Fig 1.
Spatio-temporal distribution of avian-origin H9N2 influenza viruses in Asia between 1976 and 2014.
(a) Spatial distribution where the size of circle represents the number of virus isolates obtained in the affected countries/locations. The value of circle indicates double square roots of virus isolate quantity in each location. Colors in the circle represent different fraction of virus hosts. Most of the isolates originate from chickens and from China. The source of map is http://www.naturalearthdata.com. (b) The number of H9N2 HA isolates that are deposited in NCBI through time. The bars represent the annual number of H9N2 isolates sampled in each region. (c) Inferred effective population sizes of H9N2 in Asia by using Bayesian skyline plot. The effective population size of H9N2 viruses increased in 1996 when more isolates began to be sampled in more countries representing multiple outbreaks.
Fig 2.
Time scaled phylogenetic trees of H9N2 influenza viruses in Asia.
(a) estimated using DTA model and (b) using MASCOT model. The colour of tree branches indicates location (see legend) with the maximum probability. A colour change on a branch indicates a virus migration event. Numbers on branches represent posterior probability of displayed location. Numbers in parentheses represent 95% highest posterior density interval of divergence time of the nearest node. A black asterisk represents a virus sequence isolated from wild bird. UAE is short for the United Arab Emirates. Both methods place the origin of H9N2 in Hong Kong, from where it spread to East Asia. This is likely driven by a lack of samples from other locations in the 1970s and 80s. DTA and MASCOT differ in the details on how it spread to West and South Asia. Bars on the right indicate three established lineages based on the phylogenetic relationship between the virus and the representative strains in Asia. The phylogenetic cluster of isolates from domestic poultry in nearby regions indicates their roles in virus spread among neighbouring locations; whereas the dispersal distribution of isolates from wild birds on the phylogeny questions their roles in virus spread across countries and genetic groups.
Fig 3.
Asymmetric migration rate matrix of H9N2 influenza viruses between each pair of locations in Asia.
The migration rate matrix was estimated using DTA, and it describes the virus migration rates between each pair of locations. Unit is the number of migration events per lineage per year. Bayes factors on migration rate over 3 and 20 are labeled by a yellow and a red asterisk at the bottom right of the cell respectively. UAE is short for the United Arab Emirates. The largest and most well-supported rates are between neighbouring locations, suggesting the underlying factors related to geographic proximity can contribute to virus spread.
Fig 4.
Predictors of migration rates of H9N2 influenza viruses between 12 countries/locations in Asia.
The estimated coefficients and inclusion probabilities for potential predictors of migration rates in the DTA model: (a) without and (b) with isolate sample sizes included as potential predictors; in the time-dependent MASCOT GLMs: (c) without and (d) with isolate sample sizes considered as a predictor. The 50% prior mass was specified on no predictors being included. Coefficients represent the contribution of each predictor to the migration rates of H9N2 AIVs when the corresponding predictor was included in the model. Inclusion probabilities are calculated by proportion of the posterior samples in which each predictor was included in the model. Bayes factor support values of 3 and 20 are represented by a thin and thick vertical line respectively in the inclusion probabilities plot. Geographic distance, poultry trade and rainfall seasonality in destination are the most strongly supported factors to virus spread in Asia under cross-validation in these models. Sample size at origin has an effect, but it doesn’t change the support of other predictors.
Fig 5.
Predictors of time-dependent population dynamics of H9N2 influenza viruses within 12 countries/locations in Asia.
Coefficients and indicators as in Fig 4 when estimating the effective population size of H9N2 AIVs (a) without and (b) with considering the effect of virus sample size included as a distinct predictor. The 50% prior mass was specified on no predictors being included. Bayes factor support values of 3 and 20 are represented by a thin and thick vertical line respectively in the indicator plot. Poultry production positively contributes to virus population size. When the number of samples through time in each location is also used as a predictor, the effect of poultry production is much less pronounced for the virus sampling may have been approximately proportional to effective population sizes.