Sucrose is the primary carbon resource for energy supply and carbon skeletons for SNF and is supplied to legume nodules by transportation from the shoots (Gordon et al., 1998). However, sucrose is not used directly as a substrate by isolated nitrogen-fixing bacteroids to support nitrogenase activity, as demonstrated in the cases of soybean (Stovall and Cole, 1978), garden pea (Glenn and Dilworth, 1981), lupin (Tomaszewska et al., 1991), and alfalfa (Miller et al., 1988). In nodules, sucrose can be processed by one of two enzymes, sucrose synthase (SS; EC 2.4.1.13) and alkaline invertase (AI; EC 3.2.1.26), as shown in Figure 1. SS performs a reversible UDP-dependent cleavage of sucrose into UDP-glucose (UDP-Glc) and fructose (Akazawa and Okamoto, 1980). SS activity is reduced in the SNF-defective mutant, rug4, of P. sativum, indicating that SS plays an important role in nodule functions (Gordon et al., 1998, 1999). Moreover, overexpressing the antisense of MtSucS1 (encoding enzyme SS) in M. truncatula impaired the plant growth and nodulation in the MtSucS1-reduced nodules, and the contents of amino acids and their derivatives were also found to be reduced, which suggest SS is a crucial player in the establishment and maintenance of an efficient SNF process (Baier et al., 2007). Six SS isoforms have been identified, and the mutants of SS isoform sus1-1 and sus3-1 showed 38% and 67% reduction of SS activity, respectively, in nodules formed by inoculating L. japonicus with M. loti strain Tono, while the double mutant plants of sus1-1/sus3-1 had significantly impaired growth and their nodule SS activities were reduced by 94% compared to the wild type under the same growth conditions. This further shows that SS is essential in nodulation maintenance (Horst et al., 2007). In addition, when exposed to drought stress, nitrogen fixation in nodules declined, and the activity of SS was severely inhibited in various species (M. sativa, P. sativum, and P. vulgaris), and led to the further limitation of carbon availability in bacteroids (Ramos et al., 1999; Gálvez et al., 2005; Naya et al., 2007).

The other sucrose-catabolizing enzyme, AI, is a hydrolase, first identified in soybean nodules, that irreversibly cleaves sucrose into fructose and glucose (Copeland and Morell, 1985). The isolation and purification of AI was also carried out from chickpea nodules (Asthir and Singh, 1997). In L. japonicus, the expression level of gene LjInv1 encoding enzyme AI was twofold higher in mature nodules than in uninfected roots. Enzyme activity assays showed that LjInv1 contributed to the production of hexoses and other biosynthetic processes in developing nodules (Flemetakis et al., 2006). In G. max and M. truncatula root nodules, an alkaline/neutral Inv gene was found to have enhanced transcription levels in developing root nodules inferring an increased need for sucrose degradation (Flemetakis et al., 2006; Tesfaye et al., 2006). However, there is not much direct evidence to prove that AI is essential for the regulation of nodule metabolism.

The resulting UDP-Glc and free hexoses (glucose and fructose) are phosphorylated by hexokinase (EC 2.7.1.1) and enter glycolysis or oxidative pentose phosphate pathways. In nodules, the host plant provides carbon sources for bacteroid activities in the form of dicarboxylates, particularly malate and succinate (Day, 1991). Phosphoenolpyruvate carboxylase (PEPC; EC 4.1.1.31) and malate dehydrogenase (MDH; EC 1.1.1.82) convert the carbon flux from glycolysis to form malate (Figure 1). PEPC catalyzes the conversion of PEP into oxaloacetate (OAA). The active enzymes are less sensitive to malate, which act as a negative feedback control (Nimmo, 2000). The nodule PEPC was found to be activated with phosphorylation and its sensitivity was inhibited by L-malate in vitro and in vivo (Schuller and Werner, 1993; Zhang et al., 1995). In L. japonicus, abundant transcript of LjPEPC1 was found in the vascular bundles of nodules and bacteroid-infected cells, while LjPEPC2 was expressed in roots and shoots at low levels and postulated as the housekeeping isoform. Moreover, regulatory phosphorylation of PEPC is thought to be mainly controlled by the PEPC kinase (PEPC-PK) (Vidal and Chollet, 1997), and the expression of LjPEPC-PK shares a similar pattern with LjPEPC1 in mature nodules (Nakagawa et al., 2003). The OAA is then converted to malate by MDH in a reversible reaction, and this can then be uptaken by bacteroids directly (Figure 1). In alfalfa, five species of cDNAs were found to encode MDH, but only one of them had a nodule-enhanced expression pattern (Miller et al., 1998). One nodule-enhanced MDH (neMDH) was found in garden pea (Fedorova M. et al., 1999). Two MDH isoforms were found in the L. japonicus nodule by transcriptome profiling, one of which is a cytosolic MDH (cMDH) and the other a neMDH (Takanashi et al., 2012). The activity of cMDH in L. angustifolius nodules was shown to increase under P-deficient conditions, supporting the involvement of cMDH in nodules (Le Roux et al., 2014).

Besides the enzymes described above, some other carbon metabolic enzymes also play a role in legume nodules. MsSPSA, encoding sucrose phosphate synthase (SPS; EC 2.3.1.14) in M. sativa L., is involved in the first step of sucrose synthesis, and the expression of MsSPSA showed a nodule-enhanced pattern. SPS showed higher activities in the nodules formed by wild-type S. meliloti, compared to nodules inoculated with a N-deficient Fix^- strain (a Tn5 insertional mutation in nifH; Hirsch et al., 1983), suggesting an important role in maintaining a stable carbohydrate/energy supply for the actively nitrogen-fixing symbiont (Aleman et al., 2010).

Multiple environmental stresses result in the inhibition of carbon metabolism in host legumes, resulting in the suppression of SNF efficiency in nodules. Under salt stress, the concentration of sucrose was decreased, and major carbon metabolic enzymes such as SS and AI showed lower activity in both M. truncatula nodules (indeterminate nodules) and L. japonicus nodules (determinate nodules) (López et al., 2008). When garden pea plants were cultured under drought stress, the SS activity was down-regulated, indicating the normal carbon metabolic pathway was inhibited. The NADP-dependent isocitrate dehydrogenase (ICDH; EC1.1.1.42) was reported to balance C/N metabolic fluxes in leaves (Gallardo et al., 1995; Gálvez and Gadal, 1995). The enhancement of ICDH activity in nodules was probably the result of compensating for the C and N imbalance or supplying additional NADPH to neutralize the effects of the increased production of reactive oxygen species (ROS) as a consequence of drought stress (Gálvez et al., 2005).

Ammonia exported from the bacteroids diffuses into the cytosol of the infected host cells to be rapidly assimilated. Amino acids and/or ureides are then synthesized and exported from the legume nodules to the shoots (Figure 1). The determinate-nodule legumes primarily transport allantoin and allantoate (ureides) as fixed-N compounds, while the indeterminate nodules assimilate amides in the form of asparagine (Asn) and glutamine (Gln) (Sprent, 2009). The first step of ammonium assimilation is apparently the same in amide- and ureide-exporting legume nodules (Todd et al., 2006). In both cases, nitrogen fixed in the bacteroids is exported into the host cell cytosol, where ammonia is converted to Gln and glutamate (Glu) by the enzymes Gln synthetase (GS; EC 6.3.1.2) and Glu synthase (NADH-GOGAT; EC 1.4.1.14). In indeterminate nodules, Glu and Gln are further converted to aspartate (Asp) and Asn by Asp aminotransferase (AAT EC 2.6.1.1) and Asn synthetase (AS; EC 6.3.5.4). While for determinate nodules, Gln further entry purine synthesis pathway and convert to ureides as final product exported in to host plant (Figure 1).

Glutamine synthetase, as the enzyme for the first step of ammonia assimilation in the infected root cells, is a key component of nitrogen metabolism in nodules. It catalyzes the ATP-dependent amination of Glu to Gln. GS possesses two major isoforms largely separated by their subcellular localization, plastidic GS2 and cytosolic GS1 (Temple et al., 1998; Lea and Miflin, 2003). In M. truncatula nodules, three GS genes have been identified: cytosolic MtGS1a and MtGS1b, and plastid MtGS2, where MtGS1a accounts for 90% of the GS activity in nodules (Carvalho et al., 1997, 2000). Impairment of GS activity using phosphinothricin resulted in the inhibition of nodule growth, and the promotion of nodule senescence. The reduced activity of GS also suppressed Asn biosynthesis, as demonstrated by the lower expression of AS transcripts and lower Asn content (Seabra et al., 2012). NADPH-GOGAT is another enzyme involved in the assimilation of NH4+, by transferring the amide group from Gln to α-ketoglutarate (Figure 1). In transgenic M. sativa expressing antisense NADPH-GOGAT, GOGAT activities were much reduced in the nodules. The transgenic M. sativa inoculated with S. meliloti exhibited moderate chlorosis, lower nodule weight, and reduction of N₂ fixation efficiency compared to the control (Schoenbeck et al., 2000).

In addition to primary nitrogen assimilation by converting ammonia into amino acids, ureide biosynthesis is also a common pathway among tropical legumes such as soybean, common bean, and cowpea, which generally form determinate nodules (Tajima, 2004). The allantoin and allantoate are the final nitrogen forms exported from soybean nodules to the shoots (McClure and Israel, 1979). Before ureide synthesis, ammonia enters a purine synthesis pathway, which is catalyzed by xanthine oxidase and xanthine dehydrogenase into urate, whose activities have been detected in both infected cells and uninfected cells in cowpea nodules. While uricase, which catalyzes the irreversible conversion of urate to allantoin, was detected in uninfected cells but not infected cells (Atkins et al., 1997). Moreover, the genes GmALN3 and GmALN4, encoding the enzyme allantoinase (EC 3.5.2.5) that converts allantoin to allantoate, were expressed in soybean nodules (Duran and Todd, 2012). Gln may also be enzymatically transformed into xanthine through the de novo purine synthesis pathway (Werner and Witte, 2011; Figure 1).

Phosphate is the second most limiting macronutrient element required for crop growth after nitrogen, and there is a particularly high demand for phosphates in the nodules of N₂-fixing legumes (Gunawardena et al., 1992; Vance et al., 2000). Phosphate, as an essential macronutrient, is incorporated into organic compounds such as nucleic acids (DNAs and RNAs), enzymes, ATP, sugar phosphates, phospholipids, and so on. Such organic phosphates are involved in many plant biochemical processes such as nutrient transport and photosynthesis. As a component of chromosomes, phosphates play an important role in cell division and organogenesis, and help to transfer the genetic information from one generation to the next (Ahemad and Oves, 2011). The soluble inorganic phosphate (Pi) is the most abundant metabolite in Bradyrhizobium japonicum cells. The Pi concentration in free-living B. japonicum cells is 8.1 μmol g^-1 FW and 29.2 μmol g^-1 FW in bacteroids. Massive amounts of Pi are transported into bacteroids through phosphate transporters and are highly involved in the metabolism of bacteroids. Indeed, SNF expends a great quantity of energy involving phosphorylated intermediates (Vauclare et al., 2013).

Most rhizobia are obligate aerobes. The C₄-dicarboxylic acids supplied by the host plant must be metabolized through the tricarboxylic acid (TCA) cycle in the bacteroid. Malate, the primary carbon source of bacteroids, is converted to pyruvate and CO₂ via the NAD⁺-dependent malic enzyme (DME). Pyruvate is subsequently decarboxylated by pyruvate dehydrogenase to form acetyl-coenzyme A (acetyl-CoA) and enters the TCA cycle. DME is required to provide pyruvate to the TCA cycle in S. meliloti and Azorhizobium caulinodans (Zhang Y. et al., 2012). S. meliloti possesses two distinct malic enzymes, an NAD(P)⁺-dependent enzyme (DME) (EC 1.1.1.39) and a strictly NADP⁺-dependent enzyme (TME) (EC 1.1.1.40) (Driscoll and Finan, 1993, 1996; Voegele et al., 1999). In addition, acetyl-CoA can be produced alternatively by phosphoenolpyruvate carboxykinase (PCK) which catalyzes the decarboxylation of OAA to PEP. Symbiotic phenotypes of pck mutants vary depending on the host plant. pckA mutants of R. leguminosarum MNF3085 fix nitrogen at rates comparable to wild type, while pckA mutants of S. meliloti show a reduced N₂ fixation capacity (McKay et al., 1985; Finan et al., 1991). In R. leguminosarum and S. fredii strain NGR234, DME or a pathway involving PCK and pyruvate kinase (PYK) can synthesize the precursors required for SNF (Zhang Y. et al., 2012). Enzymes participating in the TCA cycle have been identified in B. japonicum strain USDA110 (Sarma and Emerich, 2006), R. leguminosarum (McKay et al., 1985), S. meliloti (Djordjevic, 2004), and R. tropici (Romanov et al., 1994). However, TCA cycle involvement probably varies among rhizobia. Evidence shows that S. meliloti, R. tropici, and R. leguminosarum utilize the full oxidative TCA cycle to provide ATP, precursors of amino acid synthesis, as well as reducing equivalents for N₂ fixation. Enzymes of the TCA cycle appear to be essential for nitrogen fixation in S. meliloti since mutations in succinate dehydrogenase (sdh), malate dehydrogenase (mdh), isocitrase dehydrogenase (icd), 2-oxoglutarate dehydrogenase, aconitase (acnA), and citrate synthase (gltA) abolish N₂ fixation, despite that nodules were formed (Duncan and Fraenkel, 1979; McDermott and Kahn, 1992; Mortimer et al., 1999; Dymov et al., 2004; Koziol et al., 2009). In contrast, B. japonicum shows a higher metabolic plasticity. Mutations in fumarase (fumC), ICDH (idhA), alpha-ketoglutarate dehydrogenase (agdA), and acotinase (acnA) in B. japonicum USDA110 still show phenotypes capable of fixing nitrogen (Acuña et al., 1991; Thöny-Meyer and Künzler, 1996; Green and Emerich, 1997; Shah and Emerich, 2006). Transcriptome analyses of R. etli bacteroids have suggested that the TCA cycle is inactive in these bacteroids (Vercruysse et al., 2011), but additional experiments are required to explain how the bacteroids obtain the ATP necessary for nitrogen fixation in these cases. A comparative metabolic profiling study between free-living R. leguminosarum and pea bacteroids showed that the TCA cycle is not the only path to oxidize dicarboxylic acids derived from the host plant in the bacteroids (Terpolilli et al., 2016). Metabolic profiling and flux analysis revealed that pea bacteroids divert acetyl-CoA into TCA and the production of lipid or polyhydroxybutyrate (PHB). These findings suggest new pathways for electron allocation in nitrogen-fixing bacteroids where lipogenesis may be a requirement in legume nodules.

The legume–rhizobium symbiosis determines the accumulation of specific metabolites inside the microsymbiont. Excess carbon and reducing power provided by the host plant can be stored as polymers, glycogen, or lipids in bacteroids (Figure 1). PHB is accumulated in large cytoplasmic granules in the bacteroids that form determinate nodules (such as in common bean and soybean), but not in the indeterminate nodules of alfalfa, garden pea, and chickpea (Table 1; Tombolini and Nuti, 1989; Kim et al., 1996).

The PHB biosynthesis pathway consists of three steps. β-Ketothiolase (PhaA) catalyzes the first step in this pathway with the formation of acetoacetyl-CoA from acetyl-CoA. Acetoacetyl-CoA is then reduced to D-β-hydroxybutyryl-CoA by an NADH-dependent acetoacetyl-CoA (PhaB), and then PHB synthase (PhaC) catalyzes the final formation of PHB (Lodwig and Poole, 2003). In S. meliloti and R. etli, phaA and phaB form an operon, phaAB.

Fix⁺ symbiotic phenotypes have been observed in S. meliloti when mutations were introduced to disable the synthesis of PHB or alter the ability to utilize PHB cycle intermediates to support growth (Povolo et al., 1994; Aneja and Charles, 1999; Cai et al., 2000). Effective nodules were also observed when R. leguminosarum mutant strains defective in the phaC gene were used to inoculate bean (determinate nodules) or pea (indeterminate nodules) (Lodwig et al., 2005). However, a reduced nitrogen fixation capability was observed when PHB synthesis was abolished in A. caulinodans. In A. caulinodans, impaired PHB synthetase activity results in the loss of nitrogen fixation capacity both ex planta and in symbiosis with the tropical legume Sesbania rostrata (Mandon et al., 1998). These evidences suggest that PHB may play certain roles in nitrogen fixation of some legume–rhizobium systems.

Polyhydroxybutyrate synthesis is regulated in R. etli and S. meliloti by the PHA (polyhydroxyalkanoate) regulator (PhaR), previously called AniA (Povolo and Casella, 2000; Encarnación et al., 2002). PhaR homologs in other bacterial species bind to their own promoters and to the promoter of phaP (Maehara et al., 2001). PhaP, a phasin, binds to the surface of PHB granules and can control the size of the PHB granules (Kuchta et al., 2007; Mezzina et al., 2014). In Bradyrhizobium diazoefficiens, the regulation of PHB synthesis is mediated through the interaction between PhaR and FixK₂, regulators associated with the synthesis of PHB and microoxic metabolism (Quelas et al., 2016).

Glycogen is co-produced with PHB in free-living rhizobia such as R. leguminosarum and S. meliloti under nutrient-limiting conditions (Povolo et al., 1994; Lodwig et al., 2005). The role of glycogen in SNF has been studied in some rhizobia. A glycogen synthase mutant (glgA) of R. tropici showed an enhanced symbiotic performance as measured by an increase in plant dry weight and nodule number, but the mechanisms behind this phenotype are uncertain (Marroquí et al., 2001). In garden pea bacteroids, a glgA mutant of R. leguminosarum did not alter nitrogen fixation rates, as indicated by the acetylene reduction assay for nitrogenase activity. Plants inoculated with a phaC/glgA double mutant resulted in similar phenotypes to wild-type-inoculated plants (Lodwig et al., 2005). In contrast, the glgA mutant of S. meliloti resulted in lower levels of N₂ fixation in both M. truncatula and M. sativa (Wang et al., 2007). Thus, the SNF capability of S. meliloti is affected by both PHB and glycogen availability.

In Bradyrhizobium sp. ORS278, studies have shown that mutations in genes involved in the Calvin–Benson–Bassham (CBB) cycle result in deficiency in nitrogen fixation. The CBB cycle consumes 3 mol of ATP and 2 mol of reducing equivalents per mol of carbon dioxide. The symbiotic phenotypes suggest that Bradyrhizobium sp. ORS278 uses the Calvin cycle as a reductant store to regulate carbon flux. Specifically, phosphoglycerate kinase (CbbK) and ribulose 1,5-biphosphate carboxylase (RuBisCO) large chain (CbbL1) have been identified as being involved in the symbiotic process (Bonaldi et al., 2010; Gourion et al., 2011). Thus, the relevance of different carbon storage compounds in the symbiotic process varies depending on the rhizobial strain.

Constraint-based modeling enables the determination of the metabolic capabilities of an organism (Feist et al., 2009; Oberhardt et al., 2009). These metabolic models have been used successfully to bridge the gap between current knowledge and metabolic phenotypes (Schellenberger et al., 2011). Metabolic reconstructions are based on physiological and biochemical information on primary and specialized metabolic pathways, genome information, and available -omics data. These models have allowed for the prediction of the metabolic behavior of microorganisms under different nutrient availabilities and simulated gene deletions or over-expressions (Contador et al., 2015; Razmilic et al., 2018). In an attempt to get new insights into the genetic interactions that orchestrate the complex metabolic interactions in nitrogen-fixing bacteria, metabolic models have been used to study the rhizobial metabolism. Genome-scale models have been combined with experimental data to identify essential genes and to simulate the metabolic behaviors of rhizobia under nitrogen fixation conditions in legume nodules. High-throughput experimental data aimed to identify key genes, proteins and metabolites have been used in these studies to validate the model predictions.

At present, few manually curated rhizobial reconstructions are available despite their important roles in sustainable agriculture. These reconstructions cover different metabolic tasks from only the symbiotic process that takes place in the rhizobium to the complete set of metabolic reactions of the bacteria cell. iOR363, a reconstruction for R. etli CFN42, was the first model available for any rhizobium (Resendis-Antonio et al., 2007). This model describes the SNF process inside the determinate nodules of P. vulgaris and includes the main reactions associated with the nitrogen fixation process and the metabolic pathways, such as PHB synthesis, that allow simulating the accumulation of this polymer during the symbiotic process (Resendis-Antonio et al., 2007). Also, a symbiotic reaction was defined to represent this specific legume–rhizobia symbiosis. The symbiotic reaction describes the exchange of nutrients between the host plant and the bacteroid to establish symbiotic relationships and to obtain or produce chemical compounds required by the process of nitrogen fixation. A symbiotic auxotrophy hypothesis and nitrogenase requirements were included in its definition. This model has been updated to include new data sets available of this system (Resendis-Antonio et al., 2011, 2012).

On the other hand, the metabolic reconstructions for S. meliloti have different scopes. iHZ565 was designed to represent the SNF of S. meliloti 1021 with host plants such as alfalfa (Zhao et al., 2012). The reconstruction process gave insight about the missing context-specific information and genome annotation errors. Additionally, a new symbiotic reaction was proposed to capture specific SNF mechanisms of S. meliloti in indeterminate nodules. The first representation of the whole metabolism of a rhizobium cell was built for S. meliloti (iGD1575) to enable the characterization of the metabolic capabilities of rhizobia in bulk soil, rhizosphere, and nodule (di Cenzo et al., 2016). iGD1575 encompasses 94% of the genes present in iHZ565. Recently, iGD1575 and gene essentiality experiments were used to build a model of the core metabolism of S. meliloti (di Cenzo et al., 2018). Pfau et al. (2018) have described the symbiotic relationship of M. truncatula with S. meliloti through constraint-based modeling (Pfau et al., 2018). Metabolic models were used to study the metabolic exchange between the host plant and the rhizobial symbiont. An in silico representation of the metabolic network of B. diazoefficiens USDA110 (iYY1101) has also been constructed (Yang et al., 2017). This reconstruction was used to build context-specific models to describe the metabolic differences observed between the free-living state of B. diazoefficiens USDA110 and symbiotic bacteroid. These models will provide a starting point for building metabolic reconstructions of closely related organisms such as other rhizobial strains and modeling frameworks to study these symbiotic interactions (Baumler et al., 2011; Monk et al., 2013; Ong et al., 2014). Table 3 summarizes the available metabolic models of rhizobia.