Samples from freshly forming iron oxides, pools of water perched on iron duricrust and the gut of termites that build their mounds in and on the iron duricrust in the iron ore regions of Western Australia and Brazil, were collected into sterile containers in the field and stored at 4°C before later cultivation attempts.

The base medium used to enrich iron reducers was prepared anaerobically, and contained per liter: 2.5 g NaHCO₃, 1.5 g NH₄Cl, 0.68 g KH₂PO₄, 0.1 g KCl and was pH adjusted to pH ∼6.8. Iron was usually provided as goethite powder (0.5 g/100 ml or 0.1 g/10 ml) or as crushed canga (1 g/10 ml), added to serum vials or Hungate tubes individually before flushing with nitrogen and adding liquid in an anaerobic chamber of gas phase N₂:H₂ (95:5). Prior to microbial inoculation, electron donors were added from sterile anaerobic stock solutions to final concentrations of 30 mM sodium lactate, 30 mM sodium acetate, and/or 10 mM glucose unless otherwise indicated. Stable enrichment cultures were routinely grown on the base medium with goethite (1.6 g/80 ml; ∼225 mmol/L) and 30 mM glucose in 160 ml serum vials. Culture characterization was performed in triplicate cultures, after the 5th transfer from the original inoculum. Cultures were routinely transferred after plateau of ferrous iron production (any time after 2 weeks since inoculation).

Cultures were monitored for the production of ferrous iron spectrophotometrically using the ferrozine assay modified from Lovley and Phillips (1986). To measure total ferrous iron, a sample of culture (0.1 ml) was mixed with 5 ml of 0.5 N HCl for 15 min, then 0.2 ml of this mixture was added to 2 ml ferrozine (1 g/L in 50 mM HEPES) and filtered through a 0.2-μm pore size filter. To determine soluble ferrous iron, the procedure was the same except that a sample of culture was filtered through a 0.2-μm pore size filter before 0.1 ml of the filtrate was acidified and mixed using ferrozine. The absorbance was measured at 562 nm and compared to known standard concentrations of Fe(II) prepared anaerobically ferrous chloride.

A Shimadzu Prominence HPLC system (Kyoto, Japan) was used to determine the presence of formic, lactic, acetic, butyric, propionic, succinic, isovaleric, and n-valeric acids and glucose in culture supernatant. Separation and identification of acids from 20 μl sample was achieved using an Agilent Hi-Plex H Column (8 μm, 7.7 × 300 mm, Agilent Technologies, Australia) at 35°C using a mobile phase of 14 mM H₂SO₄ at 0.5 ml/min.

For cultures, 0.5 ml of each of three late log phase cultures was centrifuged (13,000 × g × 5 min) for DNA extraction using the DNEasy Powersoil kit (Qiagen, Singapore) as per manufacturer’s instructions. For the water sample from which the cultures originated, 40 ml of water had been collected onto a Sterivex filter in the field and the filter was cut into pieces for DNA extraction using the same kit. The V6–V8 region of the 16S rRNA gene was amplified using primers 926F and 1392r and sequenced on a MiSeq Sequencing System (Illumina, San Diego, CA, United States) [see Gagen et al. (2018) for primer sequences, library preparation, and sequence analysis using MOTHUR (Schloss et al., 2009) and the Silva reference database SSU Ref NR 99 v132 (Quast et al., 2012; Yilmaz et al., 2013)]. A heatmap was generated to visualize dominant OTUs in the samples using the package ComplexHeatmap (ver 2.1.0) in R (ver. 3.5.2) (Gu et al., 2016). Sequences have been submitted to GenBank Sequence Read Archive under BioProject number PRNJA561022.

DNA concentration was measured using a Qubit 3.0 high sensitivity assay (ThermoFisher Scientific, Waltham, MA, United States) and was adjusted to 5 ng/μl. A metagenome library was prepared using Nextera XT Library Preparation Kit (Illumina, San Diego, CA, United States, #FC-131-1096) according to the manufacturer’s protocol except adjusting the reaction volume to allow for processing in 96-well plate format. Library preparation and bead clean-up were performed using the Mantis Liquid Handler (Formulatrix, Bedford, MA, United States) and Epmotion (Eppendorf, Hamburg, Germany, #5075000301) automated platform. These programs cover “Tagment Genomic DNA” to “Amplify DNA” in the protocol (Mantis – Nextera XT library prep protocol) and “Clean Up Libraries” in the protocol (Epmotion – Library Clean Up protocol). On completion of the library prep protocol, the library was quantified and quality control was performed using the Quant-iT^TM dsDNA HS assay kit (Invitrogen, Carlsbad, CA, United States) and Agilent D1000 HS tapes (Agilent, Santa Clara, CA, United States, #5067-5582) on the TapeStation 4200 (Agilent, Santa Clara, CA, United States, #G2991AA) as per manufacturer’s protocol. The library was pooled with other libraries at equimolar amounts of 2 nM per library, to create a sequencing pool. The library pool was quantified in triplicate using the Qubit^TM dsDNA HS Assay Kit (Invitrogen, Carlsbad, CA, United States). Library QC was performed using the Agilent D1000 HS tapes on the TapeStation 4200 as previously. The library was prepared for sequencing on the NextSeq500 (Illumina, San Diego, CA, United States) using NextSeq 500/550 High Output v2 2 × 150 bp paired end chemistry, at the Australian Centre for Ecogenomics, according to manufacturer’s protocol.

Raw reads were processed with Trimmomatic (Bolger et al., 2014) (ver. 0.36; ILLUMINACLIP:NexteraPE-PE.fa:2:30:10, LEADING:3, TRAILING:3, SLIDINGWINDOW:4:15, HEADCROP:0, MINLEN:50) for adapter removal and quality filtering. Reads were then assembled using MetaSPAdes (Nurk et al., 2017) (ver. 3.11.1) with default parameters. Contigs whose length was <500 bp were removed using BBMap¹ (ver. 38.41). Raw and assembled reads have been submitted to GenBank Sequence Read Archive under BioProject number PRNJA561022.

Quality controlled reads for each sample were mapped onto their respective assemblies using CoverM “make”² (ver 0.2.0, B. Woodcroft, unpublished). Low quality mappings were removed with CoverM “filter” (minimum identity of 95% and minimum aligned length of 75%). Assemblies for each sample were binned by providing the contigs for each sample and filtered BAM files as input to UniteM³ (ver. 0.0.15, D. Parks, unpublished) and using a minimum contig length of 1,500 bp and Maxbin (ver. 2.2.4), MetaBAT (ver. 0.32.5), and MetaBAT2 (ver. 2.12.1) binning methods (max40, max107, mb2, mb_verysensitive, mb_sensitive, mb_specific, mb_veryspecific, and mb_superspecific). Bins were dereplicated using dRep (Olm et al., 2017) (ver. 2.2.3, sa = 0.95, comp = 40, con = 15). Bin completeness and contamination were evaluated using CheckM (Parks et al., 2015) (ver. 1.0.12). Taxonomies were assigned to each bin using GTDB-tk⁴ (Parks et al., 2018) (ver. 0.3.0).

To calculate the relative abundance of each bin, reads from each sample were mapped and filtered as was done for the unbinned assemblies. After filtering, 80% of the reads mapped to the bins. The mean coverage of each bin was calculated with CoverM. The relative abundance of each bin, among those recovered, was calculated as its coverage divided by the total summed coverage of all bins. Abundance values were multiplied by the fraction of reads that mapped to all bins after filtering to produce the relative abundance of each bin within the entire sample.

Bin assemblies were translated and functionally annotated using a combination of Prokka (Seemann, 2014) (ver 1.13.3), Prodigal (Hyatt et al., 2010) (ver. 2.6.3), and EnrichM⁵ (ver. 0.4.15, J. Boyd, in preparation), the latter using annotation options –ko_hmm, –pfam, –tigrfam, –orthologs, –clusters, –cazy, and –ec. In EnrichM, for a query gene to be considered for annotation, the minimum fraction aligning to a reference, and vice versa, was set to 0.5, with a minimum percent identity of 30% also required.

Potential c-type cytochromes were identified by determining encoded proteins with heme-binding motifs. For each bin, heme-binding motifs reported in the literature (Ferguson et al., 2008; Onley et al., 2018) were identified in all translated coding genes (amino acid sequences) outputted by Prokka, through pattern matching and counting. Hydrogenases were grouped by type as per Greening et al. (2015). Genes encoding proteins important in iron reduction in S. oneidensis (CctA, CymA, FccA, MtrA, MtrB, MtrC, MtrD, MtrE, MtrF, MshA, MshB, MshC, MshD, OmcA, and PilA) Geobacter sulfurreducens (ImcH, OmcB, OmcC, OmcE, OmcS, and PilA), and Desulfuromonas acetoxidans (cytochrome c7) were determined by BLAST analysis of the publicly available (UniProt) amino acid sequences against deduced amino acid sequences from contigs from the metagenome. Matches were considered significant (i.e., worth further investigation) at e-values <1e^–20.

Goethite and canga-reducing cultures prepared from various iron duricrust-associated samples collected in Western Australia and Brazil were initially enriched using a mixture of lactate, acetate, and glucose as per Lentini et al. (2012). When electron donors were examined individually, the cultures were found to produce equivalent ferrous iron on glucose alone, as they did on the mixture of carbon sources. Compared to when the electron donor was lactate alone, glucose-oxidizing cultures produced maximum ferrous iron much faster (6–12 days compared to 40–70 days) and to higher concentration (e.g., ∼15 mM compared to ∼5 mM, data not shown). In light of these preliminary observations from a range of cultures, and given the biotechnological potential for an effective goethite-reducing microbial culture, subsequent enrichments were prepared by inoculating freshly collected samples from iron ore regions, onto goethite in a sulfate-free basal medium, with glucose (30 mM) as the electron donor. The culture found to reduce the most goethite was characterized in more detail and will be referred to as GluFeRB. It had been enriched from a pool perched on canga (see the section “Microbial Diversity in Iron Duricrust-Associated Sample From Which GluFeRB Was Obtained”) and was passaged five times before detailed culture characterization and DNA-based analyses (sections “Characterization of Culture GluFeRB,” “16S rRNA Gene Data From Glucose-Oxidizing, Goethite-Reducing Culture GluFeRB on Sulfate-Free Medium,” and “Metagenomic Analysis of Glucose-Oxidizing Goethite Reducing Enrichment Culture on Sulfate-Free Medium”).

GluFeRB was enriched from water collected from a small pool perched on canga near Lake Violão, Carajás, Brazil (S06°24′07.5″ W050°21′05.6″). Ferrous iron was detected in the pool (purple color reaction of ferrozine assay) in the field, March 2016. Bacterial and archaeal 16S rRNA genes in the environmental sample clustered into 3960 OTUs (at <0.03% distance). Rarefaction analysis indicated that further microbial diversity would be recovered with deeper sequencing (data not shown). The microbial community was dominated by lineages that typically contain anaerobes (e.g., Methanobacteriaceae, Veillonellaceae, Clostridiaceae, Ruminococcaceae, Methanospirillaceae, Supplementary Figure 1) suggesting anaerobic fermentation and methanogenesis were major processes in the pool perched on canga. However, sequences from a phototrophic oxygenic lineage were also abundant (OTU 16, Supplementary Figure 1, 1.6% of sequences) and a Curvibacter (OTU 22, Supplementary Figure 1), a lineage recently implicated in neutrophilic iron oxidation (Gülay et al., 2018) was also abundant (1.17% of sequences). Known iron-reducing lineages were not abundant in the community profile. An Anaeromyxobacter OTU was the most abundant known iron-reducing microbial lineage detected, at 0.1% of the total sequences. A Geobacter OTU was also detected at 0.06% of total sequences. An OTU present at 0.55% (OTU 39, Supplementary Figure 1) was most similar to Azonexus WE2-4 which has been shown to be capable of extracellular electron transfer under anaerobic conditions (Jangir et al., 2016) and potentially could be contributing to iron cycling in the natural ecosystem.

In the presence of goethite (1 g/50 ml = 225 mM), on sulfate-free medium, culture GluFeRB consumed glucose (30 mM) completely within 13 days and produced acetate (27.7 mM), propionate (11.4 mM), ferrous iron (16.9 mM), and a minor amount of butyrate (0.94 mM, Figure 2). A subsample from one of the cultures showed that the pH dropped from ∼6.9 to 6.24 across this period. Acetate and ferrous iron continued to increase slightly after the consumption of glucose (Figure 2). Ferrous iron production could be re-stimulated in GluFeRB cultures that had reached a ferrous iron plateau simply by the addition of more glucose (dashed line, Figure 2A), indicating that ferrous iron toxicity was not responsible for cessation of iron reduction. The proportion of measured ferrous iron that was soluble (i.e., not adsorbed to the goethite) in GluFeRB cultures was found to be 76.3 ± 1.86% (measured at end point, in later transfers of the culture). Formate (2.1 mM), lactate (0.19 mM), and succinate (0.38 mM) were produced in the first 6 days and consumed within the next 7–14 days. In comparison, when GluFeRB was grown on glucose only without goethite, glucose was consumed much slower (<42 days) and the major end product was propionate (16.19 mM) with a smaller amount of acetate (10.8 mM) and minor butyrate (1.3 mM; Figure 2B). Lactate was never detected (Figure 2) and acetate:propionate was much higher in the presence of goethite (Ac:Pr 2.42) compared to cultures not provided with goethite (Ac:Pr 0.67; Figure 2).

GluFeRB was dominated by two OTUs at almost equal proportions, one demonstrating 95% 16S rRNA gene identity across the sequenced region to the nearest named isolate Elstera litoralis and the other demonstrating 98% 16S rRNA gene identity to Paludibacter jiangxiensis, neither of which are reported to be able to use ferric iron as a terminal electron acceptor (Rahalkar et al., 2012; Qiu et al., 2014). The five next most abundant OTUs indicated the likely presence of a hydrogenotrophic methanogen [100% 16S rRNA gene identity to Methanobacterium lacus (Cadillo-Quiroz et al., 2014)], a sulfate reducing bacterium (100% 16S rRNA gene identity to various strains of Desulfovibrio desulfuricans), two closely related fermentative bacteria (both OTUs > 99% 16S rRNA gene identity to Bacteroides xylanolyticum), and a Clostridium species [100% 16S rRNA gene identity to Clostridium aciditolerans (Lee et al., 2007)]. The dominant members of culture GluFeRB were not detected in the amplicon data from the parent material and the other members of the culture were also low abundance (Methanobacterium 0.67%, Desulfovibrio 0.25%, Clostridium OTUs 0.56–0.005% of total sequences, Supplementary Figure 1).

Six metagenome assembled genomes (MAGs) were reconstructed from metagenomic data from culture GluFeRB (Table 1). These represent population genomes for the main organisms (80% of total metagenome reads mapped to these MAGs) in the glucose-oxidizing, goethite-reducing culture on sulfate-free medium. MAGs were taxonomically classified using the Genome Taxonomy Database toolkit (GTDB tk) version 3.0 (see text footnote 4) as falling within the genera Telmatospirillum, Paludibacter, Desulfovibrio, Hungatella, Methanobacterium, and the family Clostridiaceae (Table 1). Genome phylogeny for these six MAGs was consistent with the most abundant OTUs by 16S rRNA gene analysis, except that the two OTUs most closely related to B. xyanloyticum at the 16S rRNA gene level, are likely represented by a single MAG. Relative abundances of the MAGs were consistent with those observed for correspondingly classified OTUs except for Methanobacterium which was overrepresented in the 16S rRNA gene data (11.5% of 16S rRNA gene sequences, 3rd most abundant OTU) compared to the metagenomic data (1.0% relative abundance, 6th most abundant genome).

A search for proteins involved in iron cycling, based on those characterized in S. oneidensis or G. sulfurreducens, revealed putative iron reduction capabilities within the Telmatospirillum MAG (Table 2). Two putative MtrA proteins (with 10 and 9 predicted heme-binding motifs, respectively) demonstrated high homology (Table 2) to MtrA in S. oneidensis and are likely DmsE family decaheme c-type cytochromes. Both genes were neighbored by genes encoding proteins demonstrating homology to transmembrane porin MtrB/PioB family proteins (Table 2). An extensive search for MtrC and OmcA homologs in the Telmatospirillum MAG did not return any significant matches (the genes encoding putative MtrA were the most significant to MtrC at e-values ∼1e^–7). However, a cytochrome c-552 family protein with two predicted heme-binding motifs directly neighbors one of the putative MtrA genes (Table 2) and is predicted to be present in the extracellular space based on BUSCA analysis (Savojardo et al., 2018). It is the most likely candidate for extracellular electron transfer from the membrane-spanning MtrAB complex to goethite and/or to electron shuttles (Figure 3).

A homolog to CymA, the tetraheme cytoplasmic membrane protein that transfers electrons from the quinol pool to periplasmic proteins [in the case of iron reduction, to the carrier FccA or to MtrA directly (Liu et al., 2018)] is also present in the Telmatospirillum MAG (Table 2). It exists within an operon of genes for nitrate reduction, reflective of the role this protein plays in electron transport to various electron acceptors, e.g., iron, fumarate, nitrate, and manganese (Myers and Myers, 2000). The Telmatospirillum MAG also encodes a protein demonstrating high homology to FccA in S. oneidensis (Table 2). None of the other MAGs in culture GluFeRB contained homologs for characterized outer membrane metal reduction proteins. A search for cytochrome c7 homologs [cyt c7 is involved in metal reduction in some sulfur reducing bacteria (Banci and Assfalg, 2006)] in the metagenome did not return any significant matches.

Respiratory complexes I–IV are present in Telmatospirillum (Figure 3), as well as genes encoding the high-affinity cytochrome cbb3 type and cytochrome bd complex terminal oxidases that are typically expressed under microaerobic conditions (Pitcher and Watmough, 2004; Borisov et al., 2011). The Paludibacter and Desulfovibrio MAGs also encode respiratory complexes I, II, and cytochrome bd complex (Supplementary Figures 2, 3).

The Desulfovibrio and Telmatospirillum MAGs demonstrate evidence for nitrate reduction, and the Desulfovibrio also encodes the proteins for sulfate reduction (Figure 3 and Supplementary Figure 3). Both Telmatospirillum and Desulfovibrio MAGs encoded multiple multi-heme (>2 hemes) cytochromes (at least 11 and 14, respectively) further indicative of respiratory flexibility, while multi-heme cytochromes were not predicted in other bacterial MAGs in GluFeRB.

The four most abundant organisms in the culture seemed dependent on heterotrophy (Embden–Meyerhof–Parnas or Entner Doudoroff pathways) for carbon, while the lower abundance Clostridiaceae and Methanobacterium MAGs contained genes for carbon fixation via the Wood–Ljungdahl pathway. Genes encoding enzymes involved in mixed acid fermentation were encoded in all bacterial species, with major end products predicted to be acetate, propionate, and lactate (Figure 3 and Supplementary Figures 2, 3). All bacterial MAGs contained both hydrogen-consuming and hydrogen-evolving hydrogenases (Greening et al., 2015) with more FeFe hydrogenases (likely hydrogen evolving) in the predicted fermenters Hungatella and Clostridiaceae, as expected. All MAGs were predicted to encode molybdenum nitrogenases for nitrogen fixation and Methanobacterium also encoded an iron-containing nitrogenase. Hungatella, Clostridiaceae, Telmatospirillum, and Desulfovibrio are likely motile based on the presence of proton driven flagellar assembly machinery.