Non-canonical resource allocation in heterotrophically growing Thermoanaerobacter kivui

Abstract

Allocation of resources in the costly proteome reflects trade-offs between cellular functions. For example, proteome composition of Escherichia coli is significantly regulated by growth rate. An increasing anabolic, especially ribosomal, proteome fraction correlates with a decreasing catabolic proteome fraction at faster growth, which then leads to changes in catabolism. Our systems-level studies of the thermophilic acetogen Thermoanaerobacter kivui when growth rate is varied over two orders of magnitude revealed a different strategy: proteome allocation is only partially controlled by growth rate, and metabolic rates are primarily controlled posttranslationally. At slower growth, ribosome numbers are controlled by rRNA concentrations with an excess of ribosomal proteins. Composition of the catabolic proteome is uncoupled from catabolic rates as indicated by flux analysis. This study adds to the understanding of acetogenic Clostridia, which are of interest for biotechnological processes in a carbon-neutral economy, and points to a complex landscape of microbial ecophysiological strategies.

Proteome allocation to anabolic and catabolic functions is significantly regulated by growth rate in the model bacterium Escherichia coli. By contrast, this article shows that proteome allocation is only partially controlled by growth rate, and metabolic rates are primarily controlled post-translationally, in the thermophilic acetogen Thermoanaerobacter kivui.

Introduction

Microbial growth physiology has been intensely studied for the last 100 years, because of its medical, environmental, and biotechnological relevance^1–6. Most studies have been conducted using fast-growing chemoorganotrophic E. coli, Pseudomonas, or Salmonella species as model systems^1,2,6–9. These microbes have in common, besides their experimental convenience, a substantial diversity of catabolic pathways for metabolism of organic substrates that confers ecological success in high-flux and fluctuating environments¹⁰. In conjunction with quantitative proteomics and metabolic models, a detailed understanding of their specific growth physiology has been developed on a population level (Supplementary Fig. 6a and Supplementary Table 1)^8,11. From these studies, several general concepts have emerged: i) that microbial growth is determined predominantly by tight regulation of gene expression, e.g., via catabolite repression, as well as stringent control for controlling ribosome content under high flux conditions^12–14 and ii) a general ‘law’ of microbial growth, in which growth rate is determined primarily by the level of ribosomes operating at maximum translational rate⁸. Within this framework, growth rate was found to be a determining factor in cell physiology, and the main determinant for these growth laws: For example, cell shape and proteome composition as well as ribosome numbers are strongly correlated with growth rate for E. coli when grown under otherwise differing conditions and cultivation methods (different carbon sources, minimal vs. complex media, batch vs. chemostat cultures)^1,2,6–9. These growth laws are applicable to several other, including phylogenetically distant, bacteria^{1,2,6–9,15,16}. The more recently introduced concept of coarse-grained allocation of flux-coordinated catabolic and anabolic proteomes provides a compelling mechanistic explanation for these growth-rate dependent regulation^11,17 and can also be used to accurately model responses of E. coli to other conditions such as anabolic limitation or ribosome inhibition^11,17,18. Therefore, it has been assumed that E. coli-type growth laws represent a universal principle underpinning a general microbial growth physiology.

While other microorganisms follow these predictions at least in part^15,19,20, recent studies have challenged the universality of these growth laws for ecophysiologically different microorganisms, as well as for slow-growing E. coli, and on a single-cell level. For example, at comparatively slow growth, ribosome content in E. coli and Corynebacterium glutamicum is higher than expected and, therefore, translation rates below maximum were observed^15,21. At low substrate flux and growth rate, E. coli also expresses ‘useless’ catabolic genes while no corresponding substrates are present¹⁸, indicating that proteome allocation strategies might differ for extremely slow growth. Additionally, a recent study on single-cell physiology of E. coli showed that these population-level derived growth laws do not hold up at the single cell level to the same degree²². Extensive systems-level studies in the archaeon Methanococcus maripaludis, which is limited to hydrogenotrophic methanogenesis (4H₂ + CO₂→CH₄ + 2H₂O) as its sole catabolic pathway, revealed a different strategy for resource allocation during catabolic limitation (Supplementary Fig. 6b and Supplementary Table 1)²³. Proteome allocation, including ribosome content and the methanogenic proteome, was relatively invariant over two orders of magnitude in growth rate despite differences in methanogenesis rates during growth. However, it is unclear if this constant proteome allocation is because of the low growth rates studied (0.002–0.2 h⁻¹), because of its constrained metabolism and resulting ecophysiology, or if it is only found in archaea.

Besides these studies on growth-rate dependent microbial physiology, additional deviations from the expectation of a strictly regulated proteome to match necessary metabolic fluxes were reported: In anaerobic lactic acid bacteria¹⁹ and several anaerobic, acetogenic Clostridia, shifts in metabolism without corresponding proteome reallocation have been observed. For example, Clostridium autoethanogenum produces acetate and ethanol from H₂ and CO₂ and switches to higher ethanol/acetate ratios at higher acetate levels²⁴, and Clostridium ljungdahlii switches from acetogenic to solventogenic catabolism under nutrient limitation during growth on CO-rich gas, which allows overflow of electrons from CO into ethanol²⁵.

However, comprehensive studies on growth-rate dependent physiology, including extensive proteome analyses, data on cell composition and cell shape, as well as metabolic rates, which are necessary for a more complete systems-level understanding, are rare. Additionally, many studies rely on exponentially growing cells in batch cultures, growing at different maximum rates using different nutrient sources, which intrinsically include accumulating end products. Chemostat cultivations, on the other hand, interrogate more specifically only growth-rate-dependent cellular physiologies in the absence of other compounding environmental variables. Lastly, many studies focus on high growth rates as found, for example, in E. coli, while only very few studies analyzed balanced growth at lower growth rates (<0.1 h⁻¹).

To examine more broadly the extend of alternative growth strategies, especially under slow-growth conditions, we performed a comprehensive systems-level analysis of growth in the thermophilic, acetogenic Thermoanaerobacter kivui²⁶. As other acetogenic Clostridia, T. kivui can grow chemolithoautotrophically using H₂ and CO₂ to form acetate via the reductive Wood-Ljungdahl Pathway (WLP) and is therefore of interest for gas fermentation processes in the Green Transition toward a carbon-neutral economy^27–29. As a member of the Clostridia, T. kivui is also of interest regarding alternative regulation of catabolic fluxes. T. kivui also grows chemoorganotrophically (Fig. 1a), oxidizing glucose to two acetate and two CO₂ via the glycolytic Embden-Meyerhoff-Parnas pathway (EMP) in conjunction with pyruvate:ferredoxin oxidoreductase (PFOR). The resulting NADH and reduced ferredoxin are re-oxidized with the two CO₂ via the WLP to form a third acetate. Acetate is either excreted as a product or used for biomass synthesis³⁰. With these limited catabolic capacities and being able to use H₂ and CO₂, the catabolism of T. kivui is also closer than many model organisms to that of M. maripaludis, thus allowing comparisons of an archaeon and bacterium with similar ecophysiology. To isolate the effect of growth rate in our study, cells were grown in chemostats using the same carbon source for all conditions. For analyzing the growth strategy of T. kivui, we first describe the growth, second, analyze the catabolic proteome and fluxes, and third, compare the cellular composition and protein synthesis machinery to known examples.

Fig. 1. Growth of T. kivui in anoxic, glucose-limited chemostats at different growth rates.

For reference, data from exponentially growing batch cells (orange) are included. Cultures were in steady state, and growth rate was considered equivalent to dilution rate under these conditions. a overview of catabolism in T. kivui and carbon atom patterns in degradation of ¹³C₁ glucose (labeled carbon represented as open circles). Black, metabolites; red, electron carriers (only shown in reduced state); blue, enzymes and pathways. b biomass yield as dry cell weight per mol ATP. As dry cell weight was calculated from OD₆₀₀ using a factor from batch cultures instead of individual measurements for each culture, this is only an approximation to calculate biomass yields. However, dry cell weight was calculated to include biomass of lysed cells as determined by analyzing DNA in the culture supernatant. ATP production is calculated from the amount of extracellular acetate, assuming an ATP yield of 2.9 mol ATP per 3 mol extracellular acetate. c maintenance energy, estimated from the slope of the linear regression of $\frac{1}{Y}$ versus $\frac{1}{\mu }$; d coarse-grained proteome fractions of catabolic (open symbols) and anabolic (filled symbols) proteome sectors; e steady state glucose concentration. Values indicated as 0 mM do not contain detectable glucose (detection limit of 0.05 mM); f fraction of acetate carbon in total carbon of products; g carbon balance as percent of total carbon of glucose degraded that was found in products and biomass. Red, chemostats; orange, exponentially growing cells from batch cultures. EMP Embden-Meyerhoff-Parnas pathway, PFOR pyruvate ferredoxin oxidoreductase, WLP Wood-Ljungdahl pathway, DCW dry cell weight. Source data are provided as a Source Data file.

The findings of our study identified a proteome allocation strategy clearly inconsistent with the canonical growth laws of E. coli at fast growth as well as the strategy of M. maripaludis. This further points to more variety in microbial growth physiologies and resource allocation strategies than thought so far, and challenges the generality of our current understanding based on ‘unified’ growth laws.

Results

T. kivui growth in chemostats at varying growth rates

We studied the systems-level biology, including proteome allocation, of T. kivui ΔpyrE (hereafter, T. kivui) growing heterotrophically in glucose-limited, anoxic chemostats, which provides constant and controlled growth rates (Fig. 1, Supplementary Fig. 7 and Supplementary Data 1) spanning two orders of magnitude (0.001 h to 0.101 h⁻¹, equivalent to doubling times of 30 d to 7 h). For many discussions below, growth rates were divided into four groups: very slow (<0.003 h⁻¹), slow (up to 0.03 h⁻¹), medium (up to 0.04 h⁻¹), and fast-growing (>0.04 h⁻¹) chemostat cultures in comparison to exponentially growing batch cultures (0.173 h⁻¹). Before systems-level measurements, chemostat cultures were grown for at least four doubling times to ensure that they reached steady state (Supplementary Fig. 8). These growth rates overlap only partly with the growth rates studied for E. coli (>0.035 h⁻¹) and, thus, expand the range of well-studied growth rates. Additionally, it is very similar to the range of growth rates studied for M. maripaludis²³.

Carbon source, products, and biomass production

As expected for glucose-limited chemostats, glucose was nearly completely consumed in steady state (<1 mM remaining in 62 of 73 reactors, ≤4.4 mM in eleven reactors; Fig. 1e, Supplementary Notes 1 and 2). The main product in all reactors was acetate (acetate as the only product in 25 of 73 reactors, 90%-99.9% of product carbon in acetate in 25 reactors, 60-90% in 19 reactors, <60% only in four reactors; Fig. 1f and Supplementary Fig. 9a). The additional products were primarily lactate and butyrate to variable extents (up to 24.6% and 18.2% of product carbon in lactate or butyrate, respectively; Supplementary Fig. 9b, c); their formation did not correlate with any known growth parameter. H₂ (≤15% in the gas phase) and CO₂ (≤45 %) as well as traces of CH₄ (≤2%) were also detected (Supplementary Fig. 9d–f). There was a slight trend towards higher H₂ concentration at higher growth rates; CO₂ and CH₄ levels appeared uncorrelated. As the gas phase of the chemostats was open to release gas pressure, no associated production rates could be quantified. For determining carbon balance and biomass yields, we estimated biomass from the cell density and calculated cell lysis from DNA content in the supernatant (Supplementary Fig. 10). The carbon balance accounted for 88%-112% of the total glucose carbon consumed (Fig. 1g), indicating that the mass of carbon in our chemostats was balanced within experimental error as there are similar deviations for additional carbon and missing carbon. It also shows that the carbon not found in the abovementioned products was present in biomass. Biomass yield correlated with growth rate (Fig. 1b), presumably due to a higher fraction of energy required for maintenance during slow growth; maintenance energy was estimated³¹ (Fig. 1c) to be 0.5 mmol_ATP g_{dry weight} ⁻¹ h⁻¹.

Overview of proteome analysis and coarse-grained sectors

In our proteome analyses, we detected 79% (1252 of 1577) of all predicted T. kivui proteins across all samples (Supplementary Data 2). Ninety of the 210 reliably quantified proteins changed significantly as a function of growth rate (linear regression, P < 0.05): expression of 72 proteins increased with growth rate, and expression of 14 proteins decreased (Supplementary Data 3). It is likely that the analysis is biased towards abundant proteins, as proteins with low abundance are more difficult to quantify reliably due to low spectral count numbers. Nonetheless, this shows that the abundance of many prevalent proteins is adjusted to some extent as growth rate changes. The fraction of anabolic proteins increased only from 40% to 50% when growth rate increased by almost two orders of magnitude in very slow-growing chemostats compared to batch cultures, correlating with a decrease of the fraction of the catabolic proteome from about 30% to 20% (Fig. 1d). These changes strongly contrast with the 300-fold increase in cell mass production rate, indicating changes in turnover rates. An overview of changes in coarse-grained functional proteome sectors is shown in Fig. 2.

Fig. 2. Mass fractions of proteome sectors in T. kivui cells grown with glucose in chemostat reactors at different growth rates.

For reference, data from exponentially growing batch cells (orange) are included. Asterisks indicate significantly changing sectors (P < 0.05, Benjamini-Hochberg false discovery rate correction, for exact P-values see Supplementary Data 2—P-values sectors). Datapoints, individual reactors; lines, linear regressions; shading, confidence intervals using the linear regression as center. Linear regressions were used to calculate significantly changing proteins and are depicted as guidance for trends in data with scatter, without implying a linear relationship. Red, chemostats; orange, exponentially growing cells from batch cultures. Source data are provided as a Source Data file.

Changes in the catabolic proteome

The catabolic proteome of heterotrophically growing T. kivui consists of two pathways: enzymes of the EMP plus PFOR of glucose oxidation to acetate and of the WLP of CO₂ reduction to acetate, as well as energy-conserving hydrogenase (Ech), ATP synthase, and electron-bifurcating hydrogenase (Hyd). We observed different trends in these two main components, (Fig. 2). The proteome fraction of glycolysis increased non-linearly with growth rate from 8% to 12% at up to about 0.03 h⁻¹, and then decreased to 8% at the highest tested growth rates (Fig. 3a). Notably and in contrast, WLP proteins decreased with increasing growth rate from 20% of the total proteome at the lowest growth rates to only 10%–15% at all other growth rates (Fig. 3b). As expected, the growth rate-dependent increase in glycolysis proteins is reflected in significant trends for many individual catabolic proteins of that pathway (Fig. 4). Interestingly, the protein fraction of the main³² PFOR decreased with increasing growth rate. All proteins of the WLP were present at decreasing fractions in faster-growing cells, similar to the NfnA-subunit of Nfn, the Hyd complex, and ferredoxin. The proteome fractions of the Ech and ATP synthase complex showed overall no significant change. No changes were observed for the levels of acetate kinase and phosphotransacetylase.

Fig. 3. Catabolic proteome of T. kivui grown in glucose-limited chemostats at different growth rates.

For reference, data from exponentially growing batch cells (orange) are included. a proteome fraction of glycolysis (EMP and PFOR); b proteome fraction of WLP; c ratio of the proteome fractions of WLP and glycolysis sectors; d turnover in glycolysis normalized to pathway proteins; e turnover in WLP normalized to pathway proteins. Red, chemostats; orange, exponentially growing cells from batch cultures. WLP Wood-Ljungdahl pathway. Source data are provided as a Source Data file.

Fig. 4. Catabolism and proteomic changes of T. kivui grown in glucose-limited chemostats at different growth rates.

Relative protein abundance compared to exponentially growing reference (ordinate) against growth rate (abscissa) on a linear scale. 2–2.5 acetate per glucose is excreted; the rest is present in biomass. Datapoints, reactors, lines, linear regressions, shading, and confidence intervals using the linear regression as the center. Colored graphs, significant (P < 0.05, Benjamini-Hochberg false discovery rate correction, for exact P-values see Supplementary Data 3—slopedata-reliable) linear changes with growth rate; gray graphs, no significant change; text, not quantified reliably or detected. Linear regressions show trends in data with scatter. GK glucokinase, PGI phosphoglucose isomerase, PFK phosphofructokinase, FBA fructose-1,6-bisphosphate aldolase, TPI triosephosphate isomerase, GAPDH glyeraldehyde-3-phosphate dehydrogenase, PGK phosphoglyerate kinase, PGM phosphoglyerate mutase, ENO enolase; PK pyruvate kinase; PFOR pyruvate:ferredoxin oxidoreductase; PTA phosphotransacetylase, ACK acetate kinase, HDCR hydrogen-dependent carbon dioxide reductase, FTHFS formyltetrahydrofolate synthetase, MTHFC methenyltetrahydrofolate cyclohydrolase, MTHFD methylenetetrahydrofolate dehydrogenase, CODH carbonmonoxide dehydrogenase, ACS acetyl-CoA synthase; MET methyl transferase, CFeSP corrinoid FeS protein. Protein locus tags are abbreviated. Source data are provided as a Source Data file.

Decoupling of catabolic fluxes and proteome fractions

Flux of glucose to three acetate through both pathways is expected to be kinetically coupled via the flux of reducing equivalents and of CO₂. However, the proteome fractions of both pathways are inversely correlated resulting in a varying ratio especially at the very low growth rates (Fig. 3c). Importantly, the changes of the WLP and glycolysis proteome fractions did not match the in situ rates of glucose consumption and acetate production, which increased linearly with growth rate (Fig. 3a, b and Fig. 5a, b). NMR analyses of end products of ¹³C₁-labeled glucose metabolizing chemostat cells confirmed that one third of the acetate produced contained a ¹³C-carbon (Fig. 5c–e), matching the prediction that three acetates are produced from one glucose under these conditions. Therefore, the specific catabolic turnover rates (calculated from production rates via both pathways and the number of respective proteins per cell) changed at least tenfold over the two orders of magnitude change in growth rate, suggesting a kinetic overcapacity of enzymes at slow growth (Fig. 3d, e).

Fig. 5. Glucose catabolism of T. kivui grown in glucose-limited chemostats at different growth rates.

a–e glucose catabolism in chemostats during steady-state cultivation; f–j glucose catabolism capacities of T. kivui cells grown in chemostats at different growth rates with unlimited access to glucose after stopping medium flow (converted to batch system). (a) and (f), rates of glucose consumption; (b) and (g), rates of acetate production; c and h, ratio of moles of extracellular acetate found in the supernatant per mole of glucose degraded; (d) and (i), ratio of ¹³C-labeled acetate produced from ¹³C₁ glucose; (e) and (j), specific rates of acetate produced via WLP and via glycolysis as determined by addition of ¹³C₁-labeled glucose and NMR analysis of acetate, gray line indicates equal rates via both pathways. For (a–e), red, chemostats. For (f–j), red, metabolic rates after addition of 20 mM glucose without changes of the gas phase; open blue, addition of 50% N₂ at glucose spiking; filled blue, addition of 50% H₂ at glucose spiking; gray, during cultivations in chemostats as shown in (a–e). WLP Wood-Ljungdahl pathway. Source data are provided as a Source Data file.

Catabolic capacities with unlimited catabolic substrate

To test whether slow-growing cells indeed have catabolic overcapacities, we investigated ex situ the kinetic capacity of cells for utilizing non-limiting glucose (referred to as capacities, compared to rates during chemostat cultivation) by adding 20 mM glucose to the chemostat reactor immediately after stopping the medium flow (Fig. 5f–j and Supplementary Note 3). For cells grown at rates of up to 0.04 h⁻¹, glucose consumption capacities matched or exceeded the rates during chemostat cultivation. Surprisingly, in faster-growing cells, the apparent glucose consumption capacities were lower than the rates during cultivation (Fig. 5f and Supplemental Note 3). Consistent, congruent trends were also observed for acetate production (Fig. 5g), suggesting apparent inhibition under these conditions. The ratio of acetate formed per glucose consumed was ~2 (Fig. 5h). This is similar to the ratio during chemostat cultivation, for which the carbon balance showed that the rest of the three produced acetate is used for biomass synthesis. ¹³C₁-glucose degradation in these capacity-exploring experiments confirmed that WLP and glycolysis were kinetically balanced in stopped reactors if the headspace contained 50% H₂ (Fig. 5i, j), indicating that the different ratios of protein abundances did not strongly affect the coupling of the pathways. In contrast, in control chemostats containing 50% N₂, leading to only half of the original H₂ and CO₂ concentrations, flux through the WLP was less than through EMP, pointing to a pH₂ threshold for WLP activity.

Cell composition

The macromolecular composition of T. kivui cells only partially mirrored the growth rate-dependent compositional changes of E. coli but clearly differed from the invariant composition of M. maripaludis. While total normalized RNA content of T. kivui cells increased linearly for all cells grown in chemostats, no further increase was observed including in exponential phase cells of batch cultures (Fig. 6a). Notably and in contrast to fast-growing E. coli⁶, DNA content decreased with growth rate (two chromosome copies in slow- and very slow-grown cells versus one copy in batch cultures; Fig. 6b). Moreover, protein content of T. kivui was inversely correlated with growth rate: a protein concentration of nearly 300 μg OD₆₀₀⁻¹ at a growth rate in very slow-growing cells decreased to ~100 μg OD₆₀₀⁻¹ in cells from batch cultures (Fig. 6c), in contrast to the constant protein concentration in E. coli¹³. The RNA/protein ratio, an indicator of ribosome content⁸ (Fig. 6d), increased only 10-fold (0.01–0.11) over a 100-fold increase in growth rate. It was also much lower in slow-growing T. kivui cells compared to E. coli, considering absolute growth rates (50% of the RNA/protein in E. coli at 0.035 h⁻¹) and mostly lower than the constant RNA/protein of 0.13–0.22 of M. maripaludis (Fig. 6e). Generally, these data seamlessly fit into and extend the overall trend of RNA/protein as observed in E. coli. When analyzing relative growth rates (actual growth rate normalized to the maximum growth rate for each microorganism), the RNA/protein ratio showed similar trends, but different absolute values when comparing E. coli and T. kivui (Supplementary Fig. 11), pointing to absolute growth rate as a decisive factor for cellular composition. Due to the low RNA content, we were unable to analyze polysomes, which would give a better understanding of actively translating ribosomes.

Fig. 6. Composition of T. kivui cells grown in a glucose-limited chemostat at different growth rates.

For reference, data from exponentially growing batch cells (orange) are included. a RNA content per biomass in μg per OD₆₀₀ and ml; b number of chromosome copies per cell in μg per OD₆₀₀ and ml; c protein content per biomass in µg per OD₆₀₀ and ml; d RNA/protein ratio as a measure of ribosome content; e RNA/protein ratio in comparison to other organisms. Red, chemostat cells; orange, exponentially growing cells from batch cultures; green triangles, M. maripaludis²³; blue squares, E. coli⁶⁴ (filled, batch cultures; open, chemostat cultures); black squares, Corynebacterium glutamicum¹⁵ (batch cultures). Source data are provided as a Source Data file.

Ribosome content and translation rates

The mass fraction of ribosomal proteins increased from ~3.4% in very slow-growing cells to ~6% in batch cultures (Fig. 7a), matching the trend of proteome fraction of ribosomal proteins in E. coli (Fig. 7c) despite the difference in RNA/protein ratio, and clearly differing from the trend in M. maripaludis. Again, the trends of T. kivui and E. coli matched well when considering absolute growth rate, leading to a smooth transition, but differed more notably considering relative growth rates (Supplementary Fig. 11). The RNA/protein ratio and proteome fraction of ribosomal proteins in T. kivui scaled linearly, however, with an offset so that an RNA/protein of zero would correspond to a proteome fraction of ribosomal proteins of about 2.5% (Supplementary Fig. 12), indicating an overcapacity of ribosomal proteins at low growth rates.

Fig. 7. Ribosome concentrations as well as translation rates in T. kivui cells grown with glucose in chemostat reactors at different flow rates.

For reference, data from exponentially growing batch cells (orange) are included. a proteome fraction of ribosomal proteins in T. kivui; b ribosome concentration in T. kivui cells calculated from the amount of ribosomal proteins (filled circles) or total RNA content (open circles, these numbers were calculated assuming all detected RNA is rRNA and are therefore likely over-estimations); c proteome fractions of ribosomal proteins in comparison to other microorganisms; d translation rates based on ribosome numbers derived from RNA concentrations (assuming 85% of total RNA is rRNA, as found for M. maripaludis²³); e translation rates from (d) in comparison to other microorganisms, all translation rates were calculated as the average rates across all ribosomes as for T. kivui using the following equation: $\frac{\mathrm{protein}\mathrm{per}\mathrm{cell}\times \mathrm{mass}\mathrm{rRNA}}{\mathrm{RNA}\mathrm{per}\mathrm{cell}\times \mathrm{ratio}\mathrm{of}\mathrm{rRNA}\mathrm{in}\mathrm{total}\mathrm{RNA}\times \mathrm{mass}\mathrm{amino}\mathrm{acid}\times \mathrm{doubling}\mathrm{time}}$ using the RNA/protein ratio to determine protein per cell/RNA per cell, the weight of the respective rRNAs in each organism, ratio of rRNA as described in the legend, an average amino acid weight of 109 g mol⁻¹, and doubling time in s. Red, chemostats; orange, exponentially growing cells from batch cultures; green triangles, M. maripaludis²³, translation rates calculated similar to T. kivui; blue squares, E. coli⁶⁴ (filled, batch cultures; open, chemostat culture, assuming 85% rRNA⁶⁴); black squares, Corynebacterium glutamicum¹⁵ (filled, batch cultures, also assuming 85% rRNA as found for E. coli). Source data are provided as a Source Data file.

Ribosome numbers were calculated via two different methods: from RNA or from ribosomal protein by comparing to the mass of each compound per ribosome to the cellular concentrations (Fig. 7b). Both calculations revealed for fast-growing cells ~20,000 ribosomes per μm³ cell volume but differed for cells with lower growth rates. Cells grown at very low rates were predicted to contain 5000 ribosomes per µm³ based on RNA content but >20,000 ribosomes per μm³ based on ribosomal proteins, indicating a four-fold excess of ribosomal protein over available RNA during slow growth. To exclude that this discrepancy is due to single proteins with moonlighting functions, all ribosomal proteins were compared to the maximum number of ribosomes formed from RNA. No single protein was present at a higher fraction, but many proteins were present in up to 10-fold excess (Supplementary Fig. 13a). Neither the abundance of RNA polymerase nor RNases correlated with this mismatch (Supplementary Fig. 13b). Assuming that 85% of total RNA is rRNA^8,23, predicted translation rates of T. kivui ribosomes (using growth rate, protein content of cells, and ribosome numbers calculated from RNA content) decreased from ~10 aa s⁻¹ at a growth rate in batch cultures to 0.7 aa s⁻¹ at very low growth rates (Fig. 7d). The latter rate is lower than the translation rates observed so far in E. coli (2 aa s⁻¹ average translation rate at 0.035 h⁻¹) but matches the trend of decreasing translation rate with decreasing growth rate in E. coli and again smoothly fits to and extends the curve of E. coli. The translation rate of T. kivui, however, is higher than in M. maripaludis.

Cell shape and cell volume

T. kivui mean cell volume was inversely correlated with growth rate, increasing from ~0.5 μm³ at high growth rates to >1 μm³ at low and very low growth rates, while the surface area/volume ratio remained relatively constant (Fig. 8a–d and Supplementary Fig. 14). This trend is opposite of that of E. coli, where cell volume is directly correlated with growth rate³³ and different from the constant cell size of M. maripaludis (Fig. 8e). Higher variation in cell volume at lower growth rates suggests increased morphological heterogeneity. Indeed, fast-growing cells were homogeneously rod-shaped, while very slow-growing cells grown were morphologically more heterogeneous, with mostly more elongated cells being present at lower growth rates (Fig. 8f, g and Supplementary Note 3). This shows that the apparent increase in cell size is rather due to this heterogeneity than an overall increase in cell size across the entire population. No indication of the presence of a subpopulation of elongated cells was found (Supplementary Note 3). Although T. kivui was described to be non-spore forming, some spore-like structures were observed (frequency <0.1%; Supplementary Fig. 15).

Fig. 8. Cell shape of T. kivui cells grown in glucose-limited chemostats at different growth rates.

For reference, data from exponentially growing batch cells (orange) are included. a cell length; b cell width; c cell volume; d surface area/volume ratio; e cell volume in comparison to other organisms; f representative microscopy images of T. kivui cells grown at 0.002 h⁻¹; g representative microscopy images of T. kivui cells grown at 0.07 h⁻¹. For (a–c), n = 39–1055 (Cellular dimensions are mean values of ≥39 cells measured from microscopy images as indicated in Supplemental Data 1), symbols indicate mean values, error bars indicate standard deviation. For (f–g), 25 microscopy images were acquired for each datapoint and representative images from the slowest and fastest growth rates are shown. Red, chemostats; orange, exponentially growing cells from batch cultures; green triangles, M. maripaludis²³; blue squares, E. coli³⁴. Source data are provided as a Source Data file.

Macromolecular densities

To probe the relationship between cell size and biomass composition, we calculated concentrations of macromolecules (protein, DNA, and RNA). The macromolecule concentration in the cytoplasm decreased with increasing growth rate, mostly due to the decrease in protein concentration, which, at the highest growth rates, matched the protein concentration in the cytoplasm of E. coli of ~0.3 pg μm⁻³ (versus ~1 pg μm⁻³ at the very low growth rates). No comparable data is available for M. maripaludis. Notably, protein concentration is invariant in E. coli cultures oxidizing glucose^14,34 (Fig. 9a, b). The fraction of envelope proteins in T. kivui (~15%) was lower than that in E. coli (~20%) and decreased slightly with increasing growth rate (Fig. 9c). The membrane proteome of T. kivui consisted of about 40% transport proteins, 5-10% of each Ech complex and ATP synthase proteins, as well as 10–15% surface layer protein, which covers the surface of T. kivui cells (Supplementary Fig. 16). Despite a decreasing protein density in the cytoplasmic membrane with increasing growth rate (Fig. 9d), its relative proteome composition remained constant (Supplementary Fig. 16). Compared to E. coli, no electron transport chain proteins are present in the cytoplasmic membrane of T. kivui. However, the protein density in the cytoplasmic membrane of Gram-positive T. kivui was 5- to 40-fold higher than the protein density in the inner membrane of Gram-negative E. coli at high growth rates (Fig. 9d), which could point to novel differences in membrane/cell wall biology between Gram-negative and Gram-positive microorganisms in general.

Fig. 9. Macromolecular densities in T. kivui cells grown in glucose-limited chemostats at different growth rates as well as the distribution of proteins among the cytoplasm and envelope compared to E. coli.

For reference, data from exponentially growing batch cells (orange) are included. a Cytoplasmatic density of macromolecules (protein, RNA, DNA per cell volume); b cytoplasmic protein concentration in comparison to E. coli; c proteome mass fraction of envelope proteins compared to E. coli; d protein density in the membrane of T. kivui and in the inner membrane of E. coli. Red, chemostats; orange, exponentially growing cells from batch cultures; blue squares, E. coli³⁴; gray, T. kivui chemostat with unusually low cell counts per OD₆₀₀, which affect the calculation of intracellular macromolecule densities. Due to the outlier in cell counts, we would consider this datapoint an outlier (Supplementary Fig. 17). Source data are provided as a Source Data file.

Discussion

The response of microorganisms to changing nutrient fluxes does not categorically follow the long-held E. coli paradigm consisting of tight gene regulation and strict correlation of ribosome numbers with growth rate, but can differ depending on the ecophysiology of the respective microbe. In contrast to E. coli growing aerobically using glucose, the hydrogenotrophic methanogen M. maripaludis does not change catabolic and anabolic systems-level parameters depending on growth rate under catabolic limitation. Our present study in T. kivui revealed another biological strategy: while cellular processes are regulated only to a limited extent on the proteome level, the microbe’s metabolic rate as a function of growth rate relies largely on posttranslational regulation (Supplementary Fig. 6c and Supplementary Table 1). The growth rate-dependent change in ribosome numbers is nonlinear and mostly due to limiting RNA content rather than ribosomal proteins. These findings suggest that growth rate-dependent coordination of major cell-physiological processes in microbes can follow multiple distinct strategies, which challenges the general belief that strict regulation of the costly proteome is essential for competitive growth in microbes.

First, the catabolic proteome of T. kivui was not regulated proportionally to the catabolic fluxes at different growth rates, therefore leading to reduced turnover rates and overcapacities in the proteins of glycolysis and WLP at slower growth. Due to the changes in the abundances and the ratio of catabolic enzymes for glycolysis and WLP, we expected a significant shift in the rate and spectrum of end products, e.g., production of CO₂ and H₂ as overflow due to reduced WLP capacity. We did not observe any such change, but observed balanced metabolic fluxes despite changing enzyme ratios. Additionally, the catabolic rates also did not match the protein abundances when adding unlimited glucose. This decoupling of protein abundance and metabolic flux strongly suggests the presence of posttranslational regulation in T. kivui, which was also reported in other acetogenic Clostridia^24,25,35,36. Lastly, despite no clear correlated changes in the proteome, we observed formation of lactate and butyrate under some conditions, indicating activities of other pathways for re-oxidizing reduced electron carriers and further pointing to the decoupling of metabolic fluxes and proteome composition. The lower activity of WLP when pH₂ was decreased, together with the constant pH₂ in most cultures, points to the pH₂ for controlling WLP activity in T. kivui. Regulation of catabolic proteins on a posttranslational level compared to transcriptional regulation might allow for faster metabolic response to changing conditions such as nutrient availability. This might point to an ecophysiological strategy of a very flexible and rapidly responding metabolism, which is especially interesting for microorganisms that are able to utilize gaseous substrates at highly varying availability, for example, other Clostridia and M. maripaludis.

Second, we observed ribosomal overcapacities leading to reduced translation rates at lower growth rates and, more strikingly, an overcapacity of ribosomal proteins compared to ribosomal RNA. This further contradicts the belief that strict regulation on the ribosomal protein level is necessary to ensure competitive growth. A well-matched, decreasing translation rate with decreasing growth rate was also found in C. glutamicum and E. coli^15,21. In E. coli and C. glutamicum, the reduction was due to a generally lower rate of protein synthesis or a higher fraction of inactive ribosomes as determined by direct measurements of translation rates, respectively^15,21, while the down-regulation of translation rate in T. kivui was not determined here. In general, a ribosomal overcapacity seems to be a prevalent strategy in microbes. However, while ribosomal protein fractions in T. kivui were comparable to the fractions in E. coli, the RNA/protein ratio of T. kivui continued to decrease at growth rates <0.2 h⁻¹, at which the ratio remains constant in E. coli¹⁹. (It is important to note that the E. coli data were partly derived from slow-growing batch cultures that might not have reached equilibrium²¹ and might, therefore, not be comparable.) This indicates that T. kivui adjusts ribosomal content i) even under conditions, at which E. coli has an overcapacity of inactive ribosomes, and ii) based on RNA content instead of on the proteome level. The resulting diverging trends of ribosomal proteins and RNA lead to a reservoir of unincorporated ribosomal proteins, which could be advantageous once limiting environmental conditions improve. While rRNA can be produced rapidly and at lower metabolic cost by already present RNA polymerases, a costly synthesis of new ribosomal proteins would compete with the synthesis of proteins for other cellular functions. Similarly, the low translation rates at slow growth, which might indicate the lowest ribosome concentration needed to effectively cope with diffusion limitation in a cell, can be advantageous for inexpensively restarting growth. This scenario could also explain the well-matched ribosomal protein fractions in T. kivui and E. coli, leading to a smooth transition between the trends. This continuous trend is striking considering the differences in ecophysiology, growth rate range and phylogenetic distance between these organisms. Indeed, this is unlikely to be incidental and points to a universal biological principle affecting the ribosome concentration needed for growth and shows the importance of absolute growth rate for some aspects of growth physiology. This study, therefore, significantly extends the recently found common ribosomal protein allocation strategy for the phylogenetically distant, but ecophysiologically more similar (fast growth, flexible catabolism) E. coli, Vibrio natriegens, and Bacillus subtilis⁹.

Another aspect commonly considered as a bacterial growth law is the increase of cell size with increasing growth rate as found in E. coli, B. subtilis, and other microbes^16,33,37. T. kivui, however, showed a trend toward elongated cells at very low growth rates. In E. coli, the cell size increase is thought to be due to higher DNA content because simultaneous replication of multiple chromosomes is required to compensate for generation times shorter than chromosome replication times^6,33. In T. kivui, the cell doubling time is sufficiently long to accommodate replication of the smaller, 2.4 Mbp genome, and only one chromosome copy was found per cell at the highest tested growth rates. Instead, elongated cells might result from attenuated cell division at low nutrient flux at the lower growth rates (medium growth rate and lower, <0.045 h⁻¹), which might point to a transition from slower growth to starvation. Aspects of a starvation response would also be consistent with the increase in cytoplasmatic protein density³⁸. However, the fact that the changes in cell size are also found at much higher growth rates (medium growth, 15 times faster than the very slow growth rates) could also indicate that the changes in cell size are connected to other cellular strategies and metabolic responses. Strikingly, our study showed protein and macromolecule concentrations are much higher in T. kivui than in E. coli. While the observed variation of the T. kivui data makes exact comparisons unreliable, the general trend is towards higher concentrations than E. coli, especially at slower growth. Additionally, the macromolecular concentration in E. coli can be quite variable and seems to be connected to growth rate: Faster-growing variants of E. coli in a long-term evolution experiment had a 50% lower dry mass concentration³⁹, and biomass concentration was also shown to increase over two-fold during starvation⁴⁰. While higher cytoplasmic protein density reduces diffusion of macromolecules, a more crowded cytoplasm can be advantageous for stabilizing RNA, enzyme complexes for enhanced metabolic fluxes, or enhancing formation of protein aggregates with additional functions⁴¹.

In contrast to most model microorganisms, T. kivui is thermophilic, which might affect maintenance energy, e.g., by increased recycling of macromolecules. The biomass yield as well as the maintenance energy coefficient were within the range found for other microbes such as E. coli and M. maripaludis²³. As for other thermophilic Clostridia³⁸, this finding indicates that thermophilic growth can be sustained by thermostable macromolecules instead of increased turnover.

The overcapacities of catabolic and ribosomal proteins at low growth rates challenge the universal idea that strict regulation of protein synthesis is important for metabolic efficiency^14,42. However, less transcriptional regulation can also be advantageous due to the metabolic and genomic cost of regulatory machineries as well as necessary adjustment intervals, particularly when changes are too small or too fast to justify re-organization of the proteome⁴³. In fact, the T. kivui genome encodes 1.5 annotated transcriptional regulators per 100 kb of genomic DNA, while E. coli encodes 6 regulators per 100 kb⁴⁴. A metabolic and ribosomal reserve (i.e., overcapacity) helps in scavenging substrates once they become available and in restarting growth quickly. As evidence for metabolic regulation that is different from the E. coli model is emerging for many microbes, in particular in acetogens^{24,25,35,36,45,46}, systems-level insights such as the comprehensive data presented here point to the relevance of other strategies and can help refine metabolic models for microbial ecosystems as well as biotechnological production strains^27–29.

By studying the growth of T. kivui at very low growth rates (>0.002 h⁻¹), we were able to expand the range of well-studied growth rates for bacteria and show that, in contrast to fast-growing bacteria with similar ecophysiology⁹, bacteria adapted to different growth rates and with less flexible catabolism can have different resource allocation strategies. Our results also show that the limited regulation of the proteome composition in M. maripaludis is not due to the growth rate range, and that at least the bacterium T. kivui has a more differentiated proteome regulation depending on growth rate. To further untangle the ‘periodic system’⁹ of growth rate-dependent regulation, metabolism, and ecophysiology, more detailed data from additional organisms will be needed.

Methods

Strains and cultivation

The uracil-auxotrophic strain T. kivui ΔpyrE (TKV002)⁴⁷ was grown anaerobically in chemostats using glucose as a carbon and energy source. For T. kivui, a buffered minimal medium (MTki) was used, consisting of 100 mM 2-(N-morpholino) ethanesulfonic acid (MES), 13.7 mM NaCl, 0.8 mM MgSO₄, 18.7 mM NH₄Cl, 1.3 mM KCl, 0.1 mM CaCl₂, 0.7 mM KH₂PO₄, 1X trace element solution SL10⁴⁸, 11 nM Na₂SeO₃, 13.6 nM Na₂WO₄, 0.04 mM uracil, 1 mg ml⁻¹ resazurin, 1.5 mM cysteine, and 1 mM Na₂S, adjusted to pH 7.0 using NaOH with 20 mM glucose. Cysteine, Na₂S, and glucose were added anoxically after autoclaving. Batch cultures and cultures for strain maintenance were cultivated in anoxic 60 ml serum bottles with butyl stoppers containing 25 ml of medium unless stated otherwise. For serum bottles, the medium was boiled and cooled under N₂ atmosphere before autoclaving. For medium reservoirs, the medium was autoclaved oxically, sparged with N₂ while still hot after autoclaving, and cooled under N₂. Sampling and inoculation were performed using sterile needles and syringes. For strain maintenance, T. kivui was transferred to new cultures every 7–14 days, and new aliquots were grown from cryostocks before experimental series. Cell density was measured as optical density at 600 nm (OD₆₀₀) (Ultraspec 2100 pro, Amersham Biosciences, for cuvettes and CO8000 cell density meter, Biochrom WPA, for reaction tubes and side arm reactors). All T. kivui cultures were cultivated at 65 °C. HPLC samples were withdrawn using the sampling tube for chemostats or syringes for batch cultures, and centrifuged (5 min, 16,000 g, room temperature) or filtered (0.2 µm pore size, 13 mm diameter, PVDF, Agilent) before measurements. Gas sampling was performed using gas-tight syringes (VICI).

Chemostat setup and cultivation

T. kivui cells were grown in media in 100-ml wide-neck bottles with a side arm for OD₆₀₀ measurements (Adams and Chittenden) with butyl stoppers (Glasgeraetebau Ochs) (Supplementary Fig. 18). The stoppers had a PEEK sampling tube, closed by a two-way stopcock with Luer connector and sterile filter, as well as a temperature probe (NTC (10 K 0.5%) Waterproof Sensor, matching the temperature control). The temperature probe was connected to a temperature control (type W1209, 12 V digital thermostat) outfitted with two polyimide heating strips (adhesive polyimide heating strips, 25 mm × 50 mm, 12 V 7 W) that were taped (heat-resistant polyimide tape) onto the reactor and connected to an adequate power source. Reactors were autoclaved before usage, and stoppers with temperature probes were sterilized using 70% ethanol. All reactors contained magnetic stir bars with the same shape and size and were stirred at 300 rpm using tenfold stir plates (RO10 magnetic stir plate, IKA). Medium was added to the reactors from a reservoir (2–5 l bottles with butyl stopper) that were connected to a N₂ line, keeping the gas phase at an overpressure of 5–15 psi and anoxic. The medium reservoir was connected to the reactors using PEEK (Idex-HS), Viton (most tubing, 1/8” inner diameter, 96412–16, Masterflex), or Norprene (in pumps, A-60-G, 1/16” inner diameter, AFL00003, Tygon) tubing and polypropylene connectors that were autoclaved before usage. Tubing was connected to the reactors by needles. Medium was pumped into reactors using peristaltic pumps (205S, Watson Marlow) at a set pump speed. The culture volume was maintained at approximately 50 ml via an outlet needle (18 g × 3”, Air-tite) connected to Viton tubing by polypropylene connectors and to waste bottles by one-way valves to prevent backflow. Waste tubing was autoclaved, and one-way valves were sterilized using 70% ethanol before assembly. After assembly of the reactors, the gas phase was changed to N_2, and the reactors were connected to the medium. In case of oxygen contamination, additional Na₂S was added, and the gas phase was exchanged. During cultivation, the gas phase was monitored, but not exchanged, and gas formation was not quantified. pH was monitored to be stable at pH 6 in the reactors during steady-state cultivation. Up to 18 reactors were grown in parallel using up to three media in parallel. While most chemostats were cultivated using the medium as described above, chemostats at the highest flow rate (>0.05 h⁻¹, setting 2 rpm) were cultivated using medium with only 10 mM glucose to ensure low glucose concentration in the chemostat reactors. After inoculation from batch cultures, cultures were grown without medium flow until growth was visible in all reactors before the medium flow was started. For most flow rates, flow was immediately started at the final flow rate. For the highest flow rates, flow was first started at 50% of the final flow rate for several days before increasing to the final flow rate to ensure active growth. All chemostats were grown for at least four retention times before chemostat cultures were sampled, so that the growth rate could be assumed to equal the dilution rate. Therefore, growth rate and dilution rate are used interchangeably in this manuscript. During cultivation, OD₆₀₀ was determined via the side arm, and HPLC and gas samples were analyzed regularly. Cell morphology and numbers were determined by microscopy. For final sampling, samples were withdrawn, cooled on ice, and either used for cell fixation or harvested by centrifugation (2 ml culture per pellet, 5 min, 4 °C, 16,000 g). Cell pellets were frozen in dry ice/ethanol baths without further washing and stored at −80 °C. For cell fixation, 1 ml of culture was mixed with 1 ml of fixing solution (3% formaldehyde and 3% glutaraldehyde in Milli-Q water (H₂O_MQ)). Cells were fixed overnight, washed twice with 500 µl of phosphate-buffered saline (PBS) (3 min, 4 °C, 16,000 g) and resuspended in 100 µl of ethanol/PBS 1:1 for storage at −20 °C. Exponentially growing cells as controls were harvested from batch cultures using the same protocol. Dilution rates were calculated from the exact volume of culture waste collected in waste bottles over different time intervals, as well as the exact culture volume.

Determination of biomass

OD₆₀₀ was used as an initial indicator for biomass concentrations of chemostat cultures. As OD₆₀₀ in reactors was always measured undiluted in the side arms, a correlation curve of this undiluted OD₆₀₀ and a more exact OD₆₀₀ after dilution was determined (corrected OD₆₀₀; Supplementary Fig. 19a) to correct for the underestimation of OD₆₀₀ at high OD₆₀₀ values. While the OD₆₀₀ of all chemostat cultures was determined via the side arm (same material and diameter as Hungate tubes), other cultures were grown in serum bottles, and OD₆₀₀ was determined in cuvettes (d = 1 cm). To compare these cultures, we established that OD₆₀₀ measurements in Hungate tubes and cuvettes were correlated (Supplementary Fig. 19b). For the calculation of yields, dry biomass concentrations were calculated from OD₆₀₀ measurements. The dry biomass concentration of T. kivui cultures was determined using cells that were harvested from 250 ml of batch cultures at various growth stages (exponential phase, late exponential phase, and stationary phase), collected in pre-weighed and pre-dried Eppendorf tubes, and dried at 100 °C until the weight did not decrease anymore. In this manner, dry biomass concentration was determined to be 0.23 mg ml⁻¹ OD₆₀₀⁻¹ measured in tubes (0.34 mg ml⁻¹ OD₆₀₀⁻¹ measured in cuvettes). We noticed that very slow-growing chemostat cultures had a darker color than other cultures, probably due to precipitation, which might lead to an overestimation of cell density by OD₆₀₀.

Determination of catabolic rates and capacities

Steady-state glucose consumption and acetate production rates in the reactors were calculated from in situ concentrations and flow rates. To determine the maximal capacity of chemostat cells for glucose degradation and acetate production, the medium flow of the chemostats operating at different dilution rates was stopped, and the chemostats were immediately spiked with anoxic 20 mM glucose. Glucose and acetate concentrations, as well as OD₆₀₀ were measured over a period of at least 6 h. This type of in situ analysis of capacities was chosen rather than using cell suspensions to avoid introducing perturbations by harvesting and washing, and thereby delaying the onset of glucose degradation, which might also allow for altered protein production. To prevent de novo protein synthesis when high glucose concentrations were added, kanamycin (50 µg ml⁻¹) was added to some cultures. Where indicated, we added 50% H₂ or N₂ to some reactors to analyze the effect(s) of the gas phase composition on glucose degradation rates. For this, 50 ml of the respective gas was added to the reactor using sterile filters. The reactors were shaken several times to ensure mixing, and overpressure was released via the outlet needle. The outlets of these reactors were then closed, and gas phase composition was monitored in addition to HPLC and OD₆₀₀ measurements.

To determine the fractions of acetate produced by glycolysis and the WLP ¹³C₁-labeled glucose was added to some reactors and glucose degradation experiments. Samples were then analyzed by both HPLC and NMR.

To analyze the effect of H₂ on glucose degradation rate in exponentially growing batch cultures, 50 ml batch cultures in 120 ml serum bottles were inoculated and monitored for growth, glucose degradation and product formation, as well as gas phase composition. When the culture reached ~50% of the maximum respective OD₆₀₀ in batch, 30 ml of H₂ were added to the test cultures, resulting in a gas phase of approximately 50% H₂, while control cultures were not treated. These batch cultures were started with either 100% N₂ or 80% N₂/20% CO₂ as gas phase.

Determination of substrate and product concentrations

Concentrations of glucose, acetate, and other products were analyzed via HPLC (1260 Infinity, Agilent) using an Aminex HPX-87H column (BioRad) together with an UV detector set to 210 nm and a refractive index detector as described⁴⁹. An isocratic method with 100% 5 mM H₂SO₄ as mobile phase and 0.7 ml min⁻¹ flow rate for 30 min was used for all measurements. Compound concentrations were determined using respective standards.

Gas phase composition was determined by two separate GC measurements. CO₂ concentration was measured using GC (6890 N with 30 m × 0.530 mm GS-Q capillary column, Agilent) with He as carrier gas (22 PSI, 7.4 ml min⁻¹, 250 °C) and thermal conductivity detection (150 °C) as described⁵⁰. H₂ and CH₄ concentrations were analyzed using a multiple gas analyzer (8610 C with 6’ Hayesep D column, SRI) with N₂ as carrier gas (50 °C, 12 PSI N₂) and thermal conductivity detection (150 °C) as described⁴⁹. Dissolved inorganic carbon concentrations in batch cultures were determined as described⁵⁰ by adding 150 μl of sample to 850 μl of 800 mM HCl in a 10 ml glass vial (Agilent) with N₂ atmosphere at atmospheric pressure before measuring CO₂ in the headspace.

NMR spectra were acquired on a Bruker Avance NEO 500 MHz instrument in the Stanford Sarafan ChEM-H Knowledge Center. Samples were transferred to NMR tubes (Wilmad 5-mm high-throughput NMR tubes) and spiked with D₂O to 10% v/v. One-dimensional 1H spectra were acquired with a spectral width of 10416.667 Hz with 16 scans, 8 s acquisition time, and 6 s relaxation delay. MestreNova was used for spectra analysis using 0.1 Hz exponential apodization and zero-filling to 512 K data points.

Determination of cell lysis rate

To estimate cell lysis, DNA and protein concentration in the culture supernatants were measured. Extracellular free DNA was determined by measuring total DNA using the Qubit dsDNA HS assay (Invitrogen) and after addition of DNAse I (NEB) and incubation at 37 °C for 24 h. Protein concentration was measured using the Pierce Bradford protein assay (Thermo Scientific). To infer the number of lysed cells from these concentrations, the concentrations were compared to the protein and DNA content of cells determined as described below.

Cell size quantification

The size of individual cells was determined using a method similar to previous studies^23,51. In brief, fixed cells were diluted to a final OD₆₀₀ of ~0.1 in Tris-EDTA (TE) buffer containing 0.25% Tween-20, 200 µg ml⁻¹ RNAse A (NEB). One hundred microliters of each sample were added to wells of a 96 well-plate and incubated for >2 h before microscopy. Analyses were performed using duplicates of all samples. Samples were transferred from the 96-well plates to 1% agarose pads made with 0.85X PBS using a 96-well pin array (Singer Instruments, REP-001) for single-cell imaging. Phase contrast and fluorescence images were acquired with a Ti-E inverted microscope (Nikon Instruments) using a 100X (NA 1.40) oil immersion objective and a Neo 5.5 sCMOS camera (Andor Technology). Images were acquired using μManager v.1.4⁵². High-throughput imaging was performed using the MATLAB-based imaging protocol SLIP⁵³. Phase-contrast images were initially segmented using the deep learning-based software DeepCell⁵⁴, and segmentation was checked manually to ensure accuracy. Segmented images were then analyzed using Morphometrics⁵¹ to obtain cell contours at sub-pixel resolution. A local mesh was generated for each cell contour to determine the longitudinal axis using a method adapted from MicrobeTracker⁵⁵. Cellular dimensions, including length, width, and cell area, were calculated from the contours for each growth condition. Cell volume was calculated from the cell contour assuming axial symmetry.

Cell number and DNA content determination

Cell counting and DNA content determination were performed via FACS at the Stanford Shared FACS Facility using a BD Influx cell sorter (BD Biosciences) or Acea Novocyte Quanteon (Agilent). Fixed cells were diluted to ~10⁵ cells ml⁻¹ in 500 μl of TE buffer with 50 µL of CountBright absolute counting beads (Invitrogen/Molecular Probes), 0.25% Tween-20, 200 μg ml⁻¹ RNAse A (NEB), and 1X Pico488 (Lumiprobe). Sample mixtures were incubated for >2 h before FACS analysis. Rifampicin-treated E. coli control cells were used as a reference for DNA content determination. These E. coli cultures were grown in LB medium⁵⁶ to the early exponential phase and then treated with 0.2 g l⁻¹ rifampicin for several hours, so that each cell only contained one or two chromosome copies. Flow cytometry data were analyzed using FlowJo v.10 software (FlowJo LLC), and gating was performed in SSC versus FSC to exclude noise as found in the medium control (Supplementary Fig. 20). In addition, cells grown at different flow rates were counted after dilution using a counting chamber (Petroff-Hausser counting chamber, EMS).

Quantification of total RNA

Extraction and quantification of total RNA from T. kivui cells was performed as previously described^57,58. In brief, cell pellets were washed once with 0.3 ml of ice-cold 100 mM HClO₄ and resuspended in 0.3 ml of 300 mM KOH. After incubation at 37 °C with constant shaking for 1 h, 0.1 ml of 3 M HClO₄ were added. Samples were centrifuged (4 °C, 5 min, 16,000 g) and the pellet was washed once with 1 ml of 0.5 M HClO₄. Supernatants in HClO₄ were combined and stored at 4 °C until measurement. RNA concentration was measured at 260 nm photometrically (Ultraspec 2100 pro, Amersham Biosciences) using quartz glass cuvettes (average molar extinction coefficient of RNA 10.5 mmol⁻¹ cm⁻¹, average molecular weight of E. coli RNA nucleotide 324 g mol⁻¹ used as best estimates for calculations).

Quantification of total protein

Cell lysis and protein quantification were performed by hot alkaline lysis and the Pierce Bradford protein assay (Thermo Scientific). Sample pellets were washed twice with 0.6 H₂O_MQ and cells were resuspended in 0.3 ml of 1 M NaOH. For cell lysis, suspensions were subjected to three freeze-thaw cycles (room temperature water bath versus dry ice in ethanol) and then incubated at 100 °C for 10 min. After cooling in a water bath for 5 min, cell debris was pelleted (5 min, 4 °C, 16,000 g). The pH of the supernatant was adjusted using 1 M HCl, and protein concentration in the supernatant was determined using the Bradford assay in 96 well-plates. Protein standards were subjected to the same lysis protocol of freeze-thaw cycles and incubation at 100 °C under alkaline conditions as the samples.

Proteome analyses

¹⁵N-labeled cells were added to all samples for proteome analysis to ensure more quantitative proteome comparisons^59,60. For generating these cells, T. kivui was grown in batch cultures using Na¹⁵NH₄. Cells were harvested in the exponential phase by centrifugation as described for chemostat samples. Protein content of samples and ¹⁵N-labeled controls was determined as described above, and samples were mixed to 1:1 protein content of sample to protein content of ¹⁵N-labeled control after resuspending the labeled control in ~10 µl of H₂O_MQ.

Each cell pellet was resuspended in ice-cold water to a protein concentration of ~50 μg μl⁻¹. One hundred microliters of the pellet solution were removed, mixed with 25 µl of trichloroacetic acid, and incubated overnight at 4 °C. Precipitated cell debris was centrifuged, the supernatant was removed, and the pellet was washed twice with HPLC-grade acetone and subsequently dried at 50 °C for 10 min. The pellet was then resuspended in 40 μl of freshly prepared 8 M urea in 1 M ammonium bicarbonate, vortexed vigorously and placed in a sonication bath for 10 min. Protein disulfide bonds were reduced by adding 2 μl of 200 mM dithiothreitol and incubating at 37 °C for 15 min. Then, cysteine residues were alkylated by adding 8 μl of 200 mM iodoacetamide and incubating at room temperature in the dark for 30 min. The alkylation reaction was quenched by adding 4 μl of 200 mM dithiothreitol and continuing incubation for another 30 min. Urea was diluted by adding 200 μl of water containing 1 mM CaCl₂ to support enzymatic digestion. Ten microliters of 0.1 μg μl⁻¹ MS grade trypsin protease in 50 mM acetic acid were added and incubated overnight at 37 °C for 16 h. Digestion was boosted by adding 3 μl of the trypsin solution and incubating for another 4 h. Tryptic peptides were purified and concentrated using Pierce C18 Spin Columns (Thermo Fisher Scientific) following the manufacturer’s protocol. Desalted lyophilized peptides were resuspended in 30 μl of formic acid and analyzed with a Triple TOF 5600 (Sciex) mass spectrometer.

The raw data from the mass spectrometer were converted into separate mzML and mzXML file formats at the MS1 and MS2 levels, respectively, using MSConvert (ProteoWizard). The MS2 spectra were searched against a computed peptide database using the X!Tandem search engine⁶¹. The resulting peptide spectral matches were used as guides to extract the corresponding MS1 envelopes using an in-house software suite, Massacre, and fit using the previously published algorithm Isodist⁶². Unreliable fit results were filtered out using a previously trained classifier model. The ratio of ¹⁴N/¹⁵N labeling for each peptide obtained from fitting was then normalized by the total ¹⁴N/¹⁵N amplitudes in the sample to correct for any deviation from a 1:1 sample: reference mixing ratio. The relative ¹⁴N/¹⁵N abundance (or unlabeled fraction ¹⁴N/(¹⁴N + ¹⁵N)) for each protein was calculated as the median over its quantitated peptides. The uncertainty of relative abundance was estimated as the 68% interquantile range of deviations from protein median for all the peptides in the sample ( ≈ 0.08 with unlabeled fraction), or all the peptides within the protein of interest if it had at least ten peptides, then normalized by the square root of the number of peptides. An uncertainty cutoff at 0.08 was applied before further analysis of the data.

Proteome sectors were quantified by adding the mass fraction of corresponding proteins determined by spectral counting, and individual proteins were quantified by relative quantification against the ¹⁵N-labeled reference. Growth rate dependence was determined using a linear model in R v.4.0.2 to test for significant nonzero slopes (P-values were adjusted for multiple comparisons using the Benjamini–Hochberg false discovery rate correction⁶³). Mass spectrometry proteomics data are available on at the MassIVE database (https://massive.ucsd.edu) under accession number MSV000097351.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

F.M.M., A.L.M. and A.M.S. conceived the project and designed the experiments. F.M.M. and A.L.M. conducted the experiments and formal analysis. The investigation was carried out by F.M.M., A.L.M., W.G., F.A.N., J.S. and J.K.A. Resources were provided by K.C.H., J.R.W. and A.M.S. Data curation was handled by F.M.M., A.L.M., F.A.N. and J.S. FMM and A.M.S. wrote the original draft of the manuscript, and F.M.M., A.L.M., W.G., K.C.H. and A.M.S. contributed to review and editing. Visualization of the data was done by F.M.M. and A.L.M. A.M.S. supervised the research, and funding acquisition was managed by A.M.S. and F.M.M.

Peer review

Peer review information

Nature Communications thanks Srividya Iyer-Biswas, who co-reviewed with Charles Wright, and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The proteome dataset generated in this study has been deposited in the MassIVE database (https://massive.ucsd.edu) under accession code MSV000097351 and the analyzed datasets are available in Supplementary Data 2, and 3. All other study data are included in the article, the corresponding supporting material, and Supplementary Data 1. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-63432-z.