Salinization and sedimentation drive contrasting assembly mechanisms of planktonic and sediment‐bound bacterial communities in agricultural streams

Agriculture is the most dominant land use globally and is projected to increase in the future to support a growing human population but also threatens ecosystem structure and services. Bacteria mediate numerous biogeochemical pathways within ecosystems. Therefore, identifying linkages between stressors associated with agricultural land use and responses of bacterial diversity is an important step in understanding and improving resource management. Here, we use the Mississippi Alluvial Plain (MAP) ecoregion, a highly modified agroecosystem, as a case study to better understand agriculturally associated drivers of stream bacterial diversity and assembly mechanisms. In the MAP, we found that planktonic bacterial communities were strongly influenced by salinity. Tolerant taxa increased with increasing ion concentrations, likely driving homogenous selection which accounted for ~90% of assembly processes. Sediment bacterial phylogenetic diversity increased with increasing agricultural land use and was influenced by sediment particle size, with assembly mechanisms shifting from homogenous to variable selection as differences in median particle size increased. Within individual streams, sediment heterogeneity was correlated with bacterial diversity and a subsidy‐stress relationship along the particle size gradient was observed. Planktonic and sediment communities within the same stream also diverged as sediment particle size decreased. Nutrients including carbon, nitrogen, and phosphorus, which tend to be elevated in agroecosystems, were also associated with detectable shifts in bacterial community structure. Collectively, our results establish that two understudied variables, salinity and sediment texture, are the primary drivers of bacterial diversity within the studied agroecosystem, whereas nutrients are secondary drivers. Although numerous macrobiological communities respond negatively, we observed increasing bacterial diversity in response to agricultural stressors including salinization and sedimentation. Elevated taxonomic and phylogenetic bacterial diversity likely increases the probability of detecting community responses to stressors. Thus, bacteria community responses may be more reliable for establishing water quality goals within highly modified agroecosystems that have experienced shifting baselines.

In the MAP, we found that planktonic bacterial communities were strongly influenced by salinity. Tolerant taxa increased with increasing ion concentrations, likely driving homogenous selection which accounted for ~90% of assembly processes. Sediment bacterial phylogenetic diversity increased with increasing agricultural land use and was influenced by sediment particle size, with assembly mechanisms shifting from homogenous to variable selection as differences in median particle size increased.
Within individual streams, sediment heterogeneity was correlated with bacterial diversity and a subsidy-stress relationship along the particle size gradient was observed.
Planktonic and sediment communities within the same stream also diverged as sediment particle size decreased. Nutrients including carbon, nitrogen, and phosphorus, which tend to be elevated in agroecosystems, were also associated with detectable shifts in bacterial community structure. Collectively, our results establish that two understudied variables, salinity and sediment texture, are the primary drivers of bacterial diversity within the studied agroecosystem, whereas nutrients are secondary drivers. Although numerous macrobiological communities respond negatively, we observed increasing bacterial diversity in response to agricultural stressors including salinization and sedimentation. Elevated taxonomic and phylogenetic bacterial diversity likely increases the probability of detecting community responses to stressors. Thus, bacteria community responses may be more reliable for establishing water quality goals within highly modified agroecosystems that have experienced shifting baselines.

| INTRODUC TI ON
Conversion of natural landscapes to agricultural land, which currently covers ~40% of Earth's land surface, is accelerating globally to meet growing demands for food, fiber, and energy production (Alexandratos & Bruinsma, 2012). Although humans depend on agricultural resources for survival, agricultural practices alter ecosystems and supporting ecosystem services including water quality and nutrient cycling that are also necessary for sustaining civilization (Dale & Polasky, 2007). Streams in agriculturally dominated ecosystems or agroecosystems experience excess nutrient runoff (e.g., organic carbon [OC], nitrogen [N], and phosphorus [P]) from fertilizers that cause eutrophication, hypoxia, and harmful algal blooms resulting in fish kills and contaminated drinking water (Danial et al., 1998;Watson et al., 2016). Excess fertilizer application can also increase concentrations of dissolved ions in streams, known as freshwater salinization, which can affect biodiversity and drinking water quality Kaushal, 2016). Enhanced soil erosion from agricultural practices like tillage further increases nutrient concentrations and degrades habitat by increasing turbidity which reduces light availability for primary producers and buries benthic habitats that support biodiversity of communities like fishes and macroinvertebrates. Additionally, soil-bound bacteria can be transported to streams via erosion, which may alter both bacterial diversity and functional potential of the stream community (Le et al., 2020). The reliance of civilization on both agricultural products and ecosystem services presents a significant challenge for environmental management of increasingly modified agroecosystems. Striking a balance between increasing agricultural production while maintaining or enhancing ecosystem integrity is thus a critical concern that requires a comprehensive understanding of key drivers of ecosystem processes. Bacterial communities drive many important functions that support ecosystem services, but the effects of agriculture on stream bacterial communities remains understudied.
Bacteria are involved in many basal ecosystem functions including numerous nutrient cycling pathways that drive Earth's biogeochemical cycles (Falkowski et al., 2008). Ecosystem resilience to anthropogenic stressors thus depends largely on the phylogenetic and functional diversity of inhabitant bacterial communities. In aquatic systems, bacterially-mediated nutrient cycling pathways are particularly important in regions prone to elevated nutrient runoff like agroecosystems. For example, bacteria influence the bioavailability of key nutrients like N by controlling a diverse set of N cycling pathways (Burgina & Hamilton, 2007;Canfield et al., 2010). Excess bioavailable N may be removed from streams through complete microbial denitrification, which converts bioavailable nitrate to inert N 2 gas (Mulholland et al., 2008;Schlesinger, 2009). In contrast, nutrient loading to agricultural water bodies in the spring can lead to high algal production and eventual limitation in the summer, potentially stimulating N-fixation (Marcarelli & Wurtsbaugh, 2009;Nifong et al., 2022). Decomposition of herbicides and pesticides can also be accelerated by bacteria that possess the unique cellular machinery required to metabolize these compounds, helping reduce toxic effects to wildlife and lower concentrations in drinking water (Aislabie & Lloyd-Jones, 1995). However, certain bacterially produced pesticide metabolites can actually be more toxic than their parent compounds (Caillon & Schelker, 2020;Ji et al., 2020;Ruiz-Gonzales et al., 2015). Additionally, bacteria have higher biodiversity and their occurrence is typically thought to be less affected by habitat than other stream assemblages like fishes or macroinvertebrates given their high metabolic diversity, making them potentially useful for monitoring and establishing biologically based water quality goals in regions where habitat is limiting (Falkowski et al., 2008;Locey & Lennon, 2015). Given the diverse structural and functional roles that bacteria play in stream ecosystems, identifying factors that modulate regional patterns of bacterial diversity in agricultural watersheds is a necessary step for improving stream management efforts.
Deterministic processes include selection by abiotic factors that constrain community membership based on fitness as well as species interactions. Selection can either increase or decrease regional diversity depending on whether selective pressures are consistent or variable across the landscape, respectively (Dini-Andreote et al., 2015;Stegen et al., 2015;Vellend, 2010). Conversely, stochastic processes include probabilistic colonization, dispersal, and ecological drift (Hubbell, 2001). High rates of dispersal can homogenize regional diversity and even mask the effects of selection and drift whereas low dispersal rates can increase regional diversity by allowing selection and/or drift to differentiate communities across space and time (Dias, 1996;Leibold et al., 2004). Collectively, the relative influence of these processes dictates community membership at both local and regional scales and can potentially alter ecosystem functions when functional redundancy cannot compensate (Louca et al., 2018;Stegen et al., 2012Stegen et al., , 2013. Currently, our understanding of bacterial community assembly in dendritic systems like stream networks is not well understood. Determining how agricultural practices impact the balance of deterministic and stochastic community assembly processes could improve mechanistic understanding of how agriculture modulates stream ecosystem functions. The goal of this research was to characterize stream bacterial community responses to agriculture by (1) determining the main drivers of bacterial diversity across streams in the Mississippi Alluvial Plain (MAP), an intensely cultivated region in the United States, and (2) inferring how the main drivers of bacterial diversity K E Y W O R D S 16S amplicon sequencing, agriculture, assembly mechanisms, bacterial diversity, salinization, sedimentation impact assembly mechanisms of planktonic and sediment bacterial communities. Given that agriculture is one of the most widespread land uses globally and is predicted to increase with a growing human population (Ramankutty et al., 2018), the MAP provides an agroecosystem endpoint for studying responses to future land use change and agricultural intensification (Tank et al., 2021). By Identifying linkages between agriculture stressors and bacterial diversity, we aim to improve predictions of when, where, and how agricultural practices influence stream ecosystems.

| Study site
The Mississippi Alluvial Plain (MAP) ecoregion within Mississippi (U.S.), commonly referred to as "the Delta", is the historic floodplain of the Mississippi River. Prior to European colonization in the late 1800s, this region was described as dense, bottomland hardwood forest consisting of cypress (Taxodium distichum) and tupelo gum species (Nyssa aquatica and Nyssa sylvatica). During the late 1800s and early 1900s, this region was clear cut for valuable timber and subsequently drained for agriculture due to the nutrient-rich soil deposited by the Mississippi River (Ochs et al., 2023). There are no data regarding water quality or stream ecosystem properties from this region prior to the expansion of agriculture. Currently, the MAP is ~65% agriculture dominated by row crops including cotton, corn, soybean, and rice. The remaining land is ~25% wetland that is too wet for agriculture and 10% primarily consisting of small cities and towns (Yasarer et al., 2020). Compared to other regions in Mississippi, intensive agriculture in the MAP has resulted in elevated concentrations of dissolved ions, OC, and P, but not N in surface waters potentially due to high denitrification rates that remove excess N from agricultural fertilizers (Douglas Shields et al., 2011;Taylor et al., 2015Taylor et al., , 2023.
The hydrology of the MAP has also been altered substantially.
Levees were built in the late 1800s and expanded throughout the 1900s to prevent flooding, significantly reducing connectivity of streams in the MAP with the Mississippi River (Alexander et al., 2012).

Four main rivers in the adjacent US EPA level III Mississippi Valley
Loess Plains (MVLP) ecoregion (Chapman et al., 2004) including the Little Tallahatchie River, the Yocona River, the Coldwater River, and the Yalobusha River, have also been dammed to control the movement of water and prevent flooding of MAP agricultural lands (Reuss, 1982). A significant amount of stream flow during the summer also stems from irrigation water, much of which is pumped from groundwater (Yasarer et al., 2020).

| Site selection
Wadable streams were selected based on historical nutrient concentrations collected by the U.S. Geological Survey and the Mississippi Department of Environmental Quality (MDEQ). Water quality data used to support this study are publicly available from the USGS National Water Information System (U.S. Geological Survey, 2023).
The U.S. Geological Survey station names and numbers of the 28 sampling stations used for analyses are given in Table S1. Streams were chosen to represent the widest range of chemical properties including C, N, and P concentrations and specific conductance. Sites were also chosen to adequately cover the latitudinal gradient in the MAP. To better understand the effects of adjacent ecoregions on MAP bacterial communities, a subset of samples was collected from streams located within the MAP but with watershed area

| Water chemistry
Surface water samples for total organic carbon (TOC), total nitrogen (TN), and total phosphorus (TP) were collected and transported to the laboratory for analysis according to MDEQ surface water-quality monitoring standard operating procedures (MDEQ, 2017). Surface water sampling consisted of a vertical instantaneous grab sample collected from the channel thalweg. Nutrient samples were acidified on collection with H 2 SO 4 , and all water samples were stored on ice and shipped to the laboratory within 24 h, where they were maintained at 4°C prior to analysis. Water temperature, dissolved oxygen, specific conductance (SC, a proxy for salinity), and pH were measured and recorded from the same depth as water samples using a calibrated multi-parameter sonde. Water samples were analyzed at the MDEQ laboratory in Pearl, Mississippi, in accordance with EPA Methods for Chemical Analysis of Water and Wastes (EPA, 1983) and the Standard Methods for Examination of Water and Wastewater (Eaton et al., 2005). Turbidity was measured using a Hach 2100 portable turbidimeter (Hach) following manufacture protocols. Samples for ash-free dry weight were collected by filtering 100-400 mL of stream water, depending on suspended particle concentrations, onto duplicate pre-combusted (500°C for 4 h) and pre-weighed 0.7 μm GF/F filters. Filters were weighed after drying at 105°C in a drying oven overnight. Dried filters were subsequently ashed at 550°C for 4 h and weighed. Ash-free dry weight was calculated by subtracting the ashed filter weight from the dried filter weight and dividing by the filter volume.

| Bacterial sampling
We collected water for characterization of planktonic bacterial communities in acid-washed, autoclaved, and triple-rinsed 1-L HPDE bottles. Water samples were filtered on site through 0.22 μm polycarbonate filters using a vacuum pump apparatus. Filters were placed in sterile Eppendorf tubes and kept on dry ice until returned to the US Department of Agriculture-National Sedimentation Laboratory (Oxford, MS) where they were stored at −80°C until analysis. Although field blanks were not collected, the filtering apparatus was thoroughly cleaned between sites using 70% ethanol, 10% bleach and DI water. Forceps were also flame-sterilized between uses and a sterile nitrile glove was placed over the filter apparatus during filtration to prevent contamination from airborne bacteria.
Mississippi Alluvial Plain (MAP) stream sediments consist primarily of variable proportions of sands, silts, and clays (Wren et al., 2008).
For characterization of sediment-bound bacterial communities, we collected five sediment subsamples from each stream reach that were representative of the relative proportions of sands, silts, and clays within the reach. We collected sediment samples by inverting 50-mL LDPE falcon tubes (Sterile, DNA free, SA = 6.66 cm 2 ) directly into the top 5 cm of sediment. Subsamples were pooled into sterile whirlpacks, homogenized, and stored on dry ice until we returned to the US Department of Agriculture-National Sedimentation Laboratory where they were kept at −80°C until analysis.

| DNA extraction and sequencing
We extracted and purified DNA from water and sediment samples using the Qiagen PowerWater and PowerSoil Pro kits (Qiagen, Inc.), respectively, following manufacturer's protocols. We amplified the V4 region of the 16S rRNA gene with polymerase chain reaction (PCR) according to the Earth Microbiome Project (Thompson et al., 2017) protocol using indexed 515F (5′-GGA CTA CNV GGG TWT CTA AT-3′) (Parada et al., 2016) and 806R (5′-GTG YCA GCM GCC GCC GTA A-3′) (Apprill et al., 2015) primers. Briefly, each reaction well contained 23.5 μL Hot Start Mastermix (Thermo Fisher Scientific), 0.5 μL of the 515F forward primer and 806rb reverse primer and 1 μL of DNA template. We amplified each sample in triplicate in separate wells, and a non-template control with Mastermix and primers but no DNA template was run for each sample to ensure no contamination occurred. The thermocycler was initially held at 94°C for 3 min to denature DNA. After the denaturing step, the thermocycler ran 35 cycles of 94°C for 45 s, 50°C for 60 s, and 72°C for 90 s. After 35 cycles, the thermocycler was held at 72°C for 10 min. After PCR, we pooled triplicate reactions from each sample and purified PCR products using the Qiagen Quiquick purification kit following manufacture protocols (Qiagen, Inc.). We measured the DNA concentration of purified 16S rRNA gene amplicons using a Qubit fluorometer following manufacture protocol (Thermo Fisher Scientific

| DNA sequence quality filtering and processing
DNA sequences were quality filtered in R using DADA2 (v1.22.0) (Callahan et al., 2016). Upon visual inspection of quality scores, we determined that averaged quality scores for both forward and reverse reads for all samples were high and did not warrant trimming or truncating before error modeling. Individual reads with expected errors higher than 2 (MaxEE = 2) were discarded and remaining reads were truncated at the first instance of a quality score ≤2 (truncQ = 2) using the filterAndtrim function. After quality filtering, error rates were learned using the learnErrors function and errors were removed using the dada function. Following sample inference, paired reads were merged, and chimeras were removed using the consensus method of the removeBimeraDenovo function to return a final amplicon sequence variant (ASV) table. We taxonomically classified ASVs using the assignTaxonomy function which uses a naïve Bayesian classifier (Wang et al., 2007) and the Silva training set (v138.1) (Quast et al., 2013). Lastly, we built a phylogenetic tree using an internal maximum likelihood method with the phangorn (v2.9.0) (Schliep, 2011) and DECIPHER (v2.22.0) (Wright, 2016) packages in R (R Core Team, 2021).

| Sediment characterization
In addition to sediment samples being used for DNA sequencing, samples were also analyzed for nutrient concentrations and texture at the US Department of Agriculture-National Sedimentation Laboratory. Prior to analyses, sediments were dried, ground, and pre-sieved through a 2 mm sieve to remove any coarse material. We measured sediment P (Sed P) concentrations using a Melich-3 extraction method. Briefly, 1 g of dried sediment was digested using an extract composed of 0.2 M acetic acid, 0.25 M ammonium nitrate, 0.015 M ammonium fluoride, 0.013 M nitric acid, and 0.001 M ethylene diamine tetraacetic acid. Phosphorus content of extracted sediment samples was then determined using the standard ascorbicammonium molybdate method using a MULTISKAN SkyHigh microplate reader (Thermo Fisher Scientific). We determined percent carbon (C) and N of 1 g of dried sediment using a CN analyzer (Elementar). We determined sediment texture and particle size distribution using a digital hydrometer and the integral suspension pressure method (ISP+) (Pario, METER AG) following manufacture protocols (see Durner & Iden, 2021 for theory). Briefly, dried sediments were shaken overnight in 5% sodium hexametaphosphate to disperse aggregates. The dispersed sample was transferred to a 1-L graduated cylinder with a valve for draining effluent. We homogenized the sample for 1 min prior to inserting the digital hydrometer into the sample tube. The density of the sediment solution was continuously measured for 2.5 h. After 2.5 h, we drained the top portion of the effluent containing the clay fraction through the valve into a pre-weighed, 250-mL beaker to determine the mass of the clay fraction in the sample. We wet-sieved the sand fraction through 500-, 250-, 125-, and 63μm sieves to determine the mass of different sand fractions. Based on the mass of the clay fraction in the effluent and the mass of different sand fractions, an inverse model of suspension pressure was fit using the Pario software to determine the continuous size distribution and percent sand, silt, and clay. Percent sand, silt, and clay in each sample were used to calculate a metric of sediment evenness based on Renyi's entropy equation (analogous to Simpson's evenness index) (Joust, 2007). Higher values indicate that sediments were composed of a more even amount of sand, silt, and clay, while lower values indicate that sediments were dominated by one texture class. Sediment data for each site are property of USDA-ARS and are publicly available (DeVilbiss et al., 2023).

| Analysis of bacterial communities
2.9.1 | Bacterial diversity and community structure Prior to phylogenetic alpha diversity analyses, samples were rarefied to the number of reads in the sample containing the lowest read number (21,443) using the rarefy_even_depth function in the phyloseq package (McMurdie & Holmes, 2013). Rarefaction curves indicate that diversity was well sampled for both sediment and planktonic samples ( Figure S1). Faith's phylogenetic diversity (PD) (Faith, 1992) was calculated based on the phylogenetic tree using the PD function in the picante package (1.8.2) (Kembel et al., 2010). We used multiple rarefaction to estimate ASV richness and Shannon diversity using the phyloseq_mult_raref_div function in the metagMisc package (https://github.com/vmikk/ metag Misc). For multiple rarefaction, samples were rarefied 1000 times to the sample containing the lowest read number (21,443) and estimated diversity values are reported as the mean ± 1 SD.
We visualized differences in bacterial community structure of both planktonic and sediment communities separately using nonmetric multidimensional scaling (NMDS) using Bray-Curtis dissimilarity. Prior to visualization, differences in sampling depth were normalized using cumulative sum scaling in the metagenomSeq package (Paulson, Olson, et al., 2013;Paulson, Stine, et al., 2013).
To determine which environmental variables explained the most variation in community structure, we fit generalized additive models (GAMs) for water column specific conductance (SC), pH, and C, N, and P concentrations and sediment C, N, P, and median sediment particle size (median D p ) to the NMDS axes of planktonic and sediment bacterial ordinations, respectively, using the ordisurf function in the vegan package (v2.6.2) (Osaken et al., 2022) and plotted the model predictions on the ordinations as 2-dimensional contour lines. We also tested water column SC and pH as explanatory variables for sediment bacteria community structure because dissolved ions and pH can impact sediment properties (Abolfazli & Strom, 2022;Wu et al., 2014). All environmental variables were evaluated with GAMs, but only significant model fits (p < .05) are reported. Both geographic distance over the land and river network distance (using the riverdist package in R, Tyers 2022) were also analyzed but did not show any significant relationships with bacterial diversity (Figures S2 and S3).

| Threshold indicator taxa analysis (TITAN)
We used threshold indicator taxa analysis (TITAN) (Baker & King, 2010) to identify individual ASV and community responses to different environmental gradients represented by our field sites.
TITAN combines Dufrene and Legendre's (1997) indicator species analysis (IndVal) and non-parametric breakpoint analysis to identify the region along environmental gradients with the highest probability of inducing single-taxa responses. Single-taxa change points, which identify the region across an environmental gradient with the highest probability of change in taxa abundance, are summed to identify the portion of the gradient where synchronous changes in the abundance of many taxa occur, that is, a community-level response. Briefly, TITAN arranges observations along the gradient and identifies potential biological change points as mid-points between observed values. Next, samples are partitioned above and below the potential change points and IndVal scores are calculated for each group of observations. The gradient value of the observation associated with the highest IndVal score is taken as the biological change point. We also used TITAN to determine whether taxa were sensitive or tolerant based on whether they responded negatively or positively across increasing stressor gradients, respectively. Importantly, individual taxa can either be positive or negative responders, not both. Thus, positive and negative community-level responses are driven by unique taxa. We ran TITAN with 500 permutations and 500 bootstrapped replicates. IndVal scores were then standardized as z-scores by the mean and standard deviation of permuted scores. Only taxa with observed IndVal scores greater than 95% of the permutation scores and that consistently responded in the same direction (i.e., positive or negative response) along the gradient in at least 95% of the 500 bootstrapped runs were considered indicator taxa. Individual taxa responses were then summed (summed z-score) to determine where along the gradient there is the highest probability of observing a community-level change.
Lastly, we ranked variables based on response strength (maximum summed z-score) to identify which variables were most likely to induce community-level responses. TITAN has been used recently in other systems to identify bacterial responses to stressor gradients using amplicon sequencing data (Pilgrim et al., 2022;Simonin et al., 2019). Because median D p generally decreased with increasing agricultural land use while sediment nutrients and dissolved ions increased, the negative response for median D p is shown next to the positive responses for other variables to better represent the impacts of increasing agriculture.

| Bacterial community assembly mechanisms
To determine the relative importance of deterministic versus stochastic assembly mechanisms including dispersal, selection, and drift, we leveraged a null modeling framework (Stegen et al., 2012) based on both phylogeny and relative abundance using the microeco package in R (Liu et al., 2021). To test if there was phylogenetic niche partitioning (i.e., phylogenetic signal) in our study system, which is required for making ecological inferences based on sequencing data (Cavender-Bares et al., 2009;Losos, 2008), we calculated Mantel correlograms which indicated high ASV habitat correlations across short phylogenetic distances ( Figures S4 and   S5). To determine the contribution of selection, β-nearest taxon index (βNTI) was calculated. First, we calculated β-mean nearest taxon distance (βMNTD) for all pairwise comparisons (βMNTD obs ).
Next, we calculated a null distribution of βMNTD (βMNTD null ) by randomizing the tips of the phylogenetic tree and calculating βMNTD 999 times to determine what beta diversity would be under purely stochastic conditions. βNTI was calculated using Equation (1)

| RE SULTS
Explanations and units of acronyms frequently used throughout the methods and discussion are listed in Table 1.

| Drivers of bacterial community structure
Generalized additive models of environmental variables fit to NMDS scores based on Bray-Curtis dissimilarity indicated that SC explained the most variability in planktonic community structure and median D p explained the most variability in sediment community structure (Table 2 and Figure 3). Total organic carbon, TN, and pH also explained a substantial amount of variability in planktonic community structure but are correlated to SC ( Figure S7). Other variables that explained significant variability in sediment communities were Sed C, Sed N, Sed P, SC, and pH (Tables 1 and 2). Both Sed C and Sed N are correlated to median D p ( Figure S7). GAM results indicate that both SC and median D p exhibited a non-linear relationship with planktonic and sediment bacterial communities, respectively (Figure 3a,b).

| Sediment texture, nutrients, and relationships with bacterial alpha diversity
Sediment C and N decreased exponentially as median D p increased, with several high percent sand samples (>95% sand) having undetectable N (Figure 5a,c). The steepest decline in sediment C and N occurred across median D p of ~1-20 μm in sediments with at least 35% clay by mass. However, streams with higher water column C and N did not necessarily have higher sediment C and N concentrations (Figures 5b,d). Conversely, sediment P was weakly related to median TA B L E 2 GAM summary statistics for models fit between environmental variables and NMDS scores.

F I G U R E 3 NMDS ordinations of bacterial community structure based on
Bray-Curtis dissimilarity for (a) planktonic bacterial communities and (b) sediment bacterial communities. Generalized additive model predictions for the variable that explained the most deviance for each community is overlaid on each ordination. Median D p , median particle diameter (μm), SC, specific conductance (μS cm −1 ). Point size corresponds to SC and median D p for planktonic and sediment community plots, respectively.
D p (Figure 5e) and was more variable in sediments with a high clay fraction and small median D p . Sediment P was, however, strongly related to water column TP and increased logarithmically with increasing water column concentrations (Figure 5f).
Bacterial PD demonstrated a classic subsidy-stress response along the median D p gradient (Odum et al., 1979), with diversity peaking near a median particle size of 10 μm. The lowest bacterial diversity was observed in sandier sediments with median D p > 100 μm and lower nutrient concentrations. There was considerable variability in bacterial diversity in sediments with a median D p < 10 μm.
Variability in bacterial diversity in these clay-dominant sediments was explained by the additive effect of sediment P concentration ( Figure 6a). Overall, there was a strong polynomial relationship between median D p and sediment heterogeneity indicating that sediments with a median D p ~ 10 μm were composed of the most even mix of sands, silts, and clays ( Figure 6b). Collectively, these two unimodal relationships manifest in a linear response of bacterial diversity to sediment heterogeneity (Figure 6c).

| Community assembly processes
Community assembly mechanisms differed substantially between planktonic and sediment bacterial communities. Planktonic bacterial community assembly was dominated by homogenous selection, which accounted for 89.9% of community assembly across the MAP. Undominated assembly accounted for 9.0% with minor contributions from dispersal and variable selection ( Figure 7a).
Conversely, sediment community assembly was controlled by an even mix of dispersal limitation (36.5%), homogenous dispersal (15.1%), variable selection (19.6%), homogenous selection (8.5%), and undominated assembly (20.3%) (Figure 7b). There was no relationship between β-Nearest Taxon Index (βNTI-metric of how different observed PD was from null expectations) and pairwise SC differences in planktonic communities (Figure 7c), which was the most important variable impacting planktonic bacterial diversity based on both GAM fits and TITAN. There was, however, a significant relationship between sediment βNTI and pairwise differences in median D p (p < .05), with βNTI values shifting from predominately undominated and homogenous selection across small differences in median D p to primarily variable selection across large differences in median D p (Figure 7d).
Sediment texture also impacted community assembly process between planktonic and sediment communities within the same stream (i.e., sediment and planktonic samples collected at the same site). Streams with smaller median D p had sediment and planktonic communities that were more different than expected under null conditions, indicating variable selection is differentiating bacteria that live in the water column versus bacteria in sediments. However, as median D p increased into the sand range (<100 μm), planktonic and sediment communities within the same stream became more similar to each other, indicating homogenous selection is the primary assembly mechanism in sand-dominated streams (Figure 8).

F I G U R E 4
Maximum sum(z) values from threshold indicator taxa analysis (TITAN) for both positive and negative bacterial responses for planktonic (a, c) and sediment (b, d) communities. Maximum sum(z) scores can be interpreted as a variable importance metric describing the probability that a variable will induce a communitylevel change. Note that x-axis ranges are different among plots. Also note that positive and negative community responses are driven by unique bacterial ASVs, as no ASV can be both a positive and negative responder for the same variable. median D p , median particle diameter; SC, specific conductance; Sed C, sediment carbon; Sed N, sediment nitrogen; Sed P, sediment phosphorus; TOC, total organic carbon; TN, total nitrogen; TP, total phosphorus; Turb, turbidity.

| Bacterial community responses to increasing agricultural land use
A novel finding of this research is that sediment bacterial diversity in MAP streams increased with increasing agricultural land use. In streams, agricultural stressors including deposition of fine sediments and increased nutrients and ions can decrease the diversity of multiple biological assemblages including fishes and macroinvertebrates (Camargo & Alonso, 2006;Jones et al., 2012;Kemp et al., 2011;Timpano et al., 2018;Wang et al., 2019). In the MAP, however, more stream bacterial ASVs of both planktonic and sediment-bound bacteria increased than decreased in relative abundance across agriculturally-associated stressor gradients, especially with increasing salinity and sedimentation. Thus, although agriculturally associated stressors tend to decrease the abundance and biodiversity of macro-biological communities, increased ions, nutrient concentrations, and fine sediments increase microbial diversity, especially sediment-associated communities. Although tolerant components of both planktonic and sediment communities and diversity of sediment communities increased with greater agricultural land use or agriculturally associated stressors, there were stark differences in assembly mechanisms with implications for local and regional diversity.

| Salinization homogenizes planktonic bacterial communities
Salinity (measured as SC by proxy) had the largest influence on planktonic bacterial community structure in streams across the MAP relative to other measured water quality variables like C, N, and P. Although bacterial PD neither increased nor decreased along the SC gradient in MAP streams, TITAN revealed a net positive response of tolerant planktonic ASV's to increasing SC that was substantially stronger than responses to other variables. Salinity is one of the strongest variables modulating bacterial community structure across diverse habitats (Lozupone & Knight, 2007;Tang et al., 2021). Previous work demonstrates declines in bacterial diversity associated with increasing SC (Bier et al., 2015;Simonin et al., 2021;Timpano et al., 2018). However, these studies were conducted in watersheds impacted by mining which also increases con- There are multiple potential mechanistic explanations for the increase in relative abundance of many bacterial ASVs in responses to elevated ion concentrations. In environments with low extracellular ion concentrations like some freshwaters, certain bacteria may be unable to compensate for the rapid influx of water and increased cell volume resulting in hypoosmotic stress, reduced growth, and reproduction rates, and in extreme instances, cell death (Booth & Louis, 1999;Wood, 1999). Elevated ion concentrations can increase the survival of bacteria and alleviate hypoosmotic stress, supporting more diverse bacterial communities in agricultural streams (DeVilbiss et al., 2021).

Specific ions associated with agriculturally induced salinization can
also have specific positive effects on bacteria. For example, Mg 2+ , which is often added to agricultural soils to promote crop yields, is also a critical component of bacterial cellular membranes and facilitates TA B L E 3 Bacterial community change points and 95% confidence intervals (CI) identified using TITAN. Note: Variables shown were selected based on maximum Z-scores for positive and negative responses for planktonic and bacterial communities. Change points represent the location along an environmental gradient where synchronous changes in the abundance of many taxa occur.
Calcium is another ion commonly added to agricultural soils and is a key component in cell structure maintenance, motility, and signal transduction (Dominguez, 2004). Mississippi Alluvial Plain streams are also enriched with Fe 3+ due to extensive irrigation using groundwater from the Mississippi River Valley alluvial aquifer, which generally has concentrations exceeding 6000 μg L −1 (Bednar, 1988). Iron is an enzyme co-factor that is required to synthesize ATP and is thus essential for cellular respiration, growth, and OC mineralization (Koedooder et al., 2018). Regardless of whether salinity has a general osmotic effect or specific ion effect, elevated ion concentrations in MAP streams appear to homogenize regional diversity and increase local diversity of planktonic bacteria by creating a more ideal environment that supports a larger percentage of the regional community at each site.
F I G U R E 5 Relationships between median particle diameter (Median D p ) of sediments and sediment carbon (Sed C), nitrogen (Sed N), and phosphorus (Sed P) concentrations (a, c, e) and water column and sediment nutrient concentrations for total organic carbon (TOC), total nitrogen (TN) and total phosphorus (TP) (b, d, f).

F I G U R E 6
Panel (a) shows the relationship between median particle diameter (Median D p ) and sediment bacterial phylogenetic diversity (PD). The black curve is a second-order polynomial that describes the relationship between median D p and PD across the full median D p gradient. The dashed and solid blue lines show the additive effect of sediment phosphorus concentrations (Sed P) and median D p on PD. Panel (b) shows the relationship between Median D p and sediment heterogeneity (Sed Heterogeneity) and panel (c) shows the relationship between Sed Heterogeneity and PD.

| Dispersal and sediment characteristics influence diversity of sediment bacterial communities
Over half of community assembly processes of sediment bacteria could be attributed to dispersal, the major component (36.5%) being dispersal limitation. Low rates of dispersal were expected in sediment communities given that sample streams were located within six distinct drainage basins. Low movement rates of bacteria, which are often only microns per second in the absence of active transport like flowing waters (Wisnoski & Lennon, 2022), likely further limit widespread dispersal of sediment bacteria across large geographic regions like the MAP. There was, however, no significant relationship (p > .05) between geographic or river network distance and community dissimilarity which has been observed in other stream networks.
In mountainous watersheds, for example, benthic communities become more dissimilar across greater distances and dispersal plays a minimal role in community assembly relative to selection . The MAP, however, does not represent a typical dendritic stream network. Substantial hydrologic modifications, irrigation, and regional variability in soil erosion and deposition in streams from agricultural practices throughout the MAP could explain why geographic and river network distance are not important factors affecting bacterial diversity (Logan, 1993;Yasarer et al., 2020). For example, a large fraction of bacteria in sediment communities could be soil-derived and transported across the land to surface waters during runoff events (Caillon & Schelker, 2020;Le et al., 2020;Ruiz-Gonzales et al., 2015). Thus, spatial variability in soil properties and erosion might also contribute to patters of bacterial diversity across the region. Although soil-derived bacteria might be present in stream sediments, it is possible that they are not metabolically active due to such a drastic shift in environmental conditions . Stream sediment properties, however, were strong predictors of bacterial diversity.
F I G U R E 7 Community assembly processes for (a) planktonic and (b) sediment bacterial communities. Panel (c) shows the relationship between pairwise differences in specific conductance (SC) and beta-nearest taxon index (βNTI) and Panel (d) shows the relationship between pairwise differences in median particle diameter (median D p ) and βNTI. Values >2 indicate variable selection while values <−2 indicate homogenous selection. Points within the red lines did not differ significantly from null expectations.

F I G U R E 8
Beta-nearest taxon index (β-NTI) for pairwise comparisons of planktonic and sediment communities from the same stream along the median particle diameter (median D p ) gradient. The regression line shows how sediment and planktonic communities transition from being more different from each other than expected under stochastic conditions, that is, variable selection, to more similar than expected under stochastic conditions, that is, homogenous selection, as Median D p increases.
At the local scale (i.e., within-stream), there was a strong effect of sediment texture on bacterial diversity. We observed a subsidy-stress response along the sediment particle size gradient. Bacterial diversity was lowest in sediments with the smallest and largest median D p , and greatest in sediments with intermediate median D p of ~10 μm.
Sediment nutrients, which also induced strong, positive bacterial responses, may have contributed to increased diversity in sediments with smaller median D p by alleviating nutrient limitation. However, nutrient concentrations declined exponentially with increasing particle size, not in a subsidy-stress relationship as observed between bacterial diversity and particle size. Thus, other factors besides associated nutrients must be modulating bacterial diversity in MAP stream sediments. As median D p increases, the relative proportion of sand, silt, and clay also increases, resulting in increased sediment heterogeneity (i.e., habitat heterogeneity from a microbial perspective). There is a broad spectrum of ecological research documenting positive relationships between habitat heterogeneity and biodiversity (McCoy & Bell, 1991;Stein et al., 2014;Tews et al., 2003). In general, agriculture homogenizes habitat and decreases biodiversity (Benton et al., 2003;Donald et al., 2001;Gamez-Virues et al., 2015). Our results support a novel deviation from this well-established link between agriculture and habitat homogenization, with elevated sediment deposition increasing habitat heterogeneity from a microbial perspective. Increased sediment heterogeneity across the landscape could support higher regional bacterial diversity by increasing variability of nutrient concentrations, redox conditions, interstitial space and porosity, and hyporheic exchange (Gayraud & Philippe, 2003;Shrivastava et al., 2020;Southerland et al., 2007).
Additionally, soil erosion and subsequent deposition in streams could serve as a vehicle for transporting soil bacterial communities into aquatic habitats further increasing regional diversity of sediment communities in the MAP (Caillon & Schelker, 2020;Le et al., 2020).
Within the clay-dominated sediments with smaller median D p , there was considerable variability in bacterial diversity that can be explained by sediment P concentrations. In sediments with a high clay fraction (<35%), we observed an additive effect of P on bacterial diversity, where increasing sediment P resulted in decreased diversity.
Bacterial responses to altered nutrient concentrations and stoichiometry are highly variable across studies (Horner-Devine et al., 2003;Logue et al., 2012), but P enrichment has been linked to decreased sediment bacterial diversity and function in other soil and freshwater ecosystems as a result of shifts in nutrient stoichiometry (Delgado-Baquerizo et al., 2017;Lee et al., 2017). Controlled laboratory studies have also observed increased bacterial richness in P depleted environments (Leflaive et al., 2008). Lower sediment P concentrations may promote greater niche partitioning and PD by supporting bacterial communities consisting of species with different strategies for utilizing a less abundant resource to meet requirements for maintaining productivity (Ceulemans et al., 2017;Sandipan et al., 2019;Scott et al., 2008). Key bacterial genes associated with the carbonphosphorus lyase pathway tend to be elevated in both terrestrial and aquatic environments with less available P (LeBrun et al., 2018a(LeBrun et al., , 2018bOliverio et al., 2020;Sosa et al., 2019). It is also possible that alterations in nutrient stoichiometry favor specific bacterial groups that outcompete species less adapted to low N:P environments. For example, elevated P may favor diazotrophs that are able to acquire their own N from biological N fixation under low N:P conditions (Jabir et al., 2020). Changing N:P conditions may also alter the breakdown of organic nutrients via extracellular enzyme activity (EEA) which is generally higher in stream sediments with lower nutrients (Hill et al., 2012). However, enzyme producing organisms can occur as ecologically or physiologically distinct populations within phylogenetically related groups (Zimmerman et al., 2013), such that variation in EEAproducing bacteria may not contribute much to our overall observed changes in phylogenetic diversity of stream sediment communities associated with increasing P. Although the specific mechanisms behind the negative relationship between sediment P and bacterial diversity in clay-dominated sediments are unknown, it appears to be driven by water column P concentrations. While sediment C and N are poorly related to water column C and N, there is a strong, positive relationship between sediment P and water column P. There appears to be a dynamic relationship between water column and sediment P where elevated water column P drives increased sediment P and subsequently reduces bacterial diversity in sediments. Sediment-water P equilibrium has been well documented, especially in agricultural catchments, but certain variables like iron concentrations can alter the potential for sediments to store P (Palmer-Felgate et al., 2009;Smith et al., 2005;Taylor & Kunishi, 1971). Streams in the MAP can become enriched with iron due to irrigation with naturally ferrous groundwater, increasing the capacity for Sed P retention via the formation of iron-P complexes (Bednar, 1988;Bostrom et al., 1988). At the regional scale, sediment properties drive community assembly via variable selection, which accounted for ~20% of assembly processes. Sediment bacterial communities in streams with different sediment texture (e.g., sand vs. clay dominated) were more phylogenetically distinct than null expectations (i.e., variable selection).
Streams with similar sediment texture, however, contained sediment bacterial communities that were more phylogenetically similar than expected under null conditions (i.e., homogenous selection).
Sediment texture alone also explained ~90% of sediment bacterial community variability and decreased sediment particle size and increased nutrients also induced a strong, positive response of many bacterial ASVs. The 95% confidence interval for negative bacterial responses to increasing median D p determined by TITAN was wide, ranging from 2.72-127.51 μm, indicating that bacterial responses occur gradually as particle size increases rather than abruptly at a specific particle size. Thus, agricultural and best management practices that substantially alter erosion and sediment deposition could induce considerable shifts in bacterial community assembly mechanisms, regional diversity, and function.

| Sediment texture drives differentiation of sediment and planktonic bacterial communities
Differences between planktonic and sediment bacterial communities within the same stream (and sample location) were modulated by sediment properties. As median D p decreases, sediment nutrient concentrations increase exponentially. Higher nutrient concentrations associated with clays and silts appear to cause a divergence in planktonic and sediment communities via variable selection which increases overall stream diversity and possibly biogeochemical potential. High nutrient environments like MAP sediments with smaller particle sizes tend to favor conditionally rare, fast-growing bacteria that bloom by rapidly assimilating P for ribosomal RNA production (Elser et al., 2000). A shift in life-history traits towards faster growing bacterial species in nutrient rich sediments has been observed across diverse coastal environments and explains differences between sediment and slower growing, persistently rare oligotrophic planktonic bacterial communities in marine ecosystems (Dai et al., 2022 (Lynch & Neufeld, 2015).

| Secondary drivers of bacterial diversity
Although salinity and sediment texture were the dominant drivers of bacterial diversity in MAP streams, other variables including nutrients (C, N, and P) and turbidity were also related to changes in bacterial structure. Genetic approaches like 16S rRNA amplicon sequencing have been used in other systems to better understand planktonic bacterial responses to increasing nutrient concentrations. In other less modified watersheds prone to P enrichment, bacteria begin to respond to increasing TP at 0.052 mg L in other watersheds compared to our observed range of 0.420-1.020 mg L −1 . Collectively, these comparisons with less altered systems suggests that bacterial communities in highly modified agroecosystems like the MAP may consist of more copiotrophic taxa that are better adapted for higher nutrient conditions which could have implications for biogeochemical cycling. For example, organic C enrichment not only alters bacterial community structure, but stimulates bacterial growth resulting in enhanced biological N assimilation, respiration, and decreases uptake lengths of NO 3 − and NH 4 + (Bernhardt & Likens, 2002;Johnson et al., 2012). Bacterial P-cycling genes also respond strongly to elevated TP suggesting the potential for P enrichment to alter the bioavailability of P, which could impact multiple trophic levels (LeBrun et al., 2018b).
Many planktonic ASVs also increased in relative abundance as turbidity increased. Given high erosion in the MAP, this response could be driven by soil bacteria that are flushed into streams along with suspended sediments during precipitation or irrigation events. Stream bacterial communities can comprise substantial portions of soil-derived bacteria in other environments (LeBrun et al., 2018a;Ruiz-Gonzales et al., 2015;. Particle-associate bacterial communities in the Mississippi River are distinct from free-living communities and may explain why certain ASVs were strongly associated with elevated turbidity (Jackson et al., 2014). Turbidity also impacts light availability which could have direct, negative effects on photoautotrophic organisms (Davies-Colley et al., 1992;Lloyd et al., 1987) and alter species interactions with heterotrophic bacteria. Conversely, turbidity could have been driven by high phytoplankton production in response to high nutrient concentrations and warm water temperatures (range: 19.7-32°C) (Klemas, 2012). There was a positive relationship between turbidity and ash-free dry weight (r 2 = .43, p < .0001) suggesting that a significant portion of turbidity was driven by organic particles, potentially phytoplankton. Utilization of phytoplankton-derived carbon by heterotrophic bacteria as well as bacterially-regenerated nutrient uptake by phytoplankton have been well documented (Cole, 1982;Coveney & Wetzel, 1989;Scott et al., 2008).

| Linkages between bacterial diversity and agricultural practices: Implications for ecosystem management
Bacterial communities in the MAP were most strongly influenced by variables that tend to be elevated or strongly modulated by agricultural practices. Sedimentation, which is exacerbated by agricultural practices like soil tillage, is a major water quality concern globally (Walling, 1990).
Shifting land use from forest to agriculture increases fine sediments in streams (Kreiling et al., 2021). Prior to land clearing, streams in the MAP were largely sand bottom but land clearing and the subsequent expansion of agriculture induced substantial increases in sedimentation of fine clays and silts in MAP water bodies that has remained 50-fold higher than pre-agricultural times (Wren et al., 2008). Variability in agricultural and best management practices (BMPs) can cause regional variation in the percentage of sands, silts, and clays deposited in streams across the MAP (Cooper et al., 1987;McKergow et al., 2003) resulting in regional differentiation of sediment bacterial communities. If PD affects the biogeochemical functional capacity of sediment bacterial communities, then tillage practices and BMPs designed to mitigate sedimentation could influence stream ecosystem functions by altering patterns of variable selection and bacterial diversity and function in streams.
Bacterial communities may also be useful for establishing reliable biological stressor-response relationships in intensely cultivated ecosystems. Although diversity of many macro-scale organisms declines in response to expanding agricultural land use activities, bacterial community diversity increases. In the MAP, there are currently no established biological indices for water quality management and monitoring. Reduced local and regional diversity of macroinvertebrate communities has prevented the development of a traditional multi-metric index for monitoring purposes (Stribling et al., 2016).
Bacterial communities may provide the basis for additional biological support for nutrient reduction goals because, contrary to macroinvertebrates, bacterial diversity increases in response to agricultural stressors increasing the probability of detecting responses with greater certainty. Thus, the information presented here on the ecology and assembly mechanisms of bacterial communities exposed to intense agriculture could inform water quality management and improve our ability to monitor ecosystem responses to best management practices, which could be implemented at a global scale to minimize negative impacts of expanding agriculture on water quality.

ACK N OWLED G M ENTS
We any discrimination in all its programs and activities on the basis of race, color, national origin, age, disability, sex, marital status, familial status, parental status, religion, sexual orientation, genetic information, political beliefs, reprisal, or because any part of an individual's income is derived from any public assistance program.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare that they have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
DNA sequencing data that support the findings of this study are openly available in the National Center for Biotechnology Survey station numbers for NWIS water quality data are in Table S1.