Synchrony of Bird Migration with Global Dispersal of Avian Influenza Reveals Exposed Bird Orders

Since 2014, HPAI H5, especially clade 2.3.4.4 of the Gs/GD lineage, has caused unprecedented mortality in wild birds and persistent spillovers into domestic poultry, raising concerns for public health despite no sustained human-to-human transmission. Ecological mechanisms proposed for global HPAIV movement include live poultry trade and wild bird migration. Prior work suggests long-distance dispersal via migratory birds, with clade 2.3.4.4 exhibiting properties (e.g., lower pathogenicity to some species but higher transmissibility) that could enable intercontinental spread by infected migrants. However, limited bird movement and HPAIV prevalence data impede broad inference, and the timing and locations where different bird taxa are exposed remain unclear. This study investigates how seasonal bird migration facilitates global HPAIV H5 dispersal and which bird orders are likely exposed at origin and destination regions of virus lineage movements. The authors reconstruct global diffusion histories for clades 2.3.4.4 (2010–2017; 2018–2023) and 2.3.2.1 using HA gene phylogeography, quantify the contribution of seasonal migration versus poultry-associated predictors, model monthly distributions of bird orders using environmental and tracking data, and test synchrony between lineage movement and bird order distributions to identify exposed orders.

The paper situates its research within evidence linking wild bird migration to the spread of HPAIV, including the 2005 Qinghai Lake event and subsequent global dispersal of clades 2.3.2.1 and 2.3.4.4. Clade 2.3.4.4 has been associated with multiple reassortments, greater transmissibility, and widespread intercontinental movements, with phylogenomic analyses supporting introductions to Europe and North America via migratory birds. Previous studies have highlighted roles of Anseriformes and Charadriiformes and suggested that flyways shape transmission dynamics, while poultry may act as evolutionary sinks. Persistent limitations include sparse, uneven bird movement data, limited and localized HPAIV prevalence data compared to LPAIV, and geographic and host sampling biases in virus genomes. Recent work has advocated for higher taxonomic-level analyses and integration of biodiversity datasets, but a gap remains in linking empirical bird movement with viral phylogeography at global and seasonal scales.

Viral genomics and phylogenetics: HA, NA, and six internal gene segments were retrieved from GISAID. Maximum-likelihood trees were inferred (FastTree v2.1.4, GTR model). Time-scaled phylogenies for HA were estimated in BEAST v1.10.4 (with BEAGLE) using SRD06 substitution model and Bayesian Skyride coalescent, with MCMC runs of 300–400 million steps (10% burn-in), sampling every 10,000, and convergence assessed in Tracer v1.7.5. Datasets included 1163 HA sequences for clade 2.3.2.1 and 1845 HA sequences for clade 2.3.4.4 (2018–2023) after down-sampling to balance locations (up to 262 sequences per location). A distribution of 1000 posterior trees per clade was retained for downstream analyses. Discrete trait phylogeography: Countries were aggregated into 10 regions (Africa, Central Asia, Europe, Japan–Korea, North China, South China, Qinghai, Russia, Southeast Asia, USA–Canada). A non-reversible continuous-time Markov chain with BSSVS inferred ancestral locations and lineage diffusion rates. Markov jump counts quantified expected lineage movements between regions and supported seasonal density of lineage movement throughout the year. Predictor modeling for diffusion: A GLM parameterization within a Bayesian phylogeographic framework assessed contributions of predictors to inter-regional diffusion, including a binary wild bird migration network (constructed from Movebank tracking) and poultry-related predictors (international live poultry trade from UN Comtrade; poultry population sizes; within-China trade accessibility scaling). An epoch model allowed time heterogeneity. Competing models were compared by inclusion probabilities and Bayes factors. Wild bird movement and distribution modeling: Tracking data were assembled from Movebank (53 studies; multiple device types; 3542 individuals; ~95 species). Additional tracking data (2006–2019) from China were integrated, yielding data for 10 orders and 96 species. Monthly species distribution models (SDMs) were built at 1-km resolution using environmental covariates (WorldClim, ECOSTRESS NDVI3g, Global Land Cover, SRTM-derived variables). For each order, an XGBoost binary classifier predicted monthly probability of occurrence using presence and pseudo-absence data (67% training/33% test). Reported AUCs included: Pelecaniformes 0.97, Gruiformes 0.97, Passeriformes 0.97, Suliformes 0.98, Ciconiiformes 0.92, Falconiformes 0.95, Charadriiformes 0.94, Anseriformes 0.90, Acipenseriformes 0.90. Synchrony analyses: For routes supported by phylogeography (Bayes factor > 3), monthly time series of Markov jump lineage movement counts were cross-correlated with modeled monthly order-level bird occurrence probabilities at origin and destination regions. Block bootstrapping generated confidence intervals and two-tailed p-values, with Holm’s sequential Bonferroni correction for multiple comparisons. Positive or negative correlations were interpreted as synchrony indicating potential exposure of bird orders at origin/destination.

- Global diffusion patterns: Before 2015, clade 2.3.4.4 showed few, largely unidirectional lineage movements with restricted regional retention. After 2018, movements were more frequent and mixed among regions; trunk diversity was maintained in Russia, Africa, and Europe. - Seasonal migration as key driver: In Bayesian GLM phylogeography, wild bird migration was strongly supported as a predictor of inter-regional HPAIV H5 diffusion, outperforming international live poultry trade and poultry population size, which were not strongly or consistently associated across periods and clades. - Seasonality of lineage movements: For clade 2.3.4.4, lineage movement timings exhibited pronounced seasonality that overlapped spring northward migration/wintering and fall southward migration. Before 2018, most routes (14/20) had a single temporal peak; southward peaks concentrated in fall; some northward peaks coincided with spring. - Exposed bird orders (2018–2023, clade 2.3.4.4): • Ciconiiformes in Europe-to-Africa route: Southern migration from Europe synchronized with increased viral dispersal Europe→Africa (r = -0.47; 97.5% CI: [-0.65, -0.29]; p = 3.58×10^-6), suggesting exposure during southbound migration. • Pelecaniformes in Japan/Korea-to-Russia route: Northern migration from Japan/Korea synchronized with increased dispersal Japan/Korea→Russia (r = -0.32; 97.5% CI: [-0.46, -0.03]; p = 9.86×10^-5), suggesting exposure during northbound migration. - Exposed orders (pre-2018, clade 2.3.4.4): Evidence for Anseriformes and Accipitriformes associated with export routes in Asia. - Clade differences: For clade 2.3.2.1, neither bird migration nor live poultry trade emerged as dominant predictors; no significant bird-order associations with lineage movements were detected, and seasonal bias was less evident (with discussion also noting non-seasonal winter movements for clade 2.3.1.2). - Under-studied hosts: Results point to exposure risks in orders beyond Anseriformes, including Accipitriformes, Gruiformes, Passeriformes, and Pelecaniformes, highlighting broader potential wild-bird reservoirs and interfaces.

The integration of bird migration networks into viral phylogeography reinforces the central role of migratory wild birds in global dissemination of HPAIV H5 clade 2.3.4.4 and helps explain temporal patterns of lineage movement that align with annual migration cycles. While previous studies inferred introductions to Europe and North America via migratory birds, this work directly links an empirically derived migration network and modeled seasonal order-level distributions to virus movement histories. The findings expand attention beyond traditional hosts (e.g., Anseriformes) to include under-studied orders such as Ciconiiformes and Pelecaniformes, suggesting broader ecological interfaces that may facilitate transmission or exposure during migration. Differences between clades (e.g., weaker seasonality and host associations for clade 2.3.2.1/2.3.1.2) imply varying roles of wild birds versus poultry in genetic diversification and maintenance, with evidence that clade 2.3.4.4 may be endemic in wild birds across wider regions post-2018. These insights can refine surveillance by identifying times, routes, and taxa where exposure risk is elevated, thereby informing targeted sampling and monitoring strategies.

This study provides a phylogenomic framework that couples viral phylogeography with empirical bird migration data and modeled seasonal distributions to explain the global dispersal of HPAIV H5 clade 2.3.4.4. Seasonal bird migration best predicts inter-regional virus movement, with lineage movement timings synchronized to migration phases, and identifies potential exposure of multiple bird orders, including under-studied taxa. The approach advances understanding of host–virus ecology and supports prioritizing surveillance at specific times, routes, and taxonomic groups. Future work should: (i) expand and diversify bird tracking to species-level analyses and incorporate migration volume; (ii) reduce genomic sampling bias geographically and taxonomically; (iii) explicitly model time lags and inter-species/environmental transmission pathways; and (iv) adopt structured coalescent methods when appropriate to mitigate sampling biases. Enhanced global wild-bird surveillance and integrated eco-phylogenomic analyses can improve preparedness for increasingly transmissible HPAIV H5 lineages.

- Bird tracking data limitations: Sparse, uneven coverage across species and regions; potential capture bias; coarse spatial resolution for some orders (e.g., Passeriformes using GLS ~200 km). Migration volume not incorporated; migration network treated as binary. - Viral genomic sampling bias: Geographic and host biases may distort inferred routes (e.g., under-sampling in central Eurasia). Inclusion of poultry-derived sequences to preserve genetic diversity may reflect poultry–wild bird transmission, potentially complicating attribution to migration alone. - Analytical assumptions: No explicit time-lag modelling between bird movement and virus lineage movement; limited knowledge of inter-species transmission mechanisms in wild birds. Binary migration network in counterfactual scenarios does not represent absence of migration but normalized relative rates. - Gene focus: Analyses centered on HA; reassortment involving NA and internal genes may influence evolution and diffusion patterns. - Down-sampling constraints: Host species imprecision limited options to down-sample without losing genetic diversity, and sample size at origin influenced model estimates.