Evolving dynamics of H5Nx avian influenza in China revealed by long-term wild bird surveillance Authors: Xiang Li, Xinru Lv, Yi Li, Mengdan Fei, Sipei Zhang, Qing An, Siyuan Qin, Peng Peng, Jinghao Li, Hongrui Liang, Yang Xiu, Zhuoyan Li, Yulong Wang, Yajun Wang, Xiangwei Zeng, Zhijun Hou, Heting Sun, Yu Xu, Yanbing Li, Hongliang Chai Published: 2026 ABSTRACT H5Nx highly pathogenic avian influenza viruses pose persistent threats to poultry, wildlife, and public health. Over the past two decades, their geographic and host ranges have expanded across migratory networks whose epidemiological connectivity has become increasingly apparent through recent surveillance and genomic analyses. To elucidate these dynamics, we conduct long-term nationwide wild-bird surveillance in China, integrating active and passive monitoring. Our analyses reveal the maintenance, reassortment, and transmission of H5Nx viruses in wild birds, highlighting the value of sustained surveillance in capturing viral evolution. We identify distinct ecological patterns among major clades, with 2.3.4.4b showing the widest distribution and acting as the main lineage mediating intercontinental spread. Since 2020, most 2.3.4.4b viruses detected in wild birds in China have clustered with lineages originating outside China, consistent with repeated reintroduction rather than sustained local circulation. This shift underscores the growing role of migratory connectivity in shaping global viral exchange and the need for coordinated international active surveillance. Introduction The H5Nx subtype of highly pathogenic avian influenza virus (HPAIV) first emerged in 1959 in Scotland, marking the beginning of a global challenge for public health and agriculture1. Early outbreaks in Europe and North America were sporadic and largely confined to poultry holdings2. In the mid-1990s, a novel H5N1 lineage emerged in domestic waterfowl and subsequently spread to wild birds and poultry across continents3,4. Through repeated reassortment and intercontinental transmission, this lineage established persistent circulation in multiple regions, giving rise to the globally distributed H5Nx viruses observed today5–9. One of the earliest signals of the virus’s extensive reach was the 2005 outbreak at Qinghai Lake, which resulted in the death of over 6000 wild birds10. This event dramatically underscored the role of migratory birds in the virus’s dissemination, setting the stage for a series of global spread events11–13. Over the following years, outbreaks of various H5Nx clades and subclades continued to spread across multiple continents, with wild birds acting as both vectors and amplifiers of the virus. This pattern was particularly evident during the spread of clade 2.3.2.1c H5N1 (2009-2010), clade 2.3.4.4c H5N8 (2014–2015), and clade 2.3.4.4b H5N8 (2016–2017), which witnessed rapid cross-border transmission through migratory routes5,14–16. These major outbreaks highlighted the virus’s adaptability, particularly as it transitioned from a localized phenomenon to a global threat. For example, in 2016, a novel H5N8 virus from clade 2.3.4.4b re-emerged in wild waterfowl in North-Central Asia, swiftly disseminating into Europe, Africa, South Asia, and East Asia15,17,18. The virus became enzootic in Europe, causing recurrent epidemics in 2016–2017, 2019–2020, and 2020–2021, highlighting the shift to a more complex global viral landscape19–22. The subsequent emergence of a novel 2.3.4.4b H5N1 subtype in 2020 marked another critical phase, one that demonstrated the virus’s continued evolution and reassortment. This novel strain, originating from the reassortment of H5N8 and European HxN1 low pathogenic avian influenza viruses (LPAIV), quickly replaced H5N8 as the dominant circulating subtype and spread across the Atlantic to North and South America, even reaching Antarctica23–28. During the early years, H5Nx HPAIV circulation was largely concentrated in East and Southeast Asia, where several major clades first emerged14,29–32. However, recent epidemiological and genomic studies indicate that the geographic pathways of H5Nx spread and reintroduction are now more widely distributed than in earlier periods, extending beyond East and Southeast Asia20,33–36. This emerging pattern indicates a shift in the virus’s global spread, involving a multitude of continents and migratory bird routes, with significant outbreaks now being reported in Asia, Europe, Africa, and the Americas6,24,25,37. As H5Nx viruses are increasingly detected across a broader geographic and host range, it becomes increasingly important to understand how these shifts are represented within long-term wild-bird surveillance in China. By leveraging data from long-term wild-bird surveillance, we seek to elucidate the pathways of global H5Nx redistribution and offer a clearer understanding of China’s current position in the transmission networks. This insight is critical not only for understanding the past spread of the virus but also for enhancing our predictive capabilities to better anticipate future outbreaks. Results Nationwide surveillance reveals emergent HPAI H5Nx dynamics in wild-bird reservoirs Between February 2003 and February 2023, 219,445 samples were collected from wild birds across 28 provincial-level regions of China. A total of 1344 avian influenza viruses (AIVs) were isolated, yielding an overall isolation rate of 0.61% (95% CI: 0.58%–0.65%). Prior to 2014, surveillance was limited both spatially and in intensity (8155 samples from six provinces), with no detection of H5Nx HPAIVs (Supplementary Table 1). To address these gaps, the Bird Flu Monitoring and Alert System (BFMAS) was launched in 2014, combining passive surveillance of dead or sick birds with active sampling along major migratory flyways. From 2014 to early 2023, 211,290 samples were collected, resulting in 1079 AIV isolations (0.51%; 95% CI: 0.48%–0.54%) (Figs. 1a, 1b and Supplementary Table 1), including 368 H5Nx HPAIVs (121 H5N1, 1 H5N2, 26 H5N6, and 220 H5N8) (Fig. 1c).Fig. 1Spatiotemporal epidemiology of AIV in wild birds in China, 2014–2023.a Geographical distribution of surveillance efforts and AIV isolates. Proportional red circles indicate AIV isolation counts, while color gradient shows total samples collected per province. Blank areas represent unsampled regions. The maps were generated by the authors using ArcGIS v10.2 from administrative boundary vector data. b Monthly patterns of wild-bird sampling and AIV isolation. c Temporal trends of AIV isolation. Top panel: Annual positivity rates for (i) all AIV subtypes (Rate A) and (ii) non-H5Nx subtypes (Rate B). Solid lines indicate annual point estimates of isolation rates, and shaded ribbons indicate the corresponding 95% confidence intervals calculated using the Agresti–Coull method. Differences between Rate A and Rate B were assessed using two-sided two-proportion Z-tests, with Benjamini–Hochberg adjustment for multiple comparisons. Significant differences are indicated by *** for adjusted P values < 0.001. Bottom panel: H5Nx subtype dynamics, with bubble size proportional to annual counts and colors distinguishing subtypes. Source data are provided as a Source Data file. Spatiotemporal epidemiology of AIV in wild birds in China, 2014–2023. a Geographical distribution of surveillance efforts and AIV isolates. Proportional red circles indicate AIV isolation counts, while color gradient shows total samples collected per province. Blank areas represent unsampled regions. The maps were generated by the authors using ArcGIS v10.2 from administrative boundary vector data. b Monthly patterns of wild-bird sampling and AIV isolation. c Temporal trends of AIV isolation. Top panel: Annual positivity rates for (i) all AIV subtypes (Rate A) and (ii) non-H5Nx subtypes (Rate B). Solid lines indicate annual point estimates of isolation rates, and shaded ribbons indicate the corresponding 95% confidence intervals calculated using the Agresti–Coull method. Differences between Rate A and Rate B were assessed using two-sided two-proportion Z-tests, with Benjamini–Hochberg adjustment for multiple comparisons. Significant differences are indicated by *** for adjusted P values < 0.001. Bottom panel: H5Nx subtype dynamics, with bubble size proportional to annual counts and colors distinguishing subtypes. Source data are provided as a Source Data file. Seasonal variation was pronounced, with peak viral activity from November to January and marked declines from June to August (Fig. 1b). Notably, from June 2017 to September 2020, no H5Nx HPAIVs were detected, despite continued circulation of LPAIVs, suggesting possible suppression of H5Nx transmission during this period (Fig. 1b). The overall annual AIV isolation rate ranged from 0.10% to 1.08%, but decreased to 0.02%–0.85% after excluding H5Nx viruses, suggesting that the higher annual detection rates observed in 2015, 2016, 2020, and 2021 were largely driven by H5Nx activity (Fig. 1c, top panel). H5Nx subtype dynamics showed clear temporal transitions (Fig. 1c, bottom panel). H5N1 dominated during 2014–2015, accounting for 75.93% (41/54) of H5Nx detections. Successive waves in 2016–2017 and 2020 marked a shift to H5N8, which comprised 92.75% (192/207) of H5Nx isolations. From 2021–2023, H5N1 resurged, representing 73.83% (79/107) of isolates. These subtype turnovers closely paralleled global transitions of clade 2.3.4.4b HPAIVs and likely reflect lineage replacement influenced by antigenic drift, reassortment, and host immune pressures. Spatiotemporal H5Nx HPAIV detection shaped by surveillance strategies A total of 368 H5Nx HPAIVs were isolated from 46 discrete detection events, revealing pronounced temporal and spatial heterogeneity (Supplementary Fig. 1 and Supplementary Table 2). Here, a “discrete detection event” refers to an epidemiologically linked detection episode defined by a shared location, sampling period, and outbreak context; therefore, multiple isolates obtained from the same site and time window were counted as one event rather than independent events. In 2014–2015, H5N1 predominated (5/6 events) and was mainly detected through passive surveillance during mortality episodes (83.33%, 5/6 events). In 2016–2017, the emergence of H5N8 coincided with an increased role of active surveillance, which identified 57.14% (8/14) of events. After a three-year lull, H5Nx reappeared in late 2020, initially detected passively as H5N8, but rapidly overtaken by H5N1 (16/26 events. The proportion of detections from active surveillance rose from 20% (2/10 events) during the H5N8-dominant phase to 56.25% (9/16) during the H5N1 resurgence (Supplementary Table 2), reflecting a strategic shift toward proactive monitoring of subclinical infections. Geographically, detections clustered along major migratory flyways, notably the Central Asia flyway (Tibet, n = 8) and the East Asian–Australasian flyway (Yellow River Basin: Inner Mongolia, Henan, Ningxia; Yangtze River Basin: Hubei; n = 5 each) (Supplementary Fig. 1 and Supplementary Table 2). Seasonally, 76.09% (35/46) of events occurred between September and February, corresponding to autumn migration and overwintering (Supplementary Table 2). Spatiotemporal analyses revealed recurrent southward viral spread from the Yellow to the Yangtze River basins in 2016–2017, 2020–2021, and 2021–2022, and an unusual coastal-to-inland trajectory from Heilongjiang in 2022–2023 (Supplementary Fig. 1 and Supplementary Table 2). Tibet exhibited a distinctive spring peak, with half of detections (4/8) in May, aligning with early breeding of high-altitude bird species (Supplementary Table 2). These patterns underscore the central role of migratory birds in shaping H5Nx HPAIV dissemination across China. Phylogenetic clade dynamics and genotypic diversification of H5Nx HPAIV in wild birds in China Phylogenetic analysis of 298 H5Nx HPAIVs sequenced in this study, together with all publicly available HA sequences from GISAID/NCBI (2005–2023), identified three major clades circulating in China: 2.3.2.1c (n = 26), 2.3.4.4e (n = 24), and 2.3.4.4b (n = 245) (Fig. 2a and Supplementary Data 1). Three additional isolates fell into clades 2.3.4.4 [A/bean goose/Inner Mongolia/H217/2015 (H5N6)], 2.3.4.4g [A/Bar-headed goose/Tibet/S131/2016 (H5N1)], and 2.3.4.4h [A/Sentinel goose/Hubei/HBQ5/2020 (H5N6)] (Fig. 2a and Supplementary Data 1).Fig. 2Phylogenetic Analysis of H5Nx HPAIVs.a Phylogenetic structure of global H5Nx HPAIV HA genes. The maximum-likelihood tree was reconstructed using 298 H5Nx HA sequences generated in this study (indicated by colored circles), alongside all publicly available HPAIV HA sequences (n = 10,430) collected between 2005 and February 2023, which were retrieved from the GISAID EpiFlu and NCBI GenBank databases. Peripheral concentric ribbons indicate temporal distribution by host type and subtype. Epidemiologically dominant clades are annotated with light gray shadows. b Gene constellations of H5Nx isolates are presented by the order of the HA maximum likelihood (ML) tree. The group of each gene segment is indicated by different colored bars. Detailed phylogenetic trees are available as Supplementary Fig. 2. Phylogenetic Analysis of H5Nx HPAIVs. a Phylogenetic structure of global H5Nx HPAIV HA genes. The maximum-likelihood tree was reconstructed using 298 H5Nx HA sequences generated in this study (indicated by colored circles), alongside all publicly available HPAIV HA sequences (n = 10,430) collected between 2005 and February 2023, which were retrieved from the GISAID EpiFlu and NCBI GenBank databases. Peripheral concentric ribbons indicate temporal distribution by host type and subtype. Epidemiologically dominant clades are annotated with light gray shadows. b Gene constellations of H5Nx isolates are presented by the order of the HA maximum likelihood (ML) tree. The group of each gene segment is indicated by different colored bars. Detailed phylogenetic trees are available as Supplementary Fig. 2. To assess genomic diversity, all gene segments of 285 complete genomes were analyzed (2.3.2.1c, n = 24; 2.3.4.4e, n = 13; 2.3.4.4b, n = 245; others, n = 3). Segment-based phylogenies grouped viruses into distinct lineages and genotypes (Fig. 2b, Supplementary Fig. 2–3, and Supplementary Data 1). In 2020, H5N8 isolates within clade 2.3.4.4b diverged into two sub-lineages, b1 and b2, reflecting substantial genotypic changes compared with the 2016–2017 H5N8 strains, including ongoing evolutionary divergence and reassortment of other gene segments. Within China, b1-like viruses were detected only transiently in 2020 and showed a more restricted distribution, whereas b2-like viruses were more widely distributed and remained detectable into early 2023. Correspondingly, 2.3.4.4b2, but not 2.3.4.4b1, underwent continued reassortment and gave rise to multiple H5N1 genotypes. Clade 2.3.2.1c viruses formed a single genotype. Nine H5N6 isolates recovered during a single detection event in Heilongjiang in November 2016 (2.3.4.4e) shared a conserved genotype, with HA, NA, and PB2 genes closely related to a 2015 strain from Inner Mongolia, whereas an H5N8 virus from Tibet in December 2016 was a reassortant, carrying an HA gene from clade 2.3.4.4e and the remaining segments from clade 2.3.4.4b (Fig. 2b and Supplementary Data 1). Clade 2.3.4.4b viruses exhibited dynamic genotypic shifts over time (Supplementary Fig. 3 and Supplementary Data 1). During 2016–2017, two genotypes predominated: b.G1 (n = 62) and b.G2 (n = 8), differing in PA and NP segments. In October 2020, two novel H5N8 genotypes emerged: b.G3 (n = 12, 2.3.4.4b1-like) and b.G4 (n = 78, 2.3.4.4b2-like), with b.G4 persisting through mid-2021. A reassortant H5N1 genotype, b.G5 (n = 13), combining PB1 of b.G3 with HA/M of b.G4, was detected in May 2021 but was soon replaced by b.G6 (n = 50), which circulated widely from late 2021 through 2022. By early 2023, b.G6 had diversified into four sub-genotypes (b.G6.1–b.G6.4), reflecting ongoing evolution and genetic diversification. Epidemic dynamics of global H5Nx outbreaks To evaluate global H5Nx transmission trends, we analyzed outbreak data reported to the Food and Agriculture Organization (FAO) from 2005 to 2023, revealing substantial spatiotemporal heterogeneity across host types, viral subtypes, and regions. Domestic poultry accounted for most outbreaks throughout the study period; however, wild bird–associated outbreaks increased sharply after 2016. Notably, from autumn 2020 onward, wild birds often outnumbered poultry in outbreak frequency, indicating a shift toward broader ecological dissemination (Fig. 3a and Supplementary Fig. 4a). Early outbreaks were dominated by H5N1, primarily in poultry. After 2016, a subtype shift occurred, with H5N8 driving large-scale outbreaks in both wild and domestic birds. Following a relative decline in 2018–2019, global H5Nx activity surged again in 2020–2023, led by a resurgence of H5N1 and sporadic detections of H5N5 and H5N6 (Fig. 3b and Supplementary Fig. 4b).Fig. 3Epidemiological of global H5Nx HPAI monthly outbreaks and viral sequences, 2005–2023.a Host distribution of H5Nx HPAI outbreaks. b Subtype distribution of H5Nx HPAI outbreaks. c Geographic distribution of H5Nx HPAI outbreaks. d Geographic distribution of H5Nx HPAIV HA sequences. H5Nx HPAI outbreaks were obtained from the EMPRESS-i+ Global Animal Disease Information System, Food and Agriculture Organization of the United Nations (FAO) (https://empres-i.apps.fao.org/). H5Nx HPAIV HA sequences were obtained from GISAID and NCBI. ESEA, East and Southeast Asia; EU, Europe; AF, Africa; AM, Americas; OACR, Other Asian countries and regions. Source data are provided as a Source Data file. Epidemiological of global H5Nx HPAI monthly outbreaks and viral sequences, 2005–2023. a Host distribution of H5Nx HPAI outbreaks. b Subtype distribution of H5Nx HPAI outbreaks. c Geographic distribution of H5Nx HPAI outbreaks. d Geographic distribution of H5Nx HPAIV HA sequences. H5Nx HPAI outbreaks were obtained from the EMPRESS-i+ Global Animal Disease Information System, Food and Agriculture Organization of the United Nations (FAO) (https://empres-i.apps.fao.org/). H5Nx HPAIV HA sequences were obtained from GISAID and NCBI. ESEA, East and Southeast Asia; EU, Europe; AF, Africa; AM, Americas; OACR, Other Asian countries and regions. Source data are provided as a Source Data file. Regionally, East and Southeast Asia and Africa experienced the earliest and heaviest outbreaks. Europe, initially affected only sporadically, saw recurrent seasonal epidemics beginning in 2016, which intensified after 2020. In the Americas, a brief wave occurred in 2014–2015, followed by re-emergence in late 2021, driven by intercontinental spread of clade 2.3.4.4b H5N1, leading to widespread and sustained outbreaks (Fig. 3c and Supplementary Fig. 4c). Sampling bias remains a concern. Although the temporal distribution of genome submissions broadly tracked outbreak peaks, early sequence numbers were low. This gap persisted until after 2014, when sequencing capacity expanded, allowing submissions to better reflect outbreak intensity (Fig. 3d and Supplementary Fig. 4d). Phylogeographic pathways and host shifts of major H5 clades To investigate the origins, maintenance, cross-border spread, and host dynamics of major H5 clades, we performed discrete phylogeographic analyses of HA genes from clades 2.3.2.1c, 2.3.4.4e, and 2.3.4.4b. Analyses incorporated customized geographic partitions and stratified host categories (domestic birds, wild waterfowl, others), enabling reconstruction of lineage-specific transmission routes and host transitions. Clade 2.3.2.1c was initially maintained in wild waterfowl with limited regional movement in Asia between 2009 and 2011. During this period, the virus spilled into domestic birds, though activity remained sporadic at first. After 2013, however, it established stable, regionally distinct sub-lineages in poultry, with occasional spillbacks to wild waterfowl. These spillback events did not establish stable, long-term transmission cycles in wild populations. Most sub-lineages remained regionally confined, though one lineage disseminated from China to North Asia and Africa (Supplementary Fig. 5a). Bidirectional viral exchange occurred between China and both North and Southeast Asia, strongly supported by Bayes factors (>1000; Fig. 4a and Supplementary Table 3). Spatiotemporal reconstruction highlighted North Asia as a key hub for cross-border dissemination (Fig. 4a and Supplementary Table 3).Fig. 4Spatial diffusion of H5Nx HPAIVs.a 2.3.2.1c. b 2.3.4.4e. c 2.3.4.4b. Significant epidemiological unidirectional pathways from one location to another are indicated on the maps. CN, China; EA, East Asia; SEA, Southeast Asia; NA, North Asia; ME, the Middle East; EU, Europe; AF, Africa; SCN, Southern China; NCN, Northern China; JPN, Japan; SKR, South Korea; VTM, Vietnam; AM, Americas. The maps were generated by the authors using ArcGIS v10.2 from administrative boundary vector data. Spatial diffusion of H5Nx HPAIVs. a 2.3.2.1c. b 2.3.4.4e. c 2.3.4.4b. Significant epidemiological unidirectional pathways from one location to another are indicated on the maps. CN, China; EA, East Asia; SEA, Southeast Asia; NA, North Asia; ME, the Middle East; EU, Europe; AF, Africa; SCN, Southern China; NCN, Northern China; JPN, Japan; SKR, South Korea; VTM, Vietnam; AM, Americas. The maps were generated by the authors using ArcGIS v10.2 from administrative boundary vector data. Clade 2.3.4.4e circulation was largely confined to East Asia, with its initial emergence in domestic birds in southern China. A major spillover into wild waterfowl in autumn 2016 triggered geographic expansion, spreading the virus from southern China to Japan (Supplementary Fig. 5b). Japan subsequently acted as a regional hub, facilitating further dissemination to southern and northern China, as well as South Korea (Fig. 4b and Supplementary Table 4). Reverse transmission from Japan back to southern China indicated dynamic bidirectional exchange. Two additional cross-border diffusion routes were supported, linking southern China to Vietnam (BF = 3.93) and Japan to Vietnam (BF = 287.67) (Fig. 4b and Supplementary Table 4). These patterns highlight the role of wild waterfowls in long-distance spread and show how regional epicenters can amplify and redirect transmission across East and Southeast Asia. Clade 2.3.4.4b showed the broadest phylogeographic distribution among the major H5 clades analyses. In the broader multi-host dataset, viruses sampled before 2016 were primarily associated with domestic-bird circulation in China. The emergence of 2.3.4.4b viruses in migratory wild waterfowl in autumn 2016 was followed by rapid intercontinental spread along migratory flyways (Supplementary Fig. 5c). During the 2016–2017 wave (clade 2.3.4.4b0), two major spatial epicenters were evident. The Qinghai–Tibet Plateau was associated mainly with onward spread across Asia, whereas Ubsu-Nur Lake in North Asia was more closely linked to dissemination into Europe (Fig. 4c and Supplementary Fig. 5c). These contrasting regional patterns were consistent with dissemination mediated by migratory wild waterfowl. Despite their different geographic impacts, both pathways were linked to subsequent introductions into African poultry. After 2017, two distinct clades persisted in African poultry populations, consistent with local cryptic circulation and onward evolution6 (Supplementary Fig. 5c). Clade 2.3.4.4b displayed a globally expansive phylogeographic pattern, reflecting its strong capacity for cross-regional spread and its association with an expanded host detection range. However, our data do not allow us to distinguish whether the broader host spectrum reflects intrinsic viral adaptation, increased exposure opportunity, or both. A subsequent 2020 spillover generated two divergent sub-clades: 2.3.4.4b1 and 2.3.4.4b2. Clade b1 caused localized outbreaks in European poultry, then spread via wild birds to East Asia (BF = 9.16) and from East Asia to China (BF = 476.32) (Fig. 4c, Supplementary Fig. 5c, and Supplementary Table 5). In parallel, b2 expanded rapidly across Eurasia, with North Asia serving as a major transmission hub. Notably, clade b2 reached North America through two intercontinental routes: first via a transatlantic introduction from Europe in late 2021, and later through a transpacific introduction from East Asia in late 2022 (Fig. 4c and Supplementary Fig. 5c). In phylogeographic reconstruction, wild-waterfowl-associated viruses are more often linked distinct geographic regions, consistent with long-distance dispersal, whereas domestic-bird-associated viruses more commonly formed regionally clustered lineages indicative of more localized circulation. MCC-based temporal visualization of H5Nx HPAIV migration patterns To visualize host- and region-specific transmission patterns across major H5 clades, we summarized migration load and directionality on maximum clade credibility (MCC) trees. These plots are interpreted as descriptive representations of transition-bearing branches rather than posterior estimates of migration timing. In these MCC-based summaries, clade 2.3.2.1c was initially associated mainly with wild-waterfowl lineages and later became represented predominantly by poultry-associated branches, consistent with more sustained circulation in poultry during the later phase (Supplementary Fig. 6a). Clade 2.3.4.4e showed a comparatively simpler pattern, with poultry-associated branches predominating after 2014 and a wild-waterfowl-associated expansion evident in 2016 (Supplementary Fig. 6b). By contrast, clade 2.3.4.4b showed more heterogeneous host-associated transition patterns, with repeated alternation between poultry- and wild-waterfowl-associated branches, consistent with a more episodic transmission pattern (Supplementary Fig. 6c). Regionally, clade 2.3.2.1c showed prolonged representation in China and Southeast Asia, with recurrent exchanges between the two regions, whereas North Asia was represented more intermittently and was frequently followed by onward spread across Asia and into Africa (Supplementary Fig. 7). For clade 2.3.4.4e, the MCC-based regional summaries were consistent with an origin in southern China and suggested that Japan represented an important intermediate node in subsequent cross-border spread (Supplementary Fig. 8). Clade 2.3.4.4b underwent two major waves of global expansion, first in 2016–2017 and again from late 2020 onward (Supplementary Fig. 9). In the wild-bird-associated dataset, clade 2.3.4.4b was detected in China before 2016 with only limited onward spread, and viruses detected again after 2020 clustered with lineages introduced via East Asia and North Asia following an apparent interruption in local detection. (Supplementary Fig. 9a,b, and d). In contrast, following its dissemination to North Asia, Europe, and Africa, the lineage became endemic in these regions (Supplementary Fig. 9d,f, and g). Since 2020, North Asia and Europe have emerged as the dominant epicenters driving the global spread of clade 2.3.4.4b (Supplementary Fig. 9d and f). Discussion Early influenza surveillance in China involved multiple institutions, primarily focusing on poultry and human populations. For example, the Chinese Academy of Sciences (CAS) reported the first Qinghai-like clade 2.2 H5N1 HPAIV outbreak in wild birds at Qinghai Lake in 200538. However, despite this discovery, surveillance and response systems established by CAS and other institutes remained largely centered on domestic poultry and humans, with limited systematic monitoring of wild birds39. To fill this gap, the BFMAS, established in 2014 with the attention and support of the forestry and grassland authorities, specifically focused on wild-bird populations, establishing comprehensive surveillance along key migratory flyways. BFMAS combined active field sampling with passive mortality surveillance, creating a dual-modality framework capable of detecting both overt outbreaks and silent circulation. This system allowed for the prompt and effective identification of novel gene constellations and subtype shifts, including the transition from H5N1 (2014–2015) to H5N8 (2016–2017, 2020) and the subsequent resurgence of H5N1 (2021–2023). These findings are broadly consistent with recent global studies indicating that clade 2.3.4.4b viruses are increasingly maintained and disseminated through wild-bird-associated transmission networks, and that the geographic center of H5 resurgence is no longer confined to East and Southeast Asia35,40. Our long-term surveillance extends this framework by providing direct evidence from wild birds in China, showing how these broader shifts are manifested locally through subtype turnover, changes in detection pattern, and repeated external reintroductions after 2020. Within the wild-bird surveillance record from China, H5Nx HPAIVs were not detected between 2017 and 2020 despite continued recovery of other AIV subtypes. Subsequent detections reappeared during periods of large-scale seasonal migration, consistent with renewed introduction into the Chinese wild-bird system. Early detections, such as during the 2014–2015 clade 2.3.2.1c events, primarily relied on passive surveillance during mortality events, which suggested high pathogenicity in wild birds41. However, from the 2016–2017 H5N8 wave onward, active surveillance became more prevalent. During the 2021–2023 H5N1 resurgence, over half of detection events were identified through active sampling. The growing dependence on active surveillance indicates that recent H5Nx variants may cause fewer large-scale die-offs in wild birds, making detection increasingly reliant on systematic sampling. It allows detection of low-prevalence or asymptomatic infections and raises the question of whether these viruses have lost some pathogenicity in wild hosts, a point that still needs to be tested experimentally. Wild-bird surveillance for avian influenza remains uneven internationally6,42–44. While the European Union mandates both active and passive monitoring, efforts still rely primarily on passive detection of moribund or dead birds. Active sampling, though expanding, remains inconsistently applied and often voluntary across countries42. As viral ecology shifts and host ranges expand, proactive monitoring of apparently healthy wild birds is becoming a crucial component of effective surveillance systems45,46. In recent years, active surveillance has proven increasingly vital for early detection of H5Nx HPAIVs across Europe, providing timely signals of viral incursions, informing risk-based policy responses, and serving as an early warning system for poultry outbreaks47–50. The pronounced seasonal pattern of H5Nx HPAIV detection, peaking during late autumn and winter months, aligns with key phases of wild-bird migration and aggregation in China, reinforcing the role of migratory dynamics in shaping viral transmission opportunities51. The complete absence of H5Nx HPAIVs between mid-2017 and late 2020, despite continued LPAIV activity, is particularly striking and may reflect ecological or immunological suppression of H5Nx lineages following the 2016–2017 H5N8 wave52. Notably, in years with higher isolation rates, the overall AIV burden was overwhelmingly driven by H5Nx strains, as evidenced by the sharp decline in detection when these viruses were excluded (Fig. 1c). This highlights the dominant role of H5Nx HPAIVs in avian influenza dynamics among wild birds and underscores the importance of subtype-specific tracking to evaluate epidemic risks and the effectiveness of surveillance. Consistent detection hotspots, such as Tibet, the Yellow River Basin, and the Yangtze River Basin, pinpoint critical migratory interfaces for viral amplification. Additionally, spatiotemporal data revealed recurrent southward viral movement and atypical inland incursions, such as the 2022–2023 outbreak in Heilongjiang, emphasizing the need for adaptive, geographically targeted surveillance strategies. In the broader phylogeographic dataset, the geographic structure of H5Nx transmission shifted over time. By contrast, the surveillance record from wild birds in China provides host-specific evidence showing how these broader changes were reflected locally in detection patterns, subtype turnover and repeated external reintroduction. In the early years, East Asia contributed substantially to the emergence and dispersal of several clades, including 2.3.2.1c, 2.3.4.4e, and earlier waves such as 2.2 and 2.3.4.4c10,53. By 2016–2017, however, transmission had become more geographically distributed, with major outbreaks detected both on the Qinghai–Tibetan Plateau and in North Asia at Lake Uvs-Nuur along the Russia–Mongolia border17. Following the global spread of clade 2.3.4.4b, poultry populations in Africa played an increasing role in maintaining viral circulation between 2018 and 2019. Phylogenetic analyses of African H5N8 isolates revealed long internal branches, indicating continued evolution in regions with limited surveillance. Two African-derived H5N8 subclades (2.3.4.4b1 and b2) re-emerged in 2020, demonstrating that sustained underdetected circulation can act as an incubator of viral diversity and a source of new lineages for subsequent global dissemination. This raises an important consideration: uneven surveillance efforts across regions can obscure the evolutionary history and origin tracing of novel viruses9. A striking example is the Gs/Gd-lineage H5Nx virus, whose phylogenetic analysis suggests that although it was traced to the H5N1 viruses detected in turkeys in the United Kingdom in 1991, the long internal branch indicates that it had circulated undetected for several years in regions lacking surveillance54,55. Such gaps remain a major obstacle to understanding H5Nx evolution at the global scale. They can generate long phylogenetic branches and obscure lineage relationships, giving the false impression of abrupt emergence. Sustained active surveillance, particularly in underrepresented regions and along major migratory flyways, is therefore essential for reconstructing viral origins and anticipating future influenza emergence28. From 2020 to 2023, successive waves of H5Nx outbreaks swept across Eurasia, establishing endemic circulation in Europe and North Asia. These regions have become persistent epicenters of HPAIV activity, periodically reseeding outbreaks in Asia, Africa, and other continents36,51,56,57. Intercontinental transmission events exemplify this shift in global connectivity. For instance, H5N1 spread from Asia to North America via the Pacific Flyway in 2015, but during 2021–2022 it crossed the Atlantic from Europe to Newfoundland—marking the first HPAI H5 incursion into North America by this route58,59. Such patterns indicate that viral maintenance and long-distance dissemination are now primarily driven by sustained circulation in northern and western Eurasia, supported by migratory networks that link continents. In the broader phylogeographic dataset, East Asia, including China, contributed prominently to the early emergence and outward spread of several H5Nx clades. By contrast, from 2020 onwards, long-term circulation in North Asia and Europe increasingly shaped the global H5Nx transmission network, with these regions acting as recurrent sources of onward spread across Eurasia and beyond. Within the wild-bird surveillance record from China, recent H5Nx detections have been dominated by incursions in migratory birds, whereas evidence for sustained local circulation in this host system has become less frequent. Consistent with this pattern, H5Nx genotypes detected in Chinese wild birds—including H5N8 (G3, G4) in 2020, H5N1 (G5) in 2021, and H5N1 (G6) in 2022—were genetically closer to viruses circulating in Europe and North Asia than to locally derived variants, supporting repeated reintroduction into the Chinese wild-bird system rather than continued endemic evolution within that host context. This shift may reflect multiple interacting factors, including strengthened biosecurity and long-term H5 vaccination in domestic poultry, both of which may have reduced opportunities for local amplification and reassortment in poultry populations2,60–66. However, the relative contribution of these factors cannot be resolved directly from the present dataset. These findings indicate that recent detections in China largely represent reintroductions via migratory birds rather than ongoing endemic evolution. Collectively, this transition underscores both the effectiveness of domestic control measures and the growing importance of coordinated, cross-border surveillance to mitigate future intercontinental spread. In practice, enhanced wild-bird surveillance supports control in at least three ways. First, active surveillance can detect virus circulation before conspicuous mortality events occur, thereby extending the warning window for poultry biosecurity responses. Second, recurrent hotspots identified through surveillance, such as the Qinghai–Tibetan Plateau, the Yellow River Basin, and the Yangtze River Basin, can inform geographically targeted monitoring and farm-level risk reduction. Third, sustained genomic surveillance enables timely identification of subtype replacement, reassortments, and external reintroductions, which is essential for risk assessment across wildlife, poultry, and public health sectors. Methods Ethics and biosafety statement This study complied with all relevant ethical regulations and was approved by the Research Ethics Committee of Northeast Forestry University. The study was based on routine wild-bird surveillance and did not involve experimental manipulation of live animals. The surveillance protocol was reviewed and approved by the National Forestry and Grassland Administration of China. The experiments with live H5Nx viruses were conducted in the enhanced animal biosafety level 3 (ABSL3+) facility in the Harbin Veterinary Research Institute of the Chinese Academy of Agricultural Sciences, which is approved for such use by the Ministry of Agriculture and Rural Affairs of China. Sampling strategy Surveillance of AIVs in wild birds in China began in February 2003. From 2003 to Feb 2023, a total of 219,445 samples were collected from 28 provincial-level regions. Sampling before 2013 was largely restricted to northeastern China, especially Heilongjiang Province. In 2014, the Bird Flu Monitoring and Alert System was established to enhance detection of H5Nx and H7Nx HPAIVs. Active surveillance consisted of systematic sampling along major migratory flyways, covering Northern China (Heilongjiang, Liaoning, Hebei), the Yellow River Basin (Inner Mongolia, Ningxia, Henan), the Yangtze River Basin (Hubei), and the Qinghai–Tibetan Plateau (Qinghai, Tibet). Sampling on the Qinghai–Tibetan Plateau was suspended for one year in 2020 due to the COVID-19 pandemic67. Most specimens were fresh fecal droppings, with fewer oropharyngeal and cloacal swabs. Passive surveillance was performed through a nationwide network of 749 national and >1000 provincial-level wildlife disease monitoring stations coordinated by the Biological Disaster Prevention and Control Center (formerly General Station for Surveillance of Wildlife Disease & Wildlife Borne Diseases). Swab or organ samples from moribund or dead wild birds were submitted for virological testing. To ensure sample freshness, field teams conducted site surveys before collection. Using binoculars, observers recorded species composition and estimated flock size. After birds departed, fresh, moist, and spatially discrete fecal droppings were collected. To improve representativeness and avoid redundancy, fecal samples did not exceed 20% of the estimated number of birds in each flock. To monitor potential transmission between wild and domestic birds, unimmunized sentinel poultry were free-ranged near wild-bird habitats in Hubei Province and used as indicators of avian influenza virus circulation within shared ecological niches. Sampling, virus isolation, and sequencing All samples were placed in RNase-free tubes with viral transport medium immediately after collection. In the field, samples were stored in portable refrigerators at 4 °C and transported to the laboratory within 24 h whenever possible; for longer transport, they were preserved on dry ice. Upon arrival, samples were stored at –80 °C until processing. Tissue samples were homogenized by liquid nitrogen grinding, then vortexed and centrifuged, while fecal and swab samples were directly vortexed and centrifuged. Supernatants were inoculated into the allantoic cavity of 9-day-old specific pathogen–free (SPF) embryonated chicken eggs. After 72 h incubation, allantoic fluids were harvested and tested for hemagglutination activity. Negative samples were re-passaged up to two times in SPF eggs to maximize virus recovery. HA subtypes were preliminarily identified by hemagglutination inhibition assays with subtype-specific chicken antisera. Viral RNA was extracted from HA-positive allantoic fluids using the QIAamp Viral RNA Mini Kit (Qiagen, Germany) and reverse transcribed with the Uni12 primer. Full genomes were amplified from viral RNA using segment-specific primers targeting the eight influenza A virus gene segments and sequenced by Sanger sequencing68. Primers were synthesized by Comate Bioscience Co., Ltd. (Changchun, China), and primer sequences are provided in Supplementary Table 9. Host species of AIV-positive samples were identified by amplifying mitochondrial cytochrome b or cytochrome c oxidase I gene sequences69–71. Global data sources and processing Data on global H5Nx HPAI outbreaks from 2005 to 2023 were obtained from the EMPRESS-i+ Global Animal Disease Information System maintained by the Food and Agriculture Organization of the United Nations (https://empres-i.apps.fao.org/). Outbreaks were summarized monthly by host type, viral subtype, and geographic location. To complement outbreak records and evaluate discrepancies between reported H5Nx events and sequencing efforts, all available HA sequences of H5Nx HPAIVs collected between 2005 and 2023 were retrieved from the GISAID EpiFlu database and the NCBI Influenza Virus Resource (accessed on 30 June 2024). Duplicate entries, laboratory-derived strains, and records lacking month-specific collection dates were excluded to ensure data quality and temporal resolution. Phylogenetic analysis To assess the global genetic structure of H5Nx HA sequences and assign clades to our isolates, we analyzed all available HA sequences collected between 1 January 2005 and 28 February 2023. This period matched the endpoint of our field sampling, ensuring temporal consistency between genomic and surveillance datasets. Low-quality sequences and redundant entries, defined as identical sequences with the same metadata (host, location, subtype, and collection date) but differing only by laboratory identifiers, were excluded to improve accuracy and reduce computational load. Multiple sequence alignment was performed with MAFFT72 in PhyloSuite v1.2.373. A maximum-likelihood phylogenetic tree was reconstructed using IQ-TREE v1.6.8 with 10,000 ultrafast bootstrap replicates under the best-fit nucleotide substitution model, selected by ModelFinder with the Bayesian Information Criterion74–76. The tree was annotated and visualized in iTOL (https://itol.embl.de/) or ChiPlot (https://www.chiplot.online/) to facilitate interpretation of clade structures and evolutionary relationships77. Following the phylogeny-based genotyping system established for H9N2 LPAIVs, we reconstructed ML phylogenetic trees for each of the eight gene segments of the H5Nx isolates78. Isolates were assigned to genetic groups for each segment based on tree topology, nucleotide identity, and bootstrap support. Genotypes were defined by the combination of segmental groupings, with those represented by fewer than two isolates classified as others to reduce the impact of rare or spurious lineages. Tree construction followed the approach described above. Phylogeography analyses Representative strains from clades 2.3.2.1c, 2.3.4.4e, and 2.3.4.4b, identified in the HA phylogeny, were used to establish three independent datasets. Each dataset was analyzed with tailored geographic partitions and host categories to reconstruct lineage-specific transmission routes and host transitions. Hosts were grouped as domestic birds, wild waterfowl, and others. Geographic regions were defined according to clade distribution: 2.3.2.1c—China, East Asia, Southeast Asia, North Asia, the Middle East, Europe, and Africa; 2.3.4.4e—southern China, northern China, Japan, South Korea, and Vietnam; 2.3.4.4b—China, East Asia, Southeast Asia, North Asia, the Middle East, Europe, Africa, and the Americas. Sequences lacking geographic or host information were excluded to ensure reliable spatiotemporal reconstruction. The temporal signal of each dataset was assessed by root-to-tip regression of genetic distances against sampling dates using TempEst v1.579. Preliminary Bayesian time-scaled trees were reconstructed with BEAST v1.10.480 and guided a subsampling scheme to reduce sampling bias. Long-branch outliers with disproportionate impact on topology were also excluded. For clade-specific datasets, genetically clustered strains with identical host, country, subtype, and month were represented by at least one sequence. Specially, for clade 2.3.4.4e, where Chinese isolates were overrepresented, at least one sequence per host, province, subtype, and month was retained. To reduce the influence of severe sampling imbalance across regions and epidemic waves, structured subsampling was applied before discrete phylogeographic analysis. For clade 2.3.4.4b, where European sequences were substantially overrepresented relative to several other epidemiologically informative regions, we used a baseline cap of 10 sequences per host × subtype × year stratum for Europe as a pragmatic bias-mitigation strategy. To assess the robustness of the resulting phylogeographic inferences to this choice, we repeated the analysis using more stringent and more permissive European caps of 5 and 20, respectively, and compared whether the principal conclusions were preserved across schemes. The subsampled datasets were used for asymmetric discrete-trait phylogeographic analyses with Bayesian stochastic search variable selection in BEAST v1.10.480. The best-fit nucleotide substitution models were identified with ModelFinder76 using the Bayesian Information Criterion. For each dataset, six combinations of molecular clock models (strict or uncorrelated lognormal relaxed) and coalescent priors (constant size, exponential growth, or Bayesian skyline) were tested, and the optimal combination was selected based on effective sample size (ESS) values. Each analysis consisted of at least three independent Markov chain Monte Carlo (MCMC) runs of 80 million iterations, sampled every 20,000 steps. Convergence and mixing were assessed in Tracer v1.7.1 after discarding 10% of samples as burn-in, with all estimated parameters showing ESS > 200. Maximum clade credibility (MCC) trees were summarized in TreeAnnotator, and diffusion rates were extracted to compute Bayes factors (BFs) using SpreaD381. Migration routes were considered supported when BF ≥ 3 and the mean indicator ≥0.4 and further categorized as decisively supported (BF ≥ 1000), very strongly supported (100 ≤ BF < 1000), strongly supported (10 ≤ BF < 100), or supported (3 ≤ BF < 10). For clade 2.3.4.4b, spatial diffusion was additionally reconstructed for sub-lineages 2.3.4.4b0, 2.3.4.4b1, and 2.3.4.4b2 using the same framework. Migration load and direction through time were visualized using a custom Python script (https://github.com/admiralenola/globall4scripts)82,83, in which transition events were mapped to internal nodes on MCC trees. Because this approach is based on a single representative topology, the resulting plots are intended as descriptive visual summaries of transition-bearing branches rather than formal posterior estimates of migration timing. Statistical analysis The 95% confidence intervals for virus isolation rates were calculated using the Agresti–Coull method. Yearly differences were assessed using two-sided two-proportion Z-tests. Standard errors were estimated from binomial variance. P values were adjusted for multiple comparisons using the Benjamini–Hochberg method. Statistical significance was defined as adjusted P < 0.05. All analyses were performed in R v4.4.1. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description Of Additional Supplementary File Supplementary Data 1 Reporting summary Transparent Peer Review file Supplementary Information Description Of Additional Supplementary File Supplementary Data 1 Reporting summary Transparent Peer Review file Source data Source Data Source Data