The episodic resurgence of highly pathogenic avian influenza H5 virus

Highly pathogenic avian influenza (HPAI) H5N1 activity has intensified globally since 2021, increasingly causing mass mortality in wild birds and poultry and incidental infections in mammals1–3. However, the ecological and virological properties that underscore future mitigation strategies still remain unclear. Using epidemiological, spatial and genomic approaches, we demonstrate changes in the origins of resurgent HPAI H5 and reveal significant shifts in virus ecology and evolution. Outbreak data show key resurgent events in 2016–2017 and 2020–2021, contributing to the emergence and panzootic spread of H5N1 in 2021–2022. Genomic analysis reveals that the 2016–2017 epizootics originated in Asia, where HPAI H5 reservoirs are endemic. In 2020–2021, 2.3.4.4b H5N8 viruses emerged in African poultry, featuring mutations altering HA structure and receptor binding. In 2021–2022, a new H5N1 virus evolved through reassortment in wild birds in Europe, undergoing further reassortment with low-pathogenic avian influenza in wild and domestic birds during global dissemination. These results highlight a shift in the HPAI H5 epicentre beyond Asia and indicate that increasing persistence of HPAI H5 in wild birds is facilitating geographic and host range expansion, accelerating dispersion velocity and increasing reassortment potential. As earlier outbreaks of H5N1 and H5N8 were caused by more stable genomic constellations, these recent changes reflect adaptation across the domestic-bird–wild-bird interface. Elimination strategies in domestic birds therefore remain a high priority to limit future epizootics. Recent resurgences of highly pathogenic avian influenza H5 viruses have different origins and virus ecologies as their epicentres shift and viruses evolve, with changes indicating increased adaptation among domestic birds.

caused unprecedented outbreaks in diverse wild-bird species across five continents and a significant rise in incidental infections in wild carnivores 3 , mink farms 16 and marine mammals 17 .
In this Article, we aim to address the uncertain origins of recent HPAI resurgences and the underlying evolution of HPAI H5Nx viruses in wildand domestic-bird populations.We analysed outbreak data reported to the Food and Agricultural Organization of the United Nations (FAO) and World Organization for Animal Health (WOAH) since 2005 alongside more than 10,000 whole genomes to identify epidemic trends in HPAI ecology and evolution.
Although the 2014-2015 resurgence is notable for spreading across Asia to Europe and North America, resulting in a loss of more than 50 million poultry in the USA 19 , poultry outbreaks in Europe and wild-bird detections globally were relatively minor (Fig. 1a).The 2016-2017 epidemic in wild birds lasted five months, with nearly 400 outbreaks per month at its peak (Fig. 1a).The 2017-2018 season saw fewer outbreaks, but a greater number of wild birds were affected across several regions.Following sporadic detections from 2018 to 2020, more than 200 outbreaks per month were reported in 2020-2021, and more than 400 outbreaks per month were recorded during the 2021-2022 season.Substantial outbreaks have continued into 2023.Additionally, the number of infected wild-bird species increased across all affected regions to varying degrees during the 2020-2021 and 2021-2022 seasons (Extended Data Fig. 2).The semiannual estimate of confirmed HPAI H5 cases in wild birds (predominantly dead birds) peaked at 34,000 during the second half of 2021, although in many instances the number of wild-bird cases reported to WOAH/FAO includes only birds tested and positive for HPAI and is therefore a substantial underestimate (Fig. 1a).

Article
The number of domestic-bird outbreaks generally corresponded to increases in wild-bird outbreaks.Between January and June 2022, more than 69 million domestic birds were culled.Notably, substantial numbers of poultry outbreaks were also recorded in early 2020, just before wild-bird epizootics spiked during April 2020 (Fig. 1a).
Wild-bird resurgence events exhibit seasonal patterns, with widespread outbreaks beginning in November (Fig. 1a and Extended Data Fig. 3).This seasonality is attributed to the arrival of migratory birds from their arctic breeding grounds 20 , coinciding with the 0 °C isotherm 21 .Consistent seasonality has led reporting agencies to describe annual events in waves, starting in September; however, cases in Europe continued through the summer of 2022.

Resurgent HPAI H5 origins
Analysis of HPAI H5 genomes indicates the 2014-2015 and 2016-2017 epidemics originated from independent viral lineages in China (Figs. 1d  and 2a).By contrast, all eight genes involved in the 2020-2021 outbreaks evolved from clade 2.3.4.4bH5N8 viruses first detected in Egyptian poultry in 2016-2017 (Fig. 3, Extended Data Figs. 4 and 5 and Supplementary Data 1).Despite limited HPAI H5 surveillance in surrounding poultry networks, continuous detection of ancestral lineages in Egypt strongly implies regional evolution of the 2020-2021 resurgent viruses.The H5N1 viruses responsible for the 2021-2022 epizootic emerged from H5N8 viruses in Europe in 2020 (mean time of most recent common ancestor (tMRCA) of HA gene 17 August 2020, 95% higher posterior density (HPD) 13 June 2020, 14 October 2020), with an N1 NA and five internal genes (PB2, PB1, PA, NP and NS) derived through reassortment with LPAI circulating in European wild birds since 2019 (Supplementary Data 1).Estimates of tMRCA showed ancestral genes of all three main resurgent lineages (clade 2.3.4.4bH5N8 in 2016-2017 and 2020-2021 and clade 2.3.4.4bH5N1 in 2021-2022) circulated in wild birds for at least a year before causing widespread outbreaks from November onwards.
A separate clade 2.3.4.4bH5N8 virus lineage that derived from the 2016-2017 resurgence (hereafter termed JKE-2019) was initially reported in domestic poultry in central and southeastern Europe during 2019 and early 2020 and later emerged in wild birds in Japan and Korea in 2020-2021 (Fig. 2a).JKE-2019 retained six of eight genes of clade 2.3.4.4b viruses found in west and southern Africa during 2018-2019, whereas the PB1 and NP genes were obtained from LPAI viruses through reassortment 22 (Extended Data Fig. 5).
We reconstructed the diffusion of HA clade 2.3.4.4b using discrete and continuous phylogeographic analyses and found that the initial migration of 2.3.4.4bH5N8 from its inferred ancestors in Egyptian domestic poultry (posterior probability, 0.97, Extended Data Fig. 6) was through the Black Sea-Mediterranean flyway towards western Russia and eastern Europe during 2019 (Fig. 2c and Extended Data Fig. 7), following a pattern observed in previous resurgence events 23 .During mid to late 2019, 2.3.4.4bH5N8 was introduced to the northern coastal regions of central Europe (Fig. 2c).Epidemiological reports indicate local circulation before the resurgence in late 2020 (Extended Data Fig. 3).As sampling in non-endemic regions is often limited to outbreak response, early dissemination in Europe is unclear.Nevertheless, HPAI H5N8 spread rapidly across Eurasia and became established in wild birds and poultry during 2020.By October, clade 2.3.4.4bH5N8 viruses with an HA T204I mutation migrated east to China, acquiring an N6 NA and forming an Asian lineage (tMRCA 9 June 2020, 95% HPD 3 April, 11 August) responsible for 41 of the 65 known human cases of H5N6 infection so far 24 .Continuous phylogeography indicates HPAI H5N1 emerged in eastern Europe during mid-2020, diversifying into two geographically separate lineages (Fig. 2a,d).The lineage that circulated across the northern coastal regions of central Europe in late 2020 (mean tMRCA of HA gene 27 November 2020, 95% HPD 30 September 2020, 24 January 2021) seeded North American outbreaks by means of the East Atlantic flyway with two more HA mutations (L120M and I526V) in mid-2021 (mean tMRCA of HA gene 24 July 2021, 95% HPD 27 May, 19 September).The other European lineage seeded introductions to Africa along the Adriatic flyway around the Mediterranean Sea (Fig. 2d).It acquired a further HA substitution (M548I) before causing outbreaks across Eurasia during 2021-2022, indicating the potential importance of this site for adaptation to rapid dispersal among wild birds.Discrete phylogeography confirmed that eastern Europe predominantly seeded HPAI H5N1 outbreaks in other regions (posterior probability, 0.83), with 20.1% Markov rewards (denoting time spent in the region) and 42.2% of Markov jumps (denoting region transitions) (Supplementary Table 1).More than 20% of Markov jumps with definitive support (Bayes factors (BF) > 100) were between western Europe and northern Europe along the East Atlantic flyway (Supplementary Table 1 and Extended Data Fig. 6).From northern Europe to North America, two Markov jumps had sufficient evidence (BF = 75) (Supplementary Table 1), consistent with our continuous phylogeographic inference and previous studies 25,26 .

Phylodynamics of clade 2.3.4.4b viruses
The emergence of epizootic lineages in Africa and Europe signifies a shift away from Asia as the HPAI H5 epicentre.To infer changes

Article
in dispersion and spillover dynamics, we applied an asymmetric discrete-trait model with Bayesian stochastic search variable selection to reconstruct diffusion of principal HPAI H5 clades among eight geographic regions (Africa, China, Europe, Japan and Korea, North America, South Asia, Southeast Asia and West Asia) and different hosts (domestic and wild birds and humans).In contrast to earlier HPAI H5 clades (2.2, 2.3.2.1 and other 2.3.4.4 subclades including a,c-f, collectively 2.3.4.4x), which primarily circulated regionally in domestic birds with brief periods of rapid dispersal by means of wild birds (Fig. 3), clade 2.3.4.4b viruses showed a considerably longer mean duration of persistence in wild birds (51.7% Markov rewards) than domestic birds, with 76.2% of Markov jumps going from wild birds to domestic birds and an epicentre shift from Asia to Europe (Fig. 3, Extended Data Fig. 8 and Supplementary Tables 2 and 3).and 8).Following the early dispersal of clade 2.2 to Europe and Africa, dispersal from West Asia to Africa by means of Europe was mainly associated with domestic-bird networks, although wild-bird surveillance in Africa was sparse at the time.Wild birds were primarily responsible for transmission between West Asia and Europe.Most of the regional migration of clades 2.2, 2.3.2.1 and 2.3.4.4x probably occurred through domestic poultry networks and associated human activity (Extended Data Fig. 7); however, short-range spread through unsampled wild birds cannot be excluded.
These results indicate that although a predominantly clonal population of HPAI H5N8 caused the 2020-2021 outbreaks, the 2021-2022 H5N1 virus has attained relaxed constraints for reassortment with wild-bird LPAI viruses.

Discussion
Analysis of HPAI H5 episodic resurgence reveals significant shifts in HPAI ecology and evolution, including geographic and host range expansion, increased dispersion velocity and increased reassortment (Fig. 6).
The continuing epizootics are being caused by 2.3.4.4bH5 viruses with origins traced to poultry samples collected in Africa, following a significant wild-bird resurgence of HPAI H5N8 in 2016-2017.The dispersion velocity in wild birds increased substantially in 2020, spreading HPAI H5N8 viruses across Eurasia and leading to the subsequent emergence of HPAI H5N1.Central to these findings is the observation that the 2.3.4.4bH5N8 virus that caused the 2020-2021 outbreaks across Eurasia and Africa was of a stable genotype, whereas the novel panzootic 2.3.4.4bH5N1 viruses have attained relaxed constraints for extensive genetic reassortment, indicating potential for enhanced transmission and maintenance across the domestic-bird-wild-bird interface.Whereas mammalian influenza viruses typically maintain a limited number of stable genome constellations, LPAI viruses in migratory aquatic birds undergo frequent reshuffling, resulting in transient genome constellations 32 .Poultry-adapted LPAI viruses exhibit genetic linkage of the HA and NA segments (for example, H9N2, H7N9), with limited internal gene-segment exchange 33 .The propensity for HPAI H5 to reassort with other domestic poultry viruses, as was recently observed between clade 2.3.4.4 and H9N2 in Burkina Faso 34 , and the recent surge in mammalian HPAI H5 infections are significant causes for concern.
Our findings show clade 2.3.4.4b emergence is shifting from Asia to Africa, underscoring the need for expanded surveillance capacity.Globally, genomic surveillance is limited and biased across time, geography and host species.Despite the substantial number of HPAI H5 outbreaks reported to FAO and WOAH, only 50% of outbreaks and 0.2% of cases were sequenced (Extended Data Fig. 10).Although HPAI H5 sequence availability generally corresponded to regional outbreaks (Fig. 1a and Extended Data Fig. 10), sampling bias was observed at the country level, especially in Africa (Supplementary Data 2).Some countries with endemic HPAI H5 conduct routine poultry surveillance, but many only sample following mass-mortality events.Wild-bird sampling is of lower priority and is often biased towards species that are easily sampled.Despite efforts to subsample equitably (Supplementary Tables 4-8), biased sampling could affect ancestral reconstruction 35 , especially for long-distance dispersal events.Additionally, the use of centroid coordinates may skew continuous phylogeography inferences.Nevertheless, our findings are consistent with the timing of reports emerging from new areas.
Despite the sampling bias towards poultry, we found that the dissemination of clade 2.3.4.4bH5 viruses by wild birds was similar to that of domestic birds, which were previously inferred as key drivers of HPAI H5 spread 29 .Domestic birds' dispersal velocity reflects and is constrained by poultry trade and associated human activity, whereas that of migratory birds is more influenced by ecological determinants such as seasonality, migratory flyways and climate.For example, the primary wild-bird resurgence events in Europe are consistently seasonal (Fig. 1a and Extended Data Fig. 3) and correspond with increased wild-bird HPAI H5 dispersion velocity (Fig. 4).Further, several migration routes of HPAI H5 virus were identified along the Black Sea-Mediterranean, East Atlantic and Adriatic flyways, with reports increasing along the Pacific flyway in South America since late 2022 36 .There is also evidence that migratory behaviour and breeding patterns have recently been affected by climatic variation 37,38 .Expansion and persistence of HPAI H5 viruses in wild birds may therefore continue to be exacerbated by climate change.Future studies will benefit from the integration of host ecology such as species composition, population size and demography and migration patterns (for example, using banding or telemetry data) to inform risk 39,40 .Sustained HPAI H5N1 wild-bird epizootics repeatedly spill over and spill back from domestic birds, increasing zoonotic and pandemic risks.Culling is commonly used to curb HPAI spread in poultry 41,42 , and millions of poultry have been culled to stamp out HPAI H5 outbreaks (Fig. 1).However, continuous culling is unsustainable as HPAI H5 is increasingly persistent in wild birds.Mass mortality in wildlife raises concern around a loss of biodiversity and disruption of ecosystem homoeostasis 43 .Subsidized poultry vaccination is therefore increasingly being considered for outbreak prevention in Europe and North America.Several countries in Asia and Africa use vaccination with variable effects 44 , whereas Hong Kong and parts of China and Southeast Asia combine vaccination with other control methods 45,46 .A key concern is the role of poultry vaccination in driving endemicity and evolution of HPAI H5 lineages 47 .The proximity of poultry networks to significant flyways in northern Africa, the Middle East and eastern Europe, where vaccination practices are diverse and incomplete, is also concerning.To address these issues, it is necessary to enhance global surveillance and improve multifaceted mitigation strategies for outbreak prevention and response.A better understanding of the ecological properties that enhance and sustain transmission in wild birds and the consequences of poultry vaccination with variable uptake will be crucial to mitigate future HPAI outbreaks, which pose unpredictable epizootic, zoonotic and pandemic threats.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06631-2.Avian hosts were classified as domestic and wild birds using strain names, associated metadata and original publications (https://doi.org/10.5281/zenodo.8251324).Location coordinates were used to analyse diffusion in continuous space and were accurate at least to the province/state level for China, Russia and the United States.For discrete analysis and visualization, locations were classified into countries and regions according to the country-region list in NextStrain 52 .
To mitigate sampling bias for phylogeographic analyses, curated datasets were subsampled by either epidemiological information (including country, host and sampling date) or phylogenetic relationship.HPAI H5 HA datasets were subsampled randomly with at most two sequences for clade 2.3.4.4b, five sequences for clade 2.3.4.4x, three sequences for clade 2.3.2.1 and six sequences for clade 2.2 per country per host per season of each year (season 1, January-March; season 2, April-June; season 3, July-September; and season 4, October-December).The final subsampled HA datasets contained 715, 617, 579 and 563 sequences for the four clades.

Phylogenetic analyses
Gene sequences were aligned using MAFFT v. 7.490 (ref.53) and trimmed using trimAL v. 1.4 (ref.54) with a 50% gap threshold followed by manual trimming to the open-reading frame.Maximum-likelihood (ML) trees were generated using IQ-TREE v.2.1.4 (ref.55) with the best-fit nucleotide substitution model.The ML phylogenies were used to check each dataset for molecular clock outliers (sequences that have disproportionately too much or too little root-to-tip divergence for their sampling time) using TempEst v. 1.5.3 (ref.56).The time-scaled ML trees in this study were generated with TreeTime v.0.9. 1 (ref.49).HA phylogeny showing sampling location (country/region), host and clade details of all sequences are available as NextStrain 52 builds at https://nextstrain.org/community/vjlab/episodic-h5/H5.

HPAI H5 poultry and wild-bird outbreaks
All reported and confirmed infections of HPAI H5 viruses in humans were obtained from WHO (https://www.who.int/emergencies/disease-outbreak-news).Confirmed detections/outbreaks in domestic and wild birds globally were obtained from World Animal Health Information System, WOAH (https://wahis.woah.org/)and EMPRES-i+ Global Animal Disease Information System, Food and Agriculture Organization (https://empres-i.apps.fao.org/).

Bayesian evolutionary inference
Divergence times and evolutionary rates were estimated using an uncorrelated relaxed clock model under a Bayesian framework using Markov chain Monte Carlo (MCMC) sampling in BEAST v. 1.10.4(ref.48) and the BEAGLE high-performance library 58 .A flexible Gaussian Markov random field skyride coalescent model and a general time reversible nucleotide substitution model were used with a gamma distribution of substitution rates.At least two independent MCMC chains with 100 million states were performed for each lineage, sampling every 20,000 and discarding 10% as burn-in.Runs were combined to ensure an effective sample size of more than 200 in Tracer v.1.7.1.A subset of 500 trees randomly selected from the posterior distribution was used to generate an empirical tree distribution used in the subsequent phylogeographic analysis, an approach that reduced computational time and burden 28,29 .We used the BaTS 59 to investigate the degree of phylogeographic structure (phylogenetic clustering by sampling location and host), which was compared to a null hypothesis generated by tip randomization with 1,000 replicates.

Discrete phylogeography
To reconstruct spatial diffusion among a set of eight geographic regions (Africa, China, Europe, Japan and Korea, North America, South Asia, Southeast Asia and West Asia) and different hosts (domestic birds, wild birds and humans), we conducted asymmetric discrete-trait phylogeographic analyses with Bayesian stochastic search variable selection in BEAST v. 1.10.4(ref.48).SpreadD3 v.0.9.6 (ref.60) was used to estimate BF to determine statistical significance ('definitive' (BF > 100) or 'sufficient' (100 > BF > 3)).To count all the transitions between states and the time spent in states between transitions, we again applied the continuous-time Markov chains model 61 to complete the Markov jump history over time.We combined three independent chains with five million MCMC steps for each lineage and sampled every 10,000 states.The first 10% of each run was discarded as burn-in, resulting in 1,350 posterior trees with estimates of the ancestral region and host for each internal node.The trunk region/host through time and persistence were measured from these posterior phylogenies using PACT v.0.9.5 (https://github.com/trvrb/PACT),where the trunk consists of all branches ancestral to a virus that was sampled within one year of the most recent sample and the persistence is measured by how long a tip takes to leave its sampled location, counting backwards in phylogenetic trees 62 .To further reveal and quantify the transmission patterns of the panzootic-2020 clade (including 2020-2021 and 2021-2022 resurgences) in Europe, we partitioned Europe and repeated the analysis among eight geographic regions (Africa, Asia outside West Asia, West Asia, eastern Europe, western Europe, southern Europe, northern Europe, Africa and North America).

Phylodynamics incorporating geography and host
To complement the discrete spatial diffusion analysis and reconstruct a more detailed geographic history, we estimated the HPAI H5 diffusion in continuous space (latitude and longitude of country level) using a Cauchy relaxed random walk model 63 with 0.001 jitter window size.Moreover, following the Bayesian method proposed by ref. 28, we incorporated a continuous spatial diffusion process and a discrete host transmission process in a single Bayesian analysis to quantify host-specific diffusion rates and geographic expansion (wavefront distance) for each lineage (https://github.com/vjlab/episodic-h5).At least two independent MCMC chains were performed, sampling every 10,000 steps and discarding 10% as burn-in to ensure effective sample size >200 for each parameter.The continuous phylogeographic analysis was visualized using the R package 'seraphim' 64 and codes from refs.65,66 using the Python library 'matplotlib'.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Fig. 1 |
Fig. 1 | HPAI H5 outbreak reports, subtypes and clades.a, Time series comparing HPAI H5 outbreaks in wild birds (top) and poultry (bottom) by geographic region reported to the FAO.Semiannual counts of HPAI H5affected birds reported to the WOAH are plotted on the right y axis.M, million.b, Temporal changes in HPAI H5 HA clade prevalence estimated using sample collection dates of sequences submitted to the GISAID and NCBI Influenza Virus Resource databases from January 2004 to June 2022.c, Temporal changes in HPAI H5Nx subtype prevalence estimated using observation dates

Fig. 2 |
Fig. 2 | Evolution of clade 2.3.4.4bHA genes and early migration patterns of resurgent HPAI.a, Maximum clade credibility tree with branches coloured by discrete geographic region.Colour bars indicate host and subtype.The posterior distribution of the tMRCA is shown as bar charts on specific nodes.b, Continuous phylogeographic reconstruction of the spread of H5N8 from mid-2014 to 2017.c, Early spread of H5N8 from 2017 to mid-2020 before the

Fig. 4 |
Fig. 4 | The contrasting spatial epidemiology among HPAI H5 clades 2.3.4.4b, 2.3.4.4x, 2.3.2.1 and 2.2.a, Median viral diffusion velocity (km per year) of domestic (brown) and wild birds (blue) over time for each clade.b, Median viral wavefront distance (km) of wild and domestic birds over time for each clade.We recalculated the wavefront distance in the panzootic-2020 clade (including 2020-2021 and 2021-2022 resurgences, shown in a thick line), in which Egypt was regarded as the epidemic's origin.c, Median viral diffusion coefficient (km 2 per day) over time for each clade.The diffusion velocity of wild birds in clade 2.3.2.1 during 2014 is not shown in the plot owing to abnormal estimation value, possibly caused by insufficient sampling.The shaded areas denote the 95% confidence intervals, in which some extreme values are not shown.

Fig. 5 |
Fig. 5 | Circulation of HPAI gene segments in resurgent 2020-2021 and 2021-2022 viruses.Line thickness indicates the number of sequences in a monophyletic clade (with a minimum of five sequences).Internal genes of H5N1 and H5N8 are labelled with yellow and green snowflakes, respectively.tMRCA is shown for monophyletic clades with more than 100 sequences, estimated with BEAST v.1.10.4 (ref.48) for the HA gene and TreeTime 49 for the remaining genes.

Fig. 6 |
Fig. 6 | Graphical summary of the episodic resurgence of HPAI H5N1 in wild birds.HPAI H5N1 initially emerged in domestic birds (first detected in geese) and then became established in poultry through HA diversification and internal gene reassortment with LPAI viruses endemic in poultry.Shaded regions indicate host ecology and avian influenza pathogenicity, with HPAI shown in red and LPAI in wild aquatic birds (blue background) and domestic

. 1 |
Number and proportion of HPAI H5 outbreaks.Data according to the EMPRESS-i+ Global Animal Disease Information System, Food and Agriculture Organization of the United Nations (FAO) (https://empres-i.apps.fao.org/)from January 2004 to May 2022 coloured by subtype.

Fig. 2 |. 3 |. 4 |
Reported to FAO in HPAI H5Nx Outbreaks Extended Data Comparison of affected wild bird species by taxonomic order.Data reported to the EMPRESS-i+ Global Animal Disease Information System, Food and Agriculture Organization of the United Nations (FAO) (https://empres-i.apps.fao.org/) in HPAI H5Nx outbreaks between 2016/17, 2020/21, and 2021/22.Avian orders with fewer than ten outbreaks are not shown.Spatial distribution of HPAI H5 outbreaks in Europe during 2020-2022.EMPRESS-i+ Global Animal Disease Information System, Food and Agriculture Organization of the United Nations (FAO) (https:// empres-i.apps.fao.org/).Outbreaks are coloured by subtype, and maps were generated using the R package "rnaturalearth".Temporal changes in HPAI H5 lineage predominance.(a) The number of HA sequences coloured by lineage since 2020.(b) Proportional lineage distribution by month inferred from (a).
In comparison to earlier clades, dissemination routes of 2.3.4.4b viruses were notable.Clade 2.3.4.4b frequently transmitted between Asia and Europe and endured the longest migration route from China to the USA by means of Europe (Extended Data Fig.7).Clade 2.3.4.4x disseminated frequently from China to Southeast Asia, sporadically to Japan and Korea (over five Markov jumps) and once to North America (Supplementary Table2).Clade 2.3.2.1 migrated between China and Southeast Asia (over five Markov jumps) and, to a lesser extent, to Japan and Korea and Africa by means of West Asia (Extended Data Figs.7 clade 2.3.4.4b lineages persisted in Europe and Africa for around two years, compared to roughly one year in other regions.Furthermore, whereas domestic birds in China occupied the phylogenetic trunk until 2015 and were the probable source of the 2016-2017 wild-bird outbreaks, Europe represented the primary phylogenetic trunk since 2016, with transmission primarily among wild birds.A notable exception occurred around 2019, when more than 50% of the phylogenetic trunk belonged to lineages circulating in domestic birds in Africa (Fig.3).Several HA sublineages of the 2016-2017 resurgence spread from Europe to Africa and became established in both domestic-and wild-bird populations in South Africa and Nigeria.Iran, Denmark and Bulgaria also detected sustained transmission of 2016-2017-like viruses into 2019, whereas non-2.3.4.4b clades predominated in Asia until 2020.Notably, of the nine lineages detected in 2019-2020, five can be traced to poultry viruses detected in Egypt (Fig.2a, https://nextstrain.org/community/vjlab/episodic-h5/H5).

Article Methods Data source and preparation
50AI H5 genomes with at least HA gene and sample information including collection date, location and host species, were obtained from the Global Initiative on Sharing All Influenza Data (GISAID) (https:// platform.epicov.org/)andNationalCenterforBiotechnology Information (NCBI) Influenza Virus Resource (https://www.ncbi.nlm.nih.gov/genomes/FLU/)databases on 11 July 2022.After removing duplicate isolates, laboratory-derived and mixed-subtype isolates, sequences with less than 85% gene coverage and sequences with incomplete collection dates, H5 clades on the basis of the World Health Organization (WHO) gs/Gd H5N1 nomenclature system50were determined using LABEL v.0.6.3 (ref.51).Clade 2.3.4.4 was further assigned into subclades 2.3.4.4a-h on the basis of phylogenetic relationships to WHO H5 candidate vaccine viruses, with known clade assignment estimated using a maximum-likelihood phylogenetic tree generated using the Jukes-Cantor nucleotide substitution model in FastTree v.2.1.1.