Mitochondrial ATP generation is more proteome efficient than glycolysis

Abstract

Metabolic efficiency profoundly influences organismal fitness. Non-photosynthetic organisms, from yeast to mammals, derive usable energy primarily through glycolysis and respiration. While respiration is more energy-efficient, some cells favor glycolysis even when oxygen is available (aerobic glycolysis, Warburg effect). A leading explanation is that glycolysis is more efficient in terms of ATP production per unit mass of protein (i.e. faster). Through quantitative flux analysis and proteomics, we find however that mitochondrial respiration is actually more proteome-efficient than aerobic glycolysis. This is shown across yeasts, T cells, cancer cells, and tissues and tumors in vivo. Instead of aerobic glycolysis being valuable for fast ATP production, it correlates with high glycolytic protein expression, which promotes hypoxic growth. Aerobic glycolytic yeasts do not excel at aerobic growth, but outgrow respiratory cells in oxygen limitation. We accordingly propose that aerobic glycolysis emerges from cells maintaining a proteome conducive to both aerobic and hypoxic growth.

Introduction

Cells can generate ATP via fermentation (glycolysis) or respiration. Respiration generates around ten-fold more ATP per glucose. However, many fast-growing cells favor glycolysis (‘Crabtree effect’ leading to ethanol in yeast or ‘Warburg effect’ leading to lactate in cancer or immune cells)^1–5. This metabolic shift is often assumed to be due to glycolysis being more enzyme efficient, or proteome efficient ^6–8. Specifically, it is often believed that, compared to respiration, fermentation is capable of producing ATP faster per unit enzyme expression ^9–17. Proteome efficiency is evolutionarily advantageous because cells have limited biosynthetic capacity (e.g. ribosomes) to make enzymes and physical space to house them (constrained proteome capacity ^7,8,18,19). If cells can achieve the same metabolic flux while using less enzyme, this allows them to grow faster.

The proteome efficiency of glycolysis versus respiration has been carefully experimentally tested in E. coli. Like yeast, E. coli tend to switch from respiration to aerobic glycolysis as their growth accelerates. Unlike yeast, however, aerobic glycolytic E. coli engage in a mixture of glycolytic and respiratory ATP generation, excreting the oxidized product acetate, with 4 glycolytic NADH feeding into the electron transport chain for each glucose (versus zero NADH in yeast or mammals). This provides an aerobic glycolytic ATP yield of about 12 per glucose (versus 2 in yeast or mammals). Such acetate fermentation is favored over full respiration for its proteome efficiency^7,17.

Due to the much lower ATP yield of eukaryotic aerobic glycolysis, the proteome efficiency of glycolysis versus mitochondrial respiration remains unclear. To address this, here we integrate quantitative fluxomics and proteomics to measure the proteome efficiency of ATP generation in yeasts and mammals. The quantified proteome efficiencies reflect what the cell gains in terms of ATP for its overall protein investment in the pathway, and thus is suppressed if the pathway is not fully utilized, e.g., due to limited substrate, demand, levels of one or more gating enzymes. We carefully study naïve and activated T cells and two industrially important yeasts (Saccharomyces cerevisiae and Issatchenkia orientalis, separated by 200 million years of evolution)^20–26. Across these cell types, proteome efficiency of respiration was consistently equal or greater than glycolysis. Integrating literature data, we then expand this analysis to cancer cells and to tissues and tumors in vivo, finding superior proteome efficiency of respiration in all aerobic settings. Consistent with the high proteome efficiency of respiration, across 23 evolutionarily divergent yeast species, the fastest growing yeast were respiratory, with aerobic glycolytic yeast favored selectively in oxygen limitation. Aerobic glycolytic cells displayed a large glycolytic proteome even under fully aerobic conditions. These data support aerobic glycolysis being intrinsically inefficient from both the energy and proteome perspectives, but emerging as a consequence of some cells maintaining a glycolytic proteome that enables efficient growth also in hypoxia.

Results

Quantitative flux analysis in yeasts and T cells

We first set out to measure carefully the proteome efficiency of glycolysis versus respiration in two budding yeasts (the Baker’s yeast S. cerevisiae strain CEN.PK and a yeast used industrially for organic acid production, I. orientalis SD108), and naïve and activated murine CD8+ T cells. Analogous to specific enzyme activity, measurement of proteome efficiency requires two key inputs: reaction velocities (metabolic fluxes) and enzyme abundances.

To obtain metabolic fluxes at the systems level, we carried out ¹³C metabolic flux analysis (MFA)²⁷, developing large-scale metabolic flux models with complete atom mappings for both yeasts and for mammalian cells (Fig. 1, Extended Data Fig. 1 and 2) (also see Method, “13C metabolic flux analysis”). This approach carefully accounts for carbon fluxes devoted for both energy and biosynthesis. The models were constrained by flux balance and by experimentally measured extracellular fluxes, biomass synthesis fluxes, and extensive isotope labeling data (Extended Data Fig. 1 a-d for yeast, Extended Data Fig. 2 a-e for T cell). This enabled comprehensive eukaryotic metabolic flux analysis. The entire flux maps are available in Supplementary files and can be visualized with Escher²⁸ (https://escher.github.io) (see “13C metabolic flux analysis” section in Methods).

Figure 1. Central carbon metabolic fluxes and ATP sources in yeasts and T cells.

(a) Flux map of central carbon metabolism, in units of mmol/hr/gram dry weight (gDW), of Issatchenkia orientalis and Saccharomyces cerevisiae (strain CEN.PK) in aerobic exponential growth in yeast nitrogen base (YNB) media with 20 g/L glucose. μ, growth rate.

(b) Flux map of central carbon metabolism, in units of 0.1 mmol/hr/gDW, of naïve mouse T cells (maintained with IL-7) or activated T cells (24-hour activation by αCD3 and αCD28 with IL-2). μ, growth rate.

Inset shows ATP fluxes from glycolysis (glyc) and respiration (resp). Numbers are best-estimate fluxes from large-scale ¹³C-informed metabolic flux analysis, constrained by metabolite ¹³C labeling ([U¹³C₆]glucose and [1,2-¹³C₂]glucose for all cell types as well as [U¹³C₅]glutamine for T cells, each with n = 3 or 4 biological replicates) and metabolite consumption and excretion rate measurements (n = 3 biological replicates for yeasts and n ≥ 6 for T cells).

Metabolic fluxes diverged markedly between the two budding yeasts, with I. orientalis more respiratory and S. cerevisiae more glycolytic (Fig. 1a, Extended Data Fig. 1 c-e). I. orientalis grew faster than S. cerevisiae (growth rate μ = 0.52 vs. 0.39 h⁻¹), had faster pentose phosphate pathway (PPP) and biosynthetic fluxes, mainly generated ATP by respiration via tricarboxylic acid (TCA) cycle and oxidative phosphorylation (OXPHOS), and maintained cytosolic redox balance by feeding NADH into the electron transport chain via the quinone oxidoreductase (NDE). In comparison, S. cerevisiae engaged in prototypical aerobic glycolysis with a truncated TCA cycle. The aerobic glycolytic phenotype was yet stronger in S. cerevisiae strain FY4, which has lab-acquired mutations that impair respiration (Extended Data Fig. 1b). Consistent with these findings, Nde knockout impaired I. orientalis but not S. cerevisiae growth²⁹ (Extended Data Fig. 1f). Overall, biosynthetic fluxes account for 31% of glucose carbon in I. orientalis and only 9% in S. cerevisiae (Extended Data Fig. 1g). Among 11 key central metabolites, only α-ketoglutarate in S. cerevisiae is mainly produced to fulfill biosynthetic demand (Extended Data Fig. 1h).

Similar analyses in murine CD8+ T cells showed, upon in vitro activation, a shift from net lactate consumption towards aerobic glycolysis (Fig. 1b, Extended Data Fig. 2c). Cell mass tripled. (Extended Data Fig. 2e). Per cell weight, the median metabolic flux increased 8-fold, with glycolysis (65-fold) up more than respiration (4-fold) (Fig. 1b, Extended Data Fig. 2f-g). In both naïve and activated T cells, the TCA cycle was largely driven by glutamine not glucose³⁰ (Fig. 1b, Extended Data Fig. 2b). Resulting excess TCA four-carbon units were drained by extensive flux from malate to pyruvate, suggesting high malic enzyme activity (Extended Data Fig. 2g-h), consistent with the importance of malic enzyme in maintaining T cell redox homeostasis³¹.

Figure 2. Proteome allocation and proteome efficiency in yeasts and T cells.

(a) Mass fraction of proteome sectors of I. orientalis and S. cerevisiae in glucose batch culture as measured by quantitative LC-MS/MS proteomics. Mean ± s.e.m., n = 4 biological replicates.

(b) Naïve and activated CD8+ T cells. Culture conditions are as in Fig. 1. n = 3 biological replicates.

(c) Example calculation for proteome efficiency using I. orientalis data.

(d) Proteome efficiencies in yeasts and T cells. Mean ± s.d. with error propagated from flux confidence intervals (from n = 3 biological replicates) and proteomics (n = 4 biological replicates for yeasts, n = 3 for T cells).

The above four cell types show distinct energy profiles, including total ATP flux and how this ATP is produced (glycolysis or respiration). Both the slow ATP-burning naïve T cells and fast-burning I. orientalis generated most of their ATP via respiration (98% and 91% respectively). Activated T cells, despite excreting a majority of glucose carbon as lactate, still made 58% of their ATP aerobically. Only S. cerevisiae used glycolysis as its main ATP source, with 31% of ATP via respiration (Fig. 1).

Proteome allocation in yeasts and T cells

We next assessed absolute protein abundances using quantitative proteomics with a combination of intensity-based absolute quantification (IBAQ) and isobaric TMT tagging³². Our data on S. cerevisiae is comparable to literature reports (Extended Data Fig. 3a). We then assembled this information into a coarse-grained description of proteome composition, focused on three proteome sectors, which together account for > 80% of protein biomass: Nuclear (including proteins involved in transcription), Translation, and Metabolism (Fig. 2, Extended Data Fig. 3b-c). Within Metabolism, we separately assessed the Glycolytic and Respiratory proteome. The former contains proteins that perform 14 reactions from glucose uptake to organic waste production, and the latter includes all reactions in oxidative phosphorylation and TCA (23 annotated reactions in yeasts and 16 in T cells, and, additionally for T cells, proteins involved in mitochondrial amino acid processes and fatty acid β oxidation) (Extended Data Fig. 3b-c). The two yeasts had similar proteome composition, with the exception of repartitioning from Respiratory (in I. orientalis) to Glycolytic (in S. cerevisiae) (Fig. 2a). Pyruvate decarboxylase (ethanol fermentation) in S. cerevisiae and ATP synthase (OXPHOS) in I. orientalis are the most resource-demanding proteins in the energetic pathways, each accounting for more than 3% of the respective yeast’s total proteome mass (Extended Data Fig. 3b). Thus, proteome partitioning aligns with the metabolic preference in S. cerevisiae and I. orientalis.

In T cells, upon activation, on a per cell basis there was about 5-fold expansion of both translational and metabolic machinery (Extended Data Fig. 3c). The overall nature of the metabolic proteome was largely unchanged, and did not explain the induction of aerobic glycolysis during T cell activation (Fig. 2b). Certain gating glycolytic proteins, however, selectively increased upon T cell activation^33,34: GLUT1 (glucose transporter) and HK2 (hexokinase) (Extended Data Fig. 3c). Thus, naïve T cells come pre-loaded with most of the machinery for aerobic glycolysis, but glycolysis remains slow until energy demand increases and/or these gating proteins are expressed.

Proteome efficiency of ATP generating pathways

We next compared proteome efficiency of glycolysis versus respiration, defined as the ATP generation flux per mass of all glycolytic or respiratory enzymes (Fig. 2c). It is widely assumed that glycolysis, at the expense of being less energy efficient, is more proteome efficient than respiration^9–17. Our data on ATP flux and protein abundance, however, revealed a different picture. In I. orientalis, ~ 90% of ATP was made via respiration (Fig. 1a), with the proteome efficiency of respiration being more than 5-fold higher than for glycolysis (Fig. 2d). In aerobic glycolytic S. cerevisiae, despite its respiratory machinery being minimally engaged under glucose surplus, the proteome efficiency of respiration still slightly exceeded that of glycolysis (Fig. 2d). In highly respiratory naïve T cells, the proteome efficiency of respiration was more than 40-fold that of glycolysis, while in the more glycolytic activated T cells, respiration was 2-fold more efficient than glycolysis (Fig. 2d). Thus, across the above cell types, respiration is equally or more proteome-efficient than glycolysis.

We further assessed the proteome efficiency of ATP generation through fermentation (converting glucose to ethanol) and respiration (converting glucose to CO₂) by calculating a flux-partitioned proteome cost⁷, which counts glycolytic proteins in the cost of respiration (for providing pyruvate) and discounts flux diverted to biosynthetic precursors. In both yeasts and T cells, this flux-partitioned analysis again identified respiration as the more proteome-efficient ATP production pathway (Extended Data Fig. 4, a and b). Similarly, even if including all mitochondrial proteins as part of respiration’s proteome cost—an extreme approach that overlooks the many other functions of mitochondria— the most proteome-efficient energy generation pathway was respiration in I. orientalis, exceeding the efficiency of the best glycolytic pathway (glycolysis in S. cerevisiae) by 2.3-fold (Extended Data Fig. 4, c and d).

Proteome efficiency in nutrient-limited yeasts

We next set out to evaluate the generality of the greater proteome efficiency of mitochondrial respiration than glycolysis. As a complementary context to freely growing batch culture yeast, we explored nutrient-limited chemostat cultures, generating in-depth flux and proteomics data for both S. cerevisiae and I. orientalis under ≥ 12 chemostat conditions. Fluxes aligned closely with growth rate in both yeasts (Fig. 3a, Extended Data Fig. 5a), with the exception of increasing respiration and pentose phosphate pathway fluxes upon glucose-limitation of S. cerevisiae, which renders its metabolism similar to I. orientalis (Extended Data Fig. 5b). Under severe nutrient limitation, S. cerevisiae generates ATP mainly through respiration, switching to aerobic glycolysis with faster growth and adequate glucose (Fig. 3b). In contrast, I. orientalis consistently respires, even with rapid growth. Across nutrient limitation conditions, S. cerevisiae consistently manifested a large glycolytic proteome, and I. orientalis a large respiratory one (Fig. 3c). Across both yeasts, proteome efficiency fell with slower growth, reflecting spare enzyme capacity (Fig. 3, d and e). In I. orientalis, ATP production by respiration was always at least 5-fold more proteome efficient than by glycolysis. In S. cerevisiae, the proteome efficiency of these pathways shifted strongly as their use (but not enzyme levels) changed with environmental conditions. Overall, the best proteome efficiency for glycolysis (S. cerevisiae batch, μ = 0.39 h⁻¹) was about two-fold lower than the best proteome efficiency of S. cerevisiae respiration (fast carbon-limited growth, μ = 0.28 h⁻¹) and four-fold lower than the best proteome efficiency of respiration overall (I. orientalis batch, μ = 0.52 h⁻¹). These data support mitochondrial respiration in yeast being fundamentally more proteome-efficient than glycolysis.

Figure 3. Metabolic flux, proteome, and proteome efficiency in yeasts across different nutrient conditions.

(a) Genome-wide metabolic fluxes (from ¹³C-informed metabolic flux analysis) from yeasts grown in nutrient-limited chemostat (limiting nutrient: C, carbon/glucose; N, nitrogen/ammonia; P, phosphorus/phosphate) or nutrient-replete batch culture (B). Continuous cultures are from 4 or 5 independent chemostats grown at different growth rates (μ), each shown individually. Data are normalized to the geometric mean value across different conditions within each yeast.

(b) ATP fluxes from glycolysis and respiration across these nutrient conditions.

(c) Mass fraction of 8 proteome sectors across nutrient conditions.

(d-e) Proteome efficiency of ATP production by glycolysis and respiration in glucose deplete (C-limit) or glucose replete (all other chemostat) conditions in I. orientalis (d) and S. cerevisiae (e). Proteome efficiency is linearly regressed to growth rate and the slope (mean ± s.e.) in units of mol ATP/gProtein is shown.

Proteome efficiency in mammals

To explore whether ATP-generation by mitochondrial respiration is also more proteome efficient than glycolysis in mammals, we evaluated three contexts: cultured cancer cells, mouse tissues, and tumors in vivo. For cancer cells, we took advantage of published flux and proteomics data across 59 human cancer cell lines grown in vitro^35,36 (Fig. 4, a-b). Cancer cell lines devoted much more of their proteome to glycolysis than respiration (median 3-fold), but still made the majority of their ATP via mitochondrial respiration (median 1.4-fold of glycolysis), leading to respiration being about 4-fold more proteome efficient (Fig. 4c). Thus, despite cancer cells being highly glycolytic, respiration is their more proteome-efficient ATP generation pathway.

Figure 4. Proteome efficiency of mammalian cells, tissues and tumors.

(a-c) ATP flux (a), proteome allocation (b), and proteome efficiency (c) of glycolysis and respiration in NCI60 cancer cells cultured in vitro, using published flux and proteomics data^35,36.

(d) ATP flux from glycolysis and respiration in live mice, using data from Bartman et al based on 2-deoxyglucose and lactate assimilation kinetics³⁷. Tumors are flank-implanted k-Ras-driven pancreatic ductal adenocarcinoma (flank PDAC) and spleen infiltrated with Notch1-driven leukemia (leukemic spleen).

(e-g) ATP flux (e), proteome allocation (f), and proteome efficiency (g) of glycolysis and respiration in mouse tissues and tumors in vivo. ATP flux, mean ± s.d. reported in Bartman et al ³⁷. Proteome allocation, mean ± s.e.m., n = 3 mice (pancreas, PDAC, spleen, leukemic spleen), and mean ± s.e.m. of independent studies retrieved from PaxDb⁵¹. Proteome efficiency, Mean ± s.d. with error propagated from ATP flux and proteome fraction.

(h) Ratio of glycolytic to respiratory flux versus proteome fraction for I. orientalis, S. cerevisiae, T cells, NCI60 cancer cell lines, and mouse tissues and tumors. Results from linear regression across all contexts are shown.

(i) Corresponding proteome efficiencies to (h).

For tissues in vivo, we utilized recent measurements from our lab, in fasted mice, of respiration and glucose utilization³⁷. These measurements are based on TCA labeling dynamics and accumulation rates of 2-deoxyglucose phosphate, respectively, which together provide a good approximation of ATP production routes (Fig. 4d). Mouse tissues made > 95% of their ATP aerobically (i.e. respiratory ATP exceeds glycolytic by > 20-fold) (Fig. 4e). While they also expressed on average a 3-fold larger respiratory than glycolytic proteome (Fig. 4f), respiration is still roughly an order of magnitude more proteome efficient than glycolysis (Fig. 4g).

We used similar published data to assess glycolytic and respiratory rates in k-Ras-driven pancreatic ductal adenocarcinoma (PDAC) and spleen infiltrated with Notch1-driven leukemia (leukemic spleen) (Fig. 4d). New quantitative proteomic measurements enabled assessment of proteome efficiency (Fig. 4f, Ext Data Fig. 6). Compared to normal tissues, pancreatic cancer but not leukemic spleen involved a major proteome shift, including an increase in the glycolytic proteome (Fig. 4e, Ext Data Fig. 6), consistent with hypoxic signaling via Hif1α³⁸, whose protein level increased more than an order of magnitude in PDAC (Ext Data Fig. 6c). In both PDAC and leukemic spleen, glycolytic flux was upregulated, in leukemia apparently mediated by increased expression of key gating enzymes (GLUT3 and HK3) (Ext Data Fig. 6f). Importantly, despite upregulated glycolysis, both cancer types still produced > 80% of their ATP via respiration (Fig. 4e). The higher ATP production by respiration than glycolysis results, also for tumors, in respiration being much more proteome efficient than glycolysis (Fig. 4g).

Correlation between energy proteome and flux

Across all studied cell and tissue types, the ATP contribution from glycolytic relative to respiratory pathways was predicted by their relative proteome fraction, following a power law of order 1.5 (Fig. 4h). This implies that changes in glycolytic versus respiratory protein levels shape energy sources in a supra-linear manner. For example, cancer cell lines devoted more of their proteome to glycolysis than respiration while normal tissues did the opposite (median glycolysis/respiration proteome mass was 3 in cancer cells vs. 0.3 in mouse tissues) and the differential in ATP production routes was yet larger (median glycolysis/respiration ATP flux was 0.7 in cancer cells and 0.01 in mouse tissues). Apparently, partitioning of the energy proteome both directly impacts pathway fluxes and indirectly reflects the cell’s preference for which of the pathways to more fully engage. Proteome efficiencies, both of respiration and of glycolysis, were more than an order of magnitude lower in mammalian tissues than in cultured cells, presumably reflecting high reserve capacity in mammalian tissues to enable rapid response to organismal stressors (Fig. 4i). Across all the measured biological contexts, the superiority of respiration was striking (Fig. 4i; for flux-partitioned analysis that accounts for the glycolytic proteome required for pyruvate production to feed respiration, see Ext. Data Fig. 6g).

ATP generation routes and yeast growth rates

The superior proteome efficiency of respiration should translate into a growth advantage for respiratory over aerobic glycolytic yeast. Quantitatively, each percent of the proteome that is saved through greater efficiency produces faster growth. Based on the maximum observed proteome efficiencies of respiration ($e_R$, batch I. orientalis) and glycolysis ($e_G$, batch S. cerevisiae) of 1930 and 466 mmol ATP/hr/gProtein (Fig. 2d), respectively, we calculate that every unit increase of ATP flux made by glycolysis (in units of mmol ATP/hr/gDW) comes with a proteome cost of $\frac{1}{e_G}-\frac{1}{e_R}=0.16\text{\%}$ gProtein/gDW. Decreasing glycolytic flux from 27 mmol/hr/gDW in S. cerevisiae to 10 mmol/h/gDW in I. orientalis, a difference of 34 mmol ATP/hr/gDW (as glycolysis makes 2 ATP), corresponds to about 5% protein mass savings (or 10% of the proteome, as half of dry weight is protein). Based on this proteome savings, and literature data suggesting ~ 0.02 h⁻¹ gain per 1% proteome savings⁷, we expect a growth advantage about ~ 0.2 h⁻¹, in line with the experimental growth advantage of I. orientalis of 0.23 h⁻¹.

A growth difference between any particular pair of yeast could, however, arise for reasons unrelated to the proteome efficiency of ATP generation. As an orthogonal assessment of whether respiration favors faster yeast growth, we surveyed 23 yeast species spanning about 400 million years of evolution²⁶ (Fig. 5a). These include, in addition to S. cerevisiae and I.orientalis, two other industrially relevant species Scheffersomyces stipites and Kluyveromyces marxianus³⁹, and 19 species randomly picked from more than three hundred species across the budding yeast phylum. Comparing glucose uptake rates and growth across these yeasts (Fig. 5b in rich media, Ext. Data Fig. 7a in minimal media), faster glucose consumption did not robustly predict faster growth (Fig. 5c). Rather, growth rate correlated with glucose consumption rate until glucose consumption rate reaches about 1 g/L/hr/OD (about 10 mmol/hr/gDW), above which aerobic glycolysis set in (Ext. Data Fig. 7b) and the aerobic glycolytic yeasts grew slower than the respiratory ones. The fastest growing of the aerobic glycolytic yeast was actually S. cerevisiae, but it still grew more slowly than a cadre of respiratory yeast (Fig. 5c, Ext. Data Fig. 7a). These data are consistent with respiration, through its greater proteome efficiency, enabling faster yeast growth.

Figure 5. Growth and glucose consumption of evolutionarily divergent budding yeasts.

(a) Phylogenetic tree of the budding yeast phylum²⁶, with annotation for the studied species.

(b) Growth rates (top) and glucose consumption rates (bottom) of different budding yeasts cultured in rich YPD media containing 20g/L glucose. Mean ± s.e.m. (for n = 4 or 6 biological replicates), mean ± s.e. (error from regression of n = 4 time points for n = 2 biological replicates).

(c) Growth rate plotted against glucose consumption rate.

Other explanations for aerobic glycolysis

The proteome efficiency of respiration calls for an alternative explanation for the evolutionary persistence of aerobic glycolysis. We examined three proposed explanations for aerobic glycolysis: provision of biosynthetic precursors, need for NAD⁺ regeneration from NADH, and a limitation in energy dissipation rate^40,41. In both I. orientalis and S. cerevisiae, biosynthetic fluxes consuming glycolytic intermediates account for a minority of glucose carbon (31% in I. orientalis and 9% in S. cerevisiae) (Extended Data Fig. 1g), arguing against faster glycolysis being needed to support biosynthesis. Moreover, the respiratory yeast, I. orientalis, has higher not lower biosynthetic fluxes.

Several biosynthetic steps (such as de novo serine synthesis) use NAD⁺ as the electron acceptor, generating NADH. This NADH must be oxidized to NAD⁺ for biosynthesis to continue. Metabolomics revealed that I. orientalis has higher NAD⁺ and lower NADH concentration than S. cerevisiae, consistent with respiration being more, not less, effective at regenerating NAD⁺ (Ext Data Fig. 8a). Thus, neither biosynthetic demand nor NAD⁺ regeneration are likely drivers of aerobic glycolysis.

Another possibility is that aerobic glycolysis is beneficial because cells have limitations on the maximum rate of free energy dissipation⁴¹. While respiration is more efficient in generating ATP from glucose, in so doing it also liberates more free energy, which may be problematic for reasons including overheating. We thus compared glycolysis and respiration for energy dissipation per ATP produced and also examined the total cellular free energy dissipation rate (product of Δ_rG and flux) in S. cerevisiae and I. orientalis (Ext. Data Fig. 8b). Our analysis concurs with the prior literature regarding glycolysis being less energy dissipating than respiration for S. cerevisiae⁴¹. Notably, respiratory energy dissipation depends on the coupling of ATP synthesis to electron transport (represented by the ATP to oxygen ratio, i.e. P/O ratio). The lack of a proton-pumping Complex I in S. cerevisiae results in less coupling and thus high dissipation from respiration. However, most respiratory yeasts, including I. orientalis, have a proton-pumping Complex I. This fundamentally changes the energetics of respiratory ATP production such that free energy released per ATP produced from respiration is less than or similar as that from glycolysis. The result is that energy dissipated during ATP synthesis is indistinguishable within error for S. cerevisiae and I. orientalis, arguing against aerobic glycolytic yeast evolving to cope with energy dissipation limits (Ext. Data Fig. 8b). In addition, when ATP is used for biosynthesis, a majority of energy contained in it is eventually dissipated, therefore the gross energy dissipation (including ATP hydrolysis) is higher in I. orientalis, due to its faster overall ATP turnover, suggesting that S. cerevisiae is unlikely to be pushing up against some fundamental biological barrier of maximum dissipation.

Yeast competitive fitness

Several other explanations predict superior competitive fitness for aerobic glycolytic yeast. For example, aerobic glycolysis has been suggested to enable yeast to win the battle for limited glucose^6,42,43 or to poison competitors with ethanol^44,45. To explore such possibilities, we carried out competitive growth experiments between I. orientalis and S. cerevisiae (Fig. 6a). Under aerobic conditions, the more respiratory I. orientalis outcompeted the more glycolytic S. cerevisiae in ethanol, limited glucose, as well as surplus of glucose (batch culture and ammonia- and phosphate-limited chemostats). The same trend was even observed in batch cultures fed sucrose, which can be directly metabolized by S. cerevisiae but not I. orientalis, which apparently wins by leaching off glucose and fructose liberated by its competitor. Thus, under aerobic conditions, including ones where S. cerevisiae elects to engage in aerobic glycolysis, respiration favored competitive fitness.

Figure 6. Yeasts’ fitness under aerobic and anaerobic conditions.

(a) Relative fitness of S. cerevisiae in competitive co-culture with I. orientalis under different carbon sources (ethanol, glucose, sucrose), limiting nutrients (C, P, or N-limited chemostats) or cyclic oxygen depletion with numbers indicating the fraction of time spent in the anaerobic duty cycle. Mean ± s.e.m., n = 3 biological replicates (carbon sources) or 4 biological replicates (chemostats and cyclic oxygen depletion).

(b-c) Specific growth rates and glucose consumption rates for batch-cultured S. cerevisiae and I. orientalis with oxygen, without oxygen, or with 10 μM respiratory inhibitor, antimycin. Mean ± s.e.m., n = 4 (+O₂), 7 (+O₂+antimycin), 7 (S. cerevisiae -O₂), 10 (I. orientalis -O₂).

(d) Corresponding proteome allocation as in (b). Mean ± s.e.m., n=3 (except S. c. + antimycin, n = 2).

(e) Energy charge in response to acute addition of 10 μM antimycin in S. cerevisiae and I. orientalis cultured in glucose minimal media. Lines connect mean (n=4) at each time point.

(f) Dependence of growth rate and ethanol flux on metabolic proteome allocation. Values are calculated from a coarse-grained model of metabolism and growth rate, parameterized using the experimentally measured proteome efficiencies of glycolysis and respiration with total metabolic proteome mass optimized to maximize growth rate under a fixed mass ratio between glycolytic and respiratory proteome (Supplementary Note).

(g) Growth rates of 16 yeasts cultured in YPD and glucose with or without shaking (aerated and settled, respectively). Clear bars are aerated growth rates from Fig. 5b. Color bars are settled growth rates, mean ± s.e.m., n=4 or 8.

(h) Growth rate ratio between settled and aerated growth regressed to aerobic glucose consumption rate in 16 yeasts described in (f). Linear correlation coefficient R² = 0.63, p = 0.00014.

Hypoxia

A crucial metabolic factor is oxygen availability. Due to oxygen’s limited solubility and diffusion through water, yeasts experience oxygen limitation environmentally (Extended Data Fig. 9a). S. cerevisiae also experienced it repeatedly over centuries of beverage making⁴⁶. We examined the impact of oxygen deprivation on S. cerevisiae and I. orientalis fitness. In either fully or cyclically oxygen-depleted co-cultures, S. cerevisiae outcompeted I. orientalis (Fig. 6a). I. orientalis with an engineered mild respiratory defect (ΔNde) was outcompeted by engineered fermentation-defective I. orientalis (ΔPdc) under aerobic conditions, but the results flipped under oxygen depletion (Extended Data Fig. 9b). Glucose-fed S. cerevisiae grew at a similar rate irrespective of the presence of oxygen or the electron transport chain inhibitor antimycin (Fig. 6b) with minimal impact of oxygen availability on glucose consumption rate or proteome allocation (Fig. 6c-d). In contrast, I. orientalis grew about 60% slower when oxygen was removed or antimycin added (Fig. 6b). Even this relatively sluggish anaerobic growth required substantial metabolic and proteome remodeling: doubling of glycolytic flux and tripling of glycolytic proteome fraction (Fig. 6c-d). Moreover, acute respiratory inhibition in I. orientalis but not S. cerevisiae caused severe energy stress (Fig. 6e). Thus, S. cerevisiae’s glycolytic proteome and associated aerobic glycolysis is inefficient during aerobic growth, but enhances fitness upon decreased oxygen supply.

Balancing efficiency and robustness

Both I. orientalis and S. cerevisiae can tailor their respiratory versus glycolytic enzyme expression to environmental conditions. Glucose-limited S. cerevisiae modestly shrank its glycolytic proteome (Fig. 3c). Oxygen-deficient I. orientalis expanded its glycolytic machinery at the expense of translation machinery (Fig. 6d). This tailoring, however, is incomplete: I. orientalis partially retained respiratory proteins in hypoxia, while S. cerevisiae retained high glycolytic enzymes in aerobic conditions (Fig. 6d, Fig. 3c).

To explore the consequences of incomplete proteome tailoring, we assembled a coarse-grained quantitative model of yeast growth and metabolism, where growth is limited both by translational machinery and by ATP generation machinery (fermentative or respiratory). The model was parameterized using experimentally measured proteome efficiencies of translation and of ATP generation by fermentation and respiration, as well as the flux split ratio for biosynthesis (Extended Data Fig. 9c and Supplementary Note). In this modeling approach, more efficient ATP production results in more room for translation machinery and thus faster growth. Using it, we identified the respiratory and glycolytic proteome allocation that optimizes growth rate under fully aerobic or anaerobic conditions. This analysis revealed that both I. orientalis and S. cerevisiae’s proteomes lie between the optimal aerobic and anaerobic proteome (Extended Data Fig. 9d), suggesting an evolutionary drive towards flexibility or robustness.

Outputs of the model include predicted glucose uptake, oxygen uptake, and ethanol secretion rates. The model was set up to find, for a fixed glycolytic proteome mass (Extended Data Fig. 9e) or ratio of glycolytic to respiratory enzyme expression (Fig. 6f), the metabolic fluxes (glycolytic and respiratory) and proteome partitioned between metabolism and translation that optimizes growth. This analysis predicts that yeast optimized for anaerobic conditions will, even when given oxygen, secrete ethanol, because it is efficient for them to use their existing glycolytic capacity to make more ATP as growth accelerates (Extended Data Fig. 9e). More generally, the model predicts two alternative modes of glucose-fed aerobic metabolism, depending on the ratio of glycolytic-to-respiratory proteome: (i) complete respiration when glycolytic capacity (production of NADH and pyruvate from glucose, beyond the pyruvate needed for anabolism) is less than or equal to respiratory capacity (to oxidize those glycolytic products) and (ii) aerobic glycolysis when glycolytic capacity exceeds respiratory capacity (Fig. 6f). Given the superior proteome efficiency of respiration, metabolic optimality involves perfect balancing of glycolytic and respiratory capacity, i.e. expressing the minimum glycolytic machinery to provide required anabolic substrates and respiratory fuel. When glucose is the only carbon source, excessive respiratory relative to glycolytic capacity is perilous: Respiration fails to run maximally due to lack of fuel. In contrast, whenever glycolytic capacity exceeds respiratory capacity, both pathways can be used fully, with the outcome being aerobic glycolysis.

Aerobic glycolysis and micro-aerobic fitness

Neither I. orientalis or S. cerevisiae evolved for strict anaerobicity. Yet micro-anaerobic conditions often occur, including in “settled” liquid culture, i.e. cultures without sufficient mechanical agitation (Extended Data Fig. 9a). We examined the growth of divergent yeasts in such settled cultures. The top 16 fast growing yeasts, with aerobic growth rate of 0.3 – 0.6 h⁻¹, all showed slower growth rates in settled culture (Fig. 6g). But the growth suppression was less for aerobic glycolytic yeasts. Similarly, the adapted ΔNde grows slower than wild-type I. orientalis in aerated culture but not in settled culture (Extended Data Fig. 9f). Micro-anaerobic fitness, defined as the ratio of growth rate in settled culture versus aerated culture, correlated with basal aerobic glycolytic flux (R² = 0.63, p = 0.0001, Fig. 6h). Therefore, a benefit of aerobic glycolysis appears to be readiness for oxygen limitation.

Discussion

We generated fluxomics and quantitative proteomics data from yeasts (in total 30 physiological conditions), naïve and activated primary murine T cells, and new proteomics data for paired healthy and tumorous mouse tissues. Based on these data and literature data, we determined the proteome efficiency of both glycolysis and respiration across two yeasts, T cells, 59 cancer cell lines, 10 normal mouse tissues, mouse PDAC, and mouse leukemic spleen. Only batch-grown S. cerevisiae (consistent with prior research¹⁷), activated T cells, and a few cancer cell lines manifested similar proteome efficiencies for glycolysis and respiration (within 2-fold). In every other instance, respiration was substantially more proteome efficient (Fig. 4i).

Earlier theoretical analysis point to “rate-yield tradeoff” intrinsic to enzyme catalysis. Specifically, a low energy yield pathway would consume more thermodynamic driving force and thus decrease the amount of enzyme wasted in the reverse flux^6,47,48. The theory predicts nicely the divergent choice in bacteria of high-yield versus low-yield glycolytic pathways, both leading to lactate production. However, choice between glycolysis and respiration may not be subject to such tradeoff. The low ATP yield of glycolysis does not necessarily imply a greater thermodynamic driving force. Compared to respiration that makes low-energy CO₂ (glucose to CO₂, ΔG’ = −2800 kJ/mol glucose, where ΔG’ refers to ΔG with all metabolites at common physiological concentrations), glycolysis preserves a lot of glucose energy in ethanol (glucose to ethanol, ΔG’ = −250 kJ/mol glucose) (Ext. Data Fig. 8b). Thus, respiration can be more thermodynamically driven while glycolysis is sometimes close to equilibrium⁴⁹.

The chemistry used by respiratory and glycolytic enzymes also differ significantly. Glycolysis involves soluble enzymes colliding with substrates to make and break of carbon-carbon and carbon-oxygen bonds, with some reactions such as enolase and pyruvate decarboxylase intrinsically chemically challenging. It occurs in the cytosol and may be subject to constraints on substrate concentrations that limit reaction kinetics⁴⁸. In contrast, the reactions of respiration occur primarily in the mitochondrial matrix and on the mitochondrial inner membrane, whose two-dimensional structure confines and colocalizes the large macromolecular complexes of the electron transport chain. Moreover, from a fundamental chemical reaction perspective, the electron transport chain mainly transfers protons and electrons, which involves much lower energy barriers (i.e. is intrinsically faster) than typical covalent chemical reactions, like those of glycolysis, whose transition states involve substantial distortions of heavy atom geometries⁵⁰. These basic differences in both reaction localization and type favor the proteome efficiency of respiration.

Despite its low energy and proteome efficiency, glycolysis becomes crucial when respiratory energy production is hampered. In that case, some cells increase their glycolytic proteome, mediated by mechanisms such as hypoxia or energy sensing (e.g., in pancreas cancer, Fig. 4f, and I. orientalis, Fig. 6d, respectively). Many other proliferative cell types express copious glycolytic proteins regardless of oxygen availability and default to glycolytic energy production (e.g. S. cerevisiae and cancer cells, Fig. 4h). In normal mammalian tissues that primarily use respiratory energy, there is also substantial glycolytic enzyme expression likely to meet acute energy demands, be ready for hypoxia, and contribute to systemic blood glucose homeostasis. Such constitutive expression of glycolytic machinery aligns with a general propensity for cells of a given type to have a characteristic metabolic proteome that varies only modestly across conditions (Fig. 2b, Fig. 3c). A benefit of such proteome constancy is that cells are prepared in advance for changing metabolic environments, with the high glycolytic enzyme expression leading to preparedness for hypoxia (Fig. 6). Our quantitative modeling supports high glycolytic enzyme expression per se being sufficient to render aerobic glycolysis the preferred metabolic flux mode (Fig. 6f, Ext. Data Fig. 9 c-e). Thus, expression of a large glycolytic proteome that ensures adequate substrate to feed respiration and prepares cells for hypoxia underlies the seemingly paradoxical phenomenon of aerobic glycolysis.

Online Methods

Yeast strains and cultivation

Strains

The I. orientalis strain used in this study, SD108, is originally isolated from rotting bagasse²². Two prototrophic S. cerevisiae strains were used: CEN.PK2 (MATa) was a gift from Dr. Jose Avalos, Princeton, NJ, USA; and FY4 (DBY11069, MATa) that was derived from S288C background ⁵². S288C strain naturally carries mutations that affect the gene HAP1 involved in respiratory regulation ⁵³. I. orientalis ΔNde (Δg1781) and complex I mutant (Δg1702) was created with CRISPR/Cas9 editing ^24,54 as described in reference ²³.

I. orientalis ΔPdc was generated in this study by deleting the only pyruvate decarboxylase gene (PDC) using a previously developed CRISPR/Cas9 tool⁵⁴. To target PDC deletion, a single-guide RNA (sgRNA) with the sequence 5’-AATGCCGGCTACGAAGCTGA-3’ was designed, which contains a TGG sequence as a protospacer adjacent motif (PAM) at its 3’ end. Specificity of sgRNA was verified by BLAST analysis against the whole genome sequence available on the JGI MycoCosm ⁵⁵ (https://mycocosm.jgi.doe.gov/Issorie2/Issorie2.home.html). A 200-bp donor DNA, with sequence homology to regions upstream (100 bp) and downstream (100 bp) of the targeted site, introduces a 14-bp deletion within the PDC gene’s coding region, leading to a frameshift mutation that was verified with PCR and Sanger sequencing. Both the sgRNA and donor DNA were synthesized by Integrated DNA Technologies (Coralville, IA) and subsequently assembled into a CRISPR/Cas9 tool plasmid using the NEBuilder HiFi DNA Assembly (New England Biolabs, MA). Donor DNA sequence: TTGGTGTTCCTGGTGATTTCAATTTGGCATTGTTGGACCACGTTAAGGAAGTTGAAGGCATTAGATGGGTCGGTAACGCTAACGAGTTGAATGCCGGCTAATGCAAGAATCAATGGATTTGCATCCCTAATCACCACCTTTGGTGTCGGTGAATTGTCTGCCGTCAATGCCATTGCAGGTTCTTATGCTGAACACGTCCC. Primer sequences for verification: 5’-TGTCGTTATCCTTTTGGCATTGACG-3’ (sense), and 5’-TCTGCCTTCTTGACCATTTCAACAAC-3’ (antisense).

Other budding yeast species are obtained from the ARS culture collection (NRRL) and maintained as instructed. NRRL accession number is shown in Supplementary Table 19.

Media

If not specified, yeasts were cultured in minimal media containing 20g/L glucose 6.7g/L yeast nitrogen base (YNB) without amino acids (pH = 5) (Sigma, Y0626). For nutrient limitation, YNB without amino acid or ammonium or phosphate (MP Biomedicals, 114029622) was used as the mineral base and supplemented with nutrient specified in Supplementary Table 1 (adapted from reference ⁵⁶). For cultivation in rich media, YPD was made with 10g/L yeast extract, 20g/L peptone, and 20g/L glucose. All media were filter sterilized through a 0.22μm pore filter.

Culture

Cell density was quantified by absorption at 600nm (OD600) using a UV-VIS spectrophotometer (GENESYS 10, Thermo) after 10-fold dilution. For batch culture, the yeasts were first cultured overnight in minimal media to achieve final OD600 about 4 for I. orientalis and 3 for S. cerevisiae. For ¹³C isotope tracing, the yeast was adapted in the same ¹³C culture overnight to ensure isotopic steady state in the biomass. The overnight culture was then inoculated into 4mL media in 14mL round-bottom Falcon culture tube tilted at 45° angle or 20~40mL media in 150mL vented baffled culture flask at initial OD about 0.05~0.2, and cultured in a shaker at 250rpm and 30°C. Pseudo steady state is usually maintained below OD600 = 1.5.

For nutrient limited continuous culture, S. cerevisiae FY4 or I. orientalis SD108 was cultured in a home-built miniaturized multichannel bioreactor with a working volume of 20mL following the previous procedure ⁵⁷. 200μL overnight culture was inoculated in the culture tube and allowed to grow overnight before starting the continuous flow of media. The flow rate is controlled by a multichannel peristaltic pump (205S/CA12, Watson-Marlow, MA, US) and manifold tubing with proper internal diameter. The flow rate was calibrated each time by monitoring the effluent, and the volume of the culture was adjusted within ± 2mL to match the desired dilution rate. The cultures were mixed by sparging with 7.5 standard liters per min of water-saturated air for aerobic culture. The culture was maintained under continuous flow for at least 48hrs to achieve steady state. The final pH was measured to be about 3.5. Four dilution rates (0.08, 0.16, 0.22, 0.28 h⁻¹) were used for S. cerevisiae and five (0.12, 0.18, 0.23, 0.34, 0.45h⁻¹) for I. orientalis for each nutrient limitation.

Oxygen-depleted batch culture was done in the same home-build bioreactor with continuous sparging of 7.5 standard liters per min of water-saturated nitrogen. For antimycin treatment, a concentrated stock of antimycin (100mM in DMSO) was first diluted 100X with water and then added to the culture at 100X dilution.

Settled culture was done in 96-well deep assay plate with 1mL culture volume. The cultures were left in 30°C incubator and only mixed every 3 hours for growth monitoring. Dissolved oxygen was assayed with an oxygen-sensing plate. Specifically, exponential phase yeast culture was added to 100uL fresh media in a 96-well plate coated with phosphorescence oxygen sensor at the bottom (OxoPlate, OP96U, PreSens Precision Sensing GmbH, Germany). The culture was allowed to accommodate for 10min and then analyzed by a plate reader with or without fast shaking at 30 °C (BioTek, Synergy HT). Calibration and measurement were done following the manufacturer’s procedure.

Mice

All mouse experiments were approved by the Institutional Animal Care and Use Committee at Princeton University (protocol number 3111). C57BL/6 mice (Charles River Laboratories) were used CD8+ T cell isolation. Female mice aged between 8 and 12 weeks were used unless otherwise noted. Mice were housed under a normal light cycle (7 AM to 7 PM) at room temperature of 20–26ºC and a humidity of 40–60%, with water and food (PicoLab Rodent Diet 5053, LabDiet) provided ad libitum.

Healthy and tumorous tissues were obtained from mice described in an earlier study ³⁷, including spontaneous pancreatic adenocarcinoma (GEMM PDAC, Pdx1-cre;LSL-Kras-G12D/+;Trp53^fl/fl) mice, Syngeneic pancreatic adenocarcinoma allograft tumors (flank PDAC, established by implanting tumors from Pdx1-cre;LSL-Kras-G12D/+; LSL-Trp53-R172H/+ mice subcutaneously into the mouse flank), primary T-cell acute lymphocytic leukemia (leukemic spleen, NOTCH1-induced primary transplanted into sub-lethally irradiated recipients).

Murine CD8+ T cells

Isolation, culture, and stimulation of mouse naïve CD8+ T cells

Procedures were adapted from an earlier study⁵⁸. Briefly, mouse spleens were harvested and pooled as single-cell suspensions by manual disruption and passage through 70-μm cell strainers into RPMI 1640 media. After RBC lysis (eBioscience, 00–4300-54) and another passage through 70-μm cell strainers into PBS supplemented with 0.5% BSA and 2 mM EDTA, naïve CD8⁺ T cells were purified by magnetic bead separation using naïve CD8a⁺ T Cell Isolation Kit, mouse (Miltenyi Biotec, 130–096-543) following vendor instructions. Approximately 2~3E6 purified naïve CD8+ T cells can be obtained each mouse. Cell number was counted using Trypan blue staining and the Countess system (Invitrogen).

Cells were cultured in complete RPMI media made with RPMI 1640 (11875119, ThermoFisher), supplemented with 10% FBS, 100 U ml⁻¹ penicillin, 100 μg ml⁻¹ streptomycin and 55 μM 2-mercaptoethanol). Cells were maintained at 1E6/mL in 1mL media in 12-well unless specified. Naïve T cells were either rested in complete RPMI media supplemented with recombinant IL-7 (50 U ml⁻¹, Peprotech, 217–17) or stimulated for 24 h with plate-bound anti-CD3 (10 μg ml⁻¹, Bio X Cell, BE0001–1) and anti-CD28 (5 μg ml⁻¹, Bio X Cell, BE0015–1) in complete RPMI media supplemented with recombinant IL-2 (100 U ml⁻¹, Peprotech, 217–12). All experiments on activated T cells were performed 24h post stimulation unless otherwise noted. Within this time window cell size expanded but no obvious increase in cell number was observed.

Flow cytometry

Purity of naïve CD8+ T cells (98%) and expression of activation markers were verified by flow cytometry. Specifically, cells were collected, washed with staining buffer (PBS + 2% FBS) and stained with the viability dye Live/Dead Aqua (Thermo Fisher, L34966) according to the manufacturer’s instructions. Cells were then washed with staining buffer and stained for surface markers on ice for 30 min: CD4 (APC-Cy7, 1:100, clone RM4–5, BD Biosciences, 565650), CD8a (PerCP-Cy5.5, 1:100, clone 53–6.7, BD Biosciences, 551162), CD25 (APC, 1:100, clone PC61, BD Biosciences, 557192), CD44 (PE-Cy7, 1:100, clone IM7, BD Biosciences, 560569), CD62L (PE, 1:100, clone MEL-14, BD Biosciences, 561918) and CD69 (FITC, 1:250, clone H1.2F3, BD Biosciences, 557392). All flow cytometry was analyzed with an LSR II flow cytometer (BD Biosciences) and FCS Express 7.12 (De Novo Software). The following gating strategy was used: FSS/SSC lymphocyte gate (95%), singlet gate (98%), live/dead (98%), CD8+ (using CD4 vs CD8, 99%), naive (CD62L vs CD44, 99%), activated (CD69 vs CD25, 90%).

¹³C isotope tracing and metabolite extraction

For yeasts, two glucose tracers were used for flux analysis: [U-¹³C₆] glucose and [1,2-¹³C₂] glucose, each was mixed with unlabeled glucose to achieve 1:1 molar ratio (50% enrichment). Batch cultures were adapted in the tracer media overnight before allowing to grow in fresh media for more than 3 generations. Continuous cultures were cultured in tracer media for the whole experimental period. 13C mass isotopomer distribution in about 40 metabolites was then analyzed by LC-MS from extracted polar metabolites and biomass hydrolysates (see LC-MS assay for measuring biomass).

Isotope tracing in T cells was done with 3 isotope tracers: [U-¹³C₆]glucose, [1,2-¹³C₂]glucose, and [U-¹³C₅]glutamine in RPMI1640 media (USBio R9011, with correct supplementation, pH = 7.4) and 10% dFBS (ThermoFisher, 26400044) and 1% penicillin and streptomycin. 24hrs post isolation, T cells were pelleted, washed, and resuspended in the tracing media, and cultured for 5hrs before metabolite extraction.

Yeast metabolites were extracted by chilled solvent following rapid vacuum filtration. Specifically, a total amount of cell culture equivalent to 3mL at OD600 = 0.8 was extracted. Batch cultures were extracted at OD600 between 0.6 and 1.0. Yeast culture was vacuum filtered using Nylon membrane filters (0.5μm pore size, 1213776, GVS Magna^™) on a fritted glass support of vaccum filter flask. The membrane with the yeast pellet was then quickly immersed in 1.5mL metabolite extraction solvent (40:40:20 acetonitrile:methanol:water with 0.5% formic acid, precooled in −20°C) in a petri dish. After about 1min incubation on ice, the extract was neutralized by 132uL 15.8% (w/v) NH₄HCO₃. The extract was stored in −80 ºC before LC-MS analysis. The extract was then centrifuged at 21300Xg at 4°C to obtain supernatant ready for LC-MS analysis for polar metabolites.

T cell metabolites were extracted adapting a previous procedure⁵⁹. Specifically, 3E6 naïve cells or 1E6 activated cells were pelleted (6000rpm, 30sec, room temperature). Media was quickly removed, followed by immediate addition of 50uL cold metabolite extractionsolvent. 1min after extraction, 4.4uL NH₄HCO₃ was added to neutralize the extract. Supernatant of the extract was used for polar metabolites, while the insoluble fraction was used for analyzing isotope labeling in biomass components.

To obtain isotope labeling in biomass, yeast pellet (1OD·mL, washed with water) or the biomass remnant in T cell (insoluble fraction from T cell metabolite extract, washed with cold methanol), hydrolyzed in 100uL 2M HCl at 80ºC for 2hrs. 10uL of the hydrolysates were dried under N₂, resuspended in LC-MS solvent (40:40:20 ACN:MeOH:H2O) and analyzed by LC-MS.

Metabolite analysis by LC-MS

HILIC LC-MS for polar metabolites

Separation of polar metabolites was achieved with hydrophilic interaction chromatography (HILIC), using a Vanquish UHPLC system (Thermo Fisher Scientific, CA, US) and an XBridge BEH Amide column (2.1 mm x 150 mm, 2.5 mm particle size, 130 Å pore size; Waters, Milford, MA). The LC runs at flow rate of 150μL/min with a 25 min solvent gradient as following: 0 min, 85% B; 2 min, 85% B; 3 min, 80% B; 5 min, 80% B; 6 min, 75% B; 7 min, 75% B; 8 min, 70% B; 9 min, 70% B; 10 min, 50% B; 12 min, 50% B; 13 min, 25% B; 16 min, 25% B; 18 min, 0% B; 23 min, 0% B; 24 min, 85% B; 30 min, 85% B, where solvent A is 95:5 water:acetonitrile with 20 mM ammonium hydroxide and 20 mM ammonium acetate, pH 9.4, and solvent B is acetonitrile. Autosampler temperature was 4ºC, column temperature was 25 ºC, and injection volume was 10μL. LC was coupled to a quadrupole-orbitrap mass spectrometer (Q Exactive, Thermo Fisher Scientific, CA, US) via electrospray ionization. The mass spectrometer operates in negative and positive ion switching mode and scans from m/z 70 to 1000 at 1 Hz and 140,000 resolution, with additional selected ion monitoring scan from m/z 650 to 770 for NAD(P) cofactors. Due to small sample size of T cells, to avoid ion suppression at phosphate or pyrophosphate, scans were broken into 4 (70–96.5, 97.5–176.5, 177.5–194.5, and 195.5 – 1000), thus labeling in citrate and aconitate cannot be detected. Data was collected by XCalibur (Thermo Fisher Scientific).

Reverse-phase LC-MS for saponified fatty acid

Saponified fatty acids were analyzed by liquid chromatography (Accela U-HPLC) coupled with orbitrap mass spectrometer (Exactive, Thermo Fisher Scientific, CA, US). LC separation was done by reverse-phase ion-pairing through a Luna C8 column (150 × 2.0 mm², 3 μM particle size, 100 Å pore size; Phenomenex) with a solvent gradient of 0 min 80% B; 10 min, 90% B; 11 min, 99% B; 25 min, 99% B; 26 min, 80% B; 30 min, 80% B, where solvent A is 10 mM tributylamine + 15 mM acetic acid in 97:3 H2O:methanol, pH 4.5, and solvent B is methanol. The flow rate was 250 μL/min and column temperature 25 °C with an injection volume of 5 μL. The MS scans were in negative-ion mode with a resolution of 100,000 and scan range of m/z 120–600. Data was collected by XCalibur (Thermo Fisher Scientific).

Metabolite quantitation by LC-MS

Raw LC-MS data were converted to mzxml format by ProteoWizard (https://proteowizard.sourceforge.io)⁶⁰ (version 3). Peak picking and quantitation were done in the El-Maven software (v.0.4.1, Elucidata). For comparing across nutrient conditions, samples were extracted and analyzed in the same day to reduce batch effect. Relative fold change of each metabolite was quantified by relative peak area top in the chromatogram. For comparing between different yeasts in batch culture, ¹²C labeled S. cerevisiae was 1:1 mixed with ¹³C labeled I. orientalis, and ¹³C labeled S. cerevisiae was 1:1 mixed with ¹²C labeled I. orientalis. For each compound, ratio between labeled and unlabeled peaks was used for relative quantitation. For samples with ¹³C labeling, natural isotope abundance was corrected using AccuCor ⁶¹ (https://github.com/lparsons/accucor).

Energy charge ratio

Metabolite concentration were obtained from LC-MS measured relative metabolite change and basal intracellular concentrations reported earlier⁶² (ATP = 1.9mM, ADP = 0.45mM, AMP = 0.05mM). Energy charge ratio is calculated as ([ATP]+[ADP]/2)/([ATP]+[ADP]+[AMP]).

Determining major fluxes in yeast

Determining metabolite concentrations in spent media

Glucose, ethanol, acetate, succinate, and glycerol in spent media were measured with ¹H-NMR (500 MHz Advance III, Bruker, MA, US). 50mM TMSP-d₄ internal standard was 1:10 diluted into the spent media for internal reference. A fresh media sample was also included to calibrate TMSP. ¹H-NMR spectra were collected using the following acquisition parameters: TD = 65536, NS = 64, D1 = 5s, O1P = 4.68, P1 = 11.69, P12 = 2400, SPW1 = 0.002, SPNAM1 = Gaus1_180r.1000. Chemical shift used for quantification: 0 ppm (s, 9H) for the TMSP standard, 3.22 ppm (dd, 1H) for glucose, and 1.17 ppm (t, 3H) for ethanol, 2.07ppm (s, 3H) for acetate, 2.60 (s, 4H) for succinic acid, and 3.64 (m, 4H) for glycerol. Quantitation was done in MestReNova.

Glucose concentration was also determined by a biochemistry analyzer (2900, YSI, OH, US). Spent media with initial glucose concentration of 20g/L was measured with 4-fold dilution to be within linear range. Each sample was measured with at least two technical replicates.

Growth rate and extracellular fluxes for pseudo-steady-state batch culture

Growth rate (μ) and metabolite fluxes (j) in batch culture were determined by sampling the culture at least four times (t) during the exponential growth phase, starting from OD600 about 0.1 after allowing the culture to adapt in the fresh media (about 1 hrs in aerobic culture and 4hrs in anaerobic or antimycin treatment), to OD600 about 1.5 or before half of the glucose was consumed. At each time point, OD600 was measured and supernatant was saved for analysis of metabolite concentration (c). Growth rate (μ) was determined with linear fitting: μ = slope (ln OD ~ t), while extracellular flux (j) was the product of growth rate and the slope of c ~ OD, j = μ * slope (c ~ OD). The resulting j is in unit of mmol/L/OD600/h, which is then converted to mmol/gDW/h using the OD-to-biomass conversion factor determined in biomass analysis (e.g. for batch culture, this conversion factor is around 0.35 gDW/L/OD600 for both yeasts). Errors are determined by propagating error from the linear regression.

Oxygen consumption rate

Oxygen consumption rate was measured for batch culture with a Clark-type dissolved oxygen probe (B40PCID, 89231–624, VWR, PA, US). The culture was kept in a glass chamber and temperature was maintained by water bath. The culture was fully oxygenated first, and then sealed with the temperature-equilibrated probe. Dissolved oxygen was measured every 20sec for 5 min or until oxygen drops to 60% saturation. During the measurement, the culture was gently mixed by magnetic stirrer. The culture density used for measurement was OD600 = 0.2 ~ 0.3 for I. orientalis and OD600 = 0.6 ~ 0.8 for S. cerevisiae. Oxygen consumption rate was then calculated by linear fitting of the oxygen concentration change over time, and normalized by the cell density.

Flux determination in continuous culture

After steady state was reached, the continuous culture was sampled by collecting 1mL effluent at least three times over 12hr. For each sampling, OD600 was measured and remaining glucose concentration was determined by YSI biochemistry analyzer. The whole culture was then cooled on ice and centrifuged in 4°C. The metabolite concentration (c) was determined in the supernatant. Fluxes (j) were then calculated as j = dr * (c₀-c), where dr is the dilution rate, c₀ is the initial concentration in the media, and then normalized by biomass concentration.

Dry weight and biomass composition

We first determined the dry weight and biomass composition in a reference yeast, and then developed an LC-MS assay using this reference yeast as an internal standard to quantify other samples. Specifically we first measured DNA, RNA, protein, and carbohydrate in a reference condition, S. cerevisiae in carbon limitation at 0.1h⁻¹, using method described previously ²³. Briefly, protein content was determined using the Biuret method with BSA (23209, Thermo Fisher, MA, US) calibration. Cell pellet equivalent to 1mL of OD600 = 1 was washed and lysed in 300μL 1M NaOH at 98 °C for 5min. 100μL 1.6% CuSO₄ was then added to the lysate and absorbance at 555nm was used to quantify protein concentration. For RNA quantitation, the cell pellet was lysed in 300μL 0.3M KOH at 37°C for 60 min, and then 100μL 3M HClO₄ was added to precipitate DNA and protein. The supernatant was combined with 600uL 0.5M HClO₄ used to wash the precipitant, and RNA content was then determined by the absorption at 260nm with pathlength correction (1cm), using an extinction coefficient of 31 μg/mL/A260. For DNA, cell pellet equivalent to 10mL of OD600 = 1 was hydrolyzed with 500μL 1.6 M HClO₄ for 30 min at 70˚C, and allowed to react with 1 mL diphenylamine reagent (0.5 g diphenylamine in 50 mL acetic acid, 0.5 ml 98% H₂SO₄, and 0.125 ml 3.2% acetaldehyde water solution) at 50˚C for at least 3 hours. Absorption at 600nm was measured from the supernatant, and used to quantify DNA concentration with the calibration of a purified DNA standard (15633019, ThermoFisher Scientific, MA, US).

Protein, DNA, RNA and carbohydrate in biomass in other samples are then measured by acidic hydrolysis with reference to ¹³C labeled S. cerevisiae, which is cultured in carbon limitation with [U-¹³C₆] glucose at 0.1h⁻¹. To determine biomass in a given sample, three replicates of 1mL culture was pelleted and each combined with an aliquot of ¹³C S. cerevisiae equivalent to 1mL of OD600 = 1. The pellet was washed with water twice, and hydrolyzed in 100μL 6M HCl at 80 °C and 300 rpm for 2h with a thermomixture. The hydrolysate was then centrifuged, and 8μL of supernatant was dried with nitrogen gas, and re-dissolved in 80μL 40:40:20 acetonitrile:methanol:water for LC-MS analysis. The detected monomers were categorized into components of protein, DNA, RNA, and carbohydrate (Supplementary Table), and the ¹²C/¹³C ratio from each category was averaged to obtain concentration relative to the reference condition.

Lipid in biomass is analyzed by saponification and quantified by spiking in a mixture of ¹³C fatty acids of highest abundance in yeast. Specifically, 3 replicates of 1mL culture was pelleted, and saponified in 1mL 0.3M KOH in 10:90 water:methanol containing internal ¹³C standard of 40μM [U-¹³C₁₆] palmitate, 40μM [U-¹³C₁₈] oleate, and 20μM [U-¹³C₁₈] linoleate for 1h at 80 °C. The mixture was then acidified by 100μL formic acid and extracted twice with 1mL hexane. The upper layer was separated, dried under nitrogen gas, redissolved in 100μL 1:1 acetonitrile:methanol, and analyzed by reverse phase LC-MS. Fatty acids were quantified by ¹²C/¹³C ratio for the three with internal reference, and by MS peak intensity for other fatty acid species.

Determining major fluxes in murine T cells

Oxygen consumption rate and respiratory capacity

T cell oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were measured using a Seahorse XFe96 Extracellular Flux Analyzer using published procedure with modification⁶³. Specifically, 2.5×10⁵ naïve T cells or 1×10⁵ activated T cells were plated in poly-D-lysine-coated XF96 microplates (103729–100, Agilent) in Seahorse RPMI medium (103576–100, Agilent) supplemented with 10mM glucose and 2mM glutamine. The cells were kept in assay media for less than 4hr throughout the entire procedure. Cellular bioenergetics were assessed with the manufacturer’s Mito Stress test kit (103015–100, Agilent) through the sequential addition of pyruvate (1mM), oligomycin (oligo, 5uM), fluoro-carbonyl cyanide phenylhydrazone (FCCP, 1uM), and rotenone/ antimycin A (Rot/AA, 2 uM). The following calculations were used to obtain parameters for model (also see ED Fig. 2d):

$${\text{mitochondrial OCR: OCR}}_{\text{mito}}={\text{OCR}}_{\text{basal}}-{\text{OCR}}_{\text{Rot/AA}}$$

$$\text{cytosolic OCR}:{\text{OCR}}_{\text{cyto}}={\text{OCR}}_{\text{Rot/AA}}$$

To account for proton leak in the metabolic model, we corrected ATP yield of ATP synthase with coupling efficiency (CE) obtained from mitostress data:

$$\text{CE}=\left({\text{OCR}}_{\text{basal}}-{\text{OCR}}_{\text{oligo}}\right)/{\text{OCR}}_{\text{mito}}$$

For naïve cells, CF = 0.861 ± 0.022; for activated cells, CF = 0.727 ± 0.006.

Metabolite consumption and production

Metabolite consumption and production were measured by sampling the media with or without cells and quantitating metabolite concentration difference. Specifically, for naïve cells, cells were cultured in 0.5mL media at a density of 3E6/mL, and media were sampled at 48hr post isolation to measure consumption over 48hrs. For activated cells, cells were cultured in 1mL media at a density of 1E6/mL, replaced with fresh media at 24hr, and was sampled at 32 hr post isolation and activation to measure consumption over 8hrs. To account for evaporation and spontaneous glutamine hydrolysis (to glutamate and ammonium under neutral pH), we included replicates of control media without cells, and all the spent media were compared to media without cells under matching condition.

Metabolite concentration was determined with internal standard (1.1g/L 13C algal amino acid mixture, 11mM [6,6-²H₂]glucose, 2mM [U-¹³C₅]glutamine, 10mM [U-¹³C₃]lactate), calibrated with fresh RPMI1640 (without additional supplements). As alanine is not present in RPMI1640, alanine concentration in the internal standard was estimated from the manufacturer’s report of alanine content to be 1.24mM. For analysis, media was 1:1 mixed with the internal standard, 1:5 diluted in cold methanol (−20ºC), centrifuged (10 min, 16000 rpm, 4 °C). The supernatant was then 1:4 diluted in LC-MS solvent (40:40:20 ACN:MeOH:H₂O) before loading LC-MS.

For glucose and lactate concentration, we also obtained measurement by YSI (see yeast section), which showed agreement with the LC-MS measurement. Lactate production in activated cells was also comparable to the glycolytic proton efflux rate from Seahorse (derived from ECAR and OCR according to the manufacturer’s instruction). Last, glucose and lactate flux from orthogonal methods were averaged to constrain the flux model.

Biomass fluxes

The biomass synthetic flux $J_{biomass,i}$ (i = protein, DNA, RNA, glycerol-3-phosphate, and fatty acid synthesis flux in the form of acetyl-coenzyme A) is approximated by

$$J_{biomass,i}=\frac{E_{\text{hydrolysate}}^{13C}}{E_{\text{soluble}}^{13C}}\cdot \frac{M_i}{\text{Δ}t}$$

where $E_{\text{hydrolysate}}^{13C}$ and $E_{\text{soluble}}^{13C}$ are the ¹³C enrichment of same metabolite in the hydrolysate and the soluble pool, respectively. Fractional renewal of fatty acids was directly calculated from labeled fraction $\left(1-f_{M+0}\right)\cdot M_i$ is the mass of the biomass component, and $\mathrm{\Delta }t$ is the duration of culture with tracer. The fractional renewal is averaged if there are multiple metabolites representing the same biomass category. Here we use a reasonable assumption that precursor labeling is much faster than their labeling of biomass so the latter can be approximated by a linear kinetics. To measure mass of the biomass component, we used the same LC-MS assay for yeast biomass quantitation by mixing 1E6 naïve or activated T cells with 1OD·mL fully ¹³C labeled S. cerevisiae pellet, followed by acidic hydrolysis and LC-MS.

We noted a less than 3-fold difference between protein synthesis rate derived from uptake rate of essential amino acids and that from isotope measured renewal. We used the former to constrain the model, because it is a better reflection of net flux and does not require pre-steady-state assumption.

13C metabolic flux analysis

Yeast genome-scale carbon mapping model for 13C-MFA

We developed new carbon mapping models for S. cerevisiae and I. orientalis based on their genome-scale models, iIsor850 for I. orientalis ²³ and iSace1144 for S. cerevisiae (reformatted from yeast 8.3.4 model ⁶⁴ as described previously ⁶⁵). For model reduction, flux variability analysis ⁶⁶ was performed with constraints on measured glucose uptake and byproduct (ethanol, acetate, glycerol, and succinate) excretions, to remove reactions incapable of carrying flux under glucose utilizing conditions, e.g. degradation pathways that form ATP-consuming futile cycles with biosynthesis of nucleotide, lipid, fatty acid, and carbohydrate. We also simplified intracellular compartments by assigning non-mitochondrial reactions to cytosol. Carbon mapping of reactions were obtained from previous large-scale mapping model in E.coli ⁶⁷, or for new reactions curated from BioCyc database ⁶⁸, biochemistry textbook, and the literature. Annotation of functional groups and adjacent carbon atoms were also provided for carbon atoms (which were previously associated with only numbering indexes) to facilitate future use. The models also contain cofactor balance (e.g., ATP, NADH), charge and proton balance, proton pumping and the electron transport chain pathway, as well as growth associated ATP maintenance. Stoichiometry of the 52 precursors in the biomass reactions are updated to reflect condition-specific macromolecular composition measured in this study. The mapping model for S. cerevisiae contains 394 reactions and 354 metabolites, while the model for I. orientalis contains 386 reactions and 363 metabolites.

Murine T cell carbon mapping model for 13C-MFA

We developed new core carbon mapping models for murine T cells. The metabolic network was reconstructed using information from KEGG Pathway database and the murine genome-scale model Mouse-GEM ⁶⁹ (accessed via Metabolic Atlas platform ⁶⁹). Gene-protein-reaction mappings and compartmentalization of reactions was manually assigned using information on Mouse Genome Database ⁷⁰ and UniProt database (accession number: UP000000589). Carbon mappings were reconstructed similarly as in yeasts. We included reactions that are necessary to explain measured extracellular and biomass flux, and removed reactions that will lead to futile cycle (net flux of which is ATP wasting) if the corresponding enzymes were not detected in proteomics (eg. PPCK, FBP). For simplicity, catabolism of essential amino acids was not included because uptake of essential amino acids correlates well with their frequency in proteome, and glutamine is the dominant fuel of TCA cycle. The model also contains energy balance (ATP), redox balance (NADH and NADPH, etc.), nitrogen balance, as well as proton and charge balance. Stoichiometry of the 25 precursors in the biomass reactions are updated to reflect T cell specific stoichiometry weighted by the measured biomass synthesis flux. Namely, nucleotide stoichiometry was calculated from the genome (ID: GCF_000001635.27) (for DNA) and transcriptome ⁷¹ (for RNA), while amino acid stoichiometry is derived from quantitative proteomics from this study. The mapping model for murine T cell contains 200 reactions and 194 metabolites.

¹³C-MFA

We employed ¹³C-MFA procedure described previously ⁶⁷ (formulated using the elementary metabolite unit framework ⁷²). Briefly, a non-linear optimization formulation was used to find a flux solution by minimizing the sum of squared differences between the simulated ¹³C mass isotopomer distributions (as a function of fluxes) and the observed ones from both tracers, as well as uptake/excretion fluxes. The best-fit flux solution was chosen from 200 alternative solutions with randomized initializations. Goodness-of-fit test (chi-square) and 95% confidence interval estimation were performed as described previously ^73,74. For T cell ¹³C MFA, we used similar procedure as yeast, except that an additional G-factor is included for each metabolite to account for unlabeled fraction due to incomplete mixing with a pre-steady-state pool ⁷⁴ (likely from biomass break down). The best-fit flux solution was chosen from 600 alternative solutions with randomized initializations. We also include constraints for oxPPP and Icdh based on a previous study using ¹⁴C and deuterium tracer in naïve and activated T cells ⁵⁸. For both yeasts and T cells, we also included Escher maps in each repository for visualizing the metabolic fluxes at https://escher.github.io/.

ATP yield of electron transport chain

The stoichiometry of ATP synthase is 3 ATP produced for every 10 protons translocated by ATP synthase ⁷⁵. We use this mechanistic ratio for yeast flux analysis, which results in a ratio of ATP to atomic oxygen (P/O ratio) about 1.9 in S. cerevisiae 2.7 in I. orientalis. For T cells, we also included in the model the measured coupling efficiency described in ‘Oxygen consumption rate and respiratory capacity’.

Quantitative proteomics

Absolute protein abundance was quantified by intensity-based absolute quantitation (IBAQ) using a UPS2 internal standard. Relative protein abundance across samples was quantified using TMTpro isobaric tags ³².

Proteomics sample preparation

Yeast proteomics samples were prepared as previously described with modification ^76,77. Yeast pellet equivalent to 20mL OD600 = 1 was grounded by CryoMill (Retsch, Newtown, PA) at 25Hz for 10min and lysed in 50 mM HEPES pH 7.2, 4% SDS, and 1 mM DTT to an approximate concentration of 2mg/mL protein. For T cells, about 3×10⁶ naïve T cells or 1×10⁶ activated T cells were pelleted, washed, and lysed with 40uL above lysis buffer. For murine tissues, frozen tissues were ground into powder using CryoMill. About 10~20mg tissue powder was weighted and added 400uL lysis buffer per 10mg tissue. Protein concentrations were determined in the supernatant of the lysate by BCA assay (Pierce BCA Protein Assay Kit, Thermo Scientific). The protein concentration in the final lysate was between 1 and 2 mg/mL.

For intensity-based absolute quantification, lysates equivalent to 300 μg protein were spiked with 2.5 μg of UPS2 Dynamic Range Standard (Sigma). The sample was then reduced with 5 mM dithiothreitol for 20 min at 60 °C and alkylated with 20 mM N-ethylmaleimide for 20 min at room temperature. 5 mM dithiothreitol was added to quench the excessive alkylating reagents. Proteins were purified by methanol-chloroform precipitation. The dried pellet was resuspended in 10 mM EPPS (N-(2-Hydroxyethyl)piperazine-N’-(3-propanesulfonic acid, pH 8.5) with 6 M guanidine hydrochloride). Samples were heated at 60 °C for 15 minutes, and the protein mixture was diluted 3-fold with 10 mM EPPS (pH 8.5). The protein mixture was digested with 6 μg LysC (Wako) overnight at room temperature. Samples were further diluted 4-fold with 10 mM EPPS (pH 8.5) and digested with an additional 20 ng/μL LysC and 10 ng/μL sequencing grade Trypsin (Promega) at 37 °C for 16 hours. For samples with limited protein amount, proteins were precipitated following reduction and alkylation using the SP3 method as previously described ⁷⁸. After binding and washing the bead-bound protein, the protein-containing beads were resuspended in 2M guanidine hydrochloride and digested with LysC and trypsin in two steps as above. After digestion, the peptides were cleared by ultracentrifugation at 100,000Xg for 1 hour at 4 °C (Beckman Coulter, 343775), and then the supernatant was vacuum-dried. The dried peptides were resuspended and desalted using homemade stage tips with C18 material (Empore). The samples were resuspended in 1% formic acid to 1 μg/μL before LC-MS analysis.

For TMT labeling, pre-mixed TMTpro tags (16-plex and 18-plex, Thermo Scientific, 20 μg/μL in dry acetonitrile stored at −80 °C) were added at 5μg TMTpro: 1μg peptide ratio to the above supernatant containing 200 μg of peptides, mixed, and incubated at room temperature for 2 hours. The reaction was then quenched by addition of 5 μL of 5% hydroxylamine (Sigma, HPLC grade) at room temperature for 30 minutes. The resulting mixture was vacuum-dried, desalted, and resuspended as described above for LC-MS analysis.

All replicates of T cells and tissues and one replicate of each yeast strain from batch culture were also analyzed after prefractionation to detect a larger number of peptides. Specifically, prior to LC-MS analysis, the dried peptides were resuspended in 10 mM ammonium bicarbonate (pH 8) with 5% acetonitrile to a peptide concentration of 1 μg/uL. The dissolved peptides were separated into 96 fractions using a medium pH reverse phase separation (Zorbax 300Extend C18, 4.6 × 250 mm column) on a 1260 Infinity II LC system (Agilent) as described previously ⁷⁷. Each resulting 96-well plate was combined into 24 fractions ⁷⁹, and each fraction was desalted and resuspended for LC-MS analysis as above.

Peptide analysis by LC-MS

Samples were analyzed on an EASY-nLC 1200 HPLC (Thermo Fisher Scientific) coupled to an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) with Tune version 3.3. Data was collected by XCalibur (Thermo Fisher Scientific). Peptides were separated on an Aurora Series emitter column (25 cm × 75 μm ID, 1.6 μm C18) (Ionopticks, Australia) and held at 60°C using an in-house built column oven. Solvent A consisted of 2% DMSO (LC-MS grade, Life Technologies), 0.125% formic acid (98%+, TCI America) in water (LC-MS grade, OmniSolv, VWR), solvent B of 80% acetonitrile (LC-MS grade, OmniSolv, Millipore Sigma), 2% DMSO and 0.125% formic acid in water. The following 90 min-gradient was applied at a constant flow rate of 350 nL/min after thorough equilibration of the column to 0% B: 0% – 6%B in 5 min; 6 – 25%B for 70 min; 25% – 100% for 10 min; 100% for 5 min.

For electrospray ionization, 2.6 kV were applied between 1min and 83min of the LC gradient. The Fusion Lumos was operated in data dependent mode. The survey scan was performed at a resolution setting of 120k in orbitrap, followed by MS2 duty cycle of 1.5 s. The normalized collision energy for CID MS2 experiments was set to 30%, and the HCD collision energy was set at 24%. The ion trap detector was used for MS2 scans of label-free samples or conjugate ion quantification of TMT-labeled samples ⁷⁷ (<= 9plex). An orbitrap detector was used for MS3 scans of 18plex samples. To avoid carry-over of peptides, 2,2,2-trifluoroethanol (>99% Reagent plus, Millipore Sigma) was injected in a 30 min wash between each sample. For fractionated samples, this wash was performed between every 3 fractions from the same original sample.

Proteomics data analysis

The data was analyzed using GFY software licensed from Harvard University. Raw files were converted to mzXML using ReAdW.exe. MS2 spectra assignment was performed using the SEQUEST algorithm v.28 (rev. 12) by searching the data against the combined reference proteomes for S. cerevisiae (S288C: UP000002311, 2/24/2021; CEN.PK, UP000013192, 8/20/2021), I. orientalis (UP000029867, 11/13/2019) and M. musculus (UP000000589, 10/11/2022) acquired from UniProt merged acquired from Uniprot merged with the UPS2 Proteomics Standards FASTA file provided by the manufacturer (https://www.sigmaaldrich.com/deepweb/assets/sigmaaldrich/marketing/global/fasta-files/ups1-ups2-sequences.fasta) along with common contaminants such as human keratins and trypsin. The target-decoy strategy was used to estimate the peptide false discovery rate (FDR)⁸⁰, and 1% FDR cutoff was used for MS2 spectral assignment. A 20-ppm precursor ion tolerance with the requirement that both N- and C- terminal peptide ends are consistent with the protease specificities of LysC and Trypsin was used for SEQUEST searches. One missed cleavage was allowed, and NEM was set as a static modification of cysteine residues (+125.047679 Da). Fragment ion tolerance in the MS2 spectrum was set at 1 Th. Filtering was performed using a linear discriminant analysis with the following features: Sequest parameters XCorr and unique ΔXCorr, peptide length, missed cleavages, adjusted PPM, fraction of ions matched, and charge state. Forward peptides within three standard deviations of the theoretical m/z of the precursor were used as positive training set. All reverse peptides were used as negative training set. Linear Discriminant scores were used to sort peptides with at least seven residues and to filter with the desired cutoff. Furthermore, we performed a filtering step on the protein level by the “picked” protein FDR approach⁸¹. Protein redundancy was removed by assigning peptides to the minimal number of proteins which can explain all observed peptide, with above-described filtering criteria ⁸².

Relative quantification of TMT tagged samples was done by summing up area of TMT reporter ion belonging to each protein. The signal was then normalized to the mean across samples, and then median normalized within each sample. To quantify absolute protein abundances in label-free samples, for each protein, area of precursor ion intensity from all peptides was summed up, and then normalized by number of theoretical peptides. Signals from UPS2 proteins were used to construct a calibration curve, which was then fitted to a power law, log(intensity) = k * log(concentration) + constant, to obtain absolute concentration of yeast proteins (log-linear coefficient k = 1.25 ± 0.08 on average). For yeasts, absolute protein abundance in batch culture is reported as mass fraction in whole proteome, which is approximated by the product of concentration and amino acid sequence length normalized to the sum of all proteins. Absolute protein abundance in nutrient limitation or respiratory deficient conditions is inferred from the relative fold change to batch culture obtained with relative quantification. For T cells and mouse tissues, TMT tagged samples were spiked in UPS2 standards so that the relative abundance between samples and total abundance of all samples can be quantified simultaneously.

Pathway assignment

Each S. cerevisiae protein is assigned to a functional sector based on gene ontology from Uniprot and pathway annotation from the genome-scale metabolic model yeast 8.3.4 ⁶⁴. I. orientalis proteins were assigned based on protein sequence identity to S. cerevisiae obtained from blastp (https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE=Proteins). Functional assignment in mouse is based on pathway and gene ontology from Uniprot (UP000000589), subsystem from mouse genome-scale metabolic model⁶⁹ (https://github.com/SysBioChalmers/Mouse-GEM/tree/main/model), KEGG ontology from proteomap⁸³ (https://www.proteomaps.net/), and mitochondrial localization from MitoCarta2.0⁸⁴. Glycolytic and respiratory protein assignment for NCI60 cells were obtained from Zielinski et al³⁵. The complete list of functional assignment can be found in Supplementary Table. R code to generate pathway assignment can be found at the Github repository (https://github.com/yihuishen/metabolic_flux_regulation/).

Proteome efficiency

ATP flux

For yeasts and mouse T cells, glycolytic ATP production is calculated as PGK_c + PYK_c – HEK_c – PFK_c flux. Respiratory ATP production is represented by the flux through ADPATPt_c_m, the mitochondrial ADP/ATP transporter. 95% confidence interval [lb, ub] is obtained from ¹³C MFA, based on which the ATP flux is determined as (lb + ub)/2 with a standard error of (ub-lb)/3.84. For NCI-60 cancer cell lines, the flux data was obtained from a previous flux analysis, using a model that assumes 4 proton translation per ATP production from the ATP synthase ³⁵, and constrained by experimentally measured rates (growth, uptake and excretion) ⁸⁵. ATP production is obtained similar to yeast. Flux in mouse tissues was obtained from a recent study that measured TCA flux and glucose uptake in vivo⁸⁶. Respiratory ATP flux was calculated as 14.5 ATP per AcCoA oxidized in TCA cycle (as done in the original study ⁸⁶), while glycolytic ATP flux was based on 2 ATP per glucose. Wet tissue mass is converted to dry mass by a factor of 0.4, and protein is assumed to account for half of dry mass.

Flux partitioned proteome allocation

Since glycolysis is used to provide precursor for respiration, and both glycolysis and respiration are used to provide biomass precursors, proteome allocation required for ‘fermentation’ (converting glucose to ethanol) and ‘respiration’ (converting glucose to CO₂) was calculated based on flux partitioning ⁷. Briefly, the protein cost of enzyme $i$, $f^i$, is divided among fermentation $(f)$, respiration $(r)$, biomass $\text{(}bm\text{)}$. Its cost for function $k$, $f_k^i$, is proportional to the carbon flux its product is used for $k$, $j_k$

$$f_k^i=f^i\cdot \frac{j_k}{{\sum }_k{j}_k}$$

$j_r$ is approximated by oxygen consumption rate; $j_f$ is approximated by 3-times of ethanol excretion flux; $j_{bm}$ is derived from precursor stoichiometry in biomass equation. For simplicity, oxphos is not required for biomass precursors.

Competitive co-culture and fitness

Co-culture and genomic DNA extraction

Overnight cultures of S. cerevisiae CEN.PK and I. orientalis SD108 were mixed at 1:1 according to OD600, pelleted, and inoculated into fresh media at OD600 = 0.5. The cultures were then grown aerobically in one of the following conditions: aerobic 10 g/L ethanol, 20 g/L glucose, 20 g/L sucrose YNB with serial transfer for 12 ~ 14 hrs; or in aerobic glucose-, ammonia-, or phosphate-limited continuous culture at 0.1 h-1 dilution rate for 24 hrs. For (cyclically) anaerobic culture in glucose, anaerobic phase was achieved by sparging nitrogen into the culture at desired duty cycle (75%, 18h/24h; 87%, 21h/24; 100%, 24h/24h). Anaerobic culture was achieved by sparging nitrogen into the culture with 20g/L glucose. Relative abundance of the two yeasts was measured by qPCR at 4 to 6 time points and used to obtain fitness. Specifically, at each time point, 1mL co-cultures was pelleted, and the rest was then diluted with fresh media to keep the OD600 of the culture to approximate 1 in aerobic cultures or 0.5 in (cyclically) anaerobic cultures. Calibration curve was also prepared by mixing single cultures at different ratios. The cell pellet was lysed by lyticase (Sigma Aldrich, L4025) and genomic DNA was extracted with DNeasy^® Blood & Tissue Kit (Qiagen) following the manufacturer’s procedure.

Determine relative species abundance by quantitative PCR (qPCR)

Relative abundance of S. cerevisiae and I. orientalis was determined by qPCR of the pho2 genomic sequences for S. cerevisiae (Gene ID: 851452, NM_001180165.1) and the distant homolog for I. orientalis (Gene ID: 40382003, XM_029463910). qPCR primers were designed using OligoArchitect Online (Sigma-Aldrich, http://www.oligoarchitect.com) and checked for cross-hybridization against the genomic sequences of both species using BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi). The primers and probes used for S. cerevisiae are CTCTCTTCTTCGATCATG (sense), TCTCCTCATTATTAGCATTATG (anti-sense), and [6FAM]ATAACCAACACCAACAACGGACAAG[OQA] (probe); for I. orientalis, GAGACTAGCACCCTTAAC (sense), CGTTCACATCTACACTGA (anti-sense), and [JOE]ACAGCCTCCACAACGACTTCT[TAM] (probe) (Sigma-Aldrich). For competitive co-culture of I. orientalis ΔNde and ΔPdc strains, we used primers that target Pdc and probes that recognize wild-type and mutated Pdc, respectively. The primers are CCACGTYAAGGAAGTTGAA (sense), AGGTGGTGATTAGGGATG (antisense), and the probes are 6-FAM_ATTCTTGCATAACCATCAGCTTCGTA_BHQ-1 (wile-type) and JOE_AATCCATTGATTCTTGCATTAGCCG_TAMRA (mutant). Primers and probes were tested for unspecific cross-reactivity using iTaq^™ Universal Probes Supermix (BioRad 1725132) individually and in combination with genomic DNA from both species in various ratio. No cross-activity was observed. For qPCR, 1 to 2 ng of isolated DNA was used per 10 μL assay containing 250 nM for each primer and 125 nM each probe) in a 384 plate. Assays were performed using the Applied Biosystems ViiATM7 Real-Time PCR System. The relative abundance was quantified from the calibration curve, and fitted to log(strain1 / strain2) = fitness * t + constant to obtain relative fitness.

Extended Data

Extended Data Fig. 1. Metabolic flux analysis in yeasts.

(a) Overview of 13C metabolic flux analysis in yeasts.

(b) Experimentally measured growth rate, oxygen consumption rate, and carbon metabolite fluxes for S. cerevisiae and I. orientalis (SD108) grown in YNB with 20g/L glucose. Two strains (FY4 and CEN.PK) were used for S. cerevisiae, with FY4 derived from the widely used S288C background. Mean ± s.e. derived from fitting of 3 time points in n = 3 biological replicates.

(c) Isotopomer ratio in TCA intermediates reveals higher oxidative TCA activity in I. orientalis. Isotopomer ratios show average [M+1]/[M+2] ratio from [1,2-¹³C₂] glucose tracing (1:1 mixed with unlabeled glucose) in three TCA metabolites (Asp, aspartate; Fum, fumarate; Mal, malate), mean ± s.e.m., n = 3 or 4 biological replicates. Flux ratios (from ¹³C genome-scale MFA) are between oxidative TCA (average of citrate synthase, alpha-ketoglutarate dehydrogenase and succinate dehydrogenase) and anaplerotic TCA (pyruvate decarboxylase). Filled circles, ¹³C atom.

(d) Isotopomer ratio in pyruvate reveals higher flux through pentose phosphate pathway (PPP) in I. orientalis. [M+1]/[M+2] ratio of pyruvate from same experiment in (c), mean ± s.e.m., n = 4 biological replicates. Flux ratios (from ¹³C genome-scale MFA) are between PPP (difference between glucose-6-phosphate dehydrogenase and phosphoribosylpyrophosphate synthetase) and glycolysis (phosphoglucose isomerase). Filled circles, ¹³C atom.

(e) Consumption (−) and production (+) flux contributing to the balance of whole-cell NADH. ALCD, alcohol dehydrogenase; NADHqx, NADH quinone oxidoreductase (Nde and Ndi in S.cerevisiae, Nde in I. orientalis); GAPD, glyceraldehyde-3-phosphate dehydrogenase; Complex I, electron transport chain complex I. TCA, reactions in the TCA cycle. Fluxes are best estimate from genome-scale ¹³C MFA.

(f) Growth impact of NADH dehydrogenase deletions in S. cerevisiae and I. orientalis. Data for S. cerevisiae is from a previous study¹; data for I. orientalis is determined in this study and shows mean ± s.d. from single exponential fitting of the growth curve (n = 4 time points). p value from two-sided student t test without adjustment.

(g) Glucose uptake relative to its use to make ethanol or biomass. Fluxes are from the genome-scale flux analysis, with fluxes consuming pyruvate (e.g. ethanol production) counted as 3 carbon atoms.

(h) Fraction of flux through 11 central metabolites partitioned in biosynthesis (not including the flux to another central metabolite).

Extended Data Fig. 2. Metabolic measurement for T cell metabolic flux analysis.

(a) Overview of 13C metabolic flux analysis in mouse T cells. Primary CD8+T cells were purified from murine spleen, and kept in IL7 to remain in naïve state or activated by αCD3 and αCD28 in the presence of IL2 for 24hrs. Marker expression (CD69-FITC, CD25-APC) was evaluated by flow cytometry.

(b) Isotope enrichment in central metabolites with [U-¹³C₆]glucose or [U-¹³C₅]glutamine tracing. ¹³C enrichment shows the average ¹³C labeling per carbon atom. Mean ± s.e.m., n = 3.

(c) Media nutrient exchange flux. Positive and negative values indicate uptake and excretion, respectively. Numbers show fold change between naïve and activated T cells, with negative values reflecting change in flux direction. Mean ± s.e.m., n = 12 (O₂), n = 4 (others).

(d) Oxygen consumption rate (OCR) were measured with Mito Stress test through the sequential addition of pyruvate (pyr, 1mM), oligomycin (oligo, 5uM), fluoro-carbonyl cyanide phenylhydrazone (FCCP, 1uM), and rotenone/ antimycin A (Rot/AA, 2 uM). Mean ± s.e.m., n = 12.

(e) Biomass (DNA, protein, and RNA) renewal flux is the product of mass composition and fraction renewed measured by ¹³C enrichment in the biomass hydrolysate normalized to monomer in soluble metabolites. Mean ± s.e.m., n = 6. Flux, mean ± s.d. error propagated from mass and fraction renewed

(f) Fold change of metabolic fluxes (from ¹³C MFA) between activated and naïve T cells.

(g) Flux balance of NADH and TCA four-carbon metabolites (TCA C₄). LDH, lactate dehydrogenase; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; GLUN, glutaminase; ME, malic enzyme; PC, pyruvate carboxylase; Glu, glutamate.

(h) Isotopomer ratio (from LC-MS) reveals flux ratio (best estimate from ¹³C MFA) between malic enzyme (ME) and glycolysis. M+3 pyruvate (Pyr) is produced by ME from M+4 malate (Mal), whereas M+0 Pyr is produced from glycolysis. Note that malic enzyme flux increases in activated T cells despite a reduced isotopomer ratio. Mean ± s.e.m., n = 3.

Extended Data Fig. 3. Quantitative proteomics in yeasts and T cells.

(a) Comparison between the S. cerevisiae proteomics generated in this study and in other studies^2,3. Data show mass fraction of functional sectors. Median ± s.e.m., p value from two-sided student t test between literature data and our data without adjustment, n = 19 (literature data); n = 4 (biological replicates, this study); *, p<0.05, n.s., p>0.05. Pearson’s correlation R = 0.88 (p=1E-5) between our proteome allocation and the median of reference (n=14 sectors).

(b) Protein abundance of yeasts. Total protein mass and breakdown in each functional sectors (left) and allocation to individual reactions (right). Fold change (FC) from I. orientalis to S. cerevisiae is shown on the bottom. Enzyme abundance (sum of isozymes, if any) of individual reactions in glycolysis, TCA, and OXPHOS pathways. Mean ± s.e.m., n = 4. Two-sided student t test with Bonferroni FDR correction, n.s., p>0.05; *, p<0.05; **, p<0.005; ***, p<0.0005.

(c) Protein abundance of T cells. Total protein mass and breakdown in each functional sectors (left) and allocation to individual reactions (right). Fold change (FC) from naive to activated is shown on the bottom. Enzyme abundance (sum of isozymes, if any) of individual reactions in glycolysis, TCA, and OXPHOS pathways. Fold change of individual genes is also shown. Mean ± SEM, n = 3. Two-sided student t test with Bonferroni FDR correction, n.s., p>0.05; ***, p<0.0002.

Extended Data Fig. 4. Proteome efficiency with flux-partitioning or mitochondrial proteins.

(a) Repartitioning of glycolytic and respiratory proteomes in proportion to flux distribution to biomass, fermentation, and respiration (left), and the resultant proteome efficiency (right). Data from batch cultured yeasts, mean ± s.e.m., error propagated from 13C metabolic flux analysis and proteomics (n=4).

(b) Flux-partitioned proteome efficiency of respiration in naïve and activated T cells given by $(J_{ATP,resp}+C_{glyc\to resp}·J_{ATP,glyc})/(M_{resp}+C_{glyc\to resp}·M_{glyc})$, where $\left(C_{glyc\to resp}\right)$ is the fraction of glycolytic flux that ends up in respiration $J_{ATP}$ is the ATP flux and $M$ is the total mass of proteins in each pathway. Mean ± s.d., error propagated from 13C metabolic flux analysis and proteomics (n=3).

(c) Proteome efficiency accounting for all mitochondrial proteins in respiration in yeasts. Data from batch cultured yeasts, mean ± s.d., error propagated from 13C metabolic flux analysis and proteomics (n=4).

(d) Proteome efficiency accounting for all mitochondrial proteins in respiration in yeasts. Data from batch cultured yeasts, mean ± s.e.m., error propagated from 13C metabolic flux analysis and proteomics (n=3).

Extended Data Fig. 5. Metabolic fluxes of yeasts across different growth conditions.

(a) Dependence of metabolic flux and enzyme concentration on growth rate. Flux and enzyme concentration are normalized to maximum across nutrient conditions. For each reaction, a linear regression is done for flux ~ growth rate. % variance explained is calculated, and averaged across all reactions. 53% of flux variation in S. cerevisiae and 71% in I. orientalis can be explained by growth rate alone. 21% of enzyme variation in S. cerevisiae and 23% in I. orientalis can be explained by growth rate alone.

(b) Central carbon fluxes of yeasts across nutrient conditions. Flux (j) through TCA and glycolysis, and PPP are represented by flux through citrate synthase (CS), pyruvate kinase (PYK), and glucose-6-phosphate dehydrogenase (G6PD), respectively, and normalized to growth rate (μ). Limiting nutrient for chemostat, C, carbon; N, nitrogen; P, phosphorus; B, none.

Extended Data Fig. 6. Proteomics of mouse tissues and tumors and flux-partitioned proteome efficiency.

(a) Total protein mass and mass of each functional sectors in pancreas and k-Ras-driven pancreatic ductal adenocarcinoma (PDAC) (GEMM, genetically engineered mouse model; flank, flank implanted). Mean ± s.e.m., n = 3. Two-sided student t test between tumor and healthy without adjustment, n.s., p>0.05; *, p<0.05; **, p<0.005; ***, p<0.0005. Fold changes (FC) are shown on the bottom of each graph.

(b) Mass fraction of proteome sectors as in (a).

(c) Differential protein expression between healthy pancreas and PDAC. Top, glycolytic and respiratory proteins (fold change between PDAC and healthy pancreas), mean, n=3. Bottom, hypoxia-inducible factor 1α (Hif1α), individual replicates and boxplot (median with quartiles).

(d) Total protein mass and mass of each functional sectors in spleen and spleen infiltrated with Notch1-driven leukemia (leukemic spleen). Mean ± s.e.m., n = 3. Two-sided student t test between tumor and healthy without adjustment, n.s., p>0.05; *, p<0.05; **, p<0.005; ***, p<0.0005. Fold changes (FC) are shown on the bottom of each graph.

(e) Mass fraction of proteome sectors as in (d).

(f) Differential expression of glycolytic and respiratory proteins between healthy and leukemic spleen (fold change between leukemic and healthy spleen), mean, n=3.

(g) Flux partitioned proteome efficiency considering flux contribution from glycolysis to respiration $\left(C_{glyc\to resp}\right)$, based on data in Fig. 4. For NCI60 cancer cells, $C_{glyc\to resp}$ is quantified as the ratio between mitochondrial pyruvate carrier flux and glucose uptake flux. For mouse tissues and tumors, $C_{glyc\to resp}$ is the flux ratio between glucose oxidation (estimated as 40% of TCA cycle) and glycolysis.

Extended Data Fig. 7. Metabolism of evolutionarily divergent budding yeasts.

(a) Growth rates (top left) and glucose consumption rates (bottom left) and their relation (right) of 13 yeasts cultured in minimal YNB media containing 20g/L glucose. 14 budding yeasts with top growth rates in YPD were selected, with C. petersonii not able to grow in YNB. Mean ± s.e. (error from regression), n = 3 time points for n = 2 or 4 biological replicates.

(b) Ethanol production rate (left) and its relation with glucose consumption rate (right) of the top 16 fast-growing yeasts in YPD containing 20g/L glucose. Mean ± s.e. (error from regression), n = 3 time points for n = 2 biological replicates.

Extended Data Fig. 8. Alternative explanations for aerobic glycolysis.

(a) Fold change in metabolite abundance between I. orientalis and S. cerevisiae cultured in glucose YNB. Inset shows the ratio between NAD⁺ and NADH. Metabolite abundance was measured by LC-MS, mean, n=6, p value from student t test with Bonferroni adjustment for multiple comparisons. Inset, mean ± s.e.m.

(b) Reaction Gibbs energy under physiological concentrations (top), total Gibbs energy dissipation rate (in J/mol) for ATP synthesis and hydrolysis in I. orientalis and two strains of S. cerevisiae (bottom left), and dissipation per ATP production as a function of ATP:oxygen (PO) ratio (bottom right). In the equations for dissipation per ATP, 2 reflects 2 ATP made per glycolysis, 12 reflects 12 pairs high-energy electrons made per glucose by respiration, and 12 · PO is ATP yield per glucose by respiration. Shaded areas show experimentally obtained PO ratio (95% interval, see “Assessment of PO ratio” in Ext. Data Note) for S. cerevisiae and I. orientalis. Mean ± s.d. propagated from error of flux measurement and flux analysis.

Extended Data Fig. 9. Fitness and proteome allocation of yeasts in aerobic and anaerobic conditions.

(a) Dissolved oxygen measured at the bottom of a multi-well plate culture with OxoPlate (phosphorescent oxygen sensor). Fresh exponential culture was added to the plate at indicated density and allowed to adapt for 15min. Oxygen concentration was then measured with or without active shaking. The typical maximal cell density (OD) from fully aerated culture is about 4.

(b) Competitive fitness of lab adapted I. orientalis mutants, ΔPdc (null mutant for pyruvate decarboxylase, essential for ethanol fermentation) and ΔNde (null mutant for cytosol-facing NADH dehydrogenase, which feeds into the electron transport chain). 3 colonies of mutants were adapted for 14 days before competitive coculture. Relative fitness, mean ± s.e.m., n = 8.

(c) A coarse-grained model where yeast growth is constrained by flux balance, energy (ATP) limitation, and proteome allocation. G, glycolysis; R, respiration; T, translation. For explanation of parameters and the model, see Ext. Data Note.

(d) Maximal aerobic and anaerobic growth rate (μ) under different glycolytic (f_G) and respiratory protein abundance (f_R) (mass fractions of whole cell dry weight). Stars, optimal proteome allocation in aerobic and anaerobic conditions. Circles and triangles indicate measured proteome fractions in glucose-fed batch cultures of I. orientalis (aerobic, +O2; or anaerobic, -O2) and S. cerevisiae (aerobic, +O2).

(e) Experimental glucose consumption (J_GLC) and ethanol excretion (J_ETOH) rates (symbols, in mmol/h/gDW) and prediction from proteome-constrained model (lines) under high (S. cerevisiae) or low (I. orientalis) glycolytic proteome capacity (r_G). Literature data was obtained from Van Hoek 1998⁵.

(f) Growth rate of wilt type (WT) and adapted ΔNde mutant I. orientalis in aerated and settled culture. Data show WT and three colonies of ΔNde mutant picked after a 14-day adaptation. Media is YPD with 20g/L glucose. Mean ± s.e.m., n = 3 (aerated) n = 4 (settled).

Data availability

All source data, including metabolic flux, proteomics, proteome efficiency, are provided in Supplementary Files. The following accession numbers are used to access publicly available proteome: S. cerevisiae (S288C: UP000002311, 2/24/2021; CEN.PK, UP000013192, 8/20/2021), I. orientalis (UP000029867, 11/13/2019) and M. musculus (UP000000589, 10/11/2022). We also queried Mouse-GEM ⁶⁹ and Mouse Genome Database ⁷⁰ for mouse genome information. Some of the healthy mouse tissue proteomics data are from PaxDb⁵¹. The mass spectrometry proteomics data generated in this study have been deposited to the ProteomeXchange Consortium via the PRIDE⁸⁷ partner repository with the dataset identifier PXD048012 (I. orientalis), PXD048018 (S. cerevisiae), and PXD048041 (M. musculus).