Meneco, a Topology-Based Gap-Filling Tool Applicable to Degraded Genome-Wide Metabolic Networks

Abstract

Increasing amounts of sequence data are becoming available for a wide range of non-model organisms. Investigating and modelling the metabolic behaviour of those organisms is highly relevant to understand their biology and ecology. As sequences are often incomplete and poorly annotated, draft networks of their metabolism largely suffer from incompleteness. Appropriate gap-filling methods to identify and add missing reactions are therefore required to address this issue. However, current tools rely on phenotypic or taxonomic information, or are very sensitive to the stoichiometric balance of metabolic reactions, especially concerning the co-factors. This type of information is often not available or at least prone to errors for newly-explored organisms. Here we introduce Meneco, a tool dedicated to the topological gap-filling of genome-scale draft metabolic networks. Meneco reformulates gap-filling as a qualitative combinatorial optimization problem, omitting constraints raised by the stoichiometry of a metabolic network considered in other methods, and solves this problem using Answer Set Programming. Run on several artificial test sets gathering 10,800 degraded Escherichia coli networks Meneco was able to efficiently identify essential reactions missing in networks at high degradation rates, outperforming the stoichiometry-based tools in scalability. To demonstrate the utility of Meneco we applied it to two case studies. Its application to recent metabolic networks reconstructed for the brown algal model Ectocarpus siliculosus and an associated bacterium Candidatus Phaeomarinobacter ectocarpi revealed several candidate metabolic pathways for algal-bacterial interactions. Then Meneco was used to reconstruct, from transcriptomic and metabolomic data, the first metabolic network for the microalga Euglena mutabilis. These two case studies show that Meneco is a versatile tool to complete draft genome-scale metabolic networks produced from heterogeneous data, and to suggest relevant reactions that explain the metabolic capacity of a biological system.

Fig 1. Gap-filling of metabolic networks with different heuristics.

The reactions of the initial network are depicted as black arrows. Seeds (e.g. growth medium) and targets are S and T circles, respectively. The labels on the arrows depict the stoichiometry of the reactions. Green dotted arrows represent reactions that can be added to the network (reference database). The purple arrows represent reactions proposed by different gap-filling tools. GapFill (A) reported two reactions as a minimal completion and two different combinations to produce biomass from T₁, T₂ and T₃, {R₃, R₇} and {R₇, R₈}. fastGapFill (B) reports one unique set of seven reactions to unblock all reactions of the example: {R₁, R₂, R₃, R₄, R₆, R₇, R₉}. It also add an import/export reaction for the reactant of the reaction producing T₃. In additions, 100 runs of MIRAGE (C) without scoring of reactions reported the following set of five reactions: {R₃, R₄, R₆, R₇, R₉}. Finally, Meneco (D) reported that three reactions are needed to restore the topological producibility of the three targets, with five different combinations. Therefore, the output of Meneco is the set of six reactions {R₃, R₄, R₅, R₆, R₇, R₈}.

Fig 2. Topology and stoichiometry-based producibility of compounds in a metabolic network.

Seeds and targets are S and T circles, respectively. The objective functions are formed by a reactions consuming the ensembles {T₁}, {T₂} and {T₃, T₄}. Arrows represent reactions. The labels of the reactions S ↦ na + b and J ↦ 2k depict their stoichiometry. Crosses indicate that metabolites cannot be produced. Check marks indicate that metabolites can be produced. The compound T₁ can always be produced according to graph-based criteria whereas the variation of the stoichiometric coefficient n can block the production of T₁ according to a balanced-mass stoichiometric framework: Flux Balance Analysis (FBA). By blocking the production of T₁, a variation of n can also block the production of all metabolites downstream. The compound T₂ can be produced according to graph-based criteria whereas the fact that f cannot be accumulated blocks the production of T₂ according to a balanced-mass stoichiometric framework. On the other hand, k remains FBA-producible through the cycle involving j, k and l whereas it is not producible according to our graph-based criteria. T₃ and T₄ are producible by both criteria.

Fig 3. Classification of reactions added by Meneco and fastGapFill in functional completed GEMs.

For each degraded GEM which recovered its ability to synthesize biomass after gap-filling, the functional classification (essential, alternative, blocked with respect to to biomass production) of reactions added to the GEM was calculated. (A) Comparison of Meneco and fastGapFill over the complete benchmark in terms of biomass production restoration. (B). Classification of reactions added in functional GEMs filled by Meneco. (C). Classification of reactions added in functional GEMs filled by fastGapFill.

Fig 4. Applying the gap-filling procedure Meneco on three benchmarks built from E. coli networks of different quality.

For three different reference networks (iJR904, iAF1260 and iJO1366), the Meneco tool was applied to 3,600 pairs consisting of a degraded E. coli metabolic networks (40 different networks with levels of degradation indicated by the abscissa) and a random biomass reaction (90 different targets sets). The initial E. coli network was used as a reference database to perform the completion. The percentage of functional networks after completion is indicated on the ordinate axis.

Fig 5. Biomass restoration and recovery of essential reactions due to completion of 10,800 degraded networks by Meneco.

For the 10,800 degraded iJR904, iAF1260 and iJO1366 networks, the gap-filling results were classified according to their status: (i) restored biomass production (green and white stripes), (ii) recovery of all essential reactions (green), (iii) exactly one missed essential reaction (orange) and (iv) more than one missed essential reaction (red).

Fig 6. Study of the 83 newly producible targets when combining E. siliculosus and Ca. P. ectocarpi metabolic networks.

When merging E. siliculosus and Ca. P. ectocarpi metabolic networks, 83 targets previously non-producible by E. siliculosus became producible. The 93 essential reactions related to their production by the Ca. P. ectocarpi network as well as the E. siliculosus ESTs were studied to assess whether the new producibility may be the result of a real interaction between the two organisms.

Fig 7. Vitamin B5 biosynthesis in E. siliculosus and Ca. P. ectocarpi.

Orange labels designate enzymes from the alga, blue labels correspond to enzymes from the bacterium.

Fig 8. Photosynthesis light reactions.

This pathway is a chain of oxidoreduction reactions transferring electrons from water to NADPH which will be used as a reductor in the Calvin-Benson-Bassham cycle. It is equivalent to a succession of three cycles of two reactions each and as such is not functional for Meneco unless the appropriate seed metabolites are added. The metabolites in light blue are those corresponding to the actual nutrients in the original set of seeds. The three metabolites that had to be included in the seeds to unblock each of the successive cycles are marked with a blue circle. If these metabolites are absent from the seeds, Meneco cannot propose a solution including the first three reactions, in light green, that were missing in the initial draft. Since NADP can be synthesized from NAD, there can be a functional path leading from WATER to NADPH.