Effects of enrichment on system performance
Before enrichment, a cultivation step was performed to alleviate potential electrochemical shock to the methanogenic population by transferring Ca. G. eutrophica-containing GAC from a packed-bed bioreactor to bioelectrochemical systems amended with fresh GAC. After 20 cycles (60 days) of cultivation, the electrochemical seed (0 V vs. Ag/AgCl) showed lower methane production than the control (0.1 L/L/d vs. 0.5 L/L/d, Figure 1A) but similar COD removal of about 1200 mg/L (Figure 1B). Additionally, acetate accumulation was observed in the electrochemical seed effluent (up to 50% of the TOC, Figure 1C). The results indicate strong electrochemical shock to the methanogenic population, particularly acetoclastic methanogens, during the initial cultivation. GAC was collected from the seeds for enrichment.
In Batch 1, the methane production rate of the control reactors increased slightly to 0.6 L/L/d (Figure 1A), while that of the 0-V reactors increased significantly to 0.4 L/L/d, suggesting recovery of the methanogenic population from electrochemical shock. One of the 0-V reactors produced low biogas throughout the first batch due to rapid pH drop (average 5.2 after each cycle, Supplementary Figure S2), but the reason was not clear. The 0-V reactors exhibited more stable performance in the following batches and produced CH4 at a rate of 0.6 L/L/d. Higher potential (0.4 V vs. Ag/AgCl) did not affect methane production, and both 0-V and 0.4-V reactors produced CH4 slightly slower than the control (p <0.05). Another group of reactors were operated at 0 V but fed with ethanol as the sole carbon source, as described in pure-culture DIET studies [8]. However, the EtOH reactors produced CH4 at a much lower rate of 0.1 L/L/d in all batches.
At the early stage of the enrichment, the fructose/PEG-fed reactors (i.e., control, 0 V, and 0.4 V) showed similar COD removal of about 1200 mg/L, whereas the EtOH reactors removed only 250 mg/L COD. COD removal doubled for all reactors in the following batches, but this enhancement was not reflected by CH4 production, which increased only by <20% (Figure 1A). In the meantime, CE showed a decreasing trend and dropped by ~3% for the fructose/PEG-fed electrochemical reactors and by 10% for the EtOH reactors. High COD removal and low CE are expected as a significant proportion of the electrons will be channeled to biomass, metabolites, and CH4 when fermentable substrates are fed as the electron donor [62].
The concentrations of short-chain VFAs (acetate, propionate, and butyrate) and ethanol in the effluent were converted to TOC for comparison (Figure 1C). For the electrochemical reactors, effluent TOC dropped from 800 mg/L in the seeds to about 500 mg/L in Batch 1. The abnormally high TOC found in one of the 0-V reactors in this batch was consistent with its gas production and COD removal and was attributed to the low pH that caused reactor failure. Total VFAs in those reactors further decreased to below 200 mg-TOC/L in Batch 2 and 3 with noticeable propionate accumulation. Ethanol was detected with a low concentration (<10 mg-TOC/L), which was not in agreement with our previous study [25]. For the EtOH reactors, effluent TOC remained stable throughout the enrichment and consisted of 400 mg-TOC/L acetate followed by 100 mg-TOC/L ethanol. Acetate accumulation and low CH4 production together suggest inhibition of acetoclastic methanogenesis, likely because the presence of acetate and electrode favors the growth of electroactive acetate scavengers and potentially leads to the washout of acetoclastic methanogens [63].
Effects of enrichment on microbial population and overall activity
A clear shift in the community structure in the 0-V and 0.4-V reactors can be seen from the 16S rRNA gene-based PCoA results (Supplementary Figure S3). The communities in those reactors became significantly different from the seed communities (PERMANOVA, p <0.05) and clustered closely with the control communities in Batch 3, demonstrating successful enrichment. Similar results were reported by a previous study, in which poised potentials exerted minor impacts on Geobacter-dominating microbial communities [64]. The EtOH-fed communities, on the other hand, were separated from the fructose/PEG-fed communities since Batch 1 and remained stable throughout the enrichment. The significant difference in microbial communities agreed well with the system performance (Figure 1), highlighting the key role of the substrate as a deterministic factor that drives microbial community dynamics [34, 65].
A core population composed of 22 OTUs was selected based on the criteria of average abundance >0.5% and occurrence >50% across all DNA and RNA samples (Figure 2). In the 0-V and 0.4-V reactors, Ethanoligenens OTU682 was initially abundant (57% 16S rRNA gene abundance) and active (49% 16S rRNA abundance) and was replaced by Clostridium OTU92 during the enrichment. Rhodocyclaceae OTU196 and Desulfovibrio OTU66 were the dominant taxa in the EtOH reactors. G. sulfurreducens-related OTU268 was stimulated and enriched in all electrochemical reactors, and the correlation with CE (Supplementary Figure S4) revealed its electroactive nature [66]. Also enriched were strict hydrogenotrophic Methanobacterium spp. [67, 68].
Geobacter OTU650 is of particular interest due to its high phylogenetical similarity to Ca. G. eutrophica (Figure 2). It was outcompeted in the control but not under electrochemical stimulation. Although its abundance decreased to between 5% - 10% in the 0-V and 0.4-V reactors in Batch 3, OTU650 was still among the top three most abundant taxa. Moreover, it contributed to 15% to 20% of the community 16S rRNA and showed high activity throughout the enrichment. OTU650 was absent in the EtOH reactors and might not be a competitive utilizer of ethanol and acetate while respiring on an electrode. Instead, Activity-based RDA predicted the association between this taxon and propionate (Supplementary Figure S4). OTU650 is also predicted to be involved in CH4 production. The significant difference between the control and 0-V/0.4-V reactors (p <0.05) confirms that Ca. G. eutrophica-related OTU650 is capable of EET.
Fermentative bacteria providing substrate for methanogenesis
Metagenomic analysis yielded 28 high-quality MAGs (>98% completeness and <2% contamination, Supplementary Figure S5), whose bin coverage and percentage of mapped reads in Batches 1 and 3 agreed well with the relative abundance of the selected core population shown in Figure 2. We reconstructed the metabolic pathways for fructose utilization, propionate/propanol accumulation, ethanol production/utilization, and H2 metabolism for the four primary fermentative bacteria (Supplementary Figure S6).
Ethanoligenens co_bin 13 was abundant in the 0-V and 0.4-V reactors in Batch 1 (Supplementary Figure S5). The genes for fructokinase (scrK), mannose PTS system (manXZ), and mannose isomerase (manA) were highly expressed, suggesting that co_bin 13 could metabolize fructose directly or with mannose as an intermediate. The resulting β-D-fructose 6-phosphate was fed into glycolysis and led to the accumulation of propionate through the generation of glycerone phosphate, methylglyoxal, and propionyl-CoA (Supplementary Figure S6). It actively expressed six genes that are important for the conversion of propionate/propionyl-CoA, but the genes for downstream propionyl-CoA utilization were not detected. Similar to other Ethanoligenens species [69-71], co_bin 13 could carry out complete glycolysis to produce ethanol, as indicated by the expression of the adhE gene encoding aldehyde-alcohol dehydrogenase. However, genes for H2-evolving hydrogenase were not found in its draft genome. The results collectively suggest that Ethanoligenens co_bin 13 ferments fructose to ethanol, propionate, and propanol.
Clostridium B1_C1_bin17 was ubiquitously abundant in the fructose-fed reactors (Supplementary Figure S5). In addition to the two fructose metabolism pathways mentioned above, Clostridium B1_C1_bin17 expressed a third route with mannitol as an intermediate. It could convert the produced β-D-fructose 6-phosphate to propionate-CoA either via the glycerone phosphate-methylglyoxal route or through lactate metabolism (Supplementary Figure S6), further underpinning its metabolic flexibility. B1_C1_bin17 could also be a major propionate producer. The ability to produce 1,3-propanediol and propanol by Clostridium species was observed in B1_C1_bin17 through the expression of the dhaT gene [72, 73], a 1,3-propanediol dehydrogenase-like enzyme that potentially catalyzed NADH-dependent propanal reduction to propanol. Finally, B1_C1_bin17 metabolized ethanol by expressing three alcohol dehydrogenase-encoding genes and the genes for the periplasmic [NiFe] and [NiFeSe] hydrogenases. The expression was one order of magnitude higher in the fructose-fed reactors than in the EtOH reactors, suggesting that it could couple fructose degradation to ethanol metabolism with proton as an electron sink. Overall, Clostridium B1_C1_bin17 contributes propionate, propanol, and H2 to methanogenesis.
Rhodocyclaceae B3_EtOH1_bin9 and Desulfovibrio B3_EtOH1_bin5 preferred ethanol as the substrate (Supplementary Figure S5). Genomic analyses confirmed that the Rhodocyclaceae population was incapable of fructose metabolism, and both were deficient in propionate production (Supplementary Figure S6). The Rhodocyclaceae population showed the ability to metabolize ethanol and carried the genes for the NAD-reducing Hox complex and a periplasmic [NiFeSe] hydrogenase, which indicated its role as a major H2-donating partner. The Desulfovibrio population also actively expressed the genes for periplasmic [Fe] and [NiFe] hydrogenase complexes that allowed it to use proton as an electron acceptor and grow syntrophically with H2-scavenging partners [74, 75]. Alternatively, under electrochemical stimulation, Desulfovibrio B3_EtOH1_bin5 may use its own H2 as an electron donor to respire on the poised electrode. B3_EtOH1_bin5 showed high activity of the Hnd complex that was recently reported to transfer electrons from H2 to NADH through flavin-based bifurcation [76]. We also observed high activity of the menaquinone reductase complex (Qrc) involved in sulfate respiration [77], which might explain the EET ability found in several Desulfovibrio spp. [78, 79].
Geobacter playing a key role in methanogenesis
Geobacter B1_C1_bin15 (OTU650) was recovered with an almost complete genome (completeness >99.4% and contamination <0.6%) and confirmed to be phylogenomically nearly identical to Ca. G. eutrophica (Figure 3). Metabolic construction reveals a complete set of genes for propionate metabolism, seven copies of the dhaT gene for 1,3-propanediol dehydrogenase for propanol metabolism, and high activity of several types of alcohol and aldehyde dehydrogenases (Supplementary Figure S7). These results are consistent with the previous finding that G. metallireducens grows syntrophically with DIET partners on propionate, propanol, and ethanol [15].
Through mapping the metatranscriptomic reads to a manually curated database, we identified three outer membrane c-type cytochromes that were slightly more active (up to 2 folds, p <0.05) in the electrochemical reactors after enrichment (Figure 3). Among them, OmcF appears to be required for the transcription of the gene encoding OmcC that is directly responsible for Fe(III) reduction and potentially GAC-mediated EET/DIET in this study [80]. Three copies of the conductive pili-encoding pilA gene were also detected but did not show a noticeable difference under different conditions. One of the pilA genes, with 72% identity and 100% coverage relative to that found in G. metallireducens, was actively expressed, implying its involvement in EET/DIET.
We also observed the expression of the genes for putative H2-evolving hydrogenases (Figure 3). The Hox complex was globally active across batches and conditions, while the periplasmic [NiFe] and [NiFeSe] complexes became slightly less active after enrichment but still showed 2-fold higher expression under electrochemical stimulation. Further inspection of the B1_C1_bin15 genome revealed that it carries a quinone-reactive Ni/Fe hydrogenase and an Hnd complex in which the genes (hndABCD) are 55-81% similar (>98% coverage) to those found in Desulfovibrio fructosovorans. Both were highly active in the electrochemical reactors and could be involved in H2 uptake couple with EET to electrode [76].
Geobacter co_bin3 is phylogenomically close to Ca. G. eutrophica (Figure 3) and seems to compete with Geobacter B1_C1_bin15 for the same ecological niche based on their interactive changes in abundance (Figure 2 and Supplementary Figure S5). This is supported by the gene expression profile, which shows lower activity of the omcF gene, five copies of the pilA gene, and the genes for H2-evolving hydrogenases and the Hnd complex under electrochemical stimulation.
Unlike the other two members in this genus, Geobacter B1_C1_bin22 is ubiquitously abundant in both fructose- and ethanol-fed electrochemical reactors. Its phylogenomic similarity to G. sulfurreducens is reflected by the carriage of a number of cytochromes from the Omc family (Figure 3), including four copies of the well-characterized pili-associated OmcS that is highly upregulated during EET to Fe(III) oxide and electrode [11]. This population also showed activity of both H2-evolving and H2-uptake hydrogenases, with the latter potentially playing a central role in its growth. Similar to G. sulfurreducens, B1_C1_bin22 is not likely to grow on ethanol [81]. It expressed only one gene for putative alcohol dehydrogenase and lacked acetaldehyde dehydrogenase. Therefore, Geobacter B1_C1_bin22 may compete with hydrogenotrophic methanogens on H2 as an electron donor for EET, thereby suppressing methane production and producing a high CE in the EtOH reactors (Figure 1).
Hydrogenotrophic methanogens with DIET potential
Although the bioelectrochemical systems were designed to stimulate DIET-capable Geobacter, we observed high abundances of several Methanobacterium spp. with high phylogenomic similarity to known species (Supplementary Figure S5) and retrieved near-complete genomes (>99% completeness and <1% contamination). These strict hydrogenotrophic methanogens use three routes to recycle coenzyme M and coenzyme B at the end of the methanogenic pathway (Supplementary Figure S8). The ferredoxin:CoB-CoM heterodisulfide reductase (HdrA1B1C1), a homolog of the HdrABC complex commonly found in most methanogens, became less active in the control. On the other hand, they possess the heterodisulfide reductase [NiFe]–hydrogenase complex for flavin-based electron bifurcation coupled with ferredoxin reduction and H2 oxidation [82], and the genes encoding the complex (hdrABC/mvhADG) were highly active in all treatments. Membrane-bound heterodisulfide reductase (hdrD) and F420 dehydrogenase (fpoD), which were proposed to participate in extracellular electron uptake [6], were also present but not active after enrichment, indicating their weak roles in methane production. An interesting finding is that the Methanobacterium spp. carry up to seven copies of the mvhB gene that are actively expressed across different batches and conditions, and the encoded polyferredoxins with high iron content have been speculated to participate in electron transfer [83, 84]. It is possible that the Methanobacterium spp. use polyferredoxins to shuttle extracellular electrons to the MvhADG/HdrABC complex to complete DIET, but the specific electron uptake and transfer mechanisms need further investigations.
A predictive understanding of the methanogenic population
Bayesian network analysis was performed to understand the microbial interactions and the potential functions of key taxa in a given microbial ecosystem [33]. To investigate the effects of input data type on network training, we selected 17 and 20 OTUs from the DNA and RNA datasets (abundance >0.5% and occurrence >50% in each dataset), respectively. The modeling method was first validated by reconstructing the core populations. The Bray-Curtis similarities between the predicted and observed communities (0.62 for DNA and 0.64 for RNA) were significantly higher than those from a null model (Supplementary Figure S9). Although the predictive power might be compromised by functionally redundant taxa at a high taxonomic resolution [33], the simulation was more accurate than that yielded by artificial neural networks (constructed to predict acid mine drainage communities) [61]. The Bayesian network approach was further validated by correlating the predicted and observed system performance (Supplementary Table S1). Satisfactory predictions were achieved with an average R2 >0.61 and RMSE of 0.11. Among the parameters examined, methane production was predicted with the highest accuracy, and the R2 with the RNA dataset reached 0.94. Other parameters such as acetate and ethanol in the effluent were also better predicted with RNA than with DNA. Overall, RNA was a more robust indicator than DNA with a significantly higher average R2 (0.69 vs. 0.61) and lower RMSE (p <0.05), demonstrating the strong connection between microbial activity and system performance.
After the Bayesian network modeling approach was validated with Bray-Curtis similarities and system performance predictions, a final network was constructed with RNA at the OTU level (Figure 4). The inference direction from propionate to Ca. G. eutrophica-related OTU650 implies its potential role as a propionate utilizer, which is consistent with the results from RDA (Supplementary Figure S3) and metabolic reconstruction (Supplementary Figure S7). All three Geobacter taxa in the core population were associated with methane production. OTU650 was assigned with the highest positive coefficient (0.42) followed by OTU28 (0.36), confirming their contribution to methanogenesis via IHT and/or DIET (Figure 3). OTU268 showed a negative association (-0.12) and potentially competed with hydrogenotrophic methanogens on H2. As revealed by the metatranscriptomic profiling, this G. sulfurreducens-related taxon is incapable of complete oxidation of ethanol but actively expressed hydrogenases for H2 uptake (Figure 3). The inference could also explain the less accurate prediction of methane production at a higher taxonomic level (i.e., at the genus level, Supplementary Table S1). In the genus-level network, the three physiologically distinct Geobacter OTUs were combined under the same genus, and their individual impacts on methanogenesis were neutralized, leading to a decrease in the predictive power. The inference presents a significant step toward interpreting black-box machine learning models by providing appropriate inputs (i.e., 16S rRNA) for model training [85].