Phylodynamic Analysis of Ebola Virus in the 2014 Sierra Leone Epidemic

Abstract

Background: The Ebola virus (EBOV) epidemic in Western Africa is the largest in recorded history and control efforts have so far failed to stem the rapid growth in the number of infections. Mathematical models serve a key role in estimating epidemic growth rates and the reproduction number (R0) from surveillance data and, recently, molecular sequence data. Phylodynamic analysis of existing EBOV time-stamped sequence data may provide independent estimates of the unobserved number of infections, reveal recent epidemiological history, and provide insight into selective pressures acting upon viral genes. Methods: We fit a series mathematical models of infectious disease dynamics to phylogenies estimated from 78 whole EBOV genomes collected from distinct patients in May and June of 2014 in Sierra Leone, and perform evolutionary analysis on these genomes combined with closely related EBOV genomes from previous outbreaks. Two analyses are conducted with values of the latent period that have been used in recent modelling efforts. We also examined the EBOV sequences for evidence of possible episodic adaptive molecular evolution during the 2014 outbreak. Results: We find evidence for adaptive evolution affecting L and GP protein coding regions of the EBOV genome, which is unlikely to bias molecular clock and phylodynamic analyses. We estimate R0=2.40 (95% HPD:1.54-3.87 ) if the mean latent period is 5.3 days, and R0=3.81, (95% HPD:2.47-6.3) if the mean latent period is 12.7 days. The estimated coefficient of variation (CV) of the number of transmissions per infected host is very high, and a large proportion of infections yield no transmissions. Conclusions: Estimates of R0 are sensitive to the unknown latent infectious period which can not be reliably estimated from genetic data alone. EBOV phylogenies show significant evidence for superspreading and extreme variance in the number of transmissions per infected individual during the early epidemic in Sierra Leone.

Inferring epidemiological parameters from phylogenies using regression-ABC: A comparative study

“…In the case of outbreaks, this sampling can represent a significant proportion of infected hosts [8, 9]. A time-scaled phylogeny can readily be inferred from virus sequences with known sampling dates.…”

Ebola virus disease outbreak in Nigeria: transmission dynamics and rapid control

“…Those estimates are based on a larger number of cases and represent the average number of secondary cases for a particular country as whole. In contrast, the early phase of the EVD outbreak in Nigeria can be considered a superspreading event (Lloyd-Smith et al, 2005), similarly to the funerals that are suspected to have contributed to the early spread of EVD in Sierra Leone (Gire et al, 2014;Stadler et al, 2014;Volz and Pond, 2014). Assuming that the number of secondary infections cases caused by each case were described by a geometric distribution with a mean of 2.0, the probability that a single individual generates 9 or more secondary cases would be 2.6%.…”

Getting to the root of epidemic spread with phylodynamic analysis of genomic data

“…Consequently, nonparametric models may not always be able to reconstruct complex dynamics, as for dengue virus in southern Vietnam, where seasonality, vector dynamics, and spatial structure were all found to play important roles [10]. Similarly, simplifying epidemiological dynamics may also lead to biases, as in the 2014 West African Ebola epidemic where models that ignore the incubation period returned slightly different estimates than those accounting for an incubation period [12][13][14]. Phylodynamic inference methods have been used to infer complex dynamics of, for example, HIV, hepatitis C virus, rabies, influenza, dengue virus, and Ebola virus.…”

“…In 2009, the WHO Rapid Pandemic Assessment Collaboration was able to rapidly produce an assessment of the pandemic potential of the influenza H1N1 outbreak using phylodynamic methods [2]. During the 2014 West African Ebola virus epidemic, key epidemiological parameters were derived directly from genetic data sampled during the initial phase of the outbreak in Sierra Leone [3,[12][13][14]. Using phylodynamic methods we estimated the basic reproduction number in Sierra Leone to be 2.18 [95% confidence interval (CI) 1.24-3.55] [13], which is in line with the WHO estimate based on surveillance data of 2.02 (95% CI 1.79-2.26) [1].…”

“…However, other estimates based on incidence data alone are highly divergent, possibly due to variations in the reporting rate and sensitivity to model misspecification [13,14]. Moreover, while inferences about population structure are difficult to make from incidence data alone, phylodynamic inferences from genetic data suggest the presence of superspreaders [12,13]. Perhaps most importantly for the current epidemic, it is possible to estimate the number of unreported cases from genetic data, and our estimates for the initial phase of the epidemic agree with independent measures reporting a sampling proportion of around 0.7 during the initial phase of the epidemic in Sierra Leone [3,13].…”