Beware of recent phishing e-mails. Use our official contact addresses only.

Note 81: Rapid Bacterial Chemotaxonomy By DirectProbe/MSD

By Eric D. Butrym* , Steven M. Colby* , Jackson O. Lay1 and Jon G. Wilkes1.

*Scientific Instrument Services, Ringoes, NJ

1National Center for Cancer and Toxicological Research, Jefferson, AR 72079, Ph: (501) 543-7108 Fax: (501) 543-7686

Presented at PittCon 99, Orlando, FL, March 1999


Recent developments using mass spectrometry for rapid bacterial chemotaxonomy show great promise for the application of these techniques to microbiological screening procedures in both routine and emergent situations1,2. One obstacle to the broader application of mass spectrometry to bacterial identification is the high cost of the research-grade instrumentation used in most of the work done to date. Here, a slightly modified HP 5973 Mass Selective Detector is used to assess 25 different strains of bacteria. The simplicity of direct-probe introduction makes this method attractive as a high-throughput screening technique. Samples are introduced via the direct exposure probe, and the ionization mode is 70 eV EI. Modification of the MSD includes adding the capability of elevating the source temperature to 350C, and the use of a filament designed for high temperature. Significant improvement in sensitivity at higher mass is the main benefit of these modifications. Gold-plated probe tips are used in the analysis to improve heat transfer and avoid catalysis. Collected data are analyzed with pattern-recognition software and are subjected to Canonical Variate analysis of components used for identification. Results show that direct probe/EI mass spectrometry with a low cost benchtop instrument can provide high quality spectra for use in bacterial identification.


The concept of Pyrolysis Mass Spectrometry (Py-MS) as a tool for taxonomic identification of whole bacterial cells based on distinct chemical constituents is not new.3 Recent studies have outlined techniques and methods for rapid, reproducible identification of bacteria using Py-MS with different ionization techniques and data handling procedures1,2,4. A common factor in previous work, as well as a major obstacle to wider implementation of the technique, has been the use of expensive, research-grade mass spectrometers. These instruments were needed for the high-quality spectra they yield which are rich in information in the higher (300 - 700 amu) mass range. With few exceptions, smaller benchtop instruments have not been able to produce adequate spectra due to resolution and sensitivity issues related to many of the same design considerations that kept their size and cost down. Improvements are continuously being made; however, and newer benchtop mass spectrometers are providing performance that meets or exceeds that of their research-grade counterparts of a few years ago. The purpose of these experiments is to implement Py-MS of whole bacterial cells on an inexpensive benchtop instrument and to show that excellent spectra can be obtained, which are suitable for correlation analysis and therefore applicable to previously described identification methods.

Materials and Methods

25 strains of bacteria (see Table 1) were cultured as described previously2 and pure colonies of each strain were selected and suspended in 75% ethanol until moderate turbidity was obtained. Sample suspensions were refrigerated when not in use, and were analyzed within 10 days of collection. Samples were sonicated for 1 minute prior to analysis to resuspend the cells.

The mass spectrometer used was a Hewlett-Packard (Palo Alto CA) model 5973 Mass Selective Detector. A modified repeller/source heater assembly as well as two high-temperature filament assemblies (SIS Inc., Ringoes NJ) were installed. These modifications, along with a minor adjustment to the MS ChemStation software, allow the operation of the MSD with ion source temperatures of up to 500ºC. The source was held at 350ºC for these experiments. The instrument set to scan from 50 amu to 750 amu at 2.14 scans per second. Ionization energy was 70 eV.

An SIS Direct Insertion Probe with ProbeDirectTM control software (SIS Inc.) was used for sample introduction and pyrolysis control. Copper tips plated with gold were used in place of the standard glass vials to enhance heat transfer to the sample and minimize catalytic degradation. One microliter of the sample suspension was placed on the probe tip for analysis. Each sample was allowed to dry before insertion into the interface, and remained in the interface for exactly 30 seconds while residual solvent vapor was removed. The probe temperature program was started at the moment when the probe tip contacted the ion source. For each sample, the probe was programmed to heat ballistically from an initial temperature of 50ºC to 350ºC. The average heating rate was approximately 1100 ºC/minute. The probe remained at maximum temperature for 5 minutes, and was then cooled for removal. All samples were analyzed at least twice, and some were replicated as many as four times. One of the 25 strains analyzed was not subjected to correlation analysis, because it yielded such radically different chromatograms that the spectra could not be averaged in the same way as the other samples. The discrepancy may be related to the condition of the probe tip. All replicates of the remaining samples were included in the correlation analysis.

Data were collected with MS ChemStation software (Hewlett-Packard, Palo Alto CA). A post-run macro served to perform consistent averaging of spectra, and save the resulting spectrum as an ASCII file. All spectra from 1.0 minutes to 3.0 minutes were averaged for each run. The selection of this time frame was based on comparison of replicate total ion chromatograms (See Figure 1). After an initial set of experiments using the entire range, data in the 200 - 650 amu range were selected for analysis. A custom program was written to convert the HP formatted file to one that can be read by the RESolve pattern recognition software (Colorado School of Mines, Boulder CO).

Table 1
Bacterial strain Group
Aeromonas hydrophila C
Aeromonas macrophila W
Bacillus licheniformis I
Bacillus subtilis Y
Bacillus thuringiensis H
Clavibacter michiganense N
E. coli 1090 A
Enterobacter cloacae B
Hot Springs Bacillus O
Klebsiella oxytoca F
Klebsiella pneumoniae G
Proteus mirabilis S
Pseudomonas aeruginosa b
Pseudomonas mendocina c
Pseudomonas putida a
Rathayibacter tritici L
Salmonella typhimurium X
Shigella flexneri V
Staph. aureus Z
Staph. auricularus E
Staph. epidermidis O
Staph. hominis K
Staph. warneri J
Staph.aureus mutant D
Strep. equinus T

Results and Discussion

Results of the pattern matching showed excellent agreement with earlier studies. An 89% cross validation was achieved, indicating that the program correctly matched the same samples with a high degree of precision. The samples that were incorrectly identified came from only three of the 25 classes. These samples were noted as having been run in extra replicates due to operator error or obvious irregularity in the observed chromatogram. Several interesting and unexpected results came from the analysis of this data set. The data were originally analyzed using a wider average spectrum from a larger time range. This was an attempt to use a simple method of extracting important information from all the samples without scrutinizing each one closely; an important aspect of any system developed to rapidly identify unknown strains of bacteria. Instead of merely assuring that meaningful information was included from each sample, this practice also introduced more random noise and resulted in a less accurate analysis (82% cross validation). This is highlighted by the fact that in the first analysis, 12 Principal Components (PC's) were calculated and accounted for only 97% of the total variation in the sample, whereas in the more narrowly selected data, only 5 PC's were needed to encompass 99% of the variation. The range for the second analysis was selected by overlaying the Total Ion Chromatograms from the replicate samples and selecting a region where ion abundance was similar. Additionally, a smaller range of ions was used in the average spectrum, facilitating the data file format conversion, and circumventing a limitation of the pattern-matching software that restricts the number of ions in a sample spectrum to 500 or less. In the original analysis, this number was achieved by eliminating from the spectrum a combination of ions with the largest and smallest abundance. It is suspected that valuable information can be lost by treating the data in this manner.

Figure 1. Three Total Ion Chromatograms of the Same Bacteria (E.coli 1090, Group A). Area used for averaging for the first analysis is indicated in green (the region highlighted in pink was subtracted as background). The second analysis used the average of spectra from the yellow region without background subtraction.

A noteworthy aspect of the first correlation analysis is that although the centour score is not 100% for every sample, failure to properly identify a sample as coming from a particular group occurred in a very predictable and manageable way. When the program mis-identified a sample, there were two possible configurations: all the values were reported as 0.000% and it just chose the wrong one from among bad lots, or two values were 100% and the program chose the wrong one, but the other one was the correct answer. These results indicate that faced with a true unknown, one could recognize whether the program was correct by comparing these probabilities. If a sample was identified as "X" and the probability was 0.000, one could either conclude that it wasn't "X", or one could make up a comparison set using several authentic replicates of "X" and several of the unknown and decide whether they were the same or not. If it was identified as "X" and there was a 100% probability that it could be "X" or "Y", it would be possible to set up a similar direct comparison between the unknown and authentic spectra of the two possibilities and reach a conclusion. With a carefully prepared spectral library all these experiments could be handled in software without further PyMS analyses.

Figure 2. Representative Averaged Spectra From Two Different Strains

Another striking aspect of the data is the high quality of the spectra in general, and particularly the preservation of high mass ions. Additionally, these spectra were obtained using one half to one third of the amount of sample used in our previous work. Ions are present well into the 600 - 700 amu range. The presence of these ions and the high sensitivity are attributable to the high temperature at which the ion source was operated. At higher temperatures, any material entering the ion source is less likely to condense on metal surfaces. It is, therefore, more likely to be ionized, and the result is a greater rate of transmission of material out of the source and into the analyzer. A cleaner source is also more stable, and provides better run-to-run and day-to-day reproducibility.

Figure 3. Results of the pattern recognition analysis including plots of the two canonical variates were used.

It should be noted that a significant source of variability in these analyses lies with the operator. While elaborate mechanisms have been developed to evenly coat pyrolysis wires and probe filaments with bacterial suspensions in order to achieve more reproducible spectra, here no more effort was expended in the preparation of the samples than assuring that most of the 1 ml sample was deposited on the tip of the probe.

Figure 4. 500 - 650 amu range From Figure 2a magnified


We have demonstrated the ability of an inexpensive, benchtop mass spectrometer to generate high quality data that can be used in the identification of bacteria by pattern recognition based on Py-MS. The modifications made to the instrument were minor, and resulted in excellent sensitivity and high-mass transmission. It is possible to improve the performance of the pattern recognition analysis by selecting appropriate areas of the TIC from which to draw the spectra used, as well as restricting the spectra to those peaks which contain significant information. The processes by which these selections are made can be automated. High quality spectra combined with automated data processing and affordable instrumentation will result in greatly expanding the base of potential users of this technique. Observations incidental to our analyses indicate that if good libraries are available, accurate identification can even be made from lower quality spectra without the need to re-run samples.


The authors wish to thank Pat Perkins at Hewlett-Packard for help with the high-temperature ion source conversion, and Manuel Holcomb at NCTR for culturing the bacteria and preparing the suspensions.


1.Wilkes J. G., Sutherland J., Rafii F., Voorhees K. J., Warden C., Hall R., Freeman J. P., and Lay J. O. Factors Which Affect The Reproducibility Of Pyrolysis Mass Spectra For Biological Samples. Poster presentation ASMS 1996.

2. Wilkes J.G., Holcomb M., Rafii F., Letarte S., Bertrand M.J., and Colby S. Probe Introduction/Mab/Ms For Rapid Bacterial Chemotaxonomy. Poster presentation ASMS 1998.

3. Gutteridge C.S., Norris, J.R. Journal of Applied Bacteriology 1979, 47, 5-43.

4. Barshick, S.A., Wolf, D.A., and Vass, A.A. Analytical Chemistry 1999, 71, 633-641.