The rainforest Oxisol was sampled from a per-humid lower montane site in the Luquillo Experimental Forest, Puerto Rico, and the cropland Mollisol was sampled from a central Iowa field under long-term corn/soybean (Zea mays/Glycine max) cultivation. General properties of these soils are listed in Table 1 and were further described in a companion study (Huang et al., 2021). Replicates from each soil were subjected to each of five headspace treatments, consisting of anoxic periods of 0, 2, 4, 8, or 12 d, each followed by a 4-d oxic period; these cycles were repeated for the duration of the experiment (384 d). The total number of anoxic/oxic cycles varied among treatments according to the length of the anoxic phase. There was a total of 64, 48, 32, or 24 cycles for the 2-day, 4-day, 8-day, or 12-day treatments.

Each experimental replicate comprised 5 g soil (dry mass equivalent, amended with water to field capacity) incubated inside a separate glass jar (0.95 l) with a gas-tight aluminum lid sealed with a Viton gasket and two butyl septa for headspace flushing and venting as described below. Field moisture capacity measured 1.01 g H₂O and 0.46 g H₂O per g dry soil for the Oxisol and Mollisol, respectively (Huang et al., 2020b). Because labile C may gradually become depleted during soil incubations (Chacon et al., 2006), we amended soils with ground litter at the beginning of the experiment to simulate a C-rich surface soil microsite. Each replicate of each soil was amended with 0.5 g of dried and ground aboveground litter from Andropogon gerardii (a perennial C₄ grass) to ameliorate short-term C limitation and to simulate conditions that prevail in C-rich surface soil microsites. This plant was not present in either of these soils prior to the experiment.

Anoxic or oxic treatments were imposed by manually replacing the jar headspace with humidified dinitrogen gas (N₂) or humidified CO₂-free air, respectively, at 2-d intervals throughout the experiment. We chose to directly manipulate headspace O₂ content, rather than controlling O₂ availability indirectly via soil moisture, to achieve a more precise temporal gradient of O₂ availability independent of changes in moisture (McNicol and Silver, 2014). There is a substantial literature on responses of soil microbial communities to dry/wet cycles (e.g., Placella et al., 2012) that is not directly relevant to this study, because anoxic conditions are not necessarily achieved following wet-up of dry soils, and because traits related to drought tolerance/resuscitation may not be directly relevant to O₂ dynamics in soils from humid climates. Gas was flushed through each jar at >500 ml min⁻¹ for 15 min via a distribution manifold connected to each jar by piercing a septum with a needle; jars were vented through a separate needle placed through the other septum. Needles were removed and jars were left sealed for 2 d, at which point the headspace gas was subsampled by syringe for CO₂ and CH₄ measurements (Huang et al., 2021), and jars were flushed again with the appropriate gas (N₂ or CO₂-free air). The logistical requirements of the headspace flushing regime limited the number of sacrificial replicates which could be included in the experiment. Water vapor lost during flushing was measured gravimetrically and periodically replenished at the end of an oxic period. The ratio of soil mass to headspace volume was chosen so that CO₂ concentrations were always <6,000 ppm during oxic treatment phases. Analyses of CO₂ in control jars (no soil) indicated that leaks were negligible. Mass balance indicates that any dissolved porewater O₂ present during the oxic periods would be rapidly depleted by respiration during the anoxic periods. Porewater saturated with O₂ at 23°C at 300 m (8.3 mg O₂ L⁻¹) contains 0.11–0.26 μmol O₂ per g dry soil mass equivalent, given the moisture values listed above. Given observed mean CO₂ + CH₄ emissions of 3.3 μmol C g⁻¹ d⁻¹ (Huang et al., 2021), any remaining porewater O₂ would be depleted within minutes to hours after the jar headspace was flushed with N₂. Increased CH₄ production following flushing with N₂ (Huang et al., 2021) also indicates that anoxic conditions were achieved.

During the incubation, CO₂ and CH₄ production declined slowly over time in all treatments as available C was depleted (Huang et al., 2021). At the end of the experiment, CO₂ + CH₄ production was several-fold lower than at the beginning of the experiment, but an asymptote in cumulative decomposition was not observed, and analyses of the masses and isotope composition of cumulative CO₂ and CH₄ emissions showed that more than half of the added litter C and > 85% of the initial soil C remained (Huang et al., 2021). ¹³C NMR analyses indicated that much of this remaining C was contained in carbohydrates, and that C molecular composition was similar among treatments (Huang et al., 2020b, 2021). The sum of organic C decomposed to CO₂ and CH₄ was statistically equivalent among many of the headspace treatments and differed by <20% among all treatments from a given soil type (Huang et al., 2021). Therefore, we propose that differences in C availability per se were not likely to be the primary drivers of changes in community composition among headspace treatments.

We extracted genomic DNA from replicate subsamples after 0, 48, and 384 d (n = 3, 3, or 5, respectively per soil/headspace treatment; Supplementary Table S2) using MoBio Powersoil extraction kits (Maryland, United States). More replicates were available for analysis at 384 d than at the other timepoints because all of the replicates used for gas flux analysis (Huang et al., 2021) were sacrificed at that point; at 0 and 48 d, separate replicates not used for gas measurements were used for DNA analysis. Replicates were each incubated in separate jars. DNA was submitted to the University of Wisconsin-Madison Biotechnology Center for Illumina MiSeq 2×300 paired end sequencing of the V4 region of 16S rRNA; methodological details are described in the Supplementary Material. Raw sequences were deposited in the NCBI Sequence Read Archive under accession PRJNA693044. Sequences were processed with Quantitative Insights in Microbial Ecology (QIIME) 1.9.1 (Caporaso et al., 2010b). Prior to assembly of paired end sequences, low quality and ambiguous sequences were removed using the default QIIME parameters. Chimeric sequences were identified via de novo and reference based detection using usearch61 (Edgar, 2010). Operational taxonomic units (OTUs) were identified by de novo clustering (0.97 threshold) and were aligned against the SILVA database (Quast et al., 2013) using PyNAST (Caporaso et al., 2010a) to assign taxonomy.

Sequencing depth varied from 34,382 to 54,902 reads (median = 44,081) and two samples with <30,000 reads were discarded prior to subsequent analyses (Supplementary Table S2). Read counts did not vary systematically among headspace treatments, soils, or timepoints. We identified 40,345 OTUs in the initial dataset, which was used for calculation of diversity metrics prior to trimming rare taxa (McMurdie and Holmes, 2013). For community analyses, we discarded OTUs occurring in <3 samples and normalized counts by relative abundance. We conducted community analyses in R using phyloseq (McMurdie and Holmes, 2013) 1.28.0 and vegan 2.5.5 (Oksanen et al., 2019). We used distance-based redundancy analysis ordination using the “capscale” function with sampling date and anoxic duration as numeric predictors and tested their significance by permutation. To test responses of individual taxonomic groups to anoxic duration, we used differential expression analysis based on the negative binomial distribution implemented with DESeq2 1.24 (Love et al., 2014), correcting values of p for multiple comparisons using the Benjamini and Hochberg method. In separate DESeq analyses, we used OTUs, families, orders, classes, and phyla as response variables. Anoxic duration was coded as a continuous numeric predictor (regression design) such that the response of each OTU indicates the mean log2-fold change per day of anoxic duration. To test whether microbial groups had shared responses to anoxic duration in both soils, we coded direction of response (increase or decrease) of log2-fold change values for shared groups across different taxonomic levels, and used chi-squared tests to assess whether the overall direction of treatment response was significantly shared between soils at a given taxonomic level.

The replicates extracted for DNA at 384 d were also used for CO₂ and CH₄ measurements throughout the experiment (Huang et al., 2021), enabling us to assess relationships between cumulative CO₂ and CH₄ production and 16S composition at 384 d after accounting for headspace treatments and soil type. We used a permutational multivariate ANOVA on a Bray-Curtis distance matrix, with cumulative CO₂, cumulative CH₄, headspace treatment, and soil type as predictor variables. This analysis was conducted using the “adonis2” function in vegan 2.5.5 (Oksanen et al., 2019); we report the marginal effect of each model term.

As expected, the Oxisol and Mollisol differed substantially in microbial composition at the phylum level (Figure 2A) and especially at the family level (Supplementary Figure S1). Of the 30,356 OTUs observed in >2 samples, 5,732 occurred in both soils, and 2,292 of these shared OTUs occurred in >2 samples from each soil. These shared OTUs were disproportionately abundant, comprising 36–66% (mean 50%) of sequences among soils, treatments, and timepoints (Figure 2B). Because soil type was the dominant factor influencing microbial community composition and obscured the effects of headspace treatments when all data were analyzed together, we first conducted separate ordinations for each soil to examine anoxic treatment impacts. Community composition varied predictably with anoxic duration and sampling date in both soils; these variables were strongly correlated with the first and second axes of the ordination, respectively (adjusted R² = 0.45 and 0.47 for the Oxisol and Mollisol, respectively; p < 0.001 for anoxic duration and sampling date; Figures 3A,B). After 48 d, samples with shorter anoxic duration (0-day and 2-day treatments) remained clustered in ordination space near pre-treatment samples, whereas samples with increasingly longer anoxic duration diverged along the second ordination axis (Figures 3A,B). After 384 d, samples from all treatments shifted along the first axis but remained separated along the second axis according to anoxic duration (Figures 3A,B). A similar pattern was evident in a combined ordination of samples from both soil types which included only OTUs shared between the soils (Figure 3C). Samples separated along axes that were correlated with anoxic treatment duration and sampling date, albeit with lower R² than in the separate ordinations (adjusted R² = 0.13, p < 0.001 for sampling date and p = 0.006 for anoxic duration; Figure 3C).

The permutational multivariate ANOVA of the entire microbial community dataset measured after 384 d similarly indicated that soil type was the dominant predictor (R² = 0.40, p < 0.0001), with a secondary role for anoxic treatment duration (R² = 0.03, p = 0.005). After accounting for these variables, there were relatively weak relationships between community composition and cumulative CH₄ production (R² = 0.02, p = 0.02) and cumulative CO₂ production (R² = 0.01, p = 0.13). However, within individual soils, slightly stronger relationships emerged between community composition and C flux data. Cumulative CO₂, cumulative CH₄, and anoxic treatment duration had relatively similar marginal relationships to community composition in the Oxisol (R² = 0.08, p = 0.01; R² = 0.10, p = 0.003; R² = 0.08, p = 0.01) and in the Mollisol (R² = 0.07, p = 0.02; R² = 0.09, p = 0.005; R² = 0.14, p = 0.0002).

The consistency of response of shared microbial groups to anoxic duration varied across taxonomic levels. At levels of phylum, class, and order, there was no significant overall shared response to anoxic duration in both soils. That is, increases or decreases in particular groups at these taxonomic levels as a function of anoxic duration were not shared between soils as indicated by pairwise comparisons of log2-fold change values (calculated by differential expression analysis) or by chi-squared tests for a similar direction of response (Supplementary Figure S2). However, we did observe a weakly significant anoxic response among shared families (p = 0.02, chi-squared test; Supplementary Figure S2D) and a strongly consistent anoxic response for shared OTUs (Figure 4). There were 88 combinations of OTUs and sampling dates where significant responses to anoxic duration occured in both soils (Figure 4), and 71 displayed a consistent positive or negative anoxic response in both soils—much greater than expected under the null hypothesis of independence (p < 0.0001, chi-squared test). Of these 71 OTU/sampling date combinations, 4 OTUs had significant responses to anoxic duration in both soils at both dates (48 and 384 d), such that 67 unique OTUs displayed consistent and significant responses to anoxic duration in both soils during our experiment (Supplementary Data Sheet S1). These OTUs were also disproportionately abundant, with six-fold greater (0.027%) median relative abundance than the other OTUs (0.0044%) in a given sample, considering all sequences measured at 48 d and 384 d (p < 0.0001). Considering all shared OTUs, log2-fold change values for anoxic duration response were also significantly correlated between soils (Figure 4; regression statistics are shown in the caption).

Although many taxonomic groups did not show a consistent response to anoxic duration in both soils, many other groups did respond to anoxic duration in one soil at either 48 d or 384 d. Overall, sequences belonging to a group with a significant anoxic response accounted for 58% (phylum), 53% (class), 44% (order), 49% (family), and 34% (OTU) of sequences (Figure 5A; Supplementary Table S1). However, consistency of response of particular taxonomic groups to anoxic duration varied greatly between soils and over time. For example, the phylum Bacteroidetes increased with anoxic duration in the Mollisol but decreased in the Oxisol after 48 d, whereas Firmicutes showed no response in the Mollisol but strongly increased in the Oxisol (Figure 2). The responses of individual OTUs within these phyla generally corresponded to the phylum as a whole (Supplementary Figure S3). However, in contrast to the more abundant phyla, Nitrospirae and Euryarcheaota did respond strongly and consistently to anoxic duration in both soils after 384 d (decreasing and increasing, respectively; Supplementary Data Sheet S1). The Euryarcheaota that we detected in these soils were largely methanogens (class Methanomicrobia; Supplementary Figure S4).

Similar to the phylum level, responses to anoxic duration at the levels of class and order were also generally inconsistent between soils, with several exceptions (Supplementary Data Sheet S1). The class OPB41 (Actinobacteria) increased with anoxic duration in both soils after 48 d and 384 d, Anaerolineae increased after 48 d, and Solibacteres and Deltaproteobacteria increased after 384 d. Chthonomonadetes, Gemm-1 (Gemmatimonadetes), PAUC37f (Acidobacteria), and OM190 (Planctomycetes) decreased with anoxic duration in both soils after 384 d. No orders responded consistently to anoxic duration after 48 d, but 15 responded after 384 d. In particular, orders belonging to different classes of Proteobacteria had consistent responses of differing direction: Rhodocyclales and Myxococcales increased with anoxic duration whereas Rhodobacterales, Entotheonellales, and HTCC2188 decreased. Overall, classes with consistent responses to anoxic duration in both soils comprised 0.1–1.3% (mean 0.5%) of sequences after 48 d, increasing to 11.6–22.3% (mean 15.2%) after 384 d, while orders comprised 14.1–21.1% (mean 17.8%) after 384 d (Supplementary Table S1).

Families with a consistent response to anoxic duration had generally similar relative abundances at both 48 d and 384 d, comprising 5–16% (mean 9%) of sequences (Figure 4B; Supplementary Table S1; Supplementary Data Sheet S1). Two families responded consistently to anoxic duration in both soils at both timepoints: an unnamed family in Solibacterales decreased while Cellulomonadaceae increased. After 48 d, several families in Proteobacteria decreased with anoxic duration in both soils (Bradyrhizobiaceae, Comamonadaceae, Haliangiaceae, Coxiellaceae), whereas several families in Clostridiales increased (Christensenellaceae, Peptococcaceae, Ruminococcaceae). Of these families, only Bradyrhizobiaceae had similar relative abundance in both soils after 48 d; the other families tended to be much more abundant in one soil than the other (Figure 5B). After 384 d, several unnamed families in Planctomycetes decreased with anoxic duration in both soils, while families within Acidobacteria, Actinobacteria, and Proteobacteria showed both positive and negative responses. For example, within the Alphaproteobacteria, Methylocystaceae increased with anoxic duration while Hyphomonadaceae decreased. Of the relatively abundant families (> 0.8% of sequences) that responded consistently after 384 d, three had approximately similar relative abundances in both soils: Myxococcaceae, Methylocystaceae, and 0319-6A21, a family in the Nitrospirales (Figure 5B).

OTUs with consistent responses to anoxic duration in both soils comprised a smaller proportion of total sequences than observed at the other taxonomic levels (1–8%, mean 4%), but responses between soils were more likely to be consistent for a given OTU (Figure 4). Four OTUs (two each in Micromonosporaceae and Bradyrhizobiaceae) decreased significantly with anoxic duration in both soils at both sampling dates, with log2-fold change values of −0.2 d⁻¹ to −0.7 d⁻¹ (Supplementary Data Sheet S1). These OTUs also had relatively similar mean abundances in the static oxic (0-day) treatment in both soils at both timepoints (0.01–0.04% of sequences). Of the many other OTUs that showed consistent responses to anoxic duration in both soils at either 48 d or 384 d, most belonged to the groups discussed above (Supplementary Figure S5; Supplementary Data Sheet S1). Several OTUs with a strong anoxic response in both soils were at least occasionally abundant, comprising >1% of sequences in a given sample. These included multiple OTUs in the genus Dactylosporangium (Actinobacteria, Micromonosporaceae) which responded positively, as well as in Anaeromyxobacter (Deltaproteobacteria, Myxococcaceae), Candidatus Koribacter (Acidobacteria, Koribacteraceae), and Ruminococcaceae. Three abundant Sinobacteraceae OTUs decreased with anoxic duration. OTUs from the Fe-reducing genus Geobacter showed a mixed response to anoxic duration (positive in the Mollisol and negative in the Oxisol), while OTUs from the putative Fe-reducing genus Pelosinus increased significantly with anoxic duration only in the Oxisol.

Inverse Simpson and Shannon Index values did not show consistent trends with anoxic duration throughout the experiment (Figure 6). In the Mollisol, inverse Simpson and Shannon Index values did not vary consistently with anoxic duration after 48 d. In the Oxisol, both indices decreased with anoxic duration after 48 d (p < 0.0001, linear regression), but the decreases were most pronounced in the 8- and 12-day anoxic treatments. After 384 d, inverse Simpson and Shannon Index values significantly (p < 0.05, linear regression) declined with anoxic treatment duration in both soils. However, in the Mollisol, indices actually tended to be higher in the 2- and 4-day anoxic treatments than the 0-d treatment (i.e., oxic control), and only declined consistently under the 8- and 12-day treatments. Similarly, in the Oxisol, these indices also tended to have similar values among the 0-, 2-, and 4-day treatments, and were only markedly lower in the 8- and 12-day treatments (Figure 6).

A number of previous experiments have assessed impacts of fluctuating oxic/anoxic conditions on microbial community composition (Pett-Ridge and Firestone, 2005; DeAngelis et al., 2010; Frindte et al., 2016; Mejia et al., 2016; Randle-Boggis et al., 2018; Winkler et al., 2019; Pronk et al., 2020) but it was often difficult to discern whether increasing the relative durations of anoxic vs. oxic periods caused systematic changes, and whether these differences were sustained over time. Here, our treatments corresponded to cumulative anoxic durations of 0, 33, 50, 66, and 75% of the experiment. We found directional shifts in community composition at the level of OTUs (Figure 3) and in coarser taxonomic groups (Figures 1, 4) in both soils as anoxic period duration increased from 0 to 12 d. Temporal changes across all treatments were even larger than the effects of anoxic duration (Figure 3) and were likely influenced by gradual decreases in C availability and changes in C molecular composition over time (Huang et al., 2021), given the strong linkages between microbial composition and decomposition rate observed elsewhere (Fierer et al., 2007). Notably, however, the magnitude of community change with increasing anoxic duration was similar in both soils (Figure 3), and it persisted at both timepoints despite the large temporal change in community composition over the course of the 384-d experiment. This finding suggests that a generalizable imprint of periodic anoxia on microbial community composition, manifested by taxon relative abundance or perhaps taxon presence, might be detectable across diverse soils, even if it has a smaller overall effect than other controlling factors such as pH (Delgado-Baquerizo et al., 2018).

When high-level taxonomic groups responded to anoxic duration, responses generally differed between soils (Figure 2; Supplementary Figure S2) with a few notable exceptions. Consistent with their known physiologies, nitrifiers in the phylum Nitrospirae and methanogens in the class Methanomicrobia showed strong negative and positive relationships with anoxic duration, respectively, in both soils. Soil Nitrospirae also declined following a 6-week anaerobic disinfestation treatment (van Agtmaal et al., 2015); our results showed that even brief but periodic O₂ deprivation can suppress these organisms in favor of anaerobes. The presence of Methanomicrobia in most soils and treatments (Supplementary Figure S4) implied that these organisms could either tolerate periodic O₂ incursions or that anoxic microsites persisted throughout the experiment, consistent with observations of reduced Fe in all treatments (Huang et al., 2021) and gross methane production observed from other soils in oxic environments (von Fischer and Hedin, 2002). Putative aerobic methane-oxidizers (Methylocystaceae) also increased with anoxic treatment duration, implying that benefits of increased methane availability during the oxic periods outweighed stress during the anoxic period, possibly due to use of alternative metabolic pathways or formation of resting stages (Knief, 2019).

The general lack of consistent phylum-level responses to anoxic duration between soils resulted from contrasting responses of nested taxonomic groups, many of which did respond similarly in both soils. In particular, multiple Actinobacteria subgroups showed consistent positive or negative responses to anoxic duration. The uncultured class OPB41 strongly increased with anoxic duration in both soils at both timepoints, and elsewhere this class co-occurred with methanogens in iron-rich aquatic environments (in’t Zandt et al., 2019) and in coal beds (Kirk et al., 2015). The Actinobacteria family Micromonosporaceae was highly sensitive to anoxic duration, and several OTUs showed consistent increasing or decreasing responses in both soils. The genus Dactylosporangium even contained multiple OTUs with differing anoxic responses (positive or negative) that were consistent in both soils. This putatively slow-growing genus was found to be abundant in hypoxic but not oxygenated lake water, possibly because of increased competitive ability under hypoxic conditions (de Menezes et al., 2012). The consistent sensitivity of these Actinobacteria classes, families, genera, and OTUs to anoxic duration across our two contrasting soils highlights the possible ecological importance of this group and its potential utility as an indicator of soil redox history.

Similar to the Actinobacteria, the Acidobacteria also showed strong but occasionally differing responses to anoxic duration at multiple taxonomic levels. The class Solibacteres and order Solibacterales increased with anoxic duration in both soils after 384 d; these groups occurred in other variable redox environments including Arctic peat (Lipson et al., 2013) and wastewater treatment bioreactors (Zhang et al., 2017) and increased after a pasture soil was flooded (Randle-Boggis et al., 2018). Two OTUs in the genus Candidatus Solibacter decreased with anoxic duration at 48 d. Genomic analyses suggest that these organisms are slow-growing aerobic heterotrophs (Ward et al., 2009), but they became increasingly abundant in paddy soils following long-term organic matter amendments (Yu et al., 2019), implying that they required time to proliferate under the conditions of our experiment. Solibacterales and Koribacteraceae also increased in agricultural soils under continuous corn cultivation, which was interpreted by the authors as a response to nutrient limitation (Soman et al., 2017). The strong response of these groups to anoxic duration in our experiment suggest that differences in anoxic microsites between cropping systems might also have contributed. Candidatus Koribacter may oxidize metals at redox interfaces (Ward et al., 2009; Allward et al., 2018), and this group increased with anoxic duration after 384 d in both of our soils, indicating their possible role in oxidizing reduced iron (Fe) or manganese during the oxic phases.

Both soils reduced and oxidized Fe at rates of tens to hundreds of μg Fe per g soil per anoxic/oxic cycle, as indicated by comparisons of Fe^II concentrations between oxic and anoxic treatment phases during the first 48 d of the experiment (Huang et al., 2021). The 16S data implicated several taxa that might have contributed to Fe reduction. Deltaproteobacteria showed contrasting responses among subgroups but Myxococcaeae and 10 OTUs from Anaeromyxobacter increased strongly with anoxic duration in both soils. This genus contains taxa capable of dissimilatory Fe reduction as well as aerobic respiration (Treude et al., 2003) and it increased elsewhere following Fe addition to rice paddy soils (Wang et al., 2020). Myxococcales also increased under redox fluctuation treatments in other experiments (DeAngelis et al., 2010; Randle-Boggis et al., 2018; Winkler et al., 2019). In a separate soil sample collected at the same site where our Puerto Rican Oxisol was obtained, meta-transcriptomic analyses indicated that Geobacter, Pelosinus, and Anaeromyxobacter were likely the most important Fe-reducing organisms (Wilmoth et al., 2018). However, in our Oxisol samples, Geobacter OTUs were rare, similar to the results of DeAngelis et al. (2010), who also used samples from this same site. Pelosinus OTUs were present in both soils but only significantly responded to anoxic duration in the Oxisol. The relatively high abundance of Anaeromyxobacter OTUs and their consistent increase with anoxic duration suggest that they may have contributed to measured Fe reduction in both soils.

Microbes that increased with anoxic duration in our experiment often corresponded with groups identified previously in wetland soils, anaerobic digestors, and in anaerobic disinfestation experiments (where soils were amended with C substrates and covered with impermeable material to kill pathogens). Several Ruminococcaceae OTUs increased with anoxic duration, consistent with their dominance following reductive disinfestation (Huang et al., 2019) and their increase in rice-paddy soil following straw amendment (Wegner and Liesack, 2016). The Chloroflexi class Anaerolineae increased with anoxic duration; this group was abundant in rice-paddy soil under current-producing conditions (Cabezas et al., 2015) or following iron mineral addition (Huang et al., 2020a), and was a core member of anaerobic digestor communities (Rivière et al., 2009). Taxa that responded negatively to anoxic duration included multiple classes and orders in the Planctomycetes which declined following anaerobic disinfestation (van Agtmaal et al., 2015) and in anoxic horizons of tundra soils (Lipson et al., 2015). Several Bradyrhizobiaceae OTUs consistently decreased with anoxic duration, and this broader group (Rhizobiales) also decreased in flooded pasture soils (Randle-Boggis et al., 2018). The functional attributes of these particular Bradyrhizobiaceae are unclear, but because of the abundance and ubiquity of this group here and elsewhere (Delgado-Baquerizo et al., 2018) they deserve attention as indicators of O₂ availability.

Oxygen availability has been proposed as a core driver of soil microbial community composition (Fierer, 2017) but we are unaware of experiments that explicitly tested the impacts of temporal gradients in O₂ availability in diverse environments. Using two soils with greatly differing biological and geochemical characteristics, we identified a core guild of O₂-sensitive microbes across multiple taxonomic levels. Responses were most consistent between soils at the OTU level. Shifts in the abundance and presence of the particular taxa identified may therefore hold promise in recording the recent history of O₂ availability. Many O₂-sensitive OTUs identified here were also common and prevalent members in global soil communities (Delgado-Baquerizo et al., 2018; Supplementary Data Sheet S1). It remains to be seen, however, whether these taxa also might consistently respond to O₂ variation in other soils, and experiments comparing a broader range of soils are needed to expand on the proof-of-concept experiment we report in this paper. Although O₂ tolerance/sensitivity may be a strongly conserved trait for some taxonomic groups, other groups showed disparate responses to anoxic period duration at relatively fine taxonomic resolution (Supplementary Data Sheet S1), indicating a potential for adaptation or niche partitioning. Furthermore, the O₂-sensitive OTUs identified here which were not commonly or consistently observed in previous syntheses of soil microbes may also prove useful in delineating soils that experience periodic O₂ deficiency from consistently oxic environments. Notably, shared responses of OTUs to O₂ deficiency occurred despite large differences between our study soils in pH and other environmental attributes (Table 1) and their greatly contrasting initial community compositions (Figure 2). Along with variables such as pH, temperature, C availability, and moisture (Fierer and Jackson, 2006; Fierer et al., 2007; Delgado-Baquerizo et al., 2018; Frindte et al., 2019), our results support the hypothesis (Fierer, 2017) that time-integrated O₂ availability might serve as a core driver of microbial composition across diverse terrestrial soils from humid environments. Given the tremendous spatiotemporal heterogeneity of O₂ in terrestrial soils and the challenges of directly measuring this key biogeochemical variable at scales relevant to microbial physiology (Sexstone et al., 1985; Silver et al., 1999; Hall et al., 2013; Keiluweit et al., 2016), microbial community datasets deserve further attention as proxies for redox history.