operate. The need for resolution is particularly pressing when considering the now extensive evidence for reciprocal, and consequential, interactions between communities of microbial symbionts and their hosts (McFall-Ngai et al., 2013).

Plants host an array of symbiotic fungi that live for at least some portion of their life history without causing apparent symptoms of antagonism with their hosts (Rodriguez, White, Arnold, & Redman, 2009). These fungal endophytes are ubiquitous, hyperdiverse and may interact with their hosts as commensals, mutualists or latent antagonists (Porras-Alfaro & Bayman, 2011). Most endophytes occupy only a small portion of available host tissue (Lodge, Fisher, & Sutton, 1996) and inocula are transmitted between hosts in air, rain or by animal vectors (Rodriguez et al., 2009). However, some endophytes grow systemically throughout the host and can be transmitted transgenerationally in seeds (Clay & Schardl, 2002;Panaccione, Beaulieu, & Cook, 2014). These vertically transmitted endophytes can affect host ecology and evolution in myriad ways, including through modification of gene expression (Mejía et al., 2014) and protection against herbivorous insects (Rudgers & Clay, 2007).

Previous studies have demonstrated that variation in endophyte community structure can reflect filtering by abiotic conditions (Zimmerman & Vitousek, 2012) and interspecific differences in plant traits (Kembel & Mueller, 2014). However, relatively little is known about the influence of intraspecific variation in hosts on associated endophyte assemblages. Given that a recent meta-analysis suggests that, on average, a quarter of within-community variation in plant traits is intraspecific (Siefert et al., 2015), this represents a critical gap in knowledge. Even less is known regarding the effects endophytes have on one another, including direct inhibition, exploitative competition or indirect host-mediated effects (Agler et al., 2016;Vacher et al., 2016). To disentangle the complexities shaping endophyte communities, studies are needed that span multiple scales and simultaneously characterize the filtering effects imposed by other endophytes, intraspecific variation in hosts and abiotic conditions.

Here, we take such an approach using high-throughput sequencing to characterize endophyte assemblages of spotted locoweed, Astragalus lentiginosus (Fabaceae), in the Great Basin Desert of Western North America (Figure 1a). We conduct analyses at two levels of biological organization, both within and among populations, and at each level, examine the effect of intraspecific variation in host plants on endophyte community divergence. Spotted locoweed is a widespread forb in the arid West, with over 40 nominal varieties described (Welsh, 2007). Varietal designations reflect non-discrete, phenotypic variation in fruit and flower morphology, among other traits, which is predicted to some extent by climatic variation (Knaus, 2010). Little is known regarding the genetic structure of A. lentiginosus, however, Knaus, Cronn, and Liston (2005) report divergence among three varieties, which suggests that the phenotypic variation within this infraspecific complex is not due solely to plasticity. Spotted locoweed is also notable for its relationship with a seed-borne, vertically transmitted fungal endophyte, Alternaria fulva (Ascomycota: Dothideomycetes: section Undifilum; Baucom, Romero, Belfon, & Creamer, 2012;Lawrence, Rotondo, & Gannibal, 2016;Woudenberg, Groenewald, Binder, & Crous, 2013). Alternaria fulva synthesizes the alkaloid swainsonine (Molyneux & James, 1982;Ralphs et al., 2008), which can cause toxicosis in mammalian herbivores. "Locoism," as it is called, is costly, causing over 100 million dollars a year in damages to western ranching (Cook, Ralphs, Welch, & Stegelmeier, 2009).

Using this system, we asked: what is the relative importance of abiotic filters, host-associated factors and interspecific microbial interactions on the structure of fungal endophyte communities? Speaking generally, differences between communities of macro-organisms at the broadest scales are associated with variation in abiotic conditions (e.g. temperature, water availability; Box, 1981;Diaz, Cabido, & Casanoves, 1998;Schimper, 1898). Interspecific interactions are also important (Kraft & Ackerly, 2014;Vellend, 2010) and become more important at smaller spatial scales (e.g. Pearson & Dawson, 2003; but see Wisz et al., 2013). Taking the community ecology of macroorganisms as a starting point (Christian, Whitaker, and Clay 2015), we hypothesize that the relative ranking of ecological filters affecting fungal endophytes should proceed from abiotic filters to host-associated filters to interspecific interactions with other microbes, with the latter filter more influential at smaller spatial scales than at larger scales.

Spotted locoweed was collected from 24 locations across the Great Basin Desert (Figure 1a) during the summer of 2014. Foliar tissue was sampled from c. 15 to 20 haphazardly selected individuals at each location (Table S1). The location of each plant was used to calculate pairwise haversine distances between individual plants and populations. Plant size was represented as the product of the maximum width and height of each crown, and the width of the crown perpendicular to the distance measured at the crown's widest point.

Plant phenology was categorized as either vegetative, flowering or fruiting. DNA was extracted from three, surface-sterilized, mature leaflets from each plant (see Supplemental Methods). Swainsonine was characterized from air-dried foliar tissue using LC-MS/MS as per Gardner, Molyneux, and Ralphs (2001) (see Supplemental Methods).

To characterize the climate of each sampling location, temperature and precipitation data from 1990 onwards were downloaded from PRISM Climate Group (http://prism.nacse.org), and annual means were calculated from monthly means. The Palmer drought severity index (PDSI; a proxy for long-term moisture availability; Palmer, 1965) and the normalized difference vegetation index (NDVI; a measure of vegetation coverage and health; Rouse, Haas, Schell, & Deering, 1974) were downloaded from the Climate Engine web application (http://clim-engine.appspot.com/). We calculated the mean of the average annual PDSI for each focal location since 1990 and the average NDVI for each location from 1982 to 2015

Fungal richness Swainsonine concentration A.fungal relative abundance (Landsat 4/5/7/8). Soil data for each site were downloaded from the Soil Survey Geographic Database (accessed 16 August 2016; available from http://websoilsurvey.nrcs.usda.gov/). Data included available water holding capacity; percentage clay, silt, sand and organic material; CaCO 3 concentration; ph; cation exchange capacity; and conductivity. Soil data were reduced using a principle components analysis (PCA; R Core Team, 2015). The first principal component explained 73% of the variation in the soil data, described a gradient from clayey to sandy soils and was retained for analyses.

To characterize locoweed genetic structure, we sequenced and analysed a reduced representation library using a genotyping by sequencing method described by Gompert et al. (2012) and Parchman et al. (2012). Briefly, this method involves restriction endonuclease digestion of total genomic DNA, bar-coded adaptor ligation and sequencing of the resulting fragments on the Illumina HiSeq platform. Sequencing was conducted by the University of Texas Genome Sequencing and Analysis Facility (GSAF). Sequenced reads were aligned into homologous regions, rigorously filtered to remove possible non-host sequences and single-nucleotide polymorphism (SNP) genotypes estimated for each sample (see Supplemental Methods for details). To summarize genetic structure among all individuals, we executed a PCA on genotype estimates. We characterized genetic divergence among populations with Nei's D (Nei, 1972). Genetic distance between populations was calculated as the mean pairwise distance between all individuals from both populations. For each population, inbreeding coefficients (F IS ) were calculated using a script written by Eva Chan (https://github.com/ekfchan/evachan. org-Rscripts).

The internal transcribed spacer 1 (ITS1) locus was used to characterize fungal assemblages (Lindahl et al., 2013). Library preparation was conducted by the GSAF using a nested PCR approach.

The ITS1 region was targeted using the ITS-1F and ITS-2 primers (White, Bruns, Lee, & Taylor, 1990). For full PCR conditions and library preparation details, see the Supplemental Methods. Pairedend (2 × 250) sequencing was conducted on the Illumina MiSeq platform. Samples were split between two flow cells with half of each population randomly assigned to either flow cell. The initial portions of sequences were pruned to remove primer binding regions (22 bp). We used only forward reads for all downstream analyses (as recommended by Nguyen, Smith, Peay, & Kennedy, 2015). Reads with more than a single expected base calling error were removed using USEARCH v9.2.64 (Edgar, 2010). Duplicate sequences were removed and sequences were assigned to an operational taxonomic unit using UNOISE2 in USEARCH (Edgar, 2016b). This algorithm identifies OTUs differing by as little as a single nucleotide and thus provides much higher resolution than OTU clustering methods that bin sequences by similarity (Callahan, McMurdie, & Holmes, 2017). UNOISE also finds and removes chimeric sequences. We repeated analyses using the traditional approach of merging paired-end reads and clustering them into OTUs using a 97% similarity threshold. Results were qualitatively similar to those presented here and are not discussed further. OTUs were assigned a taxonomic status using the SINTAX algorithm (Edgar, 2016a) and the UNITE (accessed 3 May 2017; Kõljalg et al., 2005) and Warcup training sets (accessed 4 December 2017; Deshpande et al., 2016). Sequences of OTUs that were described as fungi with ≥80% confidence using either training dataset were used to query unfiltered reads. To determine which OTUs were A. fulva, OTUs were aligned to GenBank accession JX827264.1 using BLAST. As A. fulva relative abundance was a predictor variable of interest, OTUs matching this accession were omitted from all downstream calculations of response variables characterizing fungal communities (i.e. diversity, richness, etc.). For additional details regarding sequence generation and processing, including analyses confirming no influence of contamination, see the Supplemental Methods.

To account for variation in sequencing depth across samples, read counts were normalized using the trimmed mean of log expression ratios (TMM) approach with the edgeR package in r (Robinson, McCarthy, & Smyth, 2010;Robinson & Oshlack 2010). This method represents read counts in relative terms while accounting for the inherently greater variation in samples with greater sequencing depth. Read counts were then Hellinger standardized and Euclidean distances between all samples were calculated to generate a Hellinger distance matrix (Legendre & Gallagher, 2001). A distance matrix was also constructed from occurrence data using a Jaccard distance measure. Prior to matrix calculation, occurrence data were rarefied to 1,000 reads using rarefy in the vegan r package (Oksanen, Kindt, Legendre, O'Hara, & Stevens, 2016). Rarefaction is necessary for occurrence data because during TMM normalization, rare taxa are down weighted, but still remain in the data. Therefore, if data are converted to binary values, then the influence of varying library size will no longer have been minimized. Fungal distance matrices were visualized using metric multidimensional scaling (MDS; also called principle coordinates analysis) conducted using cmdscale in vegan. The final ordination was overlain with contour lines corresponding to variation in plant size (whose influence was determined from downstream analyses) using a general additive model implemented by ordisurf in vegan.

Fungal diversity was calculated using a numbers equivalent approach, which describes diversity in terms of "effective numbers" of taxa (Hill, 1973;Jost, 2006). Fungal gamma diversity and "turnover" (Jost, 2007) among plants at each population were represented using Simpson's diversity equivalent. Additionally, average fungal richness within plants was calculated for each population. The average Jaccard distance between all plants within a population was calculated as an occurrence-based analogue to turnover.

We used multiple regression to investigate the predictors of fungal richness, diversity and average between-plant turnover among locoweed populations. Model selection was performed using the leaps r package (Lumley, 2017) using a best subset approach with the adjusted r 2 as the model performance criterion. All aforementioned, z-standardized biotic and abiotic variables were considered during model selection, along with the proportion of plants sampled at a population that were fruiting (as a measure of population phenology). After a model was chosen, variance inflation factors were calculated, and collinear variables substituted for one another and Akaike's information criterion and adjusted r 2 used to determine which variable provided the best model fit. Collection date, plant volume and area sampled were included in all models as statistical controls. We partitioned the variance explained in the response among predictors using the calc.relimp function ("lmg" metric) of the relaimpo r package (Grömping, 2006).

For within population analyses, we used Bayesian hierarchical linear modelling to examine the influence of phenological state (categorical), plant volume and either A. fulva relative abundance or swainsonine concentration on fungal diversity and richness. This hierarchical approach allowed us to account for the nestedness of plants within populations while quantifying the effect of predictor variables at each population. Modelling was performed using rjags v3-15 (Plummer, 2015). For model specification, see Supplemental Methods. Mean values of posterior probability distributions were used as point estimates for beta coefficients.

We characterized fungal community dissimilarity at two levels of biological organization: between-population dissimilarity (based on population averages) and within-population dissimilarity (based on data from individual plants). Between-population differences in fungal communities were calculated as the average pairwise distance between the fungal communities at different locations.

To model these dissimilarities, we used multiple regression on distance matrices implemented via the MRM function in the ecodist r package (Goslee & Urban, 2007). The response matrix was permuted 5,000 times, and beta coefficients of predictor variables extracted at each iteration. The significance of a model term was the proportion of those beta coefficients equal to, or more extreme than, the coefficient generated using the unpermutted response matrix. For this analysis, Euclidean distance matrices were created for each of the aforementioned z-standardized, abiotic variables as well as latitude, collection date, area surveyed, percentage of plants fruiting, mean plant volume, mean A. fulva relative abundance (Hellinger standardized and TMM normalized) and mean within-population swainsonine concentration. Model selection was performed as described above, with distance matrices for collection date, plant volume and area surveyed retained as covariates in all models.

To analyse within-population dissimilarity in fungal assemblages, we used a hierarchical linear modelling approach within a Bayesian framework (see Supplemental Methods). We created separate distance matrices for each population for fungal community distance (Hellinger for relative abundance data; Jaccard for occurrence data), genetic (Nei's D) and physical distance, and plant size.

Sequencing resulted in 4,341,518 reads representing 287 fungal OTUs (for OTU table, see dryad accession). Fungal diversity, richness and between-plant turnover varied among locoweed populations across the Great Basin (Figure 1b,c). Taxonomic hypotheses were tentative for many OTUs. However, when using the UNITE database 59 OTUs were assigned to the Ascomycota and 25 OTUs to the Basidiomycota ≥80% certainty, with Dothideomycetes ( 35OTUs) being the richest fungal class (Warcup provided much less certain taxonomic assignments). Remaining OTUs were assigned a taxonomic status with less certainty, essentially only determining fungal origin (Figures S1 and S2,Supplemental Data). Six OTUs were assigned to A. fulva (3,045,290 reads in total). The relative abundance of A. fulva varied among populations (Figure 1d). Four A. fulva OTUs were recovered from all focal locations and the remaining two OTUs were only recovered from several populations each.

The second most abundant fungus was identified as Leveillula taurica (two OTUs). This fungus was also recovered from all populations but was only abundant at three locations. Leveillula taurica, a powdery mildew (Erysiphaceae), is a plant pathogen that infects leaves and that can grow endophytically in many plant taxa, including other Astragalus species (Correll, Gordon, & Elliott, 1987;Palti, 1971). MDS ordination of fungal occurrence data revealed that differentiation among individuals in terms of their fungal communities was associated with plant size (Figure 1g). MDS ordination of relative abundance data showed that the fungal communities recovered from vars. piscinensis and stramineus and a single population each of var. salinus and var. kennedyi differed from other A. lentiginosus varieties, but, otherwise, varieties overlapped in ordination space (Figure 1h).

For each locoweed individual sequenced, genotype likelihood estimates were determined for 13,366 SNPs; the average sequencing coverage depth at these loci was 8.07 reads per locus, per individual.

A PCA based on genotype likelihood estimates revealed evidence for substantial population genetic structure (Figure 1f). Principle components 1 and 2, respectively, explained 9.0% and 5.6% of the variance in genotype likelihoods (which is typical for such data); samples from different geographic areas and varieties clustered together and many formed non-overlapping clusters. Pairwise Nei's D calculated between populations ranged from 0.033 to 0.059, with a mean of 0.044.

While A. fulva was present in all spotted locoweed populations, swainsonine was not detected within some varieties (Figure 1e, Figure S3), which is consistent with previous reports (Cook et al., 2016). The relatively most abundant A. fulva OTU was significantly positively associated with variation in swainsonine content (Spearman's rank correlation, rho = 0.7, p < .001; see Supplemental Methods), consistent with previous reports (Cook et al., 2011;Cook, Gardner, et al., 2009). The remaining prevalent A. fulva OTUs were also positively associated with swainsonine content, but correlations were not significant (Figure S4). At the individual plant level, neither phenological state nor plant volume predicted A. fulva relative abundance (Table S2). However, at the population level, A. fulva relative abundance was higher in populations that had more fruiting individuals.

Multiple regression revealed that fungal richness was greater in populations with less A. fulva (Table 1; Figure 2a) that were comprised of smaller plants (Figure 2b) and that were collected earlier in the season. The negative effect of A. fulva on fungal richness was further supported by a multiple regression analysis where swainsonine concentration was used as an independent proxy of A. fulva relative abundance (Figure S5). Swainsonine was also inversely related to richness within populations in individual plants (Figure 2j). When analyses were repeated with each of the four prevalent A. fulva OTUs singly substituted for total A. fulva abundance, three of the four A. fulva genotypes were negatively and significantly associated with fungal richness.

Fungal diversity was inversely related to host plant size and was greater in populations with more flowering stems (Table 1; Figure 2d,e). Turnover (Simpson's equivalent) was greatest in less inbred populations with smaller plants in higher latitudes that were less drought stressed and sampled later in the year.

Occurrence-based turnover was greatest in populations with a high relative abundance of A. fulva, with smaller plants, that were sampled later, and which experienced cooler temperatures (Figure 2g-i). The significance of the effect of plant size on diversity and turnover was the result of an outlier population from the southern portion of the sampling area (var. stramineus). This population had the largest plants of any population sampled and had a divergent endophyte community (Figure 1h). When analyses were repeated without this population, plant volume was not a significant predictor of fungal diversity and turnover, though it remained a significant predictor of richness, and the directionality of effect remained consistent in analyses of all response variables. Within populations, fungal richness declined with plant volume for all populations surveyed (Figure 2k), though credible intervals for this effect did overlap zero. Additionally, the effect TA B L E 1 Results from multiple regression analysis of patterns in fungal endophyte richness, diversity and turnover in populations of Astragalus lentiginosus. Response variables are in the leftmost column and predictors in subsequent columns. All predictor variables are z-standardized population averages. Dashes signify unsupported variables that were not included in models. Alternaria fulva and swainsonine were not simultaneously included in models because they were strongly correlated. These variables were substituted for one another in the model, and the best performing variable retained. Percentages are estimates of explained variance in the response variable. The significant effect of volume on all response variables, except richness, was due to an outlier population, though the directionality of the effect of volume was robust to the exclusion of the outlying population for all models

Response variable Precipitation PDSI Elevation Soil NDVI Temperature A. fulva relative abundance % fruiting Swainsonine concentration F IS Plant size Latitude Date Survey area Adjusted R 2 Richness -0.31 (1.3%) ---0.60 (1.0%) ---3.50*** (25.0%) ---0.72 (8.9%) -3.32*** (36.5%) 0.77* (2.1%) -2.76*** (9.0%) 1.17** (4.1%) .80*** Simpson's diversity --1.54 (2.5%) -1.89 (3.8%) ----3.42** (9.5%) -0.88 (3.0%) --4.93*** (31.4%) -0.92 (1.0%) -1.77 (4.5%) .34* Turnover (Simpson's) --0.03* (6.7%) -----0.04** (4.9%) -0.01 (5.1%) -0.06*** (29.1%) -0.04*** (13.4%) 0.04** (6.3%) -0.04** (5.1%) 0.01 (3.0%) .57*** Turnover (Jaccard) --0.02* (2.0%) ----0.03** (10.5%) 0.06*** (37.8%) ---0.01 (1.7%) -0.05*** (25.7%) --0.03** (9.4%) 0.02** (4.5%) .88*** A. fulva relative abundance -----0.05 (6.5%) --0.13* (15.3%) ---0.14** (18.7%) 0.04 (2.9%) -0.08 (7.7%) 0.00 (0.9%) .42** ***p < .01; **p < .05; *p < .10.

of plant volume was apparent in the structuring of a MDS ordination (Figure 1g).

The predictors of fungal community dissimilarity proved difficult to uncover at any level of biological organization. Betweenpopulation dissimilarity in fungal endophyte communities was significantly associated with differences in plant size when including the outlier var. stramineus population in the analysis, but this relationship was not apparent when excluding this population (Table S3). Plant size was not predictive of within-population dissimilarity in fungal assemblages (Table S4). Both within, and among populations, the influence of host genetic and physical distance on fungal community dissimilarity was weak. Also, neither variation in swainsonine concentration, nor A. fulva relative abundance, was a significant predictor of community dissimilarity within or between populations. Additionally, while temperature, precipitation and soil variation vary dramatically across the region, none were significant predictors of between-population fungal community dissimilarity.

We report a strong negative association between the richness of the endophyte community within spotted locoweed and the relative abundance of the vertically transmitted fungus, A. fulva. We also observed a negative association between host plant size and fungal richness and diversity. Surprisingly, after for these two deterministic forces, we found relatively weak associations between abiotic conditions and variation in endophyte communities. We also found no evidence for dispersal limitation of fungal endophytes, as we did not observe an association between geographic proximity and fungal community dissimilarity (Table S3). Contrary to our expectations, these results suggest that the endophyte community of spotted locoweed is primarily shaped through ecological filtering imposed by biotic interactions, most notably by the apparent antagonism between A. fulva and other fungi, while abiotic conditions appear to be less important ecological filters.

While the negative relationship between A. fulva and the rest of the fungal community is clear, the mechanism of interaction is not yet understood. It is possible that A. fulva may affect the host phenotype in a way that hinders the colonization of host tissue by other fungi. Indeed, in other systems, endophytes have been found

to elicit shifts in the host immune response leading to pathogen resistance (Arnold et al., 2003;Busby, Ridout, & Newcombe, 2015;Mejía et al., 2014). Additionally, observations of the transcriptome of several grasses and their vertically transmitted endophytes have shown that endophyte infection induces differential regulation of many host genes (Ambrose & Belanger, 2012;Khan et al., 2010;Schmid et al., 2016), including several possibly involved in pathogen defence which could affect endophyte colonization (Dinkins, Nagabhyru, Graham, Boykin, & Schardl, 2017;Dupont et al., 2015).

Alternatively, perhaps, A. fulva is a superior competitor for space, photosynthate or other limiting resources in planta as compared with other more generalized fungi, which may be less adapted to A. lentiginosus. Finally, it is plausible that the alkaloid produced by A. fulva, swainsonine, may directly inhibit growth of other fungi that come in contact with hyphae of A. fulva. Swainsonine prevents the action of alpha-mannosidases (Molyneux & James, 1982), which occur in numerous fungi (Eades & Hintz, 2000) and are important for polysaccharide catabolism (Choi, Kim, Jeon, & Lee, 2013;de Vicente, Arrayás, Cañada, & Carretero, 2000). We observed that certain A. fulva genotypes were more strongly predictive of variation in swainsonine concentration than other genotypes (Figure S4).

This relationship between fungal intraspecific genetic variation and swainsonine production is consistent with previous reports from the congeneric fungus Alternaria oxytropis and the host plant Oxytropis sericea (Oxytropis is the sister taxon to Astragalus; Cook, Grum, Gardner, Welch, & Pfister, 2013). Additionally, the A. fulva genotype that was most predictive of swainsonine concentration had the strongest negative association with fungal richness, which suggests that landscape level genetic variation in A. fulva has important ecological consequences for its host.

Previously, Alternaria section Undifilum fungi have been hypothesized to be commensals of locoweeds (including locoweed taxa belonging to both Astragalus and Oxytropis; Creamer & Baucom, 2013), because of the lack of obvious antagonistic effects on locoweeds, and because swainsonine, despite its toxicity, does not deter some mammalian grazers (e.g. sheep; Ralphs, Panter, & James, 1990) and at least some herbivorous insects which specialize on locoweeds (Parker, 2008;Richerson, 1992;Thompson, Knight, Sterling, & Murray, 1995). However, our results support the hypothesis posed by Lu et al. (2017) that interactions between locoweeds and Alternaria section Undifilum fungi may in some cases function as a mutualism, with Alternaria spp. benefiting their hosts through constriction of endophyte richness, which should limit host exposure to pathogens.

Indeed, in a culture-based survey of the endophytes of Astragalus variabilis and Oxytropis glabra, Lu et al. (2017) report abundance of Alternaria fungi was negatively correlated with pathogen abundance. We observed that A. fulva was represented by several OTUs, which varied in relative abundance within and among focal populations. Studies in A. lentiginosus as well as other locoweed species (including members of Oxytropis and other Astragalus taxa) have reported similar intrapopulation variation in Alternaria spp.

relative abundance and swainsonine concentration (Cook et al., 2016;Gardner et al., 2001;Grum et al., 2012;Ralphs et al., 2008).

Interestingly, Alternaria spp. is almost always detectable via PCR within the tissues of the locoweeds it inhabits, and vertical transmission of the fungus appears to have a high fidelity (Ralphs, Cook, Gardner, & Grum, 2011), yet for some individuals and populations of locoweeds, A. fulva biomass is reduced. This variation in Alternaria spp. biomass in planta seems to counter the hypothesis that that these fungi function as mutualists because mechanisms, in both the host plant and the endophyte, that facilitate high endophyte F I G U R E 2 Factors shaping fungal communities at the regional, among-population scale (top panel; a-i) and at the scale of the individual plant, within populations (bottom panel; j-l). Depicted in the top panel are partial regression plots from multiple regression analyses of foliar fungal richness (a-c), diversity (Simpson's equivalent; d-f), and average between-plant turnover in fungal taxa (using incidence data; g-i). Models included additional covariates and coefficients and p-values are shown in Table 1. Predictor variables for all three models were population averages. Partial regression coefficients and significance levels are shown in each plot. The structure of swainsonine is shown in (j). Analyses were generally robust to exclusion of outliers with the exception of regressing the effect of plant volume on fungal diversity (e) and turnover (h). Depicted in the bottom panel are the posterior probability distributions for coefficients from a hierarchical linear model of fungal richness at the within population scale. Grey lines denote population specific coefficients, and black lines denote coefficients estimated across all populations (mean provided in upper right of each plot)

Density Swainsonine P lant volume Fruiting -2.83 (a) (b) (c) -0.44 -1.90 (d) (e) (f) (j) (k) (l) (g) (h) (i) -3.32*** (36.5%) -3.50*** (25.0%) -2.76*** (9.0%) Within populations ***p < .01; **p < .05;*p < .10; † sig. due to outlier 3.42** (9.5%) -4.93*** † (31.4%) 0.92 (1.0%) -0.05*** † (25.7%) 0.06*** (37.8%) -0.03** (10.5%)

biomass within plant tissues should be under positive or stabilizing selection. However, if there is a cost to the host of supporting the endophyte, then spatial or temporal variation in the benefits provided by the fungus should lead to variation in the relative abundance of these endophytes. Such a trade-off has been documented

for the vertically transmitted Neotophidium endophytes of some perennial grasses, as benefits provided by these endophytes to their hosts can vary with nutrient (Cheplick, Clay, & Marks, 1989) or water availability (Morse, Day, & Faeth, 2002). Few studies of the effects (positive or negative) of Alternaria section Undifilum endophytes on their hosts have been conducted. However, for Astragalus mollissimus, Klypina, Pinch, Schutte, Maruthavanan, and Sterling (2017) demonstrated that endophyte infection tends to reduce host growth when water availability is high, but not when water availability is low, which suggests possible trade-offs associated with endophyte colonization. Additionally, previous studies have reported that colonization by A. fulva was associated with reduced size in A. lentiginosus (Cook et al., 2017), a pattern we also report here and which could be indicative of a fitness cost associated with endophyte colonization that may partially explain the observed variation in A. fulva relative abundance. Future work should focus on exploring the costs and benefits of Alternaria section Undifilum endophyte colonization for locoweeds via manipulative experiments conducted in the field.

We observed a negative association between plant size and fungal richness both within and among host populations. This association was surprising since larger plants should represent larger targets or "islands" for fungal inocula and plant size typically increases with age, which would give more time for endophytes to colonize tissues.

Indeed, for many plants, age has been correlated with increased endophyte load (Wilson, 2000). Our results could be due to inhibition of plant growth by endophytes (Faeth & Sullivan, 2003), shifts in plant phenotype through ontogeny that influence endophyte communities or more simply could be because there may not have been enough time for the tissues of rapidly growing plants to accumulate a rich endophytic assemblage.

While we recovered evidence for considerable genetic structure in spotted locoweed, host genetic distance did not predict fungal community dissimilarity within or between populations. Previous studies have reported an effect of host genotype on fungal communities (Bálint et al., 2015;Lamit et al., 2015;Whipps, Hand, Pink, & Bending, 2008). However, the few studies that have quantified genetic distance between hosts and attempted to link genetic divergence to dissimilarity in the associated microbial community have been contradictory. Cordier, Robin, Capdevielle, Desprez-Loustau, and Vacher (2012) report divergence of foliar endophyte communities with host genetic distance in a study of beech trees, while no effect was found in spruce trees (Eusemann et al., 2016). Both of these studies were conducted at a single site. In contrast, our study spans the Great Basin and uses a host with strong population genetic structure, and yet, we found no evidence for the effect of host genetic divergence on endophyte communities. While this finding does not rule out the possibility of linked evolutionary history between locoweed and individual endophytes (such as A. fulva), it does confirm a lack of community-wide codivergence between host and symbionts.

Given past reports of Astragalus spp. genome size (1C: c. 1-2; Samad et al., 2014), the marker density of our dataset is likely around 1 SNP per 50-80 kb, and is thus not ideal for locus specific analyses of functional divergence. This does not, however, negate the efficacy of our dataset for describing genetic divergence among locoweed populations or for testing the hypothesis that endophyte communities are structured by intraspecific host divergence. We suggest that future studies should examine divergence in host genes thought to influence fungal colonization and persistence, as they may be better predictors of endophyte community dissimilarity.

While genetic distance was not an important predictor in our analyses, we report that inbreeding was associated with reduced between-plant, fungal turnover (Table 1). Interestingly, this association was observed only for turnover as calculated using Simpson's equivalent, which places more weight on abundant taxa, and not for an incidence-based analogue (Jaccard).

Microbial ecologists have traditionally focused on characterizing the abiotic factors, and host associations that filter microbial taxa from regional species pools into local hosts. Our study adds to the emerging perspective that, in addition to these filters, biotic interactions among microbes fundamentally affect microbial community assembly. Just as keystone species disproportionately influence communities of macro-organisms, we report that a dominant, heritable, plant-associated fungus mediates assembly of co-occurring symbiotic fungi at both regional and local geographical scales. Moving forward, exploring how dominant microbes in other systems interact with their hosts, and one another, to shape the symbiotic milieu, affords an opportunity to begin disentangling the complexity of the microbial world and for making conceptual and theoretical connections between macroscopic and microscopic communities.