Soil texture and environmental conditions influence the biogeochemical responses of soils to drought and flooding

Climate change is intensifying the global water cycle, with increased frequency of drought and flood. Water is an important driver of soil carbon dynamics, and it is crucial to understand how moisture disturbances will affect carbon availability and fluxes in soils. Here we investigate the role of water in substrate-microbe connectivity and soil carbon cycling under extreme moisture conditions. We collected soils from Alaska, Florida, and Washington USA, and incubated them under Drought and Flood conditions. Drought had a stronger effect on soil respiration, pore-water carbon, and microbial community composition than flooding. Soil response was not consistent across sites, and was influenced by site-level pedological and environmental factors. Soil texture and porosity can influence microbial access to substrates through the pore network, driving the chemical response. Further, the microbial communities are adapted to the historic stress conditions at their sites and therefore show site-specific responses to drought and flood. Soil texture governs chemical responses of soils to water stress while environmental conditions and disturbance history influence microbial community responses in incubations of soil cores from Alaska, Florida and Washington, USA.

T he timing and distribution of precipitation are projected to change under climate change, increasing the frequency of drought in some locations and extreme inundation events, including storms and storm surges, in others 1,2 . Such extreme moisture conditions drive key soil processes in ways we cannot currently predict 3,4 . For example, antecedent moisture conditions have been shown to exert an overwhelming effect on the moisture response of soil carbon dioxide (CO 2 ) fluxes 5-7 , including the increased CO 2 fluxes that occur during rewetting [8][9][10] . The mechanisms behind this amplified respiration response, often described as the "Birch effect", remain ill-defined, as this is a complex phenomenon arising from various physicochemical and biochemical destabilization processes that collectively increase microbial oxidation of soil organic carbon (SOC) 9,[11][12][13] . With the continued intensification of the global water cycle 14,15 , it is imperative to understand how a complex soil biogeochemical system functions under moisture extremes, such as drought and flood, which could pose water stress to certain ecosystem types.
The ionic strength of soil solution varies with moisture content and can be up to nine orders of magnitude greater in dry soils compared to saturated soils 16 . Thus, varying moisture conditions can lead to extremely different chemical environments at the pore scale, causing selective desorption of different carbon (C) molecules from minerals; or otherwise solubilize from conditionally protected physical locations 12 . Changes in soil moisture also affect soil microbial activity. Drought conditions subject soil microbes to osmotic stress in dry pores 17 ; microbes in saturated soils have better access to SOC and nutrients, because of diffusion through the well-connected soil pore network 18,19 . This free exchange is decreased in field-moist soils, in which the hydrologic connectivity is decreased, and some pores remain disconnected 20 .
The objective of this work was to seek a generalized understanding of the impacts of soil water extremes (flood and drought) on soil C cycling across diverse soils, which could be attributed to microbial, chemical, or physical soil traits. We tested two hypotheses: (i) soils subjected to a simulated drought will have greater proportions of complex aromatic C species and will have greater expression of osmoprotectants; (ii) soils subjected to extended saturation will have a greater expression of motility factors (e.g., flagella and gliding motility factors).
We sampled soil cores at three locations from different climatic regions, with different drought/flooding regimes, and with soils of varying physicochemical properties: a silty soil from Caribou-Poker Creeks Research Watershed (CPCRW; Alaska, USA); a clay loam soil from a tidal freshwater near the mouth of the Columbia River (Washington State, USA), subject to frequent inundation; and a sandy soil from the headwaters of the Everglades at the Disney Wilderness Preserve (DWP; Florida, USA), subject to drying and wetting fluctuations. Cores were shipped on ice to the laboratory in Richland, WA, and each soil core was subjected to one of three randomly assigned treatments (Table 1). One set of control cores was maintained at the field-moist water content for 30 days ("field moist"). A second set of cores was immediately saturated and incubated for 30 days ("flood"). A third set of cores was allowed to air-dry until constant weight ("drought"), and then the cores were incubated for 30 days. All treatments were conducted at 21°C in the dark. We had two additional sets of cores as controls-"baseline" field-moist cores that were immediately deconstructed, and "time-zero saturation" cores that were saturated and immediately deconstructed for analysis.

Results and discussion
Pore-water DOC responded strongly to drought and sampling effect. To examine the influence of drying on C distribution and quality, we analyzed the dissolved organic carbon (DOC) in pore waters retained by fine (6-20 µm) and coarse (>200 µm) pore throats 21 . We expected pore-water DOC to increase in the droughtincubated soils, since drought-induced increases in ionic strength are known to destabilize adsorbed C from mineral surfaces 12,22 . This effect was seen only for the Alaska soil, (drought:~150 mg L −1 vs. time zero:~28 mg L −1 , Table 2; Dunnett's test, P < 0.05 for both pore size classes). The pore-water DOC concentrations for the Florida soils appeared to decline numerically in all incubations, but these values were highly variable throughout, and there was no statistically significant pattern (Dunnett's test, P > 0.10 for both pore size classes; Table 2). The Washington soils showed increased pore-water DOC across all three treatments (Table 2), suggesting a strong sampling effect, i.e., merely sampling and incubating these soils contributed to C destabilization. For these soils, DOC did not differ significantly among the three incubation treatments (ANOVA, F = 2.58, P = 0.10), and therefore these soils may be more susceptible to disturbance of handling and laboratory incubation than to the individual moisture treatments, in terms of pore-water DOC concentrations. Molecular diversity of DOC declined significantly during incubation for all soils except drought-incubated Alaska soils ( Table 2).
All our cores were incubated in the laboratory, without litter inputs or amendments. As the incubations progressed, the microbial community would preferentially consume the more labile/biodegradable organic molecules, leaving behind a more uniform, protected SOM pool. This suggests that fresh C inputs, from litter or other sources, are needed to maintain SOM diversity 23 . Drying and rewetting have been known to destabilize protected molecules 12,24 , or even release osmolytes 16 or microbial necromass 25 , and it is likely that this newly available material increased SOM diversity in the silty Alaska soils.
Molecular DOC response varied by site/soil type. We characterized the DOC in the pore-water samples using ultrahighresolution Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR-MS). Van Krevelen compound classification of FT-ICR-MS-resolved peaks is based on molecular H/C and O/C ratios 26,27 (Supplementary Fig. S1a), and it shows that the pore-water DOC for the three sites had very different initial compositions (MANOVA, F = 14.71, P < 0.001). The Alaska soils were dominated by complex lignin-like molecules, the Florida soils by simple amino sugars and carbohydrates, and the Washington soils by protein-and lipid-like molecules (time-zero saturation in Fig. 1, and Supplementary Figs. S2 and S3).
The three soils also varied in their response to moisture changes. The Alaska soils responded most strongly to the drought treatment. Post-drought, new peaks were detected in the high-oxygen aliphatic carbohydrate-and amino sugar-like region ( Fig. 2 and Supplementary Fig. S4), which was previously unique to the native, unmanipulated Florida soils ( Supplementary Fig. S2). With a higher nominal oxidation state (NOSC, see Supplementary Fig. S1b), these Field moisture represents the soil water content when the cores were collected. Drought represents moisture content after cores were air-dried to constant weight. Flood represents moisture content after cores were saturated, using a porous ceramic plate. See "Methods" for more details. a The moisture content in Alaska and Florida drought soils was below detectable levels.
compounds are typically more energetically favorable for microbial consumption and metabolism 28,29 .
In contrast to this, in the Florida soils, simple aliphatic peaks were lost post-incubation and new peaks were detected in the low-O polyphenolic and aliphatic regions (Fig. 2, and Supplementary Tables S2 and S3), which represent complex molecules less favorable for microbial metabolism. This was seen across all three incubation treatments, indicating some consistency in the pore-scale chemical response to incubations under very different moisture conditions. This might suggest a stronger role of the sampling/incubation effect than the individual moisture treatments.
The Washington soils clustered tightly across all treatments, with no strong treatment effects (Fig. 1), suggesting that despite increases in pore-water DOC concentrations, there were no significant qualitative changes in the DOC composition.
Microbial gene expression responded strongly to drought. We analyzed the soils for impacts on the metagenomes and expressed genes (metatranscriptomes) directly linked to hydrologic shifts, i.e., motility and stress due to drought. The soil metagenomes for all treatments separated by site (PERMANOVA, F = 18.22, P = 0.001), but not by treatment (PERMANOVA, F = 0.95, P = 0.444; Supplementary Fig. S5). The metagenome represents only the functional potential and not the activity of the microbial community 30 , and our results suggest that the 30-day treatments did not alter the underlying soil functional potential for each sample site. The metatranscriptomes, on the other hand, represent microbial gene expression, which is more responsive to changes in the environment. Metatranscriptomes separated by site (PERMANOVA, F = 5.20, P = 0.007) and by treatment (PERMANOVA, F = 3.97, P = 0.006), although their interaction was not significant (PERMANOVA, F = 1.41, P = 0.238 ( Supplementary Fig. S5). The difference in the metagenome and metatranscriptome response to treatment suggests that although functional potential remained generally unchanged, the drought and flood treatments may have stimulated the activity of different genes or groups of microbes that were previously less active.
Looking at gene expression (metatranscriptomes) that differed significantly across sites (highest LDA scores, see "Methods"), there was a greater representation of anaerobic processes (reductases and nitrate respiration, Supplementary Figs. S6 and S7) in the Washington soils, whereas genes for osmoprotectant synthesis and stress responses (trehalose and ppGpp synthetases) [31][32][33] were seen in the Florida soils ( Supplementary Fig. S7). These results are consistent with the antecedent moisture history of these sites: the Washington soils experience regular inundation in the river floodplain and are often at/near saturation in the field, whereas the Florida soils Table 2 Dissolved organic carbon (DOC) concentration and molecular diversity for pore water, from coarse (>200 μm) and fine (6-20 μm) pore throats. Data are reported as mean ± standard error. Asterisks denote significant differences for a given treatment compared to the time-zero control (α = 0.05). experience periodic water fluctuations between wet and dry, and being sandy, do not retain much water. We identified differential expression of a number of sporulation and osmoprotectant synthesis genes in response to the moisture extremes of drought and flood. Notably, drought stimulated an increase in eight genes linked with sporulation (spoVID, spoIIAB, spoIIID, spoIVA, spoIIAA, spoVT, spoOA, and spoVD; Fig. 3A). Surprisingly, flood samples also showed increased expression of six genes involved in sporulation (spoIIID, spoVT, spoIVA, spoOA, sigK, and sigE; Fig. 3B), suggesting that sporulation may have been a response to both kinds of stresstoo little as well as too much water. It is also possible that the physical disturbance of sampling caused stress to the microbial communities. Drought also stimulated osmoprotectant synthesis genes, such as treC and betB (Fig. 3A), which are involved in synthesis of trehalose and glycine betaine, respectively 32,34 . These stress responses were seen in the Washington and Alaska soils. Florida soils showed no difference in gene expression due to drought or flood; however, it is important to note that these soils had fewer sequencing replicates due to low RNA recoveries from some samples.
Expression of several gliding motility genes was significantly decreased following drought, but only for the Alaska soils, supporting hypothesis H2 for this site (Fig. 3A). However, these soils also showed increased expression of flagellar motility genes, suggesting that the microbes were likely producing more "swimming" apparatus to combat the low water conditions 35 . Microbes are largely presumed to be fixed to soil minerals 36,37 , although little research has been done to evaluate the relative contributions of fixed vs. planktonic microbes in soils. The presence of gliding motility genes is consistent with the availability of water-filled pore networks that would facilitate microbial distribution through the soil 38 . The Washington soils had a high expression of gliding motility genes even in the drought treatment (Fig. 3A); these soils had a high residual water content (Table 1), and therefore it is likely that these soils may not have achieved an effective drought to elicit microbial response. The Florida soils had low expression of gliding motility at baseline conditions reflective of the low water content of these soils at baseline conditions. Texture and disturbance history drive soil response to water fluctuations. Our goal was to evaluate possible common pathways of soil response to moisture fluctuations. We intentionally selected sites with different pedology and climatic conditions to see if, despite contrasting features, these soils would show consistent responses to drying and saturating conditions. However, this was not the case, and the complex results we present here must be considered within the context of the origin and structure of the soil at each site.
The Alaskan site has limited C inputs, a history of freezing stress, and silt-loam soils with limited C inputs. The history of freezing stress at this site can be considered as similar to drought history in some respects (e.g., lack of freely available water), but freezing does not have the same physicochemical effects as drought (e.g., changes in ionic strength and desorption of C), and the annual freeze-thaw regime is highly regular and predictable, unlike drought. These soils were~15% clay (Supplementary  Table S1), which likely played a role in SOC stabilization 39,40 . As a result, these soils would have a high potential for SOM desorption/destabilization when experiencing drought conditions 12,22 . While the drought increased the abundance of more saturated and oxidized molecules, the exact origin of these molecules is uncertain. It is possible that these molecules were previously held to mineral surfaces and then desorbed 12,22,24 , or that these were formed by microbial metabolism of complex molecules 41 .
The microbial community in the Florida soil is likely adapted to moisture extremes, since the field site regularly cycles between wet and moist, but its soils also drain well due to the sandy texture. Since these microbes are presumably adapted to moisture fluctuations, it is likely that this adapted community will continue to metabolize and mineralize C, preferentially consuming simple aliphatic molecules. The Florida soils are sandy, and it is likely that microbes had greater access to simple molecules usually seen more abundantly in macropores 21 . These simple molecules were consumed in the Florida soils, causing an enrichment of ligninlike and complex aromatic molecules across all treatments. The low clay content also suggests that the contribution of organo-mineral interactions to SOM stabilization would be quite low in these soils, and thus that the SOC would be more sensitive to physical disturbances during sampling.
The Washington State soils were sampled along a river floodplain, and the microbial communities are adapted to daily fluctuating saturated conditions. In fact, the field moisture content for these soils was very similar to the saturation water content (Table 1), explaining the lack of response to the flood treatment. In a wet system, such as the Washington cores sampled, flood would not be expected to stress the system. The minimal response of the soil chemistry to drought (Fig. 2) appears to be a consequence of the physical organization and pore structure of these soils. The size distribution of macropores (>50 µm) was approximately similar for the three soils (Supplementary  Table S4), but the "dry" condition represented a residual water content of~60% w/w for Washington, in stark contrast the Alaska and Florida soils ( Table 1). The Washington soils also had greater pore and hydrologic connectivity ( Supplementary Fig. S8), allowing for stronger buffering against moisture stress. Despite minimal effect on the pore-scale chemistry, the Washington drought soils exhibited the CO 2 flush usually seen upon rewetting dry soils due to physicochemical destabilization of protected C 9,11-13 , microbial cell lysis 42 , and consequent changes in microbial activity 43 (Supplementary Figs. S9-S11). This suggests some transient response to drought, likely from the macropores, since (as noted) the micropores held~60% moisture under these "drought" conditions. There thus seems to be a disconnect between the core-scale gas flux response and the pore-scale molecular response for these soils, and may be a function of the high clay content and pore size distribution in these soils.

Conclusions and implications
Our first two hypotheses were partially supported, as (i) drought increased sporulation and osmoprotectant expression in the Alaska and Washington soils, and (ii) although flood did not significantly alter motility genes, drought did decrease the expression of gliding motility genes in the Alaska soils. Contrary to our predictions (hypothesis (i)), drought did not increase the abundance of complex aromatic compounds, but there was a distinct shift in the pore-water chemistry of the Alaska soils to more oxidized molecules.
Our experiment provides insight into potential controls on soil C processes, while demonstrating a long-recognized problem: it is difficult to unambiguously identify fully generalizable soil responses because of the complexity and heterogeneity of soils. This is true even for studies with very high numbers of samples or study sites. Nonetheless, our results provide insights on some potential pedologic and environmental controls on soil C processes. In particular, as we describe in Fig. 4, soil texture-and as a consequence, pore size distribution and spatial distribution of water-may drive the chemical response, whereas environmental conditions and disturbance history may drive the microbial response, of soils to new hydrologic regimes.
Related to the latter point is the concept of ecological stress 44 , as changes in the microenvironment may not always be stressful for the system. We operationally define drought (air-drying to constant weight) and flood (water imbibition to constant weight) in this experiment, and these terms may mean different things for different soils, and this may influence how each soil responds to the experimental treatment. For instance, even though all three soils were subjected to flood (i.e., 100% saturation), the amount of water imbibed differed by site. For Washington soils, with high field moisture content, the flooding treatment would not be expected to stress the system. Similarly, the Florida soils had a field moisture content of <10% w/w (Table 1 and Fig. 4), and would not respond strongly to a simulated drought. We are cautious not to overinterpret our results; the sample size is small, and we do not account for other variables, such as mineralogy, litter quality, etc. Nonetheless, this study allows us to propose an initial experimental framework to be built upon by future research and underlines the utility of incorporating environmental history and soil physicochemical properties, when studying biogeochemical processes to understand soil C-water dynamics at the pore-to-core scale. Each soil was subjected to one of three treatments, and each treatment was replicated on five cores: (a) A set of cores was immediately saturated from below and incubated for 30 days ("flood"). (b) A second set of cores was allowed to air-dry until constant weight was reached for five consecutive days, after which they were incubated at dry conditions for 30 days ("drought"). (c) A third set of cores was maintained at the field-moist water content, for 30 days ("field moist"). The field-moist treatment was used to determine soil C processes under partially saturated conditions occurring in the field, reflecting discontinuous resource islands in the soil cores. At the end of the 30-day incubations, the cores were saturated from below and held at saturated conditions for 24 h. This was done to capture the CO 2 fluxes that typically occur during rewetting [8][9][10] , and also to prepare the cores for pore-water extraction (see below). We had two additional sets of cores as controls-"baseline" samples that were immediately deconstructed, and a "time-zero saturation" group of cores that were saturated and immediately deconstructed for analysis. All incubations were conducted at 21°C, in the dark, under ambient relative humidity (18%). We had five replicates (n = 5) per site-treatment combination.
We monitored CO 2 and methane (CH 4 ) production every 3 days during the 30day incubations, and hourly during the post-incubation saturation. The gas concentrations were measured in the cores' headspace, using a G2301 Picarro GHG analyzer (Picarro, Sunnyvale, CA, USA). Fluxes were computed from the concentration changes according to the equation where A is the flux (µmol g soil −1 s −1 ), dC/dt the rate of change in gas concentration (mole fraction s −1 ), V the total chamber volume (cm 3 ), M dry soil Fig. 4 The three soils in this experiment, represented as a function of water content and texture. We show the range of soil moisture experienced by these soils during the experimental incubations-the larger, gray circles represent field-moist water content, and the smaller circles represent moisture content for drought (red)/flood (blue) treatments. Soil texture appears to drive the chemical response by controlling porosity and water retention, and field water content drives biological stress response to moisture fluctuations. The Washington soils had a high field moisture content, and therefore showed minimal response to the "flooding" treatment. On the other hand, the Florida soils had low field moisture content, and therefore showed minimal response to the "drought" treatment. mass (g), Pa atmospheric pressure (kPa), R the universal gas constant (8.3 × 10 3 cm 3 kPa mol −1 K −1 ), and T air temperature (K).
Pore-water extraction, core deconstruction, and analysis. At the end of the incubation, pore water was sampled following the procedure of Bailey et al. 21 . Cores were transferred onto individual 100 kPa Tempe Pressure Cell units fit with a high flow ceramic plate (Soil Moisture Equipment Corp. Goleta, CA, USA) to sequentially collect pore waters with suctions of −1.5 and −50 kPa, using a pump with a PCD Dual-Valve pressure controller (Alicat Scientific, Tucson, AZ). These suction values were used because they correspond to pore size diameters of 200 µm (coarse) and 6 µm (fine), respectively, using the Kelvin equation 48 . Pore water was collected into borosilicate vials for 24 h, or until flow ceased at each suction setting and stored at −20°C until analysis. Pore-water samples were filtered through a 0.22 µm Sterivex filter immediately prior to being analyzed. Porewater samples were analyzed for total DOC via combustion catalytic oxidation (TOC-5000A TOC Analyzer, Shimadzu). The molecular composition of the dissolved C was characterized by electrospray ionization coupled with FT-ICR-MS. The samples were desalted by solid-phase extraction with PPL cartridges 49 . Detailed description of the analysis and data handling are provided below.
Cores were deconstructed following pore-water removal and the soils were sieved through 4 mm mesh to remove roots and rocks. The sieved soil was subsampled for the measurements outlined below.
Water soluble organic carbon (WSOC) was extracted from air-dried soil (1:5 soil volume: Milli-Q water) by shaking at 4°C on a reciprocal shaker set at 200 r.p.m. high at 200 r.p.m. for 2 h, centrifuging for 5 min at 10,000 r.c.f., and then filtering through 0.22 µm Sterivex™ filters. An aliquot (1.5 mL) was stored at −20°C for downstream analyses of FT-ICR-MS-used to measure changes in soil soluble carbon chemistry. The remaining sample was stored at −20°C until measured for total dissolved carbon on a Shimadzu TOC-5000A. Soils were air-dried for total C and total organic C (TOC) and were analyzed on a VarioEL Cube Elemental Analyzer (Elementar Analysensysteme GmbH, Langenselbold, Germany).
DNA and RNA were extracted from soil using the Qiagen PowerSoil DNA and Qiagen PowerSoil RNA extraction kit respectively, according to the manufacturer's instructions. DNA concentrations were determined using a Qubit 2.0 Fluorometer using the dsDNA HS (ThermoFisher Scientific). Shotgun metagenome sequencing on an Illumina HiSeq 1500 was used to estimate microbial potential.
RNA extracts were further purified with a PowerClean Ⓡ RNA Clean-Up Kit (MO BIO, Carlsbad, CA) and analyzed for concentration, ribosomal RNA (rRNA) ratio, and RNA integrity number (RIN) on a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA), using the RNA 6000 Nano kit (Agilent Technologies, Santa Clara, CA). Only samples with a RIN (RNA integrity) value of 8 and above were prepared for sequencing. rRNA was depleted prior to sequencing using the Ribo-Zero magnetic gold kit (Epidemiology). Bacterial mRNA thus obtained was purified using AMPure RNAClean XP beads. cDNA synthesis and library preparation were carried out using random hexamers and the ScriptSeq v2 RNA-Seq Kit (Epicenter). Metatranscriptome libraries were sequenced on the HiSeq 2500 platform (Illumina, San Diego, CA) on a paired-end flow-cell at 2 × 250 cycles.
FT-ICR-MS analysis. FT-ICR-MS analysis was performed on a 21 T FT-ICR mass spectrometer at the Environmental Molecular Sciences Laboratory (EMSL) in Richland, WA 50 . Ninety-six individual scans were averaged for each sample and internally calibrated using OM homologous series separated by 14 Da (-CH 2 groups). The mass measurement accuracy was <1 p.p.m. for singly charged ions across a broad m/z range (200 < m/z < 1200). Chemical formulas were assigned using in-house software based on the Compound Identification Algorithm described by Kujawinski & Behn 51 and modified by Minor et al. 52 . Chemical formulas were assigned based on the following criteria: signal-to-noise (S/N) > 7, and mass measurement error < 1 p.p.m., taking into consideration the presence of C, H, O, N, S, and P and excluding other elements.
Further processing of these data was done using the fticrrr workflow in R 53 . FT-ICR-MS spectra were classified into eight biomolecular groups, referred to as FT-ICR compound classes, using the Van Krevelen classification based on molecular O/C and H/C ratios 26 Supplementary  Fig. S1a. FT-ICR compound classes are tentative classifications as they are solely based on the O/C and H/C ratios from the molecular formula, not the molecular structure. As such, it would be more accurate to describe compounds as lipid-like or carbohydrate-like. For simplicity, we refer to each compound class, however, as lipids, tannins, proteins, and so on. Relative abundance values were calculated from count values associated with each observed biomolecule group normalized by the total number of C molecules identified.
FT-ICR-MS data were analyzed using multivariate analysis of variance (MANOVA). Treatment effect on WSOC composition was determined using group-wise relative abundance as the response variable. Due to the significant interaction of treatment and pore size class, the 1.5 and 50 kPa pore size classes were analyzed separately. Principal components analysis (PCA) was performed using the stats package in R, and was visualized using the ggbiplot package 54 .
Metagenome and metatranscriptome data processing. Metagenomic and metatranscriptomic reads were cleaned using bbduk, and paired-end reads were merged using bbmerge. Reads were then annotated to TIGRFAMS gene families using MaxRebo (McCue, unpubl.). For DNA and RNA analyses, we first removed all ribosomal genes before normalizing by relative abundance. We then removed any genes that had <0.001% relative abundance in all samples.
Due to the nonparametric nature of metagenome and metatranscriptome data, PCAs for these data were generated with Euclidean distances using vegan v2.5-5 (ref. 55 ), and significance testing was performed using the adonis function in vegan. To identify discriminatory genes across the three sites, we used linear discriminant analysis effect size (LEfSe 56 ) to identify the top 20 genes that were significantly enriched in each location and/or treatment, and had a logarithmic score >2. For gene expression analysis comparing drought and flood to baseline, we used the DESeq2 package 57 to identify significant transcripts and calculate log2fold change. For heatmap coloring, we normalized gene abundances using the normTransform() function from the DESeq2 package, and colored the resulting values using the cividis color palette 58 .
Additional soil characterization. The time-zero soils were also analyzed for pH, electrical conductivity, basic nutrients (P, K, Ca, Mg, etc.), and particle size distribution at the Oregon State University Central Analytical Laboratory (Supplementary Table S1).
Pore size distribution was determined using X-ray computed tomography on an X-Tek/Metris XTH 320/225 kV scanner (Nikon Metrology, Belmont, CA). Data were collected at 110 kV and 265 μA X-ray power. The core samples were rotated continuously during the scans with momentary stops to collect each projection (shuttling mode), while minimizing ring artifacts. A total of 3142 projections were collected over 360°with 0.5 s exposure time and four frames per projection. Image voxel size was 28 µm. The images were reconstructed to obtain 3D data sets using CT Pro 3D (Metris XT 2.2, Nikon Metrology). Representative slice and 3D images were created using VG Studio MAX 2.1 (Volume Graphics GmbH, Heidelberg Germany). Image processing and porosity analysis (including pore volume segmentation and pore analysis) was carried out using ImageJ 1.51k (National Institute of Health, USA).
The water retention characteristic (desorption/drying curve) was determined by saturating soil samples and subsequently drying by evaporation at room temperature. Changes in weight and water tension were measured using a HYPROP Ⓡ (METER Environment Inc., USA) every 10 min for 7 days. The data were fit using the van Genuchten equation 59 to generate relationships of water tension vs. water content for the three soils.