Expanded Metabolic Reconstruction of Helicobacter pylori (iIT341 GSM/GPR): an In Silico Genome-Scale Characterization of Single- and Double-Deletion Mutants

Constraint-based modeling.The reconstructed metabolic network was represented by a stoichiometric matrix, S (m × n), where m is the number of metabolites and n is the number of reactions. Reactions within the network were mass balanced such that S · v = 0, where v was a steady-state flux vector (9, 22). Additional constraints on each reaction had the form α_i ≤ v_i ≤ β_i, where α_i and β_i represented the lower and upper limits, respectively, of the corresponding reaction flux. α_i was set to zero for irreversible reactions, whereas β_i was set to measured uptake rates for transport reactions or the V_max of the corresponding enzymes. The unit for each reaction flux was defined to be mmol/h/g (dry weight), in agreement with the typical unit of uptake rates and reaction fluxes used for microorganisms (8, 37). The growth rate was therefore reported in units of 1/h.

Reconstruction of iIT341 GSM/GPR.The previous metabolic network of H. pylori, here referred to as iCS291 (39), was used as a basis for the reconstruction of iIT341 GSM/GPR. Each gene and its associated reaction(s) in iCS291 were carefully curated and compared to the current biochemical and genomic data. Only reactions consistent with current data were retained from the previous reconstruction. These reactions were elementally and charge balanced before being included in iIT341 GSM/GPR. Reactions catalyzed by enzymes with putative gene assignments (1, 44) were also included. Reactions that did not agree with the most current data were either modified or omitted. Finally, we added reactions that were consistent with existing physiological data but for which no corresponding enzyme or loci have been found. We relied on information retrieved from the KEGG database (19, 20) for reactions gene products that have no specific biochemical characterization in H. pylori. The molecular formulas and charges for each metabolite in the model were determined based on a neutral intracellular pH (42). The reconstruction and functional analysis of iIT341 GSM/GPR were performed using the software package SimPheny (Genomatica Inc., San Diego, CA). A comprehensive list of all compounds and reactions of the metabolic network can be found in the supplemental material (http://gcrg.ucsd.edu/organisms/hpylori.html ).

Medium composition.The composition of the in silico media used in this study was adapted from that described by Schilling et al. (39) (Table 1). In order to determine the minimal medium needed by iIT341 GSM/GPR to fulfill the biomass requirement, each medium substrate was taken out of the rich medium containing all substrates for which corresponding transport systems had been defined. If the network was not capable of producing biomass without taking up a particular compound, that compound was deemed essential. The set of all essential compounds made up minimal medium I (Table 1). Minimal medium II contained d-glucose in addition to minimal medium I. Minimal medium III contained alternative carbon sources, which were also studied by Schilling et al. (39). Minimal medium IV contained all amino acids in addition to minimal medium III (Table 1). The uptake rates for protons and water were not constrained; the uptake rates of all other medium components were set to be maximally 20 mmol/h/g (dry weight). The oxygen uptake rate was set to be maximally 5 mmol/h/g (dry weight). These constraints were chosen arbitrarily, and they did not alter the observed growth phenotypes in single- and double-knockout deletions.

Deletion studies.In each single-gene-deletion experiment, we removed any reaction(s) associated with a particular gene from the network by constraining the corresponding reaction flux values to zero and then maximizing for growth. Gene deletion simulations that yielded a growth rate of less than 10⁻⁶ mmol/h/g (dry weight) in the tested medium were defined to be lethal. Each single-deletion experiment was performed with five different in silico media (Table 1). The reduction in growth rate upon deletion of a nonessential gene was determined by normalizing the growth rate of the mutant to that of the wild type in the same medium.

A double-deletion in silico experiment was performed by first identifying all reactions associated with two chosen genes and then setting the corresponding reaction fluxes to zero. Again, double-deletion mutants with growth rates less than 10⁻⁶ mmol/h/g (dry weight) in the tested medium were defined to be lethal. Double-deletion studies were carried out only in the rich medium, as the lethal double-knockout mutations found in this medium would also be lethal in the other media used in this study.

In this study, a gene was designated “essential” if its deletion resulted in a lethal mutation in all tested media. On the other hand, a “conditionally essential gene” was one whose deletion led to a lethal mutant in at least one medium but not in all. Last, we used the term “synthetic lethal genes” to describe pairs of genes whose double deletion rendered biomass production impossible but neither of whose single deletion affects cell viability.

Deletion studies on essential reactions for each biomass component.We used the updated metabolic network to identify reactions that are necessary for the synthesis of each biomass constituent in H. pylori. This in silico experiment was carried out by sequentially removing every reaction from the network and then maximizing for the production of one biomass component at a time. A reaction was declared to be necessary with respect to a particular biomass component if the network was not able to produce that component in the absence of that reaction.

All calculations for this study were done using Matlab (Mathworks, Inc., Natick, MA), with Lindo (LINDO Systems Inc., Chicago, IL) as the linear-programming solver.

Properties of the iIT341 GSM/GPR metabolic network.The updated reconstruction of H. pylori, iIT341 GSM/GPR, accounts for 341 genes, 476 intracellular reactions, 78 exchange reactions, and 485 metabolites. Thus, this network includes 50 new gene loci, 88 new reactions, and 82 more metabolites than does iCS291 (Table 2). New gene loci were identified from careful searches of the literature and the PyloriGene database (created based on the revised genome annotation [1]). Most of the newly added reactions and metabolites resulted from introducing distinct steps for reactions that were lumped together in iCS291; examples are those in the lipopolysaccharide and isoprenoid pathways. After the existing data were reevaluated, five reactions present in iCS291 were completely removed and not included in iIT341 GSM/GPR. They included one enzymatic reaction (methionine synthase) and four transport reactions, for dipeptide, oligopeptide, polypeptide, and glycerol. The presence of the methionine synthase would enable the network to synthesize l-methionine, which was found to be an essential amino acid of H. pylori (34, 38). The transport reactions were removed because iIT341 GSM/GPR does not account for peptide metabolism, due to incomplete information about further peptide conversions in H. pylori.

Each reaction in iIT341 GSM/GPR was assigned a confidence level from 1 to 4 (see Table S2 in the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html ) to reflect the strength of the supporting data. The criteria for assigning the confidence levels were designated as follows: 1, modeling purpose only; 2, sequence homology or physiological data; 3, genetic or proteomic data; and 4, direct biochemical data (47). The average confidence level of the reactions in iIT341 GSM/GPR is approximately 2.5.

A biomass equation is often used to represent all compounds that are thought to be required for the growth of an organism (45). In this study, the H. pylori biomass equation accounted for 52 metabolites (see Table S3 in the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html ). The fractional contribution of each of these compounds was adapted from the biomass composition of Escherichia coli (35), with exceptions for phospholipids (cardiolipin, phosphatidylethanolamine, phosphatidylglycerol, and phosphatidylserine) for which H. pylori-specific data were available (15). Thiamine, biotin, protoheme, and menaquinone were also included in the biomass equation, even though their fractional contributions to the biomass have not been measured. These compounds were assumed to have a fractional contribution of 6 × 10⁻⁶, which is equal to that of coenzyme A, the metabolite with one of the smallest fractional contributions in the biomass. Due to the nature of the linear-programming formulation, these assumed values may marginally affect the growth rate, but they usually do not change the result of a deletion experiment (46).

All metabolic pathways in the reconstruction were grouped into six different subsystems: central metabolism, amino acid metabolism, vitamin and cofactors, cell wall, nucleotide metabolism, and other (see Table S7 in the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html ). Analyses of gene distribution over subsystems showed that most genes can be found in the central metabolism subsystem (88 genes), followed by the vitamin and cofactor (62 genes) and the cell wall metabolism (60 genes) subsystems (Fig. 1A). The same scheme was employed to evaluate the distribution of reactions over subsystems (Fig. 1B). The vitamin and cofactor metabolism subsystem had the highest number of reactions (91 reactions). On the other hand, the central metabolism subsystem had the second-lowest number of reactions (61 reactions).

• Open in new tab
• Download powerpoint

FIG. 1.

Distributions of genes (A) and reactions (B) over subsystems. The reactions of each subsystem in panel B are additionally subcategorized into gene-associated and non-gene-associated reactions. w/o, without.

Identification and analysis of dead ends in iIT341 GSM/GPR.Metabolites that are only consumed or only produced within the metabolic network are referred to as “dead ends” (37). Dead ends usually represent missing knowledge about the underlying biochemical pathways in an organism. Experimental analyses of these dead ends can help identify new reactions and pathways (37). Moreover, compounds that are experimentally verified to be dead ends are also useful in studying pathway evolution. In particular, they may suggest that either a complete pathway existed in some ancestral bacteria or that the associated gene was recently acquired but the entire pathway has not been completed. The complete version of iIT341 GSM/GPR contains 48 dead-end metabolites (see Table S4 in the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html ). In the following paragraph, we describe the inconsistent data for one of these dead ends and our decision on how to resolve it. A more detailed discussion of other dead ends can be found in the supplemental data (see section 1 at the above-cited website).

The synthesis of biotin is well studied in various organisms. Pimelate, a biotin precursor, is converted to biotin in five steps (Fig. 2). The third step is carried out by the adenosylmethionine-8-amino-7-oxononanoate transaminase (EC 2.6.1.62; abbreviated AMAOTr) (Fig. 2), which converts 8-amino-7-oxononanoate (8aonn) to 7,8-diaminononanoate (dann) and S-adenosyl-4-methylthio-2-oxobutanoate (amob). Stoner et al. suggested that amob may undergo a spontaneous reaction with hydroxide ions under alkaline conditions to form 5′-methylthioadenosine and 2-oxobut-3-enoate (43). However, even if this reaction takes place in the cell, the further use of 2-oxobut-3-enoate by the metabolic network is not clear. Therefore, in order for the biosynthesis of biotin to be used in the network, we created a reversible artificial reaction (sink_amob) to allow the network to consume and produce amob freely.

• Open in new tab
• Download powerpoint

FIG. 2.

Schematic representation of biotin biosynthesis. The dotted line represents the hypothetical downstream conversion of S-adenosyl-4-methylthio-2-oxobutanoate (amob) (43). Metabolites a and b represent 2-oxobut-3-enoate and 5′-methylthioadenosine, respectively. Reaction and metabolite abbreviations used here can be found in Table S2 in the supplemental data.

Minimal medium requirement of iIT341 GSM/GPR.The minimal medium reported for iCS291 contained 12 metabolites (39). We tested iIT341 GSM/GPR to see whether the changes introduced in the updated model led to a different minimum set of required substrates. We found that, in contrast to iCS291, our model did not require cysteine or inorganic sulfate. Cysteine and methionine each contain a sulfur group that can be used by the network to synthesize sulfur-containing compounds such as biotin. Changes introduced in the methionine metabolism in iIT341 GSM/GPR enabled it to utilize methionine as a sulfate source (see section 1 in the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html for details). In total, the minimal medium for iIT341 GSM/GPR differed from that of iCS291 by only three compounds: l-phenylalanine and sulfate (or l-cysteine) were required only in iCS291, and pimelate was required only in iIT341 GSM/GPR, due to the changed biomass definition.

Single-deletion study: essential and conditionally essential genes.The presence of a large fraction of essential genes in a particular medium implies that H. pylori is especially adapted to its environment and has only limited ability to tolerate environmental disturbances (e.g., substrate supply, acidity, or mutagenic agents). In this study, we evaluated the fraction of genes that were essential in the five specified media. Our model predicted that a high percentage, 60%, of the total genes in the model were either essential or conditionally essential in minimal medium I. The percentage of genes found to be essential or conditionally essential decreased with increasing complexity of the medium and was lowest, 37.5%, in rich medium (Fig. 3). Notably, approximately 30% of the gene deletions were predicted to have no effect on the growth rate relative to that of the wild type in the same medium. In total, 128 genes out of 341 were predicted to be essential in all media tested. Another 75 conditionally lethal mutants were also identified in minimal medium I.

• Open in new tab
• Download powerpoint

FIG. 3.

Results of the in silico single-deletion studies. The growth rate percentages of lethal and nonlethal deletions are shown for each medium. The reduced-growth section indicates the percentage of mutants displaying reduced growth compared to that of the wild type in the same medium. See Table 1, footnote a, for medium abbreviations.

In classifying genes and reactions into subsystems, we were curious to see if a certain subsystem showed a larger number of essential genes than did others. As expected, most of the essential genes belong to two subsystems: cell wall (51 genes, 39%) and vitamin and cofactor (33 genes, 26%). The remaining 44 essential genes (45%) were distributed rather evenly among the other subsystems (Fig. 4A). Closer examinations of the distribution of conditionally essential mutations showed that a large number of conditionally essential mutations can be found in the amino acid (27%) and central metabolism (25%) subsystems, but only few in the cell wall subsystem (4%) (Fig. 4B).

• Open in new tab
• Download powerpoint

FIG. 4.

Distributions of essential genes (A) and conditionally essential genes (B) over subsystems. The number of essential or conditionally essential genes per subsystem is shown in parentheses.

Validation of predicted single-deletion growth phenotypes with experimental data.We used iIT341 GSM/GPR to compute the essentiality of every gene in the network and compared the results with reported experimental data to assess the predictive capability of the model. Experimental data were available for 72 out of 341 mutants for comparison (Table 3). These experiments were usually carried out with rich media containing an agar base with either horse or sheep blood and antibiotics, but the detailed chemical compositions of the media were not reported (Becton, Dickinson and Company, Franklin Lakes, NJ). We therefore felt that it was most appropriate to compare these data with our results with minimal medium IV and with rich medium (Table 3). The model correctly predicted 73% and 75% of gene deletion experiments for the 72 mutants when compared to the results with minimal medium IV and rich medium, respectively.

Predictive capability of iIT341 GSM/GPR compared to that of previous reconstruction iCS291.To verify that iIT341 GSM/GPR represents a more accurate description of H. pylori metabolism than does iCS291, we quantitatively evaluated its improvement on gene essentiality prediction. We first identified common genes in both reconstructions for which experimental gene essentiality data were available (Table 3). Then, we applied the same medium conditions to both models to calculate the effect of these genes on growth rates. In total, iIT341 GSM/GPR predicted 44 out of 59 (75%) deletion experiments correctly, while iCS291 was correct only in 41 out of 59 cases (69%); both results were with minimal medium IV. In rich medium iIT341 GSM/GPR predicted 42 phenotypes (71%) correctly, while iCS291 achieved only 37 (63%) correct predictions. Taken together, these results show that iIT341 GSM/GPR is more accurate at predicting gene essentiality.

Analysis of false predictions.The analysis of false predictions by the models is also informative and can be used to either update the model or reevaluate experimental data. For example, both models failed to predict the essentiality of HP1100 (6-phosphogluconate dehydratase [EDD]) and HP1099 (2-dehydro-3-deoxy-phosphogluconate aldolase [EDA]), which have been found essential by Chalker et al. (3). However, Wanken et al. were able to grow HP1100 mutants (49). The gene products of HP1100 and HP1109 are part of the Entner-Doudoroff pathway, which H. pylori can use as an alternative to the glycolytic pathway to metabolize d-glucose (25, 30). Other experimental observations (25, 31) also showed that H. pylori is not dependent on d-glucose as a carbon source. Our model predictions were found to be consistent with the last three studies mentioned.

In addition, our model did not predict HP0832 to be essential. Chalker et al. suggested that HP0020 (carboxynorspermidine decarboxylase) could replace HP0832 in catalyzing the last step of the spermidine biosynthesis pathway in H. pylori (3). However, HP0020 was not included in the reconstruction, since its function as spermidine synthetase (SPMS) was not confirmed experimentally. Last, in contrast to iCS291, our model did not correctly predict the essentiality of HP0808 (acyl carrier protein synthase [ACPS]). The reason for this is that two steps of the pantheonate biosynthesis were lumped together in iIT341 GSM/GPR (EC 2.7.87 and 3.1.4.14) to avoid including an apoprotein. As a consequence, this reaction does not account for the acyl carrier protein, rendering the gene nonessential in silico.

For certain knockout mutants, the in silico predictions were false for only a subset of tested media. In order to determine which compounds in the media rendered these phenotype switches, we carried out a more detailed analysis. Each compound was added separately to minimal medium I, and then the deletion experiments in silico were repeated. For example, HP0086 (malate dehydrogenase; EC 1.1.99.16) was essential only when l-aspartate, l-asparagine, or l-threonine was not available in the in silico medium (Table 3, last column). Interestingly, we found that the mutant with deleted pyruvate dehydrogenase (PDH2; encoded by HP1108 to HP1111) was viable when both thymidine and acetate were added to the medium. Therefore, differences in medium conditions might explain the conflicting experimental results by Chalker et al. (3) and Hughes et al. (16) for the pyruvate dehydrogenase mutants. The complete list of results of this deletion study can be found in Tables S5 and S6 of the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html ).

Nonlethal single-knockout mutants with reduced growth rates.The reduction in growth rates of the mutants relative to that of the wild type in the same medium reflects the importance of the deleted genes within the metabolic network (Fig. 5). The knockout of genes involved in the respiratory chain (genes for BC10, CYOO_HP, and ATPS4r) was not strictly essential but led to reduced growth rates of the mutants relative to that of the wild type in the same medium (Fig. 5). This observation indicated that the network was mainly using the respiratory chain for ATP production under substrate-limiting conditions, while the presence of other carbon sources enabled it to produce energy via pathways other than oxidative phosphorylation. We found that the amount of reduction in the growth rate decreased with the increasing complexity of the medium. Reduced growth of the mutant relative to the wild type in the same medium also allowed for the evaluation of substrate usage. For example, the deletion of HP1100 or HP1103, whose gene products (EDD and HEX1) are necessary to metabolize d-glucose, led to a 60% decrease in growth rate relative to that of the wild type in minimal medium II (which contained d-glucose in addition to minimal medium I). The addition of alternate carbon sources (as in minimal medium III) restored 90% of the growth rate observed in the wild type. This result indicated that the mutant, as well as the wild type, can metabolize alternate carbon sources efficiently. In fact, experimental studies found that H. pylori uses d-glucose only when most other energy sources have been depleted (31).

• Open in new tab
• Download powerpoint

FIG. 5.

Variation in growth rate observed in single knockouts of nonessential genes. The percentage growth rate was calculated relative to that of the wild type under the same medium conditions. The reaction abbreviations used here can be found in Table S2 of the supplemental data at http://gcrg.ucsd.edu/organisms/hpylori.html ). See Table 1, footnote a, for medium abbreviations.

Double-deletion mutants of iIT341 GSM/GPR.Double-deletion experiments are generally difficult to perform on a large scale in vivo. In silico models provide tools to rapidly predict the phenotype of such double-knockout mutants in various media. Here we carried out double-deletion studies in rich medium for every pair of genes identified as nonessential in single-deletion studies (Table 4). In total, more than 22,000 possible double-knockout mutants were screened. We identified 47 conditionally lethal double mutants, involving 64 different metabolic genes, for iIT341 GSM/GPR. These synthetic essential genes represented almost 18% of all genes included in the model and 30% of all nonessential genes in rich medium. Most lethal synthetic mutants were found when a gene of a particular pathway and a gene for the transport for either the precursor or product of this pathway were deleted. For example, deletion of urease (HP0067 to HP0070, HP0072, and HP0073) and the urea transporter (HP0071) together led to a lethal synthetic mutant (Table 4). This result was interesting because neither did H. pylori require urea in media nor was urease (UREA, HP0067 to HP0070, HP0072, and HP0073) found essential in rich medium (Table 3).

Required reactions for the production of each biomass compound.We carried out reaction deletion studies to determine which network reactions are necessary for the production of each biomass constituent. Here, we simulated reaction deletions instead of gene deletions, so that we could also consider non-gene-associated reactions in the network. We removed one reaction at a time from the network and then maximized the production of each biomass compound. Biomass compounds were grouped based on the reactions needed for their syntheses. The first cluster accounted for 46 out of 52 biomass compounds. The syntheses of compounds in this cluster required 90 reactions on average, with a minimum of 80 reactions (Table 5). The second cluster included mostly phospholipids. Their synthesis required a common set of 17 metabolic reactions, with an average of 19 reactions. There were two biomass compounds, 5-methyltetrahydrofolate (5mthf) and lipopolysaccharide (lps_hp), that did not have the same set of required reactions as the other two groups of biomass compounds. The syntheses of 5mthf and lps_hp required 34 and 45 reactions, respectively; these two sets of reactions did not overlap. It is notable that only two biomass compounds, spermidine (sprm) and 5mthf, required oxygen uptake (O2tr) for their production.

In some cases, only one reaction in a pathway was required for the synthesis of a particular biomass constituent (Table 5), as was the case for malate synthase (MALS) (tricarboxylic acid cycle) in the synthesis of 5mthf. This suggests that such reactions are also involved in pathways other than those with which they have been traditionally associated. In many other cases, however, all reactions in the entire pathway were needed for the synthesis of a biomass constituent. Examples are reactions in the menaquinone biosynthesis and methionine salvage pathways (Table 5).