Metabolic Model of the Phytophthora infestans-Tomato Interaction Reveals Metabolic Switches during Host Colonization

A metabolic model for the tomato-P. infestans pathosystem.To reconstruct an integrated tomato-P. infestans metabolic model, we exploited a recently constructed tomato GEM (35). This tomato model comprised 2,143 reactions and 3,410 genes, whereas the previously constructed P. infestans GEM comprised 2,394 reactions and 1,301 genes, suggesting that P. infestans has less genetic redundancy for metabolic enzymes than does tomato (Table 1). Our original P. infestans GEM was based on a framework reconstructed by the annotation of enzyme orthologs in the P. infestans genome and linking those to the biochemical reactions in KEGG (34, 36). Based on the literature, we added a minimal growth medium, a set of known biomass precursors, and cellular compartments, and simulated in vitro growth (34). The tomato GEM was originally used to simulate photorespiratory fluxes under different conditions and was reconstructed from LycoCyc (35, 37). We manually improved the P. infestans and the tomato GEMs (see Materials and Methods). For instance, according to recent insights into the mitochondrial localization of glycolytic enzymes (17), we curated the gene-reaction association of P. infestans in this pathway. In the tomato GEM, we curated the thiamine biosynthesis, since P. infestans is a thiamine auxotroph (8). The two GEMs were connected by so-called transport reactions, representing the nutrient flux from tomato to P. infestans. Although the transporter repertoire of P. infestans has been predicted (16), we could not attribute specific substrates to transporters. Most functional annotations lack specificity, and even between oomycete species, transporter substrate specificity can vary (38). Therefore, we chose an unbiased approach, connecting the two GEMs by the addition of a hypothetical unidirectional transport reaction for each of the 520 metabolites that were found to be shared between the cytosol compartments, each one representing the flux of a single metabolite from the tomato cytosol to the P. infestans cytosol (Fig. 1). We realized that for a pathogen to maintain homeostasis during infection, fluxes may occur in both directions (i.e., from the host to the pathogen and vice versa). However, the lack of knowledge of this process only allowed us to model host-to-pathogen nutrient fluxes. The 520 added transport reactions predominantly concern metabolites taking part in primary metabolic subsystems (from KEGG). These include amino acid biosynthesis (n = 75), carbon metabolism (n = 61), purine and pyrimidine biosynthesis (n = 51 and 37, respectively), and ABC transporters (n = 42), which indicates that at least 42 metabolites are known ABC transporter substrates. Of the 520 introduced transport reactions, 179 could carry a flux, meaning that the respective metabolites can both be produced by the tomato plant and assimilated by P. infestans. The integration of the two GEMs resulted in an integrated metabolic model of P. infestans-tomato interaction, comprising a total of 4,695 reactions, 4,303 metabolites, and 4,578 genes (Table 1).

• Open in new tab
• Download powerpoint

FIG 1

Integrated P. infestans-tomato model. (A) Schematic illustration. Dots are metabolites, arrows are reactions, and dotted lines represent the host-pathogen transport reactions. (B) Numbers of unique and shared cytosol metabolites.

Flux simulations pinpoint nutrients utilized by P. infestans.Given a metabolic model and a metabolic objective for the cell, optimal values for the fluxes through all reactions can be calculated to simulate the production of P. infestans biomass precursors through the import of nutrients from the tomato cytosol (39). An objective could be, for instance, to maximize the production of biomass or to produce biomass at minimal enzyme expense. To study the versatility of the model, we composed four scenarios based on different cellular objectives (Table S2). In scenario I, we calculated the minimal set of nutrients P. infestans needs to import from tomato to form all its biomass precursors. This was already possible by importing three compounds, thiamine, l-gamma-glutamyl phosphate, and cysteine. While this scenario is unlikely to occur in planta, it does illustrate that P. infestans has a comprehensive enzyme repertoire, enabling it to form its biomass precursors based on a very small pool of nutrients. In scenario II, we calculated P. infestans biomass production using the minimum number of P. infestans reactions possible, to simulate growth with minimal enzyme cost for P. infestans. In this scenario, 48 nutrients were used, including ammonium, a known growth substrate (40). However, ammonium is an unfavored nitrogen source compared to amino acids (16). Among other imported nutrients in this scenario are the tricarboxylic acid (TCA) cycle intermediates malic acid and succinic acid, for which P. infestans has an experimentally verified transporter (38). Organic acids, such as succinic and malic acids, strongly promote P. infestans growth in vitro, especially in combination with ammonium as a source of nitrogen (41). Notably, 29 of the nutrients in this scenario are direct biomass precursors for P. infestans, including 19 amino acids (aspartate being the sole exception) (Fig. S1). In scenario III, we maximized the usage of P. infestans reactions to simulate a scenario in which the pathogen maximally exploits its metabolism. This simulation predicted a pool of 29 nutrients, including the inorganic compounds nitrite and hydrogen sulfide. Interestingly, Phytophthora spp. retained the complete assimilation pathways for these compounds (16), in contrast to multiple obligate biotrophic oomycetes which lost multiple genes throughout evolution (8). In scenario IV, we combined scenarios I and II, in which we calculated the minimal nutrient uptake combined with minimal usage of P. infestans reactions to produce biomass. Since both nutrient import and assimilation into biomass require energy, these processes likely require a trade-off; hence, we anticipate that this is a more realistic scenario. Here, 38 nutrients were used, 16 of which were amino acids.

The four scenarios described above have opposite objectives (minimize/maximize the fluxes of P. infestans) and simulate extreme circumstances not likely found in nature. The true pool of imported nutrients will likely be a combination of the pools predicted in each individual scenario. A comparison of the nutrients imported in all four scenarios revealed sets of nutrients in common between scenarios. There are 14 nutrients found in three out of four scenarios, i.e., II, III, and IV (Fig. S1), including several amino acids, nucleotide precursors, and glycerol 3-phosphate as a lipid precursor. Surprisingly, aspartate is not imported in any of the scenarios, even though it is a biomass precursor of P. infestans and serves as a precursor to a variety of other amino acids (42, 43). An oomycete-specific form of aspartate aminotransferase (EC 2.6.1.1) was found that seems to play a key role in pathogenicity in Phytophthora sojae, possibly by balancing nitrogen and carbon metabolism and facilitating the interconversion between amino acids (44). The presence of this enzyme in our model explains why aspartic acid import may not be necessary. Interestingly, thiamine is only imported in scenario III (maximizing the P. infestans fluxes), while thiamine pyrophosphate (TPP) is imported in scenarios II and IV (minimizing fluxes). It is assumed that many oomycetes import thiamine to form TPP (8), an essential cofactor in carbohydrate metabolism (45) and in TPP-responsive riboswitches (46). However, related Stramenopiles were found to also grow on thiamine alternatives (47), and possibly, P. infestans can import up- or downstream compounds instead. Taken together, these simulations suggest a range of nutrients from tomato that can be effectively assimilated by P. infestans. Nutrients in common between the simulated scenarios suggest that these are likely more versatile than others and hence most useful for P. infestans to import.

Network analysis identifies dependencies of P. infestans on tomato metabolism.To identify dependencies of P. infestans on tomato, we investigated the topology of the integrated model by looking for essential reactions (i.e., reactions that are indispensable in the model to form all biomass precursors) and (inter)dependencies of reactions (i.e., the flux of a particular reaction relies on the flux of another), also known as coupled reactions. Coupled reactions within a metabolic model can be identified using a method called flux coupling analysis (FCA; File S1) (48). FCA of the model identified 77 coupled P. infestans-tomato reaction pairs (Fig. 2) involving 53 unique P. infestans reactions and 49 unique tomato reactions (Table S3). One of these P. infestans-tomato reaction couplings comprises the biomass reaction of TPP in P. infestans, which is coupled to 17 tomato reactions, illustrating that P. infestans TPP is dependent on the thiamine biosynthesis pathway in tomatoes. In addition, of the 4,695 reactions in the model, 112 P. infestans reactions and 35 tomato reactions were found to be essential for P. infestans (Fig. 2A). Some reactions are essential and coupled to many other reactions at the same time (appearing as hubs in Fig. 2A), implying that these play a central role in the model, with potentially large biological implications. For instance, there is a tomato transport reaction of aspartate into the plastid compartment, supplying aspartate as an amino group donor for the synthesis of a thiamine precursor (aminoimidazole ribotide) (49, 50), thus indirectly making this tomato reaction essential for P. infestans. Similarly, ATP and phosphate transport between P. infestans cytosol and mitochondria is essential and seems to play an important role in the model. Phosphate and sulfate uptake by the tomato plant are part of the defined growth medium, hence coupled to reactions in both species. FCA also revealed several clusters of tightly interconnected coupled essential reactions (Fig. 2A). One of these represents P. infestans fatty acid biosynthesis. In the model, this comprises a mostly linear pathway, and consequently, most reactions are coupled (interdependent). Fatty acid biosynthesis in oomycetes is associated with fungicide resistance (51) and energy storage for sporangia (8).

• Open in new tab
• Download powerpoint

FIG 2

Flux coupling between reactions in the P. infestans-tomato model. (A) Graph showing the coupled reactions in the model. Nodes represent reactions in tomato (red) or P. infestans (blue) and host-pathogen transport (green), and edges represent coupling between those reactions. Node size reflects essentiality for P. infestans biomass production. Stars represent transport reactions listed in panel B. Highly connected nodes (1 to 5) and clusters (I to V) are indicated and listed in the boxes on the right (see also main text). (B) Nutrients associated with the 12 most frequently coupled host-pathogen transport reactions. The bars are stacked and indicate the number of coupled reactions per species.

Couplings to the host-pathogen transport reactions in particular can provide information about the importance of transported metabolites; the more P. infestans reactions that are coupled to a transport reaction, the more likely that the associated substrate is important for P. infestans. We selected the most frequently coupled transport reactions from the model and assessed their couplings to other tomato and P. infestans reactions (Fig. 2B). The substrates of these transport reactions were associated with a diverse range of metabolic processes in P. infestans, i.e., pantothenate/coenzyme A (CoA) biosynthesis (2-dehydropantoate), de novo pyrimidine biosynthesis (S-dihydroorotate) (52), riboflavin metabolism [2,5-diamino-4-hydroxy-6-(5-phosphoribosylamino)pyrimidine], and inositol phosphate metabolism (myo-inositol 3-phosphate). Notably, three transporter substrates were associated with glycerophospholipid metabolism (phosphodimethylethanolamine N−methylethanolamine phosphate and choline). It is conceivable that P. infestans can take up glycerophospholipid precursors to facilitate the formation of membrane lipids (53). Seven P. infestans reactions were coupled to the reaction importing fructose, an efficient growth substrate (40). Among the most frequently coupled host-pathogen transport reactions are also those transporting dTTP, histidine, and leucine. Since these are mainly coupled to tomato reactions, the import of these nutrients seems to depend on their biosynthesis pathways in tomato. FCA can identify coupled reactions in metabolic models that are not necessarily directly connected in the network and can thus derive functionally related modules of (pathogen-host) metabolism that may otherwise be overlooked.

Transcriptome-based submodels provide insight into the dynamics of P. infestans-tomato metabolism.To obtain insight into the metabolism at subsequent developmental stages of the P. infestans-tomato interaction, we integrated the model with dual-transcriptome data obtained from RNA, isolated from tomato leaves inoculated with P. infestans strain 1306 (2 to 6 days postinoculation, sampled every 4 h). The infections showed profuse sporulation at around 4 days postinoculation (dpi). In previous studies, it was found that in infections by this P. infestans strain, marker genes for biotrophic growth peak at 2 to 3 dpi, and those for necrotrophic growth peak at 4 to 5 dpi (16), while marker genes for sporulation are increasingly expressed from 3 dpi on and peak at 4 dpi. Expectedly, as the proportion of reads of each sample mapping to the P. infestans genome steadily increased over time (from 7% to 84%), the amount of reads mapping to the tomato genome decreased accordingly (from 87% to 8%) (Fig. S2). The transcriptome data were used to generate time point-specific submodels, which are subsets of the full model according to the expression of the genes in the model. Since gene expression and metabolic activity are not directly related, we used the Integrative Network Inference for Tissues (INIT) algorithm (54) that calculates a submodel with maximal agreement to the expression of genes in the model such that biomass can be produced (see Materials and Methods). This resulted in 25 submodels containing 32 to 44% of the reactions of the full model (Fig. 3), indicating that only a subset of reactions is required to form all defined biomass precursors. Over time, the total number of reactions per submodel decreases (Fig. 3), mostly due to a reduction in the number of P. infestans reactions. The number of tomato reactions is relatively stable across the submodels, while the number of transport reactions increases slightly over time. These results suggest that as the infection progresses, P. infestans relies less on its own metabolism but increasingly imports metabolites from the necrotic tomato lesion. Conceivably, once P. infestans switches to its necrotrophic lifestyle, nutrients can be more easily obtained from the decaying leaf tissue (15). To obtain an impression of the metabolic changes in the infection over time, we calculated all pairwise distances of the reaction content of the submodels (Fig. 3). This revealed a gradient of similarity scores, clearly reflecting metabolic changes over time, possibly as a response to changing nutrient availability in the tomato leaf tissue. A relatively large distance can be observed between groups of submodels, suggesting two more profound switches in metabolism at roughly 3 days and 8 h (3d/8h) and 4 days and 12 h (4d/12h), possibly reflecting a transition from the pathogens’ biotrophic growth to necrotrophic growth and sporulation (11, 12, 16, 55).

• Open in new tab
• Download powerpoint

FIG 3

Submodels based on dual-transcriptome data from a time course covering a full infection cycle of P. infestans on tomato leaf. Submodels are representative for the sampling time postinoculation shown in days (d) and hours on the y axis (middle) and x axis (left). Left, stacked bar graph indicating the number of reactions per species and number of host-pathogen transport reactions (not part of either species). Right, Jaccard similarity (intersection divided by union) between the reaction content of the submodels.

To obtain more insight into the metabolic processes of each submodel, we performed Fisher’s exact tests on the KEGG pathways in the submodels. A variety of metabolic pathways were overrepresented in submodels compared to the full model (Fig. 4). Consistent with the previously observed gradient of similarity scores (Fig. 3), groups of overrepresented pathways can be distinguished before and after 3d/8h. Notably, in the early submodels prior to the first transition, P. infestans reactions are enriched for several primary metabolic pathways, such as amino acid biosynthesis, glycolysis, and the TCA cycle. This could be a response to sugars in the apoplast that are still produced by photosynthesizing tomato leaf cells during the biotrophic phase of infection (56, 57). In contrast, late submodels, and in particular after the second transition (4d/12h), show that in tomato, several amino acid biosynthesis pathways are enriched, suggesting that during early infection, amino acids are mostly synthesized de novo by P. infestans but are later scavenged from tomato. This corroborates with a transcriptome study of P. infestans infecting potatoes, in which a diverse but generally higher expression of amino acid transporter genes at late infection time points was observed (16). In addition, early submodels are enriched in folate-related pathways on both sides, suggesting that folate plays an important role in early infection. This is in line with the finding that P. infestans has an unusually large repertoire of putative folate-biopterin transporter genes (18), although a divergent expression pattern was observed for these genes throughout infection (16). Arachidonic acid metabolism is enriched in midinfection submodels, which is a characteristic fatty acid in P. infestans known to elicit plant defense and to act as a signaling molecule in plants (58, 59). In summary, our analyses reveal profound changes in metabolic processes during infection of tomato by P. infestans and suggest that P. infestans reduces its metabolism at the expense of its host.

• Open in new tab
• Download powerpoint

FIG 4

Enriched KEGG pathways in transcriptome-based submodels from a time course covering a full infection cycle of P. infestans on tomato. Submodels are representative for the sampling time postinoculation shown on the x axis in days (d) and hours. Enrichment is calculated based on the reaction content of each of the submodels compared to the full model. Colors scale to the adjusted P value. metab, metabolism; biosyn, biosynthesis; syn, synthesis; deg, degradation.

Import of specific nutrients becomes increasingly essential to P. infestans.The ability of a model to maintain its functionality (biomass production) under perturbations (e.g., simulated reaction deletions) is often referred to as robustness (60). We can express the robustness of each submodel as the fraction of reaction deletions that do not disturb P. infestans biomass production (60). This revealed that early infection submodels have higher robustness to reaction deletions than late infection submodels (Fig. 5). This suggests that as the infection progresses, P. infestans largely shuts down accessory/alternative pathways to synthesize its biomass precursors, rendering the remaining reactions essential. The number of essential genes for P. infestans remained stable over time, with an average of 84 ± 5 essential genes (Table S1).

• Open in new tab
• Download powerpoint

FIG 5

Robustness of transcriptome-based submodels from a time course covering a full infection cycle of P. infestans on tomato. The y axis shows the robustness as the fraction of single reaction deletions that do not disable the biomass flux for P. infestans. Submodels are representative of the sampling time postinoculation shown on the x axis in days and hours.

To evaluate the importance of nutrient transport for P. infestans while colonizing tomato, we assessed the essentiality of the host-pathogen transport reactions in each submodel (61). A host-pathogen transport reaction is essential when its deletion disables biomass production and is partially essential when deletion in combination with deletions of other reactions disables biomass production. We found no essential nutrients for P. infestans when considering the full model, yet the submodels display various patterns of nutrient essentiality (Fig. S3). Early infection submodels have just a few essential nutrients, while mid- and late-infection submodels, after the first transition point, show various essential amino acid transport reactions. This corroborates our previous observations (Fig. 4) where the first transition point marks the switch from de novo synthesis of amino acids in P. infestans to an increased uptake from tomato.

Metabolomics can be used to refine transcriptome-based submodels.To assess to what extent the transcriptome-based submodels are coherent with metabolome data, we utilized untargeted metabolome data of tomato leaves colonized by P. infestans at 2 days and 12 h (2d/12h) and 5 days and 12 h (5d/12h) postinoculation. To relate the detected metabolites to metabolic fluxes, we hypothesized that metabolites strongly decreasing in abundance (log₂ fold change <−2) between these two time points are produced in the submodel of 2d/12h, and metabolites that increase (log₂ fold change >2) are produced in the submodel of 5d/12h (Fig. S4). Subsequently, we generated two additional submodels using INIT (54) based on the transcriptome data at 2d/12h and 5d/12h postinoculation, while enforcing the presence of the detected metabolites in the respective submodels. The submodels of the transcriptome-only (T) and the metabolome-guided (T+M) submodels differed in 4% (2d/12h) and 11% (5d/12h) of the reactions and in 3% and 7% of the metabolites, respectively (Fig. 6). Overall, the addition of metabolomics data leads to a net increase of reactions and metabolites in particular at the late submodels, suggesting that the T submodels are slightly too conservative; there may be a more active metabolism than suggested by mRNA levels of enzyme-encoding genes. It should be emphasized that both GEMs and metabolomics data only provide a partial description of the cell metabolism at the different infection time points. Discrepancies may arise since transcriptome-based submodels do not take into account any posttranscriptional regulation or posttranslational modification of enzymes, and misannotations and missing/blocked reactions potentially increase uncertainty in submodels. Moreover, since detected metabolites are not necessarily continuously metabolized, steady-state fluxes are not directly linked to measured metabolite abundances (62). Vice versa, metabolic fluxes do not necessarily yield detectable metabolites, for example, when reaction products are immediately consumed in downstream reactions. To predict metabolic fluxes more reliably, higher-resolution metabolome data and sufficient data points over time are prerequisites which could provide sufficient data points in time to calculate metabolite coefficients (63). Nonetheless, our initial results suggest that the integration of high-resolution metabolome data of tomato infection has the potential to refine stage-specific patterns that are embedded in the joint metabolism of the P. infestans-tomato interaction.

• Open in new tab
• Download powerpoint

FIG 6

Overlap of reaction and metabolite content of submodels, based on transcriptome data only (T) or transcriptome data and metabolome data (T+M). Data used to generate these submodels were obtained from P. infestans-infected tomato leaves at 2d/12h and 5d/12h postinoculation.