Chemostat cultures were grown in 1.2 L laboratory bioreactors (Applikon, Delft, The Netherlands), which were stirred at 300 rpm and kept at 30°C. Anaerobic conditions were maintained by flushing the headspace with nitrogen gas, at a flow rate of 120 mL min^–1. Culture pH was controlled at 8 ± 0.1 by automatic titration (ADI 1030 Biocontroller, Applikon, Delft, The Netherlands) of 1 M NaOH. The dilution rate was 0.09 ± 0.01 h^–1 and the working volume of 0.5 L was kept constant by peristaltic effluent pumps (Masterflex, Cole-Parmer, Vernon Hills, IL, United States) coupled to electrical level sensors. Bioreactors were inoculated (10% v/v) with 50 mL samples of D-galacturonate-limited, anaerobic chemostat enrichment cultures (Valk et al., 2018), stored in 30% v/v glycerol at –20°C. Cultures were run in continuous mode and after at least 6 days (18 generations) stable product composition and biomass concentration were established. System stability was assessed by online monitoring of CO₂ production and offline monitoring of fermentation products and optical density. When measurements varied by less than 10% over multiple volume changes, without a clear upward or downward trend, samples were taken during subsequent cycles.

The cultivation medium contained (g L^–1): D-galacturonate 4.3; NH₄Cl 1.34; KH₂PO₄ 0.78; Na₂SO₄.10H₂O 0.130; MgCl₂.6H₂O 0.120; FeSO₄.7H₂O 0.0031; CaCl₂ 0.0006; H₃BO₄ 0.0001; Na₂MoO₄. 2H₂O 0.0001; ZnSO₄.7H2O 0.0032; CoCl₂.H₂O 0.0006; CuCl₂.2H₂O 0.0022; MnCl₂.4H₂O 0.0025; NiCl₂.6H₂O 0.0005; EDTA 0.10. Nineteen liter of mineral solution (mineral concentration adjusted to the final volume, 20 L) was autoclaved for 20 min at 121°C after which 1 L (86 g L^–1) D-galacturonate solution was filter sterilized (0.2 μm Mediakap Plus, Spectrum Laboratories, Rancho Dominguez, CA, United States) into the media. 1.5 mL Pluronic PE 6100 antifoam (BASF, Ludwigshafen, Germany) was added per 20 L of mineral solution to avoid excessive foaming.

To determine substrate and extracellular metabolite concentration, reactor sample supernatant was obtained by centrifugation of culture samples (Heraeus Pico Microfuge, Thermo Fisher Scientific, Waltham, MA, United States). Concentrations of D-galacturonate and extracellular metabolites were analyzed with an Agilent 1100 Affinity HPLC (Agilent Technologies, Amstelveen, The Netherlands) equipped with an Aminex HPX-87H ion-exchange column (BioRad, Hercules, CA, United States) operated at 60°C with a mobile phase of 5 mM H₂SO₄ and a flow rate of 0.6 mL min^–1. CO₂ and H₂ concentrations in the bioreactor exhaust gas were measured using a Prima BT Bench Top mass spectrometer (Thermo Fisher Scientific, Waltham, MA, United States) after the gas was cooled by a condenser (4°C).

Twenty milliliter of culture broth samples were filtered over pre-dried and pre-weighed membrane filters (0.2 μm Supor-200, Pall Corporation, New York, NY, United States), which were then washed with demineralized water, dried in a microwave oven (Robert Bosch GmbH, Gerlingen, Germany) for 20 min at 360 W and reweighed. Carbon and electron balances were constructed based on the number of carbon atoms and electrons per mole, while biomass composition was assumed to be CH_1.8O_0.5N_0.2 (Roels, 1983).

Fluorescent in situ hybridization was performed as described previously (Daims et al., 2005), using a hybridization buffer containing 35% (v/v) formamide. Probes were synthesized and 5′ labeled with either 5(6)-carboxyfluorescein-N-hydroxysuccinimide ester (FLUOS) or with one of the sulfoindocyanine dyes (Cy3 and Cy5; Thermo Hybaid Interactiva, Ulm, Germany) (Table 1). The general probe EUB338mix, labeled at both 3′ and 5′ ends with Cy5, was used to identify all eubacteria in the sample. Microscopic analysis was performed with a LSM510 Meta laser scanning confocal microscope (Carl Zeiss, Oberkochen, Germany). The qFISH analysis was based on at least 29 fields of view at 6730 × magnification, using DAIME (version 2.1) software (DOME, Vienna, Austria; Daims et al., 2006). The bio-volume fractions of “Ca. G. soehngenii” and Enterobacteriaceae populations were calculated as the ratio of the area hybridizing with specific probes relative to the total area hybridizing with the universal EUBmix probe set (Amann et al., 1990; Daims et al., 1999).

A 1 M NaH¹³CO₃ solution was used to replace the regular 1 M NaOH solution as a pH titrant in steady-state D-galacturonate-limited enrichment cultures (pH 7.8 ± 0.1, D = 0.1 h^–1, T = 30°C). Broth was collected on ice every 2 h for 8 consecutive hours and centrifuged (12,000 × g, Heraeus Pico Microfuge, Thermo Fisher Scientific, Waltham, MA, United States) before the supernatant was collected and stored at –20°C until analysis by NMR. CO₂, H₂ and ¹³CO₂ concentrations in the exhaust gas were measured by MS (Prima BT Bench Top MS, Thermo Fisher Scientific, Waltham, MA, United States) after the gas had been cooled by a condenser (4°C).

The metagenomic-assembled genome of “Candidatus Galacturonibacter soehngenii” described by Valk et al. (2018) was used as template for preparing the metagenome libraries. The DNA extraction, Illumina sequencing, metagenomic assembly and binning process is described in Valk et al. (2018). Long-read genomic DNA sequencing was conducted using 1D nanopore sequencing (Oxford Nanopore Technologies, Oxford, United Kingdom), following the manufacturer’s protocol (LSK-108), omitting the optional DNA shearing and DNA repair steps. The library was loaded on a flow cell (FLO-MIN106) and the MinION Mk1B DNA sequencer (Oxford Nanopore Technologies, Oxford, United Kingdom) was used for sequencing combined with the MinKNOW v. 1.7.3 (Oxford Nanopore Technologies, Oxford, United Kingdom) software with the 48 h sequencing workflow (NC_48h_Sequencing_Run_FLO_MIN106_SQK-LSK108.py). Albacore v. 1.2.1 (Oxford Nanopore Technologies, Oxford, United Kingdom) was used to base-call the sequencing reads.

The assembling of the contigs from the “Candidatus Galacturonibacter soehngenii” genome bin into a single scaffold based on the long Nanopore reads was done using SSPACE-LongRead scaffolder v. 1.1 (Boetzer and Pirovano, 2014). GapFiller v. 1.11 (Boetzer and Pirovano, 2012) or by manual read mapping and extension in CLC Genomics Workbench v. 9.5.2 (Qiagen, Hilden, Germany) were used to close gaps in the draft genome with the previously assembled Illumina data. Finally, manual polishing of the complete genome was done to remove SNPs and ensure a high-quality assembly. The meta-genome has been submitted to the sequence read archive (SRA)¹ with accession number SRR10674409, under the BioProject ID PRJNA566068.

The metagenome-assembled genome was uploaded to the automated Microscope platform (Vallenet et al., 2006, 2017). Manual assessment of pathway annotations was assisted by the MicroCyc (Caspi et al., 2008), KEGG (Kyoto Encyclopedia of Genes and Genomes; Kanehisa et al., 2014) and SwissProt alignment (BLASTP version 2.2.28+; Altschul et al., 1997) databases. The predicted proteome of “Ca. G. soehngenii” was submitted to InterProScan (version 5.25-64.0), to identify predictive Pfam domains (El-Gebali et al., 2018). The annotated genome sequence of “Candidatus Galacturonibacter soehngenii” has been submitted to the European Nucleotide Archive (ENA) under the BioProject ID PRJNA566068.

During pseudo-steady state, broth samples were taken from the enrichment culture, directly frozen in liquid nitrogen and subsequently stored at −80°C. Five hundred microliter samples were thawed on ice, pelleted by centrifugation (21,000 × g, 2 min, 4°C) and used for total RNA extraction with the RNeasy PowerMicrobiome Kit (Qiagen, Hilden, Germany), following the manufacturer’s instruction with the addition of phenol:chloroform:isoamy alcohol (25:25:1) and β-mercaptoethanol (10 μL mL^–1 final concentration). Cell lysis was with a FastPrep-24 bead beater (MP Biomedicals, Fisher Scientific, Hampton, VA, United States, four successive cycles of 40 s at 6.0 m s^–1, 2 min incubation on ice between cycles). Total RNA extracts were subjected to DNase treatment to remove DNA contaminants by using the DNase Max Kit (Qiagen, Hilden, Germany) and further cleaned up and concentrated with the Agencourt AMpure XP magnetic beads (Beckman Coulter, Brea, CA, United States) before rRNA depletion. Integrity and quality of purified total RNA were assessed on a Tapestation 2200 (Agilent, Santa Clara, CA, United States) with the Agilent RNA screen-tapes (Agilent, Santa Clara, CA, United States) and the concentration was measured using Qubit RNA HS Assay Kit (Thermo Scientific Fisher, Waltham, MA, United States).

Five hundred nanogram of total RNA from each sample was obtained after rRNA was depleted using the Ribo-Zero rRNA Removal (Bacteria) Kit (Illumina, San Diego, CA, United States), with 2 μg total RNA as input. Quality of extracted mRNA was checked with Agilent RNA HS screen-tapes (Agilent, Santa Clara, CA, United States) and RNA concentration was determined with a Qubit RNA HS Assay Kit (Thermo Scientific Fisher, Waltham, MA, United States). The TruSeq Stranded mRNA Sample Preparation Kit (Illumina, San Diego, CA, United States) was used to prepare cDNA sequencing libraries according to the manufacturer’s instruction. Libraries were sequenced on an Illumina HiSeq2500 using the TruSeq PE Cluster Kit v3-cBot-HS and TruSeq SBS kit v.3-HS sequencing kit (1 × 50 bp; Illumina, San Diego, CA, United States). The raw meta-transcriptome reads have been submitted to the sequence read archive (SRA)¹ with accession number SRR10674118-23, under the BioProject ID PRJNA566068.

Raw RNA reads in FASTQ format were imported into CLC Genomics Workbench v. 9.5.5 and trimmed for quality, requiring a minimum phred score of 20 and a read length of 45. Reads from each sample were hereafter mapped to CDSs obtained from the MAG of “Ca. G. soehngenii” with a minimum similarity of 98% over 80% of the read length. Reads per kilobase of transcript per million mapped reads (RPKM) were calculated based on raw read-counts and the length of each CDS. The meta-transcriptome mapped to the genome of “Ca. G. soehngenii” are shown in Supplementary Data Sheet S2.

Gene F7O84_RS11645 was codon optimized for expression in Escherichia coli with the GeneArt online tool and integrated behind the TEV recognition site of the pET151/D-TOPO expression vector by GeneArt (GeneArt GmbH, Regensburg, Germany). The resulting plasmid was transformed into a chemically competent E. coli strain BL21 according to manufacturer’s instructions (NEBuilder HiFi DNA Assembly Master Mix chemical transformation protocol (E2621), New England Biolabs, Ipswich, MA, United States) and named pUD1074. The plasmid sequence of pUD1074 has been deposited at the NCBI GenBank² with the corresponding accession number MN498128.

All E. coli cultures were performed in 120 mL capped bottles with 50 mL of mineral medium (Diender et al., 2016). Prior to inoculation, the bottles were autoclaved at 120°C after which the mineral media was supplemented with autoclaved (120°C, 20 min); glucose 5 g L^–1, peptone (BD Bacto Difco, Thermo Fisher Scientific, Waltham, MA, United States) 1 g L^–1, yeast extract (BD Bacto Difco, Thermo Fisher Scientific, Waltham, MA, United States) 2 g L^–1 and cysteine 1 g L^–1. Additionally, 0.05 g L^–1 ampicillin was added and the gas phase was exchanged with air, with a final pressure of 170 kPa. All E. coli cultures used for measurements were inoculated with overnight grown pre-cultures (1:50 v/v) and incubated at 37°C and shaken (300 rpm) until oxygen was depleted (2–3 h). Subsequently 1 mL (250 g L^–1) glucose, 1 mL reducing agent (0.4 M cysteine) and 1 mL IPTG (40 mM) were added.

After 3 h (at 30°C, unshaken) of incubation, the cells were harvested and processed anaerobically according to Diender et al. (2016). Enzymatic activity analysis was conducted using a modified method initially described by Diender et al. (2016). The essays were performed in an anaerobic environment using 100–300 μL of cell extract with both CO and hydroxylamine as substrate. To increase metal cofactor availability, 1:200 (v/v) metals solution was added to the assay buffer which contained in (g L^–1); HCl 1.8, H₃BO₃ 0.0618, MnCl₂ 0.06125, FeCl₂ 0.9435, CoC_l2 0.0645, NiCl₂ 0.01286, ZnCl₂ 0.0677, CuCl₂ 0.01335.

The sequence of the putative CODH (F7O84_RS11645) was blasted with the BLASTp (version 2.2.28+; Altschul et al., 1997) tool of the JGI-IMG/M database (Markowitz et al., 2012), with default parameter settings. Finished genomes from members of the Lachnospiraceae family in the public JGI-IMG/M database (Markowitz et al., 2012) were selected for analysis, Supplementary Table S4. The stains identified in the BLAST search, or closely related strains (Supplementary Table S5) were subsequently analyzed in KEGG (Kanehisa et al., 2014) for presences of the CODH/ACS complex with pathway map 1200.

Anaerobic, galacturonate-limited chemostat enrichment cultures were used to study the physiology of “Ca. G. soehngenii” cultures. In a previous study (Valk et al., 2018), the relative abundance of “Ca. G. soehngenii” in such cultures did not exceed 65%, based on metagenomic analysis, and formate and H₂ were detected in the liquid and gas phases, respectively. It was hypothesized that, in these experiments, a low in situ hydrogen partial pressure limited in vivo WLP activity, as it was expected that hydrogen was used as reductant for the production of acetate from formate or CO₂. To investigate this possibility, head space flushing instead of sparging was applied, using N₂ gas. This caused an increase in the hydrogen partial pressure in the media broth (De Kok et al., 2013). Additionally, the dilution rate was decreased from 0.125 to 0.1 h^–1. Analysis of the abundance of “Ca. G. soehngenii” in the resulting enrichment cultures by quantitative fluorescence in situ hybridization (qFISH) indicated that 86.5 ± 2.6% of the bio-volume of qFISH-detectable cells consisted of “Ca. G. soehngenii.” The major side population Enterobacteriaceae represented 13.8 ± 2.4% of the bio-volume. As these two subpopulations together accounted for 100.2 ± 5.0% of the bio-volume, it was assumed that any other, minor, subpopulations did not significantly influence the stoichiometry of catabolic fluxes.

Product yields and biomass-specific conversion rates of the D-galacturonate-limited anaerobic enrichment cultures dominated by “Ca. G. soehngenii” (Table 2) showed acetate as dominant catabolic product (0.57 ± 0.03 Cmol (Cmol galacturonate^–1). Carbon and electron recoveries were 94 and 92%, respectively, indicating that all major fermentation products were identified. As observed previously (Valk et al., 2018), this acetate yield on galacturonic acid was significantly higher than the combined yields of formate and hydrogen. This difference was interpreted as indicative for acetogenesis by one of the dominant organisms, of which only the “Ca. G. soehngenii” MAG was shown to harbor homologs for most WLP structural genes (Ragsdale and Pierce, 2008; Valk et al., 2018). Yields of hydrogen and formate on galacturonate (0.02 ± 0.01 mol Cmol galacturonate^–1) and 0.02 ± 0.01 (Cmol galacturonate^–1), respectively were significantly lower than found in a previous study on “Ca. G. soehngenii” (Valk et al., 2018). This observation is consistent with a higher in vivo contribution of the WLP as a result of a higher hydrogen partial pressure and/or lower specific growth rate in the present study.

A simple model was constructed to predict formation of labeled acetate, using biomass-specific conversion rates measured in pseudo-steady state enrichment cultures as inputs (Supplementary Calculations S1, S2 and Supplementary Figure S1). Model simulations predicted that, after 8 h, approximately 15% of the acetate produced by the enrichment culture should be labeled. To investigate if CO₂ was indeed incorporated into acetate via acetogenic fermentation, ¹³C-labeled bicarbonate was fed to a “Ca. G. soehngenii” enrichment chemostat culture. However, after 8 h, the fraction of ¹³C in the methyl group of acetate increased to 2.0%. This increase represented only a small increase relative to the 1% natural abundance of ¹³C (Table 3; Rumble et al., 2017). In contrast, after 8 h of ¹³C-bicarbonate feeding, the enrichment culture showed a 21.5% abundance of ¹³C in the carbonyl-group of acetate (Table 3).

In the WLP, ¹³C-labeled CO₂ incorporation into the carbonyl-group of acetate involves activity of CO dehydrogenase (COOS, EC 1.2.7.4). To investigate the presence of this key enzyme in “Ca. G. soehngenii,” an anaerobic enzyme activity assay was performed on cell extracts of enrichment cultures, using CO as electron donor and methyl viologen (MV) as electron acceptor (Diender et al., 2016). These assays revealed a CO dehydrogenase activity of 2.1 ± 0.6 μmol min^–1 (mg protein) ^–1. Reduction of MV in the absence of either CO or cell extract was below detection limit [<0.05 μmol min^–1 (mg protein) ^–1].

Previous analysis of the “Ca. G. soehngenii” MAG (Valk et al., 2018) was based on an assembly made with short-read DNA sequencing data. To identify if putative CODH/ACS complex genes had been missed in this analysis due to incomplete assembly, long-read Oxford Nanopore sequencing (Deamer et al., 2016; Jain et al., 2016) was used to improve the previously assembled “Ca. G. soehngenii” MAG. The resulting genome assembly consisted of 8 contigs and was estimated to have a 98% completeness and contained no genetic contamination with sequences from other organisms according to checkM (Table 4). As in the previous study, homologs were detected for most structural genes associated with the WLP (Table 5), but none of the annotated genes in the predicted proteome showed homology with known CODH/ACS genes (Vallenet et al., 2006; Ragsdale, 2008; Valk et al., 2018). A search in the newly assembled “Ca. G. soehngenii” MAG sequence for homologs of signature genes of the six other known pathways for inorganic carbon fixation did not point toward their involvement in carbon metabolism (Supplementary Table S2).

CO dehydrogenases contain highly conserved amino-acid motifs (Pfam or protein-family domains) associated with their nickel-iron-sulfur clusters (Eggen et al., 1991, 1996; Maupin-Furlow and Ferry, 1996; Jeoung and Dobbek, 2011; Techtmann et al., 2012; El-Gebali et al., 2018). The newly assembled “Ca. G. soehngenii” MAG sequence did not reveal hits for the Pfam domain of the CO dehydrogenase α-subunit of the CODH/ACS complex (PF18537) (Darnault et al., 2003). However, two open reading frames F7O84_RS02405 and F7O84_RS11645, harbored the PF03063 Pfam domain, which is associated with the hybrid cluster protein (HCP) and the catalytic center of the Ni-CODH family (van den Berg et al., 2000; Wolfe et al., 2002). Although HCP has been associated with hydroxylamine reductase activity, its catalytic activity has not been experimentally confirmed and, moreover, sequence motifs in HCP showed high similarity with the functional domain of Ni-CODHs making it an interesting candidate genes for the CODH function of the WLP in “Ca. G. soehngenii” (Heo et al., 2002; Wolfe et al., 2002; Aragão et al., 2003; Almeida et al., 2006). A closer inspection of the genetic context of both genes showed many flanking genes encoding hypothetical proteins in their close vicinity, but no genes previously associated with acetogenesis.

A meta-transcriptome analysis of the enrichment cultures showed significant transcript levels of most homologs of known WLP genes, which were approximately 10-fold lower than those of homologs of structural genes encoding Entner-Doudoroff-pathway enzymes involved in galacturonate catabolism (Table 6). A notable exception was the extremely low transcript level of a putative formate dehydrogenase gene (F7O84_RS07405; EC 1.17.1.9). A candidate gene for pyruvate-formate lyase (PFL, EC 6.2.1.3) was highly transcribed (F7O84_03160, Table 6). These observations suggested that formate generated by PFL, rather than CO₂, was the major substrate for the methyl branch of the WLP in “Ca. G. soehngenii.”

Homologs of Rnf cluster (F7O84_03275-3295; EC 7.2.1.2) and hydrogenase (F7O84_0945-50, F7O84_04820; EC 1.12.7.2) genes, which were previously implicated in acetogenesis (Biegel and Müller, 2010; Schuchmann and Müller, 2014, 2016), showed high transcript levels (Table 6). Of the two candidate genes for CO dehydrogenase, F7O84_RS11645 showed the highest transcript level (Table 6). As, under the experimental conditions, no hydroxylamine reductase activity was expected, this result reinforced the candidature of F7O84_RS11645 as possible CO dehydrogenase gene. In an attempt to directly investigate if F7O84_RS11645 encoded a functional CO dehydrogenase, its open reading frame was cloned into high-copy-number E. coli expression vector. However, enzyme assays with cell extracts of the resulting E. coli strain did not yield consistent evidence for either CO dehydrogenase or hydroxylamine dehydrogenase activity (Supplementary Table S3).

A protein BLAST search (Altschul et al., 1997) of the putative CODH (F7O84_RS11645) was done to investigate if presence of the putative CODH gene also coincided with an apparently incomplete WLP in other members of the Lachnospiraceae family. Indeed, 13 sequenced members of the Lachnospiraceae family showed predicted proteins with a high homology with the putative CODH (Supplementary Table S4). 9 of the 13 Lachnospiraceae members were present in the KEGG database (Kanehisa et al., 2014; Supplementary Table S5), and subsequently analyzed on the presence or absence of the CODH/ACS complex. All organisms contained only a partial WLP, with the ACS genes not identified. In seven of the members, respectively Lachnoclostridium saccharolyticum, Lachnoclostridium phytofermentans, Pseudobutyrivibrio xylanivorans, Butyrivibrio fibrisolvens, Pseudobutyrivibrio xylanivorans, and both Roseburia species the full CODH/ACS complex was not identified. Further study is required to elucidate the relevance of the putative CODH for acetogenic metabolism.