Methanosarcina Play an Important Role in Anaerobic Co-Digestion of the Seaweed Ulva lactuca: Taxonomy and Predicted Metabolism of Functional Microbial Communities

Macro-algae represent an ideal resource of third generation biofuels, but their use necessitates a refinement of commonly used anaerobic digestion processes. In a previous study, contrasting mixes of dairy slurry and the macro-alga Ulva lactuca were anaerobically digested in mesophilic continuously stirred tank reactors for 40 weeks. Higher proportions of U. lactuca in the feedstock led to inhibited digestion and rapid accumulation of volatile fatty acids, requiring a reduced organic loading rate. In this study, 16S pyrosequencing was employed to characterise the microbial communities of both the weakest (R1) and strongest (R6) performing reactors from the previous work as they developed over a 39 and 27-week period respectively. Comparing the reactor communities revealed clear differences in taxonomy, predicted metabolic orientation and mechanisms of inhibition, while constrained canonical analysis (CCA) showed ammonia and biogas yield to be the strongest factors differentiating the two reactor communities. Significant biomarker taxa and predicted metabolic activities were identified for viable and failing anaerobic digestion of U. lactuca. Acetoclastic methanogens were inhibited early in R1 operation, followed by a gradual decline of hydrogenotrophic methanogens. Near-total loss of methanogens led to an accumulation of acetic acid that reduced performance of R1, while a slow decline in biogas yield in R6 could be attributed to inhibition of acetogenic rather than methanogenic activity. The improved performance of R6 is likely to have been as a result of the large Methanosarcina population, which enabled rapid removal of acetic acid, providing favourable conditions for substrate degradation.


Introduction
While primarily a waste-treatment strategy, Anaerobic Digestion (AD) is increasingly being implemented as a viable renewable-energy technology, capable of converting diverse organic substrates into biofuels. In this respect, there is renewed interest in the use of seaweeds (macroalgae) as a substrate for biofuel production [1,2], though some technical problems associated with their use still need to be resolved [3].
In contrast to plants, seaweeds possess lower quantities of recalcitrant structural polymers (e.g. lignin, cellulose, hemi-cellulose), contain large reserves of accessible carbohydrates, and produce biomass via a rapid life cycle. However, they also possess unique compounds. U. lactuca can yield high levels of protein, sulphur and nitrogen; seaweeds typically also contain excess marine salts [4][5][6][7][8]. To improve biogas yields, pre-treatments, co-digestion, and alternative reactor configurations have been investigated for seaweeds [3]. Efficient management of AD via process parameters can also improve biogas yields, as well as helping to avoid toxic shock (e.g. rapid changes in pH, ammonia etc.), accumulation of intermediates (e.g. volatile fatty acids), or over/under-feeding of the reactor (i.e. maintaining an appropriate organic loading rate). However, these parameters provide only indirect information on biological processes within the reactor, and often must be re-evaluated at each new application, restricting informative comparisons and potentially obscuring underlying processes.
Recent reports have highlighted the need for microbial indicators of optimal AD performance as a prerequisite to allow "microbial-based management" of the process [9,10]. Thorough characterisation and a greater understanding of microbial populations and processes "driving" AD can better inform the design and operation of biogas reactors treating macroalgae and other novel feedstocks. Identifying these 'indicators' has been greatly aided by the use of molecular sequencing technologies, allowing metagenomic-based analyses of microbial community structures in various AD systems. These approaches have successfully been employed to monitor the development of AD communities over time [11,12] determine core motifs in AD community structure [13], and determine dominant methanogenic pathways which can be correlated to biogas yield [14]. Previous metagenomic studies on the use of algae as a biogas substrate have identified increases in the archaeal methanogenic order Methanosarcinales under addition of the macro-alga Saccharina latissima [15], the importance of Methanosarcinales in supporting diverse metabolic pathways in AD of the micro-alga Scenedesmus obliquus [16], and the importance of retaining methanogenic Archaea in AD of the macro-alga Laminaria hyperborea [17].
In a previous study, Allen and co-workers approached difficulties in digesting the macroalga Ulva lactuca (sea-lettuce) through co-digestion with the proven and abundant substrate, dairy slurry. Six U. lactuca-slurry feedstock ratios were trialled over a nine-month period, with five of the reactors (R1 through R5) encountering total or partial inhibition through overloading of volatile fatty acids (VFAs), which was dependant on the quantity of U. lactuca supplied [18]. A sixth reactor (R6) saw no immediate inhibition, but instead demonstrated a slow decline in biogas yield, which could not be explained through process variables [18]. Here, we present a microbial analysis of these trials, investigating how AD of U. lactuca shaped archaeal and bacterial populations in the best (R6) and worst (R1) performing reactors, with a particular focus on methanogenic processes. A taxonomic time-series was constructed which illustrates how microbial community structure and activity diverged between R1 and R6, suggesting two explanations for loss of methanogenic activity and a mechanism for Methanosarcina improving reactor stability. Constrained canonical analysis (CCA) revealed the most significant effects of U. lactuca on microbial community structure and on predicted metabolic activity. To our knowledge, this is the first application of 'next-generation' 16S community sequencing to monitor microbial community structures involved in anaerobic digestion of green seaweeds (Chlorophyta).

Biogas reactor configuration
A total of six, 5L one-step continuously stirred-tank reactors (CSTRs) were operated in parallel digesting mixes of Ulva lactuca and dairy slurry for a period up to 42 weeks at a constant temperature of 37°C. Three reactors treated dried U. lactuca in co-digestion mixes of 25, 50 and 75% with dairy slurry. A further 3 reactors co-digested fresh U. lactuca with slurry in the same ratios. Regular feeding and removal of substrate allowed a constant 4 L working volume, with an initial organic loading rate (OLR) of 2 kg VS m 3 d -1 . Of the 6 reactors, 3 failed to obtain steady state biogas production, 2 achieved steady state production profiles but incurred high levels of VFA-based inhibition, while the final reactor achieved satisfactory yields. Inhibition was characterised by variable levels of VFA and biogas yield, and an inability to maintain high rates of substrate input. Reactors were operated in the configuration represented in Fig 1. Previous work [4] assessing the optimal bio-methane potentials (BMP) for U. lactuca/slurry feedstocks allowed evaluation of reactor output.
Reactor 1 (R1: digesting 75% dried U. lactuca, 25% dairy slurry) provided the longest running example of U. lactuca-inhibited digestion, while Reactor 6 (R6: digesting 25% fresh U. lactuca, 75% dairy slurry) was the best performing reactor, with stable VFA concentrations and favourable yields at an OLR of 2.5 kg VS m 3 d -1 . R1 and R6 were subsequently chosen as best and worst case examples of U. lactuca co-digestion. Reactor R1 was operated for a total of 40 weeks. Initially an OLR of 2 kg VS m 3 d -1 was used for R1, however failure to reach the designated yields after the first hydraulic retention time (HRT) and the increase in VFA concentration resulted in the OLR being reduced to 1 kg VS m 3 d -1 , with subsequent steady-state biogas production being achieved. R6 was also operated for 40 weeks. An OLR of 2 kg VS m 3 d -1 was successfully maintained for R6 after a period of 3 HRTs, with OLR then being increased to 2.5 kg VS m 3 d -1 . Steady state biogas production was achieved throughout this period. A gradual decline was observed in the final HRT for R6 without a corresponding increase in VFA or ammonia concentrations accounting for this reduction [18]. The decision to increase OLR was determined by two factors: the relationship between VFA concentrations and reactor performance, and the biomethane conversion efficiency (B eff ). The effect of VFAs was determined using the Nordmann method [19] commonly known as the FOS:TAC ratio, measuring volatile organic acids and total inorganic carbonate. Operational ranges set out by this method dictate whether the reactor is being over, under or fed satisfactorily. The biomethane conversion efficiency (B eff ) is the specific methane yield (SMY) of that reactor in continuous digestion divided by the biochemical methane potential (BMP) yield obtained from a 30 day batch test on that exact substrate. Values closer to or higher than 1 are desirable, reflecting optimum conversion of feedstock to biogas. A comprehensive detailing of the laboratory methods used to analyse all the environmental parameters within R1 and R6 has been previously described [18].

Sampling and Molecular Methods
Reactor sludges were sampled on a weekly basis, and frozen at -80°C until further analysis. For R1, weeks 1, 5, 13, 20, 30 and 39 were selected as representative time-points, spanning five retention times. For R6, weeks 1, 5, 13, 21 and 27 were selected as time-points, spanning four retention times. Sludge from these 11 time-points was processed with the PowerSoil DNA extraction kit (MoBio, CA, USA) with the following protocol modifications: 1) initial 'wet-spin' (30 seconds at 10,000 g) to remove an excess liquid fraction prior to cell lysis; 2) 3x cycles of 10 minute bead-beating followed by 5 minutes chilling at -20°C; 3) 2x washes of elution buffer. For each time-point, triplicate sludge-samples were taken from each reactor. From each of these, three separate DNA extractions were performed, and then combined in equimolar quantities to ensure representative sampling. Extractions were quantified spectrophotometerically (ND-1000, Thermo-Fisher, DE, USA) and viewed on 1% agarose gel with ethidium bromide (1μg/ml).
16S gene sequences were amplified from the DNA extracts using 11 pyrosequencing PCR primers with the following motifs: adapter sequence (Roche-454 Lib-A and Lib-B chemistry); key sequence (TCAG); Roche-454 pyrosequencing MIDs 1-10 and 12 inclusive; and 16S universal primers U-789F (5' TAGATACCCSSGTAGTCC 3') and U-1053R (5' CTGACGRCRGC Methanosarcina in Anaerobic Co-Digestion of Ulva lactuca CATGC 3') [20]. A program of initial denaturation at 94°C for 5 minutes, followed by 26 cycles of 30 seconds denaturing at 95°C, 30 seconds annealing at 53°C, and 45 seconds of extension at 72°C, with a final extension step of 72°C held for 6 minutes was followed. Products in the expected size range were extracted using a gel extraction kit (QIAGEN, Manchester, UK), which required subsequent use of a PCR purification kit (QIAGEN, Manchester, UK). Each DNA extract was amplified in triplicate, then pooled in equimolar quantities to produce 11 community samples, which were then pyrosequenced by MACROGEN (Seoul, Republic of Korea).

Bioinformatic Analysis
Denoising was performed in Acacia [21] before import into the Quantitative Insights Into Microbial Ecology (QIIME) software pipeline [22] for de-mulitplexing, chimera removal, aligning, taxonomic assignment and exploratory analyses. Sequences were split into sample libraries; Chimera filtering was carried out using USEARCH v6.1 [23]; Alignments and taxonomic assignments were carried out with reference to the Silva 111 Database release [24] at 97% similarity using PyNast [25] and the RDP Classifier 2.2 [26]; Tree building was carried out using FastTree [27]. Beta diversity was calculated using UniFrac [28] and 3D PCoA plots generated by Emperor [29].
Sequence data was combined with reactor process data from [18] within the R statistics program [30]. R packages vegan [31] and phyloseq [32] were used to subset population abundances by sample and/or reactor environment, and to perform statistical analysis.
Greengenes release 13.5 [33] was used to perform closed-reference OTU picking in QIIME prior to generating metabolic predictions from the Kyoto Encyclopedia of Genes and Genomes (KEGG; release 73.1 [34]) with the HMP Unified Metabolic Analysis Network (HUMAnN) [35] package. Significant differences between the two reactors were calculated using the LDA Effect Size (LEfSe) resource [36] on the Huttenhower Galaxy resource [37][38][39] to analyse taxonomic and predicted metabolic data. To reduce spurious inferences on metabolic activity, a more conservative LDA threshold of 3 was used.
Sequence data was deposited in the MG-RAST database under project number 14106, and is publicly available at the URL http://metagenomics.anl.gov/linkin.cgi?project=14106.
It should be noted that although primers used in this study [20] continue to see use in similar investigations [40][41][42], primers are continuously refined to increase coverage as observed microbial diversity expands. Similarly, methodologies that minimise bias [43], and reference databases with improved taxonomic and metabolic representation continue to be developed (e.g. Silva, KEGG). As such, the characterisation of communities in this study is necessarily incomplete and likely to contain errors at lower limits of taxonomic resolution-metabolic characterisation in particular is still in its infancy, with prediction best employed as an exploratory or complementary analysis. Improved, robust characterisations of AD community members are anticipated from future studies, employing updated biological data and methodologies.

Results and Discussion
A previous study trialled continuous anaerobic digestion of varying ratios of Ulva lactuca and dairy slurry, demonstrated severely inhibited biogas production at higher U. lactuca loading levels [18]. To determine potential causes of this inhibition, the microbial community profiles of two reactors digesting contrasting ratios of U. lactuca and dairy slurry were characterised and compared, with the overall aim of identifying significant 'biomarker' species or metabolic activities which differentiated successful and inhibited digestion of U. lactuca. Detailed accounts of reactor setup and performance have been provided by [18].

Process Inhibitors
Volatile Fatty Acids. VFA accumulation can occur as a product of instability [44], can be transitional [45][46][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 (NH 3 ) or the ammonium ion (NH 4 + ). Elevated pH, temperature, and headspace partial pressure increase concentration of the uncharged NH 3 state. At sufficiently high concentrations NH 3 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][51][52][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 + , Ca 2+ , Mg 2+ , 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 Mg 2+ , Na + and Ca 2+ respectively. Monitored levels of Clinfer 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 Ca 2+ 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.

Community Composition
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.

Community Makeup
The QIIME pipeline identified 2,824 Operational Taxonomic Units (OTUs) in the 89,251 sequence reads. Singleton and doubleton OTUs (abundances < 3 reads) were discarded to reduce statistical noise, leaving 1,320 OTUs (82,914 sequence reads). Of the 1,320 OTUs, 1,057 were present in R1 and 955 in R6. Taxonomic alignments provided by Silva (release 111) identified 2 phyla, at least 4 classes, 5 orders, 7 families and 8 genera of Archaea (20 OTUs, 9,010 sequences), and at least 34 phyla/candidate phyla, 44 classes, 86 orders, 124 families and 190 unique genera of Bacteria (1,206 OTUs, 73,185 sequence reads). Lower taxonomic classifications could not be assigned to 16% of Bacteria families and 53% of Bacteria genera. A final 94 OTUs remained unidentified and were not assigned to Bacteria or Archaea. Unassigned taxa comprised 1% of sequence reads (72 OTUs) from R1, and <1% of reads (42 OTUs) from R6. A complete description of community abundances is provided as supplementary data in S1 Table. 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][53][54][55].
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 petrochemicalcontaminated 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 m 3 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.

Relating Community Makeup and Process Variables
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 m 3 d -1 in Week 7, while Cllevels 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.
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.
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. Cllevels 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 m 3 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.
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).
Constrained Correlation Analysis. Constrained Correlation Analysis (CCA) measured the relationships between community structure and time-points, and metabolism and timepoints, 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.
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 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.
doi:10.1371/journal.pone.0142603.g004 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 outcompeted 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][97][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.

Conclusion
Anaerobic digestion of U. lactuca appears to indirectly inhibit acetogenic and methanogenic processes, with ammonia showing the strongest causative correlation. At high U. lactuca volumes, decreasing OLR was not sufficient to recover the acetoclastic methanogens required to remove acetic acid and prevent overloading, nor to retain hydrogenotrophic methanogens. At lower U. lactuca volumes, the inhibition of acetogenesis caused Methanosarcina populations yields to shrink, affecting overall biogas yield. U. lactuca loading significantly affected community composition within reactors, with higher volumes characterised by diverse, facultatively anaerobic, and marine and halotolerant taxa, a lack of methanogens, and a predicted reliance on alternative carbon metabolism. The Shannon Index is sensitive to the major community members, the Chao1 Index is more sensitive to diversity of rare species. Rarefaction curves indicate the major community members are well-characterised, but a large reservoir of low-abundance taxa remains undocumented. (TIFF)   [36]) for Ulva Digestion. Taxonomic abundances are related to either R1 or R6: the effectiveness of each taxon as a marker for either state is determined via its LDA effect size, as calculated in LEfSe [36], along with the significance of that effect (alpha value). A conservative LDA effect cut-off point of 3 used. (XLSX) S3 Table. Predicted Metabolic Relative Abundances (via HUMAnN Package), and statistically significant Linear Discriminant Analysis (LDA) markers (via LEfSe). Metabolic abundances are inferred from taxonomic relative abundances via KEGG annotations [35], providing expected metabolic activities. The effectiveness of each metabolism as a marker is determined via its LDA effect size, as calculated in LEfSe [36], along with the significance of that effect (alpha value). A conservative LDA effect cut-off point of 3 used. (XLSX)