Introduction

Obligate heterofermentative lactic acid bacteria (LAB) lacking the glycolytic enzyme, fructose 1,6 bisphosphate aldolase, dissimilate carbon solely through a unique route of glycolysis known as the phosphoketolase pathway (PKP)1. This pathway is fundamentally different from the canonical Embden-Meyerhof-Parnas (EMP) pathway wherein one mole of glucose-6-phosphate is broken down into two moles of glyerone-3-phosphate at the end of the preparatory phase. In PKP, on the other hand, one mole of glucose-6-phophate is converted into one mole of each acetyl-phosphate and glyerone-3-phosphate via phophoketolase, thus leading to mixed fermentation products such as acetate, lactate and ethanol. These properties along with their natural ability to degrade xylan from lignocellulose using endogenous β-xylosidases2 make obligate heterofermentative LAB (e.g. Leuconostoc mesenteroides and Lactobacillus fermentum) one of the dominant species in vegetable fermentations such as Korean kimchi and sauerkraut3,4. The heterofermentative by-products inhibit the growth of harmful microbes and simultaneously impart taste and flavour to the final food products. In addition, the obligate heterofermentative LAB exhibit several beneficial characteristics to mammalian hosts including antimicrobial activities5,6,7, immunomodulatory effects8,9 and antioxidant activities6,10. They are also known to improve the microfloral balance in gastrointestinal tract; for example, Lb. fermentum is shown to strengthen it possibly through the increased activity of α-glycosidases and esterases11. In a different study, it is reported that a L. mesenteroides strain produces the prebiotic isomaltooligosaccharides that in turn enhance the growth of Bifidobacterium and Lactobacillus species in poultry chicken12.

However, despite their probiotic abilities and generally regarded as safe (GRAS) status, therapeutic applications of obligate heterofermentative LAB remain largely unexplored. The fastidious growth requirements and low ATP yield associated with the utilization of PKP are considered as the key bottlenecks for their practical applications. Interestingly, the heterofermentative LAB have naturally evolved to meet the cellular energy requirements under specific conditions by using alternate routes comprising (i) malolactic fermentation (MLF)13, (ii) alternate electron acceptors, e.g., fructose that yields mannitol upon reduction14, and (iii) phosphotransferase system (PTS) mediated transport1,15. Nonetheless, the underlying molecular mechanisms and their contributions to ATP yield, pH, and redox states still remain elusive. Elucidating such metabolic routes related to the unique cellular phenotype of LAB is, therefore, integral to their successful probiotic application16. For example, metabolic determinants restoring cofactor balance can be identified to modulate the relative abundance of other probiotic bacteria in the gut for imparting beneficial health effects17,18. To this end, a systems biology approach based on in silico metabolic modeling and omics data integration could greatly aid to decipher the lesser investigated metabolic landscapes of these LAB.

Constraint-based flux analysis, also known as flux balance analysis (FBA), is one of the well-established approaches for investigating cellular metabolism at systems-level under various environmental/genetic perturbations19. The availability of software applications to conveniently implement FBA and related in silico methods20 has enabled the development of genome-scale models (GEMs) for more than 100 species across all three domains of life, thereby facilitating analysis of their intracellular metabolism21. However, with respect to LAB, only a handful of GEMs are available, including Lactococcus lactis 22,23,24, Lactobacillus plantarum 25, Streptococcus thermophilus 26 and Lactobacillus casei 27,28 although more than 100 LAB genomes have been sequenced29. Note that these species with available GEMs belong to either homolactic or facultatively heterolactic fermenting groups. To enable the systems analysis of the other important group, i.e., obligate heterofermentative, herein, we present new genome-scale metabolic reconstruction of Leuconostoc mesenteroides ssp. mesenteroides ATCC 829330. The reconstructed model was subsequently validated both qualitatively and quantitatively using the fermentative substrate phenotyping and batch culture data, respectively. The model was also used together with transcriptome data for generating new hypotheses related to energy and redox metabolisms, and their effects on cellular phenotype.

Results

Reconstruction of L. mesenteroides genome-scale metabolic network

The genome-scale metabolic network of L. mesenteroides was newly reconstructed following established procedures31 (see Methods). Initially, the draft metabolic network was built based on information obtained from the genome annotation32 and biochemical databases such as MetaCyc33, KEGG34 and TransportDB35. Subsequently, the metabolite dead-ends were identified using GapFind algorithm36; new metabolic reactions supported by experimental and literature evidences were added to fill the network gaps. Reaction directionality was assessed using the available thermodynamic data calculated at physiologically relevant pH and intracellular metabolite concentrations as described earlier37,38,39 (see Methods). We then used this draft model to explore the minimal nutrient requirements of L. mesenteroides by mimicking the single nutrient-omission experiments conducted using a chemically defined medium (CDM)40. The comparison of in silico predictions with the experiments revealed some discrepancies which were then resolved through further manual curation of the network model. For example, the amino acid auxotrophy experiments confirmed asparagine as one of the non-essential amino acids. However, initial simulation of the draft model gave rise to no growth in the absence of asparagine, as the common asparagine biosynthetic enzymes, i.e., EC 6.3.5.4 and EC 6.3.1.1, were absent. From the genome-annotation evidence, we found the presence of an unique enzyme, aspartyl-tRNA-asparagine amidotransferase (EC 6.3.5.6) in L. mesenteroides, which indicated the possible existence of a non-canonical route for the biosynthesis of asparagine using aspartyl-tRNA and glutamine as substrates41. Similarly, vitamin-auxotrophy data was also used to further enhance the model quality. Folate, an indispensable precursor in the biosynthesis of DNA, was found to be only stimulatory according to the experiments. On contrary, the in silico simulations require its external supply for the cell growth, indicating possible missing annotations in the folate biosynthetic pathway. Therefore, a few non-gene associated reactions were added into the reconstruction in order to complete their biosynthetic pathways. BLASTp homology searches were then performed for the corresponding enzymes which were newly added, thereby substantiating their inclusion. Overall, the curated model was able to predict the auxotrophy of 7 out of 8 vitamins and 19 out of 20 amino acids accurately (Supplemental data 1). A central metabolic pathway map showing the biosynthetic pathways of various amino acids in L. mesenteroides is depicted in Fig. 1A. The list of reactions added during the gap-filling process is provided in Supplemental data 1.

Figure 1
figure 1

L. mesenteroides metabolic network (iLME620) reconstruction and comparison of its characteristics with other LAB. (A) Central metabolism of L. mesenteroides along with amino acid biosynthetic pathways present in iLME620. Reactions represented by ‘red’ lines are added to the network during the manual curation process. Reactions in ‘grey’ represent those absent in L. mesenteroides, but present in the other two LAB; reactions in ‘blue’ represent those present in L. mesenteroides and Lb. plantarum, but absent in Lc. lactis; reactions in ‘green’ represent those absent in L. mesenteroides and Lb. plantarum, but present in Lc. lactis; reaction with dotted line is absent in all three LAB (B) Venn diagram showing comparison of EC numbers of iLME620 with the GEMs of Lactococcus lactis subsp. cremoris MG1363 and Lactobacillus plantarum WCFS1. (C) Comparison of metabolic network characteristics of the LAB.

Full size image

The resultant genome-scale metabolic network of L. mesenteroides, iLME620, includes 620 genes, 754 metabolites, 762 metabolic and transport reactions, and one reaction representing the biomass assembly (see Methods). We measured the major elemental compositions of L. mesenteroides biomass and compared them with the molar elemental constituents of the biomass equation which is derived from the assembly of the measured macromolecular compositions (Supplemental data 1). The values in the equation are within ± 20% range of the experimental measurements (C H1.765 N0.18 O0.53 compared to that of C H2.05 N0.197 O0.455). The complete list of reactions, metabolites and associated genes accounted in iLME620, and the calculations involved in the biomass equation assembly are provided in Supplemental data 1. The model is also available as Systems Biology Markup Language (SBML) file (level 3, version 1, http://sbml.org/) in Supplemental data 2.

We compared iLME620 with the GEMs of other LAB, Lb. plantarum 25 and Lc. lactis 24, using the EC numbers in order to gain insights into their unique metabolic features (Fig. 1A–C). Expectedly, significant differences were observed in the preparatory phase of glycolysis, the non-oxidative branch of pentose phosphate pathway and the initial steps of the tricarboxylic acid cycle (Fig. 1A). We also found that the unique EC numbers of Lb. plantarum and Lc. lactis mostly fall under central metabolism and biomass biosynthetic pathways. Interestingly, most of the unique EC numbers in L. mesenteroides are unclassified and belong to the family of oxidoreductases hinting at a possible dominant redox metabolism (Supplemental Fig. 1). We further evaluated the ability of three LAB to grow under various environmental perturbations, e.g., amino acid auxotrophy and sugar utilization (see Methods and Supplemental Fig. 2). L. mesenteroides is auxotrophic to relatively less number of amino acids and can additionally utilize xylose as carbon source, which is a common characteristic of microbes associated with plant-based ecological niches42,43. We also examined the distribution of essential genes in three LAB, resulting in similar overall trends across several pathways with exceptions in certain cofactor biosynthetic pathways (see Methods and Supplemental Fig. 3).

A detailed comparison of glycolytic pathways shows that while both Lb. plantarum and Lc. lactis possess high ATP-yielding EMP pathway (2 moles of ATP per mole of glucose), the low ATP-yielding PKP (one mole of ATP per mole of glucose) is the only route of carbohydrate catabolism in L. mesenteroides (Fig. 2). Notably, it has been reported that the significant loss of ATP yields is often, at least partially, compensated by their favourable thermodynamics, i.e., larger net negative standard Gibbs free energy change values (Δr G′°)44,45. Here, we demonstrated a similar thermodynamic advantage of PKP over the EMP pathway by computing Δr G′° of each reaction in either pathways as well as the net Δr G′° for the pathways at physiologically relevant anaerobic condition (see Methods). The results showed that PKP has a larger fraction of reactions with negative Δr G′° values (Fig. 2 ) and a lower net Δr G′° (−186.8 kJ/mol in PKP vs −178.0 kJ/mol in EMP pathway), highlighting thermodynamic advantage of PKP over EMP pathway.

Figure 2
figure 2

Comparison of Gibbs free energy changes of reactions under anaerobic condition in PKP and EMP pathway. The transformed Gibbs free energy changes of reactions (highlighted in red color) in the pathways were calculated using eQuilibrator73, for pH 7.2, ionic strength (0.1 M) and intracellular metabolite concentrations (1μM-10mM). For expanded names of all metabolites (black) and reactions (blue) (except, PFK: phosphofructokinase and FBA: Fructose bisphosphate aldolase, which are absent in L. mesenteroides), see Supplemental data 1.

Full size image

Qualitative and quantitative validations of in silico model predictions

In order to validate the qualitative model predictions, we simulated fermentable substrate phenotyping for various carbon sources and compared the results with experimental literature46,47. The model predictions showed good agreement for 25 of the total 29 substrates tested. Here, the model was structured in a way that the glycoside substrates among the tested carbon sources are hydrolyzed by their respective extracellular glycosidases before the metabolizable portion of the molecule becomes available for its uptake. The nature of transport of these substrates, the enzymes involved and corresponding gene annotations, if any, are provided in Supplemental data 1.

In addition, batch growth data from literature48 was used to evaluate model predictions, quantitatively. Exponential growth phase data points were used for the in silico predictions based on constraint-based flux analysis (see Methods). In order to mimic the complex medium condition, the uptake rates of all amino acids, vitamins and nucleotides were left unconstrained during the simulations. The values of growth associated maintenance (GAM) and non-growth associated maintenance (NGAM) were constrained at 13.613 mmol gDCW−1 and 1.104 mmol gDCW−1 hr−1, respectively. These energetic parameters were estimated as described earlier25 (Supplemental data 1). The comparison of in silico predicted growth rate and product formation profiles were highly consistent with the experimental data (Fig. 3A and B). We also simulated the growth using different carbon sources including glucose, xylose and pyruvate; the resulting by-product yields when compared to those reported in literature49,50,51 were found to be fairly consistent with the observed values (Fig. 3C). It should be highlighted that the growth simulations using iLME620 showed a limited uptake of all exogenously supplied amino acids, mostly proportional to the biomass protein demand as revealed by fluxes through reactions catalysed by the corresponding amino-acyl tRNA synthetases (Supplemental data 1). In contrast, the growth simulations on other LAB models determined unrealistically high amino acid uptake rates, which is attributed to the utilization of some of the supplied amino acid as sources of energy during biomass maximization. Furthermore, the model simulated with and without excess amino acid supply showed no differences in ATP yield (1 mole ATP per mole glucose in both scenarios) which is an unique characteristic of L. mesenteroides. This clearly suggests that L. mesenteroides has restricted or no capability of utilizing amino acids as energy source, invariably requiring regular carbon sources to derive energy. It should be noted that the relatively higher uptake rates of aspartate and glutamate or glutamine (Supplemental data 1) are to satisfy the additional demands imposed by peptidoglycan biosynthetic precursors, D-alanine and glucosamine-6-phosphate. Collectively, the qualitative and quantitative model validation results presented here clearly demonstrated high fidelity of the current model, iLME620.

Figure 3
figure 3

Quantitative predictions of iLME620. (A) Comparison of batch growth rates and in silico model predictions. (B) Comparison of batch fermentation end-product profiles and in silico model predictions; Exp: Experimental value, SIM: In silico simulation. (C) Comparison of in silico and experimentally observed by-product yields from glucose, xylose and pyruvate. (D) Effect of oxygen on the fermentative product profiles and the overall NAD(H) turnover rate as computed by flux-sum. Note that glucose uptake rate was constrained at 10 mmol gDCW−1 hr−1 for simulations used in Fig. 3C and D.

Full size image

We also performed sensitivity analysis to evaluate the robustness of the model predictions, which can be affected by several fixed parameters such as the maintenance energy costs and stoichiometric coefficients of the individual metabolites in the biomass equation (see Methods). The effect of change in energy costs and macromolecular compositions on cellular growth was investigated, thereby identifying protein and GAM as critical components whose variations may affect the model prediction (Fig. 4A–G). We further accounted for potential errors in the measurement of each component by determining the change in growth objective to unit change in the stoichiometry of each component (see Methods), and observed that biomass is highly sensitive to lipid and least sensitive to protein measurements (Fig. 4H ). Hence, we should carefully estimate the energetic costs and potential errors in quantifying cellular lipid content, as these may vary across different growth environments.

Figure 4
figure 4

Sensitivity of cellular growth upon the changes in maintenance costs and macromolecular compositions. (A–G). Effect of 50% variation in the stoichiometries/composition of NGAM, GAM, Protein, Carbohydrate, RNA, DNA and Lipid, respectively on growth at different glucose uptake rates. (H) Plot of sensitivity coefficients, which are ratios of change in growth for unit changes in components for biomass equation, including GAM, Protein, Carbohydrate, RNA, DNA and Lipid, at different glucose uptake rates.

Full size image

Intracellular redox state is a key factor determining the fermentative product profiles

The current model, iLME620, was used to elucidate metabolic states leading to the unique heterolactic fermentative behavior. For example, constraint-based flux simulations showed that lactic acid production during the anaerobic growth conditions is highly modulated by the tight coupling between NAD/NADH production and regeneration routes (Fig. 1A). Under aerobic growth condition, L. mesenteroides utilizes pyruvate oxidase in combination with NADH peroxidase to regenerate additional redox equivalents. Availability of such alternate routes diverts significant amounts of flux away from alcohol dehydrogenase towards pyruvate oxidase and acetate kinase. Hence, as the O2 uptake rate increases from zero to maximum, the yield of lactate, ethanol, and acetate changes from a molar proportion of 1:1:0 to 1:0:1, recapitulating the experimentally observed trend49 (Fig. 3D). This increase in acetate production further results in additional ATP synthesis leading to higher growth rate. The step decrease in the NAD(H) turnover rate with the supply of oxygen (Fig. 3D ), as quantified by flux-sum analysis52, indicates the plausible switch of pyruvate catabolism from alcohol dehydrogenase to pyruvate oxidase. Interestingly, a similar trend involving the induction of pyruvate oxidase activity has been experimentally observed in L. mesenteroides when the culture condition is changed from anaerobic to aerobic mode49,51,53. Taken together, these results imply that the dependence of enzyme activities corresponding to the observed metabolic shifts upon oxygen supply can be solely explained on the basis of prevailing intracellular capabilities to restore redox balances.

MLF is unaffected by the participation of pyruvate and oxaloacetate as reaction intermediates

The low energetic yield of the PKP when compared to EMP pathway imposes severe energy limitations in L. mesenteroides. In this regard, MLF is an important alternate pathway used by L. mesenteroides for overcoming energy limitations under low-nutrient or unfavorable growth conditions (e.g., wine fermentation13, cucumber and cabbage fermentations3). During the MLF phase of wine production, malic acid is converted into lactic acid, thus reducing the excess acidity and enhancing organoleptic properties of wine. However, it is still challenging to control the rate of MLF during wine fermentation, which further prompts for better understanding of the underlying mechanism. A previous study has proposed the conversion of malate to lactate by malolactic enzyme in a single decarboxylation mediated reaction54. Although lactate is the only final product of the active enzyme complex, the requirement of NAD as an essential cofactor in the reaction indicates the possible participation of oxaloacetate and pyruvate as intermediates13. Hence, the MLF may be represented as the combination of three reactions catalyzed by malate dehydrogenase (EC 1.1.1.37), oxaloacetate decarboxylase (EC 4.1.1.3) and lactate dehydrogenase (EC 1.1.1.28) (Fig. 5A ). While the exact combination of MLF is still debatable55, it is not clear whether the involvement of the central metabolic intermediates, pyruvate and oxaloacetate, in the three-step MLF would affect the flux towards malate. Therefore, we analyzed the cellular phenotype under both scenarios, i.e., one-step and three-step MLF (Fig. 5B). Counterintuitively, both cases did not significantly affect the overall flux through the malolactic reaction(s), under both minimal and nutrient rich media (Fig. 5C). This analysis showed that instead of utilizing pyruvate and oxaloacetate for the synthesis of biomass precursors such as amino acids, L. mesenteroides invariably converts the entire assimilated malate to the fermentative by-product, lactate, suggesting a stronger need for regenerating redox equivalents in the organism.

Figure 5
figure 5

Effect of different malolactic fermentation (MLF) mechanisms on the MLF flux. (A) MLF and its relationship with the carbon central metabolism, amino acid biosynthesis and energy generation. Bar graphs provide the flux values (mmol gDCW−1 hr−1) of reactions part of MLF and growth rates with the supply of L-malate (M) and with no supply of L-malate (NM). (B) Two of the mechanisms of MLF that have been proposed by previous studies. (C) Effect of the differences in enzymatic routes of the two mechanisms on the malolactic flux and growth in minimal and nutrient rich media.

Full size image

Furthermore, we evaluated the role of MLF in the cell growth by characterizing metabolic flux states obtained from the model simulation13,56. The flux distributions determined with and without the malate supply revealed that while malate does not contribute to the anabolic carbon pool, a single proton excluded via lactate efflux transporter generates additional ATP for higher growth rate (Fig. 5C ). We observe that the ATP yield due to malate consumption increases by 0.25 moles per mole of malate, which is resulted from a reduced flux through F0F1-ATPase. This value is close to the energy gain due to malate uptake seen in Oenococcus oeni, 0.28 moles of ATP per mole of malate56, asserting MLF as one of the alternative energy conserving processes in heterofermentative LAB.

Existence of proton symporter-dependent glucose-phosphotransferase system activity in obligate heterofermentative LAB?

Energetics of sugar uptake systems play an important role in the cell growth of most bacteria as they can consume significant proportion of the total ATP generated. For example, sugar-phosphotransferase systems (PTS, Fig. 6A), compared to other active transport systems, confer additional benefit to the organism by coupling nutrient transport to substrate level phosphorylation, thereby increasing its ATP efficiency. Previous reports have claimed that L. mesenteroides and other heterofermentative bacteria may not utilize PTS for glucose uptake since the single molecule of phosphoenolpyruvate (PEP) produced per molecule of glucose catabolized via PKP would be used up for its own transport15,57. As a consequence, no PEP would be leftover for the biosynthesis of aromatic amino acids and N-acetylmuramic acid that are required for cell wall synthesis. However, the presence of putative genes encoding glucose-PTS components in L. mesenteroides (LEUM_0507, LEUM_0508 and LEUM_0901) as annotated in TransportDB database35 allowed us to analyze the possible cellular phenotype of L. mesenteroides growing on glucose with and without a glucose-PTS transporter. Accordingly, we tested three possible scenarios of glucose transport as illustrated in Fig. 6B. The first scenario involves a simple glucose-proton symporter, a relatively less energy efficient transport in bacteria that use glucose-PTS transporter. The second one exclusively uses glucose-PTS transporter, which turns out to be infeasible because of the reason described earlier. The last scenario utilizes both transporters in combination. Noticeably, maximum growth occurs when a small proportion of glucose is consumed via glucose-proton symporter towards PEP synthesis, while the rest is transported via PTS for net ATP production (Fig. 6B).

Figure 6
figure 6

Utilization of phosphotransferase system (PTS) for glucose uptake. (A) Sugar PTS systems and their relationship to the carbon central metabolism and biomass precursor biosynthesis. Reactions that carry differential flux with and without the usage of PTS for glucose uptake are represented by ‘red’ lines and the corresponding fluxes are provided as bar graphs (B) Different scenarios of glucose transport via usage of PTS and proton-symporter, and the corresponding growth rates. (C) Bar graph showing comparison of experimentally obtained biomass yields from glucose with the in silico values simulated with and without glucose-PTS.

Full size image

Model simulations showed that the usage of glucose-PTS results in significantly increased biomass yield from glucose i.e., the ratio of growth rate to glucose uptake rate (Fig. 6C ). Therefore, we compared the experimentally determined biomass yields for L. mesenteroides anaerobically grown on glucose as sole carbon source48 with the simulated yields for the two scenarios, with and without glucose-PTS usage. Figure 6C clearly shows that the experimentally observed biomass yields for both cases are consistent with the scenario involving glucose-PTS. In conjunction to this, a report showing glucose-PTS activity in a different obligate heterofermenting LAB, O. oeni 58, very much imply that all obligate heterofermentative LAB are capable of utilizing PTS for transporting glucose.

Integrated transcriptomic analysis reveals regulatory bias towards fluxes correlated to mannitol production

Sucrose is known to support higher growth rate in L. mesenteroides compared to the monosaccharides, glucose and fructose49 although the molecular basis for this observed difference is still unclear. Therefore, we initially performed constraint-based flux analysis to evaluate effects of two carbon sources (sucrose and glucose) on metabolic utilization. As expected, the resultant flux distributions did not reveal any significant differences apart from the initial carbon assimilatory pathways, which motivated us to explore metabolic regulation by profiling the gene expression of L. mesenteroides grown in sucrose or glucose as sole carbon source (see Methods). Subsequently, we carried out metabolic subsystem-wise enrichment analysis using iLME620. Interestingly, the presence of sucrose in the culture medium induced the upregulation of genes encoding sucrose phosphotransferase system (91-fold) and sucrose-6-phosphate hydrolase (72-fold). However, no such induction of genes involved in the substrate assimilation was found for the glucose uptake. In addition to the sugar uptake, several metabolic pathways related to biomass precursor production, including fatty acid biosynthesis, biotin biosynthesis, undecaprenyl diphosphate biosynthesis, and the overall uptake transporters were upregulated (Table 1). Moreover, the energetically costlier purine biosynthesis was selectively down-regulated upon sucrose supply, while the pyrimidine biosynthesis was simultaneously up-regulated, however, the reasons for such differential expression remains unknown. Although the upregulation of biomass precursor biosynthetic pathways observed from the enrichment analysis clearly suggest that growth on sucrose predisposes L. mesenteroides for higher biomass production capabilities, the nature of driving force(s) behind such capabilities remain elusive.

Table 1 Metabolic subsystems part of iLME620 differentially expressed among glucose and sucrose uptake conditions.
Full size table

We then examined the correlation between gene expression and metabolic fluxes, and identified the transcriptionally, post-transcriptionally, or metabolically regulated reactions under sucrose to glucose growth transition by resorting to a method that takes into account the direction of change of the in silico fluxes derived from a constraint-based sampling procedure and the corresponding changes to gene expression levels (see Methods). Table 2 summarizes the top-scoring (probability > 0.9) reactions in each category. We could not derive any meaningful inferences from the post-transcriptionally and metabolically regulated enzymes. However, through additional simulations, we observe that fluxes through reactions catalysed by all the three enzymes identified as transcriptionally regulated, including glucose-6-phosphate isomerase (PGI), acetaldehyde dehydrogenase (ACALD) and ribulose 5-phosphate 3-epimerase (RPE), were negatively correlated to mannitol production (Fig. 7 ). This result is interesting as mannitol is the preferred by-product of L. mesenteroides to relieve redox imbalances during fermentative growth conditions, such as the early stages of sauerkraut production3.

Table 2 List of transcriptionally, post-transcriptionally and metabolically regulated enzymes during glucose-sucrose growth transition.
Full size table
Figure 7
figure 7

Control of mannitol production by transcriptionally regulated enzymes. Effect of flux through PGI (glucose-6-phosphate isomerase), RPE (ribulose 5-phosphate 3-epimerase) and ACALD (acetaldehyde dehydrogenase) on mannitol production. The flux through each of these reactions was varied from a value optimal for growth to a minimum feasible value determined by flux variability analysis and the corresponding mannitol production fluxes were recorded. WT flux: Wild type flux or flux optimal for growth.

Full size image

Discussion

In this study, we presented the first genome-scale metabolic model of obligate heterofermentative LAB, L. mesenteroides ssp. mesenteroides, for better understanding of crucial metabolic modules potentially influencing its industrial and probiotic applications. The in silico growth simulations performed using the model revealed negligible capability of L. mesenteroides to utilize amino acids as energy source unlike other LAB groups59. Such observation is presumably due to the absence of two important processes, the ATP-forming deiminations and proton-motive force generating amino acid decarboxylations, which contribute to energy generation using amino acids in other LAB. They are also known to be involved in intracellular pH buffering59. The absence of such enzymes in L. mesenteroides may therefore explain the relatively low tolerance of the organism to acidic condition. In addition, the negligible amino acid utilization for energy generation explains their existence in many natural ecological niches such as green vegetables or silage43. Besides, it is suggested that, in an association with the gut microbiota, protein-rich diets may lead to the dominance of other LAB over L. mesenteroides.

The knowledge derived from model reconstruction and curation processes enabled us to speculate some unknown metabolic functionalities in L. mesenteroides. For example, with the exceptions of glutamine and valine, the amino acid auxotrophy predictions were consistent with single-amino acid omission experiments40. The glutamine discrepancy could be attributed to the presence of glutamine synthetase (GLNS) (gene locus id: LEUM_0717) in the model, which can synthesize glutamine using glutamate. However, literature evidences suggest possible in vivo repression of GLNS by GlnR, a transcriptional regulator of nitrogen metabolism whose repressor activity has been observed in low-GC containing Firmicutes, including Lactococcus lactis 60 and Streptococcus pneumonia 61 under nitrogen excess condition. Hence, with a refined constraint on GLNS repression, we could replicate the glutamine auxotrophy successfully. The in silico predictions identified valine as non-essential in contrast to the single-amino acid experiments which showed it to be essential. Such discrepancy could possibly arise due to the limitations of constraint-based modeling approach which does not capture the transcriptional and translational regulation; although L. mesenteroides has ORFs encoding for all the genes in valine biosynthetic pathway as accounted in iLME620, some of them could be impaired at translational or transcriptional levels, and thus rendering the pathway inactive.

The obligate heterofermentative LAB possess distinct metabolic pathways to generate/conserve energy. Here, we elucidated two relevant metabolic behaviors to alleviate energy limitations by proposing new hypotheses. Firstly, we suggest that the fate of L-malate consumed via the MLF is independent of the intermediate steps (Fig. 5A ). We ascribe this observation to a substrate channeling-like effect induced by the cellular redox environment, wherein surprisingly the two central metabolic intermediates, i.e. oxaloacetate and pyruvate, form lactate rather than being diverted towards the synthesis of biomass precursors. Existence of such channeling mechanism is further supported by the thermodynamic analysis of MLF: the first step converting malate to oxaloacetate has a large positive Δr G′° (>30 kJ/mol at pH ≤ 7)62, indicating the channelling of malate towards lactate formation as the best possible scenario to overcome such thermodynamic limitation. Overall, this exemplifies the dominance of redox state as a factor controlling cellular processes in L. mesenteroides. Secondly, we highlighted the possibility of PTS-based glucose uptake in L. mesenteroides. Figure 6B clearly shows that most of the glucose is metabolized through PTS to enhance the overall ATP production while only a minor portion of flux through the glucose-proton symporter is used to satisfy the biosynthetic demands of certain biomass precursors such as PEP. Hence, the utilization of PTS for glucose uptake and subsequent dissimilation of carbon flux via PKP should strictly involve its concurrent operation with alternate transporter systems, such as the glucose-proton symporter. Apart from the experimental evidences supporting this hypothesis (Fig. 6C ), discussed earlier, we also observe significant expression of the genes encoding components of glucose-PTS along with the putative glucose-permease, glcU in cells grown on glucose (top 10% of the highly expressed genes; see Supplemental data 1) indicating plausible glucose-PTS and glucose permease activities in L. mesenteroides. Additionally, we also find that the usage of PTS for glucose transport has significant effect on some of the central carbon metabolic fluxes, including those of hexokinase (HEX1), pyruvate kinase (PYK), nucleoside-diphosphate kinase (NDK) and F0F1-ATPase (Fig. 6A ). Interestingly, PYK and NDK fall under the CcpA regulon for L. mesenteroides and other Firmicutes63. Note that the transcriptional regulator, CcpA, has been known to repress glucose-PTS and NDK while activating PYK64,65. The flux changes of NDK and PYK during the cell growth with and without PTS usage, as depicted in Fig. 6A, indeed follow a trend that is consistent with the regulation detailed above, providing additional clues about the predisposal of the metabolic and regulatory structure in this LAB to the usage of glucose-PTS. Overall, these results suggest that although PEP availability could be a limiting factor for the in vivo glucose-PTS activity, L. mesenteroides could theoretically operate this transporter in a coordinated manner to achieve higher net ATP and growth yield.

The integrative analysis of transcriptome data allowed us to delineate the metabolic behaviour and regulation relevant to the higher growth rate observed in sucrose as compared to glucose: (i) upregulation of the genes in sucrose assimilation, (ii) net upregulation of genes for biomass precursor biosynthesis, and (iii) downregulation of PGI, ACALD and RPE which are negative flux control points of mannitol production (Fig. 7), thereby favouring mannitol production. Although the fructose component of sucrose could be used for the production of energy and biomass precursors, it can be partly diverted towards mannitol biosynthesis. These evidences thus suggest the existence of redox imbalances and a transcriptional regulatory bias towards metabolic pathways relieving these imbalances in L. mesenteroides.

In conclusion, our work explained several features unique to L. mesenteroides, and provided a reliable in silico chassis that comprehensively represents obligate heterofermentative metabolism. Discovery of less studied regulatory mechanisms such as the substrate channeling during MLF described here could aid in robust and efficient design of heterologous pathways66. Evidently, the intracellular redox state has been observed as the major phenotype governing factor in L. mesenteroides. Therefore, the future endeavours to evaluate metabolic capabilities of this LAB should take into account the importance of achieving redox state desirable for specific industrial and therapeutic applications.

Methods

L. mesenteroides culture

Chemostat cultivations of L. mesenteroides ssp. mesenteroides ATCC 8923 were performed in an anaerobic culture system. Briefly, cells were grown using a chemically defined medium (CDM) developed earlier40, under carbon limited conditions at a dilution rate (D) of 0.1 h−1, with no oxygen in the bioreactor inlet air. The total inlet gas (N2) flow was controlled by mass flow meters (Bronkhorst High-Tech, Netherlands) at 1.5 vvm. The pH, stirring speed and temperature were maintained at 6.8 (using 0.5 N NaOH), 700 rpm and 30 °C, respectively. Biomass harvested from the chemostat culture was subsequently used to analyze the macromolecular composition (see next section).

Biomass composition analysis

Drafting a concise biomass assembly equation representing the overall cellular composition of L. mesenteroides is an important prerequisite for analysing the intracellular metabolism at exponential phase of batch culture or at steady state in chemostat, where the primary aim of cell is to grow. Therefore, in order to formulate such biomass equation, we experimentally measured the total protein, carbohydrates, RNA and DNA content, and the detailed amino acid composition of L. mesenteroides from lyophilized cells harvested from steady state chemostat culture. In each case, at least three independent samples, from two independent cultivations, were analyzed.

Elemental analysis was performed using 1 mg of a lyophilized biomass. After adding 1 mg of V2O5, the samples were subjected to 1000 °C in an oven. The volatile compounds thus released from the sample were measured using an Elemental Analyzer (NA2000 ThermoFisher, USA).

Total carbohydrates were determined by the phenol method as described earlier67. Total protein content was determined by means of the BCA method68. Protein concentration was calculated using bovine serum albumin (BSA) as standard. The total lipid content was determined after methanol–chloroform extraction as described elsewhere69. Genomic DNA (gDNA) and total RNA was extracted from L. mesenteroides grown in the CDM at 30 °C under anaerobic conditions until the cell number reached 108–109/ml. The chromosomal DNA was extracted from the cells using a gDNA prep kit (SolGent, Korea) and RNA extracted using a RNA prep kit (MACHEREY, Germany). Quantification and purity were measured by nanodrop (Epoch, BioTek Laboratory, Shoreline, WA, USA).

Amino acid composition of the proteins extracted from L. mesenteroides cells was determined using the amino acid analyzer (Biochrom 30; using the Biochrom EZ Chrom software, for quantification) using a cation exchange chromatography column and post-derivatization with ninhydrin as described previously70. Nucleotide composition of DNA was calculated using the GC content of L. mesenteroides genome (37.8%)32. Total glycerolipid, phospholipid and peptidoglycan content, and fatty acid composition were accounted based on bibliographic information (Supplemental data 1). The stoichiometric coefficients calculated for biomass precursors are represented in terms of mmol gDCW-1 (milli moles per gram of dry cell weight) in the equation. Dry cell weight was obtained by weighing the lyophilized cells. Additional details regarding the formulation of biomass equation are provided in the Supplemental data 1.

Metabolic network reconstruction

The genome scale metabolic model of L. mesenteroides was reconstructed based on the genome annotation32 and information compiled from various databases and literature, following the well-established protocol31. Initially, a draft metabolic network was reconstructed by reconciling the L. mesenteroides specific metabolic and transport reactions collected from KEGG34, MetaCyc33 and TransportDB35 databases. Duplicate and generic reactions were subsequently eliminated from the draft reconstruction using EC numbers and NCBI locus tags as common identifiers. The reconstruction content was then manually curated for reaction directionality, elemental and charge balancing of individual reactions. We employed the thermodynamic data generated by earlier studies37,38,39 to assign reaction directionality in the current model using the approach proposed earlier38. Note that those reactions with conflicting information on directionality were assigned to be reversible. The directionality of the reactions, for which no information was available, were determined based on heuristic information available in published literature and/or from biochemical databases such as Brenda71 and MetaCyc33. Metabolic dead-ends were then identified using the GapFind algorithm36 and were subsequently filled in by the addition of new reactions from other organisms based on literature evidences. Here, in order to substantiate the existence of such reactions in L. menesteroides, protein sequence homology of the corresponding enzymes were performed prior for their inclusion. Additionally, certain non-gene associated reactions were included in the reconstruction based on either direct or indirect biochemical evidences from the literature. Finally, the gene-protein-reaction (GPR) associations were manually added to each reaction based on the enzyme subunit and isoenzyme information obtained from the biochemical databases.

Constraint-based flux analysis

Constraint-based flux analysis was employed to analyze the metabolic phenotype of L. mesenteroides while growing on a nutrient rich medium. The biomass equation was maximized while simultaneously constraining the uptake rates of various amino acids and vitamins, and secretion rates of fermentation by-products based on a previously published study40. The optimization problem can be mathematically represented as follows:

$${\rm{\max }}\,Z=\sum _{j}{c}_{j}{v}_{j}$$
(1)
$${\rm{s}}{\rm{.t}}.\sum _{j}{S}_{ij}{v}_{j}=0\forall \,{\rm{metabolite}}\,{i}$$
(2)
$${v}_{j}^{\min }\le {v}_{j}\le {v}_{j}^{\max }\,\forall \,{\rm{reaction}}\,j$$
(3)

where S ij is the stoichiometric coefficient of metabolite i participating in the reaction j; v j represents flux through the reaction j; v j min and v j max respectively represent the lower and upper bounds on the flux through the reaction j; and Z is the biomass objective, where c j denotes the relative contribution of each metabolic reaction to the biomass formation. Flux variability analysis and flux-sum analysis were performed extending the formulation shown above as described earlier52,72. Briefly, the turnover rate of metabolite i (ϕ i ) obtained using flux-sum analysis is given by

$${\varphi }_{i}=\frac{1}{2}\sum _{j}|{S}_{ij}\cdot {v}_{j}|$$
(4)

Amino acid auxotrophic simulations were performed by setting each amino acid uptake rate to zero, individually, while maximizing the growth cellular objective. In this study, amino acid auxotrophy is defined by the condition of > 90% growth reduction due to exclusion of corresponding amino acid from the media. All constraint-based flux analysis simulations including gene-essentiality analysis were performed using COBRA toolbox72 with Gurobi5 (http://www.gurobi.com) as optimization solver.

Thermodynamic analysis of metabolic pathways

In order to perform pathway-wise thermodynamic analysis and to calculate the transformed Gibbs free energy changes of the corresponding metabolic reactions under physiologically relevant pH (7.2), ionic strength (0.1 M) and intracellular metabolite concentrations (1μM-10mM), we used the web-based tool, eQuilibrator (http://equilibrator.weizmann.ac.il). This tool employs a database containing thermodynamic information along with an extended version of the group contribution method to perform the Gibbs free energy calculations73.

Sensitivity analysis

The sensitivity of cell growth to variation in maintenance energy costs (GAM and NGAM) and the macromolecular composition was evaluated using two different measures. Firstly, each parameter was individually adjusted within a ± 50% error range, and the corresponding effect on the growth objective was determined. Secondly, sensitivity coefficients were calculated as the ratio of change in the growth objective to a unit change in the stoichiometric coefficient of each macromolecular component in the biomass equation.

Microarray hybridization and expression data analysis

The gene expression of L. mesenteroides was profiled while growing on sucrose (S) and glucose (G) as sole carbon source in MRS broth. cDNA samples were prepared from the total RNA extracted using a RNA prep kit (MACHEREY, Germany) from exponentially grown cells of S and G groups in two biological replicates. The oligonucleotide probes (35–40 mers, 2033 probes) for 1871 genes, designed based on the genome sequence of L. mesenteroides (Genbank accession no. CP000414), were synthesized and printed onto the CombiMatrix 4 × 2 K custom array platform (Macrogen Inc., Korea). The probe design, spotting and analysis were conducted by Macrogen. Local-pooled-error (LPE) test and fold change were applied using R 2.4.1 software to determine and evaluate the significance of each gene’s differential expression in S and G groups.

Analysis of transcriptomic data for differentially expressed pathways using iLME620

The gene-expression data with coefficient of variance (CV) < 0.3 (less than 30% variance between replicate data points) from the quantile normalized microarray datasets of the S and G groups were employed to identify the differentially expressed pathways between sucrose and glucose uptake conditions. To do so, the gene expression values were mapped to the corresponding locus tags in iLME620 using the gene-protein-reaction associations (GPRs). Gene-expression fold-change values were converted into reaction fold-change. Lowest fold change value between two or more genes with “AND” relationship was considered to account for the proportional fold-changes between the subunits of an enzyme complex required to effectuate the overall reaction fold-change. For reactions associated with genes having “OR” relationship, the fold-change values were added to account for the additive nature of intracellular enzyme concentrations. Significance of differential expression of various subsystems in S and G uptake conditions was determined using two-sided Wilcoxon signed rank test.

Identification of transcriptionally, post-transcriptionally and metabolically regulated reactions

A Z-score-based approach was used to identify the transcriptionally, post-transcriptionally and metabolically regulated reactions74. First, the flux solution space of L. mesenteroides growing on S or G was sampled using the OptGpSampler75 implemented in COBRA toolbox. The significance of change between the expression and flux of each reaction was then Z-scored using the formula below:

$${Z}_{i}^{E}=\frac{mean(E{2}_{i})-mean(E{1}_{i})}{\sqrt{Variance(E{2}_{i})+Variance(E{1}_{i})}}$$
(5)
$${Z}_{i}^{F}=\frac{mean(F{2}_{i})-mean(F{1}_{i})}{\sqrt{Variance(F{2}_{i})+Variance(F{1}_{i})}}$$
(6)

where Z i E and Z i F are the Z-scores for expression and flux of reaction i, respectively. E1 i and E2 i are the expression intensity values and F1 i and F2 i are the flux values of reaction i in condition 1 (S) and condition 2 (G), respectively. Note that 5000 sample points using 500 steps between each point was required for good convergence, as suggested earlier74. OptGpSampler75 implemented within COBRA toolbox was used to obtain sample points for both S and G conditions. Subsequently, the Z-scores were used to calculate the probabilities of correlation between the flux and gene expression of reaction. Briefly, a reaction is said to be transcriptionally regulated if the product of the probability of expression change of a reaction and the probability of the flux of the reaction changing in the same direction is significant. A reaction is considered metabolically regulated if the probability of no change in its expression level but change in flux is significant. And, the reaction is post-transcriptionally regulated if the probability of change in its expression level not leading to a change in its flux is significant.