On 15 October 2024, we captured eight wild Mandarin ducks along the Cheongmicheon in Gyeonggi-do Province, South Korea (GPS coordinate: 37°8′31.25″N, 127°22′52.23″E) as a part of the national wild bird surveillance program in South Korea (Supplementary Figure 1). On 17 October 2024, a dead Northern pintail was found at the Yongsu reservoir on Jeju Island (GPS coordinate: 33°30′18.28″N, 126°53′33.4″E). We collected oropharyngeal and cloacal swabs from the birds. Swab samples were placed in phosphate-buffered saline (PBS) containing 400 mg/ml of gentamicin and thoroughly homogenized by vortexing for 1 min. The supernatant of samples was filtered using a 0.45-μm Minisart Syringe Filter (Sartorius, Göttingen, Germany) after centrifugation of the sample at 3,000 rpm for 10 min and inoculated into 10-day-old specific-pathogen-free (SPF) embryonated chicken eggs. After 72 h of incubation at 37°C, the allantoic fluids were harvested and tested for hemagglutination activity (HA) using 10% chicken red blood cells. RNA was extracted from the hemagglutination-activity-positive allantoic fluid using the Maxwell RSC simply RNA Tissue Kit (Promega, Madison, WI, USA) according to the manufacturer's instructions and screened for the matrix (M) and H5 genes of the avian influenza virus using real-time reverse transcription-PCR (rRT-PCR) as previously described (12).

Complementary DNA was generated using the SuperScript III First-Strand Synthesis system (Invitrogen, Carlsbad, CA, USA), and the eight gene segments were amplified using AccuPrime Pfx DNA Polymerase (Invitrogen, Carlsbad, CA, USA) as previously described (13). DNA libraries were prepared using Nextera DNA Flex Library Prep Kit (Illumina, San Diego, CA, USA), which utilizes transposon-mediated tagmentation and adapter ligation, with dual-index barcodes according to the manufacturer's instructions. The complete genome was sequenced using the paired-end 150 Illumina MiSeq sequencing-by-synthesis platform. NGS raw reads were trimmed of adapters and low-quality bases using BBDuk version 38.84 by setting the minimum quality to 30 (14). Trimmed reads were assembled de novo using the SPAdes assembler 3.15.5. Trimmed reads were mapped to the top result from the GISAID EpiFlu database, identified from contigs, using Minimap 2.24 (https://github.com/lh3/minimap2) with default options and visualized on Geneious Prime software. The assembled genome sequences produced by reference-guided genome assembly were used to generate the final consensus genome sequences. The dataset presented in this study can be found in online repositories. The names of repositories and accession IDs are available through the GISAID(https://www.gisaid.org) EpiFlu database (accession ID: EPI_ISL_19528860 and EPI_ISL_19531393). H5 clade classification was performed using an online subspecies classification tool available in the BV-BRC (https://www.bvbrc.org/app/SubspeciesClassification). The consensus genome sequences were examined to identify molecular markers associated with mammalian host adaptation, pathogenicity, and drug resistance. We utilized the FluSurver mutation tool from the GISAID Initiative (15) and performed manual screening based on known markers impacting AIV biological properties (16). Identified amino acid substitutions in the HA segment are referenced according to H5 numbering.

All eight consensus genome sequences were analyzed through the BLAST query function of the GISAID database (https://gisaid.org/). From the top 500 BLAST hits, identical sequences were filtered out using ElimDupes software (https://www.hiv.lanl.gov/content/sequence/elimdupesv2/elimdupes.html). Genome sequences were aligned using MAFFT software (17). Phylogenetic tree construction for each gene was conducted with RAxML v8.0 (18) using the general time reversible model for nucleotide substitution and the Gamma model for rate heterogeneity, with 1,000 bootstrap replicates. Interactive Tree of Life (iTOL) was used to visualize the tree of each gene (19). A cluster was regarded as distinct only when it had a bootstrap support value > 70 and a nucleotide sequence identity > 97%. The genotype G2b and G2d clade 2.3.4.4b H5N1 viruses identified from 2021 to 2022 (20, 21) were used to verify the genotypes of viruses. The A/goose/Hunan/SE284/2022(SE284) (H5N1) (2) was used to categorize genotype G2c viruses.

A Bayesian relaxed-clock phylogeny of the HA gene was reconstructed using BEAST version 1.10.4 (22), applying the Hasegawa, Kishino, and Yano substitution model with an uncorrelated log-normal distribution and a Gaussian Markov Random Field (GMRF) Bayesian skyride coalescent prior (23). The Markov Chain Monte Carlo (MCMC) process was run in parallel across three chains, each with 50 million iterations, and the results were combined after a 10% burn-in. All parameters achieved effective sample sizes >200 and were examined using TRACER v1.5 (http://tree.bio.ed.ac.uk/software/tracer/) (24). A maximum clade credibility (MCC) tree was created with TreeAnnotator and visualized using FigTree v1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/). The time to the most recent common ancestor (tMRCA) was estimated based on the height values at common ancestor nodes.

A live Mandarin duck out of eight captured on 15 October 2024, and a Northern pintail found dead on 17 October 2024 tested positive for influenza A virus via chicken embryo inoculation and rRT-PCR. We successfully isolated and sequenced the H5N1 HPAI viruses, designated as A/Mandarin duck/Korea/24WS005-2/H5N1/2024 (hereafter MD/24WS005-2) and A/Northern pintail/Korea/24WC025/H5N1/2024 (hereafter NP/24WC025). A total of 52,864 and 136,268 NGS reads were generated, respectively, resulting in complete coding genome sequences (CDSs) across all eight influenza virus segments and high average sequencing depth for each segment (>350). The viruses were identified as HPAI based on the presence of multiple basic amino acids at the HA proteolytic cleavage site (PLREKRRKR/G) (25) and classified as an H5 subtype clade 2.3.4.4b.

The NP/24WC025 virus and MD/24WS005-2 virus had different genome constellations, suggesting each virus had independently evolved and been introduced into South Korea (Figure 1). The genotype of the NP/24WC025 virus was identical to that of HPAI viruses circulating in Japan during 2023–2024. The HA gene belonged to the G2d sub-lineage (20, 21). The HA gene of NP/24WC025 virus shared a common ancestry with the clade 2.3.4.4b H5N1 HPAI viruses, A/white-tailed eagle/Hokkaido/2024, A/chicken/Hokkaido/E012/2024, which were concurrently identified in Japan. Their tMRCA was estimated to be 28 April 2024 (95% BCI: 10 January 2024–29 July 2024), suggesting that these H5N1 viruses are descendants of the G2d sub-lineage H5N1 viruses that circulated in Japan during early to mid-2024.

For the MD/24WS005-2 virus, the HA, NA, and M genes clustered with the H5Nx clade 2.3.4.4b HPAI viruses that mainly circulated in wild birds during 2022–2024 in Asia but did not form a well-supported monophyletic cluster with other viruses (Supplementary Figures 1D, F, G). The HA gene belonged to the G2c sub-lineage (2). In the Bayesian phylogenetic analysis of the HA gene, the MD/24WS005-2 virus clustered with clade 2.3.4.4b H5N6 HPAI viruses from wild birds in South Korea and Japan in the winter of 2023 but this grouping is not supported by posterior probability. All internal gene segments, except the M, clustered with LPAI viruses identified in the East-Asian Australasian migratory bird flyway, indicating the MD/24WS005-2 virus evolved via genome reassortment between the HA, NA, and M gene of clade 2.3.4.4b HPAI and the PB2, PB1, PA, NP, and NS of Eurasian LPAI viruses (Supplementary Figures 1A–C, E, H). Notably, the PB1, HA, NP, NA, and M genes shared a common ancestor with the A/eagle/Korea/22WC464/2023(H5N1) virus (Supplementary Figures 1B, D–G), which belonged to a minor genotype, the Kor22-23P, of H5N1 HPAI viruses that circulated in South Korea during the winter season of 2022–2023 (9). The PB2, PA, and NS genes are derived from the Eurasian LPAI gene pool circulating in wild bird populations (Supplementary Figures 1A, C, H). The high genetic diversity of avian influenza viruses in wild birds has contributed to the generation of multiple genotypes of clade 2.3.4.4b HPAI viruses as a donor gene pool of different genetic lineages (26). The long branch length and new genes derived from the LPAI gene pool suggest that it had circulated undetected for ~2 years and had undergone multiple reassortments with prevailing LPAI viruses in wild bird populations.

Genetic mutations associated with increased binding affinity to α-2,6 sialic acid receptors were found in the HA protein, including the N110S in MD/24WS005-2 virus, S154N in NP/24WC025 virus, and S133A and T156A in both viruses (Table 1). Both viruses carried the V588T mutation in the PB2 protein, known to enhance pathogenicity in mice. The K482R, associated with increased polymerase activity in mammalian cell lines, was identified in the NP/24WC025 virus. In the PB1 gene, mutations D3V and D622G, which are associated with enhanced polymerase activity and viral replication in avian and mammalian cell lines and increased virulence in mice, were present in both viruses. Mutations in the M protein (N30D, I43M, T215A, and P42S), known to increase pathogenicity in murine models, were identified in both viruses, along with the presence of the ESEV motif in the C-terminal of NS1 protein.