Tissue Tropism and Transmission Ecology Predict Virulence of Human RNA Viruses

Liam Brierley; Amy B. Pedersen; Mark E. J. Woolhouse

We also found viruses primarily transmitted by direct contact and respiratory routes to have a 2 0 0 higher predicted probability of severe virulence than viruses transmitted by more indirect via an indirect route such as through an arthropod vector, virulence could bring ultimate 2 0 7 fitness costs due to host mortality before encountering a vector, fomite, etc.. The relationship between virulence and transmissibility appears more complex. Firstly, the human-to-human transmissibility (level 4) (Fig 5) . This would lend support towards disease for a specific subset of four viruses in the single classification tree (Fig 2) , all 'dead-end' infections) have been predicted to result in higher virulence as without any 2 2 5 evolutionary selection, viral phenotypes within that host will be non-adapted, i.e. a 2 2 6 'coincidental' by-product [24, 25] . However, we did not observe viruses incapable of human-to-2 2 7 human transmissibility to be more virulent, the highest risk instead being observed for viruses dead-end infections, ultimate levels of virulence could also feasibly turn out to be 2 3 0 'coincidentally' low. Taxonomic family being a highly informative predictor in the random forest implies that there 2 3 3 is a broad phylogenetic signal to virulence, but it is also highly likely that the explanatory 2 3 4 power represents a proxy for many other phylogenetically-conserved viral traits that are 2 3 5 challenging to implement in comparative analyses of this scale, such as variation at the 2 3 6 proteomic, transcriptomic or genomic level; or further data beyond simple categorisations, e.g. specific arthropod vector species. Untangling these sources of variation from different scales 2 3 8 of traits will be a critical next step in predictive modelling of viral virulence. We acknowledge several limitations to the quality of our data, as with any broad comparative