Symbiont and host phylogeny construction

Phylogenetic relatedness can be a source of non-independence between effect sizes—closely related symbionts may respond in the same way to selective pressures, or closely related hosts can similarly affect symbionts (Murfin et al. 2015). To account for phylogenetic non-independence in our models, we constructed phylogenies of the symbionts and hosts included in our analyses. We pruned the tree available at the Open Tree of Life (OTL) with the R packages rotl and ape to build trees containing our species of interest and visualized them using the phytools package (Revell 2012; Michonneau et al. 2016; Paradis and Schliep 2019). When a species was not found in OTL, we found the closest relative available in the genus or family (10 instances; Table S2), then substituted it in place of the missing species. Because some species have more than one effect size, we generated trees such that each species was classified at the population level to match with their corresponding effect size. The phylogenetic distances between “populations” of the same species were <  < 0.00001 (effectively zero), but the population-level designation allow us to distinguish between populations from different studies. We then converted the phylogenies into correlation matrices assuming Brownian motion to incorporate into our phylogenetically-informed meta-analyses.