Spatiotemporal reconstruction of the North American A(H5N1) outbreak reveals successive lineage replacements by descendant reassortants

Abstract

The November 2021 introduction of highly pathogenic avian influenza A(H5N1) clade 2.3.4.4b into North America triggered a devastating outbreak, affecting more than 180 million domestic birds and spreading to more than 80 wildlife species across Canada and the US. From this outbreak, we have sequenced 2955 complete A(H5N1) viral genomes from samples collected in Canada and, in conjunction with previously published data, performed multifaceted phylodynamic analyses. These analyses reveal extensive diversification of A(H5N1) viruses via reassortment with low-pathogenic avian influenza viruses. We find evidence of repeated ancestral strain replacement by direct descendants, indicative of compounding viral fitness increases. Spatiotemporal modeling identified critical geographic areas facilitating transcontinental spread and demonstrated genotype-specific host dynamics, offering essential data for ongoing control and prevention strategies.

Phylodynamics show the reassortant progeny of Eurasian H5 clade 2.3.4.4b and American LPAI viruses outcompeted their ancestors.

INTRODUCTION

Wild birds, particularly those belonging to the orders Anseriformes (waterfowl) and Charadriiformes (shorebirds), serve as the primary natural reservoir for low-pathogenicity avian influenza viruses (LPAIVs) (1–4). Poultry infected with LPAIVs typically display mild clinical signs, including respiratory symptoms such as coughing, sneezing, and nasal discharge. However, LPAIVs with H5 and H7 hemagglutinin (HA) subtypes may evolve into highly pathogenic avian influenza viruses (HPAIVs) by acquisition of multiple basic amino acids at the HA cleavage site motif (5–9). This conversion to HPAIV generally occurs in poultry farms but has also been documented in wildlife (6). HPAIVs can cause severe localized outbreaks with high mortality rates in poultry, leading to substantial economic losses. However, their transmission, adaptation, and persistence in wild populations have traditionally been limited (10, 11). Conversely, A(H5Nx) viruses belonging to the A/Goose/Guangdong/1/96 (GsGd)–derived lineage have steadily increased their geographic and host ranges, becoming atypically persistent in wild bird populations. This persistence challenges the traditional understanding of HPAIV ecology and raises concerns about the potential for long-term circulation of these viruses in wild avian reservoirs (12–14).

The GsGd lineage H5 subtype viruses first emerged in Guangdong, China, in 1996 and have since evolved into multiple clades and subclades (12–14). Particularly, viruses from the GsGd-derived H5 subclade 2.3.4.4 have been responsible for severe global outbreaks since 2014 and are now established in European wild bird populations (14–18). The initial North American incursion of GsGd A(H5Nx) was reported in 2014–2015 when clade 2.3.4.4c A(H5N8) was introduced via the Pacific flyway (19–21). This Eurasian A(H5N8) reassorted with North American LPAIVs, generating unique A(H5N2) and A(H5N1) viruses (19–21). These viruses circulated until mid-2015, marking the first Eurasian-origin HPAI outbreak in North American poultry.

In November 2021, the reemergence of GsGd A(H5N1) viruses belonging to clade 2.3.4.4b in Atlantic Canada marked a severe epidemiological event (22, 23). This resurgence was attributed to genetically distinct H5Nx viruses that originated from reassortment between H5N8 and LPAIVs in The Netherlands in 2020 (24). By 2021, this virus had disseminated across Europe via wild bird migration, causing substantial mortality in both wild and domestic avian populations (25). The initial North American detection occurred in a great black-backed gull (Larus marinus) in St. John’s, Newfoundland (NL), with the isolate designated as A/great black-backed gull/NL/OTH-0114 (23). Subsequent recurring introductions were observed along both Atlantic and Pacific coasts throughout 2022, precipitating an unprecedented epizootic event that rapidly spread across the Americas (26–28). This outbreak has had profound ecological and economic ramifications (29). Since its inception, the clade 2.3.4.4b A(H5N1) virus has demonstrated remarkable host plasticity, being detected in 12 taxonomic orders encompassing more than 80 avian and mammalian species (30, 31). The virus has successfully dispersed across all major migratory flyways in Canada, facilitated by wild bird movements (30, 31). This rapid geographical expansion has resulted in frequent spillover events into domestic poultry populations. As of February 2025, the cumulative impact on poultry in Canada and the United States has affected more than 180 million birds, underscoring the substantial veterinary and economic challenges posed by this epizootic (32, 33).

The persistence and spread of this outbreak in North America may be supported by the extensive diversification of clade 2.3.4.4b A(H5N1) viruses, as they have frequently reassorted with North American LPAIVs (30, 31). Influenza A virus (IAV) genomes are composed of eight discrete segments of single-stranded, negative-sense RNA. This genomic arrangement enables reassortment when two (or more) viruses that infect the same cell exchange segments to create hybrid progeny with a unique genomic constellation (34). All but three segments of Eurasian-origin A(H5N1) viruses have reassorted with American-origin LPAIVs, resulting in a number of novel genome constellations (30, 31). The impact of reassortment on ecological, animal, and human health is substantial, as it may lead to host range expansion and human infections (18, 35–39). To better understand this phenomenon, we analyzed 2955 complete A(H5N1) genomes from Canadian wildlife and domestic poultry samples collected between November 2021 and September 2024, combining them with previously published data (31). We conducted phylodynamic analyses to trace the emergence, dispersal, and diversification of A(H5N1) reassortant genotypes across the Americas. By examining whole-genome sequences, collection dates, locations, and host information, we further provide valuable insights into the evolution, spatiotemporal spread, and host dynamics of A(H5N1) clade 2.3.4.4b in the Americas. This comprehensive analysis provides crucial information for understanding the ongoing A(H5N1) outbreak in the Americas and its potential implications for animal and human health.

RESULTS

Viral genome diversity

We sequenced 2955 complete A(H5N1) HPAI genomes derived from specimens obtained throughout Canada between November 2021 and September 2024. These samples were collected from all Canadian provinces and territories, spanning multiple seasonal cycles and mass bird migration events (30). While the proportion of sequenced samples relative to the total A(H5N1) infections in Canada is comparatively limited, the used sampling strategy is likely to be representative of the viral populations circulating during the study period. However, it is important to acknowledge the possibility that some A(H5N1) genetic diversity may have eluded detection in this sequencing effort.

Whole viral genome sequencing revealed extensive diversification of A(H5N1) clade 2.3.4.4b viruses through reassortment with LPAIVs naturally circulating in wild birds. The US Department of Agriculture’s genotype classification tool GenoFLU identified 28 unique genotypes among these sequences (31). However, genotypes could not be assigned to 144 of the 2955 sequences collected in Canada, suggesting the presence of reassortment patterns not detected in the United States at the time of this study. Analysis of these 144 viruses using single-sequence phylogenetic trees (fig. S1) revealed a total of 62 unique A(H5N1) clade 2.3.4.4b genotypes among the Canadian viruses (Fig. 1). Five of the detected genotypes (21.FAV0033, A1, A2, A3, and A5) are of wholly Eurasian origin, representing independent incursions of the virus into Eastern and Western Canada (23, 26). The remaining 57 genotypes are reassortant viruses containing segments derived from both Eurasian A(H5N1) and American LPAI viruses (Fig. 1, fig. S1, and table S1). Evidence of reassortment with American LPAI viruses was observed for all segments, except the surface proteins [hemagglutinin (HA) and neuraminidase (NA)] and the matrix (M) proteins. This finding is consistent with the notion that the M segment is typically inherited with HA during reassortment events (40–42). The extensive genetic diversity and frequent reassortment events observed among these viruses underscore the importance of continued surveillance and genomic analysis to monitor the evolution of A(H5N1) clade 2.3.4.4b viruses in North America.

Fig. 1. Emergence and surveillance of H5N1 clade 2.3.4.4b genotypes in Canada.

(A) Number of unique viral genotypes based on whole-genome sequence detected over time. (B) Relative sampling frequency of viral genotypes over time. (C) Reassortant network depicting the evolutionary origin of each genotype. The network is derived from a timescaled maximum likelihood tree based on concatenated nonreassorted genome segments HA, NA, and M (fig. S2) with discrete character reconstruction of viral genotypes at each node. Arrows represent reassortment events with the ancestral genotype at the base of the arrow and the derived genotype at the head. Only major reassortant genotypes (>150 samples) and their most recent Eurasian ancestor (A1) are given a unique color; all others are colored black.

The temporal distribution of A(H5N1) virus detections in Canada from November 2021 to September 2024 exhibited a distinct epidemiological pattern (Fig. 1). The primary peak in viral detection occurred in May 2022, coinciding with the initial northward avian migration postincursion (Fig. 1B). Subsequent peaks were observed in October 2022 and October 2023, correlating with autumnal migratory periods (Fig. 1B). These epidemiological peaks coincided with the emergence of new reassortant genotypes (Fig. 1A). Most genotypes were detected in relatively small numbers, as 46 genotypes were detected fewer than 25 times (Fig. 1 and table S1), while the remaining 16 genotypes comprised 95.0% of all viruses sequenced in Canada between November 2021 and September 2024 (table S1). Specifically, seven predominant genotypes were identified in more than 150 samples, collectively accounting for 72.4% of the total sequenced viruses. These findings underscore a complex interplay between viral evolution, host migration patterns, and selective pressures. The observed predominance of a limited number of genotypes within the broader viral population suggests that these strains have adaptive advantages, potentially enhancing their viral fitness and transmission dynamics.

To elucidate the evolutionary history of each genotype, we constructed a timescaled maximum likelihood phylogenetic tree using concatenated nonreassorted segments (HA, NA, and M) from 2955 viruses (fig. S2). The reconstruction of genotype as a discrete character across the tree supported previous findings that only a single reassortant originated from the incursion on the west coast of Canada (Fig. 1C) (26, 43). In contrast, the remaining 56 reassortants descended from incursions on the East Coast of North America. Notably, among the seven most frequently detected genotypes in Canada, six were reassortant viruses (B1.2, B1.3, B2.1, B3.2, B3.6, and B4.1), all of which shared a most recent common ancestor with a wholly Eurasian genotype A1. Specifically, genotypes B2.1 and B4.1 were direct descendants of this Eurasian ancestor, while genotypes B1.2 and B3.2 were secondary reassortments, and genotypes B1.3 and B3.6 were tertiary reassortments (Fig. 1C).

Viral phylogeography

We applied Bayesian continuous phylogeographic inference to investigate the emergence and dispersal of major A(H5N1) 2.3.4.4b reassortant genotypes (B1.2, B1.3, B2.1, B3.2, B3.6, and B4.1) and their most recent Eurasian ancestor (A1) across the Americas (Figs. 2 and 3 and figs. S3 and S4). As genotypes 21.FAV0033, A1, A2, and A5 are homologous to the European National Reference Laboratory genotype “C” (A/Eurasian Wigeon/Netherlands/1/2020), they were combined for phylodynamic analyses (31). Our analysis estimated the time to the most recent common ancestor (TMRCA) of all wholly Eurasian genotypes to be 4 August 2021 in NL, Canada, with a 95% highest posterior density (HPD) interval spanning 10 July to 15 September 2021 (Table 1, Fig. 3A, and fig. S4). However, the earliest documented incursion of A(H5N1) into North America occurred on 4 November 2021 in a great black-backed gull collected in St. John’s, NL. Following this initial incursion into Canada, Eurasian genotypes 21.FAV0033, A2, and A5 remained largely confined to the Maritime peninsula. In contrast, genotype A1 rapidly disseminated southwestward, reaching southern Ontario and the South Atlantic and Midwestern regions of the United States by January 2022 (Fig. 2A and figs. S3 and S4). This genotype subsequently reassorted, giving rise to genotypes B1.2 and B2.1 in southern Ontario and the Midwestern United States, respectively (Figs. 1 to 3 and figs. S3 and S4). Genotype A1 also moved north into Western Canada during spring 2022 but was not widely detected throughout this region. It continued to circulate in the northern part of the Atlantic flyway until fall 2022, with sporadic detections thereafter (Fig. 2 and fig. S4).

Fig. 2. Continuous phylogeographic inference of the three major North American genotypes with the largest contributions to the transcontinental spread of H5N1 clade 2.3.4.4b.

(A) Maps display the estimated spread for genotypes A1, B2.1, B3.2, and B3.6. Mapped points represent nodes of the MCC tree for each genotype, and lines represent transmission events, colored by the inferred median time of occurrence. Transmission lines are curved to represent the direction of dispersal, counterclockwise from origin to destination. Shaded areas represent the 80% HPD of nodes’ estimated location. (B) Number of viruses sequenced for genotypes A1, B2.1, B3.2, and B3.6 over time. Videos of the phylogeographic reconstruction presented in (A) are available in movies S1 to S4.

Fig. 3. Continuous phylogeographic inference of two major North American H5N1 clade 2.3.4.4b genotypes.

Maps display the estimated spread for genotypes (A) B1.2 and (B) B1.3. Mapped points represent nodes of the MCC tree for each genotype, and lines represent transmission events, colored by the inferred median time of occurrence. Transmission lines are curved to represent the direction of dispersal, counterclockwise from origin to destination. Shaded areas represent the 80% HPD of nodes’ estimated location. (C) Number of viruses sequenced for genotypes B1.2 and B1.3 over time. Videos of the phylogeographic reconstruction presented in (A) and (B) are available in movies S5 and S6.

Table 1. Estimated TMRCA, source host, and location for major North American A(H5N1) clade 2.3.4.4b genotypes.

Genotype	Median TMRCA	TMRCA 95% HPD	Source MRCA	Location MRCA
A1	2021-08-04	2021-07-10 to 2021-09-15	Anseriformes	NL
B1.2	2022-01-22	2021-12-29 to 2022-02-07	Anseriformes	ON
B1.3	2022-06-24	2022-05-18 to 2022-07-23	Anseriformes	ON
B2.1	2022-01-01	2021-11-27 to 2022-02-01	Anseriformes	MN
B3.2	2022-01-06	2022-01-01 to 2022-01-28	Passeriformes	ND
B3.6	2022-08-03	2022-06-26 to 2022-08-03	Anseriformes	MN
B4.1	2021-12-29	2021-11-25 to 2022-02-11	Anseriformes	BC

The Eurasian lineage A1 virus exhibited limited detection outside the Atlantic and Mississippi flyways, suggesting that it was outcompeted by its reassorted progeny across North America (Fig. 2). Genotype B2.1 emerged from the Eurasian lineage in the Midwestern United States around early 2022, with a 95% HPD interval spanning 27 November 2021 to 1 February 2022 (Table 1, Fig. 2, and fig. S4). This genotype rapidly disseminated throughout the Prairie Pothole Region during the 2022 spring migration, encompassing Alberta, Saskatchewan, and Manitoba in Canada, as well as the Dakotas and Minnesota in the United States (Fig. 2 and fig. S3). Its detection declined rapidly by 1 July 2022 before the fall migratory period (Figs. 1B and 2B), thereby contributing minimally to the transcontinental spread of A(H5N1). However, genotype B2.1 gave rise to genotype B3.2, which became the most frequently detected genotype in the Americas during the study period (table S1 and Figs. 1 and 2). Phylogeographic analysis suggests that B3.2 emerged in North Dakota on 6 January 2022, with a 95% HPD interval from 14 December 2021 to 26 January 2022 (Table 1, Fig. 2, and fig. S4), aligning with previous estimates (31).

The emergence of genotype B3.2 was characterized by its similar geographic distribution to its ancestral genotype B2.1, with B3.2 largely displacing B2.1 following its emergence (Figs. 1B and 2, A and B). Notably, B3.2 became particularly prominent in the Prairie Pothole Region, a critical area where all four major migratory flyways in the Americas converge, leading to the transcontinental dissemination of this genotype (Fig. 2) (44–46). The fall migration season facilitated the latitudinal and longitudinal spread of B3.2 across the continent, marking its initial incursion into Central and South America by the fall of 2022 (Fig. 2). In the subsequent year, bird migrations during the fall of 2023 further extended the distribution of B3.2 to remote regions, including the Falkland Islands, South Georgia and South Sandwich Islands, and Antarctica (Fig. 2). Throughout the Americas, B3.2 was the most prevalent variant of the A(H5N1) clade 2.3.4.4b virus in terms of geographic distribution and collection frequency. However, by the spring of 2023 (Fig. 2B), B3.2 began to be replaced by one of its direct descendants, genotype B3.6 (Figs. 1 and 2).

The emergence of genotype B3.6 is estimated to have occurred in Minnesota around 3 August 2022, with a 95% HPD interval spanning 26 June to 31 August 2022 (Table 1, Fig. 2, and fig. S3). This genotype subsequently spread throughout the Prairie Pothole Region before the 2023 fall migration season, eventually supplanting its ancestral genotype B3.2 as the dominant circulating strain (Fig. 2 and fig. S4). During the fall migratory period, bird movements facilitated the latitudinal and longitudinal dissemination of B3.6 across North America, mirroring the initial spread pattern of its predecessor, genotype B3.2.

Outside of the Prairie Pothole Region, genotype B1.2 emerged in southern Ontario around 22 January 2022, with a 95% HPD interval from 29 December 2021 to 7 February 2022 (Table 1 and Fig. 3). This genotype primarily circulated within Ontario but was also detected in the southern United States during the winter of 2022 and in Manitoba during the spring of 2022 (Fig. 3A). Genotype B1.3, a reassortant descendant of B1.2, emerged around 24 June 2022, with a 95% HPD interval from 18 May to 23 July 2022. B1.3 spread across the same geographic region in Canada (Fig. 3B) and largely replaced its ancestral genotype, B1.2, which was rarely detected after the emergence of B1.3 (Figs. 1B and 3C).

Genotype B4.1 evolved from the wholly Eurasian virus (Fig. 1C). This genotype was prominently detected in Northwestern North America but remained confined to this region, contributing relatively little to the transcontinental spread of A(H5N1) (fig. S5).

Estimating fitness from phylogenies

Our phylogeographic analyses indicated that successive reassortment events led to the emergence of genotypes that outcompeted their immediate ancestors. To estimate the relative fitness of these genotypes, we used two independent methods: the model developed by Lefrancq et al. (47) and local branching index (LBI) (48). Both approaches supported our observation that descendant genotypes exhibited higher fitness than their direct ancestors. Specifically, the lineage replacements of A1 by B2.1, B2.1 by B3.2, and B3.2 by B3.6 were associated with successive increases in estimated fitness (Fig. 4). However, the fitness increase from genotype A1 to B2.1 was not consistently supported by both models, as the mean LBI of genotype B2.1 was lower than that of A1 (Fig. 4). Both models also support fitness increases coinciding with the replacement of genotype A1 by B1.2 in southern Ontario, followed by B1.3 replacing B1.2 (Fig. 4). These findings suggest that the evolutionary trajectory of these genotypes was driven by enhanced fitness, leading to the displacement of less fit ancestral strains.

Fig. 4. Estimates of relative fitness between A(H5N1) genotypes.

(A) Simplified maximum likelihood phylogenetic tree representing the evolutionary history of major A(H5N1) genotypes in the Americas. Estimation of relative fitness for each genotype was conducted by two methods, a multinomial logistic fitness model (B) and LBI (C). Colors of the tree branches in (A) and dotted lines between points in (B) and (C) represent genotype lineage ancestry. Error bars represent the 95% confidence interval of each data point.

Host dynamics

To investigate the potential impact of host diversity on differential dispersal patterns across genotypes, we used Bayesian discrete trait diffusion models with sequence and metadata collected as part of Canada’s Interagency Surveillance Program for Avian Influenza Viruses. These models allowed us to estimate the number of transitions between host states (Markov jumps) and the duration spent in a particular host state between transitions (Markov rewards) for the seven most prominent genotypes circulating in the Americas. It is well established that Anseriformes (ducks, geese, and swans) and Charadriiformes (gulls, terns, and shorebirds) serve as the primary host reservoirs and dispersal vectors of IAV globally (1–4). However, within North America, the role of Charadriiformes in long-distance IAV dispersal is less clear, with Anseriformes appearing to be primarily responsible (49, 50). Our findings largely align with this understanding. Anseriformes (1) was identified as the source host for all major genotypes, except B3.2 (2), was the most frequent source of Markov jumps (3), and exhibited the highest Markov rewards across all genotypes examined, excluding the wholly Eurasian A1 genotype (Table 1, Fig. 5, and fig. S6). Genotype B3.2 presented an exception, as its initial dispersal was estimated to have been driven by Passeriformes, specifically corvids (Table 1, Fig. 5, and fig. S5). This suggests that while Anseriformes plays a dominant role in IAV dispersal, other avian orders can also contribute to the spread of specific genotypes.

Fig. 5. Host dynamics for major H5N1 genotypes in North America.

The arrow width summarizes jump counts between host groups (Markov jumps with >0.70 posterior probability and a Bayes factor >3.0), where each arrow is colored according to the source host. The size of each node represents the median relative phylogenetic time spent in each host state (Markov rewards).

Most viral samples from wild birds in this study originated from Anseriformes and Charadriiformes, which are well-known reservoirs for IAVs. However, our findings indicate that Passeriformes, specifically corvids, can also contribute to the spread of IAVs. It is important to note that these sequences were primarily derived from collections of dead or sick birds, as Passeriformes was not a primary focus in most migratory bird sampling efforts. Therefore, the interpretation of these results should be approached with caution. Given that corvids were the only Passeriformes observed as hosts for A(H5N1) in this study, it is unlikely that Passeriformes as a group plays a substantial role in IAV dispersal. Nonetheless, our findings suggest that wild bird surveillance programs aimed at tracking the evolution and transmission of the virus should consider expanding their scope to include a broader range of passerine species. This would facilitate a more comprehensive understanding of their potential involvement in IAV dynamics and could provide valuable insights for future surveillance strategies.

Our analysis reveals fundamental differences in host dynamics among viral genotypes, which can be attributed to both geographic distribution and inherent host preferences. For instance, the wholly Eurasian A1 genotype exhibited elevated Markov rewards for Suliformes hosts, likely due to its prevalence in the Atlantic flyway, where Suliformes is more abundant compared to the Mississippi and central flyways (Fig. 2A). This suggests that the observed differences in host dynamics are partially confounded by variations in the geographic and temporal distribution of hosts, as many species with high viral prevalence are restricted to coastal habitats. However, there also appears to be intrinsic differences in host dynamics between genotypes. Notably, genotypes B1.2 and B1.3, which occupied the same geographic region, displayed distinct host dynamics on the basis of Markov rewards and jumps. Similarly, genotypes B2.1, B3.2, and B3.6, despite sharing similar geographic regions during their initial dispersal (Fig. 2), exhibited differing host dynamics. These findings indicate that while geographic factors contribute to host dynamics, there are also genotype-specific host preferences that influence the spread and transmission of these viruses.

DISCUSSION

Since the resurgence of H5 HPAIVs in 2005, there have been multiple intercontinental waves of dispersal, each facilitated by different HA clades or reassortant viruses (18). For instance, the 2005–2006 epizootic was driven by clade 2.2 A(H5N1), followed by clade 2.3.2.1c A(H5N1) during 2009–2010, clade 2.3.4.4c A(H5N8) in 2014–2015, and clade 2.3.4.4b A(H5N8) in 2016–2017 and 2020–2021 (51–55). However, A(H5N1) clade 2.3.4.4b viruses, which emerged as the cause of an epizootic starting in 2021, have almost completely replaced all other circulating H5 HPAIVs globally (18, 24).

Since 2005, H5 HPAIVs have progressively expanded from their geographic epicenter in Asia to Africa, Europe, and more recently, North and South America (18). The transatlantic dispersal of IAVs into North America is likely facilitated by wild bird migration patterns in the circumpolar Arctic, with Icelandic waters serving as a key mixing zone (56). This region connects the East Atlantic and North Atlantic flyways, where migratory birds from diverse locations congregate, promoting the movement of IAVs to Northeastern Canada during westward migratory periods (23, 56).

In addition to the transatlantic route, IAVs can also migrate from East Asia to North America via the trans-Beringian route (55, 57, 58), as evidenced by past incursions along the Pacific coasts of Canada and Alaska (26, 28). However, our phylodynamic analyses indicate that East Asia–derived viruses were minor contributors to the spread of clade 2.3.4.4b viruses across the Americas, as they accounted for less than 2% of all viruses sequenced in this study. Instead, most clade 2.3.4.4b genotypes in the Americas are descended from the 2021 incursion of A(H5N1) into Atlantic Canada, which have since disseminated throughout all major migratory flyways. This highlights the importance of regional migration patterns in the spread of these viruses.

Detections of clade 2.3.4.4b viruses peaked following migratory periods in regions with high waterfowl density, such as the Atlantic coast, St. Lawrence River, and the Prairie Pothole Region (59). Accordingly, 85.5% of Markov rewards—indicating the time spent in a particular host—were attributed to wild birds, surpassing global estimates for the same virus (51.7% Markov rewards) and highlighting the critical role of wild birds in HPAIV transmission in the Americas (18). In contrast to earlier global observations where H5 HPAIV clades primarily circulated in domestic birds with episodic spillover to wild birds (18), our results show that domestic birds played a much smaller role in the dispersal of A(H5N1) (Fig. 5). In addition, nonavian hosts played a role in the spread of A(H5N1), as all seven genotypes displayed Markov rewards in mammals, with evidence suggesting spillback from mammals to avian hosts for two genotypes (Fig. 5). These findings underscore the importance of genetic surveillance in pandemic preparedness, particularly given the recent spillover of A(H5N1) into nonavian livestock, such as goats and dairy cows in the United States and into humans in both Canada and the United States (60). The genotype responsible for the dairy cow outbreak in the United States (B3.13) was not detected in Canada during the study period.

Areas with high waterfowl density, such as Atlantic Canada and the Prairie Pothole Region, are well known for maintaining IAVs (61–63). Our phylogeographic reconstructions indicate that both regions served as origins for the long-distance latitudinal expansion of clade 2.3.4.4b A(H5N1), with the Prairie Pothole Region emerging as the primary driver of longitudinal expansion. The Prairie Pothole Region, an extensive area within the northern Great Plains, is characterized by millions of shallow pools formed by receding glaciers and is one of the most prominent waterfowl breeding grounds globally, supporting ~50% of North America’s waterfowl population (64). This region is recognized as an IAV hotspot (62, 63) and lies at the intersection of the four major North American migratory flyways (44–46). Our analyses suggest that the mixing of birds from different flyways in this region facilitated the dispersion of genotype B3.2 viruses both latitudinally and longitudinally during late 2022 and early 2023 (Fig. 2). The predominance of genotype B3.2 in this region before the fall migratory season likely contributed to its widespread dispersal and common detection across the Americas and Antarctica. The emergence and dispersal of genotype B3.6 mirrored those of its ancestor (B3.2) in the following year, which it replaced temporally and spatially (Fig. 2). Although B3.6 has not been extensively detected south of the United States, this may be due to delays in publishing genomic surveillance data. These findings highlight the importance of the Prairie Pothole Region as a critical area for future surveillance to predict IAV genotype frequencies in overwintering locations. In addition, they emphasize the region’s role in the emergence and selection of reassortant genotypes with increased fitness.

The segmented nature of IAV genomes enables reassortment when two or more IAVs infect the same cell, allowing for the creation of new viral strains through the exchange of genetic segments (34, 65). Although the IAV genome consists of eight segments, which theoretically allow for 254 possible reassortant combinations, not all of these combinations are equally probable. Nonetheless, laboratory experiments have demonstrated that when cells and animal models are coinfected with multiple IAVs, there is a high likelihood of producing reassortant viruses, with 60 to 95% of the resulting progeny viruses exhibiting reassortant genotypes (66–68). Our findings indicate that reassortment between Eurasian-origin A(H5N1) and American-origin LPAIVs has led to the emergence of 57 distinct genotypes in Canada, involving all genome segments except HA, NA, and M. This pattern of reassortment is consistent with previous outbreaks in Europe during 2014/2015 and 2016/2017, where the internal segments were exchanged with minimal impact on viral fitness (69). However, as we observe little or no amino acid variation among the surface proteins of major American genotypes (table S4), our results suggest that variations within internal proteins underlie the observed fitness differences between A(H5N1) genotypes (Fig. 4). This highlights the importance of reassortment in driving viral evolution and adaptation, particularly through changes in internal proteins that may influence viral fitness without altering surface antigens.

Phenotypic variations in immunogenicity, antigenicity, replication, transmission, and virulence collectively influence the epidemiological fitness of a viral variant (70, 71). This cumulative variation determines the reproductive success and capacity of a particular variant to predominate within a population (72). IAVs are often considered more fit when they have specific amino acid substitutions, typically in one or a few genes, which have been laboratory confirmed to have a phenotypic effect (73). However, identifying and experimentally validating the fitness effects of mutations can be challenging, especially given the rapid evolution of IAVs and the presence of epistatic interactions. This approach may overlook novel contributors to fitness, particularly when genetic variation is concentrated outside the well-studied surface proteins (table S4). To address this, we used methods that use phylogenetics and epidemiological data to summarize changes in population composition over time, aligning with the classic Malthusian definition of fitness (47, 48). Our study revealed that most reassortant viruses arose from earlier reassortments rather than wholly Eurasian viruses, including four of the six most prominent reassortant genotypes in the Americas (Fig. 2C). We found that successive reassortment events were associated with increased viral fitness (Fig. 4), where derived reassortants were more fit than their direct ancestors. This cumulative reassortment led to the emergence of a genotype that spread across the Americas and resulted in the first detection of HPAIV in Antarctica, which was later supplanted by a fitter direct descendant.

Our phylogeographic analyses demonstrate the spatiotemporal replacement of A(H5N1) genotypes by their direct descendants, suggesting that a fitness advantage enabled the derived viruses to outcompete and supplant their ancestors. While we support this inference using phylogenetic-based fitness methods, these approaches assume constant selection pressure across the population (47, 48). This may not be the case in a multihost system, as the unique life history patterns of the hosts infected with A(H5N1) may impose varying selection pressures across the host range. Therefore, estimating the relative fitness of A(H5N1) genotypes with differing evolutionary histories is challenging with current data. However, given that parental and reassortant offspring genotypes co-occupied geographic, temporal, and ecological niches, it is a reasonable assumption that they were subject to similar selection pressures and stochastic events that may influence spread, and the relative fitness between these two viruses can be reliably estimated.

The data presented herein demonstrate the rapid and extensive diversification of A(H5N1) clade 2.3.4.4b viruses in the Americas, leading to successive lineage replacements that are consistent with compounding fitness increases. This emphasizes that the continued surveillance and analysis of the ongoing diversification, transmission, and dispersal of HPAIV are crucial for mitigating future impacts on ecological, animal, and human health.

MATERIALS AND METHODS

Sample collection and HPAIV screening

Wildlife samples were collected between November 2021 and September 2024 and stored for testing by multiple groups, government agencies, and laboratories across Canada as described by Giacinti et al. (30). Samples from domestic birds were collected similarly by the Canadian Food Inspection Agency following the notification of infected premises. Diagnostic testing of samples for HPAIV was performed by laboratories of the Canadian Animal Health Surveillance Network (CAHSN), National Centre for Foreign Animal Disease, Canadian Food Inspection Agency (NCFAD-CFIA), and/or provincial diagnostic labs as described previously (30). Samples positive for HPAIV were subject to sequence library preparation with the Oxford Nanopore Rapid Barcoding Kit (SQK-RBK110.96 or SQK-RBK114.96), sequenced with MinION R9.4.1 or R10.4.1 Flow Cells (FLO-MIN106D or FLO-MIN114) on an Oxford Nanopore GridION sequencer (Oxford Nanopore Technologies), demultiplexed with Guppy (version 6.3.9 to version 7.1.4) or Dorado (version 7.2.1 to version 7.4.14), and processed using nf-flu (version 3.3.5 to version 3.5.3; https://zenodo.org/records/14027084) (30). All A(H5N1) sequences generated in this study have been deposited to the Global Initiative on Sharing All Influenza Data (GISAID) under EPI_SET ID: EPI_SET_250502ga and to NCBI’s GenBank under accession numbers PV301057 to PV324692.

Reassortant genotype classification

The 2955 A(H5N1) genomes produced in this study were assigned genotypes using GenoFLU software version 1.06 (https://github.com/USDA-VS/GenoFLU). Individual viral segments (PB2, PB1, PA, HA, NP, NA, M, and NS) from the 144 genomes that could not be genotyped by GenoFLU were trimmed of regions flanking the open reading frames and aligned with GenoFLU reference sequences using MAFFT version 7.49. The resulting eight alignments were used to estimate maximum likelihood phylogenetic trees with IQ-TREE version 2.2.0 (74, 75) under the best fitting model of nucleotide substitution as determined by ModelFinder (76). Node support for the resulting tree was assessed by 5000 ultrafast bootstrap replicates (77). Bootstrap consensus trees were rooted on the best fitting root using augur-refine from the Augur bioinformatics toolkit (78). Each segment tree was divided into well-supported monophyletic subgroups on the basis of bootstrap support (>80) and pairwise tree distance, as described previously for North American A(H5N1) reassortants by Youk et al. (31). Each unique viral genome constellation was assigned a genotype name based on the name of the first isolate detected with that constellation.

Genotype evolution

Nonreassorted segments (HA, NA, and M) from the 2955 viruses were concatenated, aligned, and used to estimate a timescaled phylogenetic tree using IQ-TREE version 2.2.0, as described above. The emergence of each genotype was estimated by a discrete character reconstruction of the genotype labels (e.g., “B3.2”) of each virus across all nodes in the tree using the Augur bioinformatics toolkit (78). The bootstrap consensus tree was rooted using augur-refine, and the genotype was reconstructed across the refined tree using augur-traits. Genotype transitions based on the refined tree were inferred by StrainHub (79) and visualized with the visNetwork package in R.

Phylogeographic reconstruction

Segments from each of the 2955 viral genomes generated in this study were concatenated in the order of PB2, PB1, PA, HA, NP, NA, M, and NS to create a whole-genome sequence. These data were combined with all A1, B1.2, B1.3, B2.1, B3.2, B3.6, and B4.1 A(H5N1) genomes collected in the Americas and Antarctica between November 2021 and September 2024 available on GISAID and were concatenated similarly. Viruses were grouped by genotype, separately aligned, and used to estimate maximum likelihood phylogenetic trees, as described above (see the “Reassortant genotype classification” section). Identical sequences were collapsed before tree inference. Each phylogenetic tree was assessed for temporal signal with a linear regression of root-to-tip divergence as a function of sampling time that was estimated in TempEst version 1.5.3 using the best fitting root based on heuristic mean-squared residuals (80). Samples with a residual greater than 3 standard deviations from the residual mean were removed. To ease computational load, genotypes composed of more than 600 viruses were subset on the basis of time and location, wherein viruses were clustered into 50-km radii for each epidemiological month, and the three most genetically diverse samples from each cluster were chosen by the phyloprunr function in the R package PDcalc. Timescaled phylogenetic trees were estimated using BEAST version 1.10.4 under the best fitting model of nucleotide substitution as determined by ModelFinder (76), a relaxed molecular clock with lognormal distribution, and a Bayesian coalescent SkyGrid tree prior (81). The time at last transition for each tree estimation was set to the root height estimated by TempEst with one grid point per 2 weeks. Sample collection locations (as latitudinal and longitudinal coordinates) were added as a continuous trait partition to the tree estimation. For samples without collection coordinates, the centroid of the smallest administrative region in the collection location metadata was used. Samples from the United States without state-level location were excluded. The ancestral location of each internal node was reconstructed jointly with tree inference using the Cauchy relaxed random walk continuous trait model (82) with a random jitter (window size, 0.0001) added to samples from identical locations. At least two independent Markov chain Monte Carlo (MCMC) chains (400,000,000 steps, sampled every 40,000) were run for each genotype tree. Each MCMC chain was assessed for convergence (effective sample size >200 as assessed in Tracer), 10 to 20% burn-in discarded, and combined to produce a maximum clade credibility (MCC) tree using TreeAnnotator (83). Visualization and annotation of the spatiotemporal viral spread were done with R package Seraphim (84).

Host switching

Host species were binned into eight groups: Anseriformes, Charadriiformes, domestic bird, mammal, Passeriformes, Suliformes, raptors, or other (table S3) (30). Whole-genome sequences from each major Canadian genotype (A1, B1.2, B3.2, B3.6, B2.1, B4.1, and B1.3) that were collected as part of Canada’s Interagency Surveillance Program for Avian Influenza Viruses were used to estimate a timescaled phylogenetic tree using BEAST version 1.10.4 (83) under the best fitting model of nucleotide substitution as determined by ModelFinder (76), a relaxed molecular clock with lognormal distribution, and an exponential growth coalescent tree prior. Host transitions were estimated by adding the host group of each tree tip as a discrete character trait. Host transitions were reconstructed for all internal nodes of the tree by the asymmetric substitution model and social networks inferred by Bayesian Stochastic Search Variable Selection. BEAST xml files were manually edited to log Markov rewards and the complete Markov jump history for each host group (see data S1 for an example). Each MCMC chain was assessed for convergence following burn-in, combined, and used to produce an MCC tree, as described above. All host transition events and occurrence dates were extracted from the MCC tree. The posterior distribution of indicator values from the Bayesian Stochastic Search Variable Selection procedure was used to conduct Bayes factor tests to quantify statistical support for host transitions using SpreaD3 (85). Host transitions with Bayes factor <3.0 were excluded from the dataset. Transitions where the host probability at either the parent or child node in the MCC tree was below 0.70 were also excluded.

Estimation of relative fitness

The relative fitness dynamics of major North American A(H5N1) genotypes were estimated using the model developed by Lefrancq et al. (47). A maximum likelihood phylogenetic tree was constructed using whole-genome sequences from genotypes (A1, B1.2, B3.2, B3.6, B2.1, B4.1, and B1.3) collected in the Americas between November 2021 and September 2024 (see the “Phylogeographic reconstruction” section). The sequences were aligned as above, and a phylogenetic tree was constructed with IQ-TREE version 2.2.0 (74, 75). Each segment was designated as a separate partition and allowed its own best fitting model of nucleotide substitution (as determined by ModelFinder) (76). Node support for the resulting tree was assessed by 5000 bootstrap replicates. The bootstrap consensus tree was time calibrated and rooted on the branch that minimized the squared deviation of the root-to-tip regression using augur-refine from the Augur bioinformatics toolkit (78). The timescaled tree was used as input for automatic lineage detection and relative fitness estimation models (47) with genome_length = 13,136, mutation_rate = 4.94 × 10⁻³ (calculated by augur-refine, above), timescale = 0.15, and wind = 15. Automatic lineage detection partitioned the tree into the same seven previously defined genotypes. The multinomial logistic fitness model was fit with min_year = 2021.5 and window = 30/365 to quantify the fitness of each lineage relative to the ancestral, wholly Eurasian genotype (A1).

Concatenated sequences of nonreassorted segments (HA, NA, and M) for same viruses described above were aligned and used to build a maximum likelihood phylogenetic tree with IQ-TREE version 2.2.0 (74, 75) with the best fitting model of nucleotide substitution (as determined by ModelFinder) (76), as described above. Augur-traits was used to reconstruct the genotype across all nodes of the refined tree using default settings. Augur-lbi was used to calculate the LBI at all nodes of the refined tree with tau set to 0.0625 times the average pairwise distance, as recommended by Neher et al. (48). Estimated LBI values were extracted from all nodes of the tree and separated according to the reconstructed genotype for each node. The mean LBI and 95% confidence interval were calculated for each genotype. Model details for all phylogenetic inferences are presented in table S3.

Supplementary Materials

The PDF file includes:

List of members for the British Columbia Wildlife AIV Surveillance Program

Figs. S1 to S5

Tables S1 to S4

Legends for movies S1 to S7

Legend for data S1

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S7

Data S1