Volatile Fatty Acids.

VFA accumulation can occur as a product of instability [44], can be transitional [45–47] and can even have little to no effect on biogas production [48]. Initial accumulation of iso-valeric and acetic acids was seen in both reactors: the relative difference between build-ups (initially three-fold higher in R1; higher thereafter) suggests this was due to hydrolysis and fermentation of the most accessible fractions of U. lactuca.

NH3.

The recommended ratio of carbon to nitrogen (C:N ratio) for anaerobic digestion is between 20:1 and 30:1. C:N ratios for U. lactuca range between 7:1 [18] and 14.5:1 [5]. C:N ratios for feedstocks in this study were 10.2:1 for R1 and 17.1:1 for R6, with higher values reflecting addition of slurry (C:N ratio often >20:1 [49]). Proteins contribute nearly all of the nitrogen in U. lactuca [8], entering solution as free ammonia (NH₃) or the ammonium ion (NH₄⁺). Elevated pH, temperature, and headspace partial pressure increase concentration of the uncharged NH₃ state. At sufficiently high concentrations NH₃ diffusion across cell membranes can inhibit the biogas process by causing loss of cellular potassium, de-potentiating the cell membrane, and accumulating in the cytoplasm [50]. Ammonia inhibition is well documented in methanogens [50–53], affecting other taxa to a greater or lesser extent. Pure cultures of methanogens remain viable at TAN levels up to 10,000 mg/L but have been documented to decline at a range of TAN levels between 1,700 to 6,000 mg/L when a part of a reactor community. Differential responses between hydrogenotrophic and acetoclastic methanogens are documented but contradictory (see reviews [54] and [55]).

Mineral salts.

An inhibitory role for salts has long been recognised in anaerobic digestion [56]. Cations (e.g. Na⁺, Ca²⁺, Mg²⁺, K⁺) affect biogas production in a charge-dependent manner, possibly by inhibiting a Na⁺ export channel necessary for the final methanogenic reaction [57]. However, complex and proportionate mixes of cations can offset the inhibitory effects of one another [56,58], as well as ameliorating inhibition of the biogas process due to ammonia [53] and VFA inhibition [59]. Pre-trial characterisations showed slurry to have low (< 2,000 mg/L) total mineral content, while fresh U. lactuca provided 5,220, 5,310 and 9,950 mg/L of Mg²⁺, Na⁺ and Ca²⁺ respectively. Monitored levels of Cl^- infer that salt-loading was significantly higher in R1, with a two-fold difference between R1 and R6 at close of trial (~10,300 and ~5,400 mg/L respectively). Reported inhibitory levels of Na⁺ and Ca²⁺ vary, with lower estimates of inhibition registering from 5,000 mg/L upwards [54]. Community acclimatisation and/or later inhibitory onset are likely, due to gradual accumulation and the variety of salts.

Sequencing results and diversity measures.

Pyrosequencing generated 270,111 raw sequences, which following denoising in Acacia and processing in QIIME resulted in 89,251 sequence reads (average length: 244bp) being produced, with an average of 8,114 reads per trial time-point. To ensure representative samples from both reactors, diversity metrics were calculated to estimate sensitivity to species diversity (Chao1 index) and species abundances (Simpson's Index). Rarefaction curves of these indices indicate that the most abundant species were thoroughly characterised in this study (see S1 Fig). However, rarefaction curves also indicate that a large number of low-abundance Archaea, Bacteria and unidentified taxa remain undetected due to insufficient depth of sequencing. Finally, both diversity indices (Chao1, Simpson's) decreased in later samples, suggesting the maturation of trophic systems in both reactors, where 'surplus' diversity is marginalised beyond the sequencing threshold.

Archaeal communities.

Methanosarcina was the most abundant genus in this study (7 OTUs, 9.7% of all sequence reads), the majority of which originated from R6 (9.5% of all sequence reads). Large Methanosarcina populations are known to effectively buffer against fluctuations in substrate availability, preventing accumulation or shock loading of acetic acid [60,61]. Methanosarcina has a documented tolerance for acetic acid up to 15,000 mg/L, and a higher tolerance for changes in pH and salt (see review in [62]) than hydrogenotrophic counterparts. Methanothrix, an obligate acetoclast [63], was scarce or absent in this study, likely out-competed by the higher growth rate of Methanosarcina at non-limiting acetate concentrations [47,60,64], or inhibited by salt [54] or ammonia [52–55].

Hydrogenotrophic methanogens (Methanoculleus, Methanobrevibacter, Methanobacterium, Methanocorpusculum, Methanospirillum and Methanosphaera in this study) are commonly found in anoxic sediments [65], as gut flora [66–68], and in AD reactors where they sometimes dominate [13,69]. However, most archaeal OTUs were observed at consistently low frequencies (<0.5% of total sequence reads respectively), often disappearing below the threshold of sequencing coverage.

Bacterial communities.

Bacterial components of these reactors are typical of biogas communities, while some key and accessory species are associated with marine or salt environments. The most abundant phylum was Firmicutes (565 OTUs, 36% of all sequence reads), containing many groups known to hydrolyse polymers (e.g. cellulose, lignin, polysaccharides, proteins: Lachnospiraceae, Peptostreptococcaceae, Ruminococcaceae), ferment carbohydrates (e.g. saccharides, amino acids, organic molecules: OPB54, Gelria, Christensenellaceae), and produce organic acids as metabolic endpoints (i.e.: acidogens: Sedimentibacter Tissierella, Syntrophomondaceae). Firmicutes are major components of anaerobic environments such as digesters [13,69,70] and alimentary tracts [71,72], and in this study accounted for over a third of sequences in both reactors: in short, they are highly diverse, widely distributed, and understood as essential components of anaerobic digestion.

The second-most abundant phylum, Bacteroidetes (126 OTUs, 16% of all sequence reads), is also frequently detected in anaerobic reactors, with important roles as fermenters and acidogens. In particular, species from the family Porphyromonadaceae (9% of all reads) are known to be involved in the degradation of proteins and amino acids, eschewing saccharides (genera Petrimonas [73] and Proteiniphilum [74]).

Phylum Proteobacteria (203 OTUs, 13% of sequence reads) comprises the most diverse known taxonomic group of the Bacteria to date. The sub-ordinate classes Alpha- and Gamma-Proteobacteria contributed 3% and 7% of reads in this study respectively, with remaining proteobacterial classes totaling 3%. Proteobacteria are typical residents of anaerobic digesters [13,69,75], known to incorporate nitrogen and/or sulphur as electron acceptors in the metabolism of varied carbohydrates (e.g.: Nitrosimonas, Nitrobacter). However, some species observed here are unexpected inclusions, with described preferences for aerobic metabolism (in some cases obligate: Rhodobacteraceae, Granulosiococcaceae, Nannocystineae;) and a high propensity for saline and marine environments (water: Rhizobacteraceae; sediments: Desulfomicrobium; seaweeds and plants: Alteromonadaceae, Nannocystinaceae, Granulosiococcaceae). As such, their presence in this study likely reflects persistent contributions from the U. lactuca feedstock alongside species typical of a biogas digester habitat.

Phylum Spirochaetes (47 OTUs and 6% of sequence reads in this study) are diverse, highly motile, frequently anaerobic bacteria, but metabolic information on this phylum in anaerobic digesters is somewhat limited despite being frequently encountered in low or medium abundances. They have been characterised both as acetogens [76,77] and acetoclasts assisting methanogenic activity (as Syntrophic Acetate-Oxidising Bacteria) [78].

Phylum Synergistetes comprised 6% of all sequence reads and 34 OTUs. Synergistetes are typically seen at lower abundances in a wide variety of environments [79], in syntrophic associations with hydrogenotrophic species (e.g. methanogens). A possible role in these reactors is likely to be oxidising amino acids as a substrate in the presence of methanogens [80,81].

Most phyla were present at much lower levels (< 2% of reads): Phylum Chloroflexi contains fermentative, acido- and acetogenic, obligate and facultative anaerobes seen in anaerobic digesters and hot springs respectively, and requires removal of hydrogen which suggests syntrophic roles [82]. Phylum Tenericutes is represented by Acholeplasma spp.- poorly characterised sugar fermenters [83]; Species from Phylum Actinobacteria contain many heterotrophic fermenters including lipidophiles, and obligate marine-associated species [84]; Phylum Acidobacteria species are uncharacterised but similar to sequences recovered from petrochemical-contaminated aquifers (isolate BPC102, NCBI accession AF154083.1); Taxa from Phylum Armatimonadetes are expected to be chemo-heterotrophs, and are suggested to associate with degradation of photosynthetic biomass [85].

Although the eleven phyla outlined above describe over 94% of all sequence reads, the remaining Bacteria (only 6% of reads, 135 OTUs) correspond to at least a further 26 phyla/candidate phyla, again reflecting the huge diversity in anaerobic reactor communities.

Changes in R1 community makeup.

Week 1 conditions were initially favourable for R1 at an OLR of 2 kg VS m³ d^-1, albeit with slightly elevated TAN and VFA levels (~2,000mg/L apiece). Community abundances were relatively balanced between hydrolysers, fermenters and acido-/acetogens (Clostridiales, Bacteroidales, Desulfovibrionales, Synergistales), with environmental inclusions (Rhizobiales, Rhodobacterales, Myxococcales) characteristic of slurry, U. lactuca, or marine sources.

Until Week 5, Methanosarcina abundances held at half (~1%) of all R1 archaeal sequence reads (~2%), suggesting conditions for acetoclasts were initially favourable. Canonical cellulose and protein degraders proliferated (Ruminococcaceae, Lachnospiraceae, Proteiniphilum). As TAN approached 3,500 mg/L, early accumulation of acetic and iso-valeric acid shifted to a sudden peak in iso-valeric acid (3,500mg/L) and depletion of acetic acid after Week 5. To reduce VFA content, OLR was reduced to 1 kg VS m³ d^-1 in Week 7, while Cl^- levels passed 5,000mg/L.

Week 13 sequence reads showed a sharp rise in abundance of the Pseudomonadales genus Psychrobacter to 25%, alongside catabolism of accumulated iso-valeric acid to propionic and acetic acid. Associated with cold marine environments, Psychrobacter is likely to reduce amino and organic acids to acetic acid [86], suggesting an important role in continuous digestion of U. lactuca and slurry. However, Methanosarcina abundances were negligible (<0.1% of sequence reads) and not detected at end of trial, despite stable reactor conditions (FOS:TAC 0.21–0.31 until Week 26), a lack of inhibitory VFAs (<4,000 mg/L [62], and favourable levels of acetic acid for that genus (1100–1300 mg/L [62]; evidenced by similar concentrations in R6, Week 13). Hydrogenotrophic Methanobrevibacter and Methanoculleus were then the dominant Archaea in R1, at <1% of sequence reads.

Metabolism of accumulated propionic acid by Week 21 coincided with receding Psychrobacter abundance and expansion of hydrolysing and fermenting populations emphasising protein/amino acid metabolism and acetogenesis (OPB54, Ruminococcaceae, Peptostreptococcaceae, Proteiniphilum, Aminobacterium). TAN continued to increase (~4,700mg/L) alongside steady levels of acetic acid (~1,000 mg/L) as the main VFA. Hydrogenotrophic methanogens persisted at low levels (<1% sequence reads).

After peaking at Week 23 (~5,000mg/L), TAN stabilised by Week 30 (~4,000mg/L) while acetic and propionic acid had re-accumulated (~2,300 mg/L and ~500 mg/L respectively). Despite receding TAN, hydrogenotrophic methanogens declined further, with small shifts in bacterial populations from likely peptide (Aminobacterium, Proteiniphilum, Psychrobacter, Peptostreptococcaceae) to polysaccharide metabolisers (Acholeplasmataceae, Ruminococcus, OPB54).

Increasing OLR to 1.5 kg VS m³ d^-1 at Week 34 exacerbated accumulation of TAN (+5,000mg/L), Cl^- (~6,800mg/L), and VFAs (chiefly acetic and propionic acid: ~3,200 and ~700mg/L respectively; FOS:TAC >0.4; declining biogas output). By Week 39, OPB54 (36% of sequence reads), Proteiniphilum (13%) and Acholeplasmataceae (9%) represented the most relatively abundant populations while Archaea contributed only 0.3% of sequence reads.

Changes in R6 community makeup.

Initial levels of VFA and TAN in R6 were similar to R1, with accumulation of acetic and iso-valeric acid at lower levels, and large hydrolysing, fermenting and aceto-/acidogenic populations (Clostridiales: 32% of sequence reads, Bacteroidales: 10%, Synergistales: 8%). Notably, Methanosarcina was considerably more abundant at Week 1 (10% of sequence reads, as compared to 1% in R1). This may reflect a rapid acclimatisation to substrate (uncharacteristic of methanogens), or contribution from the three-fold higher slurry portion. R6 Archaea were also more diverse, including Methanspirillum, Methanocorpusculum, Methanomasciliicoccus.

Week 6 saw TAN rise above 2,000mg/L, with iso-valeric acid quickly metabolised to acetic acid. Methanosarcina relative abundance doubled to 22% of sequence reads, while Clostridiales and Synergistales taxa showed some decline in relative abundance.

Cl^- levels passed 5,000mg/L at Week 10. Week 13 represented the high point in biogas production, acetic acid availability, and Methanosarcina abundance (24% of sequence reads), alongside diverse bacterial populations with low, evenly-distributed abundances. The largest populations were acetogenic gut-associated saccharide fermenters (Christensenellaceae, Rikenellaceae: 4–6%), cellulose (Ruminococcaceae, Lachnospiraceae: 4–5%) and peptide (Peptostreptococceae, Proteiniphilum, Sedimentibacter: ~3%) degraders. Crucially,. Subsequent rises in propionic (700mg/L) and iso-caproic acids (600mg/l) were rapidly catabolised to acetic acid.

With TAN rising (~2,500mg/L) and a decrease in B_eff at Week 20, initially abundant bacterial taxa (Peptostreptococcaceae, Lachnospiraceae, Christensenellaceae, Rikensellaceae) were replaced by functionally similar populations (Ruminococcaceae, Proteiniphilum, Psychrobacter, OPB54) while Methanosarcina relative abundance decreased (18%) in conjunction with limiting acetic acid, similar to perturbation in the R1 community. An otherwise stable methanogen population (1.4%) suggests biogas obstruction prior to methanogenesis; sudden elevation of valeric acid (~500mg/L) implicates disrupted acetogenesis. Cl^- levels peaked at Week 21 (~6,800mg/L), but decreased thereafter (~6,000mg/L).

TAN peaked at 3,000mg/L in Week 25 before stabilising to ~2,000mg/L by Week 27, despite an increased loading rate of 2.5 kg VS m³ d^-1. Abundances shifted towards larger, mono-typic populations of fermenters and acidogens, displacing degraders of cellulose and proteins, possibly in response to increased substrate availability. Relatively ideal reactor conditions (FOS:TAC 0.22–0.24; free ammonia and chloride below inhibitory levels; VFA concentrations below inhibitory levels despite an increased OLR [54,62]) and stable levels of Methanosarcina, combined with accumulated higher VFAs despite limiting acetic acid again suggest some inhibition of acetogenesis rather than methanogenesis is responsible for the decreasing yield seen in later R6 time-points.

Taxonomic characteristics.

To improve characterisation of the different microbe communities digesting slurry/U. lactuca mixes, the LDA (Linear Discriminant Analysis) Effect Size package (LEfSe, [36]) was used to detect taxa characteristic of digestion at high (R6) or low rates (R1), acting as potential 'biomarkers' for either setup. A complete LDA output for taxonomy is provided as supplementary data in S2 Table.

Taxa characteristic of R1 show a strong affinity for marine environments and/or halotolerance. Additionally, most were originally isolated from marine sources; three from Ulva species or other seaweeds (Maritalea, Arenibacter, Alteromonadaceae). Several are aerobes or facultative aerobes (Nitratireductor, Altermonadaceae) and many show degrees of fermentative and/or acidogenic activity. The most significantly associated taxa (LDA effect ≥4, α ≤0.05) are from the Actinobacteria (Micrococcales), Alpha-Proteobacteria (Devosia, Nitratireductor, Rhizobium and Rhodobacteraceae sp.), Beta-Proteobacteria (Hydrogenophaga and Limnohabitans), Bacteroidetes (Proteiniphilum) and Firmicutes (Alkaliphilus, Bacillales, Lutispora, Syntrophomonadaceae, Tepidanaerobacterales, Tissierella) phyla. As well as known fermenters, acidogens (Proteiniphilum, Firmicutes) and syntrophs (Firmicutes), these taxa suggest diverse saccharide use, and use of alternate electron acceptors (nitrogen, sulfur) detrimental to biogas production (Alpha- and Betaproteobacteria).

Indicators of the R6 environment were more closely linked to anaerobic digestion, but retained some associations with marine and saline habitats. The most significantly associated taxa (LDA effect ≥ 4, α ≤0.05) are more commonly anaerobic and documented as hydrolysers (Alkaliflexus, Caldilineae, Lachnospiraceae, Proteiniphilum, Ruminococcaceae), fermenters (Caldilineae, Desulfomicrobia) and acetogens (Alkaliflexus, BPC102, Caldilinea, Christensenellaceae, Syntrophomonas, etc.), as well as including three Archaea: the acetoclastic Methanosarcina and hydrogenotrophic Methanobacterium and Methanobrevibacterium. Most methanogens were not significant indicators, as abundances were similar between reactors.

Predicted metabolic characteristics.

Attributing reactor performance to specific microbial populations is problematic, partially due to resource-intensive technologies necessary to profile metabolic activity, which may be unsuited to industrially scaled applications (e.g. mRNA/cDNA libraries, metabolic isotope analysis). A novel compromise afforded by metagenomics is to cross-reference taxonomic information (e.g. 16S sequence data) with a database of known metabolic capabilities, and compute inferred metabolic profiles which may help explain activities in a microbial community. Characterisation of functionality through inferred metabolism has been demonstrated in medical, ecological and biofuel contexts: identifying microbial metabolisms likely to improve dietary dysfunction [87]; demonstrating differential microbial activities in healthy and compromised habitats [88]; and predicting and confirming enriched cellulolytic activity in microbial lignocellulose degradation [89]. By highlighting the metabolic capabilities of an inoculum or sludge, the same approach applied to AD has the potential to provide a more informed characterisation of biogas conditions, helping to “de-mystify” the roles of microbial populations. Using the HUMAnN package [35], taxonomic abundances for R1 and R6 were used to infer metabolic processes for the two communities. Predicted features characterising either reactor were then identified using LEfSe [36], with complete metabolic HUMAnN and LDA outputs provided as supplementary data in S3 Table.

Diverse carbohydrate metabolism is likely to characterise R1, with the highest LDA effect scores (4.1–3.9, α = 0.006) for central carbohydrate metabolism and saccharide transport. Although carbohydrates are fundamental to all metabolism, the variety of metabolic pathways represented in these categories suggests that the R1 community utilises a more opportunistic and varied range of carbon sources, with significantly elevated predictions for the Entner-Doudoroff Pathway, Pentose Phosphate Pathway and Citrate Cycle (LDA effects: 3.18–3.42, α<0.05). Predicted markers for R1 also include transport of putrescine and spermidine, key components [90] in the formation and regulation of biofilms (LDA effect: 3.47–3.71, α = 0.006–0.011); and Type VI secretion systems which are likely to be used in competition for resources (LDA effect: 3.7, α = 0.034).

Metabolism of methane is a strong recurring prediction for R6 (LDA effect: 3.53–3.98, p = 0.006) with the emphasis on methanogenesis via methanol and acetate (LDA effect: 3.64 and 3.58 respectively, α = 0.006). However, the strongest predicted characteristics of R6 metabolism are transport of cobalt (LDA effect: 4, α = 0.006) and nickel (LDA effect: 4.2, α = 0.006). Cobalt is required for methylotrophic methanogenesis [57], while nickel is central to the final step of all methanogenic pathways [91,92]. There is good evidence in the literature indicating that methane production increases substantially when nickel and cobalt are added [93–95]. Increased archaeal ribosome metabolism (LDA effect: 3.64, α = 0.006) and reduction of quinones in energy metabolism (LDA effect: 3.52, α<0.02) are also predicted to differentiate metabolism in R6 from R1.

Constrained Correlation Analysis.

Constrained Correlation Analysis (CCA) measured the relationships between community structure and time-points, and metabolism and time-points, in the context of specified ('constraining') process variables. Several process variables were inter-correlated, describing the same source of variation in the dataset. In particular, levels of TAN, alkalinity and total dissolve solids (TDS) were strongly inter-correlated (R = 0.80–0.95), as were B_eff, biogas output and specific methane yield (SMY) (R = 0.81–0.97); and chloride, total salinity, chemical oxygen demand (COD), volatile solids (VS) and duration of trial (R = 0.81–0.97). As such, three governing aspects described the reactor communities: inhibitor accumulation, biogas activity, and trial duration.

CCA of community abundances.

CCA showed that levels of ammonia (specifically total ammoniacal nitrogen, TAN), chloride and raw biogas output had the strongest correlations with community make-up, with the most significant and non-redundant effects on taxonomic abundances (R = 0.50, significant after 999 permutations, VIF< 8). Together, these 3 parameters described 49.8% of variation in community abundance and allowed the major interactions defining these communities to be visualised via bi-plot (Fig 3) showing clear segregation between the two reactors. Although initial community and process similarities cause both Week1 samples to cluster, R1 and R6 time-points diverged along X and Y axes respectively, with clustering of later time-points showing established communities. Despite low OLR in R1, accumulation of TAN exceeded 5,000 mg/L in later time-points, and was the most strongly correlated inhibitor of biogas process (X axis). R6 time-points show negligible interaction with ammonia levels or overloading along the X axis, indicating the R6 community was not inhibited by TAN levels up to 3,000 mg/L. Instead, R6 correlates strongly with increasing biogas output, seen as distribution along the Y axis. Note that Week 13 of R1 correlated with biogas production (movement on Y axis) before R1 reached higher ammonia levels. Rising chloride concentrations correlate with both reactor setups, relating trial duration and a gradual accumulation of dissolved content. A stronger association with R1 is explained through a higher U. lactuca loading, with no clear inhibitory effects.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Levels of ammonia (TAN) and biogas best differentiate microbial communities between the two reactors.

Microbial community structures diverged over time despite initial similarities (lower left quadrant), with R1 communities showing a stronger correlation with levels of ammonia across the X axis and R6 communities showing a stronger correlation with increasing biogas along the Y axis. The perpendicular relationship between biogas and ammonia (total ammoniacal nitrogen, ‘TAN’) suggests the two parameters act on community structure independently. Chloride (‘Cl’) levels show a weaker interaction with community structure, likely reflecting the accumulation of material and maturation of the reactor as the trial progresses.

https://doi.org/10.1371/journal.pone.0142603.g003

Correlations with OLR, reactor alkalinity (Alk) and total ammoniacal nitrogen (TAN) were up to 1.5 times stronger for Archaea, while pH, salinity, COD, VS% and Cl correlated to Bacteria more strongly (1.5–2 times). Curiously, the bacterial community was more than twice as correlated to B_eff as the archaeal community (R: 0.21 v 0.12), reflecting the specialised bacterial community involved in methanogenesis and relatively consistent methanogen components. A negative correlation between biogas output and biodiversity indices (R>-0.6) could potentially be explained through 'niche exclusion', where taxa unsuited to anaerobic digestion are out-competed by “better-equipped” taxa, causing a decrease in diversity. Excluded taxa are known to persist at low abundances and form important reservoirs of metabolic capability, invoked during shifts in reactor conditions [96–98].

CCA of predicted metabolic activities.

CCA using predicted metabolic abundances showed strongest non-redundant correlations with TAN and B_eff (R = 0.50, VIF = 1, significant after 999 permutations). Ordination under these constraints (Fig 4) showed differences in energy metabolism along the X axis, with methanogenesis predictions related to R6 segregating from predicted alternative anaerobic metabolic pathways (Entner-Doudoroff, ethylmalonyl, and pentose-phosphate pathways) and carbon uptake pathways (multi-saccharide transport system) related to R1. Samples differentiated along the Y axis as reactors matured, with earlier metabolic diversity (e.g. sulphate reduction and transport, methane oxidation) absent in later samples as overall diversity decreased. Methanogenesis (acetate and methanol metabolism) and archaeal translation and transcription clearly associated with R6, while negatively correlating with TAN levels. Predictions for nickel and cobalt transport also associate with R6 time-points.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Ammonia levels (TAN) and biomethane conversion efficiency (B_eff) best differentiate predicted metabolisms between R1 and R6.

Carbon metabolisms segregate along the X axis, reflecting divergent environments under the contrasting reactor setups. R1 samples ordinate more closely with diverse carbon metabolism (Entner-Doudoroff: EntDu p/w; Pentose-Phosphate: PentP p/w; ethymalonyl: Emal p/w), while R6 samples ordinate strongly with methanogenic activities (Methanogenesis: AcO, MeOH → CH4; Co-Enzyme M biosynthesis: Co-M b/s)) and the uptake of trace elements (Cobalt: Co t/s; Nickel: Pep-Ni t/s). More diverse activities (Citric Acid Cycle: CitAc Cyc; sulphate reduction: SO4->H; methane oxidation: CH4 Ox) ordinate closer to earlier samples, suggesting metabolic activities detrimental to biogas production were excluded as reactor communities developed. Activities in green represent strongest predicted biomarkers for R6, activities in red represent strongest predicted biomarkers for R1.

https://doi.org/10.1371/journal.pone.0142603.g004