycelium grown on agar shows multiple zones that correspond to different ages or developmental stages. These variations may thus influence the spectral reproducibility of the same isolate and a comprehensive database of filamentous fungi should include MS fingerprints of several different developmental forms to guarantee high yields and accuracy of identification as demonstrated by Alanio et al. (2010).

Direct identification from samples

Given the accuracy of MALDI-TOF for bacterial identification, this technology might be directly applied to some clinical samples, such as blood, urine, cerebrospinal fluid, pleural fluid, peritoneal liquid, and synovial fluid. The major limitation is the amount of bacteria present in the samples and the limit of detection of current MALDI-TOF protocols. To circumvent this difficulty, large volumes are used for blood and urine and an additional enrichment by culture is available for blood (see paragraphs below). Regarding cerebrospinal fluid, Nyvang Hartmeyer et al. (2010) successfully identified S. pneumoniae 30 min after receiving a sample supporting the proof-of-concept. However, practically, bacterial identification from cerebrospinal fluid, strongly limited by the low bacterial load and the limited volume available, is yet not applicable in routine diagnostic laboratories.

Blood (hemoculture)

Bloodstream infection, septic shock, and endocarditis represent severe diseases with important mortality and morbidity. Blood culture represents the best way to establish the etiology of such infections and to guide antimicrobial treatment. This is important since rapid and appropriate antimicrobial therapy is pivotal to reduce poor outcome (Kollef, 2000). Indeed, the fatality rate was 20% for bloodstream infection patients treated with appropriate therapy and 34% for patients treated with inappropriate therapy (Leibovici et al., 1998). The rapid notification of the Gram stain result from positive blood culture has also a positive impact for adaptation of antimicrobial regimen (Munson et al., 2003). Consequently, the precise identification of a microorganism isolated from positive blood culture early after Gram stain notification will likely help clinician to better adapt the antimicrobial therapy. As an example, the impact on the choice of the antibiotic will likely be significant when Gram-positive cocci are identified from blood cultures, as the antibiotic susceptibility of Enterococcus faecium is clearly different from that of alpha-hemolytic streptococci.

Quantitative blood cultures have demonstrated that bacterial load during bloodstream infection is very low in adults, often < 1–10 colony forming unit mL−1. In practice, blood samples are inoculated into bottles containing broth media and incubated in automated instruments monitoring CO2 concentrations released during bacterial growth. At the automated growth detection time, the bacterial load may reach a heavy growth up to 106–108 colony forming units mL−1. In our laboratory, Enterobacteriaceae, P. aeruginosa, and aerobic Gram-positive cocci were generally detected when present at 107 bacteria mL−1. Such bacterial concentration might be adequate to allow accurate bacterial identification using MS.

However, the blood culture bottle fluid represent a complex solution with multiple nonbacterial proteins isolated from patient's blood and nutrient growth media. These proteins alter the specific bacterial MS profile obtained by MALDI-TOF and have a detrimental effect on the performance of algorithm used to query the database containing bacterial MS profiles.

The preparation of a bacterial pellet from positive blood culture includes a differential centrifugation step to discard blood cells, an erythrocyte lysis step and a subsequent washing step to remove additional nonbacterial components. Application of this protocol allows the identification in < 1 h when compared with overnight growth of bacteria required to obtain pure colonies for biochemical identification.

Recent studies (Table 3) have shown that a correct identification by MALDI-TOF is obtained in > 80% of cases starting from blood culture bottles. The results varied according to the bacterial pellet preparation protocol and the type of microorganism present in blood cultures. Using ammonium chloride as lysing agent, 89% of Gram-negative bacteria and 73% of Gram-positive bacteria (i.e. 90% for Staphylococci and 33% for Streptococci) were correctly identified at the species level (Prod'hom et al., 2010). No identification was observed in 21% of cases. The lower performance of MALDI-TOF for Gram-positive bacteria and particularly Streptococci has also been observed in other studies (Table 3). Similar to MALDI-TOF identifications from pure microbial isolates, several hypotheses have been suggested to explain discordant results obtained from blood cultures: (1) the close relatedness of the different species especially within Streptococci, notably within S. mitis group (i.e. S. pneumoniae, S. mitis, Streptococcus sanguinis, Streptococcus oralis) conferring closely related MALDI-TOF spectrum profiles, (2) the cell wall composition of Gram-positive bacteria conferring resistance to lysis. In addition, the presence of capsules in different species (S. pneumoniae, Haemophilus influenzae, K. pneumoniae) may also explain the lower performance of MALDI-TOF.

Table 3. Performance of MALDI-TOF identification obtained from positive blood culture
ReferenceSample
(n)
Blood culture systemConcordant identification to species levelConcordant identification to genus levelIdentification difficultyComments
  1. Adapted from Carbonnelle et al. (2010).

  2. GN, Gram-negative bacteria; GP, Gram-positive bacteria; ICM, intact cell method; PEM, protein extraction method; SDS, sodium dodecyl sulfate; ND, not determined.

Prod'hom et al. (2010)126Positive blood culture (Bactec)78%,
GN: 89%,
GP: 72%
79%,
GN: 89%,
GP: 73%
Streptococcus mitis group, Staphylococcus spp.Use of ammonium chloride to lyse erythrocyte
The presence of a capsule explain partially the low identification rate of S. pneumoniae, H. influenzae, K. pneumoniae
La Scola & Raoult (2009)599Positive blood culture (Bactec)66%
GN : 91%
GP : 49%
66%Streptococcus spp., polymicrobial samplesDuring the study, modification of the extraction protocol to improve Staphylococci identification (38–> 75%)
Stevenson et al. (2010ab)212Positive blood culture (179),
spiked bottles (33) (Bactec)
80%80%Streptococcus mitis group, Propionobacterium acnesUse of a separator device for blood cells removal
Ferroni et al. (2010)685Positive blood culture (388),
spiked bottles (312)
89%98%Streptococcus pneumoniae, Streptococcus mitis group, polymicrobial samplesUse of saponin to lyse erythrocyte
Christner et al. (2010)277Positive blood culture (Bactec)94%95%Cocci GPMismatch mostly resulted from insufficient bacterial quantity and occurred preferentially with GP bacteria
Ferreira et al. (2010abc)300Positive blood culture (Bactec)43%,
GN: 83%,
GP: 32%
72%,
GN: 97%,
GP: 66%
Streptococcus mutans, Staphylococcus spp., Staphylococcus aureusNo mixed culture
Ferreira et al. (2010abc)68Positive blood culture (Bactec)76%
ICM 47%
PEM 76%
96%
ICM 51%
PEM 93%
Staphylococcus spp.PEM improve the identification compared with ICM
Marinach-Patrice et al. (2010)48Spiked bottles with Candida (Bactec)100%100% Use of SDS as detergent to lyse erythrocytes.
Application of a new algorithm's concept as residual blood proteins and Candida share many masses in common
Szabados et al. (2011)268Positive blood culture (BacT/ALERT)
(noncharcoal containing bottles)
31%NDPolymicrobial samplesIn a preliminary study, a lower rate of identification with charcoal-containing bottle was observed (8%)
Romero-Gomez & Mingorance (2011)129Positive blood culture
Bactec (42)
Versatrek (35)
BacT/ALERT (52)
68%
76% (Bactec)
69% (Versatrek)
62% (BacT/Alert)
NDSignificant lower performance for GP bacteria 

The performance of MALDI-TOF for the identification of fungi in blood culture is low. In one study (Ferreira et al., 2010c), no fungi (0/18) were identified at the species level and only one at the genus level. This poor performance is attributed to the relatively low load of fungi observed in positive blood culture and to the presence of residual blood protein which co-migrates during the MALDI-TOF assay, which impairs the performance of the diagnostic algorithm. To circumvent this detrimental effect, a reference database of fungi obtained from blood culture spiked with fungi was established to obtain correct identification at the species level (Marinach-Patrice et al., 2010).

Limitation of the MALDI-TOF identification was also observed for mixed bloodstream infection, where only one bacterium could be identified (La Scola & Raoult, 2009; Moussaoui et al., 2010; Szabados et al., 2011).

The impact of the broth on the spectral quality and thus on the rate of identification has been analysed in several studies. The first published studies have used bottles adapted to the BACTEC system (Becton Dickinson, Franklin Lakes, NJ). More recently, bottles with and without charcoal adapted to Bact/ALERT automated instruments have been tested (bioMérieux). In this system, charcoal is used to inactivate antimicrobial agents present in the patient's blood. In one study, the rate of identification using MALDI-TOF was 30% without charcoal and decreased to only 8% when charcoal was present (Szabados et al., 2011). Another study compared the performance of identification using MALDI-TOF with positive blood culture obtained from three automated systems: BACTEC, VERSATREK (Trek Diagnostic Systems, Cleveland, OH) and BactT/ALERT. The rate of direct identification of bacteria cultured in these three automated blood culture devices were 76%, 69%, and 62% for the investigated samples, respectively (Romero-Gomez & Mingorance, 2011).

The importance of the protein extraction method was compared with the so-called intact cell method, which consist in the direct deposition on MALDI plate of bacterial pellet obtained from positive blood culture. In one study, the performance of MALDI-TOF identification at the species level was 47% for the intact cell method compared to 76% for the protein extraction method (Ferreira et al., 2010abc). The simple extraction method used in this study improves significantly the performance of MALDI-TOF identification rate. In our laboratory, we now use this approach on a routine basis, with a turnaround time estimated to be about 1 h. However, to be efficient and have such a low turnaround time, there is a need to prioritize identification of bacteria isolated from blood cultures over other routine applications of the MALDI-TOF. Thus, this activity somehow delays other microbial identifications, as shown in Fig. 4.

Figure 4.

Importance of the organization of a working day to optimize time to results. Each line represents a MALDI-TOF run, with the waiting time for equipment availability, the time required to prepare the target layout, and the time of the MALDI-TOF running process. In this example of a working day, eight MALDI-TOF target plates have been used to analyse multiple clinical samples [ICU, S: intensive care unit and surgery (one run); IM, P: internal medicine, pediatry and others (one target plate); BC: blood cultures (four runs); U: urine (two runs)]. Urgent samples such as blood cultures are directly processed while colonies identification from agar cultures are processed by batch. The processing of the samples is organized to guarantee the optimal use of the MALDI-TOF device during the working day.

In conclusion, the application of MALDI-TOF identification to microorganism pellets obtained from positive blood culture allows a rapid identification of microorganisms growing in blood culture which is important for the management of bloodstream infections.

Urine

As the bacterial amount in urine taken from patients with urinary tract infection is often ≥ 105 bacteria mL−1, the use of MALDI-TOF directly on urine has been investigated by numerous groups. However, as only about 1–2 μL of liquid may be deposited on the MALDI-TOF microplate, results were not accurate when untreated urine are directly deposited and thus various pretreatment steps have been tested with different outcomes. Using two consecutive centrifugation steps (low speed to remove leukocytes and high speed to collect the bacteria), Ferreira et al. (2010abc) could accurately identify as much as 94.2% of bacteria. However, they only included in their study urine with > 105 bacteria mL−1. When investigating a simplified protocol in our laboratory for the identification of E. coli (single centrifugation step), the results was acceptable when the bacterial concentration was of 107 and 108 bacteria mL−1 with 69% and 70% of samples being accurately identified at the species level (score > 2). However, the yield was poor with lower bacterial load (Fig. 5). Given the huge amount of urine processed on a daily basis, the low value of early identification and the requirement of bacterial isolation in pure culture for antibiotic susceptibility testing, it appears that the MALDI-TOF on urine is not cost-effective and not efficient enough to be implemented directly on urine samples. Indeed, as most urinary isolates are E. coli (> 80% among uncomplicated cystitis occurring in young women and about 50% in complicated urinary tract infections), the use of a chromogenic agar (i.e. Urid; bioMérieux) coupled with simple phenotypic tests such as indole represents a simpler way to identify most strains. MALDI-TOF will then be mainly used to identify the remaining species starting from colonies.

Figure 5.

Escherichia coli identification yield in urine samples. Five milliliters of urine samples positive by microscopy were centrifuged to concentrate and collect the bacteria. The pellet was resuspended in 200 μL water and subjected to protein extraction with ethanol and formic acid prior deposition on the microplate. The graph shows the identification yield according to the bacterial load per mL.

Maintenance and quality controls

The MALDI-TOF MS is increasingly used in clinical diagnostic laboratory for microbial identification with reliable results for bacterial identification at the species level.

MALDI-TOF results may, however, be impaired by problems arising during extraction, for instance (1) when testing encapsulated bacteria (S. pneumoniae, K. pneumoniae), (2) when testing bacteria such as Streptomyces, that exhibit a particular cell wall that reduces the yield of protein extraction, (3) when the extraction protocol is not properly conducted, or (4) when the reagents used for extraction are outdated or impaired by inadequate storage. Problems (1) and (2), due to intrinsic bacterial properties, may only be circumvented by the development of specific extraction protocols and problems (3) and (4) might be prevented by an adequate quality program. The performance of the extraction step and of the MALDI-TOF mass spectrometer may be checked by routinely testing a few selected bacterial strains, for which spectra are available in the database. This control should ideally be done in parallel with and without a specific extraction step. We thus implemented in our laboratory routine internal quality controls that test the quality of the extraction step on two different bacterial species (E. coli ATCC 25922 and S. aureus ATCC 25923). To set-up this quality control, we first investigated the reproducibility of the extraction step (see Fig. 2 and paragraph on reproducibility). This allowed us to define conserved peaks (peaks present in 10/10 replicates). Then, we routinely tested once a week both bacterial species. As score values were always above 2 and to obtain a better expression of the quality of the extraction, we decided to report the proportion of conserved peaks detected, considering a peak as present only when its intensity was above 200. Indeed, the rate of detection of conserved peaks reflects not only the quality of the sample but also the protein yield and the spectral quality (Fig. 6).

Figure 6.

Importance of the maintenance and quality controls of the MALDI-TOF device demonstrated by a follow-up of the temporal reproducibility. (a) Spectral fingerprint of Staphylococcus aureus quality controls during a 10-week control period. (b) Percentage of conserved peaks during a 10-week control period. Escherichia coli and Staphylococcus aureus quality controls fingerprints were compared with a set of conserved peaks (81 and 80, respectively). The poor performance observed during week 1–3 was likely caused by a problem of inadequate sample deposition on the MALDI-TOFMS microplates.

MALDI-TOF results may also be impaired by inadequate deposit of the sample on the microplate and by poor cleaning of the microplate between runs. Inadequate deposit of samples is relatively rare when starting from bacterial colonies and the learning curve is rapid with most laboratory technicians being already experts in depositing appropriate amount of bacteria after only a few training days. However, erroneous identification may occur due to well inversions, especially when large series are processed and when stress is increased by human resources shortage.

Poor cleaning of the microplate is a problem only encountered by Bruker users as Shimadzu users will use disposable MALDI-TOF plates. Bruker commonly propose to use trichlorofluoroacteate (TFA) or guanidium to clean microplates between usages. As TFA is associated with significant occupational hazard (eye, skin, and respiratory toxicity), we used in our laboratory an alternative protocol to clean MALDI-TOF microplate. This protocol, initially proposed by Bruker, which mainly uses ethanol and mechanical cleaning of target plates, is, however, insufficient to properly clean MALDI-TOF microplates. Indeed, when investigating cleaned plates by testing them only in presence of matrix, we could obtain some accurate identification with score > 2 and corresponding to the same bacterial species investigated the day before. Such accurate identifications obtained after plate cleaning could rarely correspond to wells where some material was still present (Fig. 7a, wells C4, E3, and E4), but also occurred in apparently clean wells. Thus, a systematic control of the microplates should be done and the cleaning protocol adapted when necessary. Noteworthy, disposable microplates are now also available for Bruker users.

Figure 7.

Quality control for MALDI-TOFMS. (a) Presence of residual materials on some wells (especially C4, E3, and E4, indicated by white arrows) of a MALDI-TOF microplate after a routine wash, highlighting the importance of careful wash after each usage; (b) mass spectra obtained with the Bruker BST control, that consists in lyophilized Escherichia coli spiked with RNAse A (black arrow at a m/z of 13 683) and myoglobin (white arrow at a m/z of 16 952); in the lower part of the panel, the list of the eight proteins that should be present to validate the run and their expected and observed m/z values.

MALDI-TOF results may also be impaired by technical problems and/or poor maintenance of the MALDI-TOF device. To identify possible technical problems and to recalibrate the MS apparatus, we routinely use in Lausanne the calibration control proposed by Bruker. This control (called BTS) consists of lyophilized E. coli extracts and two supplementary proteins, RNAse A and myoglobin, which respectively exhibit peaks at 13 683 and 16 952 m/z (Fig. 7b). Finally, in the future, external quality control should be implemented.

Appropriate maintenance (Fig. S1) is also essential to warrant accurate bacterial identification. Vacuum failure, and thus MALDI-TOF MS functional disturbance (see Fig. 1), might be observed due to the presence of dust on plastic joints or to the aging of these plastic joints (Fig. S1c and d). Dust exposure of the MALDI-TOF apparatus might be reduced by placing the mass spectrometer in a quiet area, without drafts. Carbonization of bacteria embedded in the matrix material following each laser pulse is also a source of concern, as the laser source may be soiled (Fig. S1d). With the Bruker instrument, the level of dirt present on the laser source may be indirectly estimated according to the number of shots needed to obtain a correct identification. Maintenance should ideally be done before the dirtiness rate reaches 80%, i.e. about four times a year if three to five microplates are tested per day. Of course, frequency of maintenance of the MALDI-TOF should be increased if the apparatus is heavily used or located in a crowded/dusty area.

Despite adequate maintenance and correct procedures, some microbial groups will repeatedly be misidentified due to poor content of some databases. Thus, it appears critical not only to implement a quality control program targeting routine procedure but also to incrementally improve the quality of the database. In conclusion, quality controls might help to improve the quality of proteins extraction, MALDI-TOF analysis, and completeness of databases. This will thus further improve the accuracy and usefulness of MALDI-TOF.

Accuracy, time and cost effectiveness

The MALDI-TOF MS approach represents a new tool that has the potential to replace conventional identification techniques for a majority of routine isolates encountered in clinical microbiology laboratories. The performance of the MALDI-TOF MS approach has been compared in several studies (Seng et al., 2009; Bizzini et al., 2010; Cherkaoui et al., 2010; van Veen et al., 2010) with multiple routine phenotypic identification methods, such as semi-automated Gram staining (Aerospray Wescor®; Elitech), catalase and oxidase determination tests, automated identification by Vitek® (bioMérieux), the Api Anaérobie BioMérieux® identification strip for anaerobes (bioMérieux), the Slidex Staph plus system (bioMérieux). These studies showed that the MALDI-TOF technique has a high accuracy for most microbial identifications and performed equally as well as or better than conventional techniques. For instance, van Veen et al. (2010) showed that significantly more bacterial isolates could be identified to the species level by MALDI-TOF MS, with a special higher performance for staphylococci and bacteria from the HACCEK group. Similarly, Bader et al. (2010) have shown that identification of yeast species with MALDI-TOF MS systems gave an overall species identification rate (97.9% for Bruker and 96.1% for Shimadzu) that was comparable with the one obtained with the biochemical tests (96.9%).

The most striking differences between MALDI-TOF technique and conventional identification methods are observed in the estimated time and costs required for sample identification. The cost of bacterial identification by MALDI-TOF MS was estimated to represent only 17–32% (around €1.43/sample) of the costs of conventional identification methods (around €4.6–8.23/sample) in the study performed by Seng et al. (2009), which is supported by at least two other prospective studies (Bizzini et al., 2010; Cherkaoui et al., 2010). Cherkaoui et al. (2010) have shown that the reagents required for phenotypic identification using modern automated platforms costs at least around $10 per isolate, whereas MS-required reagents do not exceed $0.50. The expensive prices of MS instruments are comparable with other common bacteriology laboratory equipment such as automated blood culture and 16S sequencing devices but the running costs are significantly cheaper than those of conventional identification methods. Thus, in our laboratory, we estimated that the reagents costs spared during a year are of approximately €40 000 (Table S2). Of course, this cost analysis did not include the cost of maintenance, neither for MALDI-TOF nor for automated phenotypic identification systems such as Vitek.

Compared with conventional identification methods, MALDI-TOF has been shown to confer in most cases a significant gain of both technician working time (preanalytical procedure to prepare samples) and turnaround time (automated analytical procedure to obtain results).

The time needed for bacterial identification from intact cells was 6–8.5 min vs. 5–48 h for conventional identification as estimated by Seng et al. (2009). When an extraction step is required, Bizzini et al. (2010) have estimated that the extraction procedure for a single sample takes approximately 6 min, a time per sample that is further reduced during batch processing. The time effectiveness gained with MALDI-TOF identification compared with classical identification approaches is even more accentuated when several isolates are analysed in parallel. As reported by Cherkaoui et al. (2010) the analysis of 10 isolates in parallel can be accomplished in < 15 min with limiting working time by MS, whereas more than 360 min would be required on an automated system with more hands-on time for each isolate, demonstrating that MALDI-TOF MS provide a reduction of both working and turnaround times. Regarding yeast identification by MALDI-TOF MS, Marklein et al. (2009) showed that this technique required minimal time for technicians to process yeast samples for analysis and to interpret the results. The MALDI-TOF identification procedure from single yeast colonies on the agar plate was generally completed within 10 min per isolate and within 3 h for 96 samples. In contrast, the identification of germ tube-negative Candida species by phenotypic methods can require incubation periods of up to 72 h, a significantly longer turnaround time compared with MALDI-TOF. Molecular approaches have been or are currently under development to provide efficient identification of yeasts with a more rapid and reliable efficiency than classical phenotypic methods. However, high-resolution DNA-based molecular techniques such as 26S rRNA or internal transcriber spacer DNA sequencing and real-time PCR assays are expensive and time-consuming and appear in most cases less convenient than MALDI-TOF MS for routine laboratory identifications (Schabereiter-Gurtner et al., 2007; Montero et al., 2008; Seyfarth et al., 2008). Nucleic acid-based identification strategies suffer problematic limitations in terms of (1) technical problems (inhibitory compounds, contamination, separate areas for sample preparation/amplification/analysis), (2) reagent and labor costs, (3) spectrum of species identification in a single assay often limited to a few individual species, and (4) much longer turnaround time than MS. Thus such lengthier, costlier and more labor-intensive alternatives to MALDI-TOF MS are usually reserved for the identification of the small minority of isolates not identified by MS alone or by other conventional phenotypic and/or biochemical identification approaches.

Taxonomy and microbial typing

Taxonomic usefulness of MALDI-TOFMS

Taxonomy is the systematic classification of organisms based on their phenotypic, genetic and phylogenetic characteristics. Thus, various phenotypic approaches (morphology, biochemical reactions, and sugar assimilation) have been used by microbiologists to classify microorganisms. However, genome analysis through sequencing of bacterial genes or of the entire genome currently represents the gold-standard of microbial taxonomy, although it should always be confronted to phenotypic traits in a polyphasic approach. MALDI-TOF MS represents an additional approach to classify microorganisms based on phenotypic traits. As previously shown in several studies mentioned in this review, MALDI-TOF MS systems give accurate and reproducible results at the species level that are in most cases concordant with genomic identification methods and consequently, MALDI-TOF has the potential to be used in polyphasic taxonomy. For instance, in contrast to housekeeping genes sequencing such as 16SrRNA that provide taxonomical data on a single gene at a time, MALDI-TOF MS fingerprints provide information about multiple protein components that characterize a microorganism. In addition, MALDI-TOF fingerprints provide data of both the presence (conservation/divergence) and the intensity (expression level) that together compose a two dimensional taxonomical asset which offer a better discriminative resolution for microorganism classification.

Microbial typing and identification at the subspecies level

MALDI-TOF MS allows identification of microorganisms at the species level and sometimes at the subspecies level but several studies have shown that the requirements for MALDI-TOF MS-dependent microbial typing are different and more complex than those required for routine microbial identification. This is a challenge for clinical laboratories that want to use MALDI-TOF MS for routine strain typing. Relatively few biomarkers (5–10 peaks) are usually required for the identification of microbial isolates at the species level, whereas a much larger number of reproducible peaks is needed for subspecies identification (Dieckmann et al., 2008). Many studies cited in this review using conventional MALDI-TOF procedure could not directly identify taxonomic entities like species or subspecies with an accuracy of 100% without the assistance of DNA-based methods. Microbial typing and thus microbial characterization at the subspecies level required very different sample preparation and analytical procedures (Murray, 2010). As mentioned previously in this review, accurate sample preparation is generally unnecessary for microorganism identification but for strain typing and subspecies identification, a rigorous optimization of testing parameters appears to be crucial. The challenge is to obtain a sufficient number of reproducible markers with specificities below species-level specificity (Rupf et al., 2005; Vargha et al., 2006; Dieckmann et al., 2008). For instance, the sample preparation procedure (whole cell or protein extraction), the protein concentration, the type of matrix, the sample : matrix ratio, the concentration of acid added to the matrix and the growth medium are examples of technical parameters that can have a significant influence on the MALDI-TOF spectral profile of biomarkers (Vargha et al., 2006; Dieckmann et al., 2008).

The choice of analysis solutions used to process mass spectra can have a significant impact on the power of discrimination and thus on the ability to distinguish closely related isolates. Maximizing reproducibility is also critical for accurate microbial characterization. One of the major limitations of MALDI-TOF-based microbial typing is primarily due to the algorithmic methods used to analyse the protein profiles. Several similarity coefficients can be used to determinate level of similarities. Some account only for peak presence/absence such as the Dice similarity coefficient, whereas others take also in consideration the peak intensities, such as cosine and Pearson product-moment correlation coefficients. The chosen similarity coefficient affects the reproducibility and the discriminatory power of the method. Several studies have demonstrated that the Pearson coefficient appears to be more adequate for the correct classification of microbial isolates. A study by Giebel et al. (2008) showed that the Pearson product-moment correlation coefficient permitted a more accurate classification of Enterococcus spp. isolates than the Dice similarity coefficient. Similarly, the use of the Pearson correlation coefficient allowed a 28% increase in the rate of correct classification of E. coli isolates (Giebel et al., 2010). Using optimal sample preparation and MALDI conditions for discrimination at the strain level and by using the Pearson coefficient, Vargha et al. (2006) have shown that MALDI-TOF MS offered a better discriminatory power than 16S rRNA gene sequencing for the classification at the subspecies level of Arthrobacter isolates. For instance, members of the A. globiformis cluster have 99–100% sequence similarity, whereas MALDI-TOF MS similarity is 60–95%.

In some cases, the identification of multiple or single unique subspecies biomarkers have been used to discriminate closely related microbial isolates exhibiting highly similar mass signatures. For instance, five unique and conserved biomarkers ions were identified in environmental E. coli isolates from avian but not from human sources (Siegrist et al., 2007). Similarly, several Listeria monocytogenes serotypes could be separated using discriminating peaks (Barbuddhe et al., 2008).

Despite an increased level of complexity required for microbial subspecies classification, several published studies support the observation that MALDI-TOF MS represents a new promising technological approach for the classification of clinical and environmental isolates. Dieckmann et al. (2008) have successfully classified 126 isolates of Salmonella at the species and subspecies levels by optimizing a procedure that allowed them to obtain more than 300 biomarker peaks ranging from 2 to > 35 kDa. They found that out of three matrix mixtures, SA produced the most informative spectra by providing a significant increase of high molecular mass peaks with important subspecies specificity. In addition, simple clustering of mass data from bacterial fingerprints did not initially provide a clear discrimination of the strains at the subspecies level and a bioinformatic approach recently published by Teramoto et al. (2007) had to be used. The approach is a new phylogenetic classification method based on ribosomal protein profiling by MALDI-TOF MS using the bioinformatics-based method for rapid identification of bacteria published by Demirev et al. (1999). Using this approach, the result of the classification of several P. putida strains including different biovars was in agreement with the gyrB gene sequences-based classification.

The determination of serotypes of Shiga toxin-producing E. coli isolates has been achieved by constructing prototype spectra representing different serotype groups (Karger et al., 2011). The prototype spectra were generated by removing masses with low discriminative significance, which is a process comparable with the generation of super-spectra proposed by the Saramis software. The generation of prototype spectra allowed a reduction of incorrect assignments down to 0.7% compared to the 31% incorrect assignments observed when unfiltered mass spectra were used. Unlike restriction fragment length polymorphism analysis, this analytical methodology could not achieve a differentiation below the serotype level.

The typing of several microorganisms, such as Staphylococcus and Listeria species, for epidemiological studies require the use of various conventional techniques such as pulsed-field gel electrophoresis (PFGE), amplified fragment length polymorphism analysis and multilocus sequence analysis (MLSA). These gold-standard techniques provide accurate classification of microorganism but suffer from important time and cost investments. For instance, only a few hours are required to obtain results by MALDI-TOF MS whereas several days are necessary to collect PFGE data. In addition, these methods are technically relatively complex and have to be usually performed by experienced technicians.

Several Staphylococcus studies have developed standardized methods to achieve reliable and reproducible species level identification and sub-typing from MALDI-TOF fingerprints (Edwards-Jones et al., 2000; Walker et al., 2002; Jackson et al., 2005). When performed under careful experimental conditions, MALDI-TOF MS has been used to discriminate between methicillin-resistant (MRSA) and methicillin-susceptible S. aureus strains and to subtype MRSA strains. Thus, compared with conventional antimicrobial susceptibility test methods or gene sequencing techniques, these studies have demonstrated that MALDI-TOF MS represents a fast and cheap approach to accurately differentiate S. aureus strains. Unfortunately, no comparison of MALDI-TOF MS with PFGE has been performed to demonstrate that these two methods would give similar results. However, Barbuddhe et al. (2008) have used MALDI-TOF to accurately identify 146 strains of different Listeria species and correctly classified all L. monocytogenes serotypes in agreement with PFGE, which is one of the most common subtyping techniques used to classify L. monocytogenes serotypes. Similarly, Fujinami et al. (2010) have demonstrated that MALDI-TOF MS and PFGE gave similar accurate identification of epidemiologic Legionella strains. Hazen et al. (2009) have demonstrated that MALDI-TOF MS could be used to discriminate between several Vibrio parahaemolyticus strains in replacement of PFGE or MLSA. Thus, these studies showed that MALDI-TOF represents a new promising alternative approach to other demanding conventional methods such as PFGE and MLSA for microbial subtyping.

Overall, the ultimate goal would be to use MALDI-TOF for a rapid prospective typing at the time of identification, which should significantly benefit to hospital epidemiology and to infection control measures that have to be applied to prevent dissemination of pathogens. Three recent studies have demonstrated that by applying subtle minor changes in the set-up generally used in routine diagnostics, MALDI-TOF MS allowed a reproducible discrimination of major MRSA lineages (Wolters et al., 2011), an identification of Salmonella enterica subsp. enterica serovars (Dieckmann & Malorny, 2011) and a differentiation between cfiA-negative and cfiA-positive Bacteroidis fragilis isolates (Wybo et al., 2011). These studies demonstrate that for several microbial species, minor changes in standardized procedures such as improved algorithm and user-friendly softwares applied in routine diagnostics will allow the use of MALDI-TOF MS for rapid and inexpensive microbial typing. This could significantly improve the approaches currently used to monitor epidemiological outbreaks and pathogen surveillance.

Conclusion

MALDI-TOF represents a very appealing new microbial identification technology that is efficient, rapid, cheap, and easy to use. This explains why MALDI-TOF MS can be successfully used in clinical diagnostic laboratory for microbial identification starting from subcultures on agar plates and broth media but also directly from positive blood cultures and to a lesser extent from clinical samples such as urine. The application of MALDI-TOF at the subspecies level in typing is promising but still needs further improvement including instrument sensitivity, database quality, and postrun analysis methods. Overall, a MALDI-TOF MS will be soon present in most diagnostic laboratories as, despite the significant cost of the instrument and for maintenance, running costs and consumables are much lower than those for other conventional methods, rendering this technology a worthy quantum leap tool.

Acknowledgements

We thank Myriam Corthesy for technical support. G.G. is supported by the Leenards Foundation through a career award entitled ‘Bourse Leenards pour la relève académique en médecine clinique à Lausanne’.

Ancillary