Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
STABLE METAL OXIDE CATALYSTS FOR METAL OXIDE LASER IONIZATION-MASS SPECTROMETRY
Document Type and Number:
WIPO Patent Application WO/2015/168337
Kind Code:
A1
Abstract:
Oxides of lanthanide metals are used to create stable surfaces for use in laser desorption ionization mass spectrometry. The disclosed lanthanide metal oxide surfaces are used to identify and characterize fatty acids from microorganism allowing profiling of the microorganism to the strain level, including antibiotic resistance. Cerium oxide, CeO2, was used to create a stable surface for obtaining fatty acid profiles of bacterial pathogens. Cross validation of results obtained using the disclosed methods, systems, and surfaces, demonstrated greater than 95% accuracy that did not suffer long-term degradation. Bacteria were identified by the disclosed methods, systems, and surfaces with greater than 98% at the species level and 96% at the strain level. Comparisons with existing technologies demonstrate that the presently disclosed methods, systems, and surfaces provide for surprisingly enhanced accuracy and reliability in profiling and identification.

Inventors:
VOORHEES, Kent J. (714 Partridge Circle, Golden, CO, 80403, US)
COX, Christopher R. (1500 Illinois Street, Golden, CO, 80401, US)
Application Number:
US2015/028368
Publication Date:
November 05, 2015
Filing Date:
April 29, 2015
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
COLORADO SCHOOL OF MINES (1500 Illinois Street, Golden, Colorado, 80401, US)
International Classes:
C01F1/00; C01F17/00; C23C30/00; H01J1/14; H01J1/142; H01J1/148; H01J27/02; H01J27/24; H01J49/26
Foreign References:
US20030205078A12003-11-06
US20120261567A12012-10-18
US4265107A1981-05-05
US20120052006A12012-03-01
US7122792B22006-10-17
Attorney, Agent or Firm:
JONSEN, Matthew D. et al. (1400 Wewatta Street, Suite 400Denver, Colorado, 80202, US)
Download PDF:
Claims:
CLAIMS

We claim:

1. A sample plate for matrix-assisted laser desorption ionization mass spectrometry analysis, the plate comprising:

a first surface;

a second surface in contact with the first surface, the second surface comprising a layer positioned on the first surface, the layer comprising an oxide form of an element selected from a lanthanide or an alkaline earth metal.

2. The sample plate of claim 1 , wherein the oxide is of a lanthanide selected from lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), promethium (Pm), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dysprosium (Dy), holmium (Ho), erbium (Er), thulium (Tm), ytterbium (Yb), and lutetium (Lu).

3. The sample plate of claim 2, wherein the oxide is cerium oxide.

4. The sample plate of claim 3, wherein the oxide is Ce02.

5. The sample plate of claim 1 , wherein the oxide is of an alkaline earth metal selected from beryllium (Be), magnesium (Mg), calcium (Ca), strontium (Sr), barium (Ba), and radium (Ra).

6. The sample plate of claim 1 comprising a plurality of second surfaces arranged as sample dots.

7. The sample plate of claim 1 , wherein the oxide surface layer comprises a surface facet selected from <111>, <110>, and <100>.

8. A method of characterizing an analyte comprising:

depositing the analyte on a surface, the surface comprising a layer of a metal oxide; irradiating the analyte with one or more pulses from a laser to create an analyte ion; allowing the analyte ion to desorb from the surface; and

detecting the analyte ion with a mass spectrometer.

9. The method of claim 8, wherein the metal oxide is selected from an oxide of beryllium (Be), magnesium (Mg), calcium (Ca), strontium (Sr), barium (Ba), radium (Ra), lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), promethium (Pm), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dysprosium (Dy), holmium (Ho), erbium (Er), thulium (Tm), ytterbium (Yb), and lutetium (Lu).

10. The method of claim 9, wherein the metal oxide is Ce02 that is stable for greater than 48 hours.

11. The method of claim 8, wherein the analyte comprises one or more lipids.

12. The method of claim 11 , wherein the lipids are derived from a fuel, a microorganism, or a plant.

13. The method of claim 12, wherein the microorganism is a bacterium, fungus, virus or alga.

14. The method of claim 13, wherein the microorganism is a bacterium.

15. A method of characterizing a bacterium comprising:

creating a sample from the bacterium, wherein the sample comprises one or more analytes;

contacting a metal oxide surface with the sample;

pulsing a laser at the sample to ionize and desorb a plurality of analyte species from the surface;

detecting the desorbed analyte species in a mass spectrometer;

determining the identity and relative abundance of the analyte species to create a profile of the bacterium.

16. The method of characterizing a bacterium of claim 15, wherein the profile is compared to a database of profiles to determine the identity of the bacterium.

17. The method of characterizing a bacterium of claim 15, wherein the desorbed analyte specie is a fatty acid.

18. The method of characterizing a bacterium of claim 17, wherein the fatty acid is one or more of a C10-30 fatty acid.

19. The method of characterizing a bacterium of claim 18, wherein the fatty acid has 0, 1 , 2, or 3 double bonds.

20. The method of characterizing a bacterium of claim 15,

wherein the metal oxide surface is Ce02,

wherein the analyte ion is a negative ion, and wherein the desorbed analyte species are fatty acids with between 10 and 90 carbon atoms and 0, 1 , 2, or 3 double bonds.

21. The method of characterizing a bacterium of claim 20, wherein the identity of the microorganism is determined at the genus, species, or strain level with greater than 70% confidence.

22. The method of characterizing a bacterium of claim 21 , wherein the identity of the microorganism is determined with greater than 85% confidence.

23. The method of characterizing a bacterium of claim 22, wherein the strain is antibiotic resistant.

24. The method of characterizing a bacterium of claim 22, wherein the strain is antibiotic sensitive.

25. The method of characterizing a bacterium of claim 22, wherein the strain is selected from MRSA and MSSA.

Description:
STABLE METAL OXIDE CATALYSTS FOR METAL OXIDE LASER

IONIZATION-MASS SPECTROMETRY

CROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. provisional patent application nos. 61/985,919 filed April 29, 2014, and 62/136,088 filed March 20, 2015, which are incorporated herein by reference in their entirety.

FIELD

[0002] The present disclosure relates to laser desorption ionization - mass spectrometry systems, methods, and metal oxide surfaces for analyzing and characterizing analytes. The disclosed surfaces provide for reliable analysis and characterization of small analytes

(<1000 Da) and samples comprising complex heterogeneous mixtures of analytes, without the need for activating the surface.

BACKGROUND

[0003] The term "matrix" in matrix assisted laser desorption/ionization mass spectrometry (MALDI-MS) refers to an organic compound or solution (usually acidic) that is selected to aid in the ionization process. In general the use of a matrix did not interfere with the spectra produced from analysis of high molecular weight molecules (carbohydrates, proteins, nucleotides, and polymers). However, MALDI used to analyze small molecules, for example compounds with a molecular weight less than about 1 kDa suffers from the presence of the organic matrix. This is because organic matrices, themselves, produce spectral peaks in this mass region.

[0004] To alleviate the problems associated with matrix contamination, systems have been investigated that are characterized as "matrix-free." However, these systems tend not to be "matrix-free," are limited to analysis of pre-ionized compounds, or rely on a functionalized MALDI surface, for example porous silicon (DIOS - desorption/ionization on silicone). Many of these systems may also lack the ability to be manufactured with a reproducible surface and to efficiently ionize non-basic compounds such as lipids.

[0005] The analysis of lipids by MALDI is desired because fatty acid profiling aids in bacterial identification. Original profiling work utilized an ex situ method involving saponification followed by derivatization of cell wall fatty acids to form fatty acid methyl esters (FAMEs). Later, an in situ thermochemical process using tetramethyl ammonium hydroxide (TMAH) was developed, which significantly reduced sample analysis time to less than three minutes for Curie-point pyrolysis mass spectrometry (PyMS). [0006] Pyrolysis has also been observed during lipid analysis by matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS). During an investigation of TMAH as a methylating agent for MALDI-TOF MS with calcium oxide (CaO) as a matrix, lipid cleavage was observed. MALDI MS-MS analysis identified the resulting cleavage products as calcium fatty acid adducts. At the time, it was not understood that a strong base such as TMAH was needed to activate CaO to facilitate catalysis. Thermal or acid activation could also be used. Using this method, monoacylglycerides, diacylglycerides, triacylglycerides, phospholipids, as well as bacterial and algal lipid extracts produced peaks for their constituent fatty acids. In addition, spectra from bacterial and algal extracts were found to be reproducible, which suggested that this technique could be used for microbial fatty acid profiling. Further studies demonstrated that Gram-positive and Gram-negative bacteria could be analyzed using calcium oxide (CaO) in positive- and negative-ion MALDI- TOF MS modes followed by principal component analysis (PCA). PCA analysis identified distinct clusters for replicates of each bacterial species with 94% cross validation correlation of positive-ion data. Group clustering showed improved intra variance for negative-ion data and cross validation of 97%. However, CaO surfaces decay over time, hampering long term reproducibility of this technique.

[0007] What is needed is a surface for use with laser desorption ionization that does not interfere with characterization of low molecular weight compounds, does not require surface activation, and provides for reproducible results.

SUMMARY

[0008] Disclosed herein are metal oxide compositions, surfaces, and methods of using same in analysis of an analyte. In various embodiments the compounds, surfaces, and methods are used for the analysis of analytes. Any analyte known in the art may be detected/analyzed by the presently disclosed compounds, surfaces, and methods. In some embodiments the analyte may be a lipid, glycolipid, phospholipid, glycerolipid, fatty acid, carbohydrate, chemical agent, phenolic compound, lignol, pyrolysis oil, protein, peptide, nucleic acid, cell, petroleum product, oil, crude oil, fuel, fuel constituent, lignin dimer, lignin trimer, poly-aromatic, FAME, acylglyceride, carbolipid, or combinations thereof. Analytes may be obtained from natural, environmental, biological, or synthetic sources and samples. In some embodiments, the analyte may be extracted from the sample. In some

embodiments, the sample may comprise one or more microorganisms, for example bacteria, alga, yeast, fungus, protozoa, viruses, etc. In some embodiments, the sample may be any complex mixture of analytes, for example vegetable oil, olive oil, or crude oil. In other embodiments, the sample is not extracted prior to analysis. [0009] In various embodiments, the analyte is selected from the group consisting of a lipid, glycolipid, phospholipid, glycerolipid, fatty acid, carbohydrate, chemical agent, phenolic compound, lignol, pyrolysis oil, peptide, nucleic acid, cell, petroleum product, oil, crude oil, fuel, fuel constituents, lignin dimers, trimers and poly-aromatics. In many embodiments, the analyte may have a carbon backbone of from about 10 - 90 carbon atoms. In some embodiments, the analyte is cleaved and/or ionized when contacted by the laser pulse, for example an N 2 , 337 nm, laser pulse.

[0010] Various metal oxides may be used. In some embodiments the metal oxide is a lanthanide or alkaline earth metal oxide. The metal oxide surface layer may be uniform or heterogeneous, comprising defined surface facets selected from, but not limited to <100>, <1 10>, and <1 1 1 >. In various embodiments, the metal oxide is an oxide of beryllium (Be), magnesium (Mg), calcium (Ca), strontium (Sr), barium (Ba), radium (Ra), lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), promethium (Pm), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dysprosium (Dy), holmium (Ho), erbium (Er), thulium (Tm), ytterbium (Yb), and lutetium (Lu). In many embodiments, the metal oxide is cerium oxide. In many embodiments, the cerium oxide is Ce0 2 .

[0011] The disclosed metal oxide may be used to characterize complex samples. The samples may be organic, non-organic, natural, or synthetic. In one embodiment, the samples comprise biologically derived molecules. In many embodiments, for example where the sample comprises one or more microorganisms, the metal oxide may aid in

characterizing and identifying microorganisms of differing genera, species, and/or strains within the sample. In many embodiments, the strain identity may include antibiotic resistance. In some embodiments, the sample may comprise lipids derived from cell walls of microorganisms. In some embodiments, the samples may be whole cell bacteria, fungi, or algae, or whole viruses.

BRIEF DESCRIPTION OF THE DRAWINGS

[0012] FIG. 1 is a graph showing CaO reproducibility as a function of time. Square, E. coli; circle, S. typhimurium; triangle, S. boydii. Red, 0 hrs; green, 8 hrs; blue, 24 hrs.

[0013] FIGs. 2A and 2B is a comparison of catalyst stability as a function of time. S.

typhimurium spectra obtained over 24 hrs using 2A) CaO and 2B) Ce0 2 .

[0014] FIGs. 3A and 3B is a PCA plots of 10 bacteria from 3A) freshly activated CaO and

3B) Three Ce0 2 time points. (®<; blue circles) A. baumannii, (®; pink circles) B. anthracis, (©

; aqua circles) C. putrefaciens, (©; It. green circles) E. coli, (m; blue circles) E. faecalis, (©; purple circles) F. tularensis, (©; green circles) L. monocytogenes, (®; red circles) S. aureus, (©; orange circles) S. typhimurium, (©; grey circles) Y. pestis. [0015] FIG. 4 is a PCA plot of Acinetobacter. Yellow, A. baumannii ATCC 17976; dark blue, A. baumannii AC54; pink, A. calcoaceticus Mx70.71 ; purple, A. calcoaceticus 75.53; grey, A. haemolyticus ATCC 17907; cyan, A. haemolyticus 2213; turquoise, A. pittii 3 ATCC 17922; orange, A. pittii ATCC 19004; red, A. nosocomialis 13 ATCC 17903; and green, A.

nosocomialis ATCC 700472.

[0016] FIG. 5 are representative Ce0 2 -facilitated fatty acid profiles of staphylococci. Fatty acids are labeled with respect to chain length and degree of unsaturation.

[0017] FIG. 6 is a species-level differentiation of 14 Staphylococcus species by PCA.

[0018] FIG. 7 is a color-coded dendrogram by species; the dendrogram was created using average linkages and Euclidean distance. Blue lines indicate that the nearest neighbor is a different species.

[0019] FIG. 8 is a FuRES classification tree of 14 Staphylococcus species. Thirteen rules indicate perfect classification.

[0020] FIG. 9 is a three-dimensional PCA plot of 18 Staphylococcus strains.

[0021] FIG. 10 is a phylogenetic dendrogram of 18 different strains of Staphylococcus. Most strains show tight clustering with the exception of S. haemolyticus.

[0022] FIG. 11 is a FuRES tree of the 18 strain classes.

[0023] FIG. 12 is a 3-dimensional PCA plots of S. aureus strains. Representation of all strains as individual color coded data points with circles to distinguish MRSA and MSSA strains.

[0024] FIG. 13 is an average of 300 FuRES discriminant weights with 95% confidence intervals. If the confidence interval does not contain zero, then that weight is significant. Negative weights correspond to larger features in MRSA and positive weights to larger features in MSSA. DETAILED DESCRIPTION

[0025] Novel systems, surfaces, and methods for laser desorption ionization are disclosed. In many cases, the disclosed systems may include or may be coupled with one or more mass spectrometers. In many embodiments, the disclosed methods, systems, and surfaces allow for reliable and reproducible identification and characterization of an analyte, for example a heterogeneous population of small molecular weight compounds. In some embodiments, the compounds are lipids or fatty acids, and the surfaces do not require activation or re-activation.

[0026] Any analyte known in the art may be detected/analyzed by the presently disclosed systems, surfaces, and methods. For example, the analyte may be a lipid, glycolipid, phospholipid, glycerolipid, fatty acid, carbohydrate, chemical agent, phenolic compound, lignol, pyrolysis oil, peptide, nucleic acid, cell, petroleum product, oil, crude oil, fuel, fuel constituent, lignin dimer, lignin trimer, poly-aromatic, FAME, acylglyceride, carbolipids, or combinations thereof. Analytes may be obtained from natural, environmental, biological, or synthetic sources.

[0027] The analyte to be analyzed may be part of a sample that is being characterized. In some embodiments, the sample may include a combination of analytes and/or a

heterorgeneous or homogeneous population of analytes. In some embodiments, the samples are obtained from microorganisms, including algae, fungi, bacteria and viruses. In some embodiments the bacteria are antibiotic resistant. In some embodiments, the bacteria are gram positive or gram negative. In one embodiment, the sample comprises fatty acids extracted from bacteria. By comparing the fatty acid spectra obtained from a sample to a database of reference spectra, the identity of the microorganism may be determined. In many embodiments, the identity may be determined at the strain level with greater than 85% accuracy. In many embodiments, the accuracy of species and/or strain level identification is greater than about 86%, 87%, 88%, 89%, 90%, 91 %, 92%, 93%, 94%, 95%, 96%, 97%, 98%, or 99%, and less than about 100%, 99%, 98%, 97%, 96%, 95%, 94%, 93%, 92%, 91%, 90%, 89%, 88%, 87%, or 86%.

[0028] The disclosed surfaces may comprise various metal oxides. In many embodiments, the metal oxide is a lanthanide or alkaline earth metal oxide. In various embodiments, the metal oxide is selected from an oxide of beryllium (Be), magnesium (Mg), calcium (Ca), strontium (Sr), barium (Ba), radium (Ra), lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), promethium (Pm), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dysprosium (Dy), holmium (Ho), erbium (Er), thulium (Tm), ytterbium (Yb), and lutetium (Lu). In many embodiments, the metal oxide is cerium oxide. In some

embodiments, the cerium oxide is Ce0 2 . In many embodiments the metal oxide is stable and does not require activation or reactivation. The metal oxide may be used in the disclosed systems and methods to produce fatty acid profiles that are reproducible. The metal oxide surface can produce fatty acid profiles from the same sample that are greater than 95% identical after storage for greater than 24h, 48h, 72h, 1 week, 2 weeks, or 3 weeks.

[0029] The disclosed surface of metal oxide may be a uniform or heterogeneous layer. In some embodiments, the metal oxide surface is homogeneous with exposed metal and exposed oxygen atoms in defined surface facets selected from, but not limited to, <100>, <110>, and <111 >.

[0030] The disclosed metal oxide compounds, metal oxide surfaces, and LDI methods and systems allow analysis of samples, analytes, and analyte mixtures with little or no overlap with, or interference from matrix-derived background-spectra. In some embodiments, metal oxide compounds, surfaces, and methods of using the disclosed surfaces and compounds, provide for direct, matrix-free, and reproducible analysis of analytes and samples of analytes. In some cases, the disclosed methods may be referred to as MOLI (metal oxide laser desorption/ionization), and may be used to produce protonated and sodiated molecular ions. MOLI can also be used with samples comprising complex mixtures of analytes, for example vegetable oil shortening and lipid extracts derived from biological, environmental, and chemical sources. In one embodiment, samples are produced from cell walls, membranes, fuels and fuel constituents, such as heavy crude oil. The sample may be an extract or a crude (unprocessed or minimally processed) sample.

[0031] In one embodiment, the disclosed systems, surfaces, and methods are used for bacterial profiling. The disclosed systems, surfaces, and methods may provide for improved identification of bacteria at strain-level resolution. In some cases, the disclosed methods may use fatty acids (FAs) as diagnostic biomarkers for bacterial identification. The disclosed systems, surfaces, and methods can be used to identify, determine, and distinguish bacterial strains, including identifying drug resistance. In many embodiments, the laser desorption ionization surface is a lanthanide oxide, cerium oxide (Ce0 2 ). In some cases, the disclosed methods, systems, and surfaces have been used as an in situ saponification catalyst and matrix replacement in laser desorption ionization mass spectrometry, wherein the data is analyzed using multivariate statistical analysis to identify, characterize, and distinguish bacteria. In some cases, principal component analysis (PCA) can be used to classify the data produced from the disclosed systems, surfaces, and methods. Use of the disclosed methods, systems, and surfaces may allow for microorganism identification greater than about 85%, 86%, 87%, 88%, 89%, 90%, 95%, 96%, 97%, 98%, or 99% at the species level and/or at the strain level (including antibiotic resistance).

[0032] Experiments, described below, demonstrate that Ce0 2 -facilitated FA profiling by MOLI-MS, coupled with multivariate statistical analysis, is useful for strain-level identification and characterization of various microorganisms, including bacteria. Rapid and reliable identification of microorganisms, especially those that are particularly virulent and/or resistant to specific therapies is important. For example, the genus Staphylococcus has several species and strains that are of particular interest to health professionals. For example, the emergence of methicillin-resistant Staphylococcus aureus (MRSA) strains and coagulase-negative staphylococcal strains (CoNS) has made the ability to differentiate Staphylococcus at the strain level more important.

[0033] The disclosed experiments confirmed the ability to differentiate MRSA from methicillin-sensitive Staphylococcus aureus (MSSA), which greatly enhances the therapeutic management and control of these infections. Experiments using supervised and

unsupervised learning provided for accurate classification data. Leave-one-spectrum-out cross-validation (LOSOCV) yielded accuracies of 100% and 96% at the species and strain level, respectively. Fuzzy rule-building expert system (FuRES) classification, with a more rigorous evaluation, achieved 98% classification consistently. These methods, systems, and surfaces allow for construction of a comprehensive bacteria database, which has been shown to be important to the identification capabilities of this and other identification methods.

[0034] Staphylococci are Gram-positive facultative anaerobes comprising 44 species, which are commonly found in the soil or on the skin of birds and mammals. S. aureus is the most problematic of the genus, being identified as an etiologic agent of septicemia, osteomyelitis, endocarditis, and skin infection. Isolates from this organism are generally susceptible to β- lactam antibiotics, such as methicillin and penicillin, but modification of penicillin binding proteins (PBP) has led to development of resistant strains. The Centers for Disease Control and Prevention (CDC) reported in 201 1 that there were 80,461 methicillin-resistant S. aureus (MRSA) infection cases in the U.S., which led to 11 ,285 deaths, emphasizing the importance of addressing the emergence of resistant strains. As a result of implemented infection control procedures in healthcare facilities, hospital MRSA infections have decreased 52% between 2005 and 2011. While the numbers have seen a decline, there remains a need to rapidly screen patients for bacterial infection, including identification of S. aureus and determine whether the strain is antibiotic resistant.

[0035] Traditionally, culture, biochemical, and molecular methods have been at the forefront of clinical MRSA detection. Culture methods offer high specificity, but the relatively lengthy turnaround time (TAT) of 24-72 h. and requirement of sequential testing for antibiotic resistance are major drawbacks. Patients diagnosed with a MRSA strain in the ICU who receive improper or lack of therapy within 48 h had a 7% increase in mortality rate per hour compared to those who received adequate treatment. Chromogenic agars have been used to slightly decrease TAT to 18-24 h, while also improving specificity, but there is still a requirement for secondary testing. Some of the most common approaches for analysis of the specific chemical characteristics of staphylococci include: coagulase activity, hemolysis, nitrate reduction, phosphatase, and aerobic acid production from sugars and carbohydrates. Kloos and coworkers reported a simplified scheme for analyzing the extensive data produced by biochemical results to characterize staphylococci. This system acts as a flowchart for clinical bacteriologists, based solely off of positive and negative test results, for phenotypic determination.

[0036] Commercially available systems, for example BioMerieux API STAPH-IDENT and American Hospital Supply Corporation MicroScan System, are based on the principles described by Kloos. The API Staph-IDENT, for example, utilizes a battery of 10 microscale biochemical tests, whereas the MicroScan System consists of 27 tests. These systems were reported to have accuracies of 88% and 86.4%, respectively, but also showed inherent limitations. Specifically, the API-STAPH-IDENT failed to identify phosphatase-negative S. epidermidis. In comparison, while the MicroScan System was able to correctly identify S. epidermidis, it had difficulties in identifying S. hominis, S. warneri, and S. sciuri. In addition, biochemical tests that rely heavily on coagulase or dumping-factor production also had limitations, which included: the failure to differentiate between S. pseudintermedius and S. delphini and, to a greater extent, the misclassification of all coagulase-positive staphylococci as S. aureus.

[0037] To address such limitations and improve specificity, molecular methods for analyzing specific genetic markers have been extensively explored. In an attempt to ID S. aureus and assay for methicillin resistance, multiplexed PCR has been used to simultaneously target the staphylococcal nuc gene, encoding the thermostable nuclease (TNase), and the meek gene, encoding the penicillin binding protein. PCR results agreed with coagulase production and agar screening tests for single-step ID of MRSA, which proved useful for clinical specimens. In an attempt to identify coagulase-negative staphylococcal strains (CoNS), a similar study targeted a 429-bp amplicon of the sodA gene encoding the manganese-dependent superoxide dismutase. Clinical isolates and ATCC reference strains were identified using the sodA gene with 83% accuracy in about 8 hours. While culturing and biochemical assays can offer comparable specificity to results obtained by hsp60 and 16S rRNA gene

sequencing, TAT is still greater than 24 h.

[0038] Turnaround time was significantly reduced using a non-molecular technique for S. aureus analysis that employed phage amplification and lateral flow immunochromatography (LFI) detectors. This work led to the FDA-approved MicroPhage KeyPath MRSA/MSSA blood culture test. The exploitation of S. ai/rei/s-specific phage amplification against clinical blood isolates allowed for simultaneous ID and methicillin resistance determination with a TAT of 5 h and 98.3% accuracy.

[0039] Published reports suggest the rise of non-S. aureus infections in clinical studies, some with developed resistance to multiple classes of antibiotics. Indeed, CoNS are among the most commonly reported bloodstream isolates (37.3% compared to only 12.6% for S. aureus). These reports place emphasis on the virulence of S. epidermidis, S. saprophytics, S. lugdunensis, and S. schleiferi further emphasizing the need for rapid, simultaneous detection at the species level. Bacterial protein analysis by matrix assisted laser

desorption/ionization-time of flight mass spectrometry (MALDI-TOF MS) has been used to identify S. aureus and CoNS in prosthetic joint infections. Although this method is relatively rapid, only 52% highly probable species-level ID was obtained. A report by Dubois and coworkers confirmed PCR-based sodA gene array results using MALDI-TOF MS protein analysis of 152 staphylococcal isolates. By using the Biotyper 2.0 Bioinformatic platform (Bruker Corporation), a database commercialized by Bruker with a comprehensive reference library of 3,000+ microbial species, a total of 151 samples were correctly identified with species level scores. Further, Rajakurna et al. analyzed a different set of strains using the MicrobeLynx database, developed by the Waters Corporation. This resulted in correct species level ID of 97%.

[0040] The presently disclosed methods, systems, and surfaces use LDI MS for bacterial profiling. Bacterial fatty acids are used as diagnostic biomarkers (rather than proteins). The use of Ce0 2 as an in situ saponification catalyst and matrix replacement allowed bacterial samples to be identified to the species level with 97% accuracy. Additional studies compared the ability of the present disclosed methods and systems, versus previous methods to identify groups of Enterobacteriaceae, Listeria, and Acinetobacter. Specifically, the presently disclosed methods and systems, using Ce0 2 , were analyzed in parallel with a commercially available system from Bruker Corporation (Biotyper). The results from this study clearly established fatty acid MALDI profiling for strain-level differentiation of phylotypes with 98-100% accuracy. This level of accuracy is significantly better than existing techniques based on protein profiling.

[0041] While multiple embodiments are disclosed, still other embodiments of the present invention will become apparent to those skilled in the art from the following detailed description. As will be apparent, the invention is capable of modifications in various obvious aspects, all without departing from the spirit and scope of the present invention.

Accordingly, the detailed description is to be regarded as illustrative in nature and not restrictive.

[0042] All references disclosed herein, whether patent or non-patent, are hereby

incorporated by reference as if each was included at its citation, in its entirety. In case of conflict between reference and specification, the present specification, including definitions, will control.

[0043] Although the present disclosure has been described with a certain degree of particularity, it is understood the disclosure has been made by way of example, and changes in detail or structure may be made without departing from the spirit of the disclosure as defined in the appended claims. EXAMPLES

EXAMPLE 1 - Surface oxide stability

[0044] Because metal oxides degrade over time, requiring re-activation, six metal oxide catalysts were evaluated for their capacity to reproducibly cleave bacterial cell wall phospholipid extracts to their constituent fatty acids as a function of time followed by negative-ion MOLI-TOF MS. CaO was found to partially degrade in less than one-hour post- activation. Surprisingly, of the six catalysts discussed below, Ce0 2 was shown to be stable over time and provided for reproducible data. In addition, Ce0 2 required no activation, unlike other metal oxides. An extended time study up to 504 hrs using the same fatty acid extract and stored catalyst was performed. These extended studies showed that little or no difference in fatty acid distribution occurred, even after the cerium oxide surface was stored for 504 hours. This was not the case with CaO, which showed reduced reproducibility after only 1 hour, post-activation. Cross-validation analysis of all data included in the Ce0 2 time study provided 99.5% correct classification.

[0045] In addition, Ce0 2 MOLI-TOF MS analysis was used to identify and characterize Acinetobacter strains that were difficult to identify with the Biotyper protein profiler. These studies revealed much better results with the presently disclosed methods, systems, and surfaces - specifically, cross validation was shown to be 98.6% at the species level and 96% at the strain level.

[0046] Bacterial strains and culture conditions. Tables 1 and 4 summarize the bacteria used in this study. All bacterial strains were cultured overnight in brain heart infusion (BHI) medium (BD-Difco, Franklin Lakes, NJ) at 37°C with the exception of the A. nosocomialis ATCC 700472 (formerly genomospecies 13), which was cultured in tryptic soy agar (TSA) and dextrose (BD-Difco, Franklin Lakes, NJ) at 30°C with continuous aeration. Following incubation of broth cultures, bacteria were streaked onto agar plates and incubated for 22 hrs. Extracts of single colonies were immediately prepared upon removal from the incubator. In order to minimize culture conditions as a variable in fatty acid analysis, these parameters were used as a carefully controlled, standardized method for all experiments.

[0047] Table 1. Bacterial phylotypes used

Gram positive Strain Source

Bacillus anthracis Steme Armed Forces Institute of Pathology

Clostridium putrefaciens ATCC 25786 ATCC

Enterococcus faecalis V583 University of Oklahoma Health Sciences Center

Listeria monocytogenes ATCC 19112 ATCC

Staphylococcus aureus ATCC 27660 ATCC

Gram negative

Acinetobacter baumannii AC54 Felix d'herelle Reference Center Viruses

Escherichia coli ATCC 15597 ATCC

Francisella tularensis LVS Colorado State University

Salmonella typhimurium ATCC 13311 ATCC

Yersinia pestis A1122 CDC Division of Vector-borne Diseases

[0048] Lipid extraction. Individual colonies were suspended in 100 μί of a 33/66 v/v% methanol/chloroform mixture (Pharmco-AAPER, Shelbyville KY and Fisher, Pittsburgh PA, respectively) and vortexed for 2 min at 3200 rpm. Phase separation was achieved by addition of 100 μΙ_ of phosphate buffer saline (PBS, pH 7.4) followed by an additional 60 sec vortexing and cell debris removed by centrifugation for 30 sec at 1400 x g. The resulting organic fraction was then removed for MALDI-TOF MS analysis.

[0049] Catalysts. The following metal oxides were investigated for their combined catalytic properties and temporal stability during bacterial phospholipid cleavage: calcium oxide (Nano-Active Inc. Manhattan KS), magnesium oxide (Nano-Active Inc,), nickel oxide (Sigma Aldrich, St. Louis, MO), cerium (IV) oxide (Cerac, Milwaukee, Wl), barium oxide (Cerac), and strontium oxide (Alfa Aesar, Ward Hill, MA). Thermal activation was performed on CaO, BaO, and SrO in an open-air tube furnace at 800°C for 3 hrs while NiO and MgO were heated to 350°C to remove surface water. Activation of Ce0 2 was not required. All catalysts were stored in a desiccator under vacuum.

[0050] An investigation of the stability of the CaO, BaO, and SrO was conducted by taking MALDI-TOF MS measurements of the same bacterial phospholipid extracts at 0, 8, and 24 hrs post activation. Ce0 2 was analyzed at 0, 8, 24, and 504 hrs.

[0051] Mass spectrometry. Samples were prepared for mass spectrometry as previously described. Briefly, 100 mg catalyst was added to one mL of n-hexane. (Sigma-Aldrich). One iL was then removed from the bottom of the resulting slurry and spotted on a stainless steel MALDI sample plate followed by 2 [iL of lipid extract.

[0052] Mass spectrometric measurements were acquired in positive- and negative-ion modes with a Bruker Ultraflextreme MALDI-TOF mass spectrometer (Bruker Daltonics, Billerica, MA) equipped with a 355 nm Nd:YAG laser. Spectra were collected in reflector mode with a grid voltage of 50.3%, a delayed extraction time of 120 ns, and a low mass cutoff of 150 Da. Five replicate spectra were collected for each analysis as 500 shot composites at a sampling frequency of 1 kHz using automated laser rastering.

[0053] Data analysis. Mass spectral data were exported from the instrument in ASCI I format and processed using an in-house program. Briefly, data were centroided, normalized to total intensity, and fatty acid peaks selected by specific mass prior to PCA. Processed data were recorded in a tab-delimited format and imported directly into R Version 3.0.2 (The R Foundation for Statistical Computing, Vienna, Austria) as a data frame. PCA was performed using the prcompQ function with an argument passed to return rotated variables; data were automatically mean-centered by prcompQ. Data were plotted using the built-in plotQ function in R. Leave-one-out cross validation was performed by passing the appropriate argument to the linear discriminate analysis function lda(), which is part of the Modern Applied Statistics with S package. Results from IdaQ returned class assignment based on supervised learning. [0054] Results and Discussion. MALDI-TOF MS using CaO as a catalyst has been shown to cleave phospholipids into their fatty acids. Cross validation results were above 94% for the previously referenced 10 bacteria. In general, it is beneficial for a metal oxide catalyst to be stable over time and capable of reproducible profiles. CaO is known to be susceptible to moisture and carbon dioxide poisoning and must be activated prior to use for catalytic pyrolysis of lipids. It was unclear if the catalyst could be stored under vacuum and maintain its original activity. Figure 1 shows a PCA plot of the first 2 principal components (PCs) for negative-ion CaO-catalyzed MALDI mass spectra of Escherichia coli, Salmonella

typhimurium, and Shigella boydii extracts. Measurements of a single extract from each bacterial type were taken over three time points from 0, 8, and 24 hrs post thermal activation. Although unique spatial groupings corresponding to each time point were observed, the relative position of distinct groupings for each bacterial species changed with time, and showed a temporal effect on catalyst deactivation. The overlapping alignment observed for the S. boydii and S. typhimurium groupings suggests that these would be difficult to differentiate using supervised learning techniques. Table 2 summarizes the variance associated with the first six PCs. A total variance of 85.6 % in PCs 1 and 2 justified describing the data with two PCs.

Table 2. Percentage of total variance for CaO reproducibility

Principal component % individual variance % cumulative variance

1 46.4 46.4

2 39.2 85.6

3 6.3 91.9

4 5.9 97.9

5 .9 98.8

6 .7 99.5

[0055] Because CaO may be deactivated over short time periods, five additional prospective catalysts were evaluated for production of fatty acids from glyceride standards and their overall stability investigated over a 24-504 hr timeframe. NiO and MgO resulted in low levels of glyceride standard cleavage. The experiments were repeated using various instrument parameters. However, molecular species of intact glycerides in both positive- and negative- ion modes were observed. Cerium oxide, BaO, and SrO resulted in cleavage of the standards (data not shown) and BaO and SrO produced both positive- and negative-ion peaks in the same manner as previously reported for CaO. However, only negative-ions were observed for Ce0 2 , in part, because the +4 oxidation state of cerium produces a +3 charge on resulting single fatty acid complexes. The low mass, non-integer peaks of these complexes places them in a spectral region that may complicate data analysis. BaO and SrO were not investigated further. Ce0 2 is listed in the MSDS as being predominantly non- reactive. [0056] To determine whether Ce0 2 demonstrated improved stability over CaO, a time study was conducted with the same 10 bacteria listed in Table 1. Figure 2 shows selected negative ion spectra of Salmonella from 0, 8, and 24 hrs for the two catalysts. The intensities of the three spectra shown in Figure 2B obtained using Ce0 2 were consistent with each other as a function of time and comparable to the CaO-derived spectrum at time zero shown in Figure 2A. CaO spectra (Figure 2B) obtained at 8 and 24 hrs showed changes in intensity of the individual fatty acids and a prominent peak at m/z 473, indicating degradation of the catalyst. Figure 3 shows PCA plots of Ce0 2 derived bacterial data and previously published CaO data for the same bacteria. Table 3 summarizes the variance associated with the first six PCs for each catalyst. The cumulative variance for the Ce0 2 data was 92.1% for first 2 PCs. In contrast, the cumulative variance for CaO was only 73% encompassing 6 PCs. In the case of the CaO data, all combinations (PC1 vs 2, PC 1 vs 3, etc) were plotted, however, PCs 1 and 2 gave the best separation. Comparison of 3D plots of the first 3 PCs for CaO (Figure 3A) and Ce0 2 (Figure 3B) show similar separation (outer variance), however reproducibility (inner variance) was better using Ce0 2 . Cross validation of Ce0 2 - derived fatty acid profiles showed 100% correct classification. In many embodiments, the disclosed method may use fewer that 9, 8, 7, 6, 5, 4, 3, or 2 principal components to identify the microorganism. In one embodiment, 3 PCs may be used.

[0057] Table 3. Comparison of percentage of total variance

[0058] Acinetobacter has previously been shown to be difficult to classify using protein profiling. A group of 10 Acinetobacter strains (Table 4) analyzed on the Bruker Biotyper gave 30% correct identification at the species level, 80% at the genus level, 58% were misidentified as an incorrect genus/species and 12% of samples failed to provide any identification. These same bacteria were analyzed using Ce0 2 negative-ion MALDI MS to produce the PCA plot shown in Figure 4. Comparing the inner and outer variance in two dimensions showed good reproducibility and uniqueness between the strains analyzed. Cross validation showed 98% correct classification at the species level and 96% at the strain level. [0059] Table 4. Acinetobacter strains used

Gram Positive Strain Source

A. baumannii ATCC 17976 ATCC

AC54 Felix d'herelle Reference Center Viruses

A. calcoaceticus Mx 70.71

75.53

A. pittii (formerly genomospecies 3) ATCC 19004 ATCC

ATCC 17922

A.nosocomialis (formerly genomospecies ATCC 17903

13)

ATCC 700472

A. haemolyticus ATCC 17907

2213 Felix d'herelle Reference Center Viruses EXAMPLE 2

[0060] Bacterial isolates. Tables 5-7 summarize the bacteria used in this study. All samples were obtained from the Colorado School of Mines collection. Bacteria were streaked on brain heart infusion (BHI) medium (BD-Difco, Franklin Lakes, NJ) from cryogenic freezer stocks and cultured at 37°C for 18 h. as specified in Bruker standard operating procedures for bacterial cultivation.

[0061] Table 5: Staphylococcus species used (All strains were obtained from an in house collection at CSM. JMI designation is from JMI laboratories (North Liberty, IA))

[0062] Table 6: Staphylococcus strains used

[0063] Table 7: Staphylococcus aureus strains used (NCTC designates the National Collection of Type Cultures. MRSA strains are denoted by)

[0064] Lipid extraction. Bacteria were extracted as previously described. Briefly, individual colonies were suspended in 50 μΙ_ of a 1 :2 v/v methanol/chloroform (Pharmco-AAPER, Shelbyville KY and Fischer, Pittsburgh PA, respectively) and vortexed to allow for disruption of the cells. To facilitate phase separation, an equal volume of phosphate buffer saline (PBS) at a pH of 7.4 was added prior to additional vortexing. Extracts were then centrifuged prior to MALDI sample preparation.

[0065] Mass Spectrometry. Sample preparation for MOLI MS analysis was carried out as previously described. Briefly, 100 mg of Ce0 2 (Cermac Inc., Milwaukee, Wl) was suspended in 1 ml_ of n-hexane (Sigma Aldrich) prior to spotting 1 μΙ_ of the resulting slurry on a standard Bruker stainless steel MALDI plate. Two iL of lipid extract was deposited directly on a dried Ce0 2 spot and allowed to evaporate prior to analysis. MOLI-MS measurements were performed with a Bruker Ultraflextreme MALDI-TOF MS (Bruker Daltronics, Billerica, MA) in negative-ion reflectron mode with a grid voltage of 50.3%, a delayed extraction time of 120 ns, and a sampling frequency on a 355 nm Nd:YAG laser of 1 kHz. Three identical biological extracts and two technical replicates of each isolate were analyzed as 500 shot composites using automated laser rastering. Technical extracts were included as internal reproducibility controls to ensure instrument stability.

[0066] Data analysis. Mass spectra were exported as ASCII files and processed using a Python algorithm that selected and centroided 24 specific fatty acid peaks (Table 8), and scaled each peak to total ion intensity. Processed data were written as .xls files for import into R (Ver. 3.0.2, R Foundation, Vienna, Austria) as a data frame. The prcompQ function mean centered and calculated PCA scores before plotting with the built-in plotQ function in R. Leave-one-spectrum-out cross-validation (LOSOCV) was performed using linear discriminant analysis to determine the classification rate.

[0067] Table 8: Fatty Acids used in Principal Component Analysis

C20:0 311

C21:l 323

C21:0 325

C22:l 337

C22:0 339

C23:l 351

C23:0 353

C24:l 365

C24:0 367

C25:l 379

C25:0 381

C26:l 393

C26:0 395

Processed fatty acid profiles were analyzed with MATLAB 2014a (Mathworks, Natick, MA). Generalized prediction rates were measured using 3 Latin partitions and 100 bootstraps. Two classifiers were evaluated: a fuzzy rule-building expert system (FuRES) and partial least squares discriminant analysis (PLS-DA). The PLS-DA algorithm used 2 Latin partitions and 10 bootstraps to calculate the average pooled prediction error. The number of components (i.e., latent variables) that minimized this error was selected and used to build a model from the set of training data, which was then used as a prediction set. Training data consisted of a set of profiles used to build the classifiers and the test data were the set of profiles used to evaluate the performance of these classifiers. Hierarchical cluster analysis was used to generate dendrograms and graphically illustrate the linkage distances

(Euclidean distances) obtained from an agglomerative algorithm. The distances were between pairs of profiles or between the averages of profiles from subclusters.

Results and Discussion [0068] Spectral Analysis. Extracts from 14 Staphylococcus species (Table 5) were analyzed by MOLI MS using Ce0 2 for in situ determination of fatty acid content. For the 14 Staphylococcus species, the spectra (data not shown) contained similar fatty acids with C15:0 as the base peak and the other saturated FAs ranging from 0-30% relative abundance. The intensities of the unsaturated FA peak distribution allowed the spectra to be visually divided into three distinct categories: S. aureus, S. auricularis, S. capitis, S. epidermidis, and S. shleiferi formed one group with similar respective C16:0, C17:0 and C18:0 ratios,, S. haemolyticus, S. hyicus, and S. saprophytics formed a second group with the highest prevalence of unsaturation consisting of 10-38% relative unsaturated FAs, and S. lugdunensis, S. lentus, S. simulans, and S. warneri formed a third group with each having a unique characteristic that allowed differentiation. Figure 5 has two representative spectra for each of the three groups. As visual examples, slight differences in the relative abundance of minor FAs for group one enhanced the differentiation. When comparing the top two spectra (top), S. aureus was differentiated from S. auricularis by the appearance of C17:2 and C20:1 in the latter. The minor FAs were crucial in separating the species in the second group (middle). For example, S. haemolyticus was differentiated from S.

saprophytics using the absence of C20:0 in the later as well as the shifting in abundance between C18:1 and C18:0. The third group (bottom) was most easily distinguished; which may be clearly seen for the case of S. lugdunensis, where C14:0 was the second most abundant FA with respect to C15:0, encompassing 20% of the relative abundance and the appearance of C21 :0 in S. lentus. Visual analysis of the respective ratios of FAs provided a qualitative basis for bivariate analysis of the spectral data, but to process complex data sets more thoroughly multivariate statistics were needed.

[0069] Species-level detection Principal component analysis (PCA) using 24 FA peaks (Table 8) was employed to classify Staphylococcus to the species-level. A score plot of the first three components, which encompassed 93.6% of total variance is shown in Figure 6. Each colored point represents an individual replicate for each bacterial species. The degree of separation is indicated by the distinct clustering of members of the same species (inner variance) and the distance between different species (outer variance). All species clearly plotted in unique space, which is supported by the 100% classification rate obtained by

LOSOCV using linear discriminant analysis (LDA). Figure 7 shows a dendrogram based on Euclidean distances between spectra which demonstrates the overall classification of the profiles into well-defined clusters. The tight grouping of biological and technical replicates as the same strains shows the reproducibility of the extraction method as well as the instrument.

[0070] The FURES tree given in Figure 8 defines 13 rules, which indicates perfect classification. The average prediction results for 100 bootstraps were 98.1 ± 0.3% for FuRES and 90.7 ± 0.3% for PLS-DA. Bootstrap Latin partition validation randomly divided the data into training and test sets such that the training set contained twice the number of data points when compared to the test set. In addition, the validation maintained the same class distribution between training and test sets so that the training set and test set would have the same proportion of objects from each class. For bootstrap analysis, three hundred models were built and evaluated. Because each profile was only used once for prediction per bootstrap, the results of the three Latin partitions were pooled and are comprehensive for all FA profiles. The results from 100 bootstraps were averaged and reported with 95% confidence intervals. This validation approach was much more rigorous than LOOCV, which is a weaker measure with respect to a model's dependence on training set composition and the accuracy of the data within the prediction set.

[0071] Strain-level detection The versatility of MOLI MS was further explored by analyzing extracts of 18 strains (Listed in Table 6). Figure 9 shows a score plot of the first three PCs for this data. A total variance of 95.8% was explained by the first three PCs. In the projection shown in Figure 5, the overlap between some of the groups appeared not to be separated; for example, S. saprophyticus J Ml 9850 and PTX 0652 as well as S.

haemolyticus JMI 5539 and 13091 do not appear to be separated. By rotating through different visual 3D planes that are displayed in 2D on the page, these groups were isolated, but the new projection complicated other separations.

[0072] Overall, CV-LDA on the first 10 PC scores identified 85/90 strains correctly, while misidentifying 5 isolates. Two were misidentified as the incorrect species: one S.

haemolyticus sample was mistaken for S. auricularis and the other was misclassified as S. saprophyticus, resulting in 98% accuracy. At the strain level, three samples were

misidentified: one S. aureus, one S. haemolyticus, and one S. saprophyticus, leading to 97% accuracy.

[0073] FuRES and PLS-DA calculations correctly classified the data into 18 strains. These results were obtained with 100 bootstraps and three Latin partitions. FuRES and PLS-DA had 92.0 ± 0.4% and 87.6 ± 0.5% prediction rates, respectively. From the PCA scores, it was shown that some strains of the same species exhibit significantly different FA

distributions, as shown by the three distinct clusters of three strains of S. hominis in Figure 9. Figure 10 showed tight clustering for most strains. Some of the strains cluster together with other strains of the same species, while the FA profiles of other strains are more closely related to strains of a different species. This trend was more obvious in the FuRES tree in Figure 11. S. hominis JMI 7922, JMI 9382, and JMI 10153 were all correctly classified as the correct strains, as indicated by all 5 replicates for each strain falling under a single rule, Rule #3, #12 and #19, respectively. But, S. hominis JMI 10153 is separated from S.

haemolyticus JMI 1309 by Rule #19 which indicates that these profiles are highly similar; in this case identification of a S. hominis strain did not guarantee a correct identification of S. hominis.

[0074] MRSA Differentiation. Protein MALDI profiling methods utilizing intact cell have shown a series of characteristic peaks for identification of S. aureus. From direct

comparison of reference strains, discrimination between MSSA and MRSA was achieved, but a uniform signature profile could not be identified to allow for unknown classification. To assess utility of MOLI MS FA profiling for antibiotic resistance profiling, 10 S. aureus strains, five ATCC and five clinical isolates, listed in Table 7 were analyzed. A score plot of the first three components explains 97.6% of the total variance is shown in Figure 12. In this projection, all samples were separated into unique groups. Strain-level classification by LDA correctly identified 47/50, leading to 96% accuracy using LOSOCV. The replicates that were misidentified were one ATCC 700563 as ATCC 49476, one JMI 6884 as JMI 7390, and one NCTC 8321 as NCTC 8292. By classifying the individual strains to determine if there were any trend in the FA profiles that would allow for antibiotic resistance classification, it was found that distinct groupings occur between MRSA and MSSA isolates.

[0075] The above data set yielded a FuRES tree with a single rule (figure not shown) indicating perfect separation of the two bacterial classes. Because each of the MRSA and MSSA groups comprised five sample strains with five replicates, the bootstrap Latin partition method partitioned all five replicates for each strain so that no profiles from the same strain would be in both the training and prediction data sets at the same time. The prediction rates for strain-level identification of S. aureus were 98.2 ± 0.2% and 88 ± 1 % for FuRES and PLS-DA, respectively. FuRES discriminant weights for MRSA and MSSA classification revealed that the even fatty acid peaks have a larger impact for MSSA and the odd fatty acid peaks have a larger impact for MRSA (Figure 13). These results are in accordance with other well-known studies that have shown differences in FA composition for antibiotic resistant Enterococcus strains.

[0076] FuRES consistently performed better than PLS-DA, because it is a nonlinear classifier ideally suited for predicting classes that are binary encoded. PLS, which is designed for calibration of continuous variables, may construct ill-conditioned models (ones with poor predictions) when trying to fit the binary encoded target matrix. This problem often occurs with complex data sets.