Predicting the global mammalian viral sharing network using phylogeography

Gregory F Albery; Evan A Eskew; Noam Ross; Kevin J Olival

Predictors of viral sharing 68 We fitted a model designed to partition the contribution of species-level effects and pairwise 69 similarity measures to mammalian viral sharing probability. We used a published database of 70 1920 mammal-virus associations (excluding humans) as a training dataset 5 . These data 71 included 591 wild mammal species, equalling 174345 pairwise host species combinations, 72 with 6.4% connectancethat is, 6.4% of species pairs shared at least one virus. We used a 73 generalised additive mixed model (GAMM) framework, including a species-level effect in 74 our model as a multi-membership random effect, capturing variation in each species' 75 connectedness and underlying viral diversity (see Methods) . Overall, our model accounted 76 for 44.8% of the total deviance in pairwise viral sharing, with 51.1% of this explained 77 deviance attributable to the identities of the species involved (i.e., the species-level effect). 78 Our model structure was effective at controlling for species-level variation in our dataset: i.e., 79 the term had a strong impact on the centrality of each species when we simulated networks 80 using just these parameters ( Figure SI1 ). This observation suggests that ~50% of the dyadic 81 structure of observed viral sharing networks (in contrast to the true underlying network) is 82 determined by uneven sampling and concentration on specific species, and the remainder by 83 macroecological processes. remaining 49% of explained model deviance ( Figure 1A -C). Geography, phylogeny, and their 99 interaction all showed strong nonlinear effects, with geographic overlap in particular driving 100 a rapid increase in viral sharing that began at ~0-5% range overlap values, peaked at 50% 101 overlap values, and then levelled off ( Figure 1B ). This effect closely mirrors previous 102 observations of strong, nonlinear effects of geographic and phylogenetic similarity 103 determining within-order viral sharing 14,16-19 . Although occupying little of the visual space 104 within the model presentation, 93% of mammal pairs had less than 5% spatial overlap ( Figure 105 1B,D). The great majority (86%) of mammal pairs in our dataset did not overlap 106 geographically and rarely shared viruses unless phylogenetic similarity exceeded ~0.5 107 ( Figure 1A ). This phylogenetic distance corresponds roughly to order-level similarity; that is, 108 if two species did not overlap in space, it was highly unlikely that they shared a virus unless 109 they were within the same taxonomic order (8% of pairs varied widely across all groups of viruses ( Figure SI2 ; Table SI1 ), while the influence of host 128 phylogenetic relatedness was more consistent ( Figure SI3 ; Table SI1 ). Generally, host 129 phylogeny was more important in determining sharing of DNA viruses than it was for RNA 130 viruses, while space sharing was more important for vector-borne RNA viruses, and less so ecology, transmission, and evolution: for example, RNA viruses are fast-evolving, allowing 133 them to more quickly adapt to novel hosts, such that phylogenetic distances are less important 134 in determining viral sharing patterns 23 . Conversely, DNA viruses are more evolutionarily 135 constrained, with an evolutionary rate typically <1% that of RNA viruses, such that 136 phylogenetic distance between hosts presents a more significant obstacle for sharing of DNA The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.