A derivate of Clostridium autoethanogenum DSM 10061 strain–DSM 23693—deposited in the German Collection of Microorganisms and Cell Cultures (DSMZ) was used in all experiments and stored as a glycerol stock at -80°C. Cells were grown either on CO (60% CO and 40% Ar; AS Eesti AGA) or syngas (50% CO, 20% H₂, 20% CO₂, and 10% Ar; AS Eesti AGA) in a chemically defined medium without yeast extract described before (Valgepea et al., 2017a). Cells were grown under strictly anaerobic conditions at 37°C and at pH 5 maintained by 5M NH₄OH. Chemostat continuous cultures were performed in 1.4 L Multifors bioreactors (Infors AG) at a working volume of 750 mL connected to a Hiden HPR-20-QIC mass spectrometer (Hiden Analytical) for online high-resolution off-gas analysis. The system was equipped with peristaltic pumps; mass flow controllers (MFCs); pH, oxidation-reduction potential (ORP), and temperature sensors. Antifoam (435530; Sigma-Aldrich) was continuously added to the bioreactor at a rate of 10 μL/h to avoid foaming. Chemostat cultures were run at three dilution rates: ∼1.0, ∼2.0, and ∼2.8 day⁻¹ (µ ∼0.04, 0.08, and 0.12 h⁻¹, respectively) with variable gas-liquid mass transfer conditions (gas flow rate and agitation) to maintain similar steady-state biomass concentrations (Table 1). All steady-state results reported here were collected after optical density (OD), gas uptake, and production rates had been stable for at least 3‒5 working volumes.

Biomass concentration in gram of dry cell weight per litre of broth (gDCW/L) was determined by measuring the OD of the culture at 600 nm after the correlation coefficient (K) between culture OD and DCW was established at 0.23, using the methodology described in Peebo et al., 2014.

Analysis of exo-metabolome was performed using filtered broth samples stored at -20°C until analysis. Organic acids and alcohols were analysed by HPLC (Shimadzu Prominence-I LC-2030 plus system) using a Rezex™ ROA-Organic Acids H⁺ (8%) 300 × 7.8 mm column (00H-0138-K0; Phenomenex) and a guard column (03B-0138-K0; Phenomenex). Twenty microlitres of the sample were injected using an auto-sampler and eluted isocratically with 0.5 mM H₂SO₄ at 0.6 mL/min for 30 min at 45°C. Compounds were detected by a refractive index detector (RID-20A; Shimadzu) and identified and quantified using relevant standards using the software LabSolution (Shimadzu). We note that cells produced 2R,3R-butanediol.

Bioreactor off-gas analysis for determination of specific gas uptake (CO and H₂) and production rates (CO₂, ethanol) (mmol/gDCW/h) has been described before (Valgepea et al., 2017a). The Faraday Cup detector monitored the intensities of H₂, CO, ethanol, H₂S, Ar, and CO₂ at 2, 14, 31, 34, 40, and 44 amu, respectively. Argon was used as the inert gas component in the feed gases instead of nitrogen as the latter interferes with the mass-spectrometer signal for CO.

Carbon recoveries and balances were determined as described before (Valgepea et al., 2017a). Briefly, carbon balancing of substrate carbon (CO, cysteine) between growth products (acetate, ethanol, 2,3-BDO, CO₂, biomass) was calculated as C-mol fraction of each product from total C-mol carbon uptake. Carbon recoveries were calculated as the fraction of summed C-mol products from total C-mol substrates. Ethanol stripping and the total soluble CO₂ fraction in culture broth were also taken into account to achieve more accurate carbon balancing.

Transcriptome analysis of 20 chemostat cultures was conducted using RNA sequencing (two and one bio-replicate of syngas cultures at ∼1 and ∼2.8 day⁻¹, respectively, indicated in Table 1 were excluded due to experimental constraints). Ten millilitres of culture were pelleted by immediate centrifugation (5,000 × g for 3 min at 4°C) and resuspended in 5 mL of RNAlater (76106; Qiagen). Samples were stored at 4°C overnight, centrifuged (4,000 × g for 10 min at 4°C), and pellets stored at −80°C until RNA extraction.

Thawed cell pellets were resuspended in 800 µL of RLT buffer (74104; Qiagen) containing 10 µL of β-mercaptoethanol and lysed with acid-washed glass beads (G4649; Merck) using the Precellys^® 24 instrument with liquid nitrogen cooling (Bertin Technologies). Total RNA was extracted using the RNeasy mini kit (74104, Qiagen) with off-column TURBO™ DNase treatment (AM2239; Invitrogen), followed by purification and enrichment using the RNA Clean and Concentrator™ kit (R1018, Zymo). The efficiency of the total RNA purification and DNA removal was verified using the NanoDrop™ 1,000 instrument (Thermo Scientific) and the quality of RNA extracts was checked using the TapeStation 2200 equipment (Agilent Technologies). Total RNA concentration was determined using the Qubit 2.0 instrument (Q32866; Invitrogen). Next, ribosomal RNA (rRNA) was removed using the QIAseq FastSelect –5S/16S/23S Kit (335925; Qiagen) and stranded mRNA libraries were prepared using the QIAseq Stranded RNA Lib Kit (180743; Qiagen). RNA sequencing was performed using the NextSeq MID150 sequencing kit (20024904; Illumina) on the NextSeq500 sequencer (Ilumina) with 2 × 75 bp paired-end dual indexed (2 × 8 bp) reads, which produced eight fastq files per sample (160 files in total).

The R-scripts used for the analysis of RNA sequencing data after read trimming includes complete details of the methodology and can be downloaded as Supplementary Files S1, S2.

Metabolic model simulations were performed using the genome-scale model iCLAU786 of C. autoethanogenum (Valgepea et al., 2018) and flux balance analysis (FBA) (Orth et al., 2010) as specified in Valgepea et al. (2018). We used the COBRA Toolbox (Schellenberger et al., 2011) as the programming platform for FBA with Gurobi (Gurobi Optimization Inc.) as the linear programming solver. We note three characteristics of our model network: a) methylene-THF reductase is NADH-specific and reducing ferredoxin (Fd); b) membrane-bound Fd-NAD⁺ oxidoreductase Rnf complex translocates 2H⁺/reduced Fd and c) 4H⁺/ATP stoichiometry for the ATP synthase. We estimated intracellular metabolic flux rates (SIM1‒23 in Supplementary Tables S8, S9) using experimentally measured constraints [e.g., μ, substrate consumption and product production rates (CO, H₂, CO₂, cysteine, acetate, ethanol, and 2,3-BDO)] and maximisation of ATP dissipation as the objective function in FBA. Prediction of optimal growth phenotypes for μ and products (SIM24‒46) was performed using experimental substrate uptake rates (CO, cysteine, and H₂ for syngas), the ATP dissipation flux calculated above, and maximisation of biomass yield as the objective function. Accuracy of growth phenotype prediction was improved (SIM47‒59) by additionally zeroing CO₂ reduction with the redox-consuming FdhA activity (reaction rxn00103_c0) and fixing the ratio between H₂ utilisation for direct CO₂ reduction (reaction rxn08518_c0), and Fd_red and NADPH generation (reaction leq000001) by the HytA-E/FdhA complex at a value corresponding to the respective experiment’s q_H2/q_CO ratio (see Valgepea et al., 2018 for details). Simulation results identified as SIMx (e.g., SIM1) in the text are reported in Supplementary Tables S8, S9, while the SBML model file of iCLAU786 is supplied in Valgepea et al. (2018).

We used continuous cultures for controlling the µ of cells as this allows to unequivocally quantify the effects of µ on cell growth compared to batch cultures where genetically modified strains (e.g., titratable substrate uptake) or variable culture parameters (e.g., growth media) need to be used to adjust µ of cells that might confound results (Adamberg et al., 2015). Here, the acetogen Clostridium autoethanogenum was grown in 23 chemostats where cells reached steady-states on CO or syngas (CO + CO₂+H₂) using a chemically defined medium in biological triplicates or quadruplicates at dilution rates ∼1.0, ∼2.0, and ∼2.8 day⁻¹ (µ∼0.04, 0.08, and 0.12 h⁻¹, respectively) (Table 1). In addition to continuous quantification of gas uptake and production rates, cultures were sampled for extracellular metabolome and transcriptome analysis. We also used genome-scale metabolic modelling to estimate intracellular flux rates.

We adjusted gas-liquid mass transfer conditions (gas flow rate and agitation) between different dilution rates to maintain similar steady-state biomass concentrations (∼1.6 ± 0.1 gDCW/L; average ± standard deviation) (Table 1; Figure 1A) as it can affect acetogen carbon distribution (Valgepea et al., 2017a). We detected secretion of acetate, ethanol, and 2,3-BDO by the cells (Figure 1A). While specific acetate production rate (q_ace; mmol/gDCW/h) profiles with increasing µ varied between the two gases, specific ethanol production rates (q_EtOH) more than doubled on both gases (Figure 1B). This also means higher productivity (g/L/h) due to similar biomass levels. Increasing trends for specific 2,3-BDO production rates (q_2,3-BDO) were also seen with higher µ. Notably, faster growth led to molar acetate/ethanol ratios dropping from ∼1.5 to 0.9 and 0.6 for CO and syngas, respectively, due to a strong decrease of acetate levels at similar ethanol values (Figure 1A). These trends are beneficial for an industrial ethanol production process as higher productivity is not comprised with lower selectivity.

CO limited our chemostats since we could maintain similar steady-state biomass concentrations across dilution rates by changing gas-liquid mass transfer (Table 1). One could thus expect the specific CO uptake rate (q_CO; mmol/gDCW/h) to increase proportionally with µ. We detected q_CO to increase ∼2.2- (32 ± 1 to 70 ± 1) and ∼2.9-fold (25 ± 1 to 73 ± 2) for CO and syngas cultures, respectively, with 2.8-fold faster growth (Figure 1C). The non-proportional change for CO cultures means a higher biomass yield (gDCW/mmol of CO consumed) at higher µ. As expected for CO-limited chemostats (Richter et al., 2013; Martin et al., 2015; Valgepea et al., 2018), simultaneous uptake of CO and H₂ were observed for syngas cultures (Figure 1C). Interestingly, while the specific H₂ uptake rate (q_H2; mmol/gDCW/h) increased ∼2.9-fold (7 ± 1 to 21 ± 1) between µ = 1.0 and 2.0 day⁻¹ on syngas, no further change was seen with faster growth, despite gas-liquid mass transfer (gas flow rate and agitation) being increased with dilution rate. Specific CO₂ production rates (q_CO2; mmol/gDCW/h) increased by ∼1.8- (24 ± 1 to 43 ± 1) and ∼3.4-fold (10 ± 2 to 34 ± 1) for the CO and syngas cultures, respectively (Figure 1C), indicating a strong effect of H₂ uptake on CO oxidation as seen before for C. autoethanogenum growing on various gas mixes (Valgepea et al., 2018).

Product ratios show the distribution of carbon between products but carbon balancing is necessary to quantify carbon flows from substrates to products (i.e. yields). Notably, carbon flux to acetate dropped with faster growth by 24% (19.2 ± 2.2% to 14.6 ± 1.3%) on CO and by 57% (32.3 ± 2.5% to 13.9 ± 1.1%) on syngas (Figure 2). While no clear µ-dependent trend for carbon flow to 2,3-BDO was detected on CO, cells diverted up to 141% more carbon into 2,3-BDO in syngas cultures (1.5 ± 0.3% to 3.6 ± 0.4%). The aforementioned doubling of q_EtOH with faster growth (Figure 1B) was supported by up to ∼21% elevated carbon flux to ethanol (Figure 2). These observations are not trivial as faster growth demands more energy while reduced products like ethanol and 2,3-BDO consume redox co-factors that acetogens could otherwise use for ATP generation (Schuchmann and Müller, 2014). In fact, cells were even able to channel more carbon into biomass formation simultaneously (increase of ∼48 and ∼24% for CO and syngas, respectively; Figure 2). The ∼60 and ∼50% losses of substrate carbon as CO₂ for the CO and syngas cultures, respectively, are close to theoretical stoichiometric and thermodynamic calculations for CO₂ dissipation with ethanol production from CO (∼67%) or from a gas mix with a CO-to-H₂ ratio around two (50%) (Wilkins and Atiyeh, 2011; Molitor et al., 2016).

We took advantage of the quantitative steady-state data collected above on carbon and redox flows entering and leaving the cells to estimate intracellular metabolic fluxes using the genome-scale metabolic model (GEM) iCLAU786 of C. autoethanogenum (Valgepea et al., 2018) and flux balance analysis (FBA) (Orth et al., 2010). Steady-state intracellular flux patterns for our CO and syngas chemostats were estimated by constraining the GEM with experimental data (exchange rates and μ) and maximising ATP dissipation as the objective function in FBA (SIM1‒23 in Supplementary Tables S8, S9).

Flux through the WLP was significantly increased as expected with faster growth, although less than the increase in CO uptake for syngas cultures as more of the fixed CO was oxidised and dissipated as CO₂ (Figure 3; Supplementary Tables S8, S9). Faster growth was also supported with elevated ATP production through the ATPase (by 82 and 125% on CO and syngas, respectively; Figure 3; Supplementary Tables S8, S9). Interestingly, the ratio between the ATP production fluxes of the ATPase and the acetate kinase (acetyl-P → acetate) changed minimally (9% in average; Supplementary Tables S8), despite significantly altered product distributions (Figure 2). At the same time, maintenance ATP costs increased by 80 and 66% to 9.9 ± 0.5 and 11.7 ± 1.9 mmol/gDCW/h on CO and syngas, respectively (Figure 3; Supplementary Tables S8, S9). The average ∼30% fraction of maintenance costs from total ATP production (Supplementary Table S8) is similar to previously studied autotrophic C. autoethanogenum cultures (Valgepea et al., 2017a, 2018). In addition to ATP, biomass synthesis during faster growth also demands an elevated supply of NADPH. Notably for C. autoethanogenum, extra NADPH was provided by increased flux through the Nfn transhydrogenase (by 66 and 190% to 12.0 ± 0.2 and 8.8 ± 0.5 mmol/gDCW/h on CO and syngas, respectively), and not by higher flux through the electron-bifurcating hydrogenase HytA-E complex on syngas (Figure 3). No increase in flux through HytA-E with faster growth on syngas suggested that production of the critical reducing power co-factor, reduced ferredoxin (Fd_red), was possibly supported by CO oxidation. Indeed, the model demonstrated an increase of the fraction of Fd_red generated by CO oxidation from 63 ± 2% to 74 ± 1% with increasing μ (Supplementary Table S8). As seen for previous CO + H₂ or CO₂+H₂-fermenting C. autoethanogenum steady-state cultures (Valgepea et al., 2017a, 2018; Heffernan et al., 2020), all the CO₂ in the WLP was reduced to formate directly using H₂ through the formate-H₂ lyase activity of the HytA-E/formate dehydrogenase (FdhA) complex (Wang et al., 2013). This saves valuable redox for the cells compared to growth on CO only and allows higher production of reduced products when H₂ is available (Figure 2). Regarding ethanol production, our simulations are consistent with the aforementioned C. autoethanogenum datasets and gene knockout experiments (Liew et al., 2017) showing that ethanol was synthesised using the aldehyde:Fd oxidoreductase (AOR) activity through acetate to couple it with ATP production (Figure 3).

GEMs can also be used for predicting growth phenotypes (O’Brien et al., 2015) and the accuracy of the model can be evaluated beforehand, for example, by comparing experimental product profiles with predicted ones if constraining only substrate uptake rates, maintenance ATP costs, and maximising biomass yield in FBA. We initially failed to predict ethanol and 2,3-BDO production (SIM24‒46 in Supplementary Tables S8, S9), as seen before (Valgepea et al., 2017a, 2018). However, when additionally constraining the model by coupling carbon and redox metabolism from H₂ utilisation for syngas cultures (see Materials and Methods), predictions improved as ethanol production became on average 28% off from experimental values (SIM47‒59 in Supplementary Tables S8, S9). Prediction of acetogen phenotypes can be further improved by considering thermodynamics and kinetics (Greene et al., 2019; Mahamkali et al., 2020), which are important for accounting for the effects from high extracellular product levels (e.g. acetate, ethanol).

Next, we performed transcriptome analysis using RNA sequencing (RNA-seq) to quantify μ-dependent gene expression changes on both gas mixes. High reproducibility of the data was demonstrated by clear clustering of bio-replicates (Figures 4A,B) and an average Pearson correlation coefficient of R = 0.96 between bio-replicates across μ values (Figure 4C). Also, significant μ-dependent expression differences for many genes could be seen for both gases (Figure 4B). Indeed, we determined 1,337 differentially expressed genes (DEGs) in total with a fold-change (FC) > 1.5 (up- or down-regulation) and a q-value < 0.05 after FDR correction in at least one pairwise comparison within the three pairwise μ-dependent comparisons (μ = 2.0 vs. 1.0, 2.8 vs. 1.0, and 2.8 vs. 2.0 day⁻¹) of both gas mixes (Supplementary Tables S4, S5). Expectedly, the bigger the difference between compared μ values, the higher the number of DEGs (Figure 4D). Also, many DEGs were shared between comparisons within both gas mixes. Interestingly, while 442 DEGs were shared between CO and syngas comparisons of μ = 2.8 vs. 1.0 day⁻¹, hundreds of unique DEGs were also detected, demonstrating the effect of the feed gas mix on μ-dependent gene expression patterns.

To understand global transcriptome changes that facilitate faster growth, we assessed DEGs through functional gene classifications. Firstly, analysis of DEGs of μ = 2.8 vs. 1.0 day⁻¹ using Cluster of Orthologous Groups (COG) classification (Tatusov et al., 2003) showed that most DEGs were classified as “Function unknown” (S) (Figure 5A; Supplementary Table S6). This highlights the need for genotype-phenotype mapping and improving gene annotations for acetogens. Genes involved in energy, amino acid, and nucleotide metabolism (C, E, and F) and replication, transcription, and translation (J, K, and L) were also abundant among DEGs and all these processes are vital for faster growth. Secondly, we used Gene Ontology (GO) terms (Ashburner et al., 2000) to attain finer resolution behind global gene expression trends. Clustering of all 1,337 DEGs across μ comparisons and gas mixes resulted in six significant expression clusters (Figure 5B). GO enrichment analysis (q-Fisher < 0.05) of the genes within the identified clusters was consistent with COG analysis as several GO terms related to translation (0003735, 0019843, 0005840, and 0044391) and metabolism (0006096, 0016051, 0043603, 0009150, 00505996, and 0006790) were enriched in cluster 3 that includes up-regulated genes with faster growth. Enrichment of GO terms within clusters 2 and 5 was detected, including genes down-regulated with faster growth. Lastly, we aimed to identify genes and GO terms whose expression was tightly controlled with changing μ, indicated by the same expression trend on both gas mixes. We identified 245 such genes that showed either continuously increasing or decreasing differential expression (DEGs, i.e., FC > 1.5 and q < 0.05) with increasing μ on both gases with the same magnitude of expression change between the two gas mixes (Figure 5C). Enrichment analysis of these genes confirmed the aforementioned results by identifying enrichment of genes (q-Fisher < 0.05) associated with translation (0003735 and 0019843), but also of the GO term 0032787 “monocarboxylic acid metabolic process” (coloured lines on Figure 5C).

Previous transcriptomics and proteomics studies have suggested that autotrophic metabolism of acetogens is not controlled through hierarchical regulation of gene expression (Richter et al., 2016; Valgepea et al., 2017a, 2018; Al-bassam et al., 2018; Mahamkali et al., 2020). We thus analysed DEGs in terms of individual genes linked to activities specifically relevant to acetogens or with high expression changes. Surprisingly, we found numerous DEGs in central metabolism, some of which can explain the metabolic rearrangements described above (Figure 6; Supplementary Figures S1, S2). For instance, increased carbon flux to 2,3-BDO with faster growth on syngas (Figure 2) was supported by 18.5-fold (q < 0.01) (all μ = 2.8 vs. 1.0 day⁻¹ unless otherwise noted) up-regulation of the 2,3-BDO dehydrogenase (CAETHG_RS01830; BDH) that reduces acetoin to 2,3-BDO and by 2.3-fold (q < 0.01) up-regulation of the pyruvate:Fd (flavodoxin) oxidoreductase (RS14890; PFOR) that converts acetyl-CoA to pyruvate (Figure 6). At the same time, elevated production of ethanol (Figure 2) could be linked to increased expression (2.1- to 2.8-fold, q < 0.01) of several alcohol dehydrogenases (RS02620 and RS02630; Adh and Adh3), including the most abundant–RS08920; Adh4—in C. autoethanogenum (Supplementary Table S4) (Valgepea et al., 2017a, 2021) on syngas (Figure 6). Acetaldehyde for the alcohol dehydrogenases was likely not supplied directly from acetyl-CoA as we quantified strong repression of two putative acetaldehyde dehydrogenases (RS08810 and RS08865; 11.5- and 17.0-fold, q < 0.01), which interestingly seem to be the result of the uniformly strong repression of a cluster of 21 genes (RS08795‒08895) linked to bacterial microcompartments (BMCs) (Supplementary Figure S1). Despite the 4.8-fold (q < 0.01) upregulation of one of the bifunctional aldehyde/alcohol dehydrogenases (RS18395; AdhE1) which catalyses ethanol production directly from acetyl-CoA, ethanol was still most likely produced through acetate using the AOR activity as AOR1 transcripts were ∼160-fold more abundant (RPKM) compared to AdhE1 (Supplementary Tables S2, S3) across all experiments and AOR2 (RS00490) was up-regulated 2.5-fold (q < 0.01) (Supplementary Tables S4, S5). This is consistent with metabolic flux data (Figure 3) and previous–omics and knockout experiments in acetogens (see above).

Faster growth demands more energy and this was supported by increased expression (1.5- to 2.2-fold, q < 0.01) of genes of the multi-subunit Fd-NAD⁺ oxidoreductase Rnf complex (RS15845-65) across the studied range of µ (Figure 6), as the Rnf complex generates the proton motive force to drive the ATPase in C. autoethanogenum (Hess et al., 2016; Tremblay et al., 2012). Notably, up-regulation of NAD(P)- (RS02275) and FAD-dependent (RS04920) oxidoreductases was also detected (7.9- to 32.9-fold and 2.0- to 3.4-fold, respectively, q < 0.01) on both gases that might play a role in the maintenance of redox homeostasis (Supplementary Figure S2). Intriguingly, despite the elevated H₂ uptake with faster growth on syngas (Figure 1C), ∼1.5-fold (q < 0.01) repression of the main H₂ oxidiser–the HytA-E complex (RS13745‒70)—(Wang et al., 2013; Mock et al., 2015; Valgepea et al., 2021) components was seen (Figure 6). Moreover, similar down-regulation was observed in CO cultures, potentially indicating µ-dependent regulation of the HytA-E complex. Repression of HytA-E also means that gene expression levels at μ = 1.0 day⁻¹ were sufficient to realise higher H₂ uptake with faster growth, which was also the case for higher CO fixation as WLP genes were not detected as DEGs (Supplementary Tables S4, S5). In contrast, increased flux throughput for biomass synthesis with faster growth was supported by elevated expression of gluconeogenesis genes (RS08495‒08515, RS12015, and RS12020) (Figure 6B). In the future, further work using genetically modified strains is needed to determine whether any of the putative transcription factors that were up-regulated with faster growth (Supplementary Figure S2) are responsible for increased gene expression in any of the cases above.