Rice (Oryza sativa L.) is a staple food consumed by more than half of the world’s population. Tremendous progress has been made in rice production during the past half century with a doubling of yield since the Green revolution in 1960s (Elert, 2014). This was achieved by the advancement and expansion of rice breeding system to develop high-yielding rice varieties, an increase in fertilizer use and a significant expansion of land area for rice cultivation. However, we are still facing the challenge of food shortage due to the sharp increase in global population as well as the negative effects of climate change on rice production (Wassmann et al., 2009; Dayton, 2014). To tackle this problem, a better understanding of rice physiology and metabolism is essential for improving rice production, for example by increasing photosynthetic efficiency and tolerance to abiotic and biotic stress conditions. Notably, rice has the unique ability to germinate and grow up to coleoptile under anoxic/prolonged soil flooding conditions. It is believed that this distinctive behavior could be due to the metabolic adaptation by active ethanolic fermentation although the underlying mechanism is still not completely well-understood. In addition, some varieties of rice have the adaptive mechanism to tolerate complete submergence with limited gas diffusion, low light and mechanical damage caused by flash flooding as well as deep water flooding (Jackson and Ram, 2003; Bailey-Serres et al., 2010). During such stress conditions, some rice cultivars have the ability to facilitate internal aeration by developing aerenchyma, leaf gas films and barrier to radial oxygen loss from the root depending on the type and depth of flooding (Nishiuchi et al., 2012). However, the distinctive mechanism that enables rice to tolerate such adverse conditions is yet to be revealed. To better understand these mechanisms, multiple approaches have been applied by rice scientists around the world starting from the elucidation of the whole rice genome sequence (International Rice Genome Sequencing Project, 2005) to high throughput -omics profiling together with advanced bioinformatics tools.

Concurrently, the recent advances in metabolic modeling approaches and the availability of genome information for numerous organisms have facilitated the reconstruction and analysis of genome-scale metabolic networks (GSMNs) in all three domains of life (Kim et al., 2012). Based on the genome of an organism of interest, a draft metabolic network can be developed by compiling the metabolic reactions catalyzed by enzymes encoded in the genome. Information on metabolites and reaction stoichiometry are typically obtained from multiple biochemical databases such as MetaCyc (Caspi et al., 2012) and Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa et al., 2012). The draft network is then further gap-filled and manually refined to improve network connectivity by checking elemental balance and reaction directionality; gene-protein-reaction (GPR) associations are also included and exchange reactions are added for allowing nutrients uptake and secretion from the system. The reconstruction process of rice GSMN is summarized in Figure 1. Detailed procedures and challenges involved in the reconstruction process have been previously described and reviewed elsewhere (Thiele and Palsson, 2010).

Once the network is reconstructed, it can be readily converted into a mathematical model, which is represented in a matrix form where the rows and columns represent the metabolites and reactions, respectively. This “stoichiometric matrix” or simply “S matrix” is the centerpiece of constraint-based model reconstruction and analysis methodology where the stoichiometric coefficients constrain the flow of metabolites from various substrates to products in the metabolic network. In addition, measured extracellular uptake and secretion rates can be used as additional constraints. Together, these constraints confine a region within the multi-dimensional solution space of allowable reaction fluxes where the actual solution exists (Bordbar et al., 2014). Several techniques have been proposed to interrogate this solution space, among which constraints-based flux analysis, also known as flux-balance analysis (FBA), is the most commonly used approach (Orth et al., 2010). Basically, FBA optimizes a particular cellular objective function for determining the possible solution state within the metabolic network in terms of metabolic fluxes. Detailed information on metabolic modeling methodology is beyond the scope of this review and the comprehensive review could be found elsewhere (Lewis et al., 2012; Baghalian et al., 2014). Overall, the simplicity and extensibility of FBA and the availability of several conveniently accessible software tools to implement it (Lakshmanan et al., 2014a), genome-scale models have been extensively used for multiple applications including (i) prediction of metabolic phenotypes, (ii) investigation of metabolic network properties, (iii) contextualization of high throughput data, (iv) studying interspecies metabolic interactions, and (v) guidance of metabolic engineering (Oberhardt et al., 2009).

In this review, we outline the current status and future prospective of the analyses and applications of rice metabolic networks and models (Figure 2). First, the existing rice metabolic networks and models are surveyed with their analyses and applications. Next, the potential and challenges in translating the modeling approaches applied and knowledge gained from other plant models to rice are discussed. Lastly, several future applications of rice genome-scale metabolic models are proposed for the improvement of rice production.

To date, two rice GSMNs (Dharmawardhana et al., 2013; Liu et al., 2013) and three rice genome-scale metabolic models (GEMs) (Poolman et al., 2013; Seaver et al., 2014; Lakshmanan et al., 2015) have been reconstructed based on multiple data sources. Here, we make a distinction between GSMNs and GEMs, similar to the definitions used in Lewis et al. (2012), where a GSMN comprises of all known metabolic reactions in an organism while a GEM is computable derivative of a metabolic network which can be analyzed using structural metabolic modeling and constraint-based modeling techniques such as FBA. The statistics of rice GSMN/GEMs are summarized in Table 1.

The two rice GSMNs, RiceCyc (Dharmawardhana et al., 2013) and RiceGEM (Liu et al., 2013), are accessible online as databases with information on genes, proteins and reactions. RiceCyc is a rice-specific metabolic pathway database constructed based on MetaCyc and plant metabolic network (PMN) (Chae et al., 2014). The initial stage of RiceCyc construction involved the functional annotation of rice genes using a number of data sources including InterPro (Mitchell et al., 2015), Gene Ontology (Ashburner et al., 2000), MetaCyc, Enzyme Commission (EC) numbers, KEGG and Gramene databases (Monaco et al., 2014). Notably, it contains information about genes, proteins and reactions across 316 metabolic pathways. However, it should be noted that RiceCyc does not provide any information on subcellular localization of metabolic reactions. In contrast, reactions in RiceGEM are assigned to nine subcellular compartments based on multiple protein localization prediction software (Liu et al., 2013). It was constructed by integrating reaction and metabolite information from various databases including RiceCyc, KEGG, Uniport (The UniProt Consortium, 2013) and Brenda (Chang et al., 2015). RiceGEM is available as a part of the RiceNetDB database¹, which is a multi-level network database of rice created by integrating RiceGEM with gene regulatory networks and protein–protein interaction network.

The first computation-ready rice GEM was built from RiceCyc, with three subcellular compartments, cytosol, chloroplast, and mitochondrion (Poolman et al., 2013). Reactions in the cytosol were derived from RiceCyc, whereas reactions in the chloroplast and the mitochondrion and metabolite transport reactions were defined manually. The model was curated and checked for energy and redox conservation (no production of energy and redox equivalent from nothing) and stoichiometric consistency with respect to carbon, nitrogen, phosphorus, and sulfur (no production or consumption of C, N, P, and S from nothing). The next available rice GEM was constructed from the PlantSEED biochemistry database (Seaver et al., 2014), which integrates information from multiple sources including ModelSEED (Aziz et al., 2012), KEGG, PMN, MetaCyc-based databases and several published GEMs. PlantSEED supports the automated construction of computation-ready, gap-filled GEMs from plant genomes, currently with models of 10 plant species, including rice, available with gene-protein-reaction associations from the PlantSEED website². Each PlantSEED model contains a comprehensive biomass reaction with 79 components and species-specific metabolic and transport reactions assigned to nine subcellular compartments mostly based on AraCyc (Mueller et al., 2003) and the Plant Proteomics Database (PPDB) (Sun et al., 2009). We tested the rice model from PlantSEED and found that the model has no stoichiometric inconsistency, but energy conservation is violated presumably due to the lack of manual curation in reaction reversibility. To the best of our knowledge, the most recent rice GEM is iOS2164 (Lakshmanan et al., 2015) reconstructed based on multiple databases including RiceCyc, PlantCyc, KEGG, and TransportDB (Ren et al., 2007). A distinct feature of this model is the detailed definition of all possible electron transport reactions in the mitochondrion, the plastid and the thylakoid including the light-driven photophosphorylation reactions in a wavelength specific manner, which allowed for the modeling of photosynthetic metabolism under different light sources. The model was manually curated for elemental balance, reaction reversibility and directions, assignment of reactions into seven subcellular compartments, gap-filling and gene-protein-reaction mappings. Energy is not conserved when considering the reaction stoichiometries, reversibility and directions of this model, but additional constraints were applied at the stage of model simulations to prevent spontaneous ATP generation (Lakshmanan et al., 2015). Here, it should be noted that while all GEMs represent rice metabolism, still the number of genes and reactions accounted vary drastically. Such inconsistencies arise due to the incompleteness in the data used for reconstructing the rice metabolic model. Therefore, in order to test the functional use of the models, we computed the number and the proportion of allowed reactions which can carry non-zero fluxes at steady-state and identified that iOS2164 has the highest proportion when compared to other three rice GEMs (Table 1), suggesting that this latest model has the broadest coverage of rice metabolism and supports the simulation of the highest number of metabolic functions in rice cells.

Apart from these rice GSMN/GEMs, a central regulatory/metabolic model of rice was also developed (Lakshmanan et al., 2013a), which covered all the central metabolic pathways. This model also included 52 direct and indirect regulatory interactions involving 12 regulatory proteins for differentiating the photosynthetic and non-photosynthetic cells based on Boolean logic formalism, controlling the activity of 40 metabolic reactions. Further, this model also has two manually curated biomass reactions, representing the rice seed/coleoptile and leaf based on literature information.

As GSMN/GEM represent systematic descriptions of the genotype–phenotype relationships, several in silico analyses were performed with/without multiple “-omics” data for investigating rice metabolism under both normal and stressed conditions. Here, we outline the applications of rice GSMN/GEMs including the contextualization of -omics data, followed by the characterization of rice metabolic responses to stress conditions and the investigation of the effect of light on rice metabolism.

Besides studies on rice metabolic models, there have been numerous publications on large-scale metabolic modeling of Arabidopsis and other plants in the past few years. Thus, we review the studies on plant metabolic modeling which are relevant and applicable to rice metabolic models.

Influenced by the history of FBA and genome-scale metabolic modeling from bacterial systems, most plant flux-balance modeling studies simulated the metabolism of cell or tissue growth by setting a constraint on the synthesis of biomass components or an objective function of maximizing biomass synthesis. While growth as a metabolic function is applicable for plant cell or tissue cultures, developing seeds and young leaves, there are many plant tissues and organs, such as mature and senescing leaves, maturating fruits, etc., which have distinct metabolic behavior and phenotypes other than growth. In order to apply genome-scale metabolic modeling to more plant tissues, the definition of tissue-specific metabolic function(s) must be extended beyond cell or tissue growth. One example is the use of an Arabidopsis GEM to simulate the metabolism of mature (non-growing) C₃ and CAM leaves by defining the metabolic function of mature leaves with model constraints of sucrose and amino-acids export into the phloem (Cheung et al., 2014). Fruit maturation is another example where growth is not applicable, demonstrated by a recent FBA study on modeling the metabolic shifts of nine sequential stages of tomato fruit development using the objective function of flux minimization, constrained by measured concentrations of biomass components and accumulated metabolites in the pericarp at each stage (Colombié et al., 2015).

One of the major challenges in GEMs is the incorporation of regulatory aspects into the modeling framework. To do so, Boolean logic formalism was incorporated in one of the studies to control the reaction activity of rice central model, accounting the regulation of seed germination under aerobic and anaerobic conditions (Lakshmanan et al., 2013b). Here, it should be noted that although such an approach was shown to be successfully applied for modeling different tissue/conditions, it still has certain limitations: it just provides two possible states for target metabolic reactions, i.e., ON or OFF. However, in reality, biological systems exhibit a wide range of responses in transcriptional regulation, from binary to even continuous. To address such issues, alternative methods such as probabilistic regulation of metabolism (PROM; Chandrasekaran and Price, 2010) have been devised for the continuous integration of cellular regulatory network onto the genome-scale model using abundant transcriptomic data and regulatory interaction information, resulting in better combined regulatory-metabolic models. As an alternative, gene expression data can also be used as surrogate for cellular regulation to consider only the active metabolic pathways. Several methods such as Gene Inactivity Moderated by Metabolism and Expression (GIMME; Becker and Palsson, 2008), Integrative Metabolic Analysis Tool (iMAT; Shlomi et al., 2008), and E-Flux (Colijn et al., 2009) have been proposed to achieve this feat. While numerous other methods are still being developed in this regard for the appropriate integration of gene expression data onto GEM, such approach has already been successfully adopted by several plant modeling studies including rice (Mintz-Oron et al., 2012; Töpfer et al., 2013; Lakshmanan et al., 2015).

Another problem that stemmed from the mentality of simulating growth is the bias in manual curation of GEMs for reactions involved in synthesis of core biomass components (i.e., carbohydrates, proteins, nucleic acids, lipids). Most plant metabolic models were tested for biosynthesis of biomass components but relatively little effort was made in their catabolism, which is important for modeling senescing plant tissues. While carbohydrate and lipid degradation were included in most plant metabolic models, this is generally not the case for catabolism of other biomass components such as amino-acids, nucleic acids and chlorophyll. With respect to secondary metabolism, although the biosynthesis of a limited set of secondary metabolites was curated for a number of plant GEMs, the vast majority of complex plant secondary metabolites could not be produced from the models. Furthermore, the biosynthesis of most secondary metabolites was neglected in model simulations, primarily due to the lack of experimental data on the production of secondary metabolites. The exclusion of secondary metabolites from the model constraints or biomass equations could have an impact on the results from various analyses such as gene essentiality and energetic costs calculations.

Plant metabolic models were applied to simulate metabolism under multiple abiotic conditions. While a number of abiotic factors directly relate to metabolism and can be captured directly by constraints in flux-balanced models, such as anoxia and light intensity, some require indirect proxies to translate the abiotic condition into model constraints. For example, Rubisco carboxylation to oxygenation ratio was used as a proxy for stomatal closure to simulate drought stress (Lakshmanan et al., 2013a). In addition many conditions have no simple relation to the metabolic system, e.g., heat and cold stresses, hyperosmotic conditions and salt stress, therefore model simulations of these conditions require additional information such as transcriptome data (Töpfer et al., 2013) or extensive experimental measurements for model constraints (Williams et al., 2010). The continuous progress in tackling these challenges in plant metabolic modeling will enable us to extend the scope of future applications of GEMs of rice and other plants.

The primary aim of rice genome-scale metabolic modeling is to enhance our understanding of rice metabolism to guide its engineering and manipulation for improved quality and quantity of rice grain production. One active research area is the engineering of rice by incorporating C₄ photosynthesis, at both biochemical and anatomical levels, into rice (Hibberd et al., 2008; Zhu et al., 2010; von Caemmerer et al., 2012). In silico engineering of C₄ rice can be performed using rice GEMs to simulate the optimal metabolic distribution for efficient photosynthesis in C₄ rice. This will allow the identification of metabolic manipulations required for optimizing C₄ photosynthesis in rice, which include not only enzymes in the C₄ pathway but also other related processes such as the shuttling and balancing of ATP and reductant to meet altered metabolic demand and the localization of carbon storage to support dark metabolism. Besides C₄, there have been suggestions in engineering CAM photosynthesis into C₃ crop plants to improve photosynthetic and water use efficiency for sustaining and improving crop productivity in a warmer and drier world (Borland et al., 2014). The effect of engineering CAM photosynthesis into rice can be modeled by applying a diel modeling framework (Cheung et al., 2014) on rice GEMs, which could guide the engineering and optimization of CAM photosynthesis in rice.

In addition to the efforts in increasing the productivity of rice, work has also been done on improving the nutritional quality of rice grains (Bhullar and Gruissem, 2013). The most widely known example is the golden rice, which is engineered to produce β-carotene (pro-vitamin A) in the endosperm to combat vitamin A deficiency (Beyer et al., 2002; Paine et al., 2005). Other examples include engineering higher levels of folate, iron and essential amino-acids, etc. in rice grains (Bhullar and Gruissem, 2013). The biosynthesis of metabolic products with high nutritional values in rice can be investigated using rice GEMs, which may give hints to possible strategies for producing rice grains with increased nutritional value.

Up until now, plant flux-balance models have mostly been applied to investigate metabolism under different abiotic conditions, with relatively little work done on modeling biotic conditions. A well-known example of plant–microbe interaction is legume–rhizobia symbiosis. Using GEMs of a legume (Medicago truncatula) and a rhizobium (Sinorhizobium meliloti), the metabolic interactions underlying the symbiotic relationship between the two organisms were modeled, suggesting that oxygen availability limits symbiotic nitrogen fixation and influences the form of nitrogen supplied to the plant (Pfau, 2013). Regarding interactions between rice and microbes, rice pathogens can cause up to 50% yield loss, resulting in large economic losses (Miah et al., 2013). To understand the manipulation of rice metabolism by pathogens during pathogenesis, metabolic models of rice pathogens such as Xanthomonas oryzae pv. oryzae and Magnaporthe oryzae can be developed and integrated with rice GEMs to investigate rice–pathogen metabolic interactions, similar to a study on modeling human–pathogen metabolic interactions (Bordbar et al., 2010). Analysis of an integrated rice–pathogen model could potentially identify strategies to counteract the metabolic manipulation by the pathogens for engineering disease-resistant rice strains.

While GEM on its own is already a powerful tool for investigating the behavior of metabolic systems, a lot more can be done by making use of the gene-protein-reaction associations in the models, illustrated by a number of plant modeling studies ( Töpfer et al., 2013, 2014; Schwender et al., 2014; Simons et al., 2014; Lakshmanan et al., 2015). As different types of -omics data are becoming more readily available, GSMN/GEMs will prove to be important tools for analyzing and contextualizing multiple -omics data. Further work on integrating multiple -omics data with rice GSMN/GEMs will lead to better understanding of the regulation of rice metabolism, which is crucial for improving crop production.

In summary, the development and application of rice GSMN/GEMs in recent years have increased our understanding of the metabolic and transcriptional regulations and the metabolic responses of rice to multiple environmental conditions and stresses through the integration of -omics data and/or constraint-based modeling. As the field of plant metabolic modeling progresses, modeling approaches developed for Arabidopsis and other plant models will undoubtedly be applied to rice models to better understand the metabolic systems of this important crop plant. While there are several limitations and challenges in applying constraint-based techniques in modeling plant metabolic systems, progress are being made in overcoming these limitations and challenges, which will extend the scope of future applications of GEMs of rice and other plants. With the ever increasing availability of multiple -omics data and experimental measurements, we envision that rice GSMN/GEMs will play a key role in contextualizing and analyzing these complex data to give novel insights into rice metabolism, which will lead to improvements in the yield and quality of rice production.

ML, CC, and D-YL have contributed in the design and write up of the manuscript. ML, BM, and D-YL have edited, reviewed, and approved the final version of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.