Parallel exploitation of diverse host nutrients enhances Salmonella virulence

Abstract

Pathogen access to host nutrients in infected tissues is fundamental for pathogen growth and virulence, disease progression, and infection control. However, our understanding of this crucial process is still rather limited because of experimental and conceptual challenges. Here, we used proteomics, microbial genetics, competitive infections, and computational approaches to obtain a comprehensive overview of Salmonella nutrition and growth in a mouse typhoid fever model. The data revealed that Salmonella accessed an unexpectedly diverse set of at least 31 different host nutrients in infected tissues but the individual nutrients were available in only scarce amounts. Salmonella adapted to this situation by expressing versatile catabolic pathways to simultaneously exploit multiple host nutrients. A genome-scale computational model of Salmonella in vivo metabolism based on these data was fully consistent with independent large-scale experimental data on Salmonella enzyme quantities, and correctly predicted 92% of 738 reported experimental mutant virulence phenotypes, suggesting that our analysis provided a comprehensive overview of host nutrient supply, Salmonella metabolism, and Salmonella growth during infection. Comparison of metabolic networks of other pathogens suggested that complex host/pathogen nutritional interfaces are a common feature underlying many infectious diseases.

Figure 1. Nutrient utilization capabilities of Salmonella in infected mouse tissues.

Colored names represent transporters and enzymes that were detected in Salmonella purified from mouse spleen (Table S1). The color shows enzyme abundance in copies per Salmonella cell. Grey proteins were not detected. Arrows represent metabolic reactions. Transport reactions are labeled with cylinders. Arrow colors show maximal catalytic capacities calculated from enzyme abundance and reported turnover numbers (Table S2). Grey arrows represent reactions, for which enzymes were not detected and/or turnover numbers were unavailable. Tsx is an outer membrane general nucleoside channel; NupC is a high affinity transporter for all nucleosides except guanosine and deoxyguanosine. An interactive map with detailed description of all detected metabolic capabilities is available at http://www.biozentrum.unibas.ch/personal/bumann/steeb_et_al/index.html.

Figure 2. Mouse spleen colonization of Salmonella mutants with metabolic defects.

The data represent competitive indices (CI) of mutants vs. wildtype Salmonella in spleen of individual mice at three (open symbols) or four days (filled symbols) post infection (Table S3). A log₂(CI) value of 0 (equivalent to a CI of 1) represents full virulence. Down triangles represent mutants with utilization defects, up triangles represent auxotrophic mutants. Grey symbols represent data from a previous study obtained in the same disease model. Red triangles represent data from an independently reconstructed glpFK gldA glpT ugpB mutant. The data provided evidence for access to a number of host nutrient which are shown in black (for detailed interpretation see Table S5). Nutrients with apparently low availability are shown in grey. Statistical analysis was carried out with the Benjamini-Hochberg false discovery rate (FDR) approach for multiple comparisons (***, FDR<0.001; **, FDR<0.01; *, FDR<0.05).

Figure 3. Nutrient limitation of intracellular Salmonella growth.

A) Schematic representation of external supplementation of intracellular Salmonella (red) in infected macrophages (grey). B) Increasing external nutrient availability accelerates intracellular Salmonella growth, and this depends on specific Salmonella nutrient utilization capabilities (open symbols, 0.5 g l⁻¹ glucose; filled black symbols, 1 g l⁻¹ glucose; filled grey symbols, 0.5 g l⁻¹ glucose 0.5 g l⁻¹ mannitol; circles, wildtype Salmonella; upward triangles, Salmonella ptsG manX galP mglB, deficient for high-affinity glucose transport; downward triangles, Salmonella mtlAD, deficient for high-affinity mannitol transport and degradation). Colony-forming units (CFU) at 10 h post infection for triplicate wells containing 300’000 RAW 264.7 cells are shown. C) Flux-balance analysis of nutrient excess scenarios. The computational model was set to incorporate various amounts of excess nutrients (beyond what was needed for cell maintenance and growth). Model parameters were adjusted to yield predictions that were consistent with experimental mutant and wildtype colonization data. Simulation of up to 18% nutrient excess was possible but required unrealistically high maintenance costs (shown in multiples of maintenance costs for axenic conditions). Simulated scenarios with nutrient excess beyond 18% were incompatible with experimental colonization data.

Figure 4. A quantitative genome-scale model of Salmonella nutrition, metabolism, and growth in infected mouse spleen.

This schematic map shows available host nutrients, their respective uptake rates represented by color and font size, and their conversion to new Salmonella biomass through the Salmonella metabolic network (see text and Tables S6, S7, S8, S9 for detailed explanation and quantitative values). Symbols represent metabolites (squares, carbohydrates; pointing up triangles, amino acids; vertical ellipses, purines; horizontal ellipses, pyrimidines; pointing down triangles, cofactors; tees, tRNAs; circles, other metabolites; filled symbols, phosphorylated metabolites) and proteins (diamonds). The connecting lines present metabolic reactions. The brown lines represent the inner and outer membranes. An interactive map with detailed annotation of all reactions and the computational model in SBML format are available at http://www.biozentrum.unibas.ch/personal/bumann/steeb_et_al/index.html. The model is also available in the supporting information (Model S1).

Figure 5. Large-scale experimental data are consistent with computational model predictions.

A) Validation of mutant phenotype predictions. The colors show the predicted gene relevance for spleen colonization (red, essential; orange, contributing; blue, non-detectable; see text for definitions). Comparison of model predictions with 738 experimental Salmonella mutant phenotypes revealed 92% prediction accuracy (inner dark colors) but also 61 discrepancies (pale outer colors). Numbers (correct/total number of experimentally validated predictions) are also given. B) Potential reasons for inaccurate phenotype predictions (redu, unrealistic redundancy; biom, incomplete biomass/maintenance issues; part, partially contributing functions; toxic, accumulation of toxic upstream metabolites; gap, missing enzyme; or exp, possibly inaccurate experimental data). For detailed descriptions see Table S10. C) Detection of enzymes with predicted differential relevance for optimal Salmonella in vivo growth. Enzyme relevance was classified by parsimonious enzyme usage flux-balance analysis (pFBA) (ess, essential enzymes; optima, enzymes predicted to be used for optimal in vivo growth; ELE, enzymatically less efficient enzymes that will increase flux if used; MLE, metabolically less efficient enzymes that will impair growth rate if used; zeroFlux, enzymes that cannot be not used in vivo). Filled bars represent enzymes that were detected by Salmonella ex vivo proteomics, open bars represent enzymes that were not detected. Statistical significance of the relationship between enzyme classes and the proportion of detected proteins was determined using the Chi square trend test. D) Feasibility of predicted reaction rates. For each reaction, the range of flux rates compatible with full Salmonella growth was determined using Flux-Variability Analysis. The circles represent the most economical state with minimal total flux (see text). Predicted reaction rates are compared to corresponding catalytic capacities calculated form experimental enzyme abundance and turnover numbers (Table S2). The reddish area represents infeasible fluxes. Reactions with substantial infeasible fluxes in the most economic simulated state are labeled (1, formyltetrahydrofolate dehydrogenase; 2, phosphoserine aminotransferase; 3, glycerol dehydrogenase). E) Predicted flux ranges and corresponding catalytic capacities after constraining all reactions to feasible fluxes (except for the three aminoacyl tRNA ligations mentioned in the text). F) Relative flux ranges of the initial unrestrained (straight line) and the enzyme capacity-restrained (dotted line) models. For each reaction, the flux range was divided by the respective flux value in the most economical state. Reactions that carried no flux in the most economical state were not considered. Statistical significance of the difference between both distributions was tested using the Mann-Whitney U test.

Figure 6. A common nutritional pattern for mammalian pathogens.

A) Presence of 254 nutrient utilization pathways in genomes of 153 mammalian pathogens (excluding all Salmonella serovars). Data were based on pathway annotations available in MetaCyc . Degradation pathways for nutrients that support Salmonella in mouse spleen were highly overrepresented among pathogen genomes (P<0.001; Mann-Whitney U test) suggesting similar nutritional preferences (filled circles; 1, purine nucleosides; 2, pyrimidine nucleosides; 3, fatty acids; 4, glycerol; 5, arginine; 6, N-Acetylglucosamine; 7, glucose; 8, gluconate). B) Depletion frequency of 118 biosynthesis pathways in mammalian pathogens. The values represent differences in pathway frequency in sets of 153 pathogens and 316 environmental bacteria (see text for explanation). Biosynthesis pathways for biomass components that Salmonella could obtain from the host were selectively depleted among pathogen genomes (P<0.0001; Mann-Whitney U test) suggesting similar host supplementation patterns (filled circles; 1, tyrosine; 2, histidine; 3, arginine; 4, cysteine; 5, methionine; 6, tryptophan; 7, threonine; 8, valine; 9 leucine; 10, isoleucine; 11, proline; 12, pyridoxal; 13, purine nucleosides; 14, pyrimidine nucleosides; 15, glutamine; 16, thiamin; 17, pantothenate).