[BMJ, 14 March 2023]

Global reports of highly pathogenic avian influenza A(H5N1) in birds are increasing, with cases reported from every region except Australasia and Antarctica since 2020.1 The global spread of these avian influenza outbreaks is unprecedented, exacting large economic losses to poultry industries and tourism, and posing a substantial threat to global health security and animal ecology.

In Europe, 2520 H5N1 outbreaks were reported in poultry between October 2021 and September 2022, and the virus was also detected in 3867 dead wild birds.2 The US reported 131 mammalian H5N1 infections among bears, foxes, raccoons, skunks, and seals between May 2022 and February 2023.3 In October 2022, an H5N1 outbreak among Spanish farmed minks was reported for the first time,4 triggering concerns that the virus might soon become transmissible between humans (mink are physiologically similar to ferrets, the animal model used to study transmissibility of influenza viruses among humans).5

On 24 February 2023, an 11 year old girl died from an H5N1 avian flu infection in Cambodia and…

[bioRxiv, 23 May 2025]

Reconstructing the states of ancestral organisms has long been central to our understanding of the evolution of a wide range of traits. Ancestral state inference tools that account for trait-dependent properties are limited, because of challenges associated with inferring past states in a manner consistent with a phylogenetic tree (and its uncertainty) and with a stochastic process describing how states change over time. In phylogeography, ancestral state inference is used to reconstruct the past locations of viruses, bacteria or other rapidly-evolving organisms, characterizing, for example, how often and when a virus moved among locations, or from one host species to another. However, such reconstructions are sensitive to differences in sampling among different locations or host species, and this can bias the reconstruction of the location of ancestors towards the more widely sampled region/species. Here, we introduce a new method, Sampling Aware Ancestral State Inference (SAASI), which builds on recent advances in state-dependent diversification models and reconstructs ancestral states, and in particular for phylogeographic applications, accounting for sampling differences. Indeed, we find that accounting for sampling changes the inferred historical location of viral lineages and the times of key viral movements. We use simulations to show that with known sampling differences, SAASI infers past viral locations considerably more accurately than standard methods. We apply our method to the spread of the H5N1 virus in the United States in 2024, and explore how robust phylogeographic reconstruction is to differences in sampling and epidemiological rates between wild bird populations, cattle, humans and other species. We find that the key transmission event from wild birds to cattle is estimated to occur later under lower sampling in wild birds (compared to other species) than when sampling is not accounted for. SAASI is rapid and readily scales to trees with 100,000 tips, making it feasible for modern phylogeographic applications.