Biomedical applications of genome-scale metabolic network reconstructions of human pathogens

Abstract

The growing global threat of antibiotic resistant human pathogens has coincided with improved methods for developing and using genome-scale metabolic network reconstructions. Consequently, there has been an increase in the number of high-quality reconstructions of relevant human and zoonotic pathogens. Novel biomedical applications of pathogen reconstructions focus on three key aspects of pathogen behavior: the evolution of antibiotic resistance, virulence factor production, and host-pathogen interactions. New methods using these reconstructions aim to improve understanding of microbe pathogenicity and guide the development of new therapeutic strategies. This review summarizes the latest ways that genome-scale metabolic network reconstructions have been used to study human pathogens and suggests future applications with the potential to mitigate infectious disease.

Graphical abstract

Introduction

Despite advancements in medicine, hygiene, and infrastructure, microbes that can survive and cause disease within a human host remain a global health concern [1]. Successful chronic infection of a human host is dependent on multiple pathogen behaviors including the development of antibiotic resistance [2,3], the production of virulence factors [4], and the exploitation of host and microbiota resources [5,6] (Figure 1). Aspects of these behaviors have been linked to pathogen metabolism [7,8], opening up exciting opportunities for the development of novel therapeutic strategies with emerging technologies [9,10]. However, identifying potential therapies, given the wide range of possible genetic mutations, host environments, and metabolic targets, remains a significant challenge [11,12]. One approach to overcome this problem has been the use of systems-level computational models, specifically genome-scale metabolic network reconstructions (GENREs).

Figure 1. Key behaviors of human pathogens during infection.

Over the course of infection, pathogens can develop resistance to antibiotic treatments (A), express different virulence factors (triangles) in response to their environment (B), and interact with host tissues (C) and resident microbiota (D). Metabolic network models of human pathogens can offer insight into each of these processes.

A GENRE is a mathematical framework of all known metabolic reactions inside an organism [13] (Box 1). The reactions and associated metabolites are formatted into what is known as a stoichiometric matrix. Functional relationships between genes and reactions can also be characterized with the model as gene-protein-reaction (GPR) rules allowing for the systematic simulation of gene knockouts [14,15]. GENREs are often evaluated with an approach called flux balance analysis (FBA) in which the network is optimized to maximize or minimize flux through an objective function such as biomass production [16]. Overall, GENREs are useful tools for probing various aspects of microbial metabolism [17,18].

Box 1. Defining flux balance analysis (FBA).

Flux balance analysis (FBA) is a linear programming approach used to analyze genome-scale metabolic network reconstructions (GENREs). Specifically, FBA predicts metabolic fluxes through the network under a given set of physiologically relevant constraints [13,16,19].

Mathematically, a network of mass-balanced metabolic reactions can be represented by the equation:

$$\frac{dc}{dt}=Sv$$ (1)

where C is a vector of metabolite concentrations, t is time, v is a vector of unknown reaction fluxes, and S is the stoichiometric matrix. The stoichiometric matrix is an m × n matrix where each row represents a metabolite and each column is a reaction. Values in S correspond to the mass-balanced stoichiometric coefficients of each metabolite for a given reaction.

A key assumption of FBA is that metabolite concentrations do not change over short time intervals of interest [13]. This steady-state assumption constrains the system and allows Equation 1 to be simplified to:

$$Sv=0$$ (2)

Because this system is underdetermined, FBA requires the definition of a metabolic goal, or objective function. An objective function is a reaction in S for which flux is minimized or maximized. Often the objective function is set to maximize flux through a biomass reaction, v_biomass

The final constrained optimization problem can be represented as:

$$\mathit{Max}{v}_{\mathit{biomas}s}$$

subject to,

$$Sv=0$$

$$v_{\mathit{min}}\le v\le v_{\mathit{Max}}$$

where flux through the biomass reaction, v_biomass, is maximized subject to the constraints that the system is at steady state (Equation 2), and flux values for each reaction are within physiologically relevant lower and upper bounds (v_min and v_max) [19].

How can v_biomass be related back to predicting pathogen behavior? Given that the objective is to maximize biomass and that the network is solvable, a solution of v_biomass = 0 indicates that the pathogen cannot grow in the given in silico conditions while v_biomass >0 indicates growth. Consequently, any gene deletion or set of constraints that results in v_biomass = 0 could be a potential therapeutic target. Alternatively, objective functions can be set to maximize or minimize outputs other than growth (e.g., virulence factor production) to explore metabolic capabilities [20,21]. Pathogen phenotypes can thus be predicted across environmental conditions or following genetic perturbations [17,18].

In recent years, GENREs have become increasingly prevalent in the study of human pathogens. GENREs are ideal for predicting cellular phenotypes [17] (e.g., growth and virulence factor production) across a wide range of environments (e.g., media or host conditions). Of particular value, such predictions can be informed by available systems-level omics data, which can be integrated into a GENRE to generate condition-specific models [18]. The increase in available annotated genomes [22] and improved methods for drafting network reconstructions [23,24] have made it easier than ever to model pathogens of interest. Recent genome-scale metabolic network reconstructions of pathogenic bacteria [20,25–29], fungi [30], and parasites [31–34] have elucidated genes and pathways essential to the survival and growth of human and zoonotic pathogens. Here, we summarize the current applications of genome-scale metabolic network reconstructions toward studying three key behaviors of human pathogens: evolution of antibiotic resistance, production of virulence factors, and host-pathogen interactions. Looking forward, we speculate how such approaches can be applied to mitigate infectious disease.

Human pathogen models and antibiotic resistance

Pathogen metabolism plays an important but poorly understood role in the development of antibiotic resistance [7,35,36]. Several approaches integrating systems-level experimental data with genome-scale metabolic network reconstructions have recently been applied to elucidate metabolic dependencies of resistance as well as potential drug targets (Table 1).

Table 1.

Novel applications of metabolic network reconstructions of human pathogens

Study	Organism(s)	Organism type(s)	Model name(s)	Description
Antibiotic/Antimalarial Resistance
Zampieri et al., 2017	Escherichia coli K-12 MG1655	Model organism	iJO1366	Development of a novel constraint-based approach to identify metabolic dependencies of antibiotic resistance
Banerjee et al., 2017	Chromobacterium violaceum	Zoonotic pathogen	iDB149	Identification of strategies to restore antibiotic susceptibility via a novel ALE/GENRE pipeline
Carey et al., 2017	Plasmodium falciparum	Human parasite	iPfal17	Integration of clinical transcriptome data to identify resistance-specific therapeutic targets
Tewari et al., 2017	Plasmodium falciparum	Human parasite	Modified from Fang et al., 2014 [37]	Analysis of chloroquine mechanism of action and potential combination therapies to overcome resistance
Virulence
Kim et al., 2013	Salmonella typhimurium	Human pathogen	STM_v1.0	Evaluation of metabolic changes associated with virulence-inducing environmental conditions
Bartell, Blazier et al., 2017	Pseudmonas aeruginosa strains PAO1 and PA14	Human pathogen	iPae1146 iPau1129	Assessment of predicted maximum virulence factor synthesis and relationship to biomass in Pseudomonas aeruginosa
Bartell, Yen et al., 2014	Burkholderia cenocepacia; Burkholderia multivorans	Human pathogens	iPY1537 iJB1411	Comparative analysis of Burkholderia virulence factor production capacities across many environments
Bosi et al., 2016	Staphylococcus aureus	Human pathogen	64 models of S. aureus strains	Classification of 64 strains of S. aureus by virulence genes and predicted growth capabilities
Ferrarini et al., 2016	Mycoplasma flocculare; Mycoplasma hyopneumoniae; Mycoplasma hyorhinis	Swine mycoplasmas	Many	Comparative metabolic analysis of swine mycoplasmas with varying virulence
Henry et al., 2017	Klebsiella pneumoniae strains KPPR1 and MGH 78578	Human pathogens	iKp1289 iYL1228	Comparative metabolic analysis of high-virulence and low-virulence strains of K. pneumoniae
Song et al., 2013	Toxoplasma gondii Types I, II and III	Human parasites	iCS382	Association of differential enzyme expression with parasite growth and virulence
Host-Pathogen Interactions
Raghunathan et al., 2009	Salmonella typhimurium	Human pathogen	iRR1083	Premiere analysis of a pathogen in the host environment
Ding et al., 2016	Escherichia coli Pangenome	Model organism/Human pathogen	iEco1712_pan	Essentiality analysis of the E. coli pan-genome in the human bloodstream, urinary tract, and macrophage
Bordbar et al., 2010	Myobacterium tuberculosis in the human alveolar macrophage	Human pathogen	iAB-AMØ- 1410-Mt-661	Premiere integration of pathogen and host metabolic network reconstructions
Huthmacher et al., 2010	Plasmodium falciparum in the human erythrocyte	Human parasite	PlasmaNet v2.0 EryNet	Integration of the malaria parasite with the human erythrocyte
Bazzani et al., 2012	Plasmodium falciparum; human hepatocyte	Human parasite	PlasmaNet HepatoNet1	Comparative analysis of metabolic drug target selectivity in the parasite and host
Gut Commensals and Pathogens
Magnúsdóttir et al., 2017	Many	Gut microbiota	Many	Development of draft models of over 700 microbes compatible with Recon2
Larocque et al., 2014	Clostridium difficile	Human pathogen	iMLTC806cdf	Reconstruction of the gut pathogen C. difficile
Veith et al., 2015	Enterococcus faecalis V583	Human pathogen	–	Reconstruction of the gut pathogen E. faecalis
Groβeholz et al., 2016	Enterococcus faecalis	Human pathogen	Adapted from Veith et al., 2015	Integration of whole-cell proteomic data with existing model of E. faecalis

Understanding metabolic dependencies on the evolution of antibiotic resistance

One such approach has been the pairing of genome-scale metabolic models with omics data collected from adaptive laboratory evolution (ALE) experiments (Figure 2A). Adaptive laboratory evolution is an experimental method used to study molecular changes and genetic mutations that arise as a population of bacteria adapts to a selective environmental pressure (e.g., nutrient or oxygen limitation, antibiotic pressure) [38,39]. Data from ALE experiments in the absence of antibiotics have been compared to genome-scale models to evaluate the relationship between phenotypic growth, omics data, and model-predicted optimal growth [40]. ALE experiments have also been used to examine multiple aspects of resistance including the role of antibiotic tolerance in facilitating resistance, the impact of selection regime (e.g., dose) on resistance mechanisms and mutations, and the influence of past treatment history on collateral sensitivity [35,41,42].

Figure 2. Metabolic network modeling approaches for studying human pathogens.

Simple examples illustrate how metabolic network reconstructions can be applied to study antibiotic resistance (A-B), virulence factor (VF) production (C-D), and host-pathogen interactions (E-F). (A) Adaptive laboratory evolution (ALE) can be paired with metabolic network reconstructions to predict metabolic dependencies on the development of antibiotic resistance. ALE to an antibiotic allows for the controlled development of resistance in vitro. Resistance in this example is quantified by an increase in the minimal inhibitory concentration (MIC) of an antibiotic over time. Metabolomics data collected throughout an ALE can be used to identify metabolites and associated reactions altered in resistant pathogens. In parallel, a curated metabolic network reconstruction can be used to predict the contribution of each reaction to pathogen growth. The experimental and network analyses enable predictions of reactions that are important for the development of resistance. (B) Resistance-specific essential genes and reactions represent targets with the potential to kill resistant pathogens without selecting for resistance in sensitive pathogens. Gene expression profiles from antibiotic-resistant and sensitive pathogens can be integrated into a manually-curated reconstruction to generate context-specific networks. In silico gene and reaction essentiality predictions can then be made and compared across resistant and sensitive networks (R = resistant; S = sensitive). (C) A single pathogen may produce different VFs depending on its environment. Gene expression profiles from pathogens grown in rich media (RM; yellow box) or virulence-inducing media (VIM; green box) can be integrated with network reconstructions to generate network models specific to a particular environmental niche. Analyses of these models can be performed to study the impact of environment on virulence (triangles = VFs). (D) Closely related strains of a pathogenic species can exhibit differential VF production, which can result in varying levels of pathogenicity in the host. Annotated genomes of closely related pathogens can be used to build strain-specific network reconstructions. The reconstructions can then be used to calculate the maximum predicted VF production capacity of each strain. (E) Pathogen behavior can be impacted by the host. To study this relationship, metabolomics data collected from host niches can be been used to construct a pathogen network reconstruction within the host environment. In silico gene and reaction essentiality predictions can then be made and validated against experimental data (m = metabolite; R = reaction). (F) Just as pathogen behavior can be impacted by the host, host behavior can be altered by the presence of a pathogen. To study the host-pathogen relationship, models of the pathogen and the host can be combined to interrogate the interactions between their metabolic networks (inspired by Figure 4A in Bordbar et al., 2010).

Two groups have recently developed methods to incorporate adaptive evolutions of resistance with genome-scale metabolic models. Zampieri et al. evolved Escherichia coli to three different antibiotics on two different carbon sources to evaluate the impact of metabolism on the development of antibiotic resistance [43]. Metabolomics data were collected throughout the evolution experiments. In parallel to ALE experiments, the authors systematically constrained reactions in a metabolic network reconstruction of E. coli to calculate shadow prices for every metabolite in the model, where ‘shadow prices’ refer to the sensitivity of biomass production to a change in a given metabolite availability. Metabolites with negative shadow prices are predicted to be limiting for biomass production, and thus reactions with a large number of limiting metabolites are considered to be important for pathogen growth. The authors predicted that a reaction was important for the development of resistance if a large number of the metabolites in it had negative shadow prices and metabolite levels were altered in ALE metabolomics data.

In a second study, Banerjee et al. developed a pipeline to identify metabolites with the potential to restore susceptibility in antibiotic resistant populations of the zoonotic pathogen Chromobacterium violaceum [44]. Phenotypic and metabolomics data from ALE experiments of C. violaceum to two separate antibiotics were incorporated into a genome-scale metabolic network reconstruction to generate wild-type and resistant models. Scaled shadow prices were calculated following flux balance analysis and antibiotic treatment was predicted to result in redox imbalance. The alteration of intracellular NAD/NADH ratios was proposed as a potential intervention to restore antibiotic susceptibility in resistant populations.

Interestingly, although in different organisms and media conditions, models from both studies predicted decreased rates of oxygen uptake in response to adaptation to the protein synthesis inhibitor, chloramphenicol. Experimentally, both groups saw mutations in acrR, a transcriptional repressor of the proton-dependent multi-drug efflux pump AcrA. Similarities and differences of in silico predictions across studies suggests that there may exist both pathogen-independent and –dependent metabolic responses to antibiotic adaptation in a particular environment. The accuracy of such model predictions likely depends on both reconstruction quality and method of data incorporation. While promising, ALE experiments are time-consuming, taking days to weeks per adaptation to a given drug in a given environmental condition; consequently, there are limits to the number of combinations of pathogens, drugs, and environments that can be feasibly tested experimentally. Using genome-scale models earlier on in the experimental design to narrow down the choice of drugs and environments as well as automating ALE experiments with continuous-culture devices such as morbidostats can expand the scope of this approach [39,45].

Identification of resistance-specific therapeutic targets

Resistance-specific metabolic profiles and potential drug targets have been identified through the integration of clinical omics data with curated metabolic models (Figure 2B). Systems-level in vitro and in vivo studies measuring gene expression of antibiotic resistant and sensitive pathogens are gaining popularity as experimental methods for omics collection becomes easier and less expensive [46,47]. Complementing this increase in available omics data are improved algorithms for integration of transcriptomic data into existing genome-scale metabolic models [48]. In a recent study, clinical microarray data from over 300 antimalarial-resistant and sensitive malaria isolates were integrated into a genome-scale metabolic model of Plasmodium falciparum to identify robust metabolic differences associated with artemisinin resistance [31]. By comparing essential genes and reactions across condition-specific models, 21 reactions were predicted to be uniquely required for growth of artemisinin-resistant parasites. Resistance-specific essential pathways are promising potential drug targets as they theoretically kill resistant pathogens without selecting for resistance in sensitive pathogens. Of note, another model of Plasmodium falciparum was also recently developed and used to identify the mechanism of action of the antimalarial chloroquine. Guided by model predictions, potential combination therapies were proposed to overcome chloroquine resistance [33]. While the discussed examples focus on antimalarial resistance, this method of identifying “resistance-specific” drug targets via transcriptomic data integration followed by in silico gene and reaction essentially screens can be applied to a wide variety of bacterial pathogens and antibiotics.

Metabolic models and pathogen virulence

Virulence factors are small molecules that promote pathogen survival and replication within the host. Virulence factors with production mechanisms independent of pathogen growth have the potential to be effective therapeutic targets with limited selective pressure for resistance [49]. Genome-scale metabolic models of bacterial and fungal human pathogens have been used to identify novel virulence genes as well as to evaluate the interconnectivity between virulence factor synthesis, pathogen metabolism, and growth [20,30,50] (Table 1).

Assessing environment-specific pathogen virulence

Experimental screens for virulence genes have been shown to yield conflicting results across studies, making it difficult to identify candidate target genes [30]. Such discrepancies could be due to choice of media condition used for the screen, as virulence factor production has been shown to vary across environmental conditions [51,52]. Metabolic network reconstructions have been shown to be useful for teasing out the impact of environment on pathogen virulence (Figure 2C). In a study of the infectious human pathogen Salmonella typhimurium, metabolites that were upregulated in virulence conditions were identified by integrating a model of S. typhimurium with transcriptomic and metabolomics data from pathogen growth in either rich media or virulence-inducing media [53]. In the absence of condition-specific metabolomics data, genome-scale models can be used to predict growth and virulence factor production across a wide range of environments, from rich media only used in vitro to conditions mimicking those measured from patient samples in vivo. Such approaches allow for more guided identification of drug target candidates and offer insight into why the same microbial pathogen may cause disease in some host environments but not others.

Predicting strain-specific pathogen virulence

Metabolic network reconstructions can also be used to better understand why closely related pathogens may exhibit different levels of infectivity within the same host. Several approaches have been applied to elucidate differences in virulence and growth across pathogenic strains and species (Figure 2D). Provided that annotated genomes of the strains of interest exist, draft metabolic networks can be reconstructed starting from an existing model of the species of interest or using a method of automated reconstruction [54,55]. Following manual curation of draft models, virulence can be compared across strains or species by creating demand reactions in each model for virulence-related metabolites and performing FBA to maximize production of each virulence factor. More virulent pathogens have been predicted to have higher virulence factor production capacities across many environments using this method [21]. While effective, this approach requires that metabolic network reconstructions include reactions for virulence factor production, which may not always be possible, especially if pathogens lack well annotated genomes.

Strain-specific models paired with known virulence information can be used to assess relationships between metabolism and pathogen virulence. In a study of Staphylococcus aureus, metabolic models were constructed for 64 unique strains and the metabolic capabilities for each strain were predicted [56]. The presence and absence of virulence factors for each strain were compared to predicted metabolic capabilities to better understand the impact of metabolism on S. aureus pathogenicity. In another example, a study of three swine respiratory mycoplasmas used metabolic models to show that the ability to produce cytotoxic hydrogen peroxide was associated with pathogen virulence [25]. Still others have used models to broadly compare metabolism of closely related pathogens to better understand differences in virulence [26]. In situations where strains are genetically identical aside from scarce sequence variants, as is the case with Toxoplasma gondii Type I, II and III, metabolic drivers of strain-specific virulence can be predicted by integrating mRNA expression data from each strain into a highly curated model of the organism’s core metabolism and evaluating differences in metabolic capabilities [57]. Taken together, metabolic models can be used to tease out the genetic, regulatory, and environmental contributions to pathogen virulence, helping to narrow in on novel therapeutic targets.

Metabolic models of host-pathogen interactions

Host-pathogen interactions play an important role in virulence and pathogenicity and therefore should be considered when using metabolic network models [58,59]. Various computational methods – including metabolic network reconstructions - for studying host-pathogen interactions have been previously reviewed [60–62], and new approaches are consistently being developed as more complicated interactions are studied (Table 1). The degree to which host-pathogen interactions need to be considered when generating model predictions depends on the research question being addressed. When studying the impact of host environment on a pathogen, it is often sufficient to incorporate host-pathogen interactions by using host conditions as inputs into a metabolic model of the pathogen of interest (Figure 2E). This approach has been used across a range of pathogens and host tissues to predict pathogen growth, nutrient utilization, and essential genes during infection [34,63,64]. For example, to better understand infectivity of E. coli across human tissues, Ding et al. used a metabolic network reconstruction to simulate E. coli growth in the human bloodstream, urinary tract, and macrophage and discovered pathogen genes uniquely essential for growth in each environmental niche [65]. The authors then highlighted the value of their approach by validating essential gene predictions against the genomes of E. coli and Salmonella strains known to infect each tissue (Figure 2E). This method is particularly useful for elucidating the role of host environment on pathogen metabolism during the course of an infection.

Of course, host-pathogen interactions are not unidirectional, and thus alternative approaches have been developed to consider both the impact of host environment on the pathogen and the impact of pathogen infection on the host within a single computational model (Figure 2F). Host and pathogen models can be directly integrated or evaluated in parallel to (1) study host and pathogen metabolism concurrently across stages of infection, or (2) predict potential drug targets for human pathogens that have little to no impact on host metabolism. Examples include the integration of a human macrophage model with tuberculosis [66] (Figure 2F), and computational analyses of Plasmodium falciparum during infection in the erythrocyte [67] and hepatocyte [68]. Although more physiologically relevant and accurate, integrated host-pathogen models require more complicated manual curation and can be difficult to experimentally validate. As human metabolic network reconstructions, such as Recon2 [69], continue to be curated and experimental methods to study host-pathogen interactions improve, the integration of host-pathogen models will likely become less time intensive and will generate more reliable predictions.

Future directions

New directions in multi-species models: microbiota-pathogen interactions

Successfully combating pathogenic bacteria such as Clostridium difficile in the human gut will require a better understanding of interactions between the pathogen, host, and resident microbiota. Methods for studying host-microbiota interactions have previously been reviewed [70]; however, no one to our knowledge has applied metabolic network models to study microbiota-pathogen interactions. In an ideal system, a gut pathogen reconstruction would be integrated into a community model of hundreds of commensal organisms in the host environment. Such a model could theoretically be used to study how various environmental or genetic alterations impact pathogen growth and virulence within the context of the host microbiome. Many of the pieces of this ideal system are coming together in published literature. With regards to the microbiome, metabolic models have been used to study single gut species and interactions between two or three commensals [71–73]. A newly developed resource, AGORA (assembly of gut organisms through reconstruction and analysis), provides draft reconstructions for over 700 members of the human gut microbiota, all compatible with Recon2, creating a platform for larger community models [74]. In parallel, others have developed highly curated models of gut pathogens including C. difficile [27] and Enterococcus faecalis [29,75] (Table 1). Simplified methods for multi-model amalgamation and further curation of commensal models are needed to bring existing work together to build a predictive genome-scale metabolic model of the gut microbiota that can be perturbed by intestinal pathogens.

Applications of condition-specific pathogen models

As the collection of omics data becomes less expensive and integration of multiple types of omics data into genome-scale metabolic network reconstructions becomes more routine, we will likely see an increase in condition-specific metabolic models of human pathogens. Improved methods of incorporating transcriptomic data [48,76,77], metabolomics data [78], or both [79] into metabolic models are rapidly being created and updated. The implications of improved methods for omics integration are huge, as many pathogen behaviors (such as antibiotic tolerance and persistence) are driven by downstream regulatory mechanisms and are not immediately obvious from the genome. Such advancements will also pave the way for personalized host-pathogen or even host-microbiota-pathogen interaction models in the future.

Toward mitigating the spread of infectious disease

An ultimate goal of the field is to aid in the development of novel strategies to mitigate infectious disease and, in doing so, better human health around the world. Although to our knowledge no strategies identified from genome-scale metabolic network reconstructions of human pathogens have made it into the clinic, many novel therapies have been suggested. In the realm of antibiotic resistance, novel supplementation strategies to restore antibiotic sensitivity [44], combination therapies to prolong the efficacy of existing compounds [33], and resistance-specific metabolic targets [31] have all been uncovered using metabolic network reconstructions. Regarding virulence and pathogenicity, models have been used to predict metabolic pathways associated with virulence and virulence factor production that could act as targets for pathogen attenuation [25]. While such predictions are promising, it is often unclear how or if model findings will be followed up on and whether or not they will be clinically relevant. Looking forward, translation of proposed therapeutic strategies to the clinic will require increased effort experimentally validating predictions and moving promising leads through the pipeline.

Conclusion

The above approaches cumulatively offer a fresh outlook on addressing many of the longstanding challenges of infectious disease. With the development of novel methods as well as the clever implementation of well-established techniques, genome-scale metabolic network reconstructions are rapidly being applied to elucidate the importance of metabolism on key pathogen behaviors such as antibiotic resistance, virulence factor production, and interactions with the host. Further advancements in multi-model amalgamation and omics data integration will facilitate research on more complicated pathogen behaviors such as microbiota-pathogen interactions and antibiotic tolerance. As the field continues to expand and methods improve, systems-level metabolic network modeling will have a transformative impact on the way the pharmaceutical and healthcare industries prevent and treat infectious disease.

Highlights.

• -Genome-scale metabolic network reconstructions capture human pathogen phenotypes.
• -Reconstructions can explore relationships between drug resistance and metabolism.
• -Environment- and strain-specific virulence can be predicted using reconstructions.
• -Multiple approaches to model host-pathogen interactions have been developed.