Michael E. Dolch, Department of Anesthesiology, University Hospital Großhadern, Ludwig-Maximilians-University of Munich, 81366 Munich, Germany. E-mail: email@example.com
Fast and reliable methods for the early detection and identification of micro-organism are of high interest. In addition to established methods, direct mass spectrometry–based analysis of volatile compounds (VCs) emitted by micro-organisms has recently been shown to allow species differentiation. Thus, a large number of pathogenic Gram-negative bacteria, which comprised Acinetobacter baumannii, Enterobacter cloacae, Escherichia coli, Klebsiella oxytoca, Pseudomonas aeruginosa, Proteus vulgaris and Serratia marcescens, were subjected to headspace VC composition analysis using direct mass spectrometry in a low sample volume that allows for automation.
Methods and Results
Ion-molecule reaction–mass spectrometry (IMR-MS) was applied to headspace analysis of the above bacterial samples incubated at 37°C starting with 102 CFU ml−1. Measurements of sample VC composition were performed at 4, 8 and 24 h. Microbial growth was detected in all samples after 8 h. After 24 h, species-specific mass spectra were obtained allowing differentiation between bacterial species.
IMR-MS provided rapid growth detection and identification of micro-organisms using a cumulative end-point model with a short analysis time of 3 min per sample.
Significance and impact of the study
Following further validation, the presented method of bacterial sample headspace VC analysis has the potential to be used for bacteria differentiation.
Infectious complications are a frequent problem in critically ill patients resulting in significant morbidity and mortality. Infection leads to an increase in mortality rate from 11% in noninfected patients to 25% in patients with infection (Vincent et al. 2009). Furthermore, in septic shock patients, a delay in initiation of effective antimicrobial therapy is associated with increased mortality (Kumar et al. 2006). Infections were found to be of respiratory origin in 64% of intensive care unit patients, and 62% of all positive microbial isolates were Gram-negative organisms (Vincent et al. 2009). However, despite the need for rapid microbial species and strain identification, currently applied standard microbiological methods are still time consuming, requiring from hours to days. Thus, the current search for rapid methods enabling microbial species and strain identification has renewed the interest in the analysis of volatile compounds (VCs) released by micro-organisms (Kuzma et al. 1995; Scholler et al. 1997; Allardyce et al. 2006; Bunge et al. 2008; Zhu et al. 2010).
In 1966, Mann et al. identified 2-aminoacetophenone (2AA) as the compound responsible for the grape-like odour of Pseudomonas aeruginosa using thin-layer chromatography (Mann 1966). However, as the concentration of VCs released by micro-organisms ranges in the low nmol l−1 range, analysis was limited to a few molecules with sufficiently high concentrations until the advent of gas chromatography coupled with MS (GC-MS). For instance, GC-MS was applied to headspace analysis of different Pseudomonas strains (Labows et al. 1980), to the differentiation of Penicillium clavigerum and P. vulpinum (Larsen and Frisvad 1994), to the identification of isoprene in the headspace of Acinetobacter calcocaceticus, Bacillus spp., Escherichia coli, P. aeruginosa and P. citronellolis (Kuzma et al. 1995) and for profiling of Pseudomonas spp., Serratia liquefaciens, as well as Enterobacter cloacae (Scholler et al. 1997). However, GC-MS analyses are typically handicapped by the need for sample preconcentration, which is time consuming and results in extended turnover times. These disadvantages can be bypassed by direct mass spectrometric (MS) techniques, allowing straightforward analysis of microbial headspace VC composition (Mayr et al. 2003). Recently, headspace analysis of Staphylococcus aureus, Candida tropicalis, E. coli, Salmonella enterica and S. flexneri was carried out using proton transfer reaction MS (PTR-MS) (Bunge et al. 2008; O'Hara and Mayhew 2009). Allardyce et al. (2006) applied selected ion flow tube MS (SIFT-MS) to headspace analysis of blood cultures spiked with E. coli, Neisseria meningitides, P. aeruginosa, Staphylococcus aureus and Streptococcus pneumoniae. Secondary electrospray ionization MS (SESI-MS) was used to analyse the headspace of liquid cultures of E. coli, P. aeruginosa, S. aureus and Salmonella enterica serovar Typhimurium (Zhu et al. 2010). However, so far only a limited number of either Gram-negative or Gram-positive micro-organisms were analysed and directly compared with each other.
This study applies ion-molecule reaction–mass spectrometry (IMR-MS) to the analysis of microbial headspace VC composition for bacterial species differentiation. IMR-MS uses soft chemical ionization for sample molecule ionization and thereby features no or only minimal fragmentation (Bassi et al. 1998). The technique is highly sensitive down to the nmol l−1 range and has the capability of measuring compounds within milliseconds (Hornuss et al. 2007; Dolch et al. 2008). The study is part of a larger programme aiming to evaluate the suitability of VC headspace analysis as an in vivo diagnostic tool for numerous infections. Therefore, Gram-negative micro-organisms with a high prevalence in patients with nosocomial pneumonia were specifically selected for analysis (Esperatti et al. 2010). Our efforts were directed towards the identification of specific VC ‘fingerprints’ for a given species rather than towards the identification of VCs produced by the bacteria. Furthermore, the methodological approach chosen was designed to allow for close proximity to the automation of the analytical process.
Materials and methods
Micro-organisms, sample preparation and growth conditions
After preparation, the samples were transferred into an incubator at 37°C and agitated until their use for analysis. Headspace measurements determining the presence of VCs were carried out after 15 min, 4, 8 and 24 h. Vials were equilibrated at a temperature of 37·0°C in the headspace sampler (G1888; Agilent technologies, Santa Clara, CA, USA), and the gaseous headspace sample was introduced into the IMR-MS via a heated transfer line. At each point in time, two technical replicates of control and bacterial culture samples were measured. Assessment of appropriate initial micro-organism content of the stock solution (15 min), micro-organism growth characteristics (4, 8 and 24 h), as well as the sterility of corresponding control vials, were routinely determined on growth plates using 1·5% (w/v) LB at any time.
Ion-molecule reaction–mass spectrometry
The mass spectrometric system used in this study was originally designed to measure trace gas components in the industrial environment (Airsense Compact.net; V&F Analyse-und Messtechnik GmbH, Absam, Austria). A detailed description of the equipment has been published previously (Hornuss et al. 2007; Dolch et al. 2008; Netzer et al. 2009). In brief, the MS system combines two mass spectrometric techniques, a conventional electron impact mass spectrometer (EI-MS) for the detection of high, for example, mmol l−1-concentrations and an ion-molecule reaction–mass spectrometer (IMR-MS). The latter provides a highly sensitive method for online and offline sampling of organic and inorganic compounds and has already been used to determine volatile compounds in exhaled breath (Morselli-Labate et al. 2007; Dolch et al. 2008; Bennett et al. 2009; Millonig et al. 2010; Hornuss et al. 2012). In IMR-MS, positively charged atomic ions interact with neutral sample gas molecules. Two-body collision processes result in the formation of product ions whenever the ionization potential of the sample molecule is less than the potential energy of the incoming primary ion. Differences in ionization potentials between primary and product ions may result in a bond rupture and hence a lower molecular weight fragment ion. However, owing to the soft ionization process, fragmentation is typically avoided.
The IMR-MS can use krypton, mercury or xenon gas to form the primary ion beam via electron-impact ionization. In our experiments, mercury ions were used as primary ions. The IMR-MS mass separation is 1 amu over the mass range, cycle time between masses is 2 ms on average, and the response time for detection is <100 ms. The concentration drift in signal (measured against a steady test gas concentration) is below 5% over 12 h. Previously published limits of detection for reference test gases were as follows: acetaldehyde 0·08 nmol l−1, acetone 0·15 nmol l−1, ethanol 0·27 nmol l−1 and isoprene 0·10 nmol l−1 (Dolch et al. 2008). Between sample measurements, the IMR-MS lines were flushed with N2 (purity 5·0) to avoid carry-over effects. To compensate for any changes in IMR-MS sensitivity, a reference gas containing 39·21 nmol l−1 of isoprene in nitrogen 5·0 was measured as a standard prior to each vial measurement, and the respective intensity was used for signal normalization during the subsequent analysis.
Based on previous experience with a series of preliminary measurements of VCs produced by Gram-negative bacteria, the following masses were selected for analysis with a dwell time of 300 ms per mass: M17, M19, M29, M34, M35, M36, M37, M38, M39, M40, M41, M42, M43, M44, M45, M46, M47, M48, M49, M50, M53, M54, M55, M56, M57, M58, M60, M61, M62, M63, M64, M65, M66, M68, M69, M70, M72, M74, M75, M76, M77, M79, M82, M84, M86, M87, M88, M89, M90, M92, M93, M94, M95, M96, M97, M98, M100, M102, M105, M106, M107, M112, M117, M123 and M135. The likely identity of masses M17 (NH3), M34 (H2S), M42 (propene), M43 (acetyl group), M44 (acetaldehyde), M45 (ethanol), M48 (methanethiol), M54 (butadiene), M57 (propionaldehyde), M58 (acetone), M60 (propanol), M68 (isoprene), M92 (toluene) and M117 (indole) was confirmed by comparing the IMR-MS spectra of chemical compounds (Sigma-Aldrich, Germany) with those obtained from sample analysis. Simultaneous measurements of CO2-concentrations were performed using the EI-MS facility available in the Airsense Compact.net system.
Processing the IMR-MS raw data included several steps. The IMR-MS raw signals – given in counts per second (cps) – were normalized to the corresponding isoprene signal in reference gas for each vial. Using these normalized spectra, the percentage deviations for each technical replicate from its corresponding broth solution signal were calculated. Finally, the average intensity of all six technical replicates per species was calculated. These data were used during subsequent analyses. Relative changes in average signalling in excess of 100% were considered as of micro-organism origin. Thus, the identified bacteria-specific relative mass signals were used in a presence–absence manner for principal component analysis (PCA). For visualization of the specific VC fingerprints, the average spectra of relative signal changes were logarithmized. In case of negative deviations of signalling, the absolute value was logarithmized and subsequently set to negative again. Statistical analysis was carried out using the software package R 2.12.2 (R Development Core Team 2010).
Figure 1 gives a representation of bacterial growth in vials after 15 min, 4, 8 and 24 h of incubation. The number of incubated bacteria after 15 min ranged from 1 × 102 for E. coli to 5 × 102 for S. marcescens, both within the targeted range of 102 CFU ml−1. After 4 and 8 h of incubation, A. baumannii, E. cloacae, E. coli and K. oxytoca were growing more rapidly than the other species. However, at the end of the incubation period, most cultures reached a bacterial count number of 109–1010 CFU ml−1. No bacterial growth was detectable in control vials at 15 min, 4, 8 and 24 h.
After 4 h of incubation, the relative mass signal intensities for all species remained below the predefined significance threshold of more than 100%. In Figure 2, the relative VC signalling intensity of all micro-organisms after 8 and 24 h of incubation is given in a colour-coded image plot. For all species studied, increases in mass signalling were ascertainable after 8 h of incubation. VCs detectable at this stage were CO2 in six of seven species with an increase at about two orders of magnitude, M17 (NH3) in five of seven species with an increase at about two orders of magnitude, M48 (methanethiol) in six of seven species with an increase up to four orders of magnitude, M50 in five of seven species with an increase above two orders of magnitude and M70 in six of seven species with an increase in the range of two orders of magnitude. Furthermore, even at this early stage, drops in the headspace VC composition emerged for A. baumannii, K. oxytoca and P. aeruginosa at M19, and for A. baumannii, E. cloacae, K. oxytoca and S. marcescens at masses M105, M106 and M107. After 24 h of incubation, multiple signals were assignable to distinct bacterial species (Figures 2 and 3). CO2, M17 (NH3), M48 (methanethiol), M50 and M86 were present in all seven micro-organisms studied, and CO2 concentration showed the largest intensity increase by 3–7 orders of magnitude. Indole, located at m/z 117, gives a clear signal in E. cloacae, E. coli, K. oxytoca, P. aeruginosa, P. vulgaris and S. marcescens. The initial decrease in signalling noted on masses M19, M37, M38, M39, M105, M106 and M107 after 8 h for some micro-organisms progressed after 24 h of growth to significant intensity losses in the mass spectra of all studied bacteria.
Using the first three principal components (PC 1, PC 2 and PC 3) resulting from PCA on the relative VC signalling pattern, a clear separation between the different species of most micro-organisms was achievable (Figure 4). Either combination of PC 1, PC 2 and PC 3 discriminated equal between A. baumannii, E. coli, P. aeruginosa, P. vulgaris and S. marcescens. In case of E. cloacae and K. oxytoca, the clusters showed a near total overlap. However, these two bacterial species become discernible when minor differences within their mass spectra – namely the inability of E. cloacae to produce the masses 29, 60 (propanol) and 79 – were taken into account in a separate PCA (Figure 5).
The application of IMR-MS to the analysis of the headspace VC composition of seven Gram-negative bacteria enabled us to identify specific mass spectra for each micro-organism studied. Using a standardized model, characterized by close control of the number of viable micro-organisms and growth conditions, mass spectra of all biological and technical replicates showed a high degree of repeatability and reliability. Processing the data by PCA allowed us to differentiate between A. baumannii, E. coli, P. aeruginosa, P. vulgaris and S. marcescens as well as E. cloacae and K. oxytoca. Furthermore, our data show that VCs are not only released during micro-organism growth but also depleted in some cases. A finding that might be either explained by constituents undergoing chemical reactions increasing low-polarity compound solubility because of bacterial surfactant secretions or by bacterial metabolization (Schulz and Dickschat 2007). This finding could be used in further studies aiming to investigate microbial metabolism and nutrient broth interaction to differentiate between species on the basis of headspace VC analysis.
Identification of mass spectra on species level formed the basis for PCA analysis and micro-organism differentiation in this and other studies applying direct mass spectrometric methods (Bunge et al. 2008; Zhu et al. 2010). In accordance with these previous findings, this study demonstrates the suitability of IMR-MS for headspace VC profiling of micro-organisms and species differentiation. Performing PCA on bacteria-specific mass spectra obtained by IMR-MS analysis resulted in close specific cluster formation underlining the high repeatability and reliability of the analytical method applied. Five of the seven Gram-negative bacteria investigated were clearly separable from each other and from E. cloacae and K. oxytoca. These latter two species were only insufficiently separated by PCA, which is a consequence of the broad analogies in their mass spectra. E. cloacae mass spectra differed from K. oxytoca by the deficiency of masses 29, 60 (propanol), as well as 79, and using this information in a separate PCA, E. cloacae and K. oxytoca became clearly distinguishable. Recently, online monitoring of microbial VC production by PTR-MS allowed the separation between E. coli, Salmonella enterica, Shigella flexneri and C. tropicalis after 24 h of incubation (Bunge et al. 2008). Applying a metal oxide–based olfactory sensor to continuous monitoring of micro-organisms' headspace ‘odour’ enabled the differentiation between Clostridium difficile, E. cloacae, Enterococcus faecalis, E. coli, Klebsiella spp., Proteus mirabilis, P. aeruginosa, Salmonella spp. and S. aureus after 8 h of growth (Bruins et al. 2009). Using SESI-MS, Zhu and co-workers were able to distinguish between cultures of P. aeruginosa, S. aureus, E. coli and S. enterica with serovar Typhimurium and serovar Pullorum after 24 h of growth (Zhu et al. 2010).
Time to detection of bacterial growth in blood culture bottles inoculated with 5 or 102 CFU ml−1 of Gram-negative or Gram-positive bacteria occurred between 10 and 23 h for the commonly used automated blood culture system, whereas the detection of VCs by SIFT-MS was accomplished in 88·3% after 8 h and 96·6% after 24 h (Scotter et al. 2006). Using the same experimental set-up with the inoculation of 10 CFU ml−1 at 6 h, a microbial VC pattern was detectable comparable with that obtained after 24 h of incubation (Allardyce et al. 2006). After 24 h of incubation and a final cell number of approximately 109 CFU ml−1, SESI-MS–based headspace analysis allowed for species differentiation (Zhu et al. 2010). In contrast to these cumulative end-point models for VC analysis, two studies used continuous VC monitoring during growth (Bunge et al. 2008; Bruins et al. 2009). Starting with initial cell numbers ranging from about 5 × 107 to 5 × 108 CFU ml−1, species differentiation was achieved after 6–8 h by means of an oxide-based olfactory sensor with a diagnostic specificity ranging from 67 to 100% (Bruins et al. 2009). Similar results were obtained after 24 h of incubation employing the PTR-MS technique (Bunge et al. 2008). Both studies noted that VCs emitted by micro-organisms underwent significant temporal changes that facilitated species differentiation. For our experiments, a cumulative end-point model with analyses after 4, 8 and 24 h of growth was used. With an initial cell number of ~2·5 × 102 CFU ml−1, significant mass signalling (above the predefined relative 100% threshold) and thus detection of bacterial growth was possible for all seven species, which is comparable to the results obtained with inoculated blood culture bottles (Allardyce et al. 2006; Scotter et al. 2006). Species differentiation was possible using the headspace mass spectra obtained after 24 h of growth. This is in contrast to the time of 6–8 h noted by Bruins et al. but is satisfactorily explained by our six orders of magnitude lower initial cell inoculum, which approximates cell numbers present in clinical specimens. Undoubtedly, we lost information regarding the temporal evolution of VC emission using a cumulative end-point experimental set-up. However, our approach showed advantages in terms of growth detection and species differentiation by being far less complex than a continuous analytical approach and its ability to measure approximately 20 samples per hour with a single-vial analysis time of about 3 min.
For our experiments, LB was chosen as broth solution. During previous experiments in our laboratory, LB provided excellent growth conditions and VC signalling for the micro-organisms, as is the case in this study here. The selection of different types of broth solution had clear effects on production and temporal evolution of VCs (Bruins et al. 2009; O'Hara and Mayhew 2009). Recent studies used either native (Bruins et al. 2009) or blood-spiked standard blood culture bottles (Allardyce et al. 2006), German Collection of Microorganisms and Cell Cultures (DSMZ) media no. 1, 186 and 681 (Bunge et al. 2008) and tryptic soy broth (Zhu et al. 2010). Therefore, these approaches consisted of either providing a standardized nutrition broth for all micro-organisms (Allardyce et al. 2006; Bruins et al. 2009; Zhu et al. 2010) or the use of specific strain-optimized nutrition broths (Bunge et al. 2008). Because both techniques allowed microbial VC profiling and species differentiation, the superiority of one approach over the other has not been convincingly demonstrated.
Using IMR-MS, we were able to perform direct comparisons between the emitted VC spectra of A. baumannii, E. cloacae, E. coli, K. oxytoca, P. aeruginosa, P. vulgaris and S. marcescens. The obtained mass signals facilitated early detection of bacterial growth already after 8 h. Species differentiation, characterized by a high repeatability and reliability, was achieved on the basis of VC mass spectra obtained after 24 h of growth. Given the short analysis time of 3 min per vial, many samples can be processed rapidly and even automation seems to be achievable with the presented set-up. Currently no such diagnostic tool exists, but we certainly hope it will in the future allowing for species differentiation even outside the usual working hours, and thus shorten the time required for adaptions of antibiotic treatment in response to microbial testing results.
The authors are grateful to Kirsten Weinert for her excellent technical assistance during the study period. The authors would like to thank Dr Luca Ciaffoni (Physical and Theoretical Chemistry Laboratory, University of Oxford) for his support in generating graphical representations. Avacta/Oxford Medical Diagnostics Ltd provided financial support, and V&F Medical Development GmbH provided the mass spectrometry system.
Conflict of interest
MD received an indirect research grant from Avacta/Oxford Medical Diagnostics Ltd. V&F reimbursed the travel expenses of M.D. and C.H. to the ASA annual meetings and the German Anaesthesia Congress 2006–2010. S.P. is a scientist employed by V&F. J.V. is owner of V&F. At the time of the study, W.D. was employed by Oxford Medical Diagnostics Ltd/Avacta Group plc. C.K., G.S. and S.S. have no conflict of interest to declare.