Most human infectious diseases are initially transmitted from animals. An analysis of all known mammalian viruses improves our understanding of such cross-species spillover, with potential benefits for public health. See Letter p.646
In ancient Rome, priests divined the future by examining animal entrails. Today, scientists attempt to predict the emergence of human pandemics by surveying the pathogens carried by animals. Most infectious human diseases, including newly emerging ones, involve pathogens initially transmitted from other animals (these diseases are called zoonoses). Viral zoonoses, such as HIV, pandemic influenza and Ebola, are particularly concerning, given their track record of devastation. On page 646, Olival et al.1 provide the most comprehensive view yet of past, present and future virus-sharing between humans and other mammals.
Efforts to understand the drivers of zoonotic risk have been dogged by concerns about biases arising from uneven research focus across species and regions, and by challenges in untangling alternative hypotheses. For instance, are we at greater risk of zoonoses from apes, which share our genes, or from rats, which share our cities? Influential studies2,3,4 have addressed subsets of these issues or particular host types. Olival et al. confront the full challenge head-on, amassing a database of more than 2,800 animal–virus associations that span all known viruses of mammals and all major orders of terrestrial mammals. The authors used statistical models to attempt to assess all hypotheses at once, while controlling for uneven research efforts. Their work paints a far-reaching picture of the factors that govern how many viruses (total and zoonotic) are carried by mammalian hosts.
Pleasingly, this broad canvas consolidates many earlier findings, often adding nuance. For instance, a host's relatedness to humans was known to affect its propensity to carry zoonotic viruses2, but the authors' analysis shows that the detectable effect arises almost entirely from varying patterns of virus carriage between primates. Bats, primates and rodents carry the highest proportions of zoonotic viruses, but only bats carry significantly more than other species after controlling for confounders, reinforcing an earlier finding3. An analysis of the viral traits linked to zoonotic potential (aimed at predicting whether a newly discovered virus can infect humans) confirmed that an ability to replicate in the cell cytoplasm increases such potential5. This analysis also solidified the intuitive idea that viral generalists, those capable of infecting a wide range of animals, are most likely to be able to infect humans.
The study's most ambitious aim was to predict where and in which host species we might find 'missing' zoonoses — those that exist, but have not yet been detected. The authors extrapolated their models to predict the number of viruses that would be found per host species if all hosts were subject to elevated research efforts, then subtracted from this the number already known for each host (echoing a strategy used to predict future zoonotic diseases from rodents4). This approach yielded enticing predictions of the species and regions that are most likely to harbour missing zoonoses, such as bats in northern South America and carnivores in East Africa. Although caveats apply, and the authors detected biases in their predictions for some regions, these findings could guide future investment in global virological surveys.
Olival and colleagues must be commended for their robust, methodical approach to tackling this immense problem, the unruly complexity of which is arguably matched only by its value for public health. The authors have brought order to our understanding of virus sharing between humans and other mammals. But the potential of these findings to shape policy and research priorities demands a clear-eyed view of the challenges ahead.
The study aimed to predict zoonotic spillover, but each word in this phrase bears careful examination. First, predict. The researchers considered total virus numbers per species, the proportion of viruses that are zoonotic per species and the zoonotic potential of each virus. For each of these outcomes, they produced models to explain observed patterns (taking into account various biological factors, as well as research effort), which can be pared down to models that predict outcomes for unstudied species (considering only biological factors, because research effort has no effect on true outcome). The authors are exemplary in stating the explanatory power of each model, including the lower power of predictive models. But none of the key outcomes can be predicted with better than 30% accuracy. We are therefore far from a predictive era in zoonotic disease epidemiology. As the authors state, their predictions are best used to prioritize research and viral-surveillance efforts, not to drive specific policy decisions.
Second, zoonotic. Imagine dividing viruses into nested groups on the basis of their potential — realized or not, and observed or not — to infect humans (Fig. 1). When we say zoonotic, which group do we mean? The authors' analysis of zoonotic potential addresses viruses that may never have encountered humans. Their missing zoonoses include viruses that have infected humans, or may imminently do so, but are not yet recognized. Meanwhile, existing data on zoonotic viruses cover only those known to have infected humans. How well can we project patterns from known zoonoses onto broader groups, given that host and viral traits can differ systematically between them? For instance, viruses that cause more-severe disease are more likely than others to be known, all else being equal. Olival and colleagues' predictions are, necessarily, based on current knowledge. They provide a benchmark against which future data can be compared to chart possible biases arising from projecting trends across these (perhaps dissimilar) groups.
Finally, spillover. The paper addresses whether a given virus is zoonotic and so can spill over from animals to humans, but this is quite different from the quantitative risk of spillover. For example, Olival and colleagues' analysis would not distinguish between Lassa virus (which spills over tens of thousands of times annually6) and Lujo virus (which has only ever spilt over once, to our knowledge6). Systematic study of quantitative spillover risks will require approaches that integrate the relevant mechanisms, which occur at scales of molecules to landscapes7.
Looking forward, Olival et al. call for investment in viral surveillance, which would expand our knowledge of potential zoonoses — particularly if it involves epidemiological metadata and rigorous data-sharing. But although most pandemics are zoonoses, most zoonoses do not cause pandemics, so it is essential not to invest in broad, shallow surveys at the expense of understanding what determines pandemic potential. The crucial factor for a pandemic is human-to-human transmission, which is governed by viral traits8 and population susceptibility9 and mobility10. Gaining insights into transmissibility will require in-depth field and laboratory studies, combined with the development of quantitative methods to integrate the diverse data streams involved. In this endeavour, data-driven mechanistic models might end up being the new animal entrails. Footnote 1
Olival, K. J. et al. Nature 546, 646–650 (2017).
Davies, T. J. & Pedersen, A. B. Proc. R. Soc. B 275, 1695–1701 (2008).
Luis, A. D. et al. Proc. R. Soc. B 280, 20122753 (2013).
Han, B. A., Schmidt, J. P., Bowden, S. E. & Drake, J. M. Proc. Natl Acad. Sci. USA 112, 7039–7044 (2015).
Pulliam, J. R. & Dushoff, J. J. Infect. Dis. 199, 565–568 (2009).
Bausch, D. G. et al. in Viral Infections of Humans: Epidemiology and Control (eds Kaslow, R. A. et al.) 147–171 (Springer, 2014).
Plowright, R. K. et al. Nature Rev. Microbiol. http://dx.doi.org/10.1038/nrmicro.2017.45 (2017).
Geoghegan, J. L., Senior, A. M., Di Giallonardo, F. & Holmes, E. C. Proc. Natl Acad. Sci. USA 113, 4170–4175 (2016).
Gostic, K. M., Ambrose, M., Worobey, M. & Lloyd-Smith, J. O. Science 354, 722–726 (2016).
Pybus, O. G., Tatem, A. J. & Lemey, P. Proc. R. Soc. B 282, 20142878 (2015).
Related links in Nature Research
Rights and permissions
About this article
Cite this article
Lloyd-Smith, J. Predictions of virus spillover across species. Nature 546, 603–604 (2017). https://doi.org/10.1038/nature23088
This article is cited by
Global estimates of mammalian viral diversity accounting for host sharing
Nature Ecology & Evolution (2019)