Reply to: Evidence of widespread sequencing errors in global avian influenza study

replying to M. Wille et al. Nature Communications https://doi.org/10.1038/s41467-026-78539-0 (2026)

The primary issue arose due to an error during the sequence submission process to GISAID. Following GISAID’s advice for bulk submissions, we used the command-line software “FluCLI” to upload our data. During reformatting of files for submission, an analysis file originally prepared for phylogenetic tree construction and containing both our study sequences and reference sequences was inadvertently uploaded. In addition, sample identifiers were reformatted incorrectly, which led to duplication and mixing of study and reference sequences. Importantly, in our original analyses, we did not find such anomalies. We would like to apologize for the confusion caused by the erroneous sequences. The sequences have been corrected both in GISAID and in GenBank, under the same accession numbers provided in the original manuscript (GISAID: EPI_ISL_19613337 to EPI_ISL_19613573/ EPI_ISL_19812857-EPI_ISL_19812874; GenBank: PV422873-PV423054, PV422241-PV422430, and PV440337-PV440372, now version 2). Additional data from this work are available from the following link: https://doi.org/10.7910/DVN/KJDNGF (In accordance with each country’s data protection laws, these data are provided strictly upon a valid request according to collaborative agreements. They must not be used to model, predict, or otherwise infer health or commercial applications. The data are not permitted for use in generative AI modeling, dashboard development, or border-related health prediction systems.
Sequence uploading error led to the appearance of implausible 100% repeated nucleotide identities and patterns of apparent evolutionary stasis that were observed by the correspondents. According to Chen and Holmes (2006) and as highlighted by the authors, substitution rates of AIV varied between 1.8 and 8.4?×?10?3 per site per year. For HA it was 3.92?×?10?3 per site per year (95% highest probability density range: 2.43–5.40?×?10?3). For the NA, the mean substitution rate was 3.61?×?10?3 per site per year (95% highest probability density range: 2.29–4.92?×?10?3). Based on the values by Chen and Holmes (2006)1, a rough estimate for the mutation rate in the HA can be 6.7 (3.5–9.1) per year, while the mutation rate for NA may be calculated as 5.1 (3.2–7) mutations per year. Kim et al. (2022)2 evaluated the effects of host on the AIV mutation rates and found that mutation rates were higher for viruses infecting wild shorebirds and chickens compared to wild and domestic ducks. Even though the host of the collected fecal samples or the exact date of dropping data is unavailable, our sequences are consistent with the expected rate of influenza virus evolution (Supplementary Fig. 1 and Supplementary Table 1-2).
We also conducted a TempEst analysis using the correct sequence data. We downloaded the available full sequences of NA from Asia collected between 1996/1/1 and 2024/12/31 and uploaded to GISAID. In total, 817 sequences were downloaded. The phylogeny was inferred using the Maximum Likelihood method and the General Time Reversible model of nucleotide substitutions. We found that the tree constructed with the correct sequences didn’t have any outliers belonging to our study sequences (Supplementary Fig. 1).

Thus, the conclusions in our published manuscript were based on correctly generated data, while the erroneous sequences in GISAID were the result of a submission mistake. We sincerely apologize for this error and for the confusion it has caused. We have worked with GISAID to withdraw the affected entries and replace them with the corrected sequence data. We are currently reviewing and strengthening our internal submission and data-handling processes to ensure that such errors do not occur again in the future. We sincerely thank the authors for their detailed work in identifying this issue.
We wish to clearly emphasize that our study does not, in any way, question Australia’s status as free from avian influenza. We never mentioned this in our study or discussed it. We fully agree with the correspondents that Australia has a strong surveillance and infection-control system and a well-documented history of rapid detection and eradication of avian influenza outbreaks. We are grateful for the correspondents’ clarification on this point and sincerely thank them for highlighting the importance of maintaining accuracy in how such findings are interpreted. The phylogenetic tree constructed based on correct sequence data clearly shows that our sequences are not 100% identical to the two sequences reported in Australia in 2020 (Supplementary Fig. 2v).

At the same time, it is important to recognize that influenza viruses, including H7 lineages, can share genetic similarities across regions, and that related or convergent genetic constellations may appear in other parts of the world due to under-sampled reservoirs. Influenza is a global virus with shared variants circulating across multiple regions, and the limited surveillance coverage in many areas means that similar or related lineages may remain undetected outside well-monitored countries such as Australia. The COVID-19 pandemic highlighted the risks of attributing viral “origins” to a single country in the absence of comprehensive global data. Such attributions are not scientifically robust and may unfairly stigmatize specific countries. Our study makes no claims that Australia is a source of infection, nor do our findings provide any evidence that would call Australia’s current avian influenza-free status into question (Our study does not even mention Australia). Furthermore, we emphasize that our research is in no way sufficient to support any theory about the origin of influenza variants, nor to provide information that would identify any specific region or country as the primary source of such variants. This limitation is clearly mentioned in our study.