Microbes are thought to maintain diversity in plant communities by specializing on particular species, but it is not known whether microbes that specialize within species (i.e., on genotypes) affect diversity or dynamics in plant communities. Here we show that soil microbes can specialize at the within-population level in a wild plant species, and that such specialization could promote species diversity and seed dispersal in plant communities. In a shadehouse experiment in Panama, we found that seedlings of the native tree species, Virola surinamensis (Myristicaceae), had reduced performance in the soil microbial community of their maternal tree compared with in the soil microbial community of a nonmaternal tree from the same population. Performance differences were unrelated to soil nutrients or to colonization by mycorrhizal fungi, suggesting that highly specialized pathogens were the mechanism reducing seedling performance in maternal soils. We then constructed a simulation model to explore the ecological and evolutionary consequences of genotype-specific pathogens in multispecies plant communities. Model results indicated that genotype-specific pathogens promote plant species coexistence-albeit less strongly than species-specific pathogens-and are most effective at maintaining species richness when genetic diversity is relatively low. Simulations also revealed that genotype-specific pathogens select for increased seed dispersal relative to species-specific pathogens, potentially helping to create seed dispersal landscapes that allow pathogens to more effectively promote diversity. Combined, our results reveal that soil microbes can specialize within wild plant populations, affecting seedling performance near conspecific adults and influencing plant community dynamics on ecological and evolutionary time scales. plant-soil feedback | Janzen-Connell hypothesis | species coexistence | community simulation | seed dispersal

650Patterns of diversity and community composition in forests are controlled by a combination of 651 environmental factors, historical events, and stochastic or neutral mechanisms. Each of these 652 processes has been linked to forest community assembly, but their combined contributions to 653 alpha and beta-diversity in forests has not been well explored. Here we use variance partitioning 654 to analyze ~40,000 individual trees of 49 species, collected within 137 ha of sampling area 655 spread across a 900 ha temperate deciduous forest reserve in Pennsylvania to ask: 1) To what 656 extent is site-to-site variation in species richness and community composition of a temperate 657 forest explained by measured environmental gradients and by spatial descriptors (used here to 658 estimate dispersal-assembly or unmeasured, spatially-structured processes)? 2) How does the 659 incorporation of land-use history information increase the importance attributed to deterministic 660 community assembly? And 3) How do the distributions and abundances of individual species 661 within the community correlate with these factors? Environmental variables (i.e., topography, 662 soils, and distance to stream), spatial descriptors (i.e. spatial eigenvectors derived from Cartesian 663 coordinates), and land-use history variables (i.e., land-use type and intensity, forest age, and 664 distance to road), explained about half of the variation in both species richness and community 665 composition. Spatial descriptors explained the most variation, followed by measured 666 environmental variables and then by land-use history. Individual species revealed variable 667 responses to each of these sets of predictor variables. Several species were associated with 668 stream habitats, and others were strictly delimited across opposing north and south-facing slopes. 669Several species were also associated with areas that experienced recent (i.e. < 100 years) human 670 land-use impacts. These results indicate that deterministic factors, including environmental and 671 land-use history variables, are important drivers of community response. The large amount of 672 'unexplained' variation seen here (about 50%) is commonly observed in other such studies 673 Murphy 3 attempting to explain distribution and abundance patterns of plant communities. Determining 674 whether such large fractions of unaccounted for variation are caused by lack of sufficient data, or 675 are an indication of stochastic features of forest communities globally, will remain an important 676 challenge for ecologists in the future. 677 678