Rapid microbial identification and antibiotic resistance detection by

Subscriber access provided by YORK UNIV

Article

Rapid microbial identification and antibiotic resistance detection by mass spectrometric analysis of membrane lipids Tao Liang, Lisa M Leung, Belita Opene, William E. Fondrie, Young In Lee, Courtney E Chandler, Sung Hwan Yoon, Yohei Doi, Robert K. Ernst, and David R. Goodlett Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.8b02611 • Publication Date (Web): 20 Dec 2018 Downloaded from http://pubs.acs.org on December 21, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Rapid microbial identification and antibiotic resistance detection by mass spectrometric analysis of membrane lipids Tao Lianga, Lisa M. Leungb,d, BelitaOpeneb, William E. Fondriec, Young InLeeb,Courtney E. Chandlerb, Sung Hwan Yoonb, YoheiDoie, Robert K. Ernstb*, David R. Goodletta* a

Department of Pharmaceutical Sciences, School of Pharmacy, University of Maryland, Baltimore, Maryland, United States Department of Microbial Pathogenesis,School of Dentistry, University of Maryland, Baltimore, Maryland, United States c Center for Vascular and Inflammatory Diseases, University of Maryland, Baltimore, Maryland, United States d Divisions of Microbiology and Molecular Biology, Laboratories Administration, Maryland Department of Health, Baltimore, Maryland, United States e Division of Infectious Diseases, School of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania, United States b

ABSTRACT:Infectious diseases have a substantial global health impact. Clinicians need rapid and accurate diagnoses of infections to direct patient treatment and improve antibiotic stewardship. Current technologies employed in routine diagnostics are based on bacterial culture followed by morphological trait differentiation and biochemical testing, which can be time-consuming and laborintensive. With advances inmass spectrometry (MS) for clinical diagnostics, the U.S. Food and Drug Administration has approved two microbial identification platforms based on matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) MS analysis of microbial proteins. We recently reported a novel and complementary approach by comparing MALDI-TOF mass spectra of microbial membrane lipid fingerprints to identify ESKAPE pathogens. However, this lipid-based approach useda sample preparation method that required more than a working day from sample collection to identification. Here, we report a new method that extracts lipids efficiently and rapidly from microbial membranes using an aqueous sodium acetate (SA) buffer that can be used to identify clinically relevant Gram-positive and –negative pathogens and fungal speciesin less than an hour. The SA method also has the ability to differentiate antibiotic-susceptible and antibiotic-resistant strains, directly identify microbes from biological specimens, and detect multiple pathogens in a mixed sample. These results should have positive implications for the manner in which bacteria and fungi are identifiedin general hospital settings and intensive care units.

The worldwide threat of infectious diseases continues to drive development of novel diagnostics and therapeutics. Rapid and accurate diagnosis of microbial infections enables clinicians to quickly initiate targeted antimicrobial therapy to reduce patient morbidity and mortality. Classicalidentificationmethods in clinical settings rely on bacterial culture, followed by microscopic examination and biochemical testing1,2. Thiscan take a significant amount of time, usually days to even weeksdepending on the microorganism‟s growth rate. Although recent PCR-based and genome sequencing technologies have transformed the field of microbial diagnosis, these techniques suffer from limitations of long processing time, high false positive rate, heavyinformatics burden, and high cost1,3,4. Currently, mass spectrometric platforms targeting microbial products represent a burgeoning technology that simplifies the clinical workflow and reduces time and cost overconventional methodologies5. Matrix-assisted laser desorption/ionization time-of-flight MS (MALDI-TOF MS) platforms currently play a central role as a clinical diagnostic5,6;two of which have been approved for a limited number of microorganisms by the USA‟s Food and Drug Administration (FDA): the VITEK MS (bioMérieux S.A., France)7 and the MALDIBiotyper (Bruker Daltonics, Billerica, MA, USA)8. Both platforms utilize a strategy of generating protein mass spectral profiles of an unknown sample that is compared

against all mass spectra in a reference library for bacterial identification9. However, these protein-based platforms have significant limitations, including: requirement for cell culture to obtain pure colonies10, poor identification at the species level for closely related species6, inability to identify pathogens directly from biofluids,anddifficulty in differentiatingantimicrobial-resistant strains in all under an FDA approved platform11.To overcome these limitations, Leung et al. reported a novel and complementary approach that employs essential bacterial membrane lipids for identification 11. Theydemonstrated that bacterial glycolipids showed speciesspecific characteristics, which could be used as a chemical barcode when analyzed by MS in much the same way that the protein-based platforms are usedto identify bacteria11.Their reported lipid library contained fifty entries including the clinically important ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.). Complex and diverse lipids are a major component of bacterial membranes. For example,in Gram-positive bacteria, the cell envelope consists of complex multi-layers of peptidoglycan enclosing a single membrane, a bilayer of lipids including cardiolipin and lipoteichoic acid (LTA)12. The general structure of LTA varies between species consisting of two or four acyl groups with different carbon chain lengths13 and can

ACS Paragon Plus Environment

Analytical Chemistry 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

be variably modified with alanine or glycosyl linkages depending on bacterial background. In contrast, Gram-negative bacteria have two membranes, one inner and one outer with a thinner layer of peptidoglycan in between12. Lipid A, a unique constituent of the Gram-negative bacterial membrane, is the hydrophobic anchor of lipopolysaccharide (LPS), which forms the majority of the outer leaflet of the outer membrane. The general structure of lipid A is comprised of a β-1‟,6-linked diglucosamine backbone, flanked by phosphate moieties on terminal positions with four to seven attached fatty acyl chains. The variations of fatty acid composition and terminal phosphate modifications result in lipid A molecules being highly diverse across species14. In fungi, the cell wall is mainly composed of polysaccharides, whereas the plasma membrane contains a high fraction of extractable lipids, including sterols, sphingolipids, and glycerophospholipids. Like bacterial cell walls, fungal cell wall composition is a dynamic system that is unique for individual fungal species and subspecies, providing numerous species-specific lipid barcodes15. Efficient extraction of membrane lipids is important to obtain high-quality microbial lipid MS profiles. Leung et al.11 used a method originally developed by El Hamidi et al.16,which isa microextraction protocol involving a single-step ammonium isobutyratereaction at high temperature. Although this method is efficient, it requiresa lyophilization step prior to obtaining MS results, which significantly increases the total processing time. In addition, the use of noxious chemicals requiring a fume hood and multiple centrifugation steps limit its use in a clinical laboratory or hospital. To address the limitations of this and other available methods17–22, we present here a novel and clinic-friendly lipid extraction protocol that reduces the time from sample collection to identification to less than an hour.For this purification method, a sample is resuspended in a sodium acetate (SA) lysis buffer and briefly heated to release the signature lipids from microbial membranes followed byan organic solvent extraction to harvest the liberated lipids before MS analysis. A Design of Experiments (DOE) process was performed to optimize the SA method for lipid extraction. In order to be useful clinically, the new method had to preserve the ability of previously reported library11 to detect antibiotic resistance and to identify microbes from biological fluids without need for culture. Thus, to evaluate our SA extraction method, we analyzed the clinicallyrelevant ESKAPE pathogens,a selection ofbacteria and fungi that are commonly identified in nosocomial infections, and colistin-resistant ESKAPE pathogens expressing the mcr-1 gene in trans23. In Gram-negative bacteria, modification of the negatively charged, terminal phosphate moieties can result in resistance to colistin (polymyxin E), a cationic antimicrobial peptide. Plasmid-mediated mcr-1, encoding a phosphoethanolamine transferase induces colistin resistance in the Gram-negative ESKAPE pathogens by addition of phosphoethanolamine(PEtN) to lipid A.Besides, some K. pneumoniae and P. aeruginosastrains accomplish this by the addition of aminoarabinose(AraN) to lipid A.Finally, clinical diagnosis would significantly benefit by removal of the need to culture organisms from patient specimens prior to pathogen identification. Finally, we used mono- and poly-microbial urine samples that were spiked with bacteria pathogens clearly showing the utility of our SA method for the direct identification of pathogens from a complex biological fluid.

Bacterial strains. Clinically relevant bacterial strains were selected for this study including Gram-negative bacteria (E. coli, K. pneumoniae, P. aeruginosa,and P. mirabilis), Gram-positive bacteria (S. aureus), and fungi (Candida albicans) were tested. Additionally, the ESKAPE pathogens and corresponding mcr-1expressing strains with the addition of intrinsically colistin-resistant species, Serratia marcescens, Morganella morganii, and Proteus mirabilis were analyzed. Strain identities were determined by culture-based phenotyping and confirmed by protein phenotyping via the Bruker MALDI Biotyper. Clinical isolates of E. coli, K. pneumoniae, P. aeruginosa, S. aureus, and C. albicans collected from UTI patients were identified by conventional culture-based phenotyping and confirmed by the MALDI Biotyper at the University of Pittsburgh Medical Center. A complete list of species and strains are included in Supplementary Table S1 and S2. All bacterial species were grown in Lysogenic broth (LB)overnight with shaking after inoculation from agar plates. All mcr-1 containing specieswere cultured in LB media with 50 µg/mL gentamicin to maintain the plasmid expressing the mcr-1 gene. Reagents and materials.Commercially available monophosphoryl lipid A (MPLA; #699800 Avanti Polar Lipids, Inc., Alabaster, AL, USA) was used as an internal standard for Design of Experiment (DOE) study. A stock solution of MPLA was prepared in a 1:2, v/v mixture of methanol/chloroform at a concentration of 1 mg/mL. SA buffer was prepared by dissolving in endotoxin-free water to give 100 mM. The pH was adjusted by adding acetic acid and checked by a pH meter. Lipid microextraction from cells using SA method.The use of SA buffer for bacterial cells was previously reported by Zhou et al.24to study lipid A structures. Here, we further optimized it to allow rapid bacterial identification in an hour or less. Overnight cell cultures (1 mL) were harvested andresuspended in400µL of 100mM SA buffer (pH 4.0) and incubated for 10 to 30 minutes at 100ºC. Samples were vortexed every 510 minutes. After incubation, samples were cooled on ice to room temperature and centrifuged at 8,000 × g for 5 minutes. Supernatants were discarded and pellets were washed with 1 mL 95% ethanol. Insoluble lipids were extracted and reconstituted in 50-100 µL chloroform/methanol/water (12:6:1, v/v/v) mixtures and centrifuged at 5,000 × g for 5 minutes. Aliquots of 0.75 µL of the supernatant were spotted onto a MALDI stainless steel target plate with 0.75 µL pre-spotted 10 mg/mL norharmane dissolved in chloroform/methanol/water mixture (v/v/v 12:6:1) for MS analysis. In the DOE study, MPLAwas spiked into extracts as an internal standard at a final concentration of 20 µg/mL. Lipid microextraction from cells using El Hamidi method. The original protocol was described in detail by EI Hamidi et al.16. Briefly, bacterial pellets were treated with 400 µL of a mixture of 70% isobutyric acid/1 M ammonium hydroxide(5:3,v/v) and incubated at 100°C for 1 hour. Reaction mixtures were spun down at 2,000 × g for 15 minutes, and supernatants were transferred to clean tubes, combined with a 1:1 ratio of distilled water, frozen, and lyophilized overnight. The lyophilized residues containing lipids of interest were washed twice with 1 mL methanol and then resuspended in 50-100 μL chloroform/methanol/water (12:6:1,v/v/v) solvent mixture for MALDI-TOF MS analyses.

MATERIALS AND METHODS


Page 2 of 8

Page 3 of 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry MALDI-TOF MS data acquisition.Mass spectra were recorded in negative ion mode using a Bruker Microflex LRF MALDI-TOF MS (Bruker Daltonics Inc.) operated in the reflectron mode. For SA method, each mass spectrum was acquired at 42% global laser power with 900 shots summed per spectrum. Each sample was acquired in triplicate for further statistical analyses. El Hamidi et al. method mass spectra were acquired in the similar manner as the SA method.Typically, 900–1000 laser shots were summed to acquire each mass spectrum11. DOE Data Analysis. E.coli strains expressing mcr-1 were used for method optimization25. Two factors with multiple levels: incubation time (10, 20, 30, 60, and 120 minutes) and pH (3.0, 3.5, 4.0, and 4.5) were considered as key parameters for optimization using a DOE approach. The lipid A ion intensity ratios were calculated by referencing intensities of bacteriallipid A signature ions, which are those ions unique to an organism,to the spiked MPLA ion intensity. Specifically, each lipid A ion intensity including fragment ions was normalized to the MPLA ion intensity and calculated usingthe EquationIas follows. Lipid A ion ratio =

lipid A ions

int .

− lipid A fragment ions MPLA ionint .

int .

The ratios were then used as y output for DOE analysis. Minitab17 (Minitab Inc., State College, PA, USA) DOE function was used to quantitatively compare ratios across combinatorial tested conditions. The optimal condition was determined by the DOE results. Monomicrobial and polymicrobial UTI model analyses.Sterile urine was purchased from LEE Biosolutions Inc. (Maryland Heights, MO, USA). Commonly identified UTI bacterial pathogens were selected and cultured in 5 mL LB broth. After overnight culture in LB, bacterial cell counts were measured by enumeration and expressed as colony forming units (CFU). One mL of liquid culture (108-9 CFU) was seeded into 9 mL of pre-warmed (37°C) sterile urine to mimic an infection condition, and spiked urine samples were directly processed using the SA method without further culture. For polymicrobial mixtures, individual isolates were spiked into sterile urine after growth in monoculture in LB. Specifically, polymicrobial infection models using three different organismsthat commonly presented in UTI sampleswere mixed together, specifically: Mixture1 (M1) - S. aureus, P. aeruginosa, andK. pneumoniaeorMixture2 (M2) - C. albicans, E. coli, and K. pneumoniae. The SA method was employed to directly process polymicrobial pellets. Statistical analysis.MALDI mass spectra were imported and processed using the MALDIquant (v1.16.4) and MALDIquantForeign (v0.11) R packages26 in R (v3.4.2). Each mass spectrum was square root transformed and smoothed with 41 point Savitzky-Golay filter27. Mass spectra were baseline removed using the Statistics-sensitive Non-linear Iterative Peakclipping algorithm28 over 100 iterations and then aligned. Lipids were detected by selecting m/z ions with signal-to-noise (S/N) ratios greater than 8, with noise calculated by the median absolute deviation and binning matching mass ions between spectra using a 0.5 m/z tolerance. For heat map comparison and correlation analysis, a Pearson‟s correlation coefficient was calculated from pairwise vectors created from a mass list of ions from each mass spectrum. All scripts used for this analysis are freely available and can be found in the fol-

lowingGitHub repository:https://github.com/Tao-LiangUMB/Rapid-Extraction-Project MALDI Biotyper Analysis.First, the MALDI OC Biotyper software (version 4.0.19) was used to construct a custom urine lipid library. Next, mass spectra were filtered using filterSpectra.R scrip (https://github.com/wfondrie/ESKAPE_Glycolipid_MS/R/filte rSpectra.R) to reject low-quality mass spectra (i.e. suboptimal numbers of ions or low signal-to-noise ratios) from inclusion in the library. Mass spectra were loaded into the Biotyper OC software; parameters were adjusted to recognize lipid mass spectra by lowering the mass range and peak picking thresholds. A Main SPectra (MSP) was created for each species that includes peak positions, intensities and frequencies based on multiple replicates per species11. A full list of species and strains isolated from UTI patients for urine library construction areincluded in Supplementary Table S2. Second, mass spectra generated by the SA method were loaded into the Biotyper OC software for identification. The urine library was used as a reference database for spectra matching. SA method mass spectra were processed by MALDI Biotyper for peak picking and generating a mass list for further database comparison. The software then computes log scores and generates a list of potential identifications with log score values sorted from high to low. To interpret the identification results, we followed the criteria indicated in the Biotyper user manual (2.00 to 3.00 considered as positive identification; 1.70 to 1.99 considered as probable identification; 0 to 1.69 considered as unreliable identification). RESULTSANDDISCUSSION Microbial membrane lipids are highly abundant and can be readily extracted, similar to ribosomal proteins used in the existing MALDI diagnostic platforms11. Here, we report results of using a DOE process to shorten thetime from sample collection to organism identification from a working day to less than an hour and in some cases, less than thirty minutes making our lipid-based identification method clinically viable. Rapid SA method optimization. Our main goal was to devise a rapid extraction method with minimal experimental steps thatmaintained the high quality of lipid mass spectra produced previously11,while shortening time from extraction to identification. During the DOE process, two modules were designated for optimization: (i) a purification module focused on selecting the most efficient wash solvent to remove cellular debris and unwanted molecules (see Screening of Purification Solvents section in Supporting Information) and (ii)an incubation module focused on optimizing use of SA buffer pH and incubation time. The choice of SA buffer circumvented the need to use ammonium isobutyrate, thus eliminating the need for a fume hood, which is not routinely available in clinical microbiology laboratories. Determination of optimal incubation condition using DOE. To determine optimal conditions,a colistin-resistant strain of E. coli, via in trans expression of the mcr-1-gene was used as the model strain to insure colistin resistancesignature ions (i.e. those with PEtN modified lipid A) were not lost during extraction.Two key parameters were selected for optimization: SA buffer pH and incubation time. In total, four pH conditions (3.0, 3.5, 4.0, and 4.5) and five incubation times (10, 20, 30, 60, and 120 minutes) were selected and analyzed using a general 4 x 5 full factorial design. With biological triplicates, a total of 60 discrete experiments were performed. Each bio-



logical replicate (20 runs) was grouped in the general full factorial design for statistical analysis. In order to quantitatively track changes, E. coli lipid A ions were normalized tothe signal from internal standard, MPLA that was added to each sample after extraction. The ion observed at m/z 1919 (∆m/z 123) represents a PEtN modificationof the hexa-acylated E. coli lipid A ion at m/z1796, and is associated with colistin resistance.The ions observed at m/z 1716 and 1839 were assigned as fragment ions due to the loss of a phosphate group (∆ m/z = 80) from precursor ions m/z 1796 and 1919, respectively25(Supplementary Figure S2). Ratios of each extraction condition, as calculated using Equation 1, were imported into Minitab software for DOE analysis. The DOE results (Figure 1) show that multiple extraction conditions (pH 4.0, 30 min; pH 4.5, 60 min; and pH 4.0, 4.5, 120 min) fall in the darkest red region, where the lipid A ion ratios are in the range of 3.0 to 4.0. The calculated lipid A ion ratio allowed us to determine a condition that would extract the highest amount of intact lipid A molecules, while preserving labile moieties such as phosphate and PEtN. In this analysis, we observed that when moreacidic conditionswereused (pH 3.0 and 3.5), intensities of lipid A fragment ions were increased due to the loss of labile moieties from intact lipid A. However,the use of SAbuffer at pH 4.0, which falls in the range of mild acid 29 was shown to efficiently liberate lipid A fromLPS after a 30-minute incubation and maintain the antibiotic resistance associated ions. While increased incubation times did increase the yield of lipid A, we chose pH 4.0 and a 30-minute incubationtime as the optimal extraction condition for further experiments (Figure 1, green arrow highlighted). Interestingly, a 10-minute incubation time at pH 4.0 also generated high quality mass spectra, includingdetection of antibioticresistance-associated ions (Supplemental Figure S2), suggesting that the overall identification time could be reduced further.While the SA buffer works well on cleavage of acid-labile Kdo2 bond of tested Gram-negative bacteria in this study, some rare species of lipid A equipped with acid-stable Ko bond may be resistant to such condition30, resulting in the need of alternative approaches for different structural models such as gas-phase release31.

Figure 1. Design of Experiment derived contour plot of lipid A ions ratio vs pH, and time. Plot reveals optimal condition for hydrolysis.Each black point indicates a mean (n

= 3) of output response from each extraction condition. Darker red regions indicate higher ratios of lipid A ions. Rapid SA Method Evaluation. We next evaluated analytical performance of the SA method to ensure it:i)had comparable sensitivity to the El Hamidi-based method (SeeDetermination ofLimit of Detection section from Supporting Information),ii)allowed differentiation between antimicrobialsusceptible and -resistant strains, and iii)generated mass spectra similar to the mass spectra from the El Hamidi-based extraction methodon which the library of Leung et al. was based11. Rapid SA method allows detection of antimicrobial resistance.Antibiotic resistance detection is a critical limitation of current commercial protein-based MALDI-TOF platform.Using the SA method, we were able to distinguish between colistin-susceptible and -resistant strains, selected and grouped according to resistance phenotype, as determined by the minimum inhibitory concentration method11,25. Four Gramnegative bacteria were examined for the presence of colistinresistance associated ions. mcr-1-containing P. aeruginosa strains show a unique ion at m/z 1569, which is not present in susceptible strains (Figure 2aand 2c). The mass shift, m/z 1446 to 1569 is the result of a PEtN (Δm/z 123) addition to lipid A. The ion at m/z 1489 represents loss of a phosphate moiety from m/z 1569 ion.We note that this ion is not observed in susceptible strains and may represent an additional diagnostic marker of resistance11,25. In addition to the mcr-1 encoded resistance phenotype, we examined additional strains (K. pneumoniae and M. morganii) that show resistance through addition of AraN, a five carbon amino sugar.Figure 2band 2d show modification of lipid A by AraN (Δm/z 131) attached on the terminal phosphate group of hexa-acetylatedK. pneumoniae lipid A (m/z1824), resulting in a modified lipid A structureat m/z 1955(Figure 2d).Similarly, Supplementary Figure S5a shows a representative mass spectrum from M. morganiiwithAraN-modified lipid A ions from m/z 1796 to 1927.Finally, Supplementary Figure S5b and S5cpresentdata showing the viability of even a10-minute incubationtime forP. aeruginosa and K. pneumoniae, antibioticresistant and –susceptible,respectively. SA method mass spectra show high similarity to El Hamidi-based method mass spectra. The lipid-based microbial identification library built within the FDA-approved MALDI Biotyper platform demonstrated an ability to identify allESKAPE pathogens11.We evaluatedcompatibility of mass spectra generated via theSA method to mass spectra in the lipid library11 thatweregenerated using the extraction method of El Hamidi et al.16 The similarity of mass spectra generated by the two methods was derived using Pearson‟s correlation, commonly used to compare two acquired mass spectra32–34. Figure 3 shows the Pearson‟s correlation between the two methods using the SA method at pH 4.0 and 30-minute incubation. This analysis showed that eight out of ten strains‟ had Pearson‟s correlation coefficients higher than 0.6, a value previously shown to be considered a „strong correlation‟35. Similar results were obtained after 10- and 20-minute incubations (Supplementary Figure S6). These results suggest mass spectra generated by the new SA method can be used to search against the existing lipid library.


Page 4 of 8

Page 5 of 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Figure 2. Mass spectra of lipids extracted by the 30-minute incubation SA method for colistin-susceptible and -resistant strains of (a) P. aeruginosa and (b) K. pneumoniae.(c)TheMCR-1 modified lipid A structure (m/z 1569) of P. aeruginosa corresponds to a PEtN addition (red highlighted moiety) to the structure at m/z 1446. Loss of phosphate group is indicated by m/z 1489 from PEtN modified lipid A m/z 1569. (d) An AraN addition (red highlighted moiety) to K. pneumoniae lipid A (m/z 1824) results in a colistin resistance-associated ion at m/z 1955.

Figure 3. Pearson’s correlation analysis of mass spectra produce by both methods differentiate the ESKAPE pathogens.Mass spectra from the 30-minute SA method were compared with our lipid library reference mass spectra generated by the El Hamidi method. Positive correlations are displayed in red circles with Pearson‟s correlation coefficients labeled, whereas negative correlations are represented in blue circles. A coefficient of 0.0 indicates that there is no correlation. To understand the strength of correlation, the guide from Evans35 suggested an absolute value of r: 0.20-0.39 as “weak”, 0.40-0.59 as “moderate” 0.60-0.79 as “strong” and 0.80-1.0 as “very strong”. The size and color intensity of the circles are proportional to the correlation coefficients. Non-statistically significant coefficients of circles were removed and left blank.

* and ** indicate colistin-resistant strains with PEtN and AraN additions to the lipid A, respectively. SA method mass spectra permit bacterial identification. We further assessed functionality of the SA method with additional bacterial and fungal species. For these analyses, mass spectra for each organismwas compared to themselves and all others using Pearson‟s correlation. A heat map of representative microbes is presented in Figure 4 including eleven Gramnegative bacteria with corresponding antibiotic-resistant strains, three Gram-positive bacteria, and two fungal species. Mass spectra obtainedusing the SA methodshowed anability to identify each organism with the highest Pearson‟s correlation coefficient obtained when compared with themselves (indicated as black squares,Figure 4).Notably, colistin-susceptible and -resistant strains were clearly differentiated, as shown using both PEtN-modified lipid A of mcr-1-expressing strains of A. baumannii, E. coli, K. pneumoniae, P. aeruginosa, and Cronobactersakazakii and the AraN-modified lipid A ofK. pneumoniae, M. morganii, P. mirabilis, and S. marcescens. To determine if alterations in growth temperature affect ability to positively identify organisms through their lipid A signature ions, Francisellanovicida and Yersinia pestis were grown at 25°C and 37ºC, which mimic insect and mammalian growth conditions, respectively. After SA method extraction, these organisms were distinguishedfrom all other organismsas growth temperature induced carbon chain length change of fatty acidsattached to the diglucosamine backbone of lipid A36,37. Finally, we also showed that eukaryotic fungi (Candida subspecies) are differentiated from each other and all the other bacterial species. Similar results were obtained after 10- and 20-minute incubations (Supplementary Figure S7).



Figure 4. Pearson’s correlation analysis of SA method mass spectra from 30-minute incubation for species differentiation. Mass spectra were acquired from the ESKAPE pathogens, intrinsically colistin-resistant bacteria, and clinically relevant fungi. Pearson‟scorrelation coefficient is considered as a mass spectral similarity score, where 1.0 represents an identical match (black squares). White squares represent a score of 0.0 where there is no correlation. (*) indicate mcr-1containing colistin-resistant strains with a PEtN addition to lipid A and (**) indicate intrinsically colistin-resistant strains with an AraN-modified lipid A. Microbial names are color coded in black for Gram-negative bacteria, cyan for Grampositive bacteria and red for fungi. Figure 4 demonstrates that the SAmethod can readily extract lipid A molecules with modifications associated with colistin resistance within an hour, specifically PEtN and AraN modified lipid A, which may have a positive impact in overcoming existing FDA-approved platforms in antimicrobial resistance detection.Importantly, the SA method showed comparable analytical performance to the El Hamidi-based method used by Leung et al. to develop their library in terms of mass spectral similarity, indicating that the existing lipid library could be coupled with the new, faster SA method for bacterial identification. Regarding sensitivity, the limit of detection of both methods was determined to be less than 1 µg dry cells orabout 106 CFU to generate detectable lipid A signals.Currently, the protein-based approach has a routinely better sensitivity (~105 CFU)38,39 than our lipid-based approach, which indicatesroom for further improvement in limit of detection. Bacterial identification directly from urine samples.We next examined applicability of the SA method for direct identificationof microorganisms from urine,as thecurrent protein-based platforms require culture on a solid growth medium prior to identification.UTIs are one of the most common bacterial infections40with over 8 million UTIs reported annually in the United States41. Gram-positive bacteria, Gramnegative bacteria, and fungi are all causative agents of UTIs, and often UTIs are polymicrobial infections42.To evaluate the SA method, we chose C. albicans, E.coli, K. pneumoniae, P. aeruginosa, and S. aureusto represent bacterial backgrounds prevalent in patients with UTIs40. Figure 5a-e shows mass spectra for each organism produced using the SA method after spiked in sterile urine. Membrane lipid signature ions were

observed from each tested species. For example, C. albicans showed a series of unique lipid ions at m/z 1087, 1103, 1423, and 1429 (Figure 5a) and a cluster of ions at m/z 1446, 1462, and 1616 that were assigned to P. aeruginosa (Figure 5d). A control mass spectrum containing sterile urine alone (Figure 5f) demonstrated that there are minimal background contaminants that would impede microbial identification within the m/z range of interest (m/z1000-2400). Shorter incubation times (10 and 20 minutes) also worked for the species tested below (Supplementary Figure S8). Importantly,the SA method successfully extracted mcr-1 modified lipid A from urine when spiked with three antibiotic-resistant species, i.e., E.coli, K. pneumoniae, and P. aeruginosa(Supplementary Figure S9), indicating a broader impact of this method in clinical diagnostic. Twenty-fiveadditional microorganisms previously directly isolated from patient urine samples were imported into MALDI Biotyper to build a urine reference library(Supplementary Table S2). Results from analysis of urine samples spiked with bacteria were searched against this urine library for identification. For this analysis, we interpreted our lipid-based MS results using the Biotyper confidence log score as manufacturer recommends for proteinbased mass spectra (see MALDI Biotyper Analysis section). Mass spectra obtained from C. albicans, E. coli, K. pneumoniae,P. aeruginosa,and S. aureus were correctly identified with log scores > 2.00 (high-confident identification) using the DOE optimized SA method (Figure 5g). Supplementary Table S3 provides a full list of MALDI Biotyper log score from other extraction conditions, such as 10- and 20-minute incubation. Here, we have shown that the SA method has the potential to directly identifymonomicrobial infectious agents extracted fromhuman urine. Importantly, for more rapid analysis, urine samples spiked with bacteria were processed without the need for any extra steps that are required in the protein-based method(e.g.differential centrifugation to remove human cells or filtration to remove salts43,44).Furthermore, the control mass spectrum generated bylipid extraction from urine showed no host lipids presence in the m/z range of interest, illustrating potential advantage of the SA method to analyze patient samples directly without need for culture to obtain a pure colony required by the protein-based methods. Polymicrobial infection analysis from UTI model.Many UTIs present as polymicrobial infections with more than one pathogen in urine42,45, which can not be individually identified using the current protein-based MS microbial identification platforms46. Thus, we evaluated whether the SA method could detect species-specific lipids direct from specimen when multiple microorganisms were present in urine. Causative agents frequently isolated in clinical specimens from UTIpatients were selected for two types of UTIs: uncomplicated and complicated40. Clinically, uncomplicated UTIs are classified as infected individuals who are healthy with neither structural norfunctional abnormalities in their urinary tracts47, whereas complicated UTI patients have risk factors of a compromised urinary tract or host defense48.To determine the efficiency of the SA method to correctly identify organisms in a polymicrobial infection, two mixtures of clinically relevant bacterial strains were selected for this study: M1 - K. pneumoniae,P. aeruginosa and S. aureus andM2 - E.coli,K. pneumoniae,and C. albicans. In this experiment, each pathogen was-


Page 6 of 8

Page 7 of 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry makes it less than ideal for analyzing the much more sparsely cultured separately and then spiked together into sterile urine populated lipid ion mass spectra. Notably, while the Bruker and extracted directly using the optimized SA method.Figure MALDI Biotyper software performs reasonably well with 6a shows themass spectrum of a polymicrobial sample (M1) mass spectra from mono-microbial samples, it generally fails containing S. aureus, P. aeruginosa, and K. pneumoniae. Sigwhen confronted with polymicrobial data. Here we have internature ions were labeled and assigned to each organism. For S. preted polymicrobial data manually by denoting the presence aureus, m/z at 1324, 1338, 1352, and 1366 (Figure 6a) were of organism specific signature ions. To address this concern, previously reported by Leung et al.11.Ions in the mass range of we are exploring the suitability of machine learning to imm/z 1400 to 1600 including ions at m/z1446, 1462 and 1616 prove the algorithm for identification of each individual sperepresent lipid A ions that are representative of P. aeruginosa. cies in polymicrobial samples49. Signature ions of K. pneumoniae were successfully detected in both UTI mixturesamples at m/z 1824, 1840, 2063, and 2079 CONCLUSIONS (Figure6a and 6b).Figure 6b shows the mass spectrum of a The need for rapid and accurate diagnostics that positivepolymicrobial sample (M2) containingC. albicans, E. coli, and ly impact patient outcomes and improve overall public health K. pneumoniae. E. colilipid A signature ion is observed at m/z has never been greater. This work presents a rapid microbial 1796; C. albicans signature ions are detected at m/z 1087, lipid extraction method that, when coupled with thepreviously 1103, and 1115, as well as at m/z 1424 and 1449. By assigning reported MS-based lipid library11allows microorganisms to be these signature ions to each individual species, we could deciidentified from mono- or poly-microbial specimens and direct pher the complicated mass spectra to identify multiple microfrom urine in under an hour. Importantly, the new extraction organisms from a single polymicrobial sample. Moreover, protocol preserves the enhanced functionality of the lipid platmass spectra from shorter incubation times (10 minutes) of form for antibiotic resistance detection, direct analysis from uncomplicated and complicated UTI models are shown to be biofluids, and identification of single microbes from polymiviable for identification (Supplementary Figure S10a and crobial samples while shortening the time-to-result from days S10b). to minutes. This improved platform may be a powerful and Here, we note that the lipid library used was built inside complementary approach to augment the performance of existthe Bruker MALDI Biotyper software platform using an algoing FDA-approved protein-based commercial systems. rithm specifically designed for protein fingerprinting. This Figure 5.Microbial identification directly from spiked urine samples by the MALDI Biotyper. Mass spectra (a - f) were generated by SA method extraction with 30-minute incubation. Signature lipid ions are labeled with m/z values for each species: (a)C. albicans,(b)E. coli, (c)K. pneumoniae, (d)P. aeruginosa, and (e)S. aureus. A sterile urine control mass spectrum is shown in (f). TheMALDI Biotyperscores were obtained by searching mass spectra from the urine samples at each incubation time point (10, 20, and 30 minutes) against the lipid urine library constructed in the MALDI Biotyper(g). The green dashed line indicates the cutoff threshold for a positive identification. Mean of log scores were calculated from triplicate data.

Figure 6. Mass spectra of polymicrobial UTI samples with a 30-minute incubation extraction. (a)M1- S. aureus, P. aeruginosa and K. pneumoniae in sterile urine. (b)M2 - C. albicans, E. coli and K. pneumoniae spiked in sterile urine. Species-specific lipid ions are assigned to each organism, respectively. Corresponding areas of spectra are color highlighted as follows: S. aureus (light blue), P. aeruginosa (light yellow), K. pneumoniae (light red), E. coli (sky blue) and C. albicans (light green). Notes

ASSOCIATED CONTENT Supporting Information Additional information as noted in text. The Supporting Information is available free of charge on the ACS Publications website.

DRG and RKE have a significant financial interest in Pataigin LLC., the company developing this diagnostic technology for rapid bacterial identification. All other authors declare no competing financial interests.

ACKNOWLEDGMENT

AUTHOR INFORMATION Corresponding Author *DRG, Email: [email protected]. Phone: 4107061490.*RKE, Email: [email protected]: 4107063622.

We thank the National Institutes of Health Grant 1R01GM111066-01 (DRG and RKE) and 5R01AI104895-05 (YD) for funding and support.

REFERENCES


Analytical Chemistry (1)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(2)

(3) (4)

(5) (6) (7)

(8) (9)

(10) (11)

(12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22)

(23) (24) (25)

Clark, A. E.; Kaleta, E. J.; Arora, A.; Wolk, D. M. Clin. Microbiol. Rev.2013, 26 (3), 547–603. Van Belkum, A.; Durand, G.; Peyret, M.; Chatellier, S.; Zambardi, G.; Schrenzel, J.; Shortridge, D.; Engelhardt, A.; Dunne, W. M. Annals of Laboratory Medicine. 2013, pp 14–27. Bertelli, C.; Greub, G. Clin. Microbiol. Infect.2013, 19 (9), 803– 813. Espy, M. J.; Uhl, J. R.; Sloan, L. M.; Buckwalter, S. P.; Jones, M. F.; Vetter, E. A.; Yao, J. D. C.; Wengenack, N. L.; Rosenblatt, J. E.; Iii, F. R. C.; et al. Clin. Microbiol. Rev.2006, 19 (1), 165–256. Sauer, S.; Kliem, M. Nat. Rev. Microbiol.2010, 8 (1), 74–82. Croxatto, A.; Prod‟hom, G.; Greub, G. FEMS Microbiol. Rev.2012, 36 (2), 380–407. Lévesque, S.; Dufresne, P. J.; Soualhine, H.; Domingo, M. C.; Bekal, S.; Lefebvre, B.; Tremblay, C. PLoS One2015, 10 (12), 1–21. Maier, T.; Klepel, S.; Renner, U.; Kostrzewa, M. Nat. Methods2006, 3 (4), 1–2. Opota, O.; Prod‟hom, G.; Greub, G. In MALDI-TOF and Tandem MS for Clinical Microbiology, First Edition; 2017; Vol. 3, pp 55–91. Patel, R. Clin. Chem.2015, 61 (1), 100–111. Leung, L. M.; Fondrie, W. E.; Doi, Y.; Johnson, K. J.; Strickland, D. K.; Ernst, R. K.; Goodlett, D. R. Sci. Rep.2017, 7 (1), 6403–6413. Silhavy, T.; Kahne, D.; Walker, S. Cold Spring Harbor perspectives in biology. May 2010, pp 1–16. Morath, S.; von Aulock, S.; Hartung, T. J. Endotoxin Res.2005, 11 (6), 348–356. Miller, S. I.; Ernst, R. K.; Bader, M. W. Nat. Rev. Microbiol.2005. Brown, L.; Wolf, J. M.; Prados-Rosales, R.; Casadevall, A. Nat. Rev. Microbiol.2015, 13 (10), 620–630. Hamidi, A. El; Tirsoaga, A.; Novikov, A.; Hussein, A.; Caroff, M. J. Lipid Res.2005, 46 (8), 1773–1778. Li, Y.; Wang, X.; Ernst, R. K. Rapid Commun. Mass Spectrom.2011, 25 (18), 2641–2648. Zhou, P.; Chandan, V.; Liu, X.; Chan, K.; Altman, E.; Li, J. J. Lipid Res.2009, 50 (9), 1936–1944. Larrouy-Maumus, G.; Clements, A.; Filloux, A.; McCarthy, R. R.; Mostowy, S. J. Microbiol. Methods2016, 120, 68–71. W.J.Dyer, E. G. B. and. Can. J. Biochem. Physiol.1959, 37 (21), 911–917. Kawasaki, K. J. Microbiol. Methods2009, 76 (3), 313–315. Chafchaouni-Moussaoui, I.; Novikov, A.; Bhrada, F.; Perry, M. B.; Filali-Maltouf, A.; Caroff, M. Rapid Commun. Mass Spectrom.2011, 25 (14), 2043–2048. Santajit, S.; Indrawattana, N. Biomed Res. Int.2016, 2016. Zhou, P.; Hu, R.; Chandan, V.; Kuolee, R.; Liu, X.; Chen, W.; Liu, B.; Altman, E.; Li, J. Mol. Biosyst.2012, 8 (3), 720–725. Liu, Y. Y.; Chandler, C. E.; Leung, L. M.; McElheny, C. L.;

(26) (27) (28) (29) (30)

(31)

(32) (33) (34)

(35) (36) (37) (38) (39)

(40) (41) (42) (43) (44)

(45) (46) (47) (48) (49)

Page 8 of 8 Mettus, R. T.; Shanks, R. M. Q.; Liu, J. H.; Goodlett, D. R.; Ernst, R. K.; Doi, Y. Antimicrob. Agents Chemother.2017, 61 (6), 00580-17. Gibb, S.; Strimmer, K. Bioinformatics2012, 28 (17), 2270–2271. Savitzky, A.; Golay, M. J. E. Anal. Chem.1964, 36 (8), 1627– 1639. Nucl. Instruments Methods Phys. Res. Sect. B Beam Interact. with Mater. Atoms1988, 34 (3), 396–402. Caroff, M.; Tacken, A.; Szabó, L. Carbohydr. Res.1988, 175 (2), 273–282. Hashimoto, M.; Ozono, M.; Furuyashiki, M.; Baba, R.; Hashiguchi, S.; Suda, Y.; Fukase, K.; Fujimoto, Y. J. Biol. Chem.2016, 291 (40), 21184–21194. Oyler, B. L.; Khan, M. M.; Smith, D. F.; Harberts, E. M.; Kilgour, D. P. A.; Ernst, R. K.; Cross, A. S.; Goodlett, D. R. J. Am. Soc. Mass Spectrom.2018, 29 (6), 1221–1229. Li, S.; Arnold, R. J.; Tang, H.; Radivojac, P. Anal. Chem.2011, 83 (3), 790–796. Degroeve, S.; Maddelein, D.; Martens, L. Nucleic Acids Res.2015, 43 (W1), W326–W330. Yilmaz, Ş.; Vandermarliere, E.; Martens, L. In Proteoin Bioinformatics; Shivakumar, K., Mathivanan, S., Eds.; Humana Press, New York, NY: New York, 2017; pp 75–100. Evans, J. D. Straightforward Statistics for the Behavioral Sciences; Brooks/Cole Pub. Co, 1996. Matsuura, M. Front. Immunol.2013, 4 (MAY), 1–9. Gunn, J. S.; Ernst, R. K. Ann. N. Y. Acad. Sci.2007, 1105, 202– 218. Kostrzewa, M. Expert Rev. Proteomics2018, 15 (3), 193–202. Ferreira, L.; Sanchez-Juanes, F.; Gonzalez-Avila, M.; Cembrero-Fucinos, D.; Herrero-Hernandez, A.; GonzalezBuitrago, J. M.; Munoz-Bellido, J. L. J. Clin. Microbiol.2010, 48 (6), 2110–2115. Flores-Mireles, A. L.; Walker, J. N.; Caparon, M.; Hultgren, S. J. Nat. Rev. Microbiol.2015, 13 (5), 269–284. Dielubanza, E. J.; Schaeffer, A. J. Med. Clin. North Am.2011, 95 (1), 27–41. Davenport, M.; Mach, K. E.; Shortliffe, L. M. D.; Banaei, N.; Wang, T. H.; Liao, J. C. Nat. Rev. Urol.2017, 14 (5), 298–310. DeMarco, M. L.; Burnham, C. A. D. Am. J. Clin. Pathol.2014, 141 (2), 204–212. Kohling, H. L.; Bittner, A.; Muller, K.-D.; Buer, J.; Becker, M.; Rubben, H.; Rettenmeier, A. W.; Mosel, F. J. Med. Microbiol.2012, 61 (3), 339–344. Kline, K. A.; Lewis, A. L. Microbiol. Spectr.2016, 4 (2). de Vos, M. G. J.; Zagorski, M.; McNally, A.; Bollenbach, T. Proc. Natl. Acad. Sci.2017, 114 (40), 10666–10671. Hooton, T. M. N. Engl. J. Med.2012, 366 (11), 1028–1037. Lichtenberger, P.; Hooton, T. M. Current Infectious Disease Reports. November 2008, pp 499–504. Fondrie, W. E.; Liang, T.; Oyler, B. L.; Leung, L. M.; Ernst, R. K. Sci. Rep.2018.

Table of the Contents Graphic ACS Paragon Plus Environment

8

Rapid microbial identification and antibiotic resistance detection by

Recommend Documents