Abstract
Genome-scale metabolic reconstruction (GEMR), along with flux balance analysis, has been widely used to study complex metabolic networks in several microbial organisms. This approach is of particular applicability in biological systems where the lack of kinetics data is typical. This is the case of plant–pathogen interactions, where these methods open the possibility of studying host metabolic network phenotype during the interaction with pathogens. Since GEMRs are based on sequenced genomes, its applicability to organisms where genomic information is lacking is limited. Here we describe an alternative approach to GEMR: targeted metabolic reconstruction, where network reconstruction is guided by transcriptomic data instead of genomic information. This approach is being applied successfully in our laboratory for the Phytophthora infestans—Solanum tuberosum pathosystem.
INTRODUCTION
The mathematical modeling of metabolism is particularly promising in plants as it offers systems approaches to analyze the structure, dynamics and behavior of complex metabolic networks. The main aim is the identification of suitable targets for metabolic engineering, having as targets the improvement of production yields and the nutritional value of crops, as well as the understanding of certain processes such as resistance or defense from pathogens. Multiple approaches have been proposed in order to model the interplay of pathogen with host (for reviews, see [1–4]). However, although the plant metabolism is continuously being characterized, to the best of our knowledge, no detailed flux balance analysis (FBA) model for the prediction of fluxes in plants challenged by microorganisms has been published to date. Consequently, we will review the most-studied plant pathosystems and the molecular tools used to characterize them to highlight the fact that there is enough information and data to start their molecular in silico reconstruction. We will revise the applicability of genome scale and targeted reconstructions for modeling plant pathosystems.
PLANT PATHOSYSTEMS: HOW WELL DO WE KNOW THEM?
The mathematical modeling of complex metabolic networks require a minimum set of information, ideally gathered using genomic, transcriptomic and regulomic tools. We revised the better-characterized plant pathosystems to show that enough information has been gathered to date in order to model the host interacting with a pathogen. Among the pathosystems that have been extensively studied, we can include the paradigm plant–bacterial interaction Arabidopsis—Pseudomonas syringae pv. tomato, systems involving oomycetes (Hyaloperonospora parasitica and Phytophthora infestans) and finally a fungus–plant (Magnaporthe grisea—rice) interaction.
It is now clear that plant–pathogen interactions can be divided in two stages in what is called the zig-zag model [5]. During the first stage, which has been called basal defense (and, in several cases, nonhost resistance), the plant recognizes general features in pathogens such as P. syringae that are universally associated with microbes. These are called MAMPs (microbe-associated molecular patterns) [6], and they can range from proteins such as flagellin [7] and the translation elongation factor (Ef-Tu) [8] to lipopolysaccharides [9]. MAMPs are recognized by membrane receptors in plants such as Arabidopsis [10–12]. This triggers the first line of defense, which for most microbes is efficient to halt growth of the pathogen. Tangible biochemical and cellular changes occur during this defense and they include a change in nitric oxide concentration, kinase activation, the synthesis of pathogenesis-related proteins and other defense compounds [13]. However, microbes that are pathogenic on a specific host (e.g. P. syringae on Arabidopsis) display an arsenal of proteins called effectors, which are injected into the host plant through specialized secretion systems, suppressing this first line of defense [14–16]. Plants in turn, have evolved to recognize these effectors using Resistance proteins (R proteins), which directly or indirectly recognize effectors secreted by the specific pathogen [17]. Biochemical changes also occur in this line of defense, and include similar but stronger calcium spikes, changes in redox potential, de novo synthesis of defense-related compounds, in addition to the most-studied form of programmed cell death in plants, the hypersensitive response (HR) [18]. Other effector proteins suppress this recognition or the plant defense response derived from it. Therefore, plant and pathogen are involved in a permanent arms race at the population level for recognition or its avoidance.
The Arabidopsis—Pseudomonas syringae pv. tomato pathosystem is by far the best-studied plant–pathogen interaction system, and it has served as a model for more obscure and probably more economically important plant–pathogen interactions. The short life cycle and the ease of generating biomass for Arabidopsis and P. syringae, together with the fact that both genomes have been sequenced [19, 20], has accelerated the discovery of key players and the construction of models and paradigms in plant–pathogen interactions. Most of the zig-zag model has been largely based on data collected for this pathosystem [21]. One of the best-studied PAMP is flagellin from P. syringae [7, 22], along with the set of signal transduction events that take place after its recognition by Arabidopsis [23, 24]. In addition, a few of the resistance proteins have been identified and thoroughly characterized through structure-function studies. These include the Rps2 protein, which recognizes AvRpt2 from P. syringae pv. tomato [25–27], Rpm1, which recognizes AvrRpm1 from P. syringae pv. maculicola [28, 29] and Rps4, which recognizes AvrRps4 in P. syringae pv. phaseolicola [25]. Besides this first line of recognition of the pathogen by the host, the downstream signaling cascades have been extensively studied in this pathosystem [31–33].
The Hyaloperonospora parasitica—Arabidopsis pathosystem also shows all the characteristics for being an ideal model for metabolic–regulomic reconstruction: two model organisms with their genomes already sequenced and a considerable knowledge of their molecular interaction. Hyaloperonospora parasitica (formerly known as Peronospora parasitica) is a natural oomycete pathogen of Arabidopsis [34]. This plant–pathogen interaction became recognized as a laboratory model in the 1990s. Since then, this interaction has allowed the elucidation of the molecular mechanisms of disease resistance and has led to the identification of several Arabidopsis resistance genes (R-genes) [34], as well as loci required for susceptibility [35]. In fact, the genetic diversity of the molecular determinants of the interaction between Arabidopsis and Hyaloperonospora makes this system an excellent one in which to study coevolution [36]. Rentel et al. [37] developed an efficient and genetically amenable system to characterize the interaction between the resistance gene RPP13 and the avirulence gene ATR13. However, genetic studies and specifically the cloning of effector genes from H. parasitica have been difficult because of its obligate biotrophic lifestyle, making the elucidation of the holistic interaction a slow process.
Research on Phytophthora pathosystems has accelerated significantly in the last few years with the advances in microbial and plant genomics and the resulting resources such as microarrays, both for the plant and the pathogen partners. Recently, in order to gain a better understanding of the compatible interaction between potato and P. infestans, several studies have focused mainly on this type of interaction [38, 39]. These studies showed that even in a compatible interaction, photosynthesis is repressed at early time points after challenging the plant with the pathogen. Additionally, genome-sequencing projects are finished or under way for four Phytophthora species [40, 41] and their hosts, potato (Solanum tuberosum) and tomato (Solanum lycopersicum) and of course the plant model Arabidopsis, which is also a host for species of Phytophthora. Several genomics databases are available for the study of Phytophthora—plant interactions [42].
Finally, Magnaporthe grisea has emerged as the principal model organism for studying the molecular basis of fungal diseases in plants because of the availability of genetic and molecular tools of both the fungus and the host plant, rice [43]. This system is probably the best studied in terms of events that occur before the invasion of the host by the pathogen, where propagules from the fungus adhere and generate structures that allow the generation of high internal pressures that are used to mechanically surpass the physical barriers of the rice leaf surface [44, 45]. The exploitation of transcriptomic resources coupled with mutant collections of rice have led to the accumulation of data associated to the molecular dialog established between the plant and the fungus [43]. Several resistance gene systems have also been discovered in this pathosystem [46–49] and they have served as models for the study of other plant–fungal interactions.
EXPERIMENTAL APPROACHES TO THE STUDY OF MOLECULAR PLANT–PATHOGEN INTERACTIONS
The best-studied plant pathosystems were analyzed first using a one-gene-at-a-time or a one-protein-at-a-time approach. However, the sequenced genomes from hosts and pathogens at the beginning of this decade has allowed the employment of more holistic approaches in the study of plant–pathogen interactions. Oligonucleotide or cDNA microarrays have been the most widely used transcriptomic tool, followed by validation approaches such as quantitative reverse transcriptase PCR (qRT–PCR) [50–52]. They have been especially used not only to characterize the resistance and defense pathways, but also to understand the role of effector proteins in suppression of plant defense. For example, transcriptomic studies using microarrays defined the role of a subset of effectors in manipulating abscisic acid (ABA) signaling in the Arabidopsis—P. syringae pv. tomato interactions [53]. In addition, independent studies have demonstrated that several effectors target the plant secretory system, presumably to suppress cell wall-based defense responses [54, 55]. Microarray transcription profiling has also been used to globally characterize mutants in defense pathways in order to further define their role in resistance [56, 57], to determine major signal transduction routes in plant defense [58–61], and to define their putative targets in the transcriptional machinery [61]. With a different approach, gene expression profiles for both the pathogen and the plant were followed using SuperSAGE to describe the rice—Magnaporthe grisea interaction [63]. Major disadvantages when using microarrays or other transcriptomic techniques for network reconstruction come from the intrinsic qualitative features.
More quantitative approaches must be taken in order to generate data that are useful in model reconstruction. New generation sequencing technologies have been used, mainly in animal systems for transcriptional profiling, in a group of techniques that are collectively called RNAseq [64]. These are expected to be more quantitative in nature than microarray transcript profiling. However, reports of these techniques for the analysis of plant–pathogen interactions have not been published so far.
In addition to studying the transcriptome, studying the proteome has reinforced the idea that an important set of changes occur at this stage upon infection. 2D gels coupled with mass spectrometry in plant pathosystems have allowed the elucidation of important changes in the presence and posttranslational modifications of proteins during disease, basal defense and resistance gene-mediated defense [65]. Specifically, several posttranslational modifications were detected in enzymes of the antioxidant group. A similar study was carried out to characterize the phosphoproteome under infection, using isobaric tag for relative and absolute quantitation (iTRAQ) [66]. This system has a higher sensitivity than the classical 2D gels and allows for the observation of more subtle differences in the proteome, as well as a higher reproducibility. These studies highlighted the importance of the characterization of changes at the protein level, in addition to changes at the RNA level. However, proteomic studies do not allow for deeply quantitative analyses, also limiting their application to network reconstruction.
The Pseudomonas—Arabidopsis and the Magnaporthe—rice interactions have also been characterized by evaluating protein–protein interactions through yeast two-hybrid and coimmunoprecipitation approaches. These have been used for the identification of targets for pathogen effectors [25, 27, 67–69] and for the identification of protein partners for signal transduction components [70, 71]. These techniques are, again, mainly qualitative, also limiting their use in network reconstruction.
We will focus from now on in the postgenomics era and the tools used to validate biologically (functionally) the results obtained during the genomics times. The goal in the postgenomics era is to link sequences to phenotypes in a rapid and efficient manner [42]. For a complete review of the classical and genome-scale functional genetic analyses of Phytophthora pathosystems that can be adopted for other pathosystems see the review by Huitema et al. [42]. VIGS is a central technique facilitating the genetic dissection of the plant pathosystems interactions [39, 41], except in the case of Arabidopsis, where the genome has been significantly covered with mutants that are available in public databases [72]. The high-throughput in plant expression screening identifies downstream components of the plant defense regulome [73]. In their study, Nasir et al. [73] used an interesting combination of techniques to determine downstream signaling cascades for HR and cell death. Loss of function mutagenesis was the preferred genetic approach to determine the regulomics of defense and cell death [74]. However, due to genetic redundancy or even lethal effects resulting from the mutagenesis, some genes could not be discovered or related/integrated into the complex networks of defense regulation. Gain of function is therefore an interesting alternative. Two ways of performing it is through T-DNA activation tagging and functional expression screening of cDNAs. In addition, Nasir et al. [73] performed a SuperSAGE transcriptional study in the leaves with overexpression of the identified genes in their initial screening to reveal the downstream targets of the selected gene. In this particular case, VIGS was even used to determine the complete function of the gene. In one study, authors expressed a set of 54 candidate RXLR effectors of P. infestans in late blight resistant Solanum plants and identified a variety of effector responses, some of which could be R-AVR interactions [75]. In a second study, 62 effectors were in planta expressed [76]. This effectoromics approach revealed that 16 of the 62 genes were expressed in planta.
In the case of the rice—Magnaporthe grisea interactions, not only one-gene studies and transcriptomics approaches are available but recently a protein–protein interaction network for the fungus was also determined [77]. More than 11 000 interactions were predicted for 3017 proteins from the fungus [77]. The interactions of the secreted and/or pathogenicity proteins are interesting for obtaining the complete picture of a plant cell under infection. For instance, this network may play an important role in choosing bait and prey in yeast two-hybrid experiments. From the plant side, targeted metabolite profiling using GC-MS confirmed the re-programming strategy by the pathogen to suppress plant defense [78].
As we have seen, several approaches have been developed to study these complex networks of interaction by means of genomic, transcriptomic and other omic tools. Recently, these networks have been characterized by the use of computational modeling. Some of these approaches require the use of kinetic information, which often is lacking even in known model organisms. This situation limits the use of systems biology approaches in most plant pathosystems. Nevertheless, similar approaches such as metabolic network reconstruction, which allows the study of biological processes at metabolic and regulatory scales yet not requiring kinetics information, can also be used.
METABOLIC NETWORK RECONSTRUCTIONS
The implementation of genomic technologies has led to the generation of more questions than answers for the description of biological systems based on their properties. Biological systems could be analyzed either by a holistic or a reductionist approach. Once the approach is chosen, one can elucidate features of the system implicated in defining its behavior in a defined environment [79]. Genome metabolic reconstructions constitute an interesting strategy for the holistic approach. However, it requires defined experimental data that are not available for every system. Genome metabolic reconstructions are models that aim to represent all the interactions among metabolites inside the cell through metabolic pathways. Methodologies for the construction of these networks are mainly classified into manual [79] and ab initio [80].
Manual reconstructions are information driven. They usually require the use of databases such as KEGG [81]. However, this information is not always reliable due to errors in genome annotations caused, for example, by misinterpretation of partial Enzyme Commission numbers [82], thus manual reconstructions also require the use of exhaustive literature sources. The final step during the reconstruction process is the experimental validation, which is commonly carried out through phenotypic characterization using parameters such as growth and substrate consumption [83]. These reconstructions have been useful for different ends. For example, they have a potential application for improving the performance of microbial fuel cells in Pseudomonas aeruginosa [84] or alcohol synthesis in Escherichia coli [85]. They have been used to carry out comparative genomics or proteomics allowing the identification of metabolic similarities among species. This is the case for Salmonella, whose metabolic network was reconstructed and compared with that of E. coli, resulting in an interesting overlap that explained similarities between commensals and pathogens [86]. Also, 3D networks have been proposed by adding either experimental or model-determined 3D protein structures to the reconstructed network of Thermotoga maritima [87]. Particularly, in the case of plant research, mostly all metabolic reconstructions fall in to the categories previously explained. Grafahend-Belau et al. [88] proposed a metabolic network composed of 257 reactions that describes the primary metabolism of the developing endosperm of barley (Hordeum vulgare). Interestingly, the model successfully established the expected correlation between oxygen depletion and metabolism and uncovered an unexpected role of the pyrophosphate metabolism under conditions of oxygen deficiency. To our knowledge this was the first metabolic reconstruction in a plant and in a particular process of development, opening a broad spectrum of possibilities for the study of other biological processes in plants, ranging from tissue development to environmental and pathogen interactions. Nitrogen fixation was also modeled using Rhizobium etli [89] in order to acquire a better understanding of one of the most important symbiotic processes in plant development. The physiological capabilities of R. etli were evaluated during the process of nitrogen fixation and, after flux balance analysis (FBA) modeling (see below), the results seemed to adequately describe experimental responses when fixing nitrogen from the environment. This model constitutes a special case, as it describes a plant–microbe interaction where the objective function (OF) was a linear combination of chemical compounds that turn to be essential during fixation process, rather that the typical use of biomass.
GENOME-SCALE RECONSTRUCTIONS
Over the last years, there has been an increasing effort in the scientific community for the development of metabolic reconstructions at the genome scale. Since the first genome-scale reconstruction model (GSRM) for Haemophilus influenza 10 years ago [90], >50 genome GSRMs have been published [91], around 30% of which come from bacterial genomes [92]. On average, metabolic network reconstructions of prokaryotes encompass 600 metabolites, 650 genes and 800 reactions, and reconstructions on eukaryotes include 1200 metabolites, 1000 genes and 1500 reactions [92]. GSMR is based on the fact that metabolite concentrations and reactions can be interpreted as the metabolic state of a cell in a given time and under particular conditions [92, 93].
A typical high confidence metabolic reconstruction is built through a four-step process [94] (which we have further extended as depicted in Figure 1):
The reconstruction is started from gene annotation, coupled with information from biological databases.
Primary literature is examined, and this information is used to curate the initial reconstruction.
The curated reconstruction is translated into a mathematical model and then analyzed by constraint-based methods.
Model predictions are validated through comparison with phenotypic data and necessary curations are performed.
Main steps in genome scale metabolic reconstruction.
Main steps in genome scale metabolic reconstruction.
A common strategy for the interrogation of these reconstructions is by means of FBA. FBA is a constraint-based optimization method that facilitates the computational prediction of systemic phenotypes in the form of fluxes of reactions [95]. For instance, given a set of available nutrients for an organism, FBA allows for the prediction of the set of fluxes of metabolic reactions that optimize the growth for that organism. The FBA model usually optimizes for a characteristic, which is commonly called the ‘Objective Function’. Typically, growth is used as OF in the form of biomass, and it is represented in the model by a specific subset of reactions [96, 97]. Biomass was also used as the OF for FBA in Barley, optimizing all the biosynthetic precursors and cofactors required for biomass production [88]. However, we hypothesize that under particular conditions for some metabolic reconstructions in plants, other OFs can be described. For instance, in potato it is feasible to use starch production and storage as OF instead of a typical biomass set of reactions which makes sense since starch storage could be seen as an indicator of growth.
In the case of most metabolic reconstructions under FBA, these fluxes of reactions correspond to intracellular biochemical networks that, due to the lack of kinetics information, are assumed to operate under steady-state conditions. This assumption has yielded reasonable approximations of flux distributions, and agreement with experimental data. For instance, this approach showed an 85% consistency of gene essentiality for the genome-scale reconstruction of E. coli [98] and 70% consistency for gene essentiality for the reconstruction of Pseudomonas aeruginosa [92].
TARGETED RECONSTRUCTION
One of the main steps on a metabolic reconstruction is to identify the ‘metabolic genotype’ [98], which is the set of ORFs that encode for proteins involved in metabolism. On average, between 6% and 13% of all ORFs in eukaryotic genomes are expected to participate in metabolism, and about 18% in prokaryotes [92].
Figure 2 depicts a phylogenetic tree showing known metabolic reconstructions for Archaea, Eukarya and bacteria domains. A lack of metabolic reconstructions for plants is obvious in Figure 2, which Oberhardt et al. [94] have clearly identified as a gap that indicates future efforts for the systems biology community. To date, the only plant GSRM is for Arabidopsis thaliana [97], which includes 2300 reactions and three different models of the reconstruction, corresponding to three different levels of confidence. These levels were established, based on the confidence of the identified reactions on semi-automatic searches on KEGG and AraCyc online databases.
![Known metabolic reconstructions in three domains (adapted from [93]). To date, the most represent domain in metabolic reconstructions is bacteria, followed by Euchayota. Among the latter, the phylum Streptophyta is the least represented, with A. thaliana, although there exists a small metabolic reconstruction for seed development in Barley.](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/bib/12/2/10.1093/bib/bbq009/2/m_bbq009f2.jpeg?Expires=1500196180&Signature=DsnK1rlYrqyoW9SxlY6MKCAQFtF~h6sLEAOqvbRH-05urTO2H4QqSGk8CIY5Y9SGF~SaFUHwwfzbeZ5fvY9Cwiz-kFQJQrp9Lg93txcWlLdFHYtD3O9rHOg6uMb0WFMFyUJ2sJSVfZ7adJakUW7dpmbn26QrABM9wVFWnPffql6uqQVmQIXsXek9wleSrnUu0ADOhVCxiDT37WOPDO7FxCROgxnuwPpCqS~zqjuAhUpErOvfv06XfKB80nyDTaj6vMMUeHOrl6AaftA6Hl0HYol~aeygoS34sMkQeSJsMMhGZVk9FQFU7UkON1GpdIhlAH1DlfMnGqcVcKZDiI1T4A__&Key-Pair-Id=APKAIUCZBIA4LVPAVW3Q)
Known metabolic reconstructions in three domains (adapted from [93]). To date, the most represent domain in metabolic reconstructions is bacteria, followed by Euchayota. Among the latter, the phylum Streptophyta is the least represented, with A. thaliana, although there exists a small metabolic reconstruction for seed development in Barley.
Known metabolic reconstructions in three domains (adapted from [93]). To date, the most represent domain in metabolic reconstructions is bacteria, followed by Euchayota. Among the latter, the phylum Streptophyta is the least represented, with A. thaliana, although there exists a small metabolic reconstruction for seed development in Barley.
This gap in genome-scale plant metabolic reconstructions can be explained by the fact that just eight complete genomes are available in plants and there are just six ongoing sequencing projects (Table 1). When these numbers are compared with 988 complete genomes and other 2081 genomes in progress for microbial organisms, it is clear that in order to study plant pathogens effect on its hosts, it is necessary to approach metabolic network reconstruction from a different point of view.
Genome sequencing projects in plants
| Completed sequencing projects | Ongoing sequencing projects |
|---|---|
| Arabidopsis thaliana | Brachypodium distachyon |
| Glycine max | Carica papaya |
| Medicago truncatula | Lotus japonicus |
| Oryza sativa | Manihot esculenta |
| Populus trichocarpa | Solanum lycopersicum |
| Sorghum bicolor | Solanum tuberosum |
| Vitis vinifera | |
| Zea mays |
| Completed sequencing projects | Ongoing sequencing projects |
|---|---|
| Arabidopsis thaliana | Brachypodium distachyon |
| Glycine max | Carica papaya |
| Medicago truncatula | Lotus japonicus |
| Oryza sativa | Manihot esculenta |
| Populus trichocarpa | Solanum lycopersicum |
| Sorghum bicolor | Solanum tuberosum |
| Vitis vinifera | |
| Zea mays |
Our laboratory approach to this issue is a metabolic ‘targeted reconstruction’ based on expression data (Figure 3). In this approach, some steps have been added to the classical four-step process depicted in Figure 1. Step one corresponds to the analysis of expression data. Since typical expression data in plant–pathogen interactions come from microarray analysis of infected plants [39], preprocessing, normalization and identification of differentially expressed genes (DEG) are carried out. In step two metabolic genotype is established, in this step DEG are re-annotated by means of sequences similarity and profiles search and those genes involved in metabolism are identified. In our research we have identified that about 10% of DEG come from the metabolic genotype, which agrees with data from GSRM [92]. Step three establishes the scope of the network reconstruction. In this step, metabolic genotype is mapped into known metabolic networks in plants using online resources as well as primary literature data. Thus, the minimal set of metabolic networks that will be part of the reconstruction is established. For instance, if a given molecule such as carbonic anhydrase is identified as part of the metabolic genotype in step two, and then in step three it is found to be involved in nitrogen metabolism in potato, all known nitrogen metabolism reactions in potato are added to the reconstruction. Step four accounts for the addition of these reactions in an iterative process based both on curated metabolism databases and primary literature.
Targeted reconstruction of metabolic networks.
Targeted reconstruction of metabolic networks.
In step five, network reconstruction is translated into a mathematical framework and flux through the network is assessed. Reconstruction is split into subnetworks and gaps that can be blocking fluxes are filled, based on literature and/or information from related species. Finally, in step six, network is studied and simulations are performed using FBA. In our approach, DEG data are also used to drive in silico experiments and predict cell behavior. Information from genes that are repressed or overexpressed, according to differential expression analysis, is used to allow or block flux through specific reactions in the reconstruction. Therefore, biological information is represented by these ‘perturbations’ on the computational model and cell phenotype can be studied in silico. Network predictions can then be confirmed (or discarded) by wet lab analyses.
Targeted metabolic reconstruction is being used by our laboratory in the study of the compatible interaction between Phytophthora infestans and Solanum tuberosum (unpublished data). A preliminary targeted metabolic reconstruction, comprising ∼350 reactions from the central metabolism, alpha-linolenic acid metabolism, hormone biosynthesis and photosynthesis was interrogated by FBA, using microarray data from an infected potato, to introduce perturbations in the flux of the network.
Some of our preliminary results are promising and do validate biological evidence. Among these results, it is important to highlight the one that shows an important flux decrease in the reaction that synthesizes l-glutamate from 2-oxoglutarate. This results are in agreement with previous experimental results [39], where a significant repression of cysteine, aspartic and 2–oxoglutarate-dependant dioxygenase, in potato plants infected with P. infestans was identified. Besides its known role as a precursor for aminoacid synthesis, l-glutamate is also known as a key precursor in chloroplasts synthesis [99]. The flux decrease through this reaction shown by our model, could explain the characteristic decrease of photosynthetic activity in infected potatoes. On the other hand, some results have not shown a complete correlation with biological evidence. For instance in silico knockout of carbonic anhydrase, which have been previously reported as a key molecule in the compatible interaction between P. infestans and S. tuberosum [39], appears to have no effect on the overall flux of the network in our model, since the set of fluxes before carbonic anhydrase knockout remain the same. Nevertheless, a more detailed reconstruction and validation must be performed.
CONCLUSIONS
In this work, we have simultaneously reviewed some of the most important plant pathosystems to date (as well as wet lab techniques to its study) and introduced a methodology for metabolic reconstruction based on microarray data, as an alternative in those cases where genomic data are not available. Although in this work our efforts are focused on plant–pathogen interactions, this approach can be easily applied to any biological system in which external perturbations are expected to modify its metabolism. Importantly, this strategy also allows researchers to approach metabolic reconstructions without the immediate necessity of a sequenced genome.
Although some other approaches can be used to interrogate metabolic networks, a clear advantage of FBA is that it allows prediction of network phenotypes on biological systems where kinetics data are incomplete or missing.
Targeted, as well as GSMRs, enable a broad spectrum of in silico experiments at the system level, and can generate hypothesis to drive wet lab research. Since GEMR have been defined as a natural extension of genome annotation, targeted reconstructions can be seen as an extension of transcriptomic research, which increases our knowledge of network phenotype of the biological system under a particular environmental stress. Thus, the application of this methodology to host–pathogen models, as the ones here reviewed, is desirable and convenient.
Targeted metabolic reconstruction has been successfully approached in our laboratory to the study of the interaction between Phytophthora infestans and Solanum tuberosum, and although to date some preliminary results are not completely in accordance with biological evidence, some others have shown those disagreements are mostly due to manual process of curation during reconstruction and not to the methodology itself.
Targeted metabolic reconstructions allow researchers to approach metabolic reconstructions without the immediate necessity of a sequenced genome.
Targeted metabolic reconstructions can be seen as a natural extension of transcriptomic research, increasing our knowledge of a given expression profile.
In the case of the compatible interaction between Phytophthora infestans and Solanum tuberosum, this methodology has demonstrated to be an useful approach which helps to drive wet lab research. Furthermore, this methodology can be applied to any biological system where transcriptomics data are available.
