DNA extractions were performed for replicate peat samples from the field, incubation setup, and days 0, 25, and 50 of the incubations, as described by Mondav et al. (2014) and Woodcroft et al. (2018). Briefly, total nucleic acids were extracted from ∼2 g peat sample using the PowerMax Total Nucleic Acid extraction kit (MoBio), retaining the LifeGuard preservation solution during lysis. DNA was purified by RNaseA digestion, phenol-chloroform-isoamyl alcohol purified and ethanol precipitated. Approximately 15 ng DNA from each sample was used as a template in PCR reactions. Libraries for low concentration DNA samples were created using 1 ng of DNA with the Nextera XT DNA Sample Preparation Kit.

To measure microbial cell abundances, based on 16S rRNA gene copies, quantitative polymerase chain reaction (qPCR) was conducted for each community. Each reaction contained 5 μl of 2X SYBR Green PCR Master Mix (Applied Biosystems, Carlsbad, CA, United States), 4 μl of template DNA, and 1 μl of primer mix. Bacterial and archaeal 16S rRNA genes were amplified using the primer set 1406F/1525R (0.4 μM, F - GYACWCACCGCCCGT and R - AAGGAGGTGWTCCARCC). For inhibition control samples, the rpsL primer pair (0.2 μM, F - GTAAAGTATGCCGTGTTCGT and R - AGCCTGCTTACGGTCTTTA) was used to amplify Escherichia coli DH10B only. Each sample and standard were run in triplicate, with three dilutions (1/100, 1/500, and 1/1000) and an inhibition control (1/100 dilution of E. coli DH10B genomic DNA spiked into a 1/100 dilution of the sample). E. coli DH10B genomic DNA dilutions of 10⁻², 10⁻³, 10⁻⁴, 10⁻⁵, and 10⁻⁶ of the 20 ng/μl stock solution were used for the standards. The ViiA7 Real-Time PCR System (Applied Biosystems, Carlsbad, CA, United States) was used for the qPCRs, with the following cycling conditions: 10 min at 95°C, 40 cycles of [15 s at 95°C, then 20 s at 55°C, then 30 s at 72°C]. To produce a melt curve, a cycle of 2 min at 95°C was run, and then a final cycle of 15 s at 60°C. The cycle threshold (Ct) values were recorded and analyzed using ViiA7 v1.2 software, and 16S rRNA gene copy numbers were calculated for each sample. These calculations accounted for the genome size (4,686,137 bp) and 16S rRNA gene copy number (7) of the standard.

The 16S rRNA gene encompassing the V6–V8 regions was targeted using the 926F (5′-AAACTYAAAKGAATTGRCGG-3′) and 1392R (5′-ACGGGCGGTGWGTRC-3′) primers (Engelbrektson et al., 2010) modified to contain Illumina specific adapter sequence (926F:5′TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGAAACTYAAAKGAATTGRCGG3′ and 1392wR:5′GTCTCGTGGGCTCGGGTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGACGGGCGGTGWGTRC3′). The universal primer pair Univ_SSU_926F-1392wR amplifies the small subunit (SSU) ribosomal RNA of prokaryotes (16S), specifically the V6, V7, and V8 regions. In E. coli, it amplifies the 926–1392 region of the 16S gene.

Preparation of the 16S library was performed as described, using the workflow outlined by Illumina (#15044223 Rev.B). In the 1st stage, PCR products of ∼466 bp were amplified according to the specified workflow with an alteration in polymerase used to substitute Q5 Hot Start High-Fidelity 2X Master Mix (New England Biolabs) in standard PCR conditions. Resulting PCR amplicons were purified using Agencourt AMPure XP beads (Beckman Coulter). Purified DNA was indexed with unique 8 bp barcodes using the Illumina Nextera XT 384 sample Index Kit A-D (Illumina FC-131-1002) in standard PCR conditions with Q5 Hot Start High-Fidelity 2X Master Mix. Indexed amplicons were pooled together in equimolar concentrations and sequenced on MiSeq Sequencing System (Illumina) using paired end sequencing with V3 300 bp chemistry in the Australian Centre for Ecogenomics according to manufacturer’s protocol.

16S amplicons were processed using the Australian Center for Ecogenomics (ACE) mitag pipeline (AMP). The first 20 bases of all fastq files are then trimmed to remove primer sequence, and quality trimmed to remove poor quality sequence using a sliding window of 4 bases with an average base quality above 15 using the software Trimmomatic. All reads are then hard trimmed to 250 bases, and any with less than 250 bases excluded. Fastq files are then converted to fasta files. Fasta files are processed using QIIME’s pick_open_reference_otus.py workflow with default parameters (97% similarity) and taxonomy assignment and alignment features suppressed. The resulting OTU table is filtered to remove any OTU with an abundance of less than 0.005%. Representative OTU sequences are then BLASTed against the reference database (Greengenes version 2013/05 for 16S).

In order to correct the amplicon-based relative abundances for variance in genome copy number of the 16S gene between different lineages, both the OTU data and qPCR data were also processed by CopyRighter (Angly et al., 2014).

From the 16S rRNA community profiles, lineages known to perform hydrogenotrophic methanogenesis, acetoclastic methanogenesis, and anaerobic CH₄ oxidation were manually identified through literature review and included those identified at this site by previous molecular characterization of Stordalen Mire microbial communities (Mondav et al., 2017; Woodcroft et al., 2018). The copy-corrected relative abundances of these lineages were calculated within each microbial community. Aerobic CH₄ oxidizing lineages (families Methylocystaceae, Methylococcaceae, Beijerinckiaceae, and Hyphomicrobiaceae) detected at Stordalen Mire by Singleton et al. (2018) were also searched for in the incubation communities. Since both methanogenesis and CH₄ oxidation are fairly taxonomically conserved, inferring function based on 16S rRNA-inferred taxonomy is relatively robust. Acetogens, on the other hand, are phylogenetically diverse, spanning at least 20 genera (Drake et al., 2008). Since homoacetogenesis is not a monophyletic trait, a curated list was developed of demonstrated homoacetogenic lineages that have been isolated, cultured, and experimentally observed to form acetate from CO₂ and H₂ (Supplementary Table S1). Putative homoacetogens in this work are lineages affiliated at either the genus or family level with experimentally verified homoacetogenic lineages. One caveat with this analysis is that some sister taxa diverge in their laboratory ability to perform homoacetogenesis, so inferring function based solely on 16S data is less robust than inferences about CH₄ production and oxidation.

The variant 2 reaction network estimated that methanogenesis in the collapsed palsa was dominated by hydrogenotrophy (i.e., acetoclasty:hydrogenotrophy < 1 in Figure 5) until the last time point in the 8–18 cm collapsed palsa incubations (Figure 5). In agreement, the microbial methanogen community at this site was dominated by hydrogenotrophs in both shallow and deep samples, to the near exclusion of acetoclasts; this was observed at all time points examined (days 0, 25, and 50) (Figure 4).

Of the three studied habitats, the collapsed palsa had the greatest abundance of putative homoacetogens (Figure 6). Estimated rates of homoacetogenesis were higher in the collapsed palsa than in the other sites over the first 25 days of the incubations; rates dropped sharply after day 25 (Figure 4) as signaled by the inflection point in the measured rates and isotope ratios (Figure 3). At the same time, the estimated ratio of hydrogenotrophy:homoacetogenesis in the collapsed palsa increased over time approaching 1:1 (Figure 5) by day 50 in the 8–18 cm depth. On day 25, the rates of acetoclasty, hydrogenotrophy, and homoacetogenesis all decreased in this incubation (Figure 4), apparently signifying that as less acetate was produced from homoacetogenesis, less acetate was available to be consumed by acetoclasty. The microbial data indicated that this shift was accompanied by an increase in hydrogenotrophic methanogens (Figure 4).

Over the first 20 days of the incubations, the model estimated that non-fractioned CO₂ production was highest in the collapsed palsa incubation compared to incubations of material from the other sites, but this rate declined over time (Figure 4). In the beginning of the collapsed palsa incubations, non-fractionated CO₂ dominated total CO₂ production, but over time, the relative proportion of non-fractionated CO₂ declined leading to an increase in the proportion of methanogenic CO₂ for both depths (Figure 4). Concomitantly, the observed and modeled CO₂:CH₄ ratio declined over time consistent with the system becoming more methanogenic (Figure 8). The change in modeled CO₂:CH₄ ratios over time was greatest in the collapsed palsa incubations (>45%) compared to other sites (Figure 8).

In the bog, estimated rates of acetoclasty decreased over time (until the last time point in the deep bog), while estimated rates of hydrogenotrophic methanogenesis increased over time (Figure 4). The relative abundance of acetoclasts in the bog did not reflect the trend in acetoclasty rates (Figure 4). A low level of acetoclasts were present in the bog incubations, ranging from <5% of methanogens in the deeper section to 8–21% in shallower bog samples (Figure 6). Thus, in this incubation, acetoclast abundance did not reflect the modeled rates of acetoclasty. For hydrogenotrophy, however, which increased over the course of the incubations there was a corresponding increase in the abundance of hydrogenotrophs (Figure 4). Indicating a good agreement between hydrogenotroph abundance and rates of hydrogenotrophic methanogenesis. The ratio of acetoclasty:hydrogenotrophy in the bog declined over time and approached 1 near the end of the incubations. At the same time, the overall the ratio of acetoclast to hydrogenotroph abundances declined from days 0 to 50 (Figure 5). The deep bog samples had lower proportions of acetoclasts:hydrogenotrophs than the shallow bog samples, though it was not clear that the deep bog had correspondingly lower acetoclasty:hydrogenotrophy rates (Figure 5).

Homoacetogenesis rates were lower in the shallow bog on day 0 than in any of the other sites, but the estimated rate increased over time in both the shallow and deep bog incubations achieving the highest homoacetogenesis rate of any of the sites by day 50 (Figure 4). Consistent with this finding, the relative abundance of genus-level homoacetogens in the bog incubations also increased over time (Figure 4). Paucity of data precluded a statistical analysis of this correlation but there was a visual correlation between homoacetogen abundances and modeled rate distributions (Figure 4).

Non-fractionating CO₂ production was low in the bog incubations compared to the collapsed-palsa incubations and was near zero throughout the deeper bog incubations (Figure 4). But, over the course of the shallow bog incubations, rates of non-fractionating CO₂ production increased and acetoclastic methanogenesis rates decreased such that the proportion of CO₂ produced by non-fractionating mechanisms increased over time (Figure 4). This is the only incubation where there was an increase in non-fractionating CO₂ production. In all other incubations the proportion of CO₂ produced by non-fractionating mechanisms was relatively stable or declined over time (Figure 4). Despite this increase in non-fractionated CO₂ production, the ratio of CO₂:CH₄ production in incubations with material from both bog depths was relatively stable over time, although it was higher in the shallow bog incubation compared to the deep bog incubation (Figure 8) indicating that the increase in non-fractionating CO₂ production was offset by a decrease in methanogenesis rates.

In the shallow bog archaeal community, the hydrogenotrophic methanogens increased in abundance over time and the acetoclastic methanogens decrease (Figure 4) consistent with the overall modeled decline in acetoclasty:hydrogentrophy over time (Figure 5). The increasing rates of homoacetogenesis over time in the bog incubations was in agreement with the increasing relative abundance of homoacetogens observed over time in the bog incubations (Figure 4).

The fen incubation had the highest estimated rate of methanogenesis of any of the sites (Figure 4). In the fen, even more than in the bog, acetoclastic methanogenesis dominated over hydrogenotrophic methanogenesis (Figure 6) although hydrogenotrophy increased over time (Figure 5). Concurrent with the rates of acetoclasty, the fen had the highest proportion of acetoclastic methanogens of any of the sites; yet the proportion of acetoclasts:hydrogenotrophs decreased during incubation from >0.20 on day 0 to just 0.12 on day 50 (Figure 5). This decrease in acetoclasts coincided with an overall decline in the estimated ratio of acetoclasty:hydrogenotrophy rates over time (Figure 5).

Estimated rates of homoacetogenesis in the fen incubation were similar to the other sites and remained fairly constant over time (Figure 4). There was no increase in homoacetogen abundance in the fen incubations over time. While the abundance of homoacetogens was highest on day 25, the relative abundance declined over time such that over the 50 days of incubation, homoacetogen abundances stayed constant or declined only slightly (Figure 4). The ratio of non-fractionating CO₂ in the fen incubation, and consequently the CO₂:CH₄ ratio, was the lowest of any incubation suggesting the fen was the most methanogenic of the sites (Figure 4, 8). The ratio of CO₂:CH₄ produced in the fen incubations declined slightly over time from 1.25 to approaching the predicted 1:1 ratio consistent with a methanogenic steady-state (Figure 8).

Peat in permafrost settings is elevated into palsas above the surrounding landscape by ice expansion and accretion. When the supporting ice thaws and drains, the palsa collapses and frequently floods creating a heavily inundated thaw feature. This collapse results in an abrupt transition from elevated aerobic conditions in the surface of the palsa, to depressed inundated, anaerobic conditions in the resulting surface peat. Thus, permafrost collapse features may best be understood in the context of other early stage organic-rich organic environments (such as landfills) (Tchobanoglous and Kreith, 2002). In our collapsed palsa site, acetate concentrations were the highest of any of the thaw features (Table 2). This may reflect an early acid-phase as is seen in other early stage anaerobic environments such as covered landfills or lake sediments (Burdige, 2006). The identification of these processes as “acid phase” is supported by the increasing rates of both acetoclastic and hydrogenotrophic methanogenesis over time in the collapsed palsa incubations after day 20 (Figure 4) and the decline in CO₂:CH₄ production rates (Figure 8) suggesting that the system was becoming increasingly methanogenic as the incubations progressed, possibly due to the depletion of microbially available organic polysaccharides and the resulting build-up of acetate. Concomitantly, excess non-fractionated CO₂ (e.g., heterotrophic respiration products) decreased over time suggesting declining fermentation rates.

Initially, homoacetogenesis rates in the collapsed palsa were greater than hydrogenotrophic methanogenesis rates, but this ratio declined over time, until the last time point when hydrogenotrophic methanogenesis rates were greater than homoacetogenesis in the shallow collapsed palsa (Figure 4). This finding was similar to results observed by Ye et al. (2014) for other peatlands and was interpreted to indicate direct competition for H₂ between homoacetogens and hydrogenotrophs. Our results differ from Ye et al. (2014) in that, while hydrogenotrophy became increasingly important relative to homoacetogenesis over time, the absolute rates of both processes also increased over time after day 20. The high homoacetogenesis rates in the collapsed palsa incubation may suggest that the partial pressure of H₂ was elevated (Ye et al., 2014) allowing homoacetogens to effectively compete with hydrogenotrophic methanogens for substrate. If this dynamic is reflective of that occurring in the field at the collapsed palsa site, such elevated H₂ could contribute to the high levels of acetate found at this habitat (Table 2), and also suggests that something may be inhibiting acetoclastic methanogenesis which was very low compared to the other habitat types (Figure 4). Ye et al. (2014) suggested that declining fermentation rates resulted in a decrease in both acetate and H₂ production which limited both homoacetogenesis and hydrogenotrophic methanogenesis. The increase in the rates of both reactions as well as the increase in homoacetogens throughout our incubations, however, cannot be easily reconciled with this interpretation. Instead, we infer, that the declining excess of non-fractionated CO₂ over time in our incubations supported increasing rates of hydrogenotrophic methanogenesis, which also drew down excess H₂ concentrations. At some point (i.e., approximately 35 days) the H₂ concentrations were low enough that hydrogenotrophy outcompeted homoacetogenesis rates and after that shift, the peats became increasingly methanogenic.

Microbial abundances and productions rates do not always correlate strongly (e.g., collapsed palsa 28–38 cm Figure 4). This may reflect that the microbial community is not comprehensively and exhaustively characterized. For example, large portions of the microbial community have not been isolated in pure culture in the laboratory and their function is unknown, or microorganisms which are classified as homoacetogens may also be able to generate methane acetoclastically even if this activity has not been documented under laboratory culture conditions. Additionally, some microorganisms may alter their activity depending on environmental conditions such as homoacetogens which reverse the reaction to produce CO₂ under certain thereby actually decreasing acetogenesis rates (Pierce et al., 2008). Taken on their own, neither the geochemical data nor the microbial data is capable of fully explaining the system, when considered as complementary data, however, a clearer picture of CO₂ and CH₄ production dynamics emerges.

At Stordalen Mire, the intermediate stage of peatland permafrost thaw is characterized by a Sphagnum-dominated bog feature. This site is elevated relative to the collapsed palsa but still below intact palsa such that the surface of the peat is intermittently inundated. The surface peat at these sites is influenced by the oscillating water table that alters O₂ availability over time, and the release of sphagnum acid from the moss that reduces the pH and inhibits microbial activity. In contrast to the collapsed palsa, the estimated rates of acetoclastic methanogenesis in the bog were higher than that for hydrogenotrophic methanogenesis at all time points. This result is in contrast to what was estimated in the field from isotopic mass balance of chamber flux measurements (McCalley et al., 2014), wherein McCalley et al. (2014) concluded that the bog was dominated by hydrogenotrophic methanogenesis. The difference between our calculations and those by McCalley et al. (2014) possibly occurred because we included fractionation from homoacetogenesis in our rate modeling which McCalley et al. (2014) did not. Consistent with this is the higher acetoclasty:hydrogenotrophy ratios in our model variants which included homoacetogenesis in the bog 8-18 cm incubation (Figure 5). However, the ratio of acetoclasty to hydrogenotrophy in the bog incubation was always lower than in the fen incubation, consistent with field findings that hydrogenotrophy was relatively more important in the bog than in the fen (Figure 5). However, in our study, the estimated acetoclasty to hydrogenotrophy ratio in the bog incubation declined over time as did the ratio of acetoclastic to hydrogenotrophic methanogens suggesting that the bog incubations were becoming more hydrogenotrophic over time. The shift from acetoclastic to more hydrogenotrophic methanogenesis is consistent with thermodynamic considerations that acetoclastic methanogenesis is more energy yielding than hydrogenotrophic methanogenesis (at standard conditions, Conrad, 1999) thus rates of hydrogenotrophy increase after more energetic pathways have been exhausted. Measured acetate levels in the bog are low: 5–9 μM (Table 2), which is at the lower limit of what is considered the threshold for acetate uptake even for the acetoclastic methanogens in Methanosaeta (Griffin et al., 1998), thus we suggest that as the incubations progressed, the organic matter became more decomposed, acetoclasty became substrate limited and diagenesis proceeded through the next energetically favorable reaction (i.e., CO₂ reduction).

In addition to substrate limitation, the declining ratios of acetoclasty:hydrogenotrophy and corresponding declines in acetoclast:hydrogenotroph abundance in the bog (Figure 5, 6) could indicate inhibition of acetoclastic methanogens in the bog. For example, the acetoclastic methanogen genus Methanosaeta (which was least abundant in the bog) is inhibited by ammonia, high concentrations of VFAs (Karakashev et al., 2006), and glycine (Sherwood, 1972), the latter thought to occur by the competitive uptake of Ca²⁺ (Sherwood, 1972). Ammonia concentrations are higher in the bog than in the fen, while Ca²⁺ concentrations are lower (Table 2), possibly indicating inhibition of Methanosaeta in the bog relative to the other sites.

Fens are typically considered the most mature thaw features at Stordalen Mire (Hodgkins et al., 2014). They are characterized by high water table and the predominance of sedge vegetation. Sedges are important conduits for conducting gasses between anaerobic peat depths and the atmosphere. They can transport O₂ to otherwise oxygen-depleted depths possibly promoting CH₄ oxidation and allowing CH₄ and CO₂ to transit directly to the atmosphere (e.g., Chanton, 1991; Whiting and Chanton, 1993; Bhullar et al., 2013). The closed incubations that we used in our study, however, prevent CH₄ loss and so the influence of vegetation as a conduit was controlled for in this experiment. However, differences in the microbial availability of organic matter deposited by sedge vegetation versus Sphagnum mosses still remains. Estimated rates of CH₄ production were highest in the fen incubations compared to the other habitat types, consistent with previous incubation studies (Hodgkins et al., 2014, 2015) and field observations of habitat emissions (McCalley et al., 2014). While the modeled rates of hydrogenotrophic methanogenesis in the fen were higher than the other sites, the fen incubations also had much higher rates of acetoclasty than the other sites (3–4 times higher during some phases of the incubations), such that the acetoclasty:hydrogenotrophy ratio was highest in the fen of all the sites (Figure 5). This is consistent with observations from other peatlands that the more biologically available organic matter in fens stimulates acetoclastic methanogenesis (Chanton et al., 2008).

Metagenomic analyses of the microbial carbon degradation capacity in the field at this same site, has indicated that not only does the fen have 2–3 times more microbial biomass per g of peat relative to the other habitats, but that the fen also has the most diverse range of microorganisms capable of degrading polysaccharides (the first step in cellulose decomposition) (Woodcroft et al., 2018). While there are still more hydrogenotrophs than acetoclasts in the fen, the fen does have the highest abundance of acetoclasts of all the sites (Figure 6). One possible explanation for the apparent discrepancy between the microbial abundances and the modeled rates is that all microbial acetoclasts and hydrogenotrophs may not yet be comprehensively and exhaustively characterized. There may exist microorganisms accomplishing either of these reactions that have not yet been discovered or microorganisms that have been characterized as hydrogenotrophic may also be able to produce methane acetoclastically, but that function has not yet been documented. However, there have been previous extensive metagenomic studies conducted at our field site from which metagenome-assembled genomes for methanogens, including deeply branching lineages were assembled (Supplementary Table S2; Mondav et al., 2014; Woodcroft et al., 2018). Given this extensive characterization, it is unlikely that there are large numbers of previously undescribed acetoclasts at the site. Alternatively, the mismatch between microbial abundances and reaction rates may have a thermodynamic explanation. Under the non-standard conditions of our experiment, it is possible that high CO₂, H₂ and or low acetate concentrations (e.g., see bog and fen concentrations Table 2) may create conditions under which hydrogentrophic methanogenesis actually becomes more energetically favorable than acetoclasty. While, at standard conditions, acetoclasty is a higher energy yielding reaction than hydrogenotrophy (on a mol per mol basis), at high CO₂, H₂, and/or low acetate concentrations (e.g., see bog and fen Table 2) hydrogenotrophy can become more energy yielding thereby favoring the proliferation of hydrogenotrophic methanogens and forcing acetoclasts to produce CH₄ at higher rates to fulfill the same energy demands. Additionally, methanogens may simply be poorly characterized and some putative hydrogenotrophs may actually be capable of disproportionating acetate as well. For example, Methanosarcina are capable of acetoclastic methanogenesis, but they can also produce CH₄ via hydrogenotrophy. On the other hand, Methanosaeta are considered obligate acetoclasts (Karakashev et al., 2006) and their exclusive presence in the fen is consistent with the high rates of acetoclasty estimated for the fen incubation relative to the others incubations (Figure 4).

Although hydrogenotrophic CH₄ production was once observed in a culture of Methanosaeta via direct interspecies electron transfer from Geobacter metallireducens to Methanosaeta harundinacea, which allowed the reduction of CO₂ (Rotaru et al., 2014), the Rotaru et al. (2014) study is the only observation of such a process, and in our incubations, there was a near-zero abundance of Geobacter spp. Methanosaeta have lower threshold acetate requirements than Methanosarcina (5–20 μM vs. 1 mM) and are only known to outcompete Methanosarcina at low acetate concentrations (Griffin et al., 1998). Thus, the low acetate concentrations in the fen (11.0 ± 2.8 μM) may have allowed Methanosaeta to compete under the conditions of the fen habitat. The estimated ratio of hydrogenotrophy to homoacetogenesis was consistently high in the fen incubation suggesting that H₂ concentrations were low. Excess unfractionated CO₂ concentrations were also low, and CO₂:CH₄ production ratios approached 1, suggesting that the system was tending toward a methanogenic steady-state. In this way, the fen most closely approximates the theoretical thermodynamic limitation for TEA-depleted anaerobic cellulose degradation. Cumulatively our results suggest that the organic matter in the fen was the most microbially available of the three sites.

There is evidence in the incubations that the collapsed palsa is not at a theoretical methanogenic steady-state (i.e., CO₂:CH₄ did not = 1). The high CO₂:CH₄ ratio in this habitat (Figure 8) suggests that the rate of fermentation is in excess of methanogenesis. High rates of fermentation without subsequent methanogenesis would lead to increased H_2(aq) which in turn would favor homoacetogens (Weimer, 1998; Drake et al., 2006; Ye et al., 2014) explaining the high rates of homoacetogenesis in this habitat type. However, the rates shifted over time such that by the end of the incubations, hydrogenotrophic methanogenesis started to dominate over homoacetogenesis (Figure 5). These results suggest that initially there were organic electron acceptors present in the collapsed palsa incubations that got used up over the course of the incubations. As respiration rates declined, the rates of both methanogenesis pathways gradually increase in the incubations. This trend is consistent with an exhaustion (titration) of organic electron acceptors and a switch to a more methanogenic system. Without a mechanism for re-oxidation, eventually all terminal electron acceptors in a system will become exhausted—except for CO₂ which is produced during anaerobic fermentation. Thus, any reaction pathway in these closed, anaerobic incubations other than CO₂ reduction is ultimately unsustainable and the system will tend toward methanogenesis. Once all of the other terminal electron acceptors have been reduced, methanogenesis rates must balance fermentation, because fermentation is self-limiting, i.e., the build-up of fermentation products thermodynamically inhibits further fermentation from occurring. In this way, the incubations of the collapsed palsa were reflecting the transition from an acid-phase-dominated (production of electron-rich hydrocarbons, e.g., butyrate, propionate, etc.) environment into the early-stage methanogenic conditions that will dominate the subsequent thaw features.

In the shallow bog incubation, there was always more CO₂ produced than could be accounted for by methanogenesis alone (i.e., CO₂:CH₄ > 1 Figure 8) suggesting the occurrence of alternative electron acceptors or that the bog, also, was not yet at steady-state (i.e., still in the acid phase). Over the course of the incubations, acetoclastic methanogenesis (particularly at 8–18 cm) decreased slightly corresponding to an increase in the non-fractionated CO₂ production over time (Figure 4). Together, these results suggest that fermentation in the shallow bog incubation was slowed relative to the fen which had already progressed through the fermentation-dominated acid phase and was approaching a methanogenic steady-state. The slowing of fermentation that held the bog in the acid phase possibly resulted from the action of Sphagnum-derived acids, which are thought to inhibit the microorganisms responsible for breaking down complex cellulose into the substrates needed for fermentation (Woodcroft et al., 2018).

In the fen incubation, the estimated CO₂:CH₄ ratio declined slightly but it was relatively stable near the predicted 1:1 ratio consistent with the fen being at methanogenic steady-state. The fen feature is similar to the collapsed palsa feature in that it is fully inundated and it has an abundance of sedge vegetation, fens develop as the active layer deepens over time (more thawing) and eventually the feature floods becoming fully inundated and connected with the surrounding water table. Thus, the fen is a more mature thaw feature than the collapsed palsa. Over time the collapsed palsa incubations proceed through the acid phase and toward the end of the incubations, approach methanogenic steady state, whereas the more mature fen material had already achieved this steady state by the start of the incubations. We suggest this progression through the acid-phase as a possible conceptual model for the field thaw transition: the newly thawed immature collapse features have high fermentation and homoacetogenesis rates as fresh organic material is suddenly subjected to anaerobic conditions. The excess substrate is rapidly depleted, however, and eventually, in the absence of other TEAs or further inputs, the system progresses toward a methanogenic steady-state as the thaw site matures. This would suggest that upon thawing initially there may be a rapid release of CO₂ coincident with the high fermentation rates, but over time (decadal timescale) there will be a shift toward increasingly methanogenic conditions. Because CO₂ and CH₄ have very different global warming potentials, the implications of this is that models may need to account for maturity of the thaw feature when estimating climate forcing.