Community analysis
The community data of Frankia OTUs was standardized by coverage-based rarefaction (Chao & Jost, 2012) and transformed to presence/absence data. Non-metric multidimensional scaling (NMDS) was performed to visualize Frankia community dissimilarity in this study.
To elucidate whether filtering force drives symbiont assembly into host plants, we generated 1000 null models of symbiont community by seven random samplings from overall rhizosphere source community with frequency distribution of OTUs in rhizosphere soil. We comparedFrankia OTU richness and composition in host’s root-nodules with the null model, using generalized linear model (GLM) with Poisson distribution and permutation MANOVA (PERMANOVA) with 9999 permutations, using the lme4 1.1-21 package (Bates et al., 2019) and the vegan 2.5-5 package of the R 3.6.1 software, respectively.
To test how abiotic environmental factors and genetic variation in hosts affect Frankia communities, we analyzed correlations ofFrankia community dissimilarity with Nei’s genetic distance of hosts, soil pH and inorganic nitrogen contents in soil, using the vegan 2.5-5 package in the R 3.6.1 software.
To quantify variation in symbiont filtering force among hosts, we defined the “symbiosis filtering” index as dissimilarity inFrankia community composition between root nodules and rhizosphere soils in each river area. This symbiosis filtering index indicates how certain Frankia genotypes were filtered from a genetic pool in each river area. In order to quantify differences in symbiosis filtering index, we calculated the symbiosis filtering indices as Jaccard dissimilarities between root-nodules of host individuals and rhizosphere soils in each river area, using 56 individuals of studied alders which had Frankia community data for both the root nodules and rhizosphere soils. Then, the symbiosis filtering indices were averaged for each site. Next, we calculated the Euclidian distance of the averaged symbiosis filtering indices among sites. We referred to the distance as “symbiosis filtering differences.” In addition, to estimate the genetic distance of alder populations among sites, we calculated Nei’s genetic distance (Nei, 1978), based on 1,077 SNPs that had been obtained by RAD-Seq (Kagiya et al., 2018). Finally, the correlation between the Nei’s genetic distance of alders and the symbiosis filtering differences was examined using Mantel tests. All the community analyses were performed with the vegan 2.5–5 package in the R software (version 3.6.1).
To quantify the relative contribution of genetic distance of hosts and other abiotic factors in explaining the spatial patterns of symbiosis filtering, accounting for spatial autocorrelation, generalized dissimilarity modeling (GDM; Blois et al., 2013; Capinha et al., 2015; Ferrier et al., 2007; Lasram et al., 2015) was performed. GDM models for the symbiosis filtering differences included Nei’s genetic distance of hosts, soil pH, and the total amount of inorganic nitrogen in soils, Jaccard community dissimilarity of Frankia in rhizosphere soils, and spatial distance as predictors. To test whether the genetic distance of alders could explain the symbiosis filtering differences without other factors, we estimated the unique and shared contribution of each predictor to the total deviance explained by partitioning the predictors with GDM models (Capinha et al., 2015; Fitzpatrick et al., 2013; Kagiya et al., 2018). To calculate the unique contribution of genetic distance and other factors, the deviance explained of the model removed each predictor was subtracted from the deviance explained of the full model. To calculate the shared contributions of both predictors, a value of deviance explained of the model that was fitted by both predictors was subtracted from the unique contributions of each predictor.