Assessing the role of live poultry trade in community-structured transmission of avian influenza in China

Yang, Qiqi; Zhao, Xiang; Lemey, Philippe; Suchard, Marc A.; Bi, Yuhai; Shi, Weifeng; Liu, Di; Qi, Wenbao; Zhang, Guogang; Stenseth, Nils Chr.; Pybus, Oliver G.; Tian, Huaiyu

Since all three AIV subtypes in Chinese poultry exhibited evidence of a continuous process of geographic spread, we sought to test which variables might be associated with AIV lineage movements in China, with the aim of informing prevention and control of the disease. To do so, we employed a Bayesian phylogeographic approach that uses a generalized linear model (GLM) to quantify the contribution of predictor variables to the among-province lineage transition rates. This method was applied to HA sequences from the three independent AIV lineages in China: H5N1, H7N9, and H5N6. To begin, we used the established procedure of considering all potential predictors concurrently. This analysis revealed that many potential predictors are not associated with viral lineage movement (SI Appendix, Figs. S2 and S3). Based on the results of fully specified models, with and without distance, and sensitivity analysis (SI Appendix, Fig. S4 ), we recomputed these phylogeographic analyses using a much smaller set of predictors, specifically those that have been hypothesized by others to affect the large-scale dissemination of AIV (6, 8, 16, 18) : 1) the intensity of the poultry trade among locations in China, 2) the migration of wild birds among locations in China, and 3) distances along the road network in China (see Methods for details). Although minimum road distance is strongly inversely associated with the dissemination of AIV lineages across a range of model configurations ( Fig. 2A) , it is likely that this predictor is simply acting as a proxy for the spatial distance component of the other two predictors. Since our primary hypothesis relates to the linkage of the poultry trade network and the wild bird migration network to AIV dissemination in China, we excluded the minimum road distance predictor from the previous exploratory analyses (Fig. 2B ). We found that for all three subtypes (H5N1, H7N9, and H5N6), the live poultry trade network is positively associated with viral lineage spread, although this association is considerably less certain for H5N1 than for the other two subtypes (Fig. 2B ). In contrast, the wild bird migration network is associated only with dissemination of H5N1 in China and not with the other two subtypes (Fig. 2B) . The robustness of this result was confirmed using a model that directly compares the model inclusion probability of the wild bird migration and poultry trade predictors (SI Appendix, Table S1 ). Following these results, subsequent network analyses of these independent datasets were undertaken to explore how the AIV diffusion process is shaped by the structure of the live poultry trade network.