Strain-Level Differentiation of Bacteria by Paper Spray Ionization Mass Spectrometry

Abstract

Paper spray ionization-mass spectrometry (PSI-MS) is a relatively new analytical technique allowing for rapid mass spectrometric analysis of biological samples with little or no sample preparation. The expeditious nature of the analysis and minimal requirement for sample preparation makes PSI-MS a promising avenue for future clinical assays with one potential application in the identification of different types of bacteria. Although past PSI-MS studies have demonstrated the ability to distinguish between bacteria of different species and morphological classes, achieving within-species strain-level differentiation has never been performed. In this report, we demonstrate the first strain-level bacterial differentiation by PSI-MS with the mammalian intestinal bacterium Oxalobacter formigenes (Oxf). This novel application holds promising clinical significance as it could be used to differentiate between pathogenic bacteria and their harmless commensal relatives, saving time and money in clinical diagnostics. Both whole cells and cell lysates of Oxalobacter strains HC1 and OxWR were analyzed using the Prosolia Velox™ 360 PSI source coupled to a Thermo Scientific Q Exactive high-resolution mass spectrometer with a rapid 30-sec analytical method. Multivariate statistical analysis followed by examination of significant features provided for and confirmed differentiation between HC1 and OxWR. We report a panel of strain-exclusive metabolites that could serve as potential strain-indicating biomarkers.

Graphical Abstract

Paper spray ionization-mass spectrometry (PSI-MS) is an ambient analytical technique that involves the direct electrospray analysis and measurement of proteins, metabolites, or lipids in biological samples on a porous substrate, often cellulose paper¹. A solvent is applied to the substrate with subsequent electrospray ionization for rapid, direct mass spectrometric analysis. The main benefits of PSI-MS are the small sample volume, minimal requirements for sample preparation and expedited analysis time². Typical methods range from 30 sec to 2 min, far less than the runtimes of liquid chromatography-mass spectrometry (LC-MS) based methods, which can extend beyond 20 min per injection for some applications³. The efficiency and user-friendly nature of PSI-MS has generated significant interest in its clinical application potential⁴, one example being bacterial identification⁵. In the clinical setting, the diagnosis of bacterial infection is often initially made inexactly by phenotypic examination of the patient or general tests for inflammation or high white blood cell count⁶, leading to the prescription of general antibiotics that damage the diversity of the microbiome⁷. When an exact determination of the causative agent of infection is necessary, further investigation is conducted to identify the pathogen. Among other methods, two of the most common approaches are standard culture techniques and serological antigen tests. Clinical cultures can take as long as 48–72 hours and can easily become contaminated with poor sampling techniques⁶. Additionally, there are many pathogens which cannot be cultured using routine methods^6,8. Serological antigen tests are usually rapid, but lack in sensitivity and cannot always conclusively rule out infection⁶. They also often have low specificity due to off-target reactivity with other antigens and can generate false-negative results if the infection is in an early stage^6,9. These issues spanning from cost to assay confidence to time requirements indicate the need for a rapid, reliable assay for the differentiation and identification of bacteria. To address this need, various MS-based applications have been investigated in regard to bacterial identification as early as 1975^10–13. Currently, matrix-assisted laser desorption ionization (MALDI) is used as a clinical MS method for bacterial analysis on the basis of protein biomarker detection ¹⁴. MALDI-MS shares many benefits with PSI-MS in being rapid, economical, and sensitive, although one advantage of PSI-MS is perhaps the absence of sample preparation compared to MALDI-MS, which requires proper application of a matrix to the biological sample for reproducible analysis^2,15. Only in the last decade has PSI-MS been applied to bacterial identification. Previous PSI-MS applications have demonstrated the ability to discern between bacteria of different species and morphological classes (Gram-positive versus Gram-negative)⁵. However, there are many species of bacteria where only one strain or a small percentage of strains within that species are pathogenic¹⁶. A well-known example is Escherichia coli (E. coli), a species with many strains that exist naturally in the intestine, but which contains a subset of pathogenic strains such as the notorious O157:H7, which causes over 73,000 illnesses per year in the United States alone^17,18. Identification of bacteria at the strain level often requires expensive and time-consuming genomic methods, and there is a great need for improved techniques in this area of diagnostic medicine^19,20. A PSI-MS method with the capability to distinguish between common pathogens and their same-species nonpathogenic relatives could revolutionize the diagnosis of infectious disease by providing a rapid, routine pipeline to determine the cause of infection. The cornerstone of such an assay would be the detection of strain-exclusive biomarkers for confident identification of a specific strain within a specific species of bacteria. To date, no PSI-MS method has demonstrated the capability to differentiate between different strains of the same species of bacteria. This report describes the first PSI-MS application achieving strain-level differentiation of bacteria. Using the commensal mammalian intestinal bacterium Oxalobacter formigenes as the experimental model, an oxalate-degrading microorganism currently being investigated by our laboratory as a future probiotic therapy for kidney stones and other oxalate diseases²¹, we demonstrate this advancement in PSI-MS application showing analytical distinction between two of its strains, HC1 and OxWR, to better characterize their biological differences.

EXPERIMENTAL SECTION

Cell Culture and Sample Generation

Pure cultures of Oxf HC1 and OxWR were grown anaerobically from frozen glycerol stocks. For each strain, 0.5 mL stock was used to inoculate 75 mL of previously published Oxf-specific media supplemented with 100 mM oxalate²². Cultures were incubated at 37°C for 72 hours, after which 8–75 mL bottles of Oxf media were inoculated with 5 mL culture, yielding 8 culture replicates per strain. Cultures were incubated for 48 hours, after which cell pellets for each strain were combined by sequential centrifugation of replicate cultures at 15,000×g, 4°C for 5 min, discarding the supernatant after each addition. Pellets were washed 3 times by resuspension in 6mL 20 mM ammonium acetate with centrifugation and discarding of supernatant between all washes. In the third wash cycle, cell resuspensions were transferred to pre-weighed 20 mL vials, centrifuged, dried, and weighed for downstream normalization. Pellets were resuspended in 6 mL 20 mM ammonium acetate and 1 mL resuspension was removed and frozen at −80°C for whole cell analysis. Remaining resuspension was transferred to 15 mL polypropylene vials chilled in an ice slurry water bath for sonication. Cells were lysed using a Sonic Dismembrator Model 500 with a Branson Sonicator Probe (Thermo Fisher Scientific, Waltham, MA, USA) by the following method: 30% amplitude for 30 sec, 1 min cool-down, 60% amplitude for 30 sec, 2 min cool-down, 60% amplitude for 15 sec. Lysates were immediately frozen at −80°C.

PSI-MS Instrumentation and Analysis

Lysate and whole cell samples for Oxf HC1 and OxWR were thawed on ice and normalized based on cell concentration, adjusting volume with 20mM ammonium acetate. For both analyses, 15 µL sample was pipetted onto Velox™ sample cartridges (Prosolia Inc., Indianapolis, IN) containing pre-cut triangular analysis paper using a custom 3D-printed stabilizing device (Prosolia Inc., Indianapolis, IN) for reproducibility in sample dispensing and positioning. Samples were loaded into the Prosolia Velox™ 360 (Prosolia Inc., Indianapolis, IN) connected to a Thermo Scientific Q Exactive Orbitrap Mass Spectrometer (Thermo Scientific, Waltham, MA) for analysis in alternating sequence order by strain. 4:1 H₂O:Acetonitrile containing 0.1% formic acid was used as the wetting and spray solvent with 80 µL applied for wetting. This solvent was chosen to match the mobile phase gradient composition from past related LC-MS experiments with sufficient elution of both hydrophobic and hydrophilic species²¹. Spray voltage for ionization was 4.5 kV. Data acquisition was performed in positive ion mode at 140,000 mass resolution for 30 sec after 9 sec equilibration. Scan range was 70–1000 m/z. Additional instrumentation and analysis parameters are provided in Table S1. Acquired data were quality-checked for analytical reproducibility using the total ion current (TIC) both across the method duration and from sample-to-sample. Raw data TIC relative standard deviation (RSD) below 10% was observed across all samples in the HC1 lysates (9.1%), OxWR lysates (9.9%), and HC1 whole cell samples (5.2%). OxWR whole cell samples showed notable variance in raw data TIC (RSD=52.3%) without clear explanation. However, this variance was reduced to <1% by downstream TIC normalization of feature intensities during data processing. TIC normalization is common in metabolomic profiling as it helps to account for variation in instrument performance²³.

Data Processing and Statistical Analysis

RawConverter was used for data format conversion to mzXML²⁴. Peak picking and feature alignment were performed using MZmine 2²⁵. Non-detected species (intensity = 0) were replaced with half the minimum value in the dataset for statistical purposes²³. Statistical analysis and figure generation was aided using MetaboAnalyst 4.0²⁶. Feature intensities were normalized to TIC and autoscaled²⁷. Significance (p≤0.05) was determined using the two-tailed, unpaired Student’s t-test assuming equal variance between groups applying the Bonferroni-Holm false discovery rate correction²⁸. Chemical formulas and identifications were predicted using CEU Mass Mediator²⁹, simultaneously screening the Human Metabolome Database³⁰, KEGG³¹, Metlin³², and Lipid Maps³³ for exact mass matches (≤5ppm) to common ESI(+) adducts of known metabolites, lipids, and peptides.

RESULTS AND DISCUSSION

Cell Lysate Analysis

A total of 839 features were detected between HC1 and OxWR lysates in positive ion mode. Figure 1 shows that clear multi-variate statistical separation between these strains was observed by Principal Component Analysis (PCA) with 61.2% of the variance explained in 2 PCs. Strain separation is primarily observed along PC1. Although the overall metabolic pool for the strains appeared to be largely conserved in terms of features detected amongst HC1 and OxWR, 688 features (82.0%) showed a significant difference in their intensity between the strains. The distribution of these significant features between strains favored HC1 with 509 features (74.0%) showing increased abundance compared to OxWR.

Figure 1.

PCA depicts complete multivariate statistical separation and analytical distinction between features detected in cell lysates of Oxf HC1 and OxWR (n=8 replicates per strain) with 61.2% of the variance explained in 2 PCs and 76.5% explained in 5 PCs.

As mentioned in the introduction, the foundation for a PSI-MS assay for bacterial strain identification would be the detection of features present in one strain and completely absent from others, serving as strain-indicating biomarkers. These markers are the focus of this report. Among the 688 significant features, 13 were strain-exclusive, as listed in Table 1. Table S2 lists predicted identifications for these features based on exact mass matching to metabolite databases. Interestingly, all strain-specific masses were detected in OxWR as we were unable to detect any features unique to HC1 in the cell lysate analysis.

Table 1.

Strain-specific features detected in cell lysates and whole cells of Oxf strains HC1 and OxWR. Strain and sample type for each detected feature are displayed. See Table S2 for predicted identifications by exact mass matching to metabolite databases.

Mass	Strain	Analysis
75.0325	OxWR	Lysates, Whole Cells
93.0384	OxWR	Whole Cells
99.0198	OxWR	Whole Cells
100.0276	OxWR	Lysates, WC
101.0988	OxWR	Lysates
105.0303	OxWR	Whole Cells
115.0145	OxWR	Lysates
121.0251	OxWR	Whole Cells
125.0545	OxWR	Lysates, Whole Cells
133.0250	OxWR	Lysates, Whole Cells
134.0647	OxWR	Whole Cells
135.0406	OxWR	Lysates, Whole Cells
135.0770	OxWR	Whole Cells
143.0649	OxWR	Lysates
157.0807	OxWR	Lysates
159.0406	OxWR	Lysates
164.1018	OxWR	Lysates
174.0516	OxWR	Lysates, Whole Cells
188.0673	OxWR	Lysates, Whole Cells
325.3667	HC1	Whole Cells
334.3476	HC1	Whole Cells
339.3823	HC1	Whole Cells
353.3978	HC1	Whole Cells

Whole Cell Analysis

A total of 2357 features were initially detected between HC1 and OxWR whole cells in positive ion mode. To simplify this dataset because of the greater number of features, we employed a feature-by-feature blank filtration algorithm to remove features with 10% or greater signal contribution from the background³⁴, leaving 958 features in the filtered dataset. Figure 2 shows that complete separation between HC1 and OxWR strains was observed by PCA with 45.2% of the variance explained by 2 PCs. As seen with the lysate analysis, the overall metabolic pool for the strains was found to be largely conserved. However, only 43 features (4.5%) showed a significant difference in their intensity between the strains. Among the 43 significant features, we detected a total of 17 strain-exclusive features, 13 of which were specific to OxWR and 4 detected only in HC1, as listed in Table 1. Table S2 lists predicted identifications for these features based on exact mass matching to metabolite databases. 7 of the 17 strain-exclusive features were assigned at least one tentative identification. Regarding the 4 features detected only in HC1, their m/z values (m/z 325.3667–353.3978) are all relatively large for small molecules. It is likely that these are uncharacterized lipids or derivatized peptide species.

Figure 2.

PCA depicts complete multivariate statistical separation and analytical distinction between features detected in whole cells of Oxf HC1 and OxWR (n=6 replicates per strain) with 45.2% of the variance explained in 2 PCs and 78.6% explained in 5 PCs.

Strain-Exclusive Biomarkers

Between the whole cell and cell lysate analyses, a total of 24 features were detected exclusively in either HC1 or OxWR, as presented in Table 1. Among these features, 7 were detected as strain-exclusive in both analyses. Although tentative identifications were assigned to some of these features using metabolite databases (Table S2), we acknowledge the limitation of compound identification purely based on exact mass. It is common knowledge that there are many biological compounds with the same chemical formula and, consequently, the same exact mass. This presents a current challenge with mass spectrometric analysis of biological samples without chromatographic separation. MS/MS identification was attempted for all strain-specific features, but efforts were unsuccessful due to the high degree of mass overlap between chemical species even within a narrow 1 m/z window. This is expected when analyzing biological samples with hundreds or thousands of analytes present. We show example spectra in Figure S1 demonstrating the difficulty in gathering useful structural information by PSI-MS/MS on biological samples. Although a previous PSI-MS bacteria study used MS/MS for unknown identification⁵, only highly-abundant features were targeted in that experiment, likely to overcome the co-fragmentation issue. In this experiment, none of the strain-specific features were in high enough abundance for fragmentation spectra to surpass the competitive signal. Instrumentation advancements allowing for tighter mass isolation windows or employing ion mobility would improve outcomes in future related experiments. Where possible, we have reported predicted identifications for these strain-specific features, but we acknowledge that alternative identifications to what metabolomics databases provide are possible and require follow-up studies to confirm. One feature we feel is worth discussing is m/z 133.0250, detected in both whole cell and cell lysate analyses and predicted to be oxaluric acid. Oxaluric acid plays a role in purine metabolism in E. coli³⁵and is a known byproduct of oxidative DNA damage³⁶. It is also related structurally to oxalic acid, the sole energy source of Oxf²², and is a by-product of biochemistry between ureidoglycine and glyoxylic acid, a human metabolic precursor of oxalic acid²¹, in some bacteria³⁷. Being that this compound was predicted to be oxaluric acid by exact mass as two distinct adducts in the whole cell analysis - [M+H]⁺ adduct m/z 133.0249 and [M+H-H2O]⁺ adduct m/z 115.0145 – as well as [M+H-H2O]⁺ adduct in the lysate analysis, we believe this is supporting evidence for this identification. Subsequent studies, perhaps utilizing LC-MS/MS with pure analytical standards of these proposed compounds, are needed to confirm the identities of these biomarker features. Post-identification, future work should also focus on defining the biological implications behind their strain-exclusivity.

CONCLUSIONS

This novel PSI-MS application demonstrates that this analytical technique is capable of differentiating between bacteria at the strain level. Using normalized cell lysates and whole cells, Oxf HC1 and OxWR were analytically distinguished by multivariate statistical approaches, and a panel of strain-specific masses were detected that could serve as strain-indicating biomarkers. Future investigations should expand the utility of PSI-MS to differentiate and identify strains within a variety of species of bacteria, particularly those species with clinically-relevant pathogenic strains.