Effects of a long‐term anoxic warming scenario on microbial community structure and functional potential of permafrost‐affected soil

Permafrost (PF)‐affected soils are widespread in the Arctic and store about half the global soil organic carbon. This large carbon pool becomes vulnerable to microbial decomposition through PF warming and deepening of the seasonal thaw layer (active layer [AL]). Here we combined greenhouse gas (GHG) production rate measurements with a metagenome‐based assessment of the microbial taxonomic and metabolic potential before and after 5 years of incubation under anoxic conditions at a constant temperature of 4°C in the AL, PF transition layer, and intact PF. Warming led to a rapid initial release of CO2 and, to a lesser extent, CH4 in all layers. After the initial pulse, especially in CO2 production, GHG production rates declined and conditions became more methanogenic. Functional gene‐based analyses indicated a decrease in carbon‐ and nitrogen‐cycling genes and a community shift to the degradation of less‐labile organic matter. This study reveals low but continuous GHG production in long‐term warming scenarios, which coincides with a decrease in the relative abundance of major metabolic pathway genes and an increase in carbohydrate‐active enzyme classes.


| INTRODUCTION
Permafrost (PF), which is classified as ground that stays frozen for at least two consecutive years, is widespread in the Arctic and subarctic regions. PF -affected soils in these regions store $1,300 Pg carbon, which equals 50% of the global belowground organic carbon, and the major fraction ($1,000 Pg) is stored in the upper 3 m of soil. 1 Over the past 30 years, high-latitude areas have warmed at a rate of 0.6 C per decade, which is twice as fast as the global average. 2 Modeled extremes predict a temperature increase of up to 7-8 C by the end of this century. 3 Consequently, the thawing of PF exposes large organic carbon stocks to decomposition by soil microorganisms. 4 This in turn could release the sequestered frozen long-term carbon stocks into the atmosphere as greenhouse gases (GHGs), carbon dioxide (CO 2 ) and methane (CH 4 ). 5,6 Although recent data on carbon isotopes imply that CH 4 derived from older carbon substrates is released relatively slowly, the increasing decomposition of PF carbon leads to a net positive contribution to the global atmospheric GHG budget. 7 This release initiates a positive climate feedback loop. [8][9][10][11] The active layer (AL) of PF-affected soils is exposed to seasonal freeze-thaw cycles, whereas the underlying PF is characterized by year-round below-zero temperatures and low water availability. The uppermost part of the PF, which is called the transition layer (TL), is more prone to thaw than deeper PF. This layer differs in cryo-features, carbon, and moisture content from the underlying PF and is irregularly exposed to thaw. 4,12 During thaw, water accumulation can lead to rapid gas diffusion limitation and a subsequent depletion of oxygen in the deep AL, whereas drainage of melted water allows oxygen penetration into deeper soils. Thus, hydrology plays an important role in regulating the soil redox potential and therefore the conditions for aerobic and anaerobic microbial metabolism. 13 In particular, anaerobic microbial carbon turnover processes remain poorly understood. 6 In addition, changes in soil temperature, nutrient availability, and vegetation influence soil organic carbon (SOC) decomposition and the resulting ratio of CO 2 to CH 4 emissions. [13][14][15] The SOC quality is important to influence carbon release from PF; however, a labile carbon pool is usually degraded immediately after PF thawing, and slow decomposing carbon fraction determines the long-term decomposition of PF SOC on a scale of 5-15 years. 16 After the initial degradation of labile organic matter fractions, microbes have access to less-labile organic matter fractions.
These organic compounds are mainly degraded by microbial guilds with cellulase and hemicellulase activity. 16,17 Therefore, the microbial community is expected to shift toward a population that degrades less-labile organic matter in the long term.
Increasing global efforts have recognized the need to understand the microbial ecology of thawing PF to better predict its role and fate in a warmer world, especially its impact on the global carbon budget. 10,18 Repeated freeze-thaw cycles modify the AL communities to conserve energy and obtain nutrients from a diversity of substrates through aerobic and anaerobic processes as well as adapt to survival under dynamic freeze-thaw conditions. 19 In contrast, microbial communities in PF layers can be very well conserved and of ancient origin. [19][20][21][22] In extreme cryogenic environments, the microbial community is likely to maintain a high level of stress tolerance due to longterm cold exposure. 18,23 PF thawing leads to shifts in microbial diversity, abundance, and activity within days to months. 20,24,25 Short-term thaw exposure (2-7 days) can cause a change in gene composition of former PF communities to resemble the AL. 20 These findings suggest that prolonged exposure to thaw will likely lead to even stronger shifts in community structures. The taxonomic and functional shifts, especially the enrichment of genes involved in the carbon and nitrogen cycle as well as respiratory processes, were identified from shortterm thaw experiments and field studies on PF thawing. 20,[26][27][28] Microbiota in a PF-affected peatland can modulate the metabolic and trophic interactions to maintain high fermentation rates and CH 4 production. 26 In addition, dominant processes shaping microbial communities in PF resulted from the stability of the PF environment, which imposed both dispersal and thermodynamic constraints. 29 This suggests pronounced differences in the microbial metabolic potential between the AL and PF soils.
Comparative studies on AL versus PF microbial communities, are, however, scarce, 20,30 and most previous studies focused on either AL [31][32][33][34] or PF microbial communities. 29,35,36 It is obvious that integrally studying these layers is important, because the deepening of the AL results in cryoturbation on the interannual scale in the PF -AL transition zone. 12 This can serve as a selective pressure on the PF microbial communities that are progressively exposed within the transient layer. The microbial community also has to adjust to a decrease in carbon availability as less-labile carbon fractions are left after easydegradable fractions are depleted within a few years after PF thaw. 37 The pool of previously frozen SOC is largely responsible for sustained carbon loss in thermokarst-active PF regions. However, the long-term consequences of PF thaw to the microbial community composition and potential are not well understood. 13,18 Understanding the microbial dynamics on PF soil thaw and warming beyond timescales of days and weeks is important to improve current predictive climate models for cold environments and to refine our knowledge of the magnitude and timing of PF carbon emissions in a warming world. Incubation experiments, even if artificial, are commonly used to inform valuable parameters for processbased modeling of the PF carbon feedback. 6,16,26,38,39 In particular, models expect a robust response under boundary conditions to help reduce uncertainty and provide a better estimate of impacts under different scenarios (e.g., references 6 and 40). In contrast to fast labile carbon that represents less than 5% of SOC at long turnover time (5-15 years), 16 the previously frozen old carbon dominates the carbon release from the northern PF regions where thermokarst landscape is extensively developed. 41 For this reason, the investigation of microbial decomposition on old carbon on climate-relevant timescales to the maximum potential can provide valuable reference for modeling study. Therefore, we conducted a long-term incubation without additional nutrient amendment to explore the response capability of microbial system at lower boundary conditions. On thaw and longterm warming of PF soils under anoxic conditions, we hypothesize that the microbial communities adapt to the depletion of available carbon and alternative terminal electron acceptors, through taxonomic and functional shifts, in spite of the heterogeneity of initial geochemistry and the microbial community between the AL and PF. A lab-scale incubation experiment was performed at 4 C for more than 5 years under anoxic conditions. Full metagenomic sequencing at the start and end of the incubation was coupled to measurements of geochemistry as well as CO 2 and CH 4 production rates. The results can provide insights into the effects of batch incubation conditions on GHG production for future modeling studies.

| Study site and sampling
The study site is located at Samoylov Island (72 22 0 N, 126 30 0 E), in the Lena River delta in Northeast Siberia, Russia. Samoylov Island developed during the Holocene and is underlain by continuous PF. The mean annual air temperature is À12. 5 C (1998-2011), the mean annual precipitation is 125 mm, and the mean soil temperatures (MST) is 4.1 C at a depth of 20 cm in the warmest month of August (mean air temperature 8.5 C in August). 42 On the island, ice-wedge polygons are extensively developed with low-lying polygon centers and elevated polygon rims on the surface. The dominating vascular plant species in the polygon centers is the sedge Carex aquatilis. 43,44 Samples were taken from a polygon center in spring 2011 when the soil was entirely frozen. The location and sampling procedures of the polygon used for this study are described earlier. 45 According to our previous study, oxygen depleted in the upper centimeters due to waterlogging. 46 Limited by remoteness, logistics, and harsh cold sampling season, the samples retrieved from the field in the frozen season are scarce, therefore limiting the use in multiple experiments and replication studies. In this study, the samples from the AL (15-22 cm), TL (33-37 cm), and PF (42-51 cm) were used for incubations. For this study, we performed GHG flux analyses (three biological replicates for GHG production), chemical analysis (two technical replicates), and a time series of molecular analyses (three samples for each time point) to follow the general trend of microbial response to long-term warming exposure of the three distinct zones of a thawaffected PF soil. Here we report the results of the first time point (5 years) after incubation.

| Anoxic incubations
Samples from AL, TL, and PF were incubated in the dark at 4 C under anoxic conditions for more than 5 years. Here, the selected incubation temperature was the MST of 4.1 C at a depth of 20 cm in warm August 42 under oxygen-depleted conditions, which are common in this area. 46 No nutrient amendment mimicked the boundary response and minimized additional biases. During the incubation, GHG production was followed for 1,163 days. 6,45 As the CH 4 production rates remained fairly constant and low in the long term, production rate measurements To obtain an overall scenario of the taxonomic differentiation at community level, we also assigned taxonomic labels to the merged paired-end QC short reads by mapping against the SILVA 132 SSU nonredundant reference database 55 using local blast (v2.6.0+). 56 The taxonomy was reported maximally at family level due to the relatively short-sequence fragment lengths (which are 203, 243, and 269 bp at 25, 50, and 75% quantiles for the merged paired-end reads, respectively). Length and similarity fraction were set to 180 and 70%, respectively, resulting in 74,765 mapped merged paired-end reads. In addition, the taxonomic composition was assigned based on the annotated ribosomal protein S3 (rpS3, which was also recently proposed as "uS3" according to the Ban Lab, https://bangroup.ethz.ch/research/ nomenclature-of-ribosomal-proteins.html). The rpS3 marker detects organisms with incomplete or unavailable SSU rRNA gene sequences and more strongly resolves the evolutionary deeper radiations. 57 Therefore, it was used as an alternative approach to analyze the general change in microbial community composition on warming. However, the reference database is less extensive for rpS3 than for the SSU rRNA, and the analysis is more sensitive to the quality of the assembly process than that of the SSU rRNA. Details of this analysis can be found in Supplementary Document 1.
Microbial community visualization was performed in R. 58 Community data were collapsed at higher taxonomic level with the R package otuSummary. 59 Ordination analyses of microbial community structure were performed with the R package vegan (v2.5-2). 60 Heatmaps were generated with the package ggplot2 (v3.1.0). 61 In addition, the fold changes incubated with respect to the initial relative abundance of taxonomic composition were used to semiquantitatively estimate the extent of abundance shifts for various microbial taxa.
The fold changes were visualized via a bubble plot with ggplot2 (v3.1.0). Principal Coordinates Analysis (PCoA) and symmetric procrustes analysis on the functional and taxonomic profiles were performed by using Bray-Curtis dissimilarity with package vegan (v2.5-2). 60 Permutational analysis of variance on the CH 4 production rates over layers was performed with the kruskal.test function in R. 58 In addition, permutation test was used to compare the changes in functional profile by layer in R.

| CO 2 and CH 4 production rates
The CO 2 and CH 4 production rates were monitored within the AL, TL, and PF layers over the course of 1,163 days to gain insights into GHG production rates under anoxic thaw conditions in the long term ( Figure 1). The CO 2 production rates peaked during the first 50-100 days of incubation, and then a rapid decrease was observed for all layers. Then, CO 2 production rates were similar in the different layers, with slightly higher levels observed in AL. The initial CH 4 production rates were about an order of magnitude lower than CO 2 production rates. In AL the rates did not show a lag phase, and production rates peaked after $ 200 days, after which a gradual decrease was observed. After $750 days, a marginal increase was observed.
The TL showed a slight initial peak associated with direct CH 4 release followed by a short lag phase of 50-75 days. The CH 4 production in TL was more stable, with a slight increase in production after $ 600 days. The PF exhibited the smallest initial release of CH 4 , followed by the longest lag phase between 20 and 150 days. The CH 4 production rates in PF were relatively constant, with lower peaks around 300 and 750 days. In PF, the highest CH 4 production rates were observed after more than 2 years of incubation. For detailed CH 4 production data, see Figure S1. Permutational test on the measured CH 4 production rate (n = 43) was statistically different between the three layers (p < 0.01). The ratio between the cumulative CO 2 and CH 4 production showed similar patterns in all the layers. All ratios were initially >8, and they approached 1 around 300 days of incubation, which indicates that the conditions became more methanogenic.

| Physicochemical pore water and bulk soil properties
Pore water analysis on initial samples indicated clear differences between AL, TL, and PF ( Figure S2). Ammonium concentrations decreased along the depth gradient (2.0, 0.7, and 0.3 mg L À1 , respectively). Overall, nitrite was very low (<0.1 mg L À1 ), and nitrate was negligible (<0.01 mg L À1 , not shown). Sulfate contents were highest in AL and TL (2.9 and 3.3 mg L À1 ) and lowest in PF (1.5 mg L À1 ). PF was F I G U R E 1 (a) CO 2 and (b) CH 4 production rates and (c) the ratio of CO 2 to CH 4 production rates over time for the active layer (AL), transition layer (TL), and permafrost (PF). The black line shows the average rate, and the blue shadow indicates the standard deviation of the replicates. The production rate was calculated based on the slope by a moving window using four measurements. The ratio of CO 2 to CH 4 was based on the average rate values of the CO 2 and CH 4 production rates. SOC, soil organic carbon [Colour figure can be viewed at wileyonlinelibrary.com] characterized by higher moisture content (80.9%) and higher total C (17.7%) and total N (0.54%). Overall, C/N ratios were similar in all layers, and they ranged between 29.6 and 32.7%. Ferrous [Fe(II)] iron content in TL was approximately 10-fold higher than in AL and PF (37.5 vs. 3.7 and 3.9 mg L À1 , respectively). In contrast, Fe(III) contents were higher in AL and PF (11.8 and 8.9 mg L À1 ) compared to TL (4.9 mg L À1 ). The pH was slightly acidic for all samples and ranged from 5.4 in AL to 5.5 in TL and 5.6 in PF ( Figure S2).

| Taxonomic and functional shifts on long-term warming
Long-term warming resulted in a pronounced shift in both the taxo-  1 and 5.8%). Acidobacteria showed a low abundance in the original samples (3.0, 3.9, and 2.6% for AL, TL, and PF, respectively) and were negligible (below 0.01%) at the end of the incubation.
A complete overview of the bacterial community changes at phylum level for all layers is provided in Table S2. On the class level, the strongest relative decreases were observed for Verrucomicrobiae, Alphaproteobacteria, Actinobacteria, and Acidobacteria (Figure 4a).

| Taxonomic shifts within the archaeal and fungal communities
The shift in the taxonomic profile is also reflected in the archaeal community structure on the phylum level ( Figure 3)  (Ceratobasidium) ( Figure S5).

| Functional shifts
Overall, the functional profiles at metabolic pathway level showed similar responses and in most cases a decrease in long-term warming ( Figure 5). The relative gene abundance for CH 4 cycling (CH 4 production and aerobic oxidation of CH 4 ) showed slight decreases. Similar patterns were also observed for carbon fixation, nitrogen, and sulfur cycling genes, despite the minor differences observed among the different layers.
To assess the functional changes in more detail, we analyzed the relative abundance changes for key functional marker genes (Table S4) (Table S5), and Bacteroidetes-associated CAZy classes showed substantial increases in all the layers ( Figure S8). In addition, the enrichment of CAZy families within fungal functional guilds is associated with the decomposition of xyloglucan and xylan (GH54, CBM42, and GH31) across all layers, galactomannan (GH26) in the TL and PF, and cellulose (GH1, GH3, and GH5) in the AL ( Figure S9).

| DISCUSSION
GHG production rates showed an immediate response in CO 2 release and a more steady response in CH 4 release over the 5-year incubation period. Therefore, initial CO 2 production rates largely outweighed CH 4 production rates. Rapid initial carbon degradation can result in high CO 2 /CH 4 ratios, 62 which coincides with our observations. High initial CO 2 production was also observed in a 4-year incubation study on deep PF deposits from the Lena River delta, Siberia, whereas CH 4 production rates were much lower, which is consistent with our study. 63 An excess of CO 2 formation indicates fermentative processes as well as anaerobic respiration using alternative terminal electron acceptors, including NO 2 À /NO 3 À , Fe(III), and SO 4 2À , that are available in the sediment. 64 Indeed, Fe(III) and SO 4 2À were detected in the soil pore water of all layers before long-term incubation ( Figure S2), whereas NO 3 À was not detected. Rapid initial activity is furthermore related to the presence of a considerable labile carbon pool in PF soils. 65,66 It has been reported that this labile organic matter is rapidly turned over under suitable conditions, 66,67 whereas the less-labile carbon was shown to fuel lower-rate CH 4 production in the long term. 7 A study on long-term carbon mineralization showed that under anoxic conditions only 25% of the aerobically released carbon was converted to CO 2 and CH 4 . 63 The calibrated carbon degradation model executed in this study predicted 15.1% (aerobic) and 1.8% (anaerobic) initial carbon release over 100 years, highlighting the slow carbon release in the long term. This was mainly assigned to the presence of more stable complex organic compounds that are more inert and are therefore degrade much slowly. 38,68 Increased water saturation coincides with ice-rich PF thaw, which stimulates wetland and thermokarst lake formation. 4  where, in the long term, anoxic conditions also lead to stable though low rates of CH 4 production. Although methanogenesis became more important relative to CO 2 production, the abundance (both relative and absolute) of methanogenic archaea decreased, even though the relative abundance of hydrogenotrophic methanogens increased. This stresses the potential limitation of methanogenesis due to reduced substrate provision by fermentative microbial groups as well as a depletion of labile carbon. In our study, the first peak in CH 4 production could therefore be related with higher abundances of labile organic carbon that was present in situ. When labile organic matter pools get depleted, more stable organic matter fractions fuel the system. The slight increase in CH 4 production over time, especially in TL and PF, may indicate a shift toward the degradation of this less-labile organic matter pool. Evidence for this is provided by the functional microbial shifts using end-point metagenome analyses. Due to the higher structural complexity, the degradation of stabile organic molecules requires more energy, which results in higher temperature sensitivities compared to labile compounds. 71 The low GHG fluxes in the long term indicate that warming to 4 C still limits the turnover of more stable organic matter. Although the degradation of the lesslabile organic matter fraction is much slower, it could still fuel a relatively stable methanogenic community.
In combination with the decreasing GHG production rates and a shift toward more methanogenic conditions, we observed substantial changes in microbial community structure and functions during the long-term warming scenario in all layers. The observed divergent shifts in community structure in the three different layers indicate stress for the microbial community, so that microbial activity was severely restricted although it did not cease. The anoxic setup also rejects the scenario of oxygen penetration into the soils, which was, however, beyond the focus of this study. Finally, the initial samples were preserved longer before DNA extraction than the incubation that could have manipulated the communities although this effect could be negligible as the cores were drilled when PF was entirely frozen and preserved in a continuously frozen state like natural PF.
Clostridia and Bacteroidia showed a positive response to thaw based on their relative abundances. These bacterial groups include many fermenting species that, for instance, play important roles in anaerobic fermentation of organic matter in anaerobic digestors. 72 Similar to our observations, Bacteroidetes were detected within the TL on a PF soil at Svalbard, Norway. 73 High occurrences of Bacteroidetes have been related with their metabolic flexibility, as well as rapid growth on easily accessible substrates. 74 Rapid growth on labile compounds could support their success on thaw, whereas their metabolic flexibility can support growth in the long term. Clostridia are fermentative bacteria that are generally well adapted to extreme conditions, 75 and they are able to ferment plant polysaccharides in soils. 76 Exposure to anoxic conditions led to an increase in relative These taxa seem to be important players in a warming Arctic.
The long-term incubations also resulted in an overall decrease in abundance of methanogenic archaea, as briefly discussed earlier in the context of GHG production rates. This observation is in contrast with several short-term incubation studies on PF soils showing an increase in methanogen abundance. 15,20 The decrease in methanogens observed here is likely explained by the much-longer incubation period that resulted in an overall (relative) decrease in many genes and pathways involved in energy metabolisms and not just by Methanosarcinales after thaw exposure, it further supports the formation of a less acetate-driven system, which is related to the degradation of labile organic matter and the formation of a more hydrogendriven system together with less-labile organic matter degradation. 87,88 Similar observations in long-term anoxic incubations were made earlier, 22 and also studies on ice-rich Yedoma deposits have suggested that acetate is a less-relevant substrate in PF deposits exposed to long-term anoxia. 64 Unlike methanogenic archaea, Bathyarchaeia showed an overall increase in the course of the long-term warming scenario, whereas Thaumarchaeota were overall poorly abundant. The phylum of Bathyarchaeia represents an evolutionary diverse microbial group that is found in a wide range of organic-rich environments. 89 Interestingly, the recent discovery of the growth of Bathyarchaeota subgroup 8 (Bathy-8) on lignin suggests that it can play a key role in the degradation of less-labile plant organic matter fractions. 90 In addition, genomic and enzymatic analyses of several Bathyarchaeal lineages showed their capacity for acetogenesis and fermentation of a wide range of organic substrates, including cellulose and aromatic compounds. 91,92 Thaumarchaeota are related to aerobic NH 4 oxidation. 93 They are abundantly detected in environments with low ammonia concentrations, like PF, where they can provide an important role in the nitrogen cycle. 18,94,95 In conjunction with the pronounced changes in GHG production potentials and microbial community structure, we observed large changes in the microbial metabolic potential. This is in line with a study on the effects of fire thaw that found that sulfate reduction genes were more abundant in anoxic deep soil layers. 27 However, in the long term it seems that dissimilatory sulfate reduction is mainly limited by sulfate availability, but experimental evidence is needed to support our observations.
The assumption that initial GHG fluxes and the short-term thaw response are associated with the labile organic matter pool whereas long-term GHG fluxes are controlled by the slower degradable organic matter pool agrees with the CAZy-based analysis. This analysis revealed a substantial increase in functional traits that are potentially involved in degrading cellulose and hemicellulose. This relates to scenarios in which the microbial community is able to adapt to anaerobic utilization of less-labile organic carbon from dead plant material after long-term thaw exposure of PF. 90,105 Similarly, an increased lignin decomposition contributed by Proteobacteria was revealed in AL soils of Arctic tundra soils after depleting soil labile C through a 975-day laboratory incubation. 106 As decomposition removes labile organic compounds, the remaining pools of soil carbon consist of a pool of less-labile or stable material, which will force the microbial community to acclimate. Through these microbial functional modulations, PF thaw and long-term exposure to warming result in steady GHG releases, even at low temperatures.
Further temperature increases may increase GHG production rates given that less-degradable carbon is particularly sensitive to warming. 107 In addition to prokaryotes, fungi are generally resistant to warming and can play an important role in decomposing the lesslabile or complex organic carbon. 108,109 The main fungal taxa in this study include many lineages able to degrade recalcitrant carbon ( Figure S5). Following the fungal enzyme sets for plant polysaccharide degradation, 110

| CONCLUSIONS
This study used an anaerobic incubation system to investigate the potential of microbial response to environmental changes in which labile carbon is steadily depleted, and older carbon fractions are solely available in the long run. These conditions are specifically relevant to feed the boundary response potential of microbial communities into climate models. However, our setup excluded the contact to meltwater fluxes that carry gases and nutrients in field conditions. The three layers exhibited a general trend with an adaptation to the depletion of labile carbon and alternative terminal acceptors within approximately 1 year, despite the heterogeneity of the initial geochemistry of the three layers, and the significantly different CH 4 production potentials that were observed. This closed-off incubation system minimizes biases induced by other factors, including temperature, water table, and availability of nutrients. The system has imposed strong stress on the microbial population to favor lineages that can survive and function under alternative electron acceptor-and labile carbon-depleted conditions. The three different layers showed that PF thaw can lead to both taxonomic and functional adaptations for metabolizing less-labile carbon in the long term, which is unlikely to be caused by stochastic processes alone. This long-term microbial mechanism allows for relatively stable but low GHG production in long-term compared to short-term scenarios.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available in European Nucleotide Archive (ENA) at EMBL-EBI under acession number PRJEB38557 (https://www.ebi.ac.uk/ena/browser/view/PRJEB38557), sample accession number SAMEA6866753-SAMEA6866758.