Metals associate with bacteria in several fundamental ways. Major constituents of the culture medium can be incorporated into cells to maintain osmotic and ionic balance, and trace metals can be assimilated as part of enzymatic cofactors. Metals may also be adsorbed to surfaces of cells during any part of an organism's growth history (3, 6) or be precipitated as a consequence of bacterial metabolism (4). In addition, B. anthracis and other spore-forming bacteria accumulate metals in the spore core, complexed to dipicolinic acid, during the sporulation process (21, 29, 30). This process is apparently correlated to heat resistance (10, 15, 26), and this phenomenon has led to interest in the metal content of bacterial spores related to food safety, general survivability in the environment, and astrobiology (1, 15, 18, 20-23).

The first report directly indicating that Bacillus spore metal content could be altered by the metal content of the growth medium dates back to 1959 (27). Under normal conditions, Ca is by far the most abundant metal found associated with the spore, potentially making up more than 3% of the spore weight (21). It is localized primarily in the spore core and to a lesser extent in the spore coat (25, 29, 30). Other metals are also associated with the bacterial spore core, and relative concentration is to some degree a function of the growth environment present during sporulation (8, 18, 21, 27).

Elemental characterization of biological materials is attractive from a microbial forensics standpoint because elemental fingerprints may indicate the type of growth medium in which the organisms were grown. Gikunju et al. (11) showed that elemental signatures could be used to differentiate between Bacillus thuringiensis cells, Bacillus globigii cells, and B. globigii spores by using inductively coupled plasma mass spectrometry. However, it is not clear in this instance whether the effect was due to differences in culture conditions or to some combination of culture conditions, species variation, and the developmental stage of the organisms studied. Furthermore, their method used 20 mg of material, as well as extensive sample preparation.

Time-of-flight secondary ion mass spectrometry (TOF-SIMS) (2, 33), which is used in this work, requires small sample sizes and minimal sample preparation and achieves good analytical sensitivity to facilitate culture medium source attribution. Under optimal conditions, TOF-SIMS is capable of achieving ppb to ppm detection limits for most elements and mass resolutions (m/Δm, where m is mass in atomic mass units) approaching 10,000, thus easily resolving most elemental and molecular ions of the same nominal mass. TOF-SIMS uses a focused primary ion beam with the capability of interrogating a single spore. Furthermore, the primary ion beam may be used to erode a sample and provide its depth profile and thus expose the core of the spore for analysis. TOF-SIMS has been used to differentiate Bacillus spores and vegetative cells based on molecular signatures of the cell surfaces, but no information regarding the growth media has been obtained (31).

Spore preparation. Four different culture media were used in this experiment: Lab Lemco agar (LL) (Fisher Scientific, New Brunswick, NJ), nutrient sporulating medium (NSM) (24), G medium (12), and a modified Schaeffer's medium (MSM) (Karen Fox, personal communication). Constituents of these media are as follows. LL (agar) consists of 23 g of Lab Lemco agar in 1.0 liter of distilled deionized water (DDI). NSM (agar) consists of a mixture of 3.0 g tryptone, 3.0 g yeast extract, 2.0 g Bacto agar, 23.0 g Lab Lemco agar, and 1.0 ml 1.0% MnCl₂ · 4H₂O in 1.0 liter of DDI. G medium (broth) consists of 1.0 g dextrose, 2.0 g yeast extract, 2.0 g (NH₄)₂SO₄, 0.005 g CuSO₄ · 5H₂O, 0.00025 g FeSO₄ · 7H₂O, 0.2 g MgSO₄ · 7H₂O, 0.05 g MnSO₄ · H₂O, 0.6 g K₂HPO₄, 0.4 g KH₂PO₄, 0.005 g ZnSO₄ · 7H₂O, and 0.08 g CaCl₂ in 1.0 liter of DDI. MSM (broth) consists of 0.1% KCl, 0.012% MgCl, 1.0 mM Ca(NO₃)₂, 0.01 mM MnCl₂, 0.001 mM FeSO₄, and 8 g of nutrient broth in 1.0 liter of DDI. NSM, G medium, and MSM were adjusted to pH 7.0 with 10 N NaOH or 2 N HCl as needed, prior to being autoclaved.

Vegetative starter cultures of B. subtilis strain ATCC 49760 were inoculated from frozen stocks into tryptic soy broth and incubated at 30°C for ~14 h on a rotary shaker at ~150 rpm. For LL and NSM agar, 150 μl of the vegetative cells was spread onto plates and incubated upside down for 3 to 5 days at 37°C. For G medium and MSM broths, 1.0 ml of vegetative starter culture was transferred into 225 ml of liquid medium. The cultures were incubated at 37°C at ~150 rpm for 5 to 7 days. Cultures were checked microscopically and harvested by washing them numerous times with 10 ml of sterile DDI when >95% spores were observed. When appropriate, all harvested cultures were centrifuged at 120 × g for 1 to 2 min, 3,000 × g for 1 to 2 min, and 14,600 × g for 6 to 7 min. This staged centrifugation protocol helped separate spores from vegetative cells and debris. Spore preparations were checked for purity by using phase-contrast microscopy and again during plate counting (after 12 h) on tryptic soy agar. The spore preparations were stored in DDI at 4°C until analysis, at which time 10-μl aliquots of each culture replicate were spotted onto high-purity graphite wafers (POCO Graphite, Decatur, TX) and allowed to dry at room temperature overnight prior to TOF-SIMS analysis. Graphite was used as a substrate to minimize potential elemental and polyatomic interferences.

TOF-SIMS analysis. TOF-SIMS was performed using an IONTOF IV instrument (IONTOF GmbH, Münster, Germany) equipped with an O₂⁺ sputter source and a polyatomic Au⁺ analysis source. Spore preparations were sputtered with a 2-kV O₂⁺ beam at a current density of about 3.3 × 10¹⁷ ions cm⁻² over a 200- by 200-μm area. This dose was shown in a separate experiment to correspond to a maximum response in signal from ⁴⁰Ca⁺ during sputter profile analyses (Fig. 1). Contact profilometry revealed that this sputter dose corresponds to a depth of about 75 ± 2.7 nm (mean ± 1 standard error, n = 3) in vitreous carbon and thus would presumably expose the core of a B. subtilis spore.

FIG. 1. Example of a 40Ca+ depth profile of B. subtilis spores. The primary O2+ sputter beam was operated at 2.0 kV and a nominal current of 600 nA over a 200- by 200-mm area. A 40- by 40-μm area centered in the sputter crater was used (more ...)

An electrodynamically bunched Au⁺ beam operated at 25 kV, 10 kHz, and 1.4 pA (~20 nA unbunched) was used for analyses in this study. Charge compensation was achieved by means of an 18-eV electron flood gun operated for about 80% of each 100-μs data acquisition cycle. Triplicate TOF-SIMS analyses were performed on each of three culture replicates for each of the four different media (36 total spectra). After O₂⁺ sputtering, high-mass-resolution analyses (m/Δm, ≥7,500 at the full-width half maximum measured at the ³⁹K peak) collecting data from 0 to 400 atomic mass units for 180 s were performed on a 40- by 40-μm area in the center of the sputter crater. The analysis ion beam dose was thus about 9.8 × 10¹³ ions cm⁻².

Data analyses. Mass spectra were dead time corrected and mass calibrated using the IONTOF software, exported as ASCII files, and integrated using a custom FORTRAN code using empirically derived integration limits (hereafter referred to as raw data). Visual inspections of the spectra were performed, and the following elemental peaks were chosen for further analyses based on their appearance in one or more of the cultures analyzed and on minimal isobaric interferences: ⁷Li, ¹¹B, Na, ²⁴Mg, Al, ²⁸Si, ³⁹K, ⁴⁰Ca, ⁵¹V, Mn, ⁵⁸Ni, ⁵⁶Fe, Co, ⁶³Cu, ⁶⁴Zn, ⁸⁵Rb, ⁸⁸Sr, and ⁹⁰Zr. Figure 2 shows selected regions corresponding to these elements of a representative spectrum of the B. subtilis spores grown in the G medium. TOF-SIMS analyses of the graphite substrate, under similar conditions to those used for spore analyses, showed only minor elemental ion counts of Na, Al, ²⁸Si, ³⁹K, ⁴⁰Ca, and ⁵¹V. ²⁸Si and ⁵¹V were excluded from further analyses, based on detection at the same order of magnitude in the graphite planchette as in samples containing spores. Aluminum was detected with an approximately 50-times-lower signal in graphite blanks than in spore samples. Sodium, ³⁹K, and ⁴⁰Ca were detected with signals approximately 3, 4, and 5 orders of magnitude lower, respectively, in graphite blanks than in samples containing spores. These contaminants in the graphite blank represent only a minor problem in cases where these elements may be of interest in discriminating growth medium types, in which case the problem may be circumvented by the use of a different substrate for analysis.

FIG. 2. Representative TOF-SIMS spectrum showing selected mass ranges of B. subtilis spores grown in G medium. Theoretical m/z values for the species of interest are indicated by filled triangles on the top axis. Note difference in scale on the y axis, particularly (more ...)

Raw data were analyzed using analysis of variance (ANOVA) and principal component analysis (PCA). TOF-SIMS ionization efficiency is highly variable, based on factors such as sample topography, analyte, and matrix chemistry. Therefore, quantitative analysis is typically performed by normalizing the element of interest to a matrix. For this reason, elemental signals were normalized to the ⁴⁰Ca⁺ signal prior to performing ANOVA. Data for ANOVA were log₁₀(x + 1) transformed, and all nine analytical replicates for each medium type were pooled. Tukey's honestly significant difference (HSD) test with a null rejection level (α) of 0.01 was used for multiple comparisons.

The data were analyzed using PCA in a variety of ways. Principal component analyses were performed on raw data, as well as on data normalized to Na, ³⁹K, and ⁴⁰Ca. Principal component analyses were performed without data preprocessing, after mean centering, and after autoscaling (14), and the magnitudes of the loading vectors were compared to provide an indication of the contribution of each element to the PCA scores.

Confusion matrices were constructed using nearest-neighbor classification (7). In particular, a single spectral measurement was omitted from the data set, preprocessing (none, mean centering, or autoscaling) was performed, and PCA scores were constructed from the remaining data. The omitted sample was then included as test data and classified using nearest-neighbor classification. This process was repeated for each sample, and the accuracy results were tabulated from the results of the classification. Confusion matrices that were constructed by including ²⁸Si and ⁵¹V (see above) were identical to those that were constructed by excluding those elements. Nevertheless, ²⁸Si and ⁵¹V were excluded from analyses presented here due to detection of their presence in the graphite substrate at a level similar to that found in spores.

We found the spore sample surfaces too irregular to acquire meaningful profilometry data. Scanning electron microscopy (SEM) images of the analyzed samples, however, reveal a more complicated interpretation than the simple removal of 75 nm of spore surface material. Figure 3 compares SEM images of undamaged spores (Fig. 3A) and an area of a bacterial spore sample that has been sputtered and analyzed (Fig. 3B). For help in interpretation, an area of spores sputtered at 4.7 × 10¹⁶ ions cm⁻² is also presented (Fig. 3C, inset). The spores that have received an ion current at an order of magnitude lower are discernible as damaged spores. It is apparent that although considerable cell material remains in the analysis crater, the spore morphology has been radically altered after analysis, leaving cone-shaped appendages. It is possible that this effect is an artifact caused by the uneven surfaces of the native spores. Using atomic force microscopy, Chada et al. (5) have shown that B. subtilis spores have surface irregularities on the order of 7 to 40 nm in size. On the other hand, this type of ion beam-induced damage is known to occur on flat surfaces and is thought to result from localized charging or from areas of surface species that are relatively recalcitrant to sputtering (28). From visual analysis, it is also apparent that more than 75 nm of material has been removed from the spores. Some of the discrepancy between the expected sputter rate based on that of vitreous C and the apparent sputter rate inferred from the SEM images may be due to differential sputter rates between organic C and vitreous C, but increased removal of material from sputtered particles in comparison to bulk samples has been noted before. For example, Gaspar et al. (9) noted an approximately 40-fold increase in sputter rate from NaCl particles over that of bulk NaCl and hypothesized that much of the difference may have been due to the effects of reduced redeposition of sputtered material along with increased surface area exposed to the primary ion beam. Finally, it is apparent that the secondary ion signal obtained from such topography is not only an integration of all inner and outer layers of the spores, but also that different individual spores contribute various relative components of inner and outer spore materials to the signal. Thus, we cannot say with certainty that a majority of the elemental signatures obtained from this analysis are derived from the spore core. Nevertheless, the information gained using this analysis protocol is sufficient to classify spores based on the medium in which they were prepared.

FIG. 3. Scanning electron micrograph images of B. subtilis spores before (A) and after (B) sputtering and TOF-SIMS analysis (3.3 × 1017 ions cm−2) and of spores that have received 4.7 × 1016 ions cm−2 (C, inset). Bar, 2 μm (more ...)

Table 1 presents mean ratios of the TOF-SIMS intensities for each element (normalized to ⁴⁰Ca⁺) ± standard errors for pooled analytical replications (n = 9) for each of the four culture types obtained during analysis. Also presented in Table 1 is the number of culture groups that the particular analyte ratio was capable of discriminating, based on ANOVA and using Tukey's HSD test with an α of 0.01 for multiple comparisons of the log₁₀(x + 1)-transformed data. It is evident from Table 1 that different elemental signals are important in separating different medium types. However, based on these criteria, no single elemental ratio was capable of clearly separating spores grown in all four media.

TABLE 1. Calcium-normalized elemental ratios and multiple-comparison groups predicted using Tukey's HSD test of TOF-SIMS analyses of B. subtilis spores grown in the four medium types

We chose to apply PCA to the data in an attempt to utilize the multidimensional nature of the elemental signatures. Principal component analysis was performed in three ways: without data preprocessing, with mean centering, and with autoscaling (14). We present an example of a score plot in Fig. 4A of ⁴⁰Ca-normalized data that has not been preprocessed. The score plot shows good visual separation of all nine analytical replicates for each of the four growth medium types, indicates good reproducibility, and suggests that PCA can be used to separate spore growth media. We present only loading score magnitudes for the first four principal components (PC 1 to 4) for the cases of no pretreatment (Fig. 4B to 4E) and for autoscaling of data (Fig. 4F to 4I). Relatively high loading scores indicate elements that contributed highly to a particular PC. As expected, only elements with relatively high normalized values contributed to the raw PCs (Table 1 and Fig. 4B to 4E). In contrast, the loadings from the autoscaled data indicate that elements with lower relative signal intensities also contributed heavily to PC scores used to discriminate between cultures when this form of data preprocessing was performed (Fig. 4F to 4I).

FIG. 4. Example of PCA score plot of 40Ca- normalized elemental data of triplicate TOF-SIMS analyses of triplicate growth replicates for each of four different media (A) and examples of PCA loading magnitude plots of raw (B to E) and autoscaled (F to I) data. (more ...)

For a demonstration of the validity of the approach, confusion matrices were constructed from leave-one-out cross validation by using the nearest-neighbor classification of the PCA scores (7). Table 2 presents the confusion matrices for ⁴⁰Ca-normalized data in which no preprocessing, mean centering, or autoscaling was performed prior to performing PCA. Each cell in Table 2 provides the number of samples grown in a given culture medium that were predicted to have been grown in each of the four media. For example, when no preprocessing was performed on the normalized data, eight MSM-grown samples were correctly predicted to have been grown in the MSM, while one MSM-grown sample was incorrectly predicted to have been grown in the G medium.

TABLE 2. Results of confusion matrices for nearest-neighbor classification of PCA of 40Ca-normalized TOF-SIMS elemental analysis of B. subtilis spores

The results from Table 2 indicate that this approach can be used to differentiate spores based on the medium in which they were grown. We present the outcomes of these analyses only for the ⁴⁰Ca-normalized case (Table 2); however, the nonnormalized, Na-normalized, and ³⁹K-normalized treatments had similar outcomes. With no preprocessing and mean centering, one sample at most from a given growth medium was misclassified. When autoscaling was performed, perfect classification was achieved. Given the limited analytical and culture replication in this study, we hesitate to infer from the statistical results which preprocessing method gives the best classification performance. However, we speculate that autoscaling will prove to be very effective because of its tendency to mitigate differences in overall element intensity in the PCA analysis. In particular, we expect that elements that have lower signal strength, but nevertheless are characteristic of a particular medium, may play a large role in the separation of media. For example, Cu is explicitly added to the G medium and is present with a higher normalized ratio in elemental signatures of spores grown in the G medium than in spores grown in the other media. Nonetheless, ⁶³Cu is detected at a ratio of only 10⁻² of that of Na in spores grown in the G medium and at an even lower ratio in spores grown in the other media. As a result, ⁶³Cu has a relatively small loading weight for PC 1 when the data are not autoscaled to remove effects of overall signal intensity (Fig. 4B). However, ⁶³Cu has a relatively high loading score for PC 1 in the autoscaled data (Fig. 4F) and may indeed be an important discriminator for this particular set of growth media. Because of the multivariate manner in which PCA builds components, testing this hypothesis would require precise manipulation of spore chemical content and is beyond the scope of this study.

An important criterion for judging the usefulness of a forensics technique is the amount of sample consumed. Using visual analysis of SEM images of NSM-grown spores and applying the formula of Loferer-Kröbbacher et al. (19), we found that these spores had a volume of 0.457 ± 0.009 μm³ (mean ± 1 standard error, n = 10). Thus, applying the dry density value for a B. subtilis spore of 1.44 g ml⁻¹ given by Tisa et al. (32), we were able to calculate an estimate of dry weight applied to the graphite planchette of about 2 mg. Further assuming an even coverage of 0.68- by 1.5-μm spores over a 40- by 40-μm area, a theoretical mass of about 1 ng of spores is required for each analytical replicate. Of course, manipulation of such small sample sizes on an analysis substrate is a logistics problem that will require further study.

From these analyses, we have shown that elemental signatures acquired using TOF-SIMS are useful for separating B. subtilis spores based on culture media, and the method may thus be applied to microbial source attribution in the future. We have focused our attention on positive elemental spectra taken after considerable sputtering to expose the spore core. A rich body of information is available in the negative spectrum as well, i.e., the halides, which we have chosen to exclude from these analyses for now. Moreover, TOF-SIMS spectra are typically comprised of thousands of organic peaks that may contain information of forensic value.

Although for convenience we applied about 2 mg of sample to the graphite disks, we calculate that an analytical replication would require only approximately 1 ng of spores, thus making this technique attractive in instances where sample amounts are limited. We are currently working to (i) optimize our protocol, (ii) extend our database to a wider range of media and organisms, and (iii) integrate this approach with other analytical techniques in order to more narrowly define spore source attribution. Further applications to astrobiology (to verify the origin of extraterrestrial spores by comparison of spore elemental signatures with that expected from their putative environment of origin) and food safety (to correlate spore survival with metal content of spores) are predicated.

This research was conducted under the Laboratory Directed Research and Development Program of the U.S. Department of Energy. A portion of the research described in the manuscript was performed at the W. R. Wiley Environmental Molecular Sciences Laboratory, a national scientific user facility sponsored by the U.S. Department of Energy's Office of Biological and Environmental Research, located at Pacific Northwest National Laboratory. The Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of Energy, under contract DE-AC05-76RLO1830. Support from the National Science Foundation (DMR-0216639) for the TOF-SIMS instrumentation at the University of Oregon is gratefully acknowledged.

Special thanks go to Shannon Goodwin and Brian Pultz for SEM analyses.