We collected samples from the Wind River-Little Wind River floodplain near Riverton, WY (42° 59′ 19.1″ N, 108° 23′ 58.6″ W; Supplementary Figure S1) about once per month from April to September 2017 (Supplementary Table S1). Hydrological data was obtained from the USGS monitoring station on the Little Wind River¹. Flooding at the site during almost the entire month of June prevented sample collection in this month, so instead two samples were collected in July (7th and 25th). Soil samples were collected using a hand-auger (30 cm long, 10 cm diameter collection head) from a new location within a ∼1 m diameter circle at the site for each sample date. The cores were taken in an area between a soil pit dug (and re-filled) in 2016 (profile description in Supplementary Figure S2) and a gallery of porewater samplers [10 cm long, 0.15 μm pore size rhizons (Rhizosphere Research Products)].

Cores were divided into ∼10 cm intervals or consistent with distinct soil layers (if noted as being smaller than 10 cm). All samples used for microbial or redox-sensitive analyses were immediately frozen on dry ice in the field and subsequently shipped on dry ice and stored in −80°C until analysis. Total elemental composition and C analyses were conducted on samples preserved and shipped on ice and stored at −20°C prior to fresh sample extractions (water-extractable concentrations), extractions inside an anoxic chamber (3–4% hydrogen, H₂, 96–97% nitrogen, N₂; redox sensitive samples), or drying in ambient conditions (total concentrations). Soil moisture was determined gravimetrically by weighing subsamples of field moist soils upon arrival in the laboratory and again after drying for 24 h at 105°C. Samples from April were not analyzed for soil moisture.

Porewater from the nearby sampling gallery (5 depths) were collected at a finer time interval than soil samples, generally once a week; however, porewater was always collected on the same occasions as sediment soil for comparison. Porewater samples were immediately injected into pre-evacuated and crimp-sealed serum vials containing nitric acid (final concentration after injection of porewater was 1–2%) and shipped on ice. Porewater samples were stored at 2°C prior to analysis.

Total C and N concentrations of soil samples were determined in duplicates on finely ground, air-dried soils by a Carlo Erba NA 1500 elemental analyzer. Total organic C was determined in the same way, but after removal of carbonate C (CO₃-C) through dropwise 1M HCl treatment of soil slurries until fizzing stopped and subsequent rinsing with MilliQ-water and drying at 30°C. Total concentrations of other elements in soils (Na, Mg, S, Fe, K, Ca, As, Mo, U) were also determined on finely ground, air-dried samples by X-ray fluorescence (XRF) spectroscopy (Spectro Xepos HE). Acid-extractable Fe(II) was determined by the ferrozine method (Stookey, 1970; Viollier et al., 2000) on MilliQ-diluted samples (to get within the analytical range for the method) from 1 g:10 mL (soil:0.5M HCl) extracts that had been shaken for 2 h and subsequently filtered (0.22 μm, syringe-filters). Water-extractable concentrations of elements (organic C, Na, Mg, S, Fe, K, Ca, As, Mo, U) were determined on 1 g:10 mL (soil:MillliQ-water) extracts after 2 h of shaking and subsequent filtration (0.22 μm, pre-rinsed syringe filters) of the supernatant. Water-extractable organic C in these extracts was analyzed by the non-purgeable organic C (NPOC) method on a Shimadzu TOC-L analyzer with in-line acidification (phosphoric acid) of samples to volatilize inorganic C. All other water-extractable element concentrations were analyzed by inductively coupled plasma (ICP) spectroscopy after acidification with nitric acid (final concentration of nitric acid in analyzed sample was 2%). Fe, S, Ca, Mg, Na, and K concentrations were analyzed by ICP optical emission spectroscopy (ICP-OES, iCAP6000, Thermo Fisher Scientific, Cambridge, United Kingdom), whereas As, Mo, and U were analyzed by ICP mass spectrometry (ICP-MS, XSERIES 2, Thermo Fisher Scientific). Porewater samples were analyzed by ICP-OES and ICP-MS in the same way as the water extracts. All chemical data is included in Supplementary Table S1.

DNA was extracted from 0.25–0.5 g of soil from 9 to 17 depths per sample date (Supplementary Table S1) using the PowerSoil DNeasy Extraction Kit (Qiagen) with slight modifications: cell lysis was performed on a FastPrep (Thermo Fisher Scientific) bead beater twice at speed 5.5 for 30 s with a 1 min incubation on ice in between; DNA was eluted after a 1 min incubation at room temperature. The concentration of DNA in each sample was determined on the Qubit system with the high sensitivity assay (Invitrogen). Samples were submitted to the Georgia Genomics and Bioinformatics Core² for Illumina library preparation and 16S rRNA amplicon sequencing, using V4–V5 primers (Parada et al., 2016). A total of 8,063,477 reads were generated from 120 samples (available in the NCBI Single Read Archive – BioProject ID: PRJNA626616). Sequences were pooled from replicate samples to a total of 83 unique samples (Supplementary Table S2). These reads were quality checked and processed through mothur (v. 1.41.3; Schloss et al., 2009) following the MiSeq SOP (Kozich et al., 2013; accessed December 2019), including trimming to 450 bp and removal of chimeras. Sequences were aligned against the SILVA SSU database (Release 132; Quast et al., 2013; Yilmaz et al., 2014) and classified. The resulting 6,828,356 reads were grouped into operational taxonomic units (OTUs) at 97% similarity (Supplementary Table S2) for further analysis, with singleton and doubleton OTUs removed (a total of 26,531 OTUs were analyzed).

Analysis of molecular variance (AMOVA), homogeneity of molecular variance (HOMOVA), and linear discriminant analysis (LDA) effect size (LEfSe) tests were run on microbial communities grouped by soil horizon and sample date in mothur (as described in Schloss, 2008). LEfSe is further described in Segata et al. (2011). All other statistical analysis was performed in R (v. 3.5.3; R Core Team, 2017). Geochemical values missing from <15% of samples in our dataset (Supplementary Table S1) were imputed prior to statistical analysis, predicted from a multivariable regression model using the Multivariable Imputation by Chained Equations (MICE) method, implemented in R with the mice package (v. 3.8; van Buuren and Groothuis-Oudshoorn, 2011) with parameters set to: maxit = 100, seed = 500. The following variables were included: total and soluble organic C; total inorganic C; total and soluble Ca and As; soluble Na, Mg, S, Mo, and U; and acid-extracted Fe(II). Soil moisture had too many missing data points for inclusion in the imputation model; therefore, it was analyzed further with a reduced dataset that excluded data from April samples.

We combined our geochemical data with microbial sequencing data (taxonomy and OTU counts) for analysis in phyloseq (v. 1.24.2; McMurdie and Holmes, 2013). Relative abundance of OTUs was calculated using the transform_sample_counts function from count data, as a percentage of total sequence counts per OTU per sample. For all ordination plots, we rarefied the sequencing data to an even number of sequences per sample using rarefy_even_depth in phyloseq. We used redundancy analysis (RDA) to analyze geochemical data. For OTU data, we used a Bray–Curtis dissimilarity matrix to evaluate community variability across samples with unconstrained [non-metric multidimensional scaling (NMDS)] and constrained [Constrained Analysis of Principal Coordinates (CAP)] ordinations. Correlation coefficients (r) for CAP were extracted from biplot scores. Figures were made in R using the ggplot2 package (v. 3.3.0; Wickham, 2016).

Based on the profile description from the nearby soil pit we dug in 2016 (Supplementary Figure S2), we identified seven major categories of soil layers: Topsoil (∼0–30 cm), Evaporite (∼30–55 cm), Sand (∼55–75 cm), Evaporite-Clay (∼75–100 cm), TRZ (∼100–140 cm), Clay (∼140–150 cm), and Aquifer (∼150–170 cm). Although each core exhibited the same distinct layers, due to slight variability in surface elevation and layer thicknesses between the core locations, the exact depth intervals varied slightly between sample dates. The grouping of samples into categories for each time point was therefore done by matching soil properties (color, texture, visual evaporite abundances, and total organic C content), rather than by exact depth. Depth designation was adjusted accordingly (i.e., assuming the same layer occurred at the same depth at each sample location) to enable direct comparisons of data for the same layer between time points.

Three organic-enriched zones (>1% organic C, wt%) were identified, largely corresponding to the Topsoil, Evaporite, and bottom of Evaporite-Clay extending through the TRZ layers (Supplementary Figure S3), i.e., mainly coinciding with a clayey texture (Supplementary Figure S2). Coarser horizons typically contained ≤0.05% organic C (Supplementary Figures S2, S3). Solid phase carbonate content varied between sampling times (Supplementary Figure S3, difference between Tot-C and Org-C), with the Evaporite-Clay and TRZ layers exhibiting the highest carbonate concentrations (up to 89% of the total C content) as well as the most variable carbonate concentrations throughout the season.

Over the period sampled, the Riverton site experienced seasonal transitions in hydrology (Figure 1A): (1) groundwater table rise in April/May due to spring snowmelt, eventually culminating in overbank flooding of the Little Wind River; (2) inundation at the sampling location for almost the entire month of June, followed by rapid and seasonal decline of the Little Wind River discharge and onset of summer drought conditions; and (3) water table recession below the bottom clay layer. As snowmelt runoff saturated the aquifer and flooded the site, a reducing front propagated upward through the soil profile, as indicated by an almost simultaneous increase in porewater concentrations of Fe and Mn (Figures 1B,C). In this period, the corresponding oxidizing front propagated downward at a much slower pace than water table recession, as evidenced by porewater concentrations of Fe and Mn, which at a given depth remained elevated for several weeks after the water table had fallen below this depth (Figures 1B,C). Notably, the HCl-extractable Fe(II) concentrations post-flood continued to accumulate with depth throughout the TRZ and Clay during the falling water table stage (Supplementary Figure S4).

Another effect of the flood was that water-soluble organic C was released and re-distributed throughout the profile (Supplementary Figure S4). This pool was then quickly depleted in the sand and (more slowly) in the TRZ, whereas it remained high in the Topsoil and the Evaporite-Clay layers. The Evaporite-Clay apparently forms a distinct moisture boundary at Riverton; the soils above dried relatively quickly whereas those at and below maintained high moisture throughout the sampling period (Supplementary Figure S4). The maintenance of high water-soluble organic C concentrations in the Topsoil is not surprising, given the large pool of solid phase organic C in this layer (Supplementary Figure S3) and the abundance of roots (Supplementary Figure S2). Moreover, the maintained pool of HCl-extractable Fe(II) into September further indicates that root and microbial respiration of infusing O₂ likely prevented re-oxidation of Fe(II) at the top of the soil profile (Supplementary Figure S4), which is consistent with a maintained supply of soluble organic C from microbial decomposition and root exudation.

Total and water-soluble U (Supplementary Figure S5) exhibited the same general trends, with the TRZ having the highest concentrations before the flood, which appears to have flushed U downward. The Evaporite-Clay and Clay layers appear to have accumulated U as it moved downward with the receding water table, but the Aquifer maintained the highest U concentrations after the flood. As the soil dried, U accumulated in the layers containing evaporites, as well as in the bottom of the profile in a similar pattern to HCl-extractable Fe(II), with a considerably higher proportion of U being water soluble in the Evaporite-Clay than in the bottom Clay and Aquifer (Supplementary Figure S5).

Overall, statistical analysis of geochemical measurements at the Riverton site strongly separated samples based on depth across the primary axis (RDA1, 41.3% variance explained; Figure 2), in accordance with the soil horizons defined above. Depth was the strongest contributor to this axis (r = 0.94), along with organic C, N, and metals, including Mo, U, and Fe(II). Though minor, RDA2 (10% variance explained) represented K (r = 0.71) and HCl-extractable Fe(II) (−0.69), followed by Mg, S, Fe, and U.

A total of 26,532 OTUs were recovered and analyzed from Riverton soils, representing 58 bacterial and 9 archaeal phyla; our dataset includes 190 different taxa at the class level, and 847 families. Microbial community composition strongly separated by depth – an indicator of soil type – at the phylum level (Figure 3A). A majority of taxa recovered belong to the Proteobacteria phylum, specifically Alphaproteobacteria, Deltaproteobacteria, and Gammaproteobacteria classes (Figure 3B). Deeper depths at our site become dominated by Proteobacteria (43–50% of total community), primarily Deltaproteobacteria. Surface layers in the soil have approximately equal proportions of Alphaproteobacteria (12–13%) and Gammaproteobacteria (10–12%), as well as Actinobacteria (18–19%) and Acidobacteria (10–12%), and to a lesser extent Gemmatimonadetes (7–11%). Other groups showed relatively consistent proportional abundance throughout the soil column, including Bacteroidetes (4–8% of total community), Chloroflexi (6–13%), Planctomycetes (5–8%), and Thaumarchaeota (1–2%). Though minor portions of the overall community, members of the phyla Rokubacteria and Firmicutes were two to three times more abundant in the upper soil layers, whereas Nitrospirae became more prominent in the deepest parts of the soil column (in Clay and Aquifer samples, Nitrospirae comprised 1–2% of total community compared to 0.4–0.6% in other layers).

We observed distinct differences between soil layers in N-cycling microorganisms, particularly with regard to nitrification, which involves the sequential oxidation of ammonia to nitrite and then nitrate by specific groups of Archaea and Bacteria. Within the ammonia-oxidizing Thaumarchaeota – which catalyze the first and rate-limiting step of nitrification – we found typical soil lineages (Nitrososphaera and Nitrosocosmicus sp.) at surface depths (Figure 4). However, deeper soil layers (TRZ, Clay, Aquifer) included groups more often observed in marine sediment (Nitrosopumilus sp.) or the major freshwater/estuarine lineage (Nitrosarchaeum sp.), both of which belong to the family Nitrosopumilaceae (Figure 4). Although comparatively minor contributors to the nitrifying community overall, ammonia-oxidizing bacteria were observed at the surface (Nitrosospira sp.; Topsoil and Evaporite) and deepest soil layers sampled (Nitrosomonas sp.; TRZ, Clay, Aquifer). Nitrite-oxidizing bacteria, namely Nitrospira sp., were found at all depths sampled, with the highest relative abundance (>25% of nitrifying microorganisms) within the Evaporite-Clay layer (Figure 4).

We found that microbial community composition varied by soil horizon (Figure 3), classified according to the intervals defined above. Overall, nearly every soil horizon at the Riverton site hosted a distinct microbial community (AMOVA, p < 0.05); only the TRZ and Clay communities were statistically indistinguishable (p = 0.09). Using the LEfSe analysis (Supplementary Table S6), we determined that OTUs representing Desulfobulbaceae (#1), MND1 (#2), and Desulfurivibrio (#7) were significantly elevated in TRZ samples relative to other taxa. The key OTUs in deeper soil layers were Desulfuromonas (#3), MBNT15 (#4), Sulfurifustis (#5), and Thiobacillus (#6) in Aquifer samples, and Methyloligellaceae (#10), Pseudomonas (#12), and Marinobacterium (#14) from the Clay (Supplementary Table S6). For the surface soil layers, a group of Acidobacteria (Subgroup 6; OTU #s 16, 17, and 22) were identified as significant to Topsoil samples, while Gemmatimonadaceae (#8) was elevated in the Sand, and the significant OTUs from Evaporite samples were Woesia sp. (#8) and an uncultured Alphaproteobacteria (#15) (Supplementary Table S6).

In contrast to the variation across depth, there was no significant difference between communities by sample date alone (AMOVA, p = 0.12). Despite the myriad changes observed in geochemistry, microbial communities in Riverton soils were considerably more consistent at the Phylum level over time (Supplementary Figure S6a). We also observed this trend at finer taxonomic resolution; for example, even at the class and family level, communities within Proteobacteria separated by depth rather than sample date (Figures 3B, 5 and Supplementary Figure S6b). With LEfSe, only five OTUs were highlighted as being significantly distinct between time points (Supplementary Table S6). In the post-snowmelt pre-flood April samples, OTU #s 8 (Gemmatimonadaceae), 11 (Pelagibius, Alphaproteobacteria), and 20 (order Actinomarinales) were significant, and OTU #30 (PLTA13, Gammaproteobacteria) was significant in May samples. The last OTU was significant in August samples after the onset of drought conditions; this OTU (#10) belongs to the family Methyloligellaceae (Supplementary Table S6).

We also analyzed samples based on a combination of soil horizon and date sampled with AMOVA; this showed that only communities from Evaporite-Clay, TRZ, and Aquifer soil samples were significantly different at the taxa level between sampling periods (Supplementary Table S3). These differences primarily occurred in April (wet) versus July (post-flood), August, and September (dry) communities. Strikingly, communities during snowmelt and the start of drying were the most distinguishable throughout the soil column (AMOVA; April and August, p < 0.05) and the TRZ community remained distinct from other layers in September (Supplementary Table S4).

Although overall separation by soil horizon showed differences in variance between communities (HOMOVA, p < 0.05), pairwise comparisons between layers showed that the center of the soil profile was not significantly different (p = 0.30–0.95) and was therefore stable. In fact, the only significant differences were found with Topsoil and Aquifer communities compared against other layers (HOMOVA, p < 0.01 for all comparisons; Table 1). Interestingly, the direct comparison between Topsoil and Aquifer communities was not significant (p = 0.15), likely caused by the overall difference in taxa between these disparate soil horizons (Figure 6).

Whereas communities showed significant differences with depth, sample date could not be used to distinguish samples (HOMOVA, p = 0.20) indicating an overall stable community over time (Supplementary Figure S7). Within individual soil horizons, we were able to see differences more clearly. For example, Evaporite and TRZ communities were overall stable across each month sampled (HOMOVA, p > 0.05), and sand samples only varied from one another significantly post-flood (July; Supplementary Table S3). In contrast, the deepest soil horizons (Clay and Aquifer) as well as the Topsoil were the least stable over time (p < 0.05; Supplementary Table S3). The most stable communities overall appeared to occur in April and September (HOMOVA, p > 0.05; Supplementary Table S4), representing the end members of seasonal hydrological shifts at Riverton, before intense flooding and at the end of the dry season, respectively. We observed the appearance of Firmicutes (Bacilli) in samples from Topsoil, Evaporite, and Sand collected in May and July (flood-influenced months; Supplementary Figure S7). Interestingly, another group within Firmicutes – Clostridia – appeared under drought conditions (August, September), but then within the deepest layers (Clay and Aquifer; Supplementary Figure S7), although overall the Firmicutes exhibited relatively low abundance (1–3% of total community). We also observed increases in relative abundance of Bacteroidetes and a group of unclassified Bacteria in the upper soil layers during this period of flooding. This “unclassified Bacteria” OTU (#40; Supplementary Table S2) does not match known sequences in databases to allow further taxonomic classification, but has been recovered in metagenome sequences from a number of similar floodplain environments (IMG database, Joint Genome Institute³).

The strongest factor influencing community composition was depth or soil horizon, observed using NMDS (Figures 6, 7A). Other geochemical measurements covarying with depth at Riverton show similar separation of taxa, including soluble U (Figure 7B), soil moisture (Figure 7C), and soluble organic C (Figure 7D).

Using CAP analysis, we were able to investigate microbial community composition patterns overlain with geochemical changes. Depth remained the most significant source of differentiation between communities (Supplementary Figure S8a; r = 0.96); even with depth excluded, a clear separation by soil horizon is still observed across CAP1 (23.7% of variance explained; Supplementary Figure S8b). Using a reduced dataset eliminating covarying variables (Figure 8A), we found that surface layers were defined by increased total and soluble organic C (r = −0.41 for both) along CAP1. Deeper soil communities grouped with higher concentrations of total Na (r = 0.41) and metals – including total and soluble U (r = 0.31 and 0.41, respectively); other factors that covaried and increased with depth included total and soluble Mo and soluble Na. Although samples did not separate as clearly along CAP2 (7.3% variance explained), it appears to indicate samples with increased Fe and U concentrations in the negative direction [HCl-extracted Fe(II), r = −0.57; total U, r = −0.65] and total Na (r = −0.43). In the positive direction, CAP2 represents inorganic C (r = 0.63), total K (r = 0.67), and soluble S (r = 0.33). Analysis of variance (ANOVA) performed on CAP results indicate a significant difference between the separated communities (p < 0.01).

In order to investigate the impact of soil moisture on communities, we analyzed a reduced dataset that excluded April samples. CAP analysis on this dataset (Supplementary Figure S9) was very similar, and soil moisture contributed primarily to CAP1 (31.6% variance explained; soil moisture, r = 0.53), increasing with depth. Separation between communities was most distinct between the Sand and the Evaporite-Clay (Supplementary Figure S9a) – which we observed to act as a moisture boundary in the Riverton soil column. Total Fe also significantly contributed to CAP1 on the reduced moisture dataset (r = 0.26), and HCl-extracted Fe(II) again comprised a portion (r = 0.53) of CAP2 (8.8% variation explained; Supplementary Figure S9a). Overall, this analysis indicates that moisture and depth play major roles in structuring microbial communities. Samples (and thus microorganisms) from deeper, wetter soil corresponded to higher metal (Fe, Mo, U) and lower nutrient (C, N) concentrations (CAP1), with an additional separation (CAP2; which is not seasonal) driven by increased inorganic C (carbonates) and K versus increased concentrations of Fe and Na.

We also used CAP analysis to investigate the influence of geochemistry on individual taxa (OTUs); this was run with the Top 100 most abundant taxa in our samples using the same variables as the reduced datasets described above. Although most taxa cluster together toward the center – indicating a more cosmopolitan presence throughout the soil column throughout the season – eight Proteobacteria OTUs appear to have been impacted by increased moisture or metal (Fe, U) concentrations (outliers in Figure 8B and Supplementary Figure S9b). Four of these OTUs are affiliated with Deltaproteobacteria – including Desulfobulbaceae and Desulfuromonadaceae families – which we found to be most abundant in deeper layers (TRZ, Clay, Aquifer; 21–24% of total community versus 7–8% in the upper soil layers; Figures 3B, 5A). Three Gammaproteobacteria OTUs (3.8% of the microbial community) were identified as belonging to the S-cycling genera Sulfurifustis (Acidiferrobacteraceae) and Thiobacillus (Hydrogenophilaceae), and the “MND1” group (Nitrosomonadales). The last distinct OTU (#8, Gammaproteobacteria; 1.2% of sequences) aligns apart from the others (Figure 9), defined instead by increased organic C, Mg, and As, and was classified to the genus level as Woesia – a member of the globally abundant marine benthic group found in coastal and deep ocean sediments (Bienhold et al., 2016; Dyksma et al., 2016). In our dataset, this OTU (#8) is most abundant in Evaporite, TRZ, and Evaporite-Clay layers with higher concentrations of soluble organic C and S.

The geochemical dynamics observed in both porewater (Figure 1) and solid phase (Supplementary Figures S4, S5) concentrations of Mn, Fe, organic C, and U were clearly linked to hydrological dynamics during the 2017 season at the Riverton site (Figure 1) and likely accentuated by temperature changes and root activity. Although we do not have direct temperature measurements from the layers sampled in this study, data from a nearby in situ sensor array indicate that soil temperature in four depths (corresponding to the Evaporite-Clay, TRZ, Clay, and Aquifer layers, i.e., below ∼1 m depth) increased steadily from April to July and then remained stable in those layers until September (Supplementary Figure S10). However, these same layers also remained moist throughout the season, which would have buffered any temperature changes, whereas shallower layers above the Evaporite-Clay dried quickly and, hence, could be expected to experience a much stronger temperature gradient during the hot summer months.

From April to October 2017, we observed seasonal changes at Riverton typical of U-contaminated floodplains in the Upper Colorado River Basin and neighboring watersheds: spring snowmelt and rain caused water table rise and flooding, producing a spike in porewater (and groundwater) concentrations of U, Fe, Mn, (Figure 1) and other elements (data not shown for this study, but similar trends have previously been reported in e.g., DOE Legacy Management, 2012). This was followed by subsequent drying and a drop in aqueous Fe and Mn concentrations through the summer and fall (Figure 1) – a clear indication of re-oxidation (at least of the pore space sampled by rhizons). However, as noted above, the oxidation of soil layers occurred several weeks after the water table dropped below them, whereas the onset of reducing conditions was almost immediate upon water saturation during the spring flood (Figure 1). This hysteresis effect indicates that the TRZ is considerably more resilient to oxidation than to reduction, as has been noted at other similar sites (Noël et al., 2017a, b, 2019; Dwivedi et al., 2018; Pan et al., 2018). The buffering capacity against re-oxidation has been explained by a combination of slow diffusivity and high oxygen consumption rates driven by microbial respiration of organic C (Pan et al., 2018) and abiotic oxidation of reduced species, such as iron monosulfides, FeS (Noël et al., 2017a, b, 2019). This is likely also the case for the Riverton TRZ where the HCl-extractable Fe(II) increased after the flood (Supplementary Figure S4), indicating an accumulation of “reactive” Fe(II) minerals, such as FeS. The accumulation of Fe(II) over time suggests an increasingly pronounced reducing regime that progressed toward sulfate reduction in soil layers that remained moist throughout the season (Supplementary Figure S4). However, the geochemical data is partially contradicted by simultaneous changes in the microbial community in the top and bottom layers of the profile (Supplementary Figure S7 and Supplementary Table S3), which suggests there may be a discrepancy in terms of perceived changes in redox regimes indicated by geochemical and microbial data. This is discussed further below (see section “Microbial and Geochemical Couplings”).

The importance of microorganisms in mediating key biogeochemical processes in soils is well-documented. Particularly in floodplain soils that host TRZs and other organic-rich layers, microorganisms have been shown to produce compounds that control redox states, and determine solubility and mobility of contaminants, including metals such as Fe and U (Davis et al., 2006; Mouser et al., 2015). Previous work has indicated that these microorganisms may be distributed based on the geochemical environment in which they reside and its respective seasonal changes (Hug et al., 2015; Danczak et al., 2016a; Nelson et al., 2019), and also that microbial metabolisms in groundwater-influenced soils are interconnected and interdependent (Wrighton et al., 2014; Anantharaman et al., 2016).

At Riverton, we observed certain taxa influenced by concentrations of U, Mo, or Fe (Figure 8B and Supplementary Figure S9b). After CAP analysis, eight Proteobacteria OTUs appeared as “outliers” relative to other taxa (Figure 9), separating out farther with increased depth and metal concentration. These include 7 of the 8 most abundant taxa in our entire dataset (over 11% of recovered 16S rRNA sequences), and all were classified as either Delta- and Gamma-proteobacteria. Four of the Deltaproteobacteria were most closely related to known sulfur-reducing families – Desulfobulbaceae (OTU #s 1, 7, 31; 3.5% of total sequences; Galushko and Kuever, 2020) and Desulfuromonaceae (OTU #3, 3.1%; Kuever et al., 2015) – commonly detected in floodplain soils (Wrighton et al., 2014; Danczak et al., 2016a, b; Lavy et al., 2019), while the fifth (OTU #4) was a less-characterized group “MBNT15” (1.6%) that has been found in sediment and bioreactors associated with S cycling and metal cycling (Repert et al., 2014; Rezadehbashi and Baldwin, 2018). Desulfuromonas sp. have also been shown to reduce Fe, in addition to S (Kim et al., 2014).

Two Gammaproteobacteria OTUs – Sulfurifustis (OTU #5) and Thiobacillus (OTU #6) – are involved in sulfur oxidation (Kojima et al., 2015; Boden et al., 2020) and are commonly observed in other aquifers and metal-contaminated soils (Liu et al., 2019; Wegener et al., 2019). A third gammaproteobacterial OTU identified as “MND1” (OTU #2) belongs to the same family as a genus of ammonia-oxidizing Bacteria (AOB; Nitrosomonas). However, it was first described in lake sediments rich in ferromanganous micronodules from Green Bay, WI (Stein et al., 2001), and was hypothesized to be involved in sulfur oxidation. Other studies since have found this group to be abundant in soils with high metal concentrations, including U (Satchanska et al., 2004; Sheik et al., 2012).

An additional OTU of interest, OTU #8 belongs to the genus Woesia and was associated with increased organic C (Figure 9). Genomic evidence from a single isolate and a small number of genomes suggest members of Woeseiaceae are facultative anaerobes with either heterotrophic or chemoautotrophic metabolisms, including partial denitrification to N₂O and potentially ferric iron respiration (Du et al., 2016; Mußmann et al., 2017). Further metagenomic sequencing in floodplain soils will be necessary to determine how similar this freshwater sediment group is to its marine relatives.

Importantly, S-cycling organisms (such as those found in the Riverton soil) and their relatives have been shown to mediate redox transformations of Fe and U (Lovely et al., 1993; Fortin et al., 1996; Kwon et al., 2014; Hansel et al., 2015), further highlighting their potential biogeochemical significance within contaminated floodplain soils. Though this cannot yet be verified with 16S rRNA amplicon data alone, it appears that Riverton hosts both established and novel groups of S-, Fe-, and potentially U-cycling organisms.

Depth, and therefore soil horizon/layer, was a stronger contributor than sample date with all analyses performed on this dataset (Figures 2, 6, 7A, 8A). Overall, microbial communities were not significantly different over time (AMOVA, Supplementary Table S3) – indicating the same taxa were present throughout the year but may have been more or less active (as 16S rRNA data does not provide information about activity). Changes we did observe occurred in specific layers, particularly the Topsoil, Clay, and Aquifer horizons (Supplementary Figure S7). This clearly indicates an effect by microorganisms on seasonal changes in geochemistry, especially since differences in microbial community composition were only observed post-flood (or in the case of the Aquifer, during the drought; Supplementary Table S3). Given the vertical locations of these layers (Supplementary Figure S2) relative to the water table, we postulate that the TRZ communities may have profoundly influenced the geochemistry in layers above (during water table rise, transport of microbial products out of TRZ to shallower layers) and below (during receding water table, flushing of microbial products through TRZ to deeper layers). This is supported with the most stable communities residing within TRZ horizons (HOMOVA, Supplementary Table S3). As we observed, the Clay-Evaporite layer forms a moisture boundary within the soil profile, suggesting that precipitation or solute export from the TRZ upward as the water table rises is the primary influence on microbial community composition in Topsoil and Clay horizons. Communities in Topsoil did show clear seasonal changes (Supplementary Table S3), especially with hydrological shifts observed at the site (Figure 1) and accentuated by the largest moisture and temperature fluxes, as well as the prevalence of roots.

Solute transport may also have played a role in the stability of Evaporite and Evaporite-Clay communities (Supplementary Table S3), as these layers already accumulate many of the elements (including metals like U and Fe) transported out of the TRZ with hydrological flux. Therefore, it is reasonable to assume that layers with evaporites consistently contain concentrations of these solutes above some critical threshold that enables microorganisms to maintain their activity. In contrast, communities responding to elevated concentrations of these elements in Sand or Topsoil only overcome such a threshold with intrusion of solutes from TRZ and other layers when the water table rises. This is further supported by the post-flood stability of TRZ, Evaporite, and Evaporite-Clay layers in contrast to all other samples when comparing July 7 and July 25 (Supplementary Table S5); between these sample points, the floodwaters fully receded and upper soil layers completely dried out with the onset of summer drought, indicating that solutes were transported out of the Topsoil and down into the bottom Clay and Aquifer. We know that U and S co-accumulate in reduced sediments including TRZs (Janot et al., 2016; Bone et al., 2017; Noël et al., 2017a) and are spatially coincident in contaminant plumes at Riverton (DOE Legacy Management, 2012). Given that the oxidized states of these elements – U(VI) and sulfate – are much more soluble, redox-active TRZs have the potential to control groundwater concentrations in the Aquifer. Since we did recover a number of abundant S-, Fe-, and U-associated microbial groups from these layers – including Desulfuromonadaceae, Desulfobulbaceae, Thiobacillus, and MND1 – our results suggest that geochemical changes observed were driven in part by microbial communities, in spite of the lack of seasonal signals in the microbial data for the same layers. This implies that geochemical measurements may be more sensitive to seasonal changes in the activity of TRZ microorganisms than 16S rRNA sequencing (which may detect DNA from both active and inactive cells); however, further meta-omic work examining microbial transcripts, proteins, and/or metabolites should provide better insight in temporal changes in these soils.

Importantly, the discrepancies in temporal dynamics between microbial and geochemical data observed in this study tells a story that would not be known unless both types of data had been collected. Based on our dataset, it appears that geochemical data is highly sensitive to detecting the onset of anaerobicity (as implicated by Fe and Mn dissolution, Figure 1), but may overestimate the longevity of reduction as indicated by the prolonged accumulation of HCl-extractable Fe(II) (Supplementary Figure S4) at the end of the season, in spite of the simultaneous increase in relative abundance of aerobes (e.g., Nitrospira sp.). Although 16S rRNA sequencing data is insensitive to rapid changes in microbial metabolic activity and relative abundance is not necessarily linked to activity, a distinct increase in relative abundance is a clear indication of a microbial response to a change in environmental conditions. 16S rRNA data is expected to be less responsive than geochemical data, which explains the lagging response in microbial compared to geochemical data during the onset of anaerobicity. However, we propose that the coincidental increase in abundance of reduced geochemical species and aerobic microbes during the drying period is a reflection of heterogeneous redox conditions occurring simultaneously within a distinct soil horizon that exhibits frequent oscillations between saturated/unsaturated (and relatedly anoxic/oxic) conditions. We further hypothesize that observations indicating both oxic and anoxic redox conditions at a given depth is due to heterogeneous pore size distribution; pores of different sizes within a soil horizon drain and become oxygenated at different “bulk” moisture conditions (i.e., pressures) (O’Geen, 2013; Hu et al., 2018). Hence, a single soil horizon – particularly in clayey and organic-rich soils that typically have a wide range of pore-sizes (O’Geen, 2013; Zaffar and Lu, 2015) – can exhibit co-occurrence of aerobic microbial metabolism with maintenance of anoxic conditions (e.g., protection of oxygen-sensitive FeS) in different pore spaces and between versus inside of aggregates (Renault and Sierra, 1994; Renault and Stengel, 1994; Feng et al., 2002; Keiluweit et al., 2016). Even in water-saturated conditions, the tortuosity and size variability of pores lead to uneven oxygen diffusion rates and, hence, variable oxygenation of pores within the same soil horizon (Fan et al., 2014; Neira et al., 2015; Hu et al., 2018). Finally, differences in seasonal dynamics and stability between microorganisms sampled from soil versus groundwater (Hug et al., 2015; Danczak et al., 2016b; Nelson et al., 2019) imply that a complete view of microbial controls on geochemistry may require both types of samples to be analyzed in floodplain soils, or further work is needed to determine which is more important.