Although considerable progress has been made to profile bacteria at the strain level with MALDI TOF MS, the approach faces several impediments to its broader and more widespread application. Library-based approaches have been limited by the fact that no single universal culture condition and sample preparation protocol has been widely adopted. Several have been proposed (Liu et al., 2007; Freiwald & Sauer, 2009). As described in an excellent review (Šedo, Sedláček, & Zdrahal, 2011) of sample-preparation methods used in MALDI profiling of bacteria, the use of diverse sample-preparation techniques has resulted in several discrepancies in the literature regarding the taxonomic limits. The reproducibility of the approach has also limited its broader application at the strain level. Reproducibility must be particularly high when comparing MS profiles of closely related bacteria that yield highly similar profiles. Similarity between replicates of the same bacterium (i.e., reproducibility) must exceed the similarity of profiles of closely related strains of bacteria. Unfortunately, very few studies (e.g., Ghyselinck et al., 2011) have rigorously quantified reproducibility. In addition, although MS profiles are often obtained via automation to leverage the high-throughput potential of MALDI TOF MS, consequent effects of automation on profile reproducibility and quality have only recently been explored (Schumaker, Borror, & Sandrin, 2012).
Bioinformatics-enabled approaches to MS profiling at the strain level might obviate many limitations associated with library-based approaches (Dieckmann et al., 2008); however, these approaches are limited by the fact that they tend to: (1) have significantly greater hardware and software requirements (i.e., top-down approaches, in particular, typically require more sophisticated and expensive TOF-TOF or other MS-MS capable instruments); (2) require more time, labor, and training (particularly those approaches that rely upon sample digestion and pre-fractionation); and (3) work best with bacteria for which complete genome sequences are available, although success with characterization on unsequenced bacteria with MALDI TOF MS (Teramoto et al., 2009; Fox et al., 2011) and other MS approaches, including an Orbitrap (Wynne et al., 2009; Wynne, Edwards, & Fenselau, 2010) have been described. While the requisite hardware and software to perform bioinformatics-enabled profiling at the strain level becomes more readily available and economical, these approaches may become more common. In addition, novel approaches to characterize bacteria without fully sequenced genomes as well as the ever-increasing number of bacteria with fully sequenced genomes will certainly facilitate more widespread use of bioinformatics-enabled approaches as has been suggested previously (Demirev & Fenselau, 2008a). Considering the more widespread application of library-based approaches to profile bacteria at the strain level, the following sections focus on challenges and limits associated with library-based approaches.
Sample Preparation and Culture Conditions
Broadly speaking, two approaches to prepare samples for analysis with library-based approaches have been employed: intact cell-based methods and cell extract-based methods (Tables 1–3). Intact cell-based methods place suspensions of intact cells on the MALDI target (Stackebrandt, Päuker, & Erhard, 2005; Moura et al., 2008; Schumaker, Borror, & Sandrin, 2012), whereas cell extract-based methods place cell extracts (supernatants) alone on the target (Vargha et al., 2006; Schmidt et al., 2009; Fujinami et al., 2011). Closer examination of sample-preparation methods reveals substantial variability in sample preparation (Šedo, Sedláček, & Zdráhal, 2011). Important sample-preparation features, including choice of matrix, have varied considerably (Tables 1–3). In many cases, multiple preparation methods were explored as an approach to maximize the taxonomic resolution of the method (Williams et al., 2003; Ruelle et al., 2004; Liu et al., 2007; Dieckmann et al., 2008); however, in other cases, a rationale to use a particular method was not provided. Interestingly, commercially available software products recommend different cell-preparation approaches. bioMérieux's SARAMIS recommends direct deposition of cells onto the MALDI target, whereas Bruker's BioTyper suggests a cell extract-based sample-preparation approach.
Most applications of MALDI TOF MS profiling of bacteria at the strain level have used cell extract-based sample preparation methods (Tables 1–3). In fact, one of the first reports of strain-level resolution employed a sample-preparation method that relied upon cell extracts (Krishnamurthy, Rajamani, & Ross, 1996). The authors used chemical and mechanical lysis techniques to prepare extracts that yielded strain-specific biomarkers of members of the genus Bacillus. Several subsequent studies reported that cell extract-based sample-preparation methods had also allowed strain-level resolution. For example, Nilsson (1999) reported strain-level resolution of H. pylori with cell extract-based sample-preparation methods. Perhaps not surprisingly, the authors noted that the choice of extraction solvent (either ACN/water or 0.1% TFA) dramatically affected which strain-specific biomarkers were detected. The authors also reported dramatic effects of matrix on peak detection. Horneffer et al. (2004) used wet-heat treatment to extract additional analytes to facilitate strain-level resolution of B. subtilis and B. cereus. Others have used enzymatic pre-treatment to facilitate more complete extraction of cell contents, often with trypsin (Krishnamurthy, Rajamani, & Ross, 1996) or lysozyme (Vargha et al., 2006; Giebel, Fredenberg, & Sandrin, 2008). More recently, Fujinami et al. (2011) relied upon cell extract-based sample-preparation methods to differentiate strains of Legionella. The authors used membrane-filtered cell extracts obtained by exposing cells to 1% TFA. This extraction approach was compared to one that involved bead beating with zirconia/silica beads. The membrane-filtered extracts yielded more useful strain-discriminating peaks than the extracts obtained with bead beating. Although there is no clear consensus on a universal cell extract-based sample-preparation method, an ethanol–formic acid extraction procedure (Sauer et al., 2008) has been recently employed by many (Barbuddhe et al., 2008; Ayyadurai et al., 2010; Dubois et al., 2010; Wensing, Zimmermann, & Geider, 2010; Ghyselinck et al., 2011).
Although strain-level resolution has been reported by many groups with cell extract-based sample-preparation methods, intact cell-based methods have also shown promise. In addition, intact cell-based preparation approaches are simpler and more rapid, because they do not require additional steps involved in chemical, enzymatic, and mechanical extraction often associated with cell extract preparation (Freiwald & Sauer, 2009). Several studies have reported strain-level resolution by depositing cells directly on the MALDI target. It should be noted that, although many studies placed intact cells directly onto MALDI targets and subsequently overlaid them with matrix and solvent mixtures (Walker et al., 2002; Jackson et al., 2005; Carbonnelle et al., 2007; Siegrist et al., 2007; Dieckmann et al., 2008), others suspended intact cells in matrix/solvent (often TFA and ACN) mixtures prior to deposition onto the target (Arnold & Reilly, 1998; Welham et al., 1998; Ryzhov, Hathout, & Fenselau, 2000; Dickinson et al., 2004; Donohue et al., 2006; Moura et al., 2008). Both approaches to intact cell-based sample preparation have yielded strain-level resolution; however, very few rigorous quantitative comparisons between direct deposition and suspension prior to deposition have been reported in the literature. Jackson et al. (2005), in their efforts to develop a standardized approach to profile MRSA at the strain level, presented a comparison of intact cell deposition (as colonies from agar plates) to deposition of suspended cells (in matrix solvent or water) onto the target. The authors found that direct deposition of strains from agar plates onto the MALDI target was more effective than suspending cells prior to deposition. Directly deposited cells yielded higher quality spectra that exhibited higher reproducibility, numbers of peak per profile, signal intensity, and S:N (Fig. 8).
Figure 8. Direct deposition of cells of MRSA onto the MALDI target (A) yielded spectra of higher quality than deposition of cells suspended in matrix solvent (B) or water (C). Adapted with permission from Jackson et al. (2005).
Download figure to PowerPoint
Despite extensive efforts to optimize a set of standard sample-preparation methods (Jackson et al., 2005; Liu et al., 2007), most attempts to profile MRSA and other bacteria at the strain level with MALDI TOF MS have not used standardized sample-preparation protocols. This lack of broad and consistent implementation of a standardized approach has resulted in conflicting reports in the literature. For example, inconsistencies in the literature on the ability of MALDI TOF MS to discriminate between strains of methicillin-resistant and methicillin-sensitive strains of S. aureus are common. Several studies reported that MALDI TOF MS can resolve strains of methicillin-resistant and methicillin-sensitive strains of S. aureus (Edwards-Jones et al., 2000; Du et al., 2002; Majcherczyk et al., 2006), whereas others have not reported such success (Bernardo et al., 2002; Walker et al., 2002; Jackson et al., 2005). These divergent results might have been caused, at least in part, by different sample-preparation methods.
As with sample-deposition methods, culture conditions have varied considerably across efforts to profile bacteria at the strain level with MALDI TOF MS. A variety of microbiological media, including solid agar and broth types, have been employed. The type of microbiological medium used has been widely reported to affect MS profiles of bacteria (Walker et al., 2002; Ruelle et al., 2004; Valentine et al., 2002), but some have suggested that effects of medium type on spectra are subtle and do not affect the overall ability of MALDI TOF MS to discriminate among bacteria (Conway et al., 2001; Bernardo et al., 2002; Vargha et al., 2006). Conway et al. (2001) reported that profiles of E. coli grown in two different broths exhibited 80% similarity. Given that strain-level profiling is more sensitive to minor differences in spectra associated with strain differences, the importance of an appropriate and consistent medium is likely far more important in strain-level applications than in species-level applications. For example, Horneffer, Haverkamp, and Janssen (2004) noted significant effects of medium type on MS profiles of spores of strains of B. subtilis. Rupf et al. (2005) attributed their success to obtaining strain-level resolution of mutans streptococci, in part, to their use of strictly controlled culture conditions. Similarly, in their efforts to discriminate strains of S. pyogenes, Moura et al. (2008) reported that medium type (blood agar or THB) affected MALDI profiles. Dieckmann et al. (2008) also noted medium-specific profiles and medium-dependent clustering (Fig. 9), but concluded that strain-specific biomarker peaks were mostly consistent across multiple media. In addition, such subtle changes did not prevent representatives of two subspecies, arizonae and houtenae, from clustering together. Similarly, Sauer et al. (2008) used media (Luria Broth) supplemented with one of two carbon sources, glycerol or glucose. Neither supplement had a significant effect on MS profiles or the ability to identify individual strains. Grosse-Herrenthey et al. (2008) reported similar, largely insignificant, effects of medium type on profiling Clostridia at the strain level. Taken together, differing results from these studies suggest that medium type has the potential to affect MALDI TOF MS profiles, but these effects might be bacterium-specific and limited. Potential effects should be thoroughly investigated and quantified before medium type is varied.
Figure 9. Medium type affected profiles of Salmonella enterica subspecies, but these effects did not prevent grouping of strains with respect to subspecies (Arizonae or Houtenae). Duplicate spectra are represented in a gel view format in which peaks and their respective intensities are represented by grayscale bands. PCA, plate count agar; SBA, sheep blood agar; MHA, Mueller–Hinton agar; MHB, Mueller–Hinton blood agar. Adapted with permission from Dieckmann et al. (2008).
Download figure to PowerPoint
In addition to culture-medium type, culture conditions (whether cells were grown on solid agar or in broth) have also been varied across attempts to profile bacteria at the strain level with MALDI. In many cases, cells cultured on solid media have been used (Rajakaruna et al., 2009; Pennanec et al., 2010). In others, broth cultures have been used (Moura et al., 2008; Schumaker, Borror, & Sandrin, 2012). Greater cell heterogeneity associated with plate cultures might explain these observations. Bacterial colonies consist of cells that vary in age, with the oldest cells found in the center of the colony and the newest colonies at the perimeter. In contrast, broth cultures tend to contain more homogenous populations of cells that are synchronized in their growth (Madigan et al., 2009). Additional research is warranted to determine the extent to which culture conditions affect MALDI TOF MS profiling of bacteria at the strain level. In addition, possible interactions among sample preparation approach (intact cells or extract), medium type, and culture conditions warrant investigation.
The assessment of effects of sample preparation and culture conditions as well as overall method reliability and efficacy requires rigorous quantification of reproducibility. Such assessment is particularly true at the strain level because often only very minor differences in MS profiles facilitate resolution of closely related bacterial strains. Unfortunately, no standardized approach to quantify and report reproducibility has been widely implemented. In addition, the term reproducibility has been used with different meaning by different authors. Rather infrequently, reproducibility has been used to refer to the accuracy with which strains were identified (Mellmann et al., 2009). Most commonly and in this review, though, reproducibility refers to how similar (or different) replicate spectra of the same strain are to one another with regard to peak presence/absence and often peak intensity (Giebel, Fredenberg, & Sandrin, 2008; Moura et al., 2008; Pennanec et al., 2010).
Several studies have reported reproducibility based upon visual inspection of spectra (Arnold & Reilly, 1998; Jackson et al., 2005; Lui et al., 2007; Wolters et al., 2011). Several efforts to more rigorously quantify reproducibility relied upon the root mean square (RMS) of replicate spectra (Keys et al., 2004; Majcherczyk et al., 2006; Rajakaruna et al., 2009). Freiwald & Sauer (2009) described the coefficient of variation (CV) of replicate spectra, but this approach requires specialized and proprietary software (ClinProTools; Bruker Daltonics). An alternative approach, not reliant upon specialized software (only MATLAB), that employed a combined analysis of variance (ANOVA)-principal component analysis (PCA) has also been proposed, but not widely adopted (Chen, Lu, & Harrington, 2008). Our group has quantified reproducibility using similarity coefficients of spectra calculated using the following Pearson correlation coefficient (r) in which xi and yi represent intensity values of peaks in two profiles, x and y, and n represents the number of peaks in each profile (Giebel, Fredenberg, & Sandrin, 2008; Devore, 2012; Schumaker, Borror, & Sandrin, 2012):
As with another approach (Freiwald & Sauer, 2009), though, these analyses are facilitated by proprietary and rather specialized software (BioNumerics; Applied Maths; Sint-Martens-Latem, Belgium). Indeed, a standard approach to quantify reproducibility that uses tools widely available to investigators will be required to enable future meaningful comparisons of reproducibility reported by different studies.
Although relatively few studies have rigorously quantified reproducibility, even fewer have implemented measures to ensure that reproducibility was at an appropriate threshold level to allow reliable resolution of bacterial strains. In this regard, the work by Rupf et al. (2005) is exemplary in its approach to quantify threshold parameters, σ1 and σ2, which quantify the similarity of replicate spectra and spectra from different strains, respectively, as described above in Section II.C. (Fig. 7). Broad implementation of a similar approach seems critically important to demonstrate the reliability of library-based methods to profile bacteria at the strain level.
Rigorous and quantitative assessments of reproducibility will likely be critically important as applications of MALDI TOF MS-based profiling of bacteria at the strain level rely more upon automated data acquisition. Automated data acquisition has been proposed to increase reproducibility in comparison to manual data acquisition, in which different operators collect spectra (Freiwald & Sauer, 2009). Most modern mass spectrometers are bundled with instrument operation software that facilitates fully automated acquisition of spectra. Instrument operators need only to specify minimum spectrum quality criteria (e.g., base peak resolution, S:N, etc.). As such, automated data acquisition should facilitate more objective acquisition than manual acquisition by a human operator. The potential benefits of automated data acquisition to throughput and reproducibility have made this type of data collection common (Seng et al., 2009; Cherkaoui et al., 2010; van Veen, Claas, & Kuijper, 2010; De Bruyne et al., 2011). A recent report, though, suggests that data-acquisition automation can reduce MS profile quality and reproducibility (Schumaker, Borror, & Sandrin, 2012; Fig. 10). Reproducibility was lower when spectra were acquired using automation (Fig. 10A) than when spectra were acquired by either an experienced (Fig. 10B) or a more novice (Fig. 10C) operator. These findings suggest that effects of automation on reproducibility must be considered, particularly at the strain level, where minor differences in profiles can have profound effects on the ability of the method to reliably resolve closely related bacterial strains. In addition, efforts should be made to ensure that parameters used in automation maximize reproducibility and spectrum quality.
Figure 10. Mode of data acquisition affected reproducibility of MALDI-MS profiles of several bacteria. Similarities of replicate spectra acquired via automation (A), an operator with 2 years of experience (B), and an operator with 6 months of experience (C) are represented by multidimensional scaling (MDS) analysis. Adapted with permission from Schumaker, Borror, and Sandrin (2012). [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com]
Download figure to PowerPoint