MERS-CoV spillover at the camel-human interface

Gytis Dudas; Luiz Max Carvalho; Andrew Rambaut; Trevor Bedford; Ali M. Somily; Mazin Barry; Sarah S. Al Subaie; Abdulaziz A. BinSaeed; Fahad A. Alzamil; Waleed Zaher; Theeb Al Qahtani; Khaldoon Al Jerian; Scott J.N. McNabb; Imad A. Al-Jahdali; Ahmed M. Alotaibi; Nahid A. Batarfi; Matthew Cotten; Simon J. Watson; Spela Binter; Paul Kellam

This procedure removes lineages coalescing rapidly in humans, which would otherwise introduce a strong signal of low effective population size. These subsampled MERS-CoV sequences from humans were combined with existing sequence data from camels to give us a dataset with minimal demographic signal coming from epidemiological processes in humans. Sequences belonging to the outgroup clade where most of MERS-CoV sequences from Egypt fall were removed out of concern that MERS epidemics in Saudi Arabia and Egypt are distinct epidemics with relatively poor sampling in the latter. Were more sequences of MERS-CoV available from other parts of Africa we speculate they would fall outside of the diversity that has been sampled in Saudi Arabia and cluster with early MERS-CoV sequences from Jordan and sequences from Egyptian camels. However, currently there are no indications of what MERS-CoV diversity looks like in camels east of Saudi Arabia. A flexible skygrid tree prior (Gill et al., 2013) was used to recover estimates of relative genetic diversity (N e τ ) at 50 evenly spaced grid points across six years, ending at the most recent tip in the tree (2015 August) in BEAST v1.8.4 (Drummond et al., 2012) , under a relaxed molecular clock with rates drawn from a lognormal distribution (Drummond et al., 2006) and codon position partitioned (positions 1 + 2 and 3) HKY+Γ 4 (Hasegawa et al., 1985; Yang, 1994) nucleotide substitution models. At time of writing advanced flexible coalescent tree priors from the skyline family, such as skygrid (Gill et al., 2013) are available in BEAST v1 (Drummond et al., 2012) but not in BEAST v2 (Bouckaert et al., 2014) . We set up five independent MCMC chains to run for 500 million states, sampling every 50 000 states. This analysis suffered from poor convergence, where two chains converged onto one stationary distribution, two to another and the last chain onto a third stationary distribution, with high effective sample sizes. Demographic trajectories recovered by the two main stationary distributions are very similar and differences between the two appear to be caused by convergence onto subtly different tree topologies. This non-convergence effect may have been masked previously by the use of all available MERS-CoV sequences from humans which may have lead MCMC towards one of the multiple stationary distributions.