Integrated profiling of Gram-positive and Gram-negative probiotic genomes, proteomes and metabolomes revealed small molecules with differential growth inhibition of antimicrobial-resistant pathogens

Abstract

Probiotics produce small molecules that may serve as alternatives to conventional antibiotics by suppressing growth of antimicrobial resistant (AMR) pathogens. The objective of this study was to identify and examine antimicrobials produced and secreted by probiotics using ‘omics’ profiling with computer-based metabolic flux analyses. The cell-free supernatant of Gram-positive Lacticaseibacillus rhamnosus GG (LGG) and Gram-negative Escherichia coli Nissle (ECN) probiotics inhibited growth of AMR Salmonella Typhimurium, Escherichia coli, and Klebsiella oxytoca ranging between 28.85 – 41.20% (LGG) and 11.48 – 29.45% (ECN). A dose dependent analysis of probiotic supernatants showed LGG was 6.27% to 20.55% more effective at reducing AMR pathogen growth when compared to ECN. Principal component analysis showed clear separation of ECN and LGG cell free supernatant metabolomes. Among 667 metabolites in the supernatant, 304 were differentially abundant between LGG and ECN probiotics. Proteomics identified 87 proteins, whereby 67 (ECN) and 14 (LGG) showed differential expression as enzymes related to carbohydrate and energy metabolic pathways. The whole genomes and metabolomes were next used for in-silico metabolic network analysis. The model predicted the production of 166 metabolites by LGG and ECN probiotics across amino acid, carbohydrate/energy, and nucleotide metabolism with antimicrobial functions. The predictive accuracy of the metabolic flux analysis highlights the novel utility for profiling probiotic supplements as dietary-based antimicrobial alternatives in the control of AMR pathogen growth.

1. Introduction

Probiotic microorganisms, naturally abundant in fermented foods, have been largely explored for their capacity to inhibit pathogen growth. Broad-acting mechanisms by which probiotics antagonize pathogens include competitive exclusion in host tissues, production of organic acids, and modulation of host immunity [1, 2]. More recently, there is evidence for probiotic species and strain-dependent differences in antimicrobial resistant (AMR) pathogen growth suppression [1, 3].

Two established probiotics used as dietary supplements include the Gram-positive probiotic Lacticaseibacillus rhamnosus GG (LGG), and the Gram-negative probiotic Escherichia coli Nissle (ECN), which have been well-characterized for inhibition of enteric pathogens [4]. Emerging research support that both LGG and ECN produce small molecules in the gastrointestinal tracts of pigs and other animals which reduced human rotavirus diarrhea, decreased Salmonella infection in the murine colon, and antagonized pathogenic Enterobacteriaceae in a murine model of gut dysbiosis [5]. Despite the evidence of antagonism between LGG and ECN on viral and bacterial pathogens, research exploring their application to drug-resistant pathogens is limited. Additionally, multi-omic comparisons between different probiotic strains in suppression of AMR pathogen growth with mechanistic understandings are lacking.

Metabolomics, the systematic examination of small molecules, has been used to support the characterization of antimicrobial compounds produced by probiotics. Multiple investigations identified bioactive small molecules, including amino acids, carbohydrates, and lipids, that are produced by LGG and ECN that contribute to anti-pathogen actions [6–10]. Proteome and genome markers in LGG and ECN include genes regulating the production of lipids, amino acids, and energy metabolites for anti-cancer, anti-inflammatory, and pathogen-suppressive capacities [6, 8, 11, 12]. The integration of genomic, proteomic, and metabolomic datasets from distinct probiotic strains and supplements is needed to enhance the global understanding of metabolism, and specifically metabolites secreted after probiotic-mediated production [13].

This study was designed to evaluate LGG and ECN metabolic diversity using a multi-omics lens to characterize antimicrobial metabolites responsible for AMR pathogen growth suppression. The hypothesis was that differential nutrient metabolism by LGG, and ECN leads to production of distinct antimicrobial metabolites with dose-dependent differences in growth suppression of AMR pathogens. Exploring metabolic differences between LGG and ECN for antimicrobial functions will provide new information for commercial dietary supplement production and for direct application to curb existing and emerging AMR pathogen strains in humans and animals.

2. Materials and Methods

2.1. Antimicrobial resistant pathogen isolation.

The Salmonella enterica serovar Typhimurium isolate used in this study was collected from the human intestinal tract at Washington State University in 2010 and provided as a generous gift from Dr. Sangeeta Rao, Colorado State University. The AMR E. coli and K. oxytoca isolates were collected from environmental water samples in Northern Colorado using published methods [14]. Briefly, water samples were collected using sterile Pyrex wide-mouth storage bottles which were immediately placed on ice and kept in a light-sensitive container until analysis approximately 1h following sample collection. Water samples were diluted into CHROMagar-ESBL (extended-spectrum beta-lactamase) and CHROMagar-KPC (Klebsiella pneumoniae carbapenemase) (DRG Diagnostics, Springfield, NJ) media to identify and isolate individual colonies. Isolated colonies were incubated in tryptic soy broth (TSB) at 37°C for ~18 h. Colony identities were made to species-level using matrix-assisted laser desorption-ionization time-of-flight analysis (MALDI) on a VITEK-MS machine (Biomerieux, Durham, NC).

2.2. Antimicrobial susceptibility testing of Salmonella Typhimurium, E. coli, and K. oxytoca.

The AMR profiles of Salmonella Typhimurium, E. coli, and K. oxytoca were established using Kirby-Bauer Disc Diffusion method established by the Clinical and Laboratory Standards Institute (CLSI) [15]. Briefly, overnight incubations of each isolate cultured in sterile TSB were diluted to a concentration of 1.5 × 10⁸ cells/mL using a 0.5 McFarland Standard. The resultant dilutants were spread onto Mueller-Hinton agar (Hardy Diagnostics, Santa Maria, CA). The following antimicrobial discs were applied onto the plates: Meropenem (MEM-10), Linezolid (LZD-30), Vancomycin (VA-30), Cefazolin (CZ-30), Ciprofloxacin (CIP-5), Gentamicin (CN-10). Ampicillin (AMP-10), Penicillin (P-10), Tobramycin (NN-10), Tetracycline (TE-30), and Amikacin (AK-30). After 18 h incubation at 37°C, the zone of inhibition was measured and reported as the radius from the center of the disc to the edge of the inhibition zone (mm). The Kirby-Bauer Disc Diffusion assay was performed in triplicate for each pathogen, and the zone of inhibition averaged across assays. These averaged antimicrobial disc inhibition zones were compared to CLSI standards for each isolate to make the determinations of “Susceptible”, “Intermediate”, and “Resistant” [15].

2.3. Probiotic cultures and cell-free supernatant preparation.

The E. coli Nissle 1917 and L. rhamnosus GG ATCC 53103 isolates used for the experiments were provided by Dr. Lijuan Yuan at the Virginia Polytechnic Institute and State University. Cell-free supernatant was prepared as described previously [16]. Briefly, 1 × 10⁷ colony forming units (CFU) of each probiotic isolate was propagated in deMan Rogasa Sharpe (MRS) broth (Beckton, Dickinson and Company, Difco Laboratories, Franklin Lakes, NJ) for 24 h at 37°C. Cultures were centrifuged at 4000 x g for 10 minutes, and the supernatant decanted from the cellular pellet. The resultant supernatant was then centrifuged and decanted again using the same conditions as the initial round and titrated to a pH of 4.5 using a 1 mol*L⁻¹ solution of NaOH (Sigma Aldrich, St. Louis, MO) with a pH meter (Corning Pinnacle 530, Cole-Parmer, Vernon Hills, IL). The titrated supernatant was then filtered through a 0.22 μM-pore filter (Pall Corporation LifeSciences, Port Washington, NY) before being stored at −80°C prior to use. Three biological replicates of supernatants were prepared independently and used in the subsequent analyses.

2.4. Pathogen growth assays and probiotic cell-free supernatant treatments.

S. Typhimurium, E. coli, and K. oxytoca isolates were thawed and grown in the presence of probiotic cell-free supernatant as described previously [16]. Frozen −80°C stocks of each pathogen were thawed and grown to early/mid exponential phase using a Cytation3 plate reader (BioTek Instruments Inc., Winooski, VT). Approximately 2 × 10⁵ CFU/mL of pathogen was inoculated into 180 μL of sterile Luria Bertani (LB) broth in a 96-well plate. The following concentrations of cell-free supernatant from LGG and ECN were added to wells inoculated with pathogen: 25% v/v (60 μL), 22% v/v (50 μL), 18% v/v (40 μL) and 12% v/v (25 μL). These supernatant concentrations were guided by previous dose-dependent treatments screened against the related pathogen, S. enterica serovar Typhimurium strain 14028s [16]. Equivalent concentrations of sterile MRS and sterile LB were used as a vehicle control and negative control respectively. To control for pH-dependent effects on pathogen growth suppression, all supernatant and control media were adjusted to a pH of 4.5 before use in all assays, and this adjustment reflects the average pH of LGG supernatant preparations used herein (data not shown). Pathogen growth in the presence of cell-free supernatant was measured every 20 minutes for 18 h on a Cytation3 plate-reader using optical density read at a wavelength of 600 nm (OD600). To quantify growth difference at each timepoint, percent growth change was calculated by comparing pathogen growth in the presence of a supernatant treatment versus the vehicle control using the following equation: Percent Growth Inhibition = ((OD600_CFS – OD600_Vehicle) / (OD600_Vehicle)) * 100

For each pathogen, the growth assay was repeated at least 3 times with three technical replicates of each probiotic supernatant concentration. A repeated measures two-way analysis of variance (ANOVA) was used to compare treatment optical densities at each time point and p-values were adjusted using a Tukey post-test to control for multiple comparisons. A p-value of p < 0.05 was defined as statistically significant. Each supernatant concentration was compared between LGG and ECN for each pathogen (e.g., 25% LGG vs ECN CFS for S. Typhimurium, E. coli, or K. oxytoca).

2.5. Probiotic cell free supernatant metabolomics.

To establish the small molecule profiles of L. rhamnosus GG and E. coli Nissle cell-free supernatants, the global, non-targeted metabolome of each was determined by Metabolon Inc © (Durham, NC) using previously described methods [16]. Briefly, three replicates each of LGG and ECN supernatant, representing three independent supernatant collections, and three replicates of sterile MRS broth were sent to Metabolon on dry ice and stored in liquid nitrogen. Prior to extraction, the protein content of each sample was removed using an 80% ice-cold (−80°C) methanol aqueous solution coupled with vigorous shaking for two minutes and subsequent centrifugation at 680 x g for 3 minutes. Samples were then divided into five parts for analysis using ultra-high-performance liquid-chromatography tandem mass-spectrometry (UPLC-MS/MS) and consisted of: two aliquots for reverse phase UPLC-MS/MS analysis with positive ion mode electrospray ionization (ESI), one aliquot for reverse phase UPLC-MS/MS analysis with negative ion mode ESI, one aliquot for hydrophilic interaction (HILIC)/UPLC-MS/MS with negative ion mode ESI, and one backup aliquot. Each aliquot was evaporated using a TurboVap ® solvent evaporation system (Zymark, Hopkinton, MA) to remove organic solvent and stored under nitrogen before subsequent analysis.

For UPLC processing, each sample was injected into a Waters ACQUITY UPLC column using solvents optimized for the five aliquot run analyses described above. For the reverse phase UPLC-MS/MS with positive ion mode ESI analysis, one aliquot of each sample was gradient-eluted using a C18 column (Waters UPLC BEH C18–2.1×100mm, 1.7μm) with a water and methanol mobile phase containing 0.05% v/v perfluoropentanoic acid and 0.1% v/v formic acid. A second aliquot for analysis using UPLC-MS/MS with positive ion mode ESI was gradient-eluted using the afore-mentioned C18 column with a mobile phase of methanol, acetonitrile, water, 0.05% v/v perfluoropentanoic acid, and 0.01% formic acid. For the reverse phase UPLC-MS/MS analysis with negative ion mode ESI, an aliquot of each sample was gradient-eluted using a separate C18 column with a mobile phase of methanol, water, and 6.5 mM of ammonium bicarbonate at a pH of 8.0. HILIC-UPLC-MS/MS with negative ion mode ESI for each sample was performed on a HILIC column (Waters UPLC BEH Amide 2.1 × 150 mm, 1.7 μm) with a mobile gradient of water, acetonitrile, and 10 mM of ammonium formate at a pH of 10.8.

Following gradient elution, all samples were subjected to MS/MS processing using a Thermo Scientific Q-Exactive high resolution/accurate mass spectrometer operating with an Orbitrap mass analyzer at 35,000 mass resolution and coupled with heated electrospray ionization source. Mass spectral scans utilized dynamic exclusion with both MS and data-dependent MSⁿ scans to detect peaks, with a scan range covering approximately 70–1000 m/z. Raw data from MS/MS scans were peak-identified and processed for quality control using proprietary Metabolon .NET systems that compared data to known sources of artifacts and background noise inherent to each UPLC-MS/MS run type. Peak identification was made by comparing peaks to an internal database of ~3,300 purified chemical standards using retention indices, m/z ratios (within +/− 10 ppm of the purified internal standards), and MS/MS forward and reverse match scores. Peaks that did not match a purified internal standard but had a retention index and m/z that was not artifact or background noise were reported as “unknown” in subsequent analyses.

2.6. Metabolite normalization, statistical analysis, and visualization.

The raw abundances for each detected metabolite were normalized using area under the curve analysis. The raw abundance of each metabolite was divided by the median raw abundance of the metabolite across the dataset. One-way ANOVA was used to compare the median-scaled abundance of each metabolite, and statistical significance was defined as p < 0.05. To account for false positives, q-values were calculated, and metabolites with q > 0.1 were excluded from downstream analysis. Pathway enrichment scores (PES) were calculated to evaluate the contribution of metabolic pathways to metabolite profile differences between treatments using the formula below, where “k” indicates the number of significant metabolites in the metabolic pathway, “m” indicates the total number of metabolites in the metabolic pathway, “n” indicates the number of significant metabolites in the entire data set, and “N” indicates the total number of metabolites in the data set:

$$PES=\raisebox{1ex}{$\left(k\right)/\left(m\right)$}\!\left/ \!\raisebox{-1ex}{$\left(n\right)/\left(N\right)$}\right.$$

Median-scaled abundances were additionally used to calculate metabolite fold differences by dividing the average median-scaled abundance of a metabolite in one sample by its average median scaled abundance in a second sample type (e.g. average metabolite abundance in ECN supernatant versus average metabolite abundance in LGG supernatant). Metabolite visualization was performed using Metaboanalyst ® (version 4.0) with R version 3.6.1, using the raw abundance of each metabolite generated by Metabolon (R-script File S1) [17]. A principal coordinates analysis plot was generated using metabolite median-scaled abundances. A heat map with hierarchical clustering analysis was generated using Euclidean and Ward differential clustering algorithms, where red boxes indicate metabolites that were elevated in ECN compared to LGG, and blue boxes indicate metabolites that were decreased in ECN versus LGG. The heat map visualizes the 50 metabolites with the largest statistical differences when comparing ECN versus LGG.

2.7. Probiotic cell free supernatant proteomics.

The non-targeted proteomes of each probiotic supernatant and sterile growth media (MRS) were generated by the Colorado State University Bioanalysis and Omics (ARC-BIO) Facility team using LC-MS/MS. Proteins were isolated from supernatant in a 1:4 v/v suspension of ice-cold (−80°C) methanol. The resultant protein pellets were washed in 100% ice-cold (−80°C) acetone and centrifuged at 15,000 x g for 10 minutes. Following two rounds of acetone washing, samples were air-dried and reconstituted in 2 M urea and bath sonicated for five minutes. To collect insoluble material, sonicated samples were centrifuged at 4000 x g for two minutes. For quantitation, aliquots were diluted 1:2 and 1:5 in 2 M urea solvent and total protein concentration was measured using a Pierce Bicinchoninic Acid Protein Assay (Thermo Scientific, Waltham, MA) following manufacturer’s instructions. Approximately 50 μg of each sample was subjected to trypsin digestion using the methods described by Schauer et al. 2013 [18]. Briefly, 50 μg of each sample was reconstituted in a solution containing 8 M urea, 0.2% v/v ProteaseMax™ surfactant trypsin enhancer (Promega, Madison, WI), 5 mM dithiothreitol, and 5 mM iodoacetic acid. Purified trypsin (Pierce MS-Grade, Thermo Scientific, Waltham, MA) was added at a 1:28 ratio to the sample proteins, and the slurry was incubated at 37°C for 3h, after which trypsin was deactivated using 5% trifluoroacetic acid. Desalting occurred using Pierce C18 spin columns following manufacturer instructions (Thermo Scientific, Waltham, MA). The eluates were dried in a vacuum evaporator and reconstituted in 5% v/v acetonitrile and 0.1% v/v formic acid. Total peptide quantification was determined for each sample resuspension on a NanoDrop (Thermo Scientific, Waltham, MA) at a wavelength of 205 nm and normalized using an extinction coefficient of 31 [19].

Reverse phase chromatography was performed using water with 0.1% formic acid and acetonitrile with 0.1% formic acid. A total of 0.75 μg of peptides was purified and concentrated using an on-line enrichment column (Waters Symmetry Trap C18 100Å, 5 μm, 180 μm ID x 20 mm column). Subsequent separation was performed using a reverse-phase C18 nanospray column (Waters, Peptide BEH C18; 1.7 μm, 75 μm x 150 nm column) at 45°C, and samples were eluted using a 30-minute mobile phase gradient of 3 to 8% formic acid over 3 minutes, followed by 8% to 35% of acetonitrile with 0.1% formic acid solution over 27 minutes, at a flow rate of 350 nL/min. A Nanospray Flex ion source (Thermo Scientific, Waltham, MA) introduced eluate directly into the mass spectrometer (Orbitrap Velos Pro™, Thermo Scientific, Waltham, MA). Spectra were collected using positive ion mode over a range of 400 – 2,000 m/z, and MS/MS was performed on ions assigned a charge state of 2+ or 3+ using a dynamic exclusion limit of 2 MS/MS spectra of a given m/z value for 30 seconds (exclusion duration of 90 s). Fourier-Transformation mode (60,000 resolution) was applied for MS detection, and ion trap mode was applied for the subsequent MS/MS with 35% normalized collision energy. Spectra were generated using Xcalibur 3.0 software (Thermo Scientific) with a S/N threshold of 1.5 and 1 scan/group.

2.7.1. Proteome identification and normalization.

MS/MS spectra for each sample were extracted, charge state deconvoluted and deisotoped by ProteoWizard MsConvert (version 3.0). All spectra were then screened for protein identities using Mascot (Matrix Science, London, UK, version 2.6.0) with a fragment ion mass tolerance of 0.80 Daltons and a parent ion tolerance of 20 ppm. Carboxymethylation of cysteine was specified in Mascot as a fixed modification. Deamidation of asparagine and glutamine, methylation of lysine and arginine, hydroxylation of proline, oxidation of methionine, dimethylation of lysine and arginine and acetylation of the n-terminus were specified in Mascot as variable modifications. The following reverse concatenated Uniprot reference proteomes were used for the search: Uniprot_Yeast_rev_022119, Uniprot_Sus_scrofa_rev_022119, Uniprot_Bovine_rev_022119. LGG supernatant samples were additionally screened with the Uniprot_Lactobacillus_rhamnosus_GG_rev_021819 database and ECN supernatant samples were also screened with the Uniprot_Escherichia_coli_Nissle_rev_021819 database. Identified spectra were further combined using the probabilistic protein identification algorithms utilized by Scaffold (version 4.8.4, Proteome Software Inc., Portland, OR) [20, 21]. The peptide probability threshold was set (90%) such that a peptide false discovery rate of 0.0% was achieved based on hits to the reverse database [22]. Protein identifications were accepted if they could be established at greater than 95% probability as assigned by the Protein Prophet algorithm and contained at least two identified peptides [23]. Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. For each protein, raw abundances were used to derive the normalized abundance factor (NSAF) within each sample, a method used to estimate the protein content within a single sample or gel band. NSAF is calculated using the number of spectra (SpC) identifying a protein divided by the protein length (L), referred to as Spectral Abundant Factor (SAF) and then normalized over the total sum of spectral counts/length in each analysis.

2.8. Metabolic modeling of E. coli Nissle and L. rhamnosus GG.

Draft metabolic models for ECN and LGG were reconstructed using apps in DoE KBase. iML1515, a published metabolic model for E. coli K-12 MG1655, was used as a reference model for ECN, and a model for L. casei ATCC 344 was used as a reference model for LGG [24–26]. The genomes for the four organisms were accessed through the KBase interface for NCBI genomes. The KBase application ‘Compare Two Proteomes’ was used to find orthologs between the strain [27]. The application ‘Propagate Model to New Genome’ was then used to translate the model for E. coli K-12 to ECN and the model for L. casei ATCC 344 to LGG, respectively. Each model was translated using ortholog comparisons, and gap-filling was performed to ensure biomass production. Further gap-filling was performed manually to include pathways for the consumption and production of the metabolites detected in the supernatant metabolome. To simulate flux distributions consistent with the supernatant metabolome data for ECN and LGG, an optimization problem was constructed to solve parsimonious flux balance analysis (pFBA) for both models simultaneously [28]. Within the models, variables controlled for included metabolite availability in the MRS media and uptake/export of a metabolite by ECN or LGG based on the relative abundance from the metabolome data. Ten thousand flux distributions were simulated under randomly sampled maximum substrate uptake (which were not measured during the experiments) using pFBA maximization of biomass production in the models as the objective function [29]. File S2 provides the complete mathematical formulation used to construct this modeling analysis. The shadow price for each metabolite was retrieved from the solution of the optimization problem. All simulations were performed in MATLAB R2017b using the COBRA toolbox, the optimization solvers GUROBI, and IBM CPLEX (Corporation, 2017; Gurobi Optimizer Reference Manual, 2020) [30]. For each metabolite, the efficacy of the model in predicting metabolites identified in the supernatant metabolome was required to meet both of the following criteria: the metabolite was present in at least 50% of the metabolic flux distributions and the metabolite was identified in LGG vs ECN supernatant.

3. Results

3.1. L. rhamnosus GG and E. coli Nissle cell free supernatants demonstrate differential inhibition of AMR pathogen growth.

E. coli, S. Typhimurium, and K. oxytoca pathogens were screened for phenotypic AMR against five representative drug classes using Kirby-Bauer disk diffusion method (Table 1). All three pathogens were resistant to the beta-lactam drugs ampicillin and cefazolin. In addition, S. Typhimurium displayed multidrug resistance by displaying additional resistance to the aminoglycoside drug gentamicin as well as tetracycline.

Table 1.

Kirby-Bauer disk diffusion for antimicrobial resistant pathogens

	S. Typhimurium	E. coli	K. oxytoca
Antimicrobial Agent	Zone Diameter (mm), (Designation)
Amikacin (AK-30)	26.9 ±2.3, (S)	24.3±2.3, (S)	31.3±2.1, (S)
Ampicillin (AMP-10)	6.0 ± 0.0, (R)	6.0 ± 0.0, (R)	0.0±0.0, (R)
Cefazolin (CZ-30)	6.0 ± 0.0, (R)	7.8±4.5, (R)	14.2±9.6, (R)
Ciprofloxacin (CIP-5)	26.3±3.8, (S)	25.8±5.3, (S)	22.9±4.4, (I)
Gentamicin (CN-10)	7.1±3.2, (R)	20.2±2.7, (S)	18.5±3.3, (S)
Linezolid (LZD-30)	6.0 ± 0.0, (NA)	6.2±0.4, (NA)	1.8±2.6, (NA)
Meropenem (MEM-10)	32.6 ± 0.79, (S)	34.3±1.0, (S)	40.2±2.5, (S)
Penicillin (P-10)	6.0 ± 0.0, (NA)	8.5±2.5, (NA)	0.8±3.4, (NA)
Tobramycin (NN-10)	17.4 ±3.9, (S)	23.7±1.2, (S)	31.0±2.4, (S)
Tetracycline (TE-30)	9.6 ±7.0, (R)	17.2±4.0, (I)	28.5±3.1, (S)
Vancomycin (VA-30)	6.3 ±0.76, (NA)	6.7±0.8, (NA)	0±0.0, (NA)

Values represented mean ± standard deviation of the zone of inhibition diameter measured in millimeters (mm).

Designations are defined as Susceptible “S”, Intermediate “I” or Resistant “R” based on standards defined by the Clinical and Laboratory Standards Institute (CLSI) for Enterobacteriaceae species.

“NA” indicates cutoff value for “S”, “I”, or “R” not defined by the CLSI.

LGG and ECN supernatants decreased AMR S. Typhimurium, E. coli, and K. oxytoca growth in a dose dependent manner from 12% v/v to 25% v/v. Figure 1 shows the minimum inhibitory cell free supernatant dose-response for each pathogen, which was defined as the dose of supernatant that reduced pathogen growth when compared to the vehicle control. Subsequently, the percent growth inhibition was calculated for each probiotic supernatant relative to the vehicle control. Across all three AMR pathogens tested and for each probiotic supernatant concentration, LGG was 6.27% to 20.55% more effective at reducing pathogen growth when compared to ECN (Figure 1). Figure S1 shows the quantification and comparisons for ECN and LGG growth reduction of each pathogen at 4 h intervals. The minimum supernatant dose at which both LGG and ECN supernatants achieved growth reduction of S. Typhimurium was 12% v/v (Figure 1A). LGG reduced S. Typhimurium growth between 4.33 h - 16.00 h (p < 0.0001), and ECN between 13.33 h - 16.00 h (p < 0.05) when compared to the vehicle control treatment. Maximal S. Typhimurium growth reduction of 41.20% was achieved at 5.33 h (p < 0.0001) for LGG and 11.48% for ECN at 13.67 h (p < 0.01). LGG supernatant was 6.27% more effective at reducing S. Typhimurium growth compared to ECN at 8.33 h (p < 0.0001). The 18% v/v supernatant dose was the lowest dose where the LGG and ECN supernatant resulted in a difference in E. coli growth compared to the vehicle control (Figure 1B). At this dose, the LGG and ECN supernatant reduced E. coli growth by 30.40% at 4.67 h (p < 0.0001) and 29.45% at 5 h (p < 0.0001) compared to the vehicle control respectively. When comparing probiotic supernatants, LGG was 20.55% more effective than ECN at 16.00 h (p < 0.0001) at inhibiting E. coli growth. For K. oxytoca, the 12% v/v supernatant was the lowest dose where LGG and ECN achieved growth reduction versus the vehicle control (Figure 1C). At this dose, LGG resulted in a growth delay of K. oxytoca growth between 3.00 h −8.00 h and achieved a maximal percent growth inhibition of 28.85% n at 7.33 h (p < 0.0001). ECN inhibited K. oxytoca growth earlier than LGG, between 3.00 h −16.00 h and reached maximal growth reduction of 23.86% at 3.33 h (p < 0.005). When comparing probiotic supernatants, LGG was 19.30% more effective than ECN at 11.00 h (p < 0.0001) at reducing K. oxytoca growth.

Figure 1.

AMR pathogen growth reduction by ECN and LGG probiotic cell free supernatants. Figures depict the growth curves of S. Typhimurium, E. coli and K. oxytoca recorded over 18 hours under the minimum inhibitory dose (supernatant volume/total volume *100) of probiotic cell free supernatant. Bacterial abundance is reported through optical density readings at a wavelength of 600 nm (OD600). The minimum supernatant doses at which both L. rhamnosus GG (LGG) and E. coli Nissle (ECN) supernatants achieved growth reduction for S. Typhimurium, E. coli and K. oxytoca were the 12%, 18% and 12% respectively. Maximal Salmonella growth reduction was achieved at 5.33 h for L. rhamnosus GG (41.20% p < 0.0001) (Dashed line- L) and at 13.67 h for E. coli Nissle (11.48%, p < 0.01) (Dashed line- E). For pathogenic E. coli maximum growth reduction for LGG supernatant was 30.40% and occurred at 4.67 h (p < 0.0001). For the E. coli Nissle supernatant, maximal pathogenic growth reduction occurred at 5.00 h at 29.45% (p < 0.0001). L. rhamnosus GG reduced K. oxytoca growth between 3.00 h-16.00 h and achieved a maximal percent growth of 28.85% reduction at 7.33 h (p < 0.0001). E. coli Nissle reduced K. oxytoca growth between 3.00 h-16.00 h and reached maximal growth reduction of 23.86% at 3.33 h (p < 0.01). Dashed lines indicate maximum growth reduction observed for LGG (black) or ECN (blue).

3.2. E. coli Nissle and L. rhamnosus GG cell metabolite profiles differ across amino acid, carbohydrate, energy, and nucleotide metabolites.

Differences in growth reduction against AMR pathogens between LGG and ECN supernatant (Figure 1) provided rationale for conducting a global, non-targeted metabolomics analysis to identify bioactive small molecules in the probiotic cell free supernatants. A total of 667 metabolites were detected in the LGG and ECN supernatant metabolomes, and the major differences between LGG and ECN are depicted in Figure 2. Statistically significant metabolites profiled in this study are shown in Table S1. Among the 667 detected metabolites, 412 metabolites had confirmed identifications. Two-hundred and fifty-five metabolites were unnamed and reported with mass to charge ratio (m/z) and retention index (RI). Principal component analysis showed separation of ECN and LGG cell free supernatant metabolomes (Figure 2A) with 494 differentially abundant metabolites (p < 0.05) (Figure 2B). Metabolites with unique representations to each probiotic (i.e., 5 metabolites in LGG and 6 metabolites in ECN) were minor contributors to the metabolite profile differences identified between LGG and ECN supernatant (Figure 2B). When organized into chemical classes and metabolic pathways, these differentially abundant metabolites included 102 amino acids, 28 peptides, 23 carbohydrates, 9 energy metabolites, 41 lipids, 55 nucleotides, 16 cofactors and vitamins, 29 xenobiotic/other metabolites, and 191 unknown/unidentified metabolites (Figure 2C). The 50 metabolites with the lowest p-values across ECN versus LGG supernatant are shown in Figure 2D.

Figure 2.

Global, non-targeted metabolomes of L. rhamnosus GG and E. coli Nissle cell-free supernatant. A. Principal component analysis of L. rhamnosus GG (LGG) and E. coli Nissle (ECN) supernatant and vehicle control media. Each circle represents a biological replicate. B. Venn diagram illustrating metabolite presence versus absence differences in ECN versus LGG along with metabolites not present in the vehicle control (MRS broth) when compared to probiotic supernatants. C. Heat map of 50 metabolites ranked according to magnitude of fold-differences between ECN and LGG. D. Pathway enrichment scores for metabolic pathways that contributed to significantly different metabolites when comparing ECN versus LGG.

Amino acids represented the largest metabolic pathway in LGG and ECN metabolomes and accounted for ~23.2% of metabolites in the supernatant. Of these amino acid metabolites, ~20.6% were differentially abundant when comparing LGG and ECN supernatant. The polyamine metabolic pathway (PES 1.10) revealed that ECN produced higher levels of polyamines compared to LGG. The polyamine cadaverine, a lysine derivative, was 37.46- times higher in ECN versus LGG supernatant (p < 1.00E-30). The polyamine agmatine, a metabolite of arginine, was 11.35-fold higher in ECN versus LGG supernatant. Additional enrichment of arginine metabolites was identified in arginine, proline, and urea cycle metabolism (PES 1.38). In this metabolic pathway, arginine (0.58-fold lower in ECN versus LGG, p < 1.00E-7), ornithine (4.73-fold higher in ECN versus LGG, p < 1.00E-11), and citrulline (2.85-fold higher in ECN versus LGG, p < 1.00E-13) distinguished LGG and ECN metabolite profiles.

Additional metabolite classes distinguishing ECN and LGG supernatant included carbohydrates, energy metabolites, and nucleotides, which together accounted for ~17.6% of differentially abundant metabolites when comparing ECN and LGG supernatants. Carbohydrate and energy metabolism differences between ECN and LGG supernatant involved glycolysis and gluconeogenesis (PES 1.21) and the tricarboxylic acid (TCA) cycle (PES 2.48). LGG had increased levels of glycolytic metabolites compared to ECN, including the glycolysis intermediate glucose 6-phosphate (0.060-fold lower in ECN versus LGG, p < 1.00E-30), and the fermentation product lactate (0.31-fold lower in ECN versus LGG, p < 1.00E-10). Downstream of glycolysis, ECN and LGG differentially produced metabolites involved in the tricarboxylic acid cycle (TCA cycle). Citrate was significantly more abundant in LGG versus ECN supernatant (0.010-fold lower in ECN versus LGG, p < 1.00E-30), while the ECN supernatant metabolome had a significantly higher abundance of succinate (8.98-fold higher in ECN versus LGG, p < 1.00E-30) and its oxidized intermediate fumarate (2.58-fold higher in ECN versus LGG, p < 1.00E-04). Among nucleotides, metabolites in the metabolic pathways for guanine metabolism (PES 2.47) and uracil (PES 2.80) contributed to the largest treatment differences between ECN and LGG. Uracil, the pyrimidine product of thymine demethylation, was 73.22-fold higher in ECN versus LGG supernatant (p < 1.00E-11), and uridine, a metabolite of uracil, was 0.01 fold decreased in ECN versus LGG ( p < 1E-30). Guanine, the purine nucleobase, was 19.57-fold higher in ECN versus LGG supernatant (p < 1.00E-11), and guanosine, a metabolite of guanine, was 0.02-fold decreased in ECN versus LGG supernatant (p < 1E-30).

3.3. Proteomic composition differences between E. coli Nissle and L. rhamnosus GG supernatants.

The non-targeted proteome of ECN and LGG cell-free supernatants was explored for mechanistic contributions to AMR pathogen growth reduction (Figure 3). Table S2 shows probiotic supernatant proteomes. The complete probiotic cell free supernatant proteome, protein accession numbers, and gene-ontology terms are available upon request. Forty-nine of these proteins were from animal origin, arising from culture media-broth, and were excluded from downstream analysis. Of the remaining proteins, 67 had an ECN origin and 14 proteins were from LGG. Only one protein, glyceraldehyde 3- phosphate dehydrogenase, was identified in both ECN and LGG supernatants (Figure 3A).

Figure 3.

Non-targeted proteome of E. coli Nissle and L. rhamnosus GG supernatants. A. Venn diagram shows the number of proteins identified in E. coli Nissle (ECN) supernatant, L. rhamnosus (LGG) supernatant, and sterile MRS broth. Percent relative abundances of proteins found in cell free supernatants of B. L. rhamnosus GG supernatant and C. E. coli Nissle supernatant.

The LGG supernatant proteome classifications are shown in Figure 3B. Notably, the glycolysis enzyme glyceraldehyde 3-phopshate dehydrogenase represented 1.47% of the LGG supernatant proteome with glycoside hydrolase enzyme representing more than 18% of the total protein abundance of the sample. Other proteins identified included the CHAP (cysteine and histidine-dependent aminohydrase/proteases) and a hydrolase domain protein (2.73%). The remaining 3 proteins detected in the LGG supernatant were uncharacterized. In the ECN supernatant-proteome, proteins involved in carbohydrate metabolism (14 proteins) and amino acid metabolism (8 proteins) were identified (Figure 3). Elevated carbohydrate metabolism proteins were glycolysis enzyme glyceraldehyde 3-phosphate (1.93% abundance) and enolase (1.81%), another glycolytic protein. Major contributors to amino acid metabolism included the aspartate metabolism enzyme aspartate ammonia lyase (3.59% of total proteome abundance) and glutamine-binding periplasmic protein (3.28% abundance) that is responsible for glutamine transport.

3.4. Metabolic modeling predictions for metabolites contributing to AMR pathogen growth reduction from E. coli Nissle and L. rhamnosus cell-free supernatants.

In silico metabolic modeling was used to explore the differential production of metabolites by ECN and LGG in standard nutrient media (MRS). Figure 4 shows the simulated flux distributions for LGG and ECN, which reflect their differential use of MRS media and the predicted metabolites each probiotic generates, and Table S3 details each metabolite and the percent of model simulations it was identified in. Of the 667 metabolites detected in the supernatant metabolome, 204 metabolites were present in at least one of the 10,000 reconstructed simulations. One hundred sixty-five metabolites were present in at least 50% of the 10,000 reconstructed simulations, including 61 amino acids, 27 peptides, 18 carbohydrates, 8 energy metabolites, 13 lipids, 27 nucleotides, 5 cofactors and vitamins, and 6 xenobiotic/other metabolites.

Figure 4.

Predicted metabolism of (A) E. coli Nissle (ECN) and (B) L. rhamnosus GG (LGG) by parsimonious Flux Balance Analysis (pFBA) under the constraints of relative consumption and production of metabolites inferred from the metabolomics dataset. The flux values shown are the average values of 10,000 simulations, normalized by the biomass production, in the unit of mmol / gram cell dry weight. The color of each reaction changes with the magnitude of the average flux as shown in the color bar. The entire dataset is available as Table S3.

Due to their ubiquity in the supernatant metabolome, amino acid, carbohydrate, energy, and nucleotide metabolite production were examined further with this predictive model. Arginine, proline, and urea cycle metabolism contributed to ~10% of the amino acid metabolites identified in the in-silico models for ECN and LGG. Within this metabolic pathway, arginine was identified as a growth-promoting substrate (i.e., a metabolite whose metabolism produces a net gain in ATP) in 82.7% of ECN models but as a growth-competing substrate (i.e., a metabolite whose metabolism results in a net loss of ATP) in 79.4% of LGG models.

Within carbohydrate metabolism, lactate was predicted to be a growth-promoting metabolite in 81.3% of ECN models. However, in LGG, lactate was predicted to be growth inhibiting in 68.8% of models. Among TCA cycle energy metabolites, succinate was present in 82.7% of ECN models as a growth-inhibiting substrate and not identified as a significant contributor to metabolism in LGG, which is consistent with a prediction that succinate would be elevated in the ECN supernatant metabolome compared to the LGG supernatant metabolome. When examining nucleotide metabolism, both guanine and uracil were predicted to be growth promoting products in 82.7% of ECN models, and in LGG, both guanine and uracil were predicted to be growth competing substrates in 82.7% of models. These predictions were suggestive of uracil and guanine being produced in abundance within ECN supernatant, as their production was associated with net ATP gain.

4. Discussion

The differential efficacy of cell free supernatants from two probiotic strains commonly used as dietary supplements was investigated in a dose-dependent manner for inhibitory effects on AMR pathogen growth. The Gram-positive probiotic LGG inhibited growth of S. Typhimurium, E. coli, and K. oxytoca with lower doses and exposure time when compared to the Gram-negative probiotic ECN. S. Typhimurium, E. coli, and K. oxytoca expressed resistance to multiple antimicrobial drug classes suggesting that use of probiotic supernatants can offer a safe, non-antibiotic targeted solution for inhibiting AMR pathogen growth in animals, people, and environmental systems. Profiling of the cell free supernatants by global, non-targeted metabolomics and proteomics revealed small molecules contributing to differential antimicrobial activities by ECN and LGG. Furthermore, in silico modeling distinguished metabolites produced by the probiotics from those dectected in growth media. This integrated model co-identified amino acid, carbohydrate, energy, and nucleotide metabolites that merit mechanistic exploration for impacting AMR pathogen growth.

Multiple amino acid metabolites distinguished the ECN and LGG metabolomes, which may account for the pathogen growth inhibition differences observed between ECN and LGG supernatants. Among these metabolite differences is arginine, which was 1.72-fold higher in LGG versus ECN supernatant (i.e. 0.58 fold decreased in ECN versus LGG as reported in Table S1). As reflected in the metabolic flux analysis, LGG supernatant may contain increased levels of arginine relative to ECN because it is a growth-limiting substrate for ATP production in LGG. In fermented food products, arginine is characterized for beneficial cardiovascular and immune-modulatory effects [31], but it has not been widely explored as a contributing source of antimicrobials from probiotics.

In the proteome, an increasing number of investigations into cationic antimicrobial peptides are revealing that arginine-containing peptides have enhanced capacities to lyse pathogens when compared to peptides containing other amino acids [32–34]. Although arginine-associated proteins were not widely identified in the LGG proteome, the multiple protein motifs identified in the ECN metabolome warrant further resolution and structural elucidation for their roles in arginine metabolism. The roles for ECN protein motifs in arginine metabolism or in the catalytic conversion into arginine-containing antimicrobial peptides can be examined with the goal of optimizing their production to improve ECN antimicrobial activity. Likewise, ECN production of arginine-derived metabolites with known antimicrobial activity including fumarate, citrulline, ornithine[35, 36] all of which were increased in ECN versus LGG supernatant, can additionally be explored in an effort to improve ECN antimicrobial activity.

In addition, carbohydrate and energy metabolite production by probiotics is also widely manipulated within the fermented food industry to influence the organoleptic properties of food products [37]. Proteomic analysis identified the decreased production of fructose 1,6 diphosphate, glucose 6-phosphate, phosphoenolpyruvate and lactate by ECN compared to LGG (Figure 3), which may have contributed to the lower efficacy of this probiotic supernatant against pathogen growth. In addition to increasing the glycolytic capacity of LGG, GAPDH has been increasingly explored in both prokaryotic and eukaryotic species to produce antimicrobial peptides that inhibit growth of Gram-negative pathogens [38–40]. Probiotic secreted GAPDH has been implicated in host adhesion as well as immunomodulation [40–43], and in conjunction with these results merit targeted investigation for dual roles in LGG-mediated pathogen growth reduction.

The coordination of metabolism across chemical classes is an opportunity for co-exploration and co-optimization of antimicrobial compounds from food and gut associated probiotics. This has been accomplished for lactate, an organic acid produced during carbohydrate fermentation, which is well-characterized for its broad-spectrum bacteriostatic activity against Gram negative pathogens when applied to foods or cultures as a single agent or in combination with other compounds [13, 44–47]. However, in-silico approaches for optimizing probiotic lactate production have not yet been readily applied in food science, human or animal health. Measurement and evaluation of glycolytic intermediates through in silico metabolic modeling approaches support the concentrations that inform on net production of the antimicrobial lactate, and thus highlights the value of in silico modeling for exploring and optimizing probiotic antimicrobial activity.

Nucleotide metabolism by probiotics has not been widely characterized for antimicrobial activity, but has been linked to antibiotic failure [48]. Nucleotides did account for major differences in the ECN versus LGG supernatant metabolomes herein. in-silico metabolic flux analysis predicted substantial differences in guanine and uracil concentrations when comparing LGG and ECN, due to synthesis of these metabolites favoring ATP production in ECN but causing a net loss of ATP in LGG. These findings were supported by the metabolomics analysis, which identified a 19.57-fold increased guanine in ECN versus LGG supernatant and 73.22-fold increased uracil in ECN versus LGG supernatant. However, LGG supernatant contained significantly increased levels of guanine and uracil metabolites including guanosine (50.0-fold increased in LGG versus ECN) and uridine (100.0-fold increased in LGG versus ECN). Multiple studies have demonstrated the broad-spectrum growth inhibitory functions of synthetic and other exogenous guanine and uracil-derived metabolites against human and animal pathogens, when in excess, interfere with normal nucleotide metabolism and cause oxidative damage to cellular components [49]. The higher circulating guanine and uracil levels in ECN compared to LGG may reflect a diminished capacity or metabolic preference of ECN to produce antimicrobial derivates from these metabolites compared to LGG.

Metabolomic profiling of LGG and ECN supernatant detected 255 metabolites of unknown identity which contribute to the AMR pathogen growth reduction of these supernatants. The large number of unidentified metabolites produced by probiotics represents a challenge to complete optimization of probiotics for antimicrobial production, as the metabolic pathway relationships to named metabolites remain unknown. The current limitations of unidentified molecules from this omics analysis highlight a knowledge gap that can be addressed by further development of high-resolution chemical detection and structural elucidation of microbial compounds, including both metabolites and proteins.

5. Conclusion

The findings reported from this multi-omics analysis of probiotic supernatants have important implications for targeting dietary supplements to poses bacteriostatic properties that inhibit growth of AMR pathogens. This study further illustrates the utility of integrated omics approaches to expand functional metabolic profiles with novel antimicrobial drug discovery potential. The metabolomic and proteomic-guided metabolic flux analysis of two genetically distinct probiotics emphasized amino acid, energy metabolites, and nucleotide metabolism as potential targets for reducing AMR pathogen growth. Notably, these metabolite classes have been minimally explored for contributions to antimicrobial functions of probiotics. We conclude that probiotic species differences in metabolism are essential to understand for antimicrobial activities and specifically for differences in addressing problems of AMR pathogen spread. Nucleotides were illustrated for novel utility by in silico modeling for estimating the relative concentration of a metabolite with relevance. This study filled knowledge gaps on probiotic capacity for differential production of antimicrobial compounds and expands opportunities for probiotic based dietary supplements to address emerging and existing antimicrobial resistant bacterial diseases.

Supplementary Material

Supplementary Material

Figure S1_ Percent difference supernatant pathogen reduction

File S1_ R script

File S2_ Metabolic modeling_mathematical methods

Table S1_ Statistically significant metabolites

Table S2_ Probiotic supernatant proteomes

Table S3_ Metabolic Flux Analysis- Metabolite Shadow Prices

Abbreviations

AMR

Antimicrobial Resistance

CFS

Cell-Free Supernatant

CFU

Colony Forming Unit

CLSI

Clinical and Laboratory Standards Institute

ECN

Escherichia coli Nissle

ESI

Electrospray Ionization

GAPDH

Glyceraldehyde-3-Phosphate Dehydrogenase

HILIC

Hydrophilic Interaction Liquid Chromatography

LB

Luria Bertani

LC-MS/MS

Liquid Chromatography-Tandem Mass Spectrometry

LGG

Lacticaseibacillus rhamnosus GG

m/z

Mass to Charge Ratio

MALDI

Matrix-Assisted Laser Desorption/Ionization

MRS

deMan Rogasa Sharpe

NAD+

Nicotinamide Adenine Dinucleotide (oxidized)

OD600

Optical Density, 600-nanometer wavelength

PES

Pathway Enrichment Score

pFBA

Parsimonious Flux-Based Analysis

RI

Retention Index

TCA

Tricarboxylic Acid Cycle

TSB

Tryptic Soy Broth

UPLC-MS/MS

Ultra-High-Performance Liquid Chromatography-Tandem Mass Spectrometry

About the authors

Petronella R. Hove. Dr. Hove is a veterinarian with a PhD in Infectious Disease and Immunology and a master’s degree in public health. She is a researcher whose current research interests are in cancer therapy, antibiotic resistance investigation, the characterization of genes, proteins, and enzyme function as well as the evaluation of small molecule metabolites and lipids in both infectious and organic disease pathogenesis and immunology.

Nora Jean Nealon: Dr. Nealon is a Postdoctoral Researcher at The Ohio State University’s College of Veterinary Medicine. Nora Jean completed her PhD in cellular and molecular biology and her DVM (Doctor of Veterinary Medicine) at Colorado State University. Her research interests include the mechanistic investigation and clinical treatment of companion animal chronic gastrointestinal diseases, especially in the space of developing sustainable and nutritious diets and supplements that will promote disease prevention and improve life quality for our animals.

Siu Hung Joshua Chan: Dr. Joshua Chan is an Assistant Professor in the Department of Chemical and Biological Engineering at Colorado State University (CSU). His research focuses on modeling and engineering microbiomes, which have a huge impact on the earth ecosystem and the life form therein, from the global geochemical cycle, soil, and plant health to human immune systems.

Shea M. Boyer: Shea Boyer is a Professional Research Assistant at the University of Colorado Anschutz Medical Campus. Shea holds a Bachelor of Science degree in Health and Exercise Science, Sport medicine and Microbiology from Colorado State University and is currently working towards his master’s degree in biostatistics from the University of Louisville.

Hannah B. Haberecht: Hannah B. Haberecht is a current medical student (M.D. candidate) at the Mayo Clinic. Hannah earned her Bachelor of Science degree in Biomedical Science from Colorado State University. Hannah’s research interests include nutritional chemistry and its applications towards human health.

Elizabeth P. Ryan: Dr. Ryan is an Associate Professor in Environmental and Radiological Health Sciences at Colorado State University, the Colorado School of Public Health, and the University of Colorado Cancer Center. Her research explores the complex interactions of food components with gut microbiota and the immune system that are important to mitigate gastrointestinal diseases across the lifespan. Her team also seeks to develop innovative solutions for enhancing food security that improve gut health.

Data availability

The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materials. The complete raw data and R script for metabolomics analysis, Metaboanalyst visualization, and raw data are available from the corresponding author [EPR] on request. The files for the metabolic modeling analysis are available at https://github.com/chan-csu/modelEcnLggExoMetabolomes.