Predicting the global mammalian viral sharing network using phylogeography

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

at the analytical 368 . CC-BY-NC-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/732255 doi: bioRxiv preprint framework and validation we describe were conducted on a global scale, while many 369 zoonotic sampling efforts occur on a national or regional scale. Restricting the focal 370 mammals to a regional pool may improve the applicability of our model in certain sampling 371 contexts, and future studies could leverage higher-resolution phylogenetic and geographic 372 data to fine-tune predictions. In particular, the mammalian supertree 28 has relatively poor 373 resolution at the species tips such that relatedness estimates based on alternative molecular 374 evidence (e.g., full host genome data) may allow more precise estimates of the phylogenetic All analyses were performed in R version 3.6.0 43 . Phylogenetic similarity was calculated 531 using a mammalian supertree 28 as previously described 5 . Pairwise phylogenetic distances 532 were defined as the cumulative branch length between the two species and were scaled to 533 between 0 and 1, and subtracted from 1 to give a measure of relative phylogenetic similarity 534 (rather than distance). Of the 4716 Eutherian species in the mammalian supertree, 591 had 535 virus association records in our fully-connected network and 4196 had known geographic 536 ranges. We used IUCN species ranges to quantify species' geographic distributions 27 . These 537 range maps are generated based on expert knowledge and only comprise species 538 presence/absence information rather than density. We converted all range polygons to 25 km 2 539 raster grids. For each species-pair, we quantified range overlap as the number of raster grid 540 squares jointly inhabited by the two species (in the Mollweide projection, which exhibits 541 equal grid size), divided by the total number of grid squares occupied by these species 542 combined, so that each value was scaled from 0-1: overlapA,B=gridA,B/(gridA+gridB-gridA,B). 543 Disease-related research effort for each host species was quantified as previously described, 544 using counts of studies including species names and disease-related terms such as "virus," 545 "pathogen", or "parasite 5 . To fit citation number as a pairwise trait, we took the smaller of a 546 pair of species' respective citations, and log-transformed the value. Domestication status was 547 defined sensu lato, again as previously described 5 , based on whether a species was ever seen 548 in a domestic setting. We fit this as a binary pairwise trait where 1=at least one of the species 549 was domesticated and 0=neither species had been domesticated. The first term ("s") represents a phylogeny effect smooth fitted across species pairs that did 562 not overlap in space (Gz=1), and "t2" represents a phylogeny:geography tensor product 563 smooth fitted to species that had geographic overlap greater than zero (Gz=0). This allowed 564 us to model these two aspects of the data separately, helping us to more effectively model the 565 large number of spatial zeroes (85% of species pairs did not overlap in space). "mm" 566 represents a multi-membership random effect, accounting for the identity of both species in 567 the pair. We implemented this multi-membership effect to control for species-level effects by 568 including a species-level effect for both the row (Species 1) and co