Thermodynamics shapes the in vivo enzyme burden of glycolytic pathways

ABSTRACT

Thermodynamically constrained reactions and pathways are hypothesized to impose greater protein demands on cells, requiring higher enzyme amounts to sustain a given flux compared to those with stronger thermodynamics. To test this, we quantified the absolute concentrations of glycolytic enzymes in three bacterial species—Zymomonas mobilis, Escherichia coli, and Clostridium thermocellum—which employ distinct glycolytic pathways with varying thermodynamic driving forces. By integrating enzyme concentration data with corresponding in vivo metabolic fluxes and ΔG measurements, we found that the highly favorable Entner-Doudoroff pathway in Z. mobilis requires only one-fourth the amount of enzymatic protein to sustain the same flux as the thermodynamically constrained pyrophosphate-dependent glycolytic pathway in C. thermocellum, with the Embden-Meyerhof-Parnas pathway in E. coli exhibiting intermediate thermodynamic favorability and enzyme demand. Across all three pathways, early reactions with stronger thermodynamic driving forces generally required lower enzyme investment than later, less favorable steps. Additionally, reflecting differences in glycolytic strategies, the highly reversible ethanol fermentation pathway in C. thermocellum requires 10-fold more protein to maintain the same flux as the irreversible, forward-driven ethanol fermentation pathway in Z. mobilis. Thus, protein investment across glycolytic pathways reflects differences in their thermodynamic favorability.

IMPORTANCE

Cells regulate metabolic fluxes to balance energy production, biosynthesis, and the efficient use of limited resources, including the finite capacity for synthesizing and maintaining metabolic enzymes. Here, we present in vivo evidence that strongly thermodynamically favorable metabolic pathways require significantly fewer enzyme resources to sustain a given flux compared to less thermodynamically favorable pathways. These findings underscore the connection between pathway thermodynamics, resource allocation, and enzyme burden, providing valuable insights for metabolic engineering strategies aimed at optimizing pathways for high flux with minimal protein cost.

INTRODUCTION

Metabolic flux is a primary driver of cellular physiology. Cells regulate fluxes to meet energy and biosynthetic demands while efficiently managing limited resources, including the finite capacity to synthesize and maintain metabolic enzymes (1–4). Multiple factors influence metabolic flux within cells, including enzyme abundance, catalytic constant (k_cat), active site saturation (governed by K_m values and substrate concentrations), and regulatory mechanisms such as allosteric inhibition and post-translational modifications (5–10). A less commonly appreciated but critical factor is the energetics of biochemical reactions, typically quantified as the change in Gibbs free energy (ΔG). This thermodynamic parameter not only determines reaction directionality but also imposes intrinsic constraints on flux (11, 12). Specifically, the ratio of forward (J⁺) to reverse (J^-) fluxes of a reaction relates to its ΔG via the following equation:

$${\displaystyle {\displaystyle \mathrm{\Delta }G=-RT\mathrm{l}\mathrm{n}\left(J^{+}/J^{-}\right)}}$$

where R is the gas constant and T is the absolute temperature in kelvin. This equation, known as the flux-force relationship, reveals the interdependence between a reaction’s thermodynamic driving force, net flux, and enzyme cost (5, 13–15). Reactions far from thermodynamic equilibrium (i.e., with a large negative ΔG) have forward fluxes that greatly exceed reverse fluxes, resulting in a high net flux (J^net = J⁺ − J⁻) and efficient enzyme utilization, as most enzyme activity is directed toward the forward reaction. In contrast, reactions operating near equilibrium have nearly equal forward and reverse fluxes (J⁺ ≈ J⁻), which leads to inefficient enzyme utilization and a reduced net flux. Consequently, thermodynamically constrained reactions incur higher enzyme costs—defined as the amount of enzyme required per unit flux—to sustain the same net flux compared to reactions with stronger thermodynamic driving forces (5, 16).

Building on these principles, a previous computational study investigated the interdependence between pathway thermodynamics, enzyme cost, and energy output (i.e., ATP production) in the two most prevalent glycolytic pathways used by bacteria: the Embden-Meyerhof-Parnas (EMP) and the Entner-Doudoroff (ED) pathways (17). By combining computationally estimated free energies with model-derived protein cost estimates, this study showed that the ED pathway is significantly less thermodynamically constrained than the EMP pathway and predicted that the ED pathway requires three to five times less enzymatic protein to sustain the same glycolytic flux as the EMP pathway. However, this reduction in enzyme cost, driven by greater thermodynamic favorability, comes at the expense of a lower ATP yield per glucose (17). Subsequent computational studies have further supported the hypothesis that thermodynamically constrained reactions and pathways impose greater protein demands on the cell as a consequence of large reverse fluxes and inefficient enzyme utilization (10, 16, 18).

While these computational predictions are compelling, they remain to be experimentally validated. Testing these hypotheses in vivo requires simultaneous measurements of metabolic fluxes and protein levels in organisms that utilize pathways with distinct thermodynamic profiles (17). In this study, we address this gap by quantifying the absolute concentrations of glycolytic enzymes in three bacterial species—Zymomonas mobilis, Escherichia coli, and Clostridium thermocellum—which employ distinct glycolytic pathways with varying thermodynamic driving forces. By integrating enzyme concentration data with corresponding in vivo metabolic fluxes and intracellular ΔG measurements, we provide strong experimental evidence that thermodynamic driving force is a key parameter influencing the in vivo enzyme burden of metabolic reactions and pathways.

RESULTS

Experimental system: energetics and flux of three distinct glycolytic pathways

We investigated the in vivo relationship between pathway thermodynamics, metabolic fluxes, and enzyme concentrations across the glycolytic pathways of three different bacteria: the ethanologenic Z. mobilis, the cellulolytic and ethanologenic C. thermocellum, and the model organism E. coli. These bacteria metabolize glucose to pyruvate via distinct glycolytic routes, which vary in key enzymatic steps, energy yield (i.e., ATP/GTP output), thermodynamics, and flux (Fig. 1). Z. mobilis exclusively relies on the ED pathway for glucose catabolism (Fig. 1A) (19–21). In contrast, C. thermocellum employs a pyrophosphate (PPi)-dependent EMP (PPi-EMP) pathway, which features a PPi-phosphofructokinase (PPi-Pfk) that utilizes PPi, rather than ATP, as a phosphate donor to convert fructose 6-phosphate (F6P) to fructose 1,6-bisphosphate (FBP) (22–25). Additionally, C. thermocellum lacks a pyruvate kinase (Pyk) to convert phosphoenolpyruvate (PEP) to pyruvate. Instead, it produces pyruvate via a PPi-dependent dikinase (Ppdk), and can also generate pyruvate via the “malate shunt,” which involves phosphoenolpyruvate carboxykinase (Pepck), malate dehydrogenase (Mdh), and malic enzyme (Me) (Fig. 1A) (22, 26, 27). Finally, E. coli primarily uses the EMP pathway to convert glucose into pyruvate, utilizing the ED pathway only under specific conditions (such as growth on gluconate or during gut colonization [28, 29]) or in parallel with the EMP pathway during carbon or nitrogen upshifts to promote rapid growth acceleration (30).

Fig 1.

Glycolytic pathways and their energetics. (A) The ED pathway in Z. mobilis (blue arrows), the EMP pathway in E. coli (pink arrows), and the PPi-EMP pathway in C. thermocellum (green arrows) utilize distinct enzymes at various steps to convert glucose into pyruvate (PYR). Reactions depicted with black arrows are common to all three pathways. In E. coli, glucose is simultaneously imported and converted to glucose 6-phosphate (G6P) using PEP as the phosphate donor via the phosphotransferase system (PTS). Enzymes are depicted as ovals, and metabolites are shown as rectangles. (B) The cumulative drop in ΔG for the glycolytic pathways in Z. mobilis (blue), E. coli (pink), and C. thermocellum (green). ΔG data are a combination of previous experimental measurements and computationally estimated values constrained by in vivo metabolite concentrations obtained in this study (Materials and Methods). Ninety-five percent confidence intervals for ΔG values are presented in Table S1, as error bars are too small to be visible in this representation. The ΔG for the pyrophosphate-dependent pyruvate dikinase (PPDK) and malate shunt (MS) (i.e., PEP carboxykinase [PEPCK], malate [MAL] dehydrogenase [MDH], malic enzyme [ME]) in C. thermocellum represents the combined reaction (Table S1; Materials and Methods). (C) Glucose consumption rates for each bacterium. The glucose consumption rate for C. thermocellum is presented as twice the calculated cellobiose uptake rate, since each molecule of cellobiose consists of two glucose moieties. Glucose uptake rates were calculated in cells grown aerobically (E. coli) or anaerobically (Z. mobilis and C. thermocellum) in minimal media (Materials and Methods). Data represent the averages of three to four biological replicates. Error bars show ±standard deviation. See Table S2 for glucose consumption rate data. Metabolite abbreviations: 6-phosphogluconate (6PG), glucose 6-phosphate (G6P), glucose 1-phosphate (G1P), 2-keto-3-deoxy-6-phosphogluconate (KDPG), fructose 6-phosphate (F6P), fructose 1,6-bisphosphate (FBP), glyceraldehyde 3-phosphate (GAP), dihydroxyacetone phosphate (DHAP), 1,3-bisphosphoglycerate (1,3-BPG), 3-phosphoglycerate (3-PG), 2-phosphoglycerate (2-PG), phosphoenolpyruvate (PEP), pyruvate (PYR), oxaloacetate (OAA), malate (MAL). Enzyme abbreviations: cellobiose phosphorylase (Cbp), glucokinase (Glk), G6P dehydrogenase (G6pdh), phosphogluconolactonase (Pgl), phosphoglucomutase (Pgmt), 6PG dehydratase (Edd), phosphoglucose isomerase (Pgi), phosphofructokinase (Pfk), KDPG aldolase (Eda), FBP aldolase (Fba), triose phosphate isomerase (Tpi), GAP dehydrogenase (Gapdh), phosphoglycerate kinase (Pgk), phosphoglycerate mutase (Pgm), enolase (Eno), PYR kinase (Pyk), PYR phosphate dikinase (Ppdk), PEP carboxykinase (Pepck), MAL dehydrogenase (Mdh), malic enzyme (Me).

A notable difference between E. coli, Z. mobilis, and C. thermocellum lies in glucose uptake and its conversion to glucose 6-phosphate (G6P). In E. coli, glucose import into the cytoplasm is coupled with its phosphorylation to G6P via the phosphotransferase system (PTS), which uses PEP as the phosphate donor and produces pyruvate as a byproduct. In contrast, Z. mobilis and C. thermocellum phosphorylate glucose or cellobiose, respectively, only after these sugars enter the cytoplasm.

The overall thermodynamic favorability and energy output of these glycolytic pathways differ greatly. In vivo ΔG measurements obtained from ¹³C and ²H metabolic flux analyses (MFA) coupled with ΔG computational estimates (Materials and Methods) show that the ED pathway in Z. mobilis is approximately three times more thermodynamically favorable than the PPi-EMP pathway in C. thermocellum and nearly twice as favorable as the EMP pathway in E. coli (Fig. 1B; Table S1). Notably, the high thermodynamic favorability of the ED pathway in Z. mobilis correlates with an in vivo glycolytic rate that is approximately sixfold higher than that of C. thermocellum and fivefold higher than that of E. coli (Fig. 1C; Table S2).

Protein resources are unevenly allocated across glycolysis

In Z. mobilis, each reaction of the ED glycolytic pathway is catalyzed by a single enzyme (31). In contrast, E. coli has multiple isoenzymes for several glycolytic reactions, including phosphofructokinase (PFK), fructose 1,6-bisphosphate aldolase (FBA), glyceraldehyde 3-phosphate dehydrogenase (GAPDH), phosphoglycerate mutase (PGM), and PYK (32–40). Similarly, C. thermocellum possesses multiple isoenzymes for FBA and PGM (Table S3) (41).

We used shotgun proteomics to identify the predominant glycolytic enzymes in each bacterium (Table S3). Z. mobilis and C. thermocellum were grown anaerobically, while E. coli was cultured under aerobic conditions. Z. mobilis and E. coli were grown using glucose as the sole carbon source, whereas C. thermocellum was grown on cellobiose (Materials and Methods). All isoenzymes with comparable expression levels, as determined by intensity-based absolute quantification (iBAQ) values (42, 43) from shotgun proteomics, were selected for direct quantitation using the absolute quantification (AQUA) method (Table S3). For each protein, two to eight isotopically labeled reference peptides were chosen based on shotgun proteomics data (Table S5) (44, 45). Isoenzymes with markedly lower expression (e.g., >15-fold difference) compared to the predominant isoenzyme were excluded from AQUA quantification (Table S3).

ED pathway enzymes in E. coli were also excluded from direct absolute quantitation using AQUA as previous MFA studies have shown negligible carbon flux (0.2%–1%) through 6-phosphogluconate dehydratase (EDD) and 2-dehydro-3-deoxyphosphogluconate aldolase (EDA) when E. coli is grown aerobically on glucose (46, 47). Similarly, although C. thermocellum possesses multiple ATP/GTP-dependent PFKs in addition to PPi-Pfk (41), enzyme assays in cell extracts revealed no ATP/GTP-PFK activity (22, 48). Consistent with these findings and other previous studies (22–24), PPi-PFK was the most highly expressed PFK isozyme in our C. thermocellum cells, leading us to exclude ATP/GTP-dependent PFKs from direct AQUA quantitation (Table S3).

Using AQUA, we determined the absolute intracellular concentrations of 13, 16, and 15 glycolytic enzymes in Z. mobilis, C. thermocellum, and E. coli, respectively (Table 1; Tables S4, S21 to S23). For Z. mobilis and C. thermocellum, both of which produce ethanol as their primary fermentation product, we also used AQUA to quantify the absolute concentrations of their ethanol pathway enzymes (Table 1; Tables S4, S21 and S23).

TABLE 1.

Absolute intracellular concentrations of glycolytic and fermentation enzymes in Z. mobilis, C. thermocellum, and E. coli quantified via AQUA

Organism	Locus tag	Protein	Abbr.	Avg. (fg/cell)	SD
Z. mobilis	ZMO0366	Glucose-facilitated diffusion protein	Glf	0.403	0.016
Z. mobilis	ZMO0369	Glucokinase	Glk	0.498	0.068
Z. mobilis	ZMO0367	Glucose 6-phosphate dehydrogenase	G6pdh	1.417	0.366
Z. mobilis	ZMO1478	6-Phosphogluconolactonase	Pgl	0.111	0.037
Z. mobilis	ZMO0368	6-Phosphogluconate dehydratase	Edd	1.625	0.144
Z. mobilis	ZMO0997	2-Dehydro-3-deoxyphosphogluconate aldolase	Eda	0.611	0.3
Z. mobilis	ZMO0177	Glyceraldehyde 3-phosphate dehydrogenase	Gapdh	4.498	0.84
Z. mobilis	ZMO0178	Phosphoglycerate kinase	Pgk	1.465	0.2
Z. mobilis	ZMO1240	Phosphoglycerate mutase	Pgm	1.003	0.059
Z. mobilis	ZMO1608	Enolase	Eno	2.899	0.11
Z. mobilis	ZMO0152	Pyruvate kinase	Pyk	4.338	0.273
Z. mobilis	ZMO1212	Phosphoglucose isomerase	Pgi	0.449	0.175
Z. mobilis	ZMO0179	Fructose bisphosphate aldolase	Fba	0.048	0.018
Z. mobilis	ZMO0465	Triose phosphate isomerase	Tpi	0.033	0.014
Z. mobilis	ZMO1360	Pyruvate decarboxylase	Pdc	6.662	0.477
Z. mobilis	ZMO1236	Alcohol dehydrogenase I	AdhA	0.067	0.013
Z. mobilis	ZMO1596	Alcohol dehydrogenase II	AdhB	1.643	0.642
E. coli	b1101	Phosphotransferase enzyme IIBC component	PtsG	0.264	0.057
E. coli	b2388	Glucokinase	Glk	0.043	0.008
E. coli	b4025	Phosphoglucose isomerase	Pgi	0.216	0.032
E. coli	b3916	6-Phosphofructokinase I	PfkA	0.112	0.009
E. coli	b1723	6-Phosphofructokinase II	PfkB	0.026	0.009
E. coli	b2097	Fructose bisphosphate aldolase class I	FbaB	0.042	0.008
E. coli	b2925	Fructose bisphosphate aldolase class II	FbaA	0.227	0.093
E. coli	b3919	Triose phosphate isomerase	Tpi	0.207	0.018
E. coli	b1779	Glyceraldehyde 3-phosphate dehydrogenase	Gapdh	1.946	0.616
E. coli	b2926	Phosphoglycerate kinase	Pgk	0.831	0.054
E. coli	b3612	Phosphoglycerate mutase, 2,3-bisphosphoglycerate independent	GpmM	0.248	0.03
E. coli	b0755	Phosphoglycerate mutase, 2,3-bisphosphoglycerate dependent	GpmA	0.219	0.084
E. coli	b2779	Enolase	Eno	1.072	0.088
E. coli	b1676	Pyruvate kinase I	PykF	0.354	0.033
E. coli	b1854	Pyruvate kinase II	PykA	0.069	0.009
C. thermocellum	Clo1313_1954	Cellobiose phosphorylase	Cbp	0.254	0.052
C. thermocellum	Clo1313_0993	Phosphoglucomutase	Pgmt	0.088	0.016
C. thermocellum	Clo1313_0489	Glucokinase	Glk	0.057	0.01
C. thermocellum	Clo1313_2015	Phosphoglucose isomerase	Pgi	0.226	0.056
C. thermocellum	Clo1313_1876	6-Phosphofructokinase, PPi dependent	Pfk1	1.333	0.431
C. thermocellum	Clo1313_1875	Fructose bisphosphate aldolase class II	Fba1	0.381	0.055
C. thermocellum	Clo1313_2093	Triose phosphate isomerase	Tpi	0.505	0.047
C. thermocellum	Clo1313_2095	Glyceraldehyde 3-phosphate dehydrogenase	Gapdh	2.133	0.445
C. thermocellum	Clo1313_2094	Phosphoglycerate kinase	Pgk	1.285	0.207
C. thermocellum	Clo1313_2092	Phosphoglycerate mutase, 2,3-bisphosphoglycerate independent	Pgm1	0.217	0.046
C. thermocellum	Clo1313_0966	Phosphoglycerate mutase, 2,3-bisphosphoglycerate independent	Pgm2	0.018	0.003
C. thermocellum	Clo1313_2090	Enolase	Eno	0.917	0.476
C. thermocellum	Clo1313_0949	Pyruvate phosphate dikinase	Ppdk	0.308	0.027
C. thermocellum	Clo1313_0415	Phosphoenolpyruvate carboxykinase	Pepck	1.206	0.477
C. thermocellum	Clo1313_1878	Malate dehydrogenase	Mdh	0.491	0.1
C. thermocellum	Clo1313_1879	Malic enzyme	Me	0.542	0.182
C. thermocellum	Clo1313_0022	Pyruvate ferredoxin oxidoreductase I, alpha domain	Pfor1-α	0.676	0.044
C. thermocellum	Clo1313_1798	Bifunctional acetaldehyde and alcohol dehydrogenase	Aldh/Adh	2.244	0.299

To estimate the absolute concentrations of proteins not quantified via AQUA, we constructed a model that relates AQUA-derived absolute protein measurements to their corresponding summed precursor peptide ion intensities (iBAQ values) (42, 43). This approach yielded strong correlations (R² ≈ 0.87–0.94) between AQUA absolute protein measurements and their respective iBAQ values across all three bacteria (Fig. S1), with low normalized root mean square error (NRMSE) (Table S4), as determined via leave-one-out cross-validation (Materials and Methods) (49–51). Using this method, we quantified the absolute concentrations of 1,634, 2,428, and 1,972 proteins, representing 85%, 56%, and 66% of the proteomes of Z. mobilis, E. coli, and C. thermocellum, respectively (Tables S6 to S8).

Figure 2 presents the absolute concentrations of glycolytic enzymes for each bacterium. Z. mobilis has approximately three times more total glycolytic enzyme per cell than E. coli and twice as much as C. thermocellum. Across all three bacteria, protein resources were unevenly distributed within glycolysis, with substantial variation in enzyme abundance at different pathway steps. In Z. mobilis, enzymes catalyzing the upper ED pathway (i.e. reactions from glucose to glyceraldehyde 3-phosphate [GAP]: glucokinase [GLK], G6P dehydrogenase [G6PDH], phosphogluconolactonase [PGL], EDD, and EDA) account for only 23% of the total glycolytic enzyme pool on a mass basis (fg cell⁻¹), while enzymes in the lower part of the pathway (i.e. reactions from GAP to pyruvate: GAPDH, phosphoglycerate kinase [PGK], PGM, enolase [ENO], and PYK) make up the remaining 77%. A similar trend is observed in E. coli, where upper glycolytic enzymes (i.e., GLK, phosphoglucose isomerase [PGI], PFK [PFKA, PFKB], FBA [FBAA, FBAB], triose phosphate isomerase [TPI]) constitute just 16% of the glycolytic enzyme pool, excluding the PTS. Notably, the PTS enzymes themselves represent a major protein investment, comprising 16% of E. coli’s glycolytic enzyme pool. In C. thermocellum, the enzymes performing the lower glycolytic reactions (i.e., GAPDH, PGK, PGM [PGM1, PGM2], ENO, PPDK, PEPCK, MDH, ME) make up a disproportionate 71% of the total glycolytic enzyme pool. Remarkably, Pfk in upper glycolysis accounts for a much larger fraction of the glycolytic enzyme pool in C. thermocellum (13%) compared to E. coli (2%).

Fig 2.

Intracellular concentrations of glycolytic enzymes. Absolute concentrations of glycolytic enzymes, expressed in femtograms (fg) per cell, were quantified in Z. mobilis (blue), E. coli (pink), and C. thermocellum (green). For E. coli, the concentration of the PTS represents the sum of the four component enzymes: PtsG, PtsH, PtsI, and Crr (Table S10). Data represent the average of four biological replicates; error bars show ±standard deviation. Some error bars are too small to be visible in this representation. See Tables S4, S21 to S23 for absolute enzyme concentration data and individual replicate data. Abbreviations: glucokinase (Glk), glucose 6-phosphate dehydrogenase (G6pdh), phosphogluconolactonase (Pgl), 6-phosphogluconate dehydratase (Edd), 2-keto-3-deoxy-6-phosphogluconate aldolase (Eda), glyceraldehyde 3-phosphate dehydrogenase (Gapdh), phosphoglycerate kinase (Pgk), phosphoglycerate mutase (Pgm, GpmM, GpmA, Pgm1, Pgm2), enolase (Eno), pyruvate kinase (Pyk, PykF, PykA), phosphoglucose isomerase (Pgi), fructose 1,6-bisphosphate aldolase (Fba, FbaB, FbaA), triose phosphate isomerase (Tpi), phosphofructokinase (PfkA, PfkB, Pfk), cellobiose phosphorylase (Cbp), phosphoglucomutase (Pgmt), pyruvate phosphate dikinase (Ppdk), phosphoenolpyruvate carboxykinase (Pepck), malate dehydrogenase (Mdh), malic enzyme (Me).

Across all three glycolytic pathways, GAPDH consistently emerged as the most abundant enzyme, representing 24%, 29%, and 21% of the total glycolytic protein pool in Z. mobilis, E. coli, and C. thermocellum, respectively.

Protein cost across glycolytic pathways reflects differences in their thermodynamic favorability

Theoretical and computational analyses predict that thermodynamically constrained reactions in glycolysis incur higher protein costs than those with larger driving forces (5, 17). These studies further suggest that glycolytic pathways with greater overall thermodynamic favorability require less protein compared to those with lower favorability. To investigate the in vivo relationship between pathway thermodynamics, metabolic flux, and enzyme concentration, we normalized the absolute protein concentration of each glycolytic reaction (i.e., the sum of all enzymes and isoenzymes involved) to its respective in vivo flux (Table S9). This approach yielded a metric of protein cost (μg protein/(mmol h⁻¹)) (52–54), enabling comparisons of protein costs across glycolytic reactions and pathways in the three organisms studied (Fig. 3; Table S20).

Fig 3.

In vivo protein costs for glycolytic reactions and pathways. (A) Protein costs, expressed as the amount of protein (μg) required per unit flux (mmol h⁻¹), for glycolytic reactions in Z. mobilis (blue), E. coli (pink), and C. thermocellum (green) (Table S20). The protein cost of the PPDK and the malate shunt (MS) reactions in C. thermocellum were grouped together because intracellular flux measurements for the conversion of PEP to pyruvate (PYR) do not distinguish the amount of flux occurring through each route (Table S9). The protein cost of the PTS in E. coli represents the sum of the four participating enzymes: PtsG, PtsH, PtsI, and Crr. The cumulative drop in ΔG (kJ mol⁻¹) for each glycolytic pathway is also shown. ΔG data are a combination of previous experimental measurements (52–54) and computationally estimated values constrained by in vivo metabolite concentrations obtained in this study (Materials and Methods). Ninety-five percent confidence intervals for ΔG values are presented in Table S1, as error bars are too small to be visible in this representation. (B) The total protein cost, expressed as the amount of protein (μg) required per unit flux (mmol glucose h⁻¹), for core glycolysis (G6P to PYR) and glycolysis including periplasmic sugar uptake in each bacterium. Periplasmic sugar uptake enzymes include the glucose-facilitated diffusion (Glf) protein in Z. mobilis, the four PTS enzymes in E. coli, and the four enzymes (CbpB, MsdB1, MsdB2, and NbdB) that comprise transporter B in C. thermocellum (Table S10). Protein costs for each glycolytic reaction (panel A) were calculated by normalizing the sum of all participating enzymes and isoenzymes to the intracellular flux of the reaction. Similarly, the protein cost for each glycolytic pathway (panel B) represents the sum of all glycolytic enzymes and isoenzymes, normalized to the glucose uptake rate of the corresponding bacterium. (Materials and Methods). For all graphs, data represent the averages of four biological replicates (Tables S21 to S23). Error bars show ±standard deviation (protein costs) or 95% confidence intervals (ΔG values). Some error bars are too small to be visible in this representation. See Tables S1, S2, S9, and S20 to S23 for ΔG values, glucose uptake, flux data, calculated protein costs, and individual replicate data for absolute concentrations, respectively. Abbreviations: glucokinase (GLK), glucose 6-phosphate dehydrogenase (G6PDH), phosphogluconolactonase (PGL), 6-phosphogluconate dehydratase (EDD), 2-keto-3-deoxy-6-phosphogluconate aldolase (EDA), glyceraldehyde 3-phosphate dehydrogenase (GAPDH), phosphoglycerate kinase (PGK), phosphoglycerate mutase (PGM), enolase (ENO), pyruvate kinase (PYK), phosphoglucose isomerase (PGI), phosphofructokinase (PFK), fructose 1,6-bisphosphate aldolase (FBA), triose phosphate isomerase (TPI), cellobiose phosphorylase (CBP), phosphoglucomutase (PGMT), malate shunt (MS).

Our analysis revealed a trend across all three glycolytic variants: early pathway reactions generally have lower protein costs than downstream reactions, suggesting that the initial steps operate at a higher enzyme efficiency (Fig. 3A). These lower protein demands align with the larger in vivo thermodynamic driving forces observed in early glycolysis (Table S1). For example, in Z. mobilis’s ED pathway, approximately 76% of the total change in free energy (−120 kJ mol⁻¹) occurs within the first four reactions (GLK to EDD). These reactions exhibit an average protein cost of 31.8 µg protein/(mmol h⁻¹), nearly threefold lower than that of the later steps. Similarly, in the PPi-EMP pathway of C. thermocellum, the first three reactions (cellobiose phosphorylase [CBP] to GLK) account for approximately 80% of the pathway’s total driving force (−35 kJ mol⁻¹) and have an average protein cost of 67.3 µg protein/(mmol h⁻¹), also about threefold lower than that of the subsequent reactions.

The E. coli EMP pathway presents a more complex scenario due to its use of the PTS, which couples glucose import to its phosphorylation to G6P while converting PEP to pyruvate. The initial three EMP reactions (PTS, PGI, and PFK) account for approximately 77% of the total free energy change (−83 kJ mol⁻¹). Due to the high concentrations of PTS enzymes, these reactions have an average protein cost of 97 µg protein/(mmol h⁻¹), comparable to the costs of lower EMP glycolytic reactions (104 µg protein/(mmol h⁻¹)). However, a proportion of the PTS protein cost is attributable to the conversion of PEP to pyruvate, confounding the distinction of protein costs between upper and lower glycolytic reactions. When excluding the PTS, the average protein cost of early glycolytic reactions (PGI to TPI) is about twofold lower than that of downstream reactions.

A central hypothesis of this study was that the higher thermodynamic favorability of the ED pathway in Z. mobilis would translate to lower protein costs compared to the less favorable EMP pathways in E. coli and C. thermocellum. This hypothesis was supported by our findings: lower protein costs were generally associated with higher thermodynamic driving forces for equivalent or analogous reactions across the three glycolytic variants. For the core glycolytic reactions from G6P to pyruvate, the more thermodynamically favorable ED pathway in Z. mobilis required approximately fourfold and twofold less protein per unit flux (μg protein/(mmol glucose h⁻¹)) than the EMP pathways in C. thermocellum and E. coli, respectively (Fig. 3B). When accounting for glucose transport systems (PTS in E. coli and membrane transporters in Z. mobilis and C. thermocellum), the ED pathway in Z. mobilis remained the most enzyme-efficient, requiring approximately five- and twofold less protein per flux than the EMP pathways in C. thermocellum and E. coli, respectively (Fig. 3B). These findings underscore the critical role of thermodynamic driving forces in shaping the in vivo protein investment required in glycolytic pathways.

Protein costs of sugar uptake

Z. mobilis, E. coli, and C. thermocellum use distinct processes for glucose or cellobiose uptake (Fig. 4A). Z. mobilis has four carbohydrate-specific porins (OprB1, ZMO0064; OprB2, ZMO0847; OprB3, ZMOp33 × 009; RpfN, ZMO1859) to transport sugars across its outer membrane into the periplasm (31, 55). Among these, OprB2 was expressed at substantially higher levels compared to the other three porins (Fig. 4B; Table S10). Consistent with prior studies, the glucose-facilitated diffusion protein (Glf), which transports glucose from the periplasm into the cytosol, was highly expressed (Fig. 4B) (56, 57). Although Z. mobilis encodes another transporter, ZMO0293, to import glucose into the cytosol, this protein is expressed at very low levels (Table S10), suggesting it may function under different growth conditions (58, 59).

Fig 4.

Protein costs of sugar uptake processes. (A) Z. mobilis (blue), E. coli (pink), and C. thermocellum (green) use distinct enzymes and mechanisms to uptake glucose or cellobiose. In C. thermocellum, cellobiose enters the periplasm without the need for a dedicated transporter (60–62). (B) Absolute concentrations (in fg per cell) of predominant sugar uptake proteins in each bacterium. Data are the averages of four biological replicates, with error bars indicating ±standard deviation. Enzymes designated with ND were not detected. Some error bars are too small to be visible in this representation. See Table S10 for the absolute concentration data for all sugar uptake proteins and Tables S6 to S8 for individual replicate data. (C) The total protein cost of glucose uptake, expressed as the amount of protein (μg) required per unit flux (mmol glucose h⁻¹). For C. thermocellum, the protein cost for glucose uptake is shown as half the calculated cost for cellobiose uptake, since each molecule of cellobiose contains two glucose moieties. Abbreviations: carbohydrate-selective porin OprB (OprB1, OprB2, OprB3), carbohydrate porin (RpfN), glucose-facilitated diffusion protein (Glf), outer membrane porin F (OmpF), outer membrane porin C (OmpC), maltoporin (LamB), cryptic outer membrane porin (BglH), phosphotransferase system (PTS) glucose-specific EIICB component (PtsG), PTS system glucose-specific EIIA component (Crr), phosphocarrier protein HPr (PtsH), phosphoenolpyruvate-protein phosphotransferase (PtsI), extracellular solute-binding protein family 1 (CbpB), binding-protein-dependent transport systems inner membrane component (MsdB1, MsdB2), ABC transporter-related protein (NbdB).

E. coli has four outer membrane porins (OmpF, b0929; OmpC, b2215; BglH, b3720; LamB, b4036) to transport sugars into its periplasm (63). Consistent with previous studies showing that OmpF and OmpC are utilized for glucose uptake, these two porins were highly expressed relative to LamB and BglH (Fig. 4B; Table S10) (64–66). Periplasmic glucose is subsequently transported into the cytosol via the PTS, which consists of four phospho-relay proteins: PtsG, b1101; Hpr, b2415; PtsI, b2416; and Crr, b2417 (67–69). Our data confirm that all four PTS components are highly expressed under the conditions tested (Fig. 4B).

C. thermocellum harbors five multi-component ATP-binding cassette (ABC) transporters to import sugars across the cell membrane: transporters A, B, C, D, and L (70). Consistent with previous research showing that C. thermocellum primarily uses transporter B to uptake cellobiose, our data show that the components of transporter B (MsdB1, Clo1313_1195; MsdB2, Clo1313_1196; NbdB, Clo1313_2554; CbpB, Clo1313_1194) are expressed at much higher levels than the other ABC transporter proteins (Table S10) (71). Notably, the cellobiose binding protein CbpB (Clo1313_1194) was the second most abundant protein in the C. thermocellum proteome, while the transmembrane (MsdB1/2) and ATP binding (NbdB) subunits were expressed at lower levels (Fig. 4B; Table S8).

Our analysis reveals that E. coli and C. thermocellum allocate significantly more protein (∼3 fg cell⁻¹ each) for glucose or cellobiose uptake than Z. mobilis (∼1 fg cell⁻¹). Furthermore, due to its higher glucose uptake rate (Table S2), Z. mobilis requires more than 10 times less protein to import an equivalent amount of glucose compared to E. coli and C. thermocellum (Fig. 4C).

Protein burden of fermentative pathways correlates with their reversibility

Z. mobilis and C. thermocellum produce ethanol and acetate via distinct metabolic routes (Fig. 5A). In Z. mobilis, over 95% of carbon is directed toward ethanol, with minimal production of other fermentation products, such as acetate, formate, and lactate (72, 73). Reflecting this, the concentrations of the ethanol fermentation enzymes pyruvate decarboxylase (Pdc) and alcohol dehydrogenase B (AdhB) are much higher compared to those involved in acetate or lactate production (Fig. 5B; Table S11) (74, 75). Notably, the levels of Pdc and AdhB are comparable to those of glycolytic enzymes, with Pdc levels exceeding those of all glycolytic enzymes on a mass basis (fg cell⁻¹) (Table S4).

Fig 5.

Protein costs and reversibility of fermentation pathways in Z. mobilis and C. thermocellum. (A) Fermentation pathways in Z. mobilis (blue) and C. thermocellum (green). (B) Absolute concentrations of fermentation enzymes and complexes, expressed in fg per cell. Data represent individual protein concentrations (e.g., Pdc, AdhA, AdhB, etc.), the sum of protein subunits forming a complex (e.g., Pfor1, PFor4, etc.), the combined protein and its activator (i.e., Pfl), or the sum of all enzymes and isoenzymes participating in the reaction (e.g., Ldh, Adh1-5). See Table S10 for average absolute concentration data of fermentation pathway enzymes, and Tables S6 and S8 for individual replicate data. (C) Labeling patterns of pyruvate and acetyl-CoA when cells are grown in the presence of isotopically labeled ethanol (EtOH) or acetate. The y-axis represents the fraction of the metabolite pool that is isotopically labeled, corrected for the natural abundance of ¹³C (Materials and Methods). M+0 indicates all atoms are ¹²C, and M+1 indicates one ¹³C atom is present. Pyruvate labeling patterns were calculated from valine to exclude unlabeled (M+0) pyruvate originating from the media. Acetyl-CoA labeling specifically refers to the acetyl group, calculated from aspartate and glutamate labeling data. (D) Protein costs of fermentation, expressed as the amount of protein (μg) required per unit flux (mmol glucose h⁻¹) in each bacterium. Protein costs were calculated by normalizing the sum of ethanol fermentation enzyme concentrations to ethanol flux (left bars, ethanol pathway) or by normalizing the sum of ethanol, acetate, formate, and lactate fermentation enzyme concentrations to total fermentation flux (bars on the right, total fermentation) (Materials and Methods). Data represent the averages of four biological replicates. Error bars show ±standard deviation. Some error bars are too small to be visible in this representation. See Table S11 for absolute concentration data for fermentation and acetyl-CoA metabolism proteins. Abbreviations: lactate dehydrogenase (Ldh), pyruvate decarboxylase (Pdc), pyruvate ferredoxin oxidoreductase (Pfor), pyruvate formate-lyase (Pfl), alcohol dehydrogenase (Adh, AdhA/B), NADP⁺-dependent acetaldehyde dehydrogenase (AldB), phosphate acetyltransferase (Pta), acetate kinase (Ack), bifunctional acetaldehyde/ alcohol dehydrogenase (Aldh/Adh).

C. thermocellum produces ethanol, acetate, and hydrogen as its primary fermentation products (76–79). Ethanol fermentation in C. thermocellum involves pyruvate ferredoxin oxidoreductase (Pfor), pyruvate formate-lyase (Pfl), and the bifunctional acetaldehyde/alcohol dehydrogenase (Aldh/Adh) enzymes (Fig. 5A) (76, 80). While C. thermocellum encodes five annotated Pfor complexes (Table S11), previous studies have shown that deletion of Pfor1 (Clo1313_0020-0023) or Pfor4 (Clo1313_1353-1356) reduces PFOR activity by 80%, suggesting these two complexes play a major role in ethanol fermentation (81, 82). Consistent with these findings, we observed that the subunits of Pfor1 and Pfor4 are the most abundant among all the Pfor complexes, but we also found that Pfor3 subunits are highly expressed as well (Fig. 5B). Notably, the abundances of these Pfor complexes and Aldh/Adh are comparable to that of highly abundant glycolytic enzymes (Table S8) and exceed the concentrations of enzymes involved in acetate and lactate production (Fig. 5B; Table S11).

Previous studies suggest that the PFOR reaction in C. thermocellum is highly reversible (83–86). In contrast, Pdc in Z. mobilis catalyzes a reaction that is considered to have limited reversibility (52, 87). We hypothesized that the ethanol and acetate fermentation pathways in C. thermocellum, which are preceded by a glycolytic pathway with limited thermodynamic driving force, are highly reversible and thermodynamically constrained. Conversely, the ethanol fermentation pathway in Z. mobilis, reliant on the PDC reaction and preceded by the thermodynamically favorable ED pathway, is expected to be largely irreversible. While previous computational thermodynamic analyses have supported these hypotheses (86), direct experimental evidence regarding the reversibility of fermentation pathways in C. thermocellum and Z. mobilis is lacking.

To investigate the reversibility of these fermentation pathways, we cultured Z. mobilis and C. thermocellum in the presence of 2-¹³C-labeled ethanol and 1,2-¹³C-labeled acetate (Materials and Methods) and tracked the incorporation of isotope labeling into upstream metabolites, including pyruvate and acetyl-CoA. In Z. mobilis, there was no detectable incorporation of ¹³C from labeled ethanol or acetate into acetyl-CoA or pyruvate, indicating that its ethanol and acetate fermentation pathways are highly irreversible (Fig. 5C). In contrast, C. thermocellum showed substantial incorporation of ¹³C from labeled ethanol and acetate into acetyl-CoA, with lesser incorporation into pyruvate (Fig. 5C). Specifically, 70% of acetyl-CoA was labeled from ¹³C-ethanol, reflecting the high reversibility of the ALDH/ADH reactions, while 20% of pyruvate was labeled, highlighting the reversibility of PFOR/PFL reactions (Fig. 5C). Similarly, when C. thermocellum was cultured on ¹³C-labeled acetate, we observed significant labeling of acetyl-CoA (23%) and a smaller fraction of labeled pyruvate (5%), indicating that the phosphate acetyltransferase (PTA) and acetate kinase (ACK) reactions are also reversible (Fig. 5C), and further supporting the reversibility of PFOR/PFL. These findings demonstrate that, similar to its glycolytic pathway, the ethanol and acetate fermentation pathways in C. thermocellum are highly reversible and thermodynamically constrained.

Given the pronounced differences in the reversibility of ethanol and acetate fermentation pathways between Z. mobilis and C. thermocellum, we predicted that the protein cost associated with these pathways would be substantially higher in C. thermocellum. Supporting this hypothesis, our analysis showed that the total protein cost for fermentation reactions was ninefold higher in C. thermocellum than in Z. mobilis to ferment an equivalent amount of glucose into ethanol and acetate (Fig. 5D). For ethanol production alone, C. thermocellum required nearly 11-fold more protein compared to Z. mobilis. Thus, similar to its glycolytic pathway, C. thermocellum incurs a higher enzymatic cost for fermentation compared to Z. mobilis due to the limited thermodynamic driving force of its pathways.

Protein costs of hydrogen production

Hydrogenase activity in C. thermocellum plays a key role in maintaining cellular redox balance, thereby influencing carbon flux toward fermentation products (84) (Fig. 6A). Hydrogen (H_₂) production shifts carbon flux toward acetate fermentation, increasing overall energy yield but reducing ethanol titers (84, 88, 89). Acting as electron sinks, hydrogenases divert reducing equivalents away from acetyl-CoA reduction to ethanol (Fig. 6A). Accordingly, disrupting hydrogenase activity—by deleting structural genes or interfering with their maturation—has been shown to increase ethanol titers (84, 89).

Fig 6.

Protein cost of hydrogenases in C. thermocellum. (A) Hydrogenase complexes in C. thermocellum and their connection to glycolytic and fermentative pathways. Arrows show the flow of electrons from glycolytic intermediates to ethanol or H₂ production. (B) Absolute concentrations of hydrogenases (HndD, HndG, and EchD) and their subunits. HndA-D and HndE-H make up the two bifurcating hydrogenases (BiF-HyD), and EchA-F make up the energy-conserving Fd-dependent [NiFe]- hydrogenase (Ech). Data are the averages of four biological replicates, with error bars indicating ±standard deviation. Subunits designated with ND were not detected. Some bars and error bars are too small to be visible in this representation. See Table S19 for average absolute concentration data of hydrogenases, and Table S8 for individual replicate data. (C) Protein cost of hydrogen production, expressed as the amount of protein required per unit flux (mmol glucose h⁻¹). Data represent the average of four biological replicates. Error bars show ±standard deviation. Abbreviations: pyruvate ferredoxin oxidoreductase (Pfor), phosphate acetyltransferase (Pta), acetate kinase (Ack), bifunctional acetaldehyde/alcohol dehydrogenase (Aldh/Adh), energy-conserving Fd-dependent [NiFe]-hydrogenase (Ech), bifurcating hydrogenase (BiF-HyD), hydrogenase (HndD, HndG, EchD), NADH dehydrogenase (HndC, HndF, EchB), ferredoxin (HndB, HndH, EchF), NADH-quinone oxidoreductase (HndA, HndE, EchC, EchE), Fe-only hydrogenase (Fe-HyD), NADH/ubiquinone/plastoquinone complex I (EchA).

C. thermocellum encodes three classes of hydrogenases, distinguished by their active site metal cofactors: [NiFe], [FeFe], and [Fe]. Specifically, it encodes one membrane-bound, energy-conserving, ferredoxin-dependent [NiFe] hydrogenase (Ech; Clo1313_0570–0575), two cytoplasmic bifurcating [FeFe] hydrogenases (BiF-Hyd; Clo1313_1790–1793 and Clo1313_1881–1883, Clo1313_1885), and one putative cytoplasmic [Fe]-only hydrogenase (Clo1313_0554) (23, 89). Our results show that the BiF-Hyds are the most abundant hydrogenases in C. thermocellum (Fig. 6B), consistent with previous reports (23). For the Ech complex, we detected only three of its subunits (EchD–F), likely reflecting its low expression and membrane abundance relative to BiF-Hyds, as previously reported (BiF-Hyds) (90). It is possible that EchA–C subunits were present at levels below our detection limit.

To estimate the protein cost of hydrogen production, we summed the concentrations of all hydrogenases and their subunits, divided by the previously reported rate of H₂ production (0.43 ± 0.05 mmol H₂/mmol hexose [91]), and normalized to the cellobiose uptake rate (Table S2). The resulting protein cost for hydrogen production (Fig. 6C) is lower than the total protein cost for glycolysis (Fig. 3B) but slightly higher than the combined protein costs of ethanol, acetate, and lactate fermentation (Fig. 5D).

Proteome-wide allocation of protein resources

To examine how the bacteria studied allocate their protein resources across cellular processes, we conducted a Cluster of Orthologous Groups (COG) analysis (Materials and Methods), classifying proteins into distinct biological functions (e.g., transcription, cell motility, carbohydrate metabolism and transport, etc.) (Tables S6 to S8 and S12) and quantifying protein allocation to each category (92, 93). Overall, protein allocation across COG-defined cellular functions was largely consistent across the organisms studied. Five major categories—translation, ribosomal structure and biogenesis; amino acid transport and metabolism; energy production and conversion; carbohydrate transport and metabolism; and cell wall/membrane/envelope biogenesis—accounted for 58%, 70%, and 59% of the proteome (on a fg cell⁻¹ basis) in Z. mobilis, E. coli, and C. thermocellum, respectively (Fig. 7).

Fig 7.

Allocation of protein resources to cellular processes. Proportion of the measured proteome dedicated to 21 COG-defined cellular processes in Z. mobilis (blue), E. coli (pink), and C. thermocellum (green). COG categories accounting for less than 0.1% of total protein mass are excluded. Protein mass was quantified in femtograms per cell (fg cell⁻¹). COG classifications for each protein are provided in Tables S6 to S8, and the corresponding letter designations for COG categories can be found in Table S12.

Despite these broad similarities, we also observed notable differences. In Z. mobilis and E. coli, the cellular process with the largest allocation of protein resources was protein biogenesis (i.e., translation, ribosomal structure/biogenesis), comprising ~23% of the proteome in both bacteria. In contrast, the most resource-demanding process in C. thermocellum was energy production and conversion, accounting for 20% of its proteome. This was followed closely by carbohydrate metabolism and transport, which constituted 17% of the proteome (Fig. 7).

Enhanced glycolytic flux during N₂ fixation in Z. mobilis

A potential drawback of the highly forward-driven ED pathway in Z. mobilis is that when cellular demand for energy or biomass increases, each of the enzyme-efficient steps in the pathway can become a kinetic bottleneck, potentially requiring an increase in enzyme concentration to increase flux. Interestingly, Z. mobilis can increase its glycolytic rate by approximately 40%, compared to growth under ammonia-replete conditions, when utilizing dinitrogen gas (N₂) as its sole nitrogen source (94–96). This increase in glycolytic rate correlates with increased thermodynamic favorability of the ED pathway (97).

Leveraging previous proteomics data (97), we compared ED pathway enzyme levels and protein costs between N₂-fixing and NH₄⁺-replete conditions (Fig. 8; Table S13). While the levels of most ED pathway enzymes remain unchanged or increased only marginally during N₂-fixing conditions, Pgl, Pgk, and Eno displayed significant increases (Fig. 8A). Notably, Pgl, the least abundant ED enzyme under NH₄⁺-replete conditions, showed a 1.8-fold increase during N₂ fixation, suggesting that this enzyme may be a rate-limiting step. Normalizing enzyme levels to intracellular fluxes revealed that most ED enzymes—including all enzymes in the lower half of the pathway—exhibited significantly lower protein costs under N₂-fixing conditions (Fig. 8B), indicating that Z. mobilis utilizes its glycolytic enzymes more efficiently when nitrogen availability is limited.

Fig 8.

Protein cost comparisons for the ED pathway in Z. mobilis grown under N₂-fixing vs NH₄⁺-replete conditions. (A) Absolute concentrations of glycolytic enzymes (fg/cell⁻¹) in Z. mobilis grown with NH₄⁺ (solid bars) or N₂ (striped bars) as the sole nitrogen source. Enzyme concentrations for Z. mobilis cells grown on N₂ were quantified by normalizing previous shotgun proteomics data to absolute values (Materials and Methods) (97). (B) Comparisons of protein costs for ED pathway reactions [µg protein/(mmol h⁻¹)] between NH₄⁺-replete conditions and N₂-fixing conditions. Protein costs for each glycolytic reaction were calculated by normalizing the sum of all participating enzymes to the intracellular flux of the reaction. (C) Total protein cost of the ED pathway in Z. mobilis grown on NH₄⁺ vs N₂. The total protein cost for the ED pathway in each condition represents the sum of all glycolytic enzymes normalized to the glucose uptake rate of Z. mobilis grown under that condition (Materials and Methods). Data for each nitrogen condition represent the averages of five biological replicates. Error bars show ±standard deviation. Some error bars are too small to be visible in this representation. Asterisks indicate statistical significance: *, P  <  0.05; **, P  <  0.01; ***, P  <  0.001. See Table S13 for absolute enzyme concentrations under N₂ conditions. Abbreviations: glucokinase (GLK), glucose 6-phosphate dehydrogenase (G6PDH), phosphogluconolactonase (PGL), 6-phosphogluconate dehydratase (EDD), 2-keto-3-deoxy-6-phosphogluconate aldolase (EDA), glyceraldehyde 3-phosphate dehydrogenase (GAPDH), phosphoglycerate kinase (PGK), phosphoglycerate mutase (PGM), enolase (ENO), pyruvate kinase (PYK).

The increased thermodynamic driving force of lower ED pathway reactions under N₂-fixing conditions likely contributes to decreased protein costs and higher flux for these enzymes; in contrast, regulatory mechanisms—such as allosteric control or post-translational modifications—might be responsible for improving enzyme efficiency and flux of the early highly thermodynamically favorable steps of the pathway (98–100). For example, the ΔG of GAPDH improves from −0.90 to −1.62 kJ mol⁻¹ during N₂-fixing conditions (97). Despite this seemingly minor increase in thermodynamic favorability, this change in free energy corresponds to a 1.8-fold higher net flux (Table S15). Similarly, the ΔG of EDA decreases by 0.46 kJ mol⁻¹ (−1.42 to −1.88 kJ mol⁻¹) (97), which corresponds to a 1.3-fold higher net flux (Table S15). These observations align with a previous study in E. coli showing that increases in glycolytic rates during nitrogen or phosphorus upshift correlate with increased thermodynamic driving force of pathway steps initially close to equilibrium (101).

However, the highly thermodynamically favorable upper ED pathway reactions (i.e., GLK, G6PDH, PGL, and EDD) already operate with very high efficiency, so increases in thermodynamic favorability are predicted to have a negligible effect on net flux. For the PGL reaction, increased flux could be explained by the ~1.8-fold increase in Pgl levels, but for the other upper ED pathway reactions, whose levels do not increase significantly, other explanations are warranted. Considering thermodynamically favorable reactions in glycolysis are known to be targets of metabolic regulation in several organisms (102), another possibility as to how Z. mobilis cells are capable of sustaining enhanced flux through glycolytic reactions under N₂-fixing conditions is by modulating enzyme activity via metabolic regulation and/or post-translational modifications (99, 100). Notably, phosphoproteomic analyses have identified phosphorylation changes in multiple Z. mobilis glycolytic enzymes under N₂-fixing conditions (98). Additional research is needed to determine if these post-translational modifications regulate enzyme activity and contribute to higher glycolytic flux.

DISCUSSION

This study provides in vivo evidence that thermodynamic driving forces are a key parameter influencing the in vivo enzyme burden of metabolic pathways. We show that the more thermodynamically favorable ED glycolytic pathway in Z. mobilis requires substantially less enzymatic protein to sustain the same flux as the less favorable PPi-EMP and EMP pathways in C. thermocellum and E. coli. Additionally, we show that the highly reversible fermentation pathways in C. thermocellum impose a markedly higher protein cost compared to the irreversible fermentation pathways in Z. mobilis. We note, however, that the relationship between thermodynamic driving force and enzyme burden is inherently interdependent: thermodynamics influences the amount of enzyme required to sustain a given flux, while enzyme abundance and activity, in turn, shape the intracellular metabolite concentrations that define a pathway’s thermodynamic landscape.

Comparison with theoretical predictions and trade-offs between ATP yield and protein cost

A previous theoretical analysis, based on computationally derived thermodynamic values and in vitro enzyme kinetics data, estimated that the canonical EMP pathway requires between 3.5- and 5-fold more enzymatic protein than the ED pathway to sustain the same glycolytic flux (17). Our experimental findings closely align with these predictions, as we found that the ED pathway in Z. mobilis requires approximately five- and twofold less protein to achieve the same glycolytic flux as the EMP pathways in C. thermocellum and E. coli, respectively (Fig. 3B).

The ED pathway in Z. mobilis generates one ATP per glucose, while the EMP pathway in E. coli produces two ATPs, and the PPi-EMP pathway in C. thermocellum yields four ATP equivalents (22, 103). Our findings provide strong experimental support for the predicted tradeoff between glycolytic ATP yield and protein costs. The higher energy yield of the PPi-EMP pathway in C. thermocellum, linked to its limited thermodynamic driving force, comes at the expense of significantly greater enzyme burden relative to the EMP and ED pathways in E. coli and Z. mobilis. This reliance on a thermodynamically constrained glycolytic pathway with increased ATP yield is likely an evolutionary adaptation to growth on cellulosic substrates. Microorganisms metabolizing soluble substrates can optimize either a high specific substrate consumption rate (grams of substrate consumed per gram of cells per hour) or a high cell yield (grams of cells produced per gram of substrate), both of which contribute to maximizing the specific growth rate. However, for microbes growing on cellulosic biomass, the specific substrate consumption rate is inherently limited, creating strong selective pressure to maximize cell yield by increasing glycolytic ATP yield (53). In contrast, highly thermodynamically favorable pathways with lower ATP yield, such as the ED pathway in Z. mobilis, are well-suited for environments rich in glucose, where rapid substrate consumption provides a competitive advantage.

High protein cost of PTS in E. coli

Across the glycolytic pathways examined, we observed a general trend in which early reactions with greater thermodynamic favorability incur lower protein costs, whereas less favorable downstream steps require greater enzyme investment. However, we identified several exceptions to this trend. One such exception is the PTS in E. coli, which, despite being highly thermodynamically favorable (Table S1), incurs one of the highest protein costs within EMP glycolysis (Fig. 3A). This elevated cost likely arises from its dual role in both catalysis and regulation. Specifically, the PTS is involved in carbon catabolite repression and inducer exclusion, regulating the uptake of preferred carbon sources (104). Additionally, only a fraction of catalytic components may be active at any given time due to feedback inhibition. For example, the E. coli PTS is inhibited by α-ketoglutarate, a tricarboxylic acid (TCA) cycle intermediate involved in nitrogen assimilation that binds non-competitively to PtsI preventing PEP dephosphorylation (105). Notably, PtsI is the most highly expressed PTS component (Table S10). These regulatory roles likely necessitate abundant expression of PTS components, thereby increasing its overall protein cost.

Another notable exception is the PGM reaction in C. thermocellum, which, despite operating near equilibrium (≤−0.48 kJ mol⁻¹) (53), exhibits one of the lowest protein costs within the pathway (Fig. 3A). C. thermocellum encodes multiple Pgm isoenzymes (Pgm1–Pgm7), with Pgm1 being the most highly expressed, but still maintains a low overall protein burden for this reaction. (Tables S3 and S4). This could be explained by high catalytic efficiency (k_cat/k_m) or high active site saturation, but direct measurements are unavailable. Alternatively, an unannotated, highly expressed enzyme might be responsible for catalyzing this step.

Metabolic engineering strategies to reduce protein costs in C. thermocellum

The use of a PPi-dependent phosphofructokinase (PPi-Pfk) in C. thermocellum’s glycolytic pathway enables the generation of one additional ATP per glucose compared to the ATP-dependent phosphofructokinase (ATP-Pfk) employed in the EMP pathway of E. coli. However, this increased ATP yield comes with a significant tradeoff: the protein cost of the PFK reaction in C. thermocellum is more than 12-fold higher than in E. coli.

Given its exceptional ability to degrade cellulose, C. thermocellum is widely regarded as a promising platform organism for consolidated bioprocessing of lignocellulosic biomass into fuels and chemicals (106). One potential strategy to improve biofuel production in this organism involves reducing the high protein cost of the PPi-EMP pathway by alleviating thermodynamic bottlenecks through metabolic engineering. A recent study demonstrated the feasibility of this approach by replacing PPi-Pfk with ATP-Pfk, deleting ppdk, and introducing genes encoding a soluble pyrophosphatase and Pyk to engineer a PPi-free glycolytic pathway in C. thermocellum (107). These modifications improved the thermodynamics of the PFK reaction and increased ethanol titers by 38%. Future studies could assess how these modifications impact enzyme efficiency by quantifying protein levels in the engineered pathway. Furthermore, given the high protein cost and limited thermodynamic driving force of the ethanol fermentation pathway in C. thermocellum, another potential strategy to enhance ethanol production could involve increasing the thermodynamic favorability of this pathway, potentially by replacing Pfor with Pdc.

Conclusion

This study provides in vivo evidence that thermodynamic driving forces constitute a key factor influencing the enzyme burden in glycolytic pathways. The insights and quantitative proteomic data generated here will serve as a valuable resource for developing constraint-based genome-scale metabolic models, such as resource balance analysis and metabolism and expression models, which explicitly account for the amount of enzyme needed to sustain metabolic flux.

MATERIALS AND METHODS

Strains and growth conditions

Zymomonas mobilis ZM4 (ATCC 31821) was streaked onto Zymomonas rich-medium glucose (ZRMG) plates (10 g/L yeast extract, 2 g/L KH₂PO₄, 20 g/L glucose, and 20 g/L agar) from 25% glycerol stocks and incubated in an anaerobic (5% H₂, 5% CO₂, 90% N₂ atmosphere, and <100  ppm O₂) chamber (Coy Laboratory) at 30°C for 3 to 4  days. Single colonies were used to inoculate 15 mL test tubes containing 10 mL liquid ZRMG. Cells were grown overnight and then subcultured into 25 mL–50 mL of Zymomonas minimal media (ZMM) [1 g/L K₂HPO₄, 1 g/L KH₂PO₄, 0.5 g/L NaCl, 1 g/L (NH₄)₂SO₄, 0.2 g/L MgSO₄·6H₂O, 0.025 g/L Na₂MoO₄·2H₂O, 0.0025 g/L FeSO₄·7H₂O, 0.02 g/L CaCl₂·2H₂O, 5  mg/L calcium pantothenate, and 20 g/L glucose]. These subcultures were incubated for 14–16 h and used to inoculate experimental cultures.

Escherichia coli RL3000 (MG1655 ilvG⁺ rph⁺ pyrE⁺), a non-hyperflagellated prototrophic derivative of MG1655 (108) was streaked onto Luria broth (LB) plates (10 g/L tryptone, 5 g/L yeast extract, 5 g/L NaCl, 15 g/L agar) from 25% glycerol stocks and incubated aerobically at 37°C for 16–18 h. Single colonies were used to inoculate 15 mL test tubes containing 10 mL of liquid LB. Cells were grown for 8–10 h at 37°C, 250 revolutions per minute (RPM), and then subcultured into 25 mL–50 mL of M9 minimal media (6  g/L Na₂HPO₄, 3  g/L KH₂PO₄, 0.5  g/L NaCl, 1  g/L NH₄Cl, 0.12 g/L MgSO₄, 0.0147 g/L CaCl₂, 0.002 g/L FeSO₄·7H₂O, and 4  g/L glucose). These subcultures were incubated for 14–16 h and used to inoculate experimental cultures.

Clostridium thermocellum DSM1313 growth was carried out anaerobically in thermophilic clostridium media (MTC) (9.39 g/L morpholine propanesulfonic acid [MOPS] sodium salt, 2 g/L potassium citrate monohydrate, 1.3 g/L citric acid monohydrate, 1 g/L Na₂SO₄, 1 g/L KH₂PO₄, 2.5 g/L NaHCO₃, 2 g/L urea, 1 g/L MgCl₂·6H₂O, 0.2 g/L CaCl₂·2H₂O, 0.1 g/L FeCl₂·4H₂O, 1 g/L L-cysteine HCl monohydrate, 0.02 g/L pyridoxamine HCl, 0.004 g/L p-aminobenzoic acid, 0.002 g/L biotin, 0.002 g/L vitamin B₁₂, 0.004 g/L thiamine, 0.5 µg/L MnCl₂·4H₂O, 0.5 µg/L CoCl₂·6H₂O, 0.2 µg/L ZnCl₂, 0.1 µg/L CuCl₂·2H₂O, 0.1 µg/L H₃BO₃, 0.1 µg/L Na₂MoO₄·2H₂O, 0.1 µg/L NiCl₂·6H₂O, and 5 g/L cellobiose). To prepare MTC media, tubes or bottles were filled with an initial base media containing MOPS solution, sealed with butyl rubber stoppers, made anaerobic via a vacuum manifold, overlaid with N₂ gas (oxygen scrubbed), and autoclaved. The additional media components were made anaerobic, autoclaved separately, and then added to the culture tubes/bottles. Before inoculating/extracting cultures, syringes were made anoxic by multiple drawings and expulsions of the headspace from an anaerobic sealed bottle containing 2.5% cysteine HCl solution. Cultures were inoculated directly from 25% glycerol stocks into 5 mL of MTC media and grown anaerobically in a 55°C water bath for 24 h. Cultures were then subcultured into 10 mL of fresh MTC and grown for 14–16 h, and the subcultured growth was used to inoculate experimental cultures. Experimental cultures for all three microbes were inoculated at an initial optical density at 600 nm (OD₆₀₀) of 0.05 to 0.06.

Protein extraction and sample preparation for proteomics analyses

When cells reached an OD₆₀₀ of 0.45–0.46, 10 mL of bacterial culture was collected in a prechilled 15 mL conical tube for four biological replicates and centrifuged at 4,255 × g for 5 min at 4°C. Cell pellets were washed with phosphate-buffered saline (8 g/L NaCl, 0.2 g/L KCl, 1.44 g/L Na₂HPO₄, 0.24 g/L KH₂PO₄) (109) and centrifuged at 16,000 × g for 5  min at 4°C, and the supernatant was discarded. Cell pellets were then stored at −80°C until proteomics analysis.

Cell pellets were thawed and resuspended in denaturing buffer (5.4 M guanidinium hydrochloride, 100 mM Tris-HCl). Samples were sonicated for 5 min in a chilled water bath (QSonica) using the following program: 20 s on, 10 s off, amplitude of 30, and temperature maintained at 14°C. Samples were then incubated in a sand bath at 110°C for 5 min, cooled at room temperature for 5 min, and incubated again in the sand bath at 110°C for 5 min. To precipitate the protein, liquid chromatography-mass spectrometry (LC-MS)-grade MeOH was added to each sample to a final solution volume of 90% MeOH (vol/vol) and vortexed. Samples were centrifuged at 14,000 × g for 2 min at 4°C to pellet precipitated protein, and the supernatant was carefully removed without disturbing the protein pellet. Protein pellets were resuspended in 8 M urea, 100 mM Tris-HCl, 10 mM Tris(2-carboxyethyl)phosphine (TCEP), and 40 mM chloroacetamide, and vortexed for 10 min at room temperature to resolubilize the protein. Samples were incubated with LysC (1 mg/mL; VWR, Radnor, PA) at a 1:50 protease-to-protein mass ratio for 4 h at room temperature with gentle rocking. Samples were then diluted with freshly prepared 100 mM Tris-HCl, pH 8.0, in order to bring the sample urea concentration to 2 M. Trypsin (Promega, Madison, WI) was added to each sample at a 1:50 protease:protein mass ratio, and samples were incubated at ambient temperature overnight while gently rocking. The digestion reaction was terminated by adding sufficient 10% trifluoroacetic acid (TFA) in H₂O to each solution to bring solution pH to <2, as verified by pH strip. Samples were centrifuged at 14,000 × g for 2 min at room temperature to pellet insoluble material. The resulting supernatant was desalted using Strata-X 33 µm polymeric reversed-phase SPE cartridges (Phenomenex, Torrance, CA). The desalted peptides were dried down in a vacuum centrifuge (Thermo Fisher Scientific, Waltham, MA). Peptides were resuspended in water to determine peptide concentration via NanoDrop One Microvolume UV-Vis spectrophotometer (Thermo Fisher Scientific, Waltham, MA). For samples used for absolute protein quantification, peptides were combined with synthetic HeavyPeptide AQUA peptide standards (Thermo Fisher Scientific, Rockford, IL). For each sample, two dilutions were prepared to ensure peptide standard concentrations were approximately close to the native peptide concentrations as estimated by shotgun proteomic analyses. The sample mixtures were dried down again, then resuspended in 40% acetonitrile in 0.2% formic acid for infusion. For samples used for LC-MS shotgun proteomics analysis, desalted peptides were resuspended in 0.2% formic acid, and peptide concentrations were quantified via NanoDrop.

Absolute proteomics and shotgun proteomics methods

Sample analysis for absolute protein quantitation was performed using the TriVersa NanoMate (Advion, Ithaca, NY) coupled to an Orbitrap Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific, San Jose, CA). The NanoMate was equipped with a 5 µm nominal internal diameter nozzle electrospray ionization (ESI) chip operated at 1.60 kV, with a gas pressure of 1.0 psi, and 10 µL injection volume, with remaining volume returned to well after an injection. The MS was operated in positive ionization mode via parallel reaction monitoring (PRM), in which the m/z values corresponding to the ions from the native and isotope-labeled peptides were targeted for MS2 spectral acquisition. Targeted precursor ions were isolated from a 0.5 Da isolation window in the quadrupole; HCD MS2 scans with 25% fixed collision energy and a normalized automatic gain control (AGC) target (%) of 200, equivalent to 1 × 10⁵ ions, were collected in the Orbitrap from 350 to 2,000 m/z with a resolution of 500,000. Maximum injection time was set to 1,014 ms for higher concentration samples or 5,000 ms for lower concentration samples.

To analyze samples for shotgun LC-MS proteomics, 2 µg of peptides was loaded onto a 75 μm-inside-diameter, 30 cm-long capillary with an imbedded electrospray emitter and packed in a 1.7 μm-particle-size C₁₈ ethylene bridged hybrid (BEH) column. The mobile phases used were as follows: phase A, 0.2% formic acid; and phase B, 0.2% formic acid–70% acetonitrile. Peptides were eluted with a gradient increasing from 0% to 75% B over 42 min, followed by a 4 min 100% B wash and 10 min of equilibration in 100% A for a complete gradient of 60 min.

The eluting peptides were analyzed with an Orbitrap Fusion Lumos (Thermo Scientific) mass spectrometer. Survey scans were performed at a resolution of 240,000 with an isolation analysis at 300 to 1,350 m/z and AGC target of 1 × 10⁶ ions. Data-dependent top-speed (1 s) tandem MS/MS sampling of peptide precursors was enabled with dynamic exclusion set to 10 s on precursors with charge states 2 to 4. MS/MS sampling was performed with 0.7 Da quadrupole isolation and fragmentation by higher-energy collisional dissociation with a collisional energy value of 25%. The mass analysis was performed in the ion trap using the “turbo” scan speed for a mass range of 200 to 1,200 m/z. The maximum injection time was set to 11  ms, and the AGC target was set to 20,000.

Absolute and shotgun proteomics data analysis

For targeted data analysis, raw data files from the PRM direct infusion-MS/MS experiments were imported into Skyline 22.2.0.351. Three to five transitions per targeted precursor ion were manually integrated to quantitate over a period of time during which the MS2 ion intensities were stable. For a given native peptide and its matching isotope-labeled peptide, selected transitions were quantitated over the same period of time.

Data were exported into Excel. For each quantitated transition, peak height was calculated by normalizing the measured area to the quantitation time. The calculated height values were summed for each set of transitions per precursor ion. If multiple charge states were tracked for a peptide (e.g., 2+ and 3+), these summed height values were added together. These summed values, as well as the known concentration of the isotope-labeled peptide spiked into the sample mixture and the total protein digest volume, were consequently used to calculate the concentration of the native peptide in the total protein digest, with corrections for dilution as necessary. The concentrations of each native peptide in the total protein digest were then normalized to a per mL of cells basis by factoring in the total amount of cells used per protein extraction (i.e., 10 mL). These concentration data were converted to copies per cell using calculated cell numbers and Avogadro’s constant or converted to a mass basis (i.e., fg/cell and μg/g per dry cell weight [gDCW]) using calculated cell volumes, gDCW values, and protein molar masses (Table S2). Finally, absolute protein concentrations and their associated error were determined by calculating the average and standard deviation, respectively, across all peptides for a protein and all four biological replicates (Tables S21 to S23). NRMSE across peptides and biological replicates was calculated using the equation $\sqrt{1/n\sum _{i=1}^n{(X}_i-X_o{)}^2}/X_o$ (Tables S21 to S24).

Raw shotgun LC-MS proteomics data were analyzed using the MaxQuant software (version v2.6.2.0) (110). Spectra were searched using the Andromeda search engine against a target-decoy database. FASTA reference proteomics for each microbe was obtained from the National Center for Biotechnology Information (NCBI) or UniProt databases. Label-free quantitation (i.e., iBAQ) was toggled on, and default values were used for all other analysis parameters. The peptides were grouped into subsumable protein groups and filtered to reach 1% false discovery rate based on the target-decoy approach. iBAQ intensities were log₂-transformed, and these values for proteins that were absolutely quantified were used to construct a quantitative model for global protein quantification (Fig. S1). Cross-validation of our quantitative model was performed using the leave-one-out cross-validation method via the R package caret (version 6.0.94).

Absolute protein concentrations for Z. mobilis grown under N₂ fixation conditions were quantified using label-free quantification proteomics data from a previous study (97). Growth conditions for NH₄-replete Z. mobilis cells grown in Martien et al. were identical to the growth conditions used in our absolute proteomic measurements (97). Thus, we normalized the NH₄-replete proteomics data from Martien et al. to absolute values and used the fold change measurements between NH₄-replete and N₂ fixation conditions to obtain absolute values for Z. mobilis proteins when grown under N₂ fixation conditions (Table S13) (97).

Cell volume measurements via microscopy

To calculate cell volumes, 1 mL of cells was collected in a tube at an OD₆₀₀ of 0.45–0.46 following the growth scheme described above under “Strains and growth conditions.” The tube containing cells was not placed on ice to minimize fluctuations in cell volume induced by temperature changes. Following sample collection, 1 µL of cells was placed on a 1.5% agarose pad made with M9, ZMM, or MTC media without carbon source to reduce cell movement, and the pad was placed cell-side down onto a coverslip. Samples were analyzed <5 min following removal from the flask via phase-contrast microscopy with a resolution of 0.1083 µm/pixel using a 60× UPlanSApo oil objective attached to an Olympus IX83 inverted microscope and an ORCA-Flash4.0 V2 digital camera (Hamamatsu, C11440-22CU). Cell images were adjusted for brightness and contrast and analyzed using the ImageJ software (111). For each bacterium, length (L) and width (W) dimensions of 100 individual cells were obtained, and cell volumes (V) were calculated by assuming cells have the shape of a cylinder capped with two half-spheres and the following formula: $V=(\pi \bullet W^2)\bullet (L-W/3)/4$ (Table S2) (112–115). To confirm the accuracy of our cellular measurements, we calculated the dimensions of microspheres (LIVE/DEAD BacLight Bacterial Viability and Counting Kit, ThermoFisher Scientific) with a reported diameter of 6 µm and obtained an average diameter of 5.94 ± 0.07 µm (N = 20).

Grams per dry cell weight to OD₆₀₀ measurements

To perform gDCW measurements, three biological replicates of each microbe were grown as described above under “Strains and growth conditions.” A total of 500 mL (Z. mobilis and E. coli) or 100 mL (C. thermocellum) of bacterial culture was collected during late-phase growth (OD₆₀₀ >0.75), and the culture was centrifuged at 4,255 × g for 20 min at 4°C. Cell pellets were washed with ddH₂O to remove salts, and this cell suspension was vacuum filtrated through a pre-weighed 0.45 µm-pore-size hydrophilic nylon filter (Millipore; catalog no. HNWP04700) applied to a sintered glass funnel. The nylon filter containing cells was placed in a glass Petri dish and oven dried at 80°C until the mass of the filter was stable (24–48 h) (52, 84). The mass of the cells on the filter was then normalized to the OD₆₀₀ at the time of collection and the culture volume that was filtered to obtain gDCW OD₆₀₀⁻¹ L⁻¹ (Table S2).

Sugar uptake and growth rate calculations

Growth rates (h⁻¹) and sugar (glucose: Z. mobilis and E. coli; cellobiose: C. thermocellum) consumption rates (mmol_sugar gDCW⁻¹ h⁻¹) were obtained by growing three biological replicates of each microbe as described above under “Strains and growth conditions.” OD₆₀₀ measurements and 1 mL culture samples were collected every hour until stationary phase was achieved. Culture samples were centrifuged at 21,000 × g for 5 min at 4°C, and the supernatant was stored at −80°C until analysis by LC-MS. Samples were diluted 1:100 (Z. mobilis) or 1:20 (E. coli and C. thermocellum) with high-performance liquid chromatography (HPLC)-grade H₂O, mixed 50:50 with 1 mM [U-¹³C] glucose (Z. mobilis and E. coli) or 1 mM [U-¹³C] cellobiose (C. thermocellum), and analyzed via LC-MS. LC-MS analysis was performed on a Vanquish ultra-high-performance liquid chromatography (UHPLC) system (Thermo Scientific) coupled to a hybrid quadrupole-Orbitrap mass spectrometer (Q Exactive; Thermo Scientific) equipped with electrospray ionization operating in negative-ion mode. The chromatography was performed at 25°C using a 2.1- × 100 mm reverse-phase C₁₈ column with a 1.7 µm particle size (Water; Acquity UHPLC ethylene-bridged hybrid). The chromatography gradient used solvent A (97:3 H₂O:methanol with 10  mM tributylamine adjusted to pH 8.2 using 10 mM acetic acid) and solvent B (100% methanol) and was as follows: 0–2.5 min, 5% B; 2.5–8 min, linear gradient from 5% B to 95% B; 8–10.5 min, 95% B; 10.5–11 min, linear gradient from 95% B to 5% B; 11–15 min, 5% B. The flow rate was held constant at 0.2  mL min⁻¹. The MS parameters used were as follows: full MS-single ion monitoring scanning between 70 and 1,000 m/z; AGC target, 1e6; maximum injection time, 40  ms; resolution of 70,000 full width at half maximum. Data analysis was performed using the MAVEN software (116). Glucose and cellobiose were identified based on retention times matched to pure standards. The ratio of ¹²C-to-¹³C peak intensities was used to calculate glucose or cellobiose concentrations, and sugar consumption rates were normalized to gDCW and growth rates (Table S2).

Cell enumeration via flow cytometry

Cell densities (cell mL⁻¹) were quantified using flow cytometry. Three biological replicates of each microbe were grown as described above under “Strains and growth conditions”. When cells reached an OD₆₀₀ of 0.45, 5 mL of bacterial culture was collected and centrifuged at 4,255 × g for 5 min at 4°C. Cell pellets were washed twice with NaCl solutions to remove media components. NaCl solutions were prepared at 0.85, 0.55, and 0.27% to match the osmolarity of M9, ZMM, or MTC media, respectively, to prevent cell lysis/plasmolysis. Cells were then diluted 1:100 in NaCl solution, equimolar amounts of SYTO 9 and propidium iodide, and 10⁶ counting beads (LIVE/DEAD BacLight Bacterial Viability and Counting Kit, Thermo Fisher Scientific). Samples were then immediately analyzed via flow cytometry.

Prior to acquisition, sample tubes were briefly vortexed. Samples were analyzed using an Attune NxT Acoustic Focusing Cytometer (ThermoFisher Scientific) with a flow rate of 12.5 µL min⁻¹. Forward scatter area (FSC-A) and side scatter area (SSC-A) were set to 300 and 325, respectively. Fluorescence was measured using the instrument’s standard detector channels: BL1-A, YL2-A, RL3-A, and VL1-A, set to 350, 500, 400, and 400, respectively. For each replicate, 50 µL equating to approximately 100,000 single cells were analyzed/counted. Data analysis was performed using the FlowJo software (BD Biosciences, version 10.9). Manual gating using an FSC-A vs SSC-A dot plot was performed to distinguish cells and beads from debris and aggregates, and an SSC-A vs SSC-H dot plot was used to account for smaller aggregates and multiplets. Cell numbers were then calculated using the following formula: no. of bacterial events × dilution factor/no. of bead events × 10⁶ (Table S2).

Protein cost calculations and in vivo flux and thermodynamic data

In vivo free energies and glycolytic fluxes were obtained from previous studies that quantified these values via MFA models (Tables S1 and S9). These flux and thermodynamic data were calculated under similar growth conditions used in this study (52–54). Intracellular fluxes and free energies under N₂ fixation conditions in Z. mobilis were also obtained from previous MFA data (96, 97). To quantify protein costs, we normalized the sum of all participating enzymes and isoenzymes to the intracellular flux of the metabolic reaction. For example, to calculate the protein cost for the PFK reaction in E. coli, we first summed the average protein concentrations of the isoenzymes PfkA (211.9 ± 8.2 µg gDCW⁻¹) and PfkB (49.0 ± 3.6 µg gDCW⁻¹) to obtain the total protein concentration of 260.9 ± 8.9 µg gDCW⁻¹, where the total standard deviation represents the square root of the summed variances from each isoenzyme. This total protein concentration and standard deviation for PfkA and PfkB was then normalized to the in vivo flux of the PFK reaction (8.23 mmol/(gDCW h⁻¹)) to obtain the PFK protein cost of 31.8 ± 1.1 µg protein/(mmol h⁻¹). See Table S20 for the calculated protein costs of glycolytic enzymes.

To quantify the total protein cost of fermentation in Z. mobilis, we took the ratio of the sum of Pdc, Adh (AdhA and AdhB), AldB, and Ldh (Ldh1 and Ldh2) enzyme concentrations (μg gDCW⁻¹) to the combined flux of Adh and AldB. Lactate flux data were unavailable but are largely considered to be negligible in Z. mobilis. The protein cost of just ethanol fermentation in Z. mobilis was determined first by normalizing the Pdc enzyme concentration to the ratio of acetate to Pdc flux, which provided the proportion of Pdc enzyme strictly dedicated to ethanol production. The ethanol fermentation protein cost was then quantified by taking the ratio of the sum of the adjusted Pdc enzyme concentration and Adh enzyme concentration to the Adh flux.

To quantify the total protein cost of fermentation in C. thermocellum, we first quantified the proportion of Pfor and Pfl (i.e., the sum of all Pfor subunits, Pfl, and Pfl-activating enzyme, see Table S11) proteins dedicated toward fermentation metabolites (i.e., acetate and ethanol). This was done by calculating the ratio of acetate and ethanol (i.e., Adh) flux to the total acetyl-CoA flux (i.e., the sum of Pfor and Pfl flux) (117). This ratio represented the proportion of Pfor and Pfl enzymes used for fermentation. These normalized Pfor and Pfl protein concentrations were combined with the protein concentrations for Pta, Ack, Ldh, and Adh (Aldh/Adh, Adh1–5), and this total protein sum was normalized to the sum of lactate, acetate, and ethanol flux. The protein cost of just ethanol fermentation in C. thermocellum was determined first by normalizing the Pfor and Pfl enzyme concentrations to the ratio of ethanol flux to the total acetyl-CoA flux, which provided the proportion of Pfor and Pfl enzyme strictly dedicated to ethanol production. The sum of these adjusted Pfor and Pfl protein levels and total Adh enzyme concentration was then normalized to the ethanol flux. All fluxes obtained from the literature were normalized to our glucose uptake rates (Tables S2 and S9).

To calculate free energies for the E. coli PTS and the C. thermocellum reactions cellobiose phosphorylase, phosphoglucomutase, glucokinase, and pyruvate phosphate dikinase/malate shunt that lack intracellular data, we combined in vivo metabolite concentration data (52, 53, 118–120) with standard Gibbs free energy estimates (87) and obtained theoretically optimized free energies for these reactions using the Max-Min driving force (MDF) computational tool (Table S1). The MDF method identifies the most thermodynamically restrictive reactions in a pathway and maximizes their thermodynamic driving force by optimizing metabolite concentrations (16). MDF analysis was performed using the Python package equilibrator-pathway (version 0.5.0) (121). Intracellular pH, pMg, and ionic strength were set to 7, 3, and 250 mM, respectively. Temperature was set to 310.15 K and 328.15 K for E. coli and C. thermocellum, respectively. Maximum and minimum metabolite concentration bounds were based on a 50% range of absolute intracellular data (52, 53). Importantly, these in vivo metabolite concentrations were quantified in E. coli RL3000 and C. thermocellum DSM1313 cells grown under equivalent conditions used in this study. Cellobiose and glucose concentration bounds were informed by cellobiose and glucose concentrations in the media at the time of protein quantification (i.e., OD₆₀₀ 0.45). Diphosphate concentration bounds for C. thermocellum were based on intracellular concentration data quantified in related Clostridia species that encode for PPi-PFKs (119, 120). Orthophosphate concentration bounds for E. coli were based on measurements performed in E. coli K-12 (118). For both microbes, the minimum bound of pyruvate was increased from the default 1 µM to 1 mM based on the intracellular concentrations of other glycolytic intermediates. SBtab files used to perform the MDF analyses for E. coli and C. thermocellum can be found in Tables S17 and S18, respectively.

COG classification of proteins

Proteins were assigned to COG-defined cellular functions using the NCBI Batch CD-Search tool (92, 93, 122, 123). Searches were performed against the COG database. Unassigned proteins were manually classified with the “Function unknown” COG category. The percentage of the proteome mass dedicated to each COG-defined cellular function was quantified on a mass basis (fg cell⁻¹). For proteins with multiple COG classifications, the protein concentration was evenly divided among each category.

Isotopically labeled ethanol and acetate experiments

To assess the reversibility of the fermentation pathways in Z. mobilis and C. thermocellum, we performed growth experiments with isotopically labeled ethanol or acetate and tracked the propagation of isotope labeling to upstream intermediates. Z. mobilis was grown as described above under “Strains and growth conditions.” When cells reached an OD₆₀₀ of 0.5, 7.5 mL of bacterial culture was collected for four biological replicates. Cells were centrifuged, the supernatant was discarded, and two replicates of cell pellets were resuspended in either 7.5 mL of fresh ZMM spiked with 2.5 g L⁻¹ of 1-¹³C-ethanol or 1-¹³C-acetate. Cells were grown for an additional 45 min before metabolites were extracted. C. thermocellum was also grown as described above under “Strains and growth conditions.” Two biological replicates were grown in either MTC media prepared with 2 g L⁻¹ of 2-¹³C-ethanol or 1,2-¹³C-acetate. Metabolite extractions were performed when cells reached an OD₆₀₀ of 0.45.

At the time of metabolite extraction, 5 mL of liquid culture was collected in the anaerobic chamber using a serological pipette. Cells were separated from the media by vacuum filtering the culture through a 0.45 μm-pore-size hydrophilic nylon filter (Millipore; catalog no. HNWP04700) applied to a sintered glass funnel. The nylon filter containing cells was immediately immersed cell-side down into a plastic petri dish (5.5 cm diameter) containing 1.5 mL cold (–20°C) extraction solvent (40:40:20 by % volume methanol-acetonitrile-water; all HPLC grade) and kept on a chilled aluminum block. This process simultaneously lysed the cells, quenched metabolism, and dissolved intracellular metabolites. The petri dish was lightly swirled to ensure complete contact of solvent with the filter. Filters remained in the cold solvent for ~15 min before being repeatedly rinsed in the extraction solvent to collect any remaining cell debris and metabolites. The cell-solvent mixture was then transferred to a 1.5 mL microcentrifuge tube, removed from the anaerobic chamber, and centrifuged at 16,000 × g for 10 min at 4°C, and the supernatant was collected for LC-MS analysis.

LC-MS analysis was performed as previously described (124, 125) but with altered chromatography. The chromatography gradient used solvent A and solvent B and was as follows: 0 to 2.5 min, 5% B; 2.5 to 17 min, linear gradient from 5% B to 95% B; 17 to 19.5 min, 95% B; 19.5 to 20 min, linear gradient from 95% B to 5% B; and 20 to 25 min, 5% B. Data analysis was performed using the MAVEN software (116). Pyruvate and acetyl-CoA were identified based on retention times matched to pure standards. Metabolite mass isotopomer distributions from ¹³C labeling samples were corrected for ¹³C natural abundance using ElemCor (126). Pyruvate labeling patterns were calculated from valine to exclude unlabeled (M+0) pyruvate in the media, and the acetyl group labeling in acetyl-CoA was calculated from aspartate and glutamate.

DATA AVAILABILITY

The mass spectrometry shotgun proteomics and absolute quantitation proteomics data sets for C. thermocellum, Z. mobilis, and E. coli, have been deposited to the MassIVE (Mass Spectrometry Interactive Virtual Environment) database under the accession number MSV000098743 and can be accessed at https://doi.org/10.25345/C54M91P8M.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/mbio.01837-25.

Figure S1. mbio.01837-25-s0001.pdf.

Correlation between AQUA quantitation and iBAQ values.

Supplemental tables. mbio.01837-25-s0002.xlsx.

Tables S1 to S23.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.