In silico and in vivo analyses reveal key metabolic pathways enabling the fermentative utilization of glycerol in Escherichia coli

Summary Most microorganisms can metabolize glycerol when external electron acceptors are available (i.e. under respiratory conditions). However, few can do so under fermentative conditions owing to the unique redox constraints imposed by the high degree of reduction of glycerol. Here, we utilize in silico analysis combined with in vivo genetic and biochemical approaches to investigate the fermentative metabolism of glycerol in Escherichia coli. We found that E. coli can achieve redox balance at alkaline pH by reducing protons to H2, complementing the previously reported role of 1,2‐propanediol synthesis under acidic conditions. In this new redox balancing mode, H2 evolution is coupled to a respiratory glycerol dissimilation pathway composed of glycerol kinase (GK) and glycerol‐3‐phosphate (G3P) dehydrogenase (G3PDH). GK activates glycerol to G3P, which is further oxidized by G3PDH to generate reduced quinones that drive hydrogenase‐dependent H2 evolution. Despite the importance of the GK‐G3PDH route under alkaline conditions, we found that the NADH‐generating glycerol dissimilation pathway via glycerol dehydrogenase (GldA) and phosphoenolpyruvate (PEP)‐dependent dihydroxyacetone kinase (DHAK) was essential under both alkaline and acidic conditions. We assessed system‐wide metabolic impacts of the constraints imposed by the PEP dependency of the GldA‐DHAK route. This included the identification of enzymes and pathways that were not previously known to be involved in glycerol metabolisms such as PEP carboxykinase, PEP synthetase, multiple fructose‐1,6‐bisphosphatases and the fructose phosphate bypass.


Introduction
Fermentative and respiratory metabolisms represent the primary modes of cellular energy metabolism. In the absence of external electron acceptors, fermentative metabolism operates utilizing internal electron acceptors to maintain overall redox balance and substrate-level phosphorylation for ATP generation (Bettenbrock et al., 2014). If an external electron acceptor (e.g. O 2 ) is available, respiratory metabolism can utilize the electron transport chain and oxidative phosphorylation for redox balancing and ATP generation (Lengeler et al., 1999;Nicholls and Ferguson, 2013). These contrasting metabolic modes often rely on different enzymes, pathways and regulatory processes, which has significant consequences on the ability of and mechanisms for microorganisms to utilize carbon and energy source(s) under varying conditions (Bettenbrock et al., 2014). For example, although many microorganisms can metabolize glycerol under respiratory conditions, few can do so under fermentative conditions (Clomburg and Gonzalez, 2013).
A critical factor in the dichotomy between fermentative and respiratory glycerol utilization is the highly reduced nature of this carbon source (j = 4.67; degree of reduction per carbon, j, a measure of the number of available electrons per unit of carbon). With more oxidized carbon sources such as glucose (j = 4), the synthesis of cell mass (j = 4.3, based on average biomass molecular formula of CH 1.9 O 0.5 N 0.2 ) results in a net consumption of reducing equivalents (Nielsen et al., 2003). However, the synthesis of cell mass from (i.e. growth on) glycerol generates reducing equivalents, tying a microorganism's ability to maintain redox poise to a sink for these excess reducing equivalents (Clomburg and Gonzalez, 2013). Under respiratory conditions, this is achieved via electron transfer to an external acceptor (Fig. 1A); however, options are much more limited under fermentative conditions where redox balance must be maintained by regenerating oxidized electron carriers using internal metabolites. The most well-established means of enabling fermentative glycerol utilization are through the synthesis of highly reduced 1,3-propanediol (1,3-PDO, j = 5.33), a model extensively studied in several species of the Enterobacteriaceae family ( Fig. 1B; Bouvet et al., 1995;Sun et al., 2003). While the absence of 1,3-PDO-producing capability was long thought to confine microorganisms to respiratory glycerol metabolism, fermentative glycerol utilization has been reported in non-1,3-PDO-producing organisms such as Escherichia coli (Dharmadi et al., 2006;Zhang et al., 2010), Paenibacillus macerans (Gupta et al., 2009) and Propionibacterium acidipropionici (Bories et al., 2004).
Fermentative glycerol utilization by E. coli has been further established and linked to its ability to produce 1,2-propanediol (1,2-PDO, j = 5.33), whose synthesis from glycerol results in the net consumption of NADH, although it also consumes ATP ( Fig. 1C; Gonzalez et al., 2008). Ethanol synthesis was also shown to play an essential role as the overall glycerol to ethanol conversion is redox-balanced and enables the synthesis of ATP through glycolysis ( Fig. 1C; Gonzalez et al., 2008).
Here, we utilize a genome-scale model (GEM) of E. coli in combination with Flux Balance Analysis (FBA) and Flux Variability Analysis (FVA) to investigate the distribution of metabolic fluxes during the fermentative utilization of glycerol in E. coli. This in silico analysis facilitated the identification of key areas within E. coli metabolism that was further investigated in vivo using genetic and biochemical approaches. Our findings reveal that E. coli possesses multiple means for achieving redox balance during fermentative glycerol utilization dependent on environmental conditions such as pH, with proton reduction to hydrogen (H 2 ) via a reversible hydrogenase critical under alkaline conditions. H 2 -dependent redox balancing can improve ATP generation compared with 1,2-PDO production previously shown to operate under acidic conditions and is enabled by the activity of the respiratory glycerol dissimilation pathway which generates electrons in a form amenable to hydrogenasemediated proton reduction. However, owing to the pH dependence of proton reduction via hydrogenases and the inability of the respiratory pathway to exclusively support glycerol fermentation, the previously established GldA-DhaKLM dissimilation route was found to be essential under both alkaline and acidic conditions. The stoichiometric coupling imposed by the PEP dependence of DHA kinase further underlines the importance of several enzymes and pathways that were not previously known to be involved in glycerol metabolism. Together, these findings demonstrate how the robust metabolic network of E. coli manages the stoichiometric, redox and ATP generation constraints associated with the fermentative metabolism of glycerol.

Results and discussion
Hydrogen evolution and 1,2-PDO synthesis both contribute to redox balance during fermentative metabolism of glycerol Utilizing a curated genome-scale model for E. coli (GEM: see Appendix S1 for details), Flux Balance Analysis (FBA) and Flux Variability Analysis (FVA) were used to predict cell growth and flux distributions and solution spaces, respectively (Fang et al., 2020). Optimization to maximize biomass resulted in a specific growth rate of 0.075 h À1 , with product solution spaces revealing the existence of multiple solutions with varying product fluxes (i.e. the edge of the curve defining the maximum specific growth in Fig. 2A). Ethanol and H 2 represent growthcoupled products, with all optimal solutions for biomass maximization requiring non-zero fluxes ( Fig. 2A). The production of and growth-coupling to ethanol are expected based on the redox-balanced and ATP-generating nature of the pathway from glycerol to ethanol and formate (or glycerol to ethanol, CO 2 and H 2 : C 3 H 8 O 3 ? C 2 H 6 O + CH 2 O 2 /CO 2 -H 2 ). However, the prediction of significant acetate production and non-essentiality of 1,2-PDO synthesis as part of optimal solution(s) maximizing biomass formation ( Fig. 2A) represents a phenotype not previously observed experimentally. To account for this deviation, acetate production was constrained in the model to a previously observed molar yield (0.012 mol mol À1 glycerol consumed; Murarka et al., 2008). Based on these previous experiments, succinate production was also constrained (0.015 mol mol À1 glycerol consumed). Furthermore, reactions enabling the transfer of electrons from NAD(P)H to the quinone pool were removed, as the expression of enzymes responsible for this conversion (NADH:ubiquinone oxidoreductase I and II) is induced by O 2 or other external electron acceptors (Bongaerts et al., 1995;Tran et al., 1997). Simulations with these constraints indicated that cycling between malate/fumarate and glycolate/glyoxylate reactions was occurring to shift electrons between NAD(P)H and quinones, which do not represent a phenotype experimentally observed, and hence, these cycles were removed as well.
With this updated model, optimization to maximize biomass resulted in single optima for ethanol and H 2 production at the maximum specific growth rate of 0.075 h À1 , with 1,2-PDO production again a non-essential product (Fig. 2B). H 2 production is higher than that of ethanol (plus acetate; Fig. 2B), indicating that its generation is not solely from formate oxidation. It should be noted that conversion of pyruvate to acetyl-CoA was predicted to take place exclusively via pyruvate formate lyase (PFL) with no flux through pyruvate dehydrogenase (PDH). While PDH is active during the fermentation of more oxidized carbon sources such as glucose (j glucose = 4; Murarka et al., 2010b) and glucuronate (j glucuronic acid = 3.33; Murarka et al., 2010a), negligible activity is expected due to PDH inhibition by NADH under the highly reduced intracellular environment generated by glycerol (j glycerol = 4.67) fermentation (de Graef et al., 1999;Sawers and Clark, 2004).
Given the role of H 2 evolution via hydrogenases (HYDs) predicted by the in silico analysis, we investigated the impact of gene deletions impairing hydrogen metabolism (genes encoding subunits of various hydrogenases). In addition to specific gene deletions, an important consideration when evaluating the role of  (Sargent, 2016), has a negative impact on glycerol fermentation under acidic pH, but negligible impact under alkaline conditions (Dharmadi et al., 2006;Gonzalez et al., 2008;Murarka et al., 2008). Furthermore, reports have demonstrated that overall HYD activity and the involvement of specific HYDs in H 2 production by E. coli cells grown in the presence of glycerol are pH dependent, with hydrogenase 1 (HYD1) and 2 (HYD2) involved at alkaline pH and hydrogenase 3 (HYD3) responsible for H 2 production at acidic pHs (Trchounian and Trchounian, 2009;Trchounian et al., 2012;Trchounian et al., 2013). Based on these previous findings and the model prediction of H 2 generation not solely from formate oxidation, we initially conducted fermentations at an alkaline pH (7.2). To facilitate the investigation into numerous gene deletions, we utilized the parent strain E. coli BW25113 and associated single-gene deletion mutants of the Keio collection . For hydrogenases, these included strains with deletions documented to impair hydrogenase 1, 2 and 3: hyaB, encoding the large subunit of HYD1 that contains the [NiFe] active site (Menon et al., 1991), hybA encoding the iron-sulphur protein shown to be essential for reversible electron transfer between HYD2 and the quinone pool (Pinske et al., 2015), hybC encoding the large, [NiFe]-containing subunit of HYD2 (Dubini et al., 2002) and hycE encoding the large subunit of HYD3 (Sauter et al., 1992). Cell growth and glycerol consumption were evaluated for fermentations conducted under strict anaerobic conditions (see Experimental procedures) with time profiles assessed in 24-h intervals based on our previous demonstration of exponential growth extending through at least the first 30 h of cultivation .
As glycerol fermentation at both acidic and alkaline pH is homoethanologenic in nature , ethanol synthesis was also utilized as a key metric to evaluate the impact of various gene deletions. Under these conditions, fermentative glycerol utilization was observed in BW25113 with ethanol representing the predominant product as expected ( Fig. 2C and D). Interestingly, estimating the average specific growth rate during the first 24 h, assuming a short-lag phase, results in a value comparable with the model prediction (~0.08 experimentally vs 0.075 h À1 from the model), which are both higher than the previously reported growth rates under acidic conditions (0.04 h À1 ; Murarka et al., 2008) where 1,2-PDO production was identified during glycerol fermentation. Analysis of gene deletions impairing hydrogen metabolism also revealed that while HYD3 disruption has a minimal impact under these alkaline conditions, aligning with previous findings, disruption of HYD1 (DhyaB) and especially HYD2 (DhybA, DhybC) resulted in reduction to cell growth, glycerol consumption and ethanol synthesis ( Fig. 2C and D). While no significant difference in growth or glycerol consumption was observed in the DhycE strain, there was a near fourfold increase in formate concentration compared with BW25113 (1.13 AE 0.04 vs 0.31 AE 0.02 g L À1 ). Similar high formate production and minimal impact to cell growth and glycerol consumption were observed in a DfdhF strain (data not shown), indicating that hycE deletion resulted in a non-functional FHL complex. To further assess the impact of individual hydrogenase disruption in the context of pH, we evaluated BW25113 and hydrogenase deletion strains at an acidic pH (6.5). As shown in Fig. 2E, under these conditions disruption of the various hydrogenases had a much less pronounced impact on glycerol utilization compared with alkaline pH, with only DhybA and DhybC strains deficient in HYD2 resulting in slight decreases to glycerol consumption levels compared with BW25113. It should be noted that the minimal impact of HYD3 disruption under acidic conditions here is likely the result of CO 2 included in the headspace, as our previous results have demonstrated the negative impact of FHL disruption on glycerol fermentation under acidic conditions is mitigated by a partial CO 2 headspace atmosphere .
The above results clearly demonstrate a role of hydrogenases, specifically HYD1 and HYD2, during the fermentative utilization of glycerol that is dependent on external pH. While HYD1 and HYD2 have commonly been thought to act as H 2 -uptake enzymes, both enzyme complexes have been shown to contribute to H 2 production depending on the external pH and carbon source, including cells grown in the presence of glycerol (Trchounian and Trchounian, 2009;Trchounian et al., 2013Trchounian et al., , 2017. Furthermore, the O 2 -sensitive HYD2 is well adapted to reduced environments to the extent that it can act as an electron release valve if the quinone pool becomes over-reduced when electron acceptors are limited (Lukey et al., 2010). This behaviour, combined with the model prediction of the importance of H 2 production and our experimental validation that glycerol fermentation significantly benefits from functional HYD1 and HYD2 complexes at alkaline pH, supports a scenario in which the highly reduced nature of glycerol and absence of external electron acceptors result in proton reduction to H 2 serving as a mechanism for redox balancing during glycerol fermentation under certain conditions. Importantly, the ability for HYD2 to catalyse H 2 evolution has been shown to occur with glycerol as the carbon source using electrons in the form of menaquinol or demethylmenaquinol (Pinske et al., 2015). Furthermore, the reversibility of HYDs, including that of HYD2, has been shown to rely on the proton motive force (PMF; Trchounian et al., 2013;Pinske et al., 2015), while the activity of HYDs both depends on and contributes to F 0 F 1 -ATPase activity (Trchounian et al., 2011(Trchounian et al., , 2013. During glycerol fermentation, it has been proposed that coupling of HYD and proton ATPase activities to regulate intracellular pH contributes to overall PMF generation, which is pH dependent (Trchounian and Trchounian, 2019). Thus, while HYD-dependent proton reduction may enable H 2 production to consume excess reducing equivalents, the reliance on the PMF may dictate pH dependence. Our findings of the importance of HYDs here support this proposed involvement and suggest proton reduction to H 2 can serve as a sink for excess reducing equivalents generated from biomass formation during glycerol fermentation at alkaline pH. An alternative mechanism such as 1,2-PDO production is required under conditions in which HYD reversibility is less favourable (e.g. acidic conditions).
To compare 1,2-PDO synthesis and HYD-dependent H 2 production as enablers of glycerol fermentation under acidic or alkaline conditions, respectively, we used in silico modelling to determine the solution space when the reversible HYD reaction involving menaquinol was removed from the reaction network. In this case, optimization to maximize biomass formation resulted in a lower specific growth rate (0.057 h À1 ), with ethanol remaining the predominant product and 1,2-PDO now representing a growth-coupled product (Fig. 2F), findings that align with both previously determined growth rates and the importance of these products at acidic pH (Dharmadi et al., 2006;Gonzalez et al., 2008;Murarka et al., 2008). While H 2 remained growth-coupled, its production was at stoichiometric levels to ethanol (plus acetate) indicating its generation solely from formate oxidation via FHL (Fig. 2F). These findings demonstrate that E. coli can use either 1,2-PDO or HYD-dependent H 2 production as a sink for excess reducing equivalents, thus achieving redox balance. While the H 2 evolution route appears more efficient and results in higher specific growth rates, the pH dependence of this mechanism limits the conditions under which it plays a prominent role. This higher efficiency could relate to energy demands of 1,2-PDO production, as its synthesis from glycerol results in the net consumption of an ATP molecule (Fig. 1C). Furthermore, in contrast to diverting a fraction of carbon to 1,2-PDO solely to consume reducing equivalents, H 2 evolution also enables increased carbon flux to ethanol synthesis with the corresponding increase in substrate-level phosphorylation via glycolysis resulting in further improved ATP yield.

Involvement of different glycerol dissimilation pathways depends on the mode of redox balancing
An interesting facet to the role of H 2 evolution and hydrogenase activity in overall redox balance is the form of electrons donor(s) that can facilitate proton reduction to H 2 . As HYD1 and HYD2 operate in conjunction with the quinone pool, their involvement during fermentative glycerol metabolism implies the generation of reduced quinone (e.g. menaquinol) to enable proton reduction. However, previous studies on fermentative glycerol metabolism under 1,2-PDO-dependent, acidic conditions have demonstrated the essentiality of an NADHgenerating glycerol oxidation pathway mediated by GldA and DhaKLM under these conditions (Dharmadi et al., 2006;Gonzalez et al., 2008;Murarka et al., 2008). To investigate how glycerol dissimilation, central carbon metabolism, hydrogen metabolism and overall redox balance are connected under various conditions, we examined in silico flux distributions for the scenarios of alkaline H 2 -enabled (Fig. 3, black text) and acidic 1,2-PDO-enabled (Fig. 3, orange text) glycerol fermentation.
In addition to the NADH-generating GldA and DhaKLM-mediated pathway (Fig. 3, shaded grey), for the H 2enabled scenario flux through the respiratory pathway composed of glycerol kinase (GlpK) and anaerobic glycerol 3-phosphate dehydrogenase (GlpABC) was also predicted (Fig. 3, shaded purple). The involvement of the GlpK-GlpABC pathway is linked to HYD-mediated H 2 evolution through the generation of menaquinol from glycerol-3-phosphate (G3P) oxidation, demonstrating how operation of a respiratory pathway under anaerobic conditions could result in electron donor(s) facilitating proton reduction (Fig. 3). In addition to these two routes, flux through a dissimilation pathway involving fructose-1,6-biphosphate (F1,6-BP) as an intermediate was also predicted (Fig. 3, shaded light blue). This pathway, herein termed the fructose phosphate (FP) bypass, involves aldol condensation of DHA, the product of glycerol oxidation via GldA and glyceraldehyde 3-phosphate forming fructose 6-phosphate (F6-P), which is subsequently phosphorylated to F1,6-BP and cleaved to glyceraldehyde 3-phosphate and DHAP (Fig. 3). While this pathway has not been implicated in glycerol dissimilation to date, the gldA operon also includes fsaB, encoding a F6-P aldolase (FsaB). Both FsaB and another F6-P aldolase, FsaA (encoded by fsaA), have been shown to catalyse the aforementioned aldol condensation (Schurmann and Sprenger, 2001). Given these findings, we conducted FVA to determine the minimum and maximum reaction fluxes that support the maximum predicted biomass flux for the H 2 -enabled and 1,2-PDO-enabled scenarios (0.075 and 0.057 h À1 , respectively). This analysis showed that glycolysis and product synthesis fluxes could not vary, expected based on the single optima observed in the product solution spaces ( Fig. 2B and F); however, a significant amount of flux variability through glycerol dissimilation pathways can be tolerated (Table S1). This involves modulating flux through the NADH-generating GldA-DhaKLM and FP bypass pathways, at the expense of one another, as the flux through the menaquinol-generating GlpK-GlpABC pathway was fixed at a value equal to HYDdependent H 2 evolution (Table S1). Thus, from a stoichiometric standpoint there is not a discernable difference in either NADH-generating glycerol dissimilation pathway; however, concurrent operation of NADH-and menaquinol-generating dissimilation pathways is required for HYD-dependent H 2 evolution. This reflects a scenario in which coupling G3P oxidation to proton reduction enables H 2 production to serve as a sink for excess reducing equivalents; however, the GlpK-GlpABC route is unable to exclusively support glycerol fermentation.
To determine the importance of these dissimilation pathways in vivo, we utilized BW25113 derivatives with deletions to gldA (encoding GldA), dhaK (encoding a subunit of DhaKLM), glpK (encoding GlpK) as well as fsaA and fsaB, encoding the two distinct F6-P aldolases (Fig. 4A). Given the lack of knowledge on the role of these F6-P aldolases, we also constructed a DfsaA DfsaB double mutant. Under alkaline conditions, disruption of either GldA or DhaKLM resulted in near abolishment of glycerol consumption ( Fig. 4B and C), confirming their essential role during glycerol fermentation. glpK deletion also had a significant impact under these conditions, with more than twofold reduction in glycerol consumption observed ( Fig. 4B and C). Further confirming the involvement of these dissimilation pathways, activity of GldA, DhaKLM, GlpK and GlpABC was all detected in strain BW25113 while GlpD activity was not (Table S2). However, under acidic conditions the glpK deletion had a less pronounced impact (Fig. S1) providing further evidence towards the link between operation of this respiratory pathway and HYD activity and the associated pH dependence.
The combined deletion of both F6-P aldolases also decreased glycerol consumption (M = 2.01 g L À1 , SD = 0.06 g L À1 ) compared with BW25113 (t(5.97) = 9.56, p < 0.01; Fig. 4C). Intriguingly, individual deletions to fsaA and fsaB did not have a significant impact on glycerol consumption, indicating that either enzyme may be able to complement for deletion of the other. It should also be noted that while gldA deletion results in disruption of both the FP bypass and the GldA-DhaKLM pathway, the lack of glycerol consumption in the DdhaK strain demonstrates the essential nature of the GldA-DhaKLM route under both acidic and alkaline conditions (Fig. 4C).   Overall, these findings demonstrate the importance of parallel operation of NADH-and menaquinol-generating dissimilation pathways under alkaline conditions in which HYD1 and HYD2 play an important role. Given the potential but non-essential role of the FP bypass (Fig. 4C), we refined our in silico model by limiting the upper bound flux through this pathway to 20% of the incoming glycerol transport flux, a value representative of the decrease in glycerol consumption observed in the DfsaA DfsaB strain. Optimization to maximize biomass formation with this refined model results in a specific growth rate of 0.075 h À1 , with the flux distribution predicting glycerol dissimilation through all 3 routes (Fig. 3, green text). Additional FVA analysis confirmed that FP bypass flux can vary in the permissible range, with associated variability through GldA-DhaKLM, while flux through GlpK-GlpABC did not allow for variability at the maximum growth rate (Table S1). These imposed constraints serve to confine glycerol dissimilation fluxes in agreement with our in vivo data, allowing us to further investigate perturbations in other areas of the metabolic network.

NADH, H +
Impact of stoichiometric constraints imposed by the PEP dependency of the essential glycerol utilization pathway, GldA-DhaKLM Despite contributions to glycerol dissimilation from three different pathways, both the in silico analysis and the experimental evidence clearly established that the GldA-DhaKLM pathway is the only one essential for glycerol fermentation. This essentiality places unique constraints on glycerol fermentation as DhaKLM uses PEP as the phosphate group donor for DHA phosphorylation. While PEP-dependent phosphorylation systems exist in nature for other substrates, such as glucose, the fact that both the substrate (glycerol) and the phosphate donor (PEP) are 3-C molecules imposes distinctive restrictions. Unlike glucose where the phosphorylated product (glucose-6-phosphate) is converted into two 3-C molecules (one of which is used to regenerate the PEP utilized for phosphorylation), for 3-C glycerol this creates a cycle in the metabolic network: every mole of DHA converted to DHAP is coupled to the conversion of one mole of PEP to pyruvate (Fig. 5A). This stoichiometric constraint greatly limits the availability of PEP and all 3-C intermediates upstream of PEP and hence the synthesis of biosynthetic precursors or products (e.g. 1,2-PDO) derived from them.
To probe this coupling, we investigated how removing various glycerol dissimilation pathways from the in silico reaction network in combination with key reactions interconverting PEP and pyruvate impacts the predicted growth rate. The latter included the pyruvate kinase reaction, converting PEP to pyruvate with concomitant ATP generation, and PEP synthetase, encoded by ppsA, a gluconeogenic enzyme which phosphorylates pyruvate to PEP requiring 2 ATP equivalents (ATP is converted to AMP; Fig. 5A; Niersbach et al., 1992). As seen in Fig. 5B, the in silico modelling predicts that removal of pyruvate kinase has a slight impact on growth rate when GlpK-GlpABC is present. However, the removal of the two PEP-decoupled glycerol dissimilation pathways (i.e. GlpK-GlpABC and FP bypass) significantly decreased the predicted specific growth rate to 0.044 h À1 with no further effect from pyruvate kinase removal. While this lower growth rate hints at the constraints placed on the metabolic network when glycerol dissimilation is fully coupled to PEP, the true implications of this are seen by the prediction that removing PEP synthetase in the absence of a PEP-decoupled pathway completely inhibits cell growth (Fig. 5B). Thus, without an additional dissimilation pathway that can uncouple glycerol dissimilation from requiring PEP, cells are forced to utilize the ATP intensive PEP synthetase to regenerate PEP from pyruvate to enable growth.
To investigate these predictions, we assessed the impact of deletion of genes encoding enzymes that catalyse the corresponding reactions, including pyruvate kinases I (pykF) and II (pykA), as well as a DpykF DpykA double mutant (Ponce et al., 1995). pykF deletion had marginal impact on glycerol consumption, while a pykA deletion decreased glycerol consumption more than twofold, with the double mutant exhibiting a similar decrease (Fig. 5C). This suggests that conversion of PEP to pyruvate via the kinase reaction is important, with PykA responsible under the examined conditions, implying the operation of a PEP-independent dissimilation pathway (i.e. GlpK-GlpABC and FP bypass, as presented in the previous section). To further demonstrate the relationship between a PEP-independent dissimilation pathway(s) and pyruvate kinase activity, we overexpressed an ATPdependent (PEP-independent) DHA kinase from Citrobacter freundii (CfDhaKL; Daniel et al., 1995) in conjunction with GldA in the DpykF DpykA strain. While minimal glycerol consumption is observed with CfDhaKL overexpression, significant consumption is seen with overexpression of E. coli DhaKLM along with GldA in this background (Fig. 5D) showing the importance of pyruvate kinase activity with a PEP-independent dissimilation pathway.
The importance of a PEP-independent glycerol dissimilation pathway(s) becomes more evident in a DppsA background. Following the in silico results, while individual disruption of PEP synthetase had minimal impact in vivo, the further deletion of glpK resulted in significant decreases in cell growth and glycerol consumption (Fig. 5C). The impact of this DppsA DglpK background is as severe as the disruption of GldA-DhaKLM (Fig. 4C) or ethanol synthesis (DadhE, Fig. S2), the only other two pathways that are essential for glycerol fermentation. This directly results from coupling dictated by the PEP dependence of E. coli DhaKLM, as overexpression of ATP-dependent CfDhaKL (which is independent of PEP) in conjunction with GldA restored glycerol consumption in this background whereas the overexpression of PEP-dependent DhaKLM and GldA did not (Fig. 5D). This result also implies that despite the detrimental impact of FP bypass deletion (Fig. 4C), flux through this pathway appears to be insufficient to serve as the PEP-independent dissimilation pathway under these conditions. Interestingly, overexpression of CfDhaKL and E. coli GldA here also resulted in significant levels of 1,2-PDO production (0.12 AE 0.04 g L À1 ) indicating that decreasing the PEP dependence of glycerol dissimilation increases the availability of DHAP for 1,2-PDO production and this route becomes significant in enabling redox balance at alkaline conditions as well. Intriguingly, we also found that at an acidic pH (6.5), a condition under which glpK deletion had a less pronounced impact on glycerol consumption, the deletion of ppsA alone resulted in a significant reduction in glycerol consumption compared with BW25113 that was not observed at alkaline pH (Fig. S1). This further demonstrates that while GlpK-GlpABC provides a PEP-independent glycerol dissimilation pathway, its coupling to HYD activity and the associated pH dependence limits its effective operation in the absence of external electron acceptors. Overall, these results provide strong evidence that operation of the PEP-independent GlpK-GlpABC pathway and PEP synthetase activity are the only two ways for E. coli to compensate for the stoichiometric constraints created by the PEP dependency of GldA-DhaKLM.
The above PEP limitations may also have implications for the operation of PEP-carboxylating enzymes. Synthesis of the required 4-carbon intermediate oxaloacetate (OAA) and associated tricarboxylic acid cycle intermediates (e.g. aketoglutarate) can proceed from either PEP or pyruvate carboxylation (Fig. 5A). PEP carboxylase (Ppc, encoded by ppc) is viewed as the primary anaplerotic route with PEP carboxykinase (Pck, encoded by pckA) and malic enzymes (encoded by maeA and maeB) primarily gluconeogenic enzymes (Sauer and Eikmanns, 2005). These roles have been established for aerobic glycerol utilization, with ppc an essential gene (Joyce et al., 2006;Martinez-Gomez et al., 2012). In contrast, we found that under alkaline fermentative conditions pckA deletion had a pronounced impact on cell growth and glycerol consumption while Dppc had a negligible impact (Fig. 5C). Deletion of maeA or maeB also did not have a significant impact, further indicating a previously unknown anaplerotic role for Pck under these conditions (Fig. 5C). To confirm this role was specific to fermentative conditions, ppc and pckA deletion strains were also tested under aerobic conditions. In agreement with previous findings, glycerol utilization was abolished in the Dppc strain while the DpckA strain exhibited glycerol consumption levels similar to BW25113 (Fig. S3).
Since the Pck-catalysed carboxylation of PEP generates ATP via substrate-level phosphorylation, this enzyme provides the cells with the ability to increase ATP yield during glycerol fermentation. In addition, Pck has a much higher affinity for PEP (K m = 0.07 mM; Krebs and Bridger, 1980) than Ppc (K m values ranging from 25 mM to 0.19 mM in the absence and presence of activators, respectively; Izui et al., 1983). The K m of Pck towards PEP is also very similar to that of DhaKLM (K m for PEP of 0.045 mM; Gutknecht et al., 2001), and hence, it is able to compete with this essential glycerol dissimilation pathway. Finally, while Pck is subject to allosteric inhibition by PEP (K i = 0.06 mM) and ATP (K i = 0.04 mM; Krebs and Bridger, 1980), both of these inhibitors are expected to be present at low concentrations during glycerol fermentation.
The affinity of Ppc (K m = 0.10 mM; Kai et al., 1999) and PckA (K m = 13 mM; Krebs and Bridger, 1980) for bicarbonate may also play a role in our findings, especially considering the closed fermentation system employed for anaerobic conditions where the initial and evolved CO 2 accumulation may enable Pck to overcome its much lower affinity compared with Ppc. The fourfold increase in formate concentration seen in DhycE and DfdhF strains implies significant amounts of CO 2 evolution during fermentations with an intact FHL complex.

Formation of 6-carbon precursor metabolites from 3-carbon intermediates
Generation of 6-C biosynthetic precursors (e.g. F6-P) from 3-C intermediates during glycerol metabolism requires the operation of both glycolytic and gluconeogenic pathways. Under respiratory conditions, 6-C intermediates are generated via condensation of DHAP and glyceraldehyde-3-phosphate to form F1,6-BP (catalysed by fructose-bisphosphate aldolase) and subsequent conversion to F6-P via fructose-1,6-bisphosphatase (FBPases; Fig. 6A). While E. coli encodes multiple FBPases, including GlpX which is part of the glpFKX operon, previous reports have demonstrated that only Fbp (encoded by fbp) is required during respiratory growth on glycerol (Donahue et al., 2000). However, our in silico modelling suggests that the FP bypass could negate the requirement for FBPase activity during fermentative glycerol utilization (Fig. 3). Only in the case of complete FP bypass removal was flux to F6-P via fructosebisphosphate aldolase and FBPase predicted (see previous section). Furthermore, even if the FP bypass appears incapable of serving as the PEP-decoupling pathway in the absence of GlpK-GlpABC in vivo, the impact observed in the DfsaA DfsaB strain (Fig. 4C) could be explained by its ability to generate F6-P (and other biosynthetic precursors derived from F6-P).  Given this dichotomy, we evaluated the impact of deleting the FBPases encoded by fbp and glpX, along with yggF and ybhA, encoding 2 additional enzymes shown to have FBPase activity (Brown et al., 2009). As seen in Fig. 6B, not only did fbp deletion have a major impact on cell growth, DglpX and DyggF genotypes also exhibited lower cell growth than BW25113. We further constructed double and triple mutants combining deletions of fbp, glpX, and yggF and found that regardless of the combination of FBPase deletions, cell growth remained at similarly low levels to those observed with a single fbp or glpX deletion (Fig. 6B). With previous reports indicating that only Fbp is required for respiratory glycerol utilization, the individual FBPase deletion strains were also tested under aerobic conditions. While as expected the fbp deletion abolished glycerol consumption, the DglpX and DyggF strains show no significant difference in glycerol consumption compared with BW25113 (Fig. 6C). These results demonstrate that, unlike respiratory conditions, fermentative metabolism of glycerol requires multiple FBPases, with both GlpX and Fbp playing critical roles.

Revised metabolic model for the fermentative utilization of glycerol in Escherichia coli
Under fermentative conditions, redox balance must be maintained by regenerating oxidized electron carriers using internal electron acceptors. Given the high degree of reduction of glycerol (j = 4.67) compared to biomass (j = 4.3), fermentative growth on this substrate results in the next generation of reducing equivalents. Therefore, the ability for microorganisms, including E. coli, to utilize glycerol in the absence of external electron acceptors has been linked to the production of organic compounds with a degree of reduction higher than glycerol (e.g. 1,3-PDO, 1.2-PDO: j = 5.33), which represents an electron sink and enables redox balance. Here, we demonstrate that E. coli possesses an additional means of achieving overall redox balance by reducing protons to H 2 through reversible hydrogenase(s). This H 2 -enabled glycerol fermentation mode in E. coli is inherently limited by physiological constraints on the reversibility of hydrogenases and hence restricted to alkaline pH. Under acidic conditions, 1,2-PDO production remains a critical mechanism for achieving redox balance. Furthermore, the H 2 evolution mode is enabled by coupling of hydrogenases to G3PDH (GlpABC) and hence to glycerol dissimilation through the GlpK-GlpABC pathway. However, this respiratory pathway is unable to support glycerol utilization under fermentative conditions in the absence of the GldA-DHAK route. These aspects, combined with additional stoichiometric and ATP generation constraints during fermentative metabolism of glycerol, result in system-wide responses in the metabolic network of E. coli dependent on the mode redox balance is achieved.
Both the alkaline, H 2 -enabled or acidic, 1,2-PDOenabled modes of glycerol fermentation require glycerol dissimilation through the NADH-generating GldA-DhaKLM route (Fig. 7). From a stoichiometric standpoint, the coupling between glycerol dissimilation and pyruvate synthesis creates a stoichiometric constraint that limits the availability of PEP and other precursor metabolites (e.g. F6-P, OAA and TCA cycle intermediates). To address these constraints, PEP must be either generated from a different source (e.g. pyruvate) or the PEP dependence decreased. Under alkaline conditions in which hydrogenase reversibility enables a portion of glycerol oxidation to be metabolized via GlpK-GlpABC (Fig. 7, purple shading), this route serves as a PEPindependent dissimilation pathway that decouples glycerol utilization from PEP consumption. However, when hydrogenase reversibility is limited, such as under acidic conditions, the use of the ATP intensive PEP synthetase to regenerate PEP is required (Fig. 7, orange shading).
While the use of an NADH-generating PEP-independent glycerol dissimilation pathway, such as an ATP-dependent DHA kinase in conjunction with GldA, alleviates this strain on PEP, despite its potential to natively serve this role the FP bypass in E. coli is unable to support glycerol fermentation under the conditions tested here. Combined with the limitations of the GlpK-GlpABC serving as the PEPindependent glycerol dissimilation pathway, the strain on PEP and implied low intracellular concentrations has profound system-wide impacts, including dictating the importance of several enzymes that under fermentative conditions supplement or replace those essential under respiratory conditions. For example, the higher affinity for PEP of Pck compared with Ppc may underlie the importance of the former during fermentative glycerol utilization. Regulatory aspects may also play a role here, as allosteric inhibition of Pck by PEP is likely alleviated. In this context, the tight regulation between glycolysis and gluconeogenesis, specifically the regulatory functions of PEP and F1,6-BP, also appears to directly contribute to several of our findings. For example, F1,6-BP activates Ppc (Donahue et al., 2000) and notably is a noncompetitive inhibitor of GlpK with high intracellular F1,6-BP levels potentially decreasing GlpK activity (Thorner and Paulus, 1973). An implied low F1,6-BP concentration could also contribute, in part, to the lack of a role of Ppc and explain the prominent role of pyruvate kinase II (PykA; Fig. 5C), as pyruvate kinase I (PykF) is strongly activated by F1,6-BP (Malcovati and Valentini, 1982).
Given its significant regulatory role, the combined action of two FBPases might be required to maintain F1,6-BP (which is the substrate of FBPases) homeostasis especially as PEP has been proposed as the physiological  (Hines et al., 2006). Fbp is extremely sensitive to AMP inhibition (50% inhibition by 15 lM AMP; Babul and Guixe, 1983); however, this inhibition is alleviated by PEP. While GlpX and YggF are less catalytically efficient enzymes than Fbp (Fig. 6A), neither have shown such sensitive inhibition by any compound. Thus, while the PEP-independent glycerol dissimilation pathway utilized under respiratory conditions (GlpK-GlpD) may help alleviate inhibition of Fbp (due to increased PEP levels), under fermentative conditions with PEP-dependent glycerol dissimilation and implied low intracellular PEP levels, Fbp may not be fully activated and F1,6-BP conversion requires a combination of Fbp, GlpX and to a lesser extent YggF. The requirement to generate ATP and maintain redox balance in the absence of external electron acceptors with a carbon source as reduced as glycerol also provides a rationale for E. coli to operate under the H 2 -enabled mode when conditions permit, despite the challenges of coupling to the GlpK-GlpABC pathway (i.e. electron donor and pathway regulation). By channelling electrons to hydrogenases, excess reducing equivalents can be consumed through H 2 evolution, which compared to 1,2-PDO production: (1) provides a more ATP efficient sink for reducing equivalents, (2) circumvents the requirement of channelling an intermediate upstream of PEP (DHAP) to 1,2-PDO which further strains PEP availability and (3) enables increased carbon flux to ethanol synthesis further improving ATP yields. This increased ATP yield may underlie our findings that suggest H 2 production from protons provides a more efficient route to achieve redox balance that facilitates higher growth rates. Additional experiments towards better understanding the conditions required and ability for H 2 evolution to support glycerol fermentation in E. coli can provide new insights on this metabolic process. This may provide a basis for identifying other organisms with this latent capability to extend the range of organisms capable of fermentative glycerol utilization beyond those with the established propanediol dependent models.

Model implementation, simulation and curation
The genome-scale model (GEM) iJO1366 for E. coli, which includes 1366 genes, 2251 reactions and 1136 unique metabolites, was used as starting point . iJO1366 was further modified to most accurately reflect important enzymes, pathways and mechanisms involved with the current knowledge of fermentative glycerol utilization as described in Appendix S1. External conditions were set to represent minimal medium under anaerobic conditions (no external electron acceptor) as defined elsewhere (Feist et al., 2007;Orth et al., 2011), and glycerol was set as the only limiting nutrient with an uptake rate value of À20 mmol gCDW À1 h À1 . The nongrowth associated ATP maintenance (ATPM) is considered as 3.15 mmol gCDW À1 h À1 . Flux Balance Analysis (FBA) and Flux Variability Analysis (FVA) were used to predict flux distribution for optimal cell growth and solution spaces, respectively. All simulations were performed using COBRA toolbox v2.0 (Schellenberger et al., 2011) in Matlab v7.11.0 (R2010b, The Mathworks, Natick, MA, USA).

Strains, plasmids and genetic methods
Escherichia coli K12 strain BW25113 and single-gene knockout mutants were obtained from the National BioResource Project (NIG, Mishima, Japan; Baba et al., 2006). Strains containing multiple gene deletions were constructed by P1 phage transduction using the single-gene  Table S3. All mutations, including single deletion mutants obtained from NIG, were confirmed by polymerase chain reaction using appropriate primers. Plasmids pZSKLMgldA and pZSKLcfgldA used in this study have been previously described Clomburg and Gonzalez, 2011). Molecular biology techniques were performed with standard methods (Miller, 1972;Sambrook et al., 2001) or by manufacturer protocol. Strains were kept in 32.5% glycerol stocks at À80°C. Plates were prepared using LB medium containing 1.5% agar with appropriate antibiotics (50 µg ml À1 kanamycin, 34 µg ml À1 chloramphenicol).

Culture medium and cultivation conditions
The minimal medium designed by Neidhardt and Bloch (1974) with 1.32 mM Na 2 HPO 4 in place of K 2 HPO 4 supplemented with 10 g L À1 glycerol and 2 g L À1 tryptone (pH 7.2) was used unless otherwise stated. The inclusion of 2 g L À1 tryptone has been found to be required for fermentative glycerol utilization but does not compromise the fermentative nature of glycerol utilization Murarka et al., 2008). The redox indicator resazurin (1 mg L À1 ; Ferguson and Cummins, 1978) and reducing agent dithiothreitol (DTT, 1 mM) were included as a means to both detect (resazurin is pink in the presence of oxygen) and eliminate oxygen from the media ensuring strict anaerobic conditions. Chloramphenicol (12.5 µg ml À1 ) and anhydrotetracycline (100 ng ml À1 ) were included for strains transformed with pZS series vectors. Chemicals were obtained from Fisher Scientific (Pittsburgh, PA, USA) and Sigma-Aldrich (St. Louis, MO, USA). Anaerobic fermentations were conducted in 17-ml Hungate tubes (Bellco Glass, Vineland, NJ, USA). Prior to inoculation, oxygen in the medium was removed through the addition of reducing agent (DTT) and storage in an oxygen-free environment (90% N 2 , 5% CO 2 , 5% H 2 ,) provided within a BACTRON I Anaerobic Chamber (Sheldon Manufacturing, Cornelius, OR, USA) until the medium was void of colour. A single colony was used to inoculate 5 ml of media in each tube, which were then incubated at 37°C with rotation in a LabQuake rotator (Fisher Scientific, Pittsburgh, PA, USA) for 96 h in an Isotemp Incubator (Fisher Scientific). Aerobic shake flask experiments were conducted in 25 ml Erlenmeyer flasks (Corning, Corning, NY, USA) using the above defined media without the inclusion of tryptone, resazurin and DTT. Pre-cultures for these fermentations were prepared by inoculating 5 ml of LB media with a single colony in Hungate tubes and incubating these tubes with rotation at 37°C until an optical density of 0.4 was reached. An appropriate volume of actively growing pre-culture was centrifuged, washed twice with minimal media and used to inoculate 5 ml of media in each flask, with a target initial optical density of 0.05. Flasks were capped with a foam plug and incubated at 37°C and 140 r.p.m. in a Lab Companion SI-600 rotary shaker (Jeiotech, Seoul, Korea) for 24 h.

Analytical methods and enzyme assays
Optical density was measured at 600 nm using a WPA BioWave CO8000 Cell Density Meter (Biochrom, Cambridge, UK). After centrifugation of the sample, the supernatant was stored at À20°C for high-performance liquid chromatography (HPLC) analysis. Quantification of glycerol, organic acids, ethanol and other compounds was determined via HPLC using a Shimadzu Prominence SIL 20 system (Shimadzu Scientific Instruments, Columbia, MD, USA) equipped with a refractive index detector and a HPX-87H organic acid column (Bio-Rad, Hercules, CA, USA) with operating conditions to optimize peak separation (0.3 ml min À1 flowrate, 30 mM H 2 SO 4 mobile phase, column temperature 42°C). See fermentation substrate, pH and the F0F1-ATPase. Crit Rev Biochem Mol Biol 47: 236-249. Trchounian, K., and Trchounian, A. (2009) Hydrogenase 2 is most and hydrogenase 1 is less responsible for H 2 production by Escherichia coli under glycerol fermentation at neutral and slightly alkaline pH. Int J Hydrogen Energy 34: 8839-8845. Yazdani, S.S., and  Engineering Escherichia coli for the efficient conversion of glycerol to ethanol and co-products. Metab Eng 10: 340-351. Zhang, X.L., Shanmugam, K.T., and Ingram, L.O. (2010) Fermentation of glycerol to succinate by metabolically engineered strains of Escherichia coli. Appl Environ Microbiol 76: 2397-2401.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.