SARS-CoV-2 Delta variant re-emerges in US farmed mink and free-ranging white-tailed deer in 2022–2023

Abstract

American mink (Neogale vison) are susceptible to SARS-CoV-2, but little is known about virus circulation in mink since 2021. Here, in the first active surveillance study of SARS-CoV-2 in apparently healthy farmed mink in the United States, we find a ~0.9% (760/85,656) RT-PCR positivity rate among nasal swabs collected in 18 farms across six states during 2022–2023. Phylogenetic analysis of 293 viral genome sequences shows that human-adapted SARS-CoV-2 variants (e.g., Omicron) repeatedly spill over into mink. Surprisingly, the detection of a Delta lineage virus (AY.39) on a mink farm four months after its last detection in humans within the same state suggests prolonged unsampled transmission. The spread of mink-adapted AY.39 viruses from a mink farm to neighboring free-ranging white-tailed deer represents a rare instance of SARS-CoV-2 transmission between livestock and wildlife. These findings demonstrate the value of active surveillance for identifying subclinical infections and interspecies transmission between humans, mink, and wildlife.

American mink are known to be susceptible to SARS-CoV-2, yet the extent of virus circulation in these animals remains unclear. Here, the authors conduct an active surveillance of asymptomatic US farmed mink, revealing mink-adapted variants and interspecies transmission, underscoring the importance of monitoring for subclinical infections and cross-species spread.

Introduction

SARS-CoV-2, the virus that causes COVID-19 in humans, repeatedly caused outbreaks on mink farms in 2020 (Fig. 1, Supplementary Table 1). Once introduced by humans, the virus spreads efficiently between mink through direct and indirect contact. Since the start of the pandemic, 168 SARS-CoV-2 outbreaks in farmed mink across 13 countries in Europe and North America have been reported to the Program for Monitoring Emerging Diseases (ProMED) and/or the World Animal Health Information System (WAHIS) (Supplementary Table 1). Based on the scientific literature, Europe alone experienced at least 400 SARS-CoV-2 outbreaks in mink farms from the start of the pandemic to January 2021¹. In the United States, 16 SARS-CoV-2 outbreaks were reported in 2020 that affected some of the country’s largest mink-producing states: Michigan², Utah³, Oregon, and Wisconsin (Supplementary Table 1). The outbreaks were associated with high mortality in mink, with farms in Utah losing up to 50% of their adult mink³.

Fig. 1. Number of reported SARS-CoV-2 outbreaks in mink by country, 2020–2024.

Countries with reported mink outbreaks are shaded, proportional to the number of outbreaks (yellow = low, orange = high). Countries shaded grey have not officially reported mink outbreaks as of August 6, 2024. As of September 20, 2024, 18 SARS-CoV-2 outbreaks have been reported in US mink. Only outbreaks reported to the Program for Monitoring Emerging Diseases (ProMED) and/or the World Animal Health Information System (WAHIS) obtained through the SARS-ANI Project⁵⁶ (https://github.com/amel-github/sars-ani). Source data are provided in a Source Data file.

The first outbreaks of SARS-CoV-2 in farmed mink were documented in Europe in 2020⁴. Millions of mink were culled after public health officials became alarmed by novel mutations in the spike protein that emerged in farmed mink and transmitted from mink to people⁵. Soon after, however, it became evident that humans, rather than non-human hosts, were the major drivers of the evolution of SARS-CoV-2 immune escape variants. In December 2020, the Alpha VOC (PANGO sublineage B.1.1.7) was detected in humans in the United Kingdom and contained numerous mutations in the spike protein⁶. Immune-evasion mutations in the Delta variant led to widespread breakthrough infections in vaccinated and previously infected individuals in the US during the summer and autumn of 2021⁷. The Omicron variant arrived in the US at the end of 2021⁸, and various Omicron sub-lineages continued to cycle in humans during 2021–2025. At first, it was suspected that the highly mutated Omicron variant might have originated from an unsampled non-human host, where the virus could have evolved for many months undetected. North American free-ranging white-tailed deer (WTD)^9–12 appear to be capable of sustaining SARS-CoV-2 transmission for more than a year, including VOCs, which appear to evolve faster in WTD compared to humans¹³. At least one cervid-to-human transmission event has been documented in North America¹⁴, demonstrating that novel mutations acquired in cervids can potentially spill back to humans. However, the paucity of spike mutations in WTD suggests that selective pressure for antigenic change is low¹³. Currently, the leading hypothesis is that the Omicron variant’s immune-selected mutations in the spike protein did not evolve in a non-human host but rather in an immunocompromised human who was chronically infected and received serial courses of monoclonal antibody therapy^8,15.

There are lingering questions as to whether SARS-CoV-2 can sustain long-term transmission in rodents, mustelids, cervids, or other wildlife. SARS-CoV-2 has been identified in a broad range of species (e.g., felines, canines, primates)⁹, but these infections tend to be “dead end,” i.e., without subsequent transmission¹⁶. VOCs were reported in mink in European countries where routine weekly active surveillance on farms was performed, but there is no clear evidence for long-term transmission¹. SARS-CoV-2 infections were identified on one farm each in the US in 2021 and 2022, but diagnosis was based only on serology, and the virus lineages were unknown¹⁷. Consequently, no VOCs have been identified on US mink farms, but SARS-CoV-2 surveillance on US mink farms is generally performed only in response to clinical outbreaks. It remains unclear whether subclinical infection occurs on mink farms when the virus actively circulates in humans.

Here, in this work, we show through a large active surveillance study (>85,000 nasal swabs) that SARS-CoV-2 viruses of human origin can circulate sub-clinically in US farmed mink, acquire host-specific adaptations, and host-switch to wildlife. We find that mink remain highly susceptible to the Omicron lineage and that further adaptation of the SARS-CoV-2 virus to humans does not appear to introduce new barriers to spillover to mink. However, the largest mink outbreak in our study involved a variant belonging to the Delta lineage that was no longer circulating in humans at the time, indicating that mink can serve as reservoirs for extinct viruses, at least for a short term.

Results

SARS-CoV-2 detected on three US farms with otherwise healthy mink

During July 2022–January 2023, a total of 750 mink (250 per farm) were longitudinally tested for SARS-CoV-2 infection by rRT-PCR on three US farms: Farms N and O in State 1 and Farm P in State 2 (Supplementary Fig. 2). Of the samples collected from these three farms, 5.9% (760/12,926) were confirmed rRT-PCR positive for SARS-CoV-2 (Fig. 2). The frequency of virus detection varied by farm: 12.1% on Farm P (517/4296), 4.8% on Farm N (238/4945), and 0.1% on Farm O (5/3685). Mink on Farm P were positive for SARS-CoV-2 on the first date of sampling (August 8, 2022), whereas mink on Farms N and O were negative at the beginning of the study. The SARS-CoV-2 outbreak on Farm N lasted from August 8–October 12, 2022, with a peak on September 13 when 26.9% (67/249 mink) of mink were positive (Fig. 2a). The outbreak on Farm P appeared to consist of two waves. The first wave during August 8–September 20, 2022, which peaked on August 31, 2022, with 49.0% (121/247) positive. A second smaller wave occurred during October 10–November 29, 2022 that peaked on November 7, 2022, with 12.9% (31/241) positive. The second wave was only observed on one of the two barns (Barn P2) on Farm P (Fig. 2f). All animals on all three farms were negative on the final sampling date of the study, indicating that any outbreaks had cleared by the end of 2022 (Fig. 2). For additional confirmation that SARS-CoV was no longer present in Farm P, 50 nasal swab samples collected in mink in April 2024 and 250 swabs from July 2024 all tested negative for SARS-CoV-2.

Fig. 2. SARS-CoV-2 rRT-PCR test results in three US mink farms.

a Percentage of PCR-positive nasal swab samples over time, stratified by farm. Labeled points indicate peak positivity dates and values for Farm N (State 1), Farm O (State 1), and Farm P (State 2). Sample counts by rRT-PCR test result for b Farm N and c Farm O (State 1), and d Farm P and its barn-level breakdown: e Barn P1 and f Barn P2 (State 2). Bars are stacked by test result: confirmed positive (magenta), presumptive positive – i.e., detecting only the N1/N2 regions of the SARS-CoV-2 N gene (light pink), and negative (gray). The magenta line in panels b–f connects the raw count of PCR-positive samples across collection dates. Source data are provided in a Source Data file and in Supplementary Data 1.

Delta and Omicron variants identified in US farmed mink

To determine which SARS-CoV-2 lineages were circulating in mink on Farms N, O, and P, high-quality whole-genome sequences were obtained from 293 samples: 48 from Farm N, 1 from Farm O, and 244 from Farm P (Supplementary Table 3). The viruses from Farm N (State 1) were classified as members of two Omicron sublineages: BF.1 (n = 44) and BA.5.2.1 (n = 4) (Fig. 3). The one virus in Farm O (State 1) belonged to Omicron sublineage BA.4.1. All 244 sequences from Farm P (State 2) were classified as belonging to the AY.39 Delta sublineage. To our knowledge, these data represent the first detection of natural infection with Delta and Omicron in US mink. All other SARS-CoV-2 sequences available in GISAID from US mink (n = 281) correspond to earlier lineages (e.g., B1.1) without WHO variant of concern (VOC) classification. Globally, very few SARS-CoV-2 sequences from mink have been published for the Delta (n = 63) and Omicron (n = 13) VOCs (Supplementary Tables 1, 4). However, the specific subvariants we detected (Delta AY.39; Omicron BA.4.1, BA.5.2.1; and BF.1) are distinct from those previously reported, indicating four new spillovers into mink (Fig. 3). Two human-to-mink introductions occurred on Farm N (BF.1 and BA.5.2.1), one introduction occurred on Farm O (BA.4.1), and one introduction occurred on Farm P (AY.39).

Fig. 3. Global phylogeny of SARS-CoV-2 from mink and humans, 2019–2023.

Midpoint-rooted maximum-likelihood phylogeny of SARS-CoV-2 sequences sampled from mink and humans worldwide between December 26, 2019 – April 24, 2023 (the collection date of the most recently available mink-derived sequence in the global dataset). Colored circles at branch tips denote mink-derived isolates, with fill color indicating country of origin (see legend); uncircled tips represent human-derived isolates. Background shading denotes WHO variant of concern (VOC) designations: Alpha (light green), Beta (magenta), Delta (beige), Gamma (green), and Omicron (cyan); pre-WHO VOC lineages are shaded gray, and WHO variants of interest are shaded light gray. Labeled clades highlight sequences generated in this study, with Pango lineage assignments and sample sizes indicated: Farm N (BF.1, n = 44), Farm N (BA.5.2.1, n = 4), Farm O (BA.4.1, n = 1), and Farm P (AY.39, n = 244).

Omicron (BF.1) outbreak on Farm N

BF.1 caused the largest Omicron outbreak seen in our study, which was detected during September 6 - 21, 2022 (Supplementary Figs. 2, 3). All Omicron viruses identified on Farm N and Farm O belonged to strains circulating (albeit at low levels) in State 1 humans at the time and likely represented recent human-to-mink transmission events (Supplementary Fig. 3). BF.1 was detected in State 1 in humans during June–July 2022, which overlaps with the estimating timing of the human-to-mink introduction of BF.1 during the summer of 2022 (time to the most recent common ancestor (TMRCA), node B, Supplementary Fig. 4a). This suggests that BF.1 was still circulating among State 1 in humans at the time of the introduction into Farm N. On Farm N, six substitutions rose to >10% among the Omicron BF.1 viruses: NSP2:N195K, NSP5:G278E, S:K478T, S:T572I (fixed), and M:T7I (fixed) (Supplementary Fig. 4b). A recurrent acquisition of reversion S:K478T was observed in this farm, emerging twice and found in 13 Farm N samples.

Delta (AY.39) outbreak in Farm P

The first AY.39 virus on Farm P mink was identified on August 8, 2022, approximately four months after the AY.39 sublineage was last detected in State 2 (April 17, 2022) and three months after any form of the Delta variant was last detected in humans in State 2 (May 17, 2022) (Fig. 4). The major Delta wave in humans occurred during the autumn of 2021, approximately one year prior to the sampling of Farm P in autumn 2022 (Fig. 4). Among Delta viruses in humans in State 2, only 4.2% of sequenced Delta viruses belonged to the AY.39 sublineage. A phylogenetic tree that includes AY.39 viruses from humans, mink, and WTD indicates that all AY.39 viruses on Farm P are monophyletic, which suggests the entire mink outbreak, including both waves, was sourced by a single virus introduction (Fig. 5, S5A). The most closely related AY.39 viruses were sampled in humans from State 2 in December 2021, eight months before AY.39 was first detected on Farm P. It is not clear in which host the AY.39 virus was persisting in State 2 during this long gap.

Fig. 4. Re-emergence of AY.39 in mink.

Epidemiological curves represent the number of weekly COVID-19 cases in humans in State 2 from January 1, 2021 to January 31, 2023. The curves are colored by the proportion of human SARS-CoV-2 sequences in State 2 that belong to the Alpha (light green), Delta (non-AY.39; peach), Delta (AY.39; blue) or Omicron (teal) sublineages, estimated weekly (n = 68,304 sequences). The blue horizontal bar below the x-axis shows the circulation time for the AY.39 Delta sublineage in humans in State 2 (last detection in humans: April 17, 2022). The vertical blue bars represent the number of sequenced AY.39 viruses from Farm P (n = 244 sequences; right y-axis). The black bar represents the time period between the last detection of AY.39 in humans and the first detection on Farm P (~4 months). Each deer icon represents an AY.39 white-tailed deer (WTD) sample collected in an adjacent county to Farm P. Source data are provided in a Source Data file.

Fig. 5. Evolution AY.39 viruses in mink and white-tailed deer (WTD).

a Time-scaled MCC tree of AY.39 viruses sequenced from Farm P mink (n = 176), WTD (n = 28) and humans (n = 389). Branches are shaded by host: human (gray), mink (blue), WTD (brown). A subsection of MCC tree with the mink clade is shown on the right. Tips are colored by barn: yellow = P1, pink = P2. The 95% highest posterior density (HPD) distribution plots for the TMRCA are shown for node A (common ancestor of mink clade and most closely related human virus) and node B (common ancestor of the mink clade). Clade A (blue) and Clade B (orange) are shaded. Select clade-defining mutations are provided. b Proportion of farm P samples (n = 176) with identified nonsynonymous substitutions, classified as derived (away from the ancestral state) or reversion (back to the ancestral state). Genome coordinates of annotated SARS-CoV-2 open reading frames correspond to the Wuhan-Hu-1 reference genome sequence (GenBank RefSeq accession NC_045512).

Two AY.39 clades on Farm P

AY.39 virus diversified into two large clades on Farm P, named clade A (n = 151 sequences) and clade B (n = 93 sequences) (Fig. 5a). Clade A is defined by two amino acid substitutions (NSP4:G196S and ORF10:T38I). Clade B is defined by NSP1:∆82–84 and NSP1:M85V (Source Data, Fig. 5a). Clade A was found primarily in Barn P1, and Clade B was found primarily in Barn P2 (Fig. 5a, S6). Both clades co-circulated during the first wave of the AY.39 outbreak on Farm P during August 8 – September 13, 2022. Most viruses in Clade A contain two amino acid substitutions (NSP9:G37E and S:Y453F) that were characterized as mink adaptive in prior studies^5,18–20, which emerged early in the Clade A outbreak (August 28, 2022). Wave 2 occurred when clade A viruses with the two mink adaptations spread to Barn P2 (first detected October 10, 2022). Viruses with the two mink adaptations were also subsequently detected in four WTD during December 2022 and January 2023 in two counties in unrecorded locations that are <60 miles from Farm P (Fig. 5). The mink adaptations NSP9:G37E and S:Y453F found on Farm P are highly associated with mink globally, with a global prevalence in mink of 21.1% and 43.3%, respectively, but have never been detected in WTD and only rarely in humans (<0.1%) (Supplementary Table 5). Notably, reversion S: K478T was observed on this farm as well, emerging 12 independent times (Supplementary Fig. 7).

Mink-related AY.39 viruses identified in local free-ranging white-tailed deer

A long branch on the phylogenetic tree separates the four AY.39 viruses collected in WTD from the most closely related mink viruses from Farm P1 (Clade A), on both the ML tree (Supplementary Fig. 5a) and the MCC tree (Fig. 5). The TMRCA estimate [95% highest posterior density (HPD): September–October 2022] suggests that AY.39 had been circulating in WTD for at least two months prior to the first detection WTD on December 12, 2022 but not prior to the AY.39 outbreak in mink. Six amino acid mutations found in all four WTD viruses were not present in mink at the start of the mink outbreak but subsequently emerged in Clade A as the virus replicated and transmitted in mink and were transmitted onward to WTD. The likelihood of WTD acquiring the same set of mutations independently, rather than being exposed, whether directly or indirectly, to a virus that originated in mink, is extremely low.

Higher evolutionary rate of SARS-CoV-2 in mink and WTD compared to humans

A host-specific local clock (HSLC) was used in our Bayesian phylogenetic analyses to account for the higher rate of SARS-CoV-2 virus evolution in deer and mink compared to humans, which was observed in the root-to-tip regression analysis of AY.39 sequences (Supplementary Fig. 5b) and in previous reports^13,21. Compared to humans (5.46 × 10⁻⁴ subs/site/year, 95% HPD: 4.88 × 10⁻⁴ – 6.07 × 10⁻⁴), AY.39 evolved ~1.5x higher in mink (8.25 × 10⁻⁴ subs/site/year, 95% HPD: 6.31 × 10⁻⁴ – 1.01 × 10⁻³) and ~3.5x higher in deer (1.94 × 10⁻³ subs/site/year, 95% HPD: 1.54 × 10⁻³–2.40 × 10⁻³, Fig. 6A). The Omicron BF.1 sublineage evolved twice as fast in mink (7.07 × 10⁻⁴ subs/site/year, 95% HPD: 4.25 × 10⁻⁴–1.01 × 10⁻³) compared to humans (3.44 × 10⁻⁴ subs/site/year, 95% HPD: 4.88 × 10^-4–6.07 × 10⁻⁴, Supplementary Fig. 8). How the elevated substitution rates in mink and WTD relates to different underlying mutation patterns compared to humans is explored below.

Fig. 6. Host-specific evolutionary and mutational dynamics of SARS-CoV-2 in mink and white-tailed deer (WTD).

a Posterior distributions of evolutionary rates (substitutions/site/year) estimated under a host-specific local clock model for human (grey), mink (blue), and WTD (brown) AY.39 sequences. Dashed vertical lines and values above indicate the mean rate for each host. b C > T substitution proportion over time for AY.39 sequences stratified by host. Lines represent fitted linear regressions and shaded areas indicate 95% confidence intervals. R² and p-values from two-sided ordinary least-squares linear regressions are shown for each host; no adjustment for multiple comparisons was applied. Asterisks denote the significance of the mink temporal trend (*** p = 3.00 × 10⁻²¹). c Divergence of the mink mutational spectrum from the human baseline over the course of the Farm P outbreak. Each point represents the Euclidean distance of an individual mink (circles) or State 2 WTD (triangles) AY.39 sequence from the centroid of human AY.39 sequences in PC1–PC2 principal component space. The vertical dotted line marks the phylogenetically estimated time to the most recent common ancestor (TMRCA: May 29, 2022). The line represents a fitted linear regression, and the shaded area indicates the 95% confidence interval. R² and p-values from a two-sided ordinary least-squares linear regression are shown; no adjustment for multiple comparisons was applied. State 2 WTD sequences are highlighted with a dashed ellipse. Source data are provided in a Source Data file.

Mutation patterns in mink

Differences in mutation spectra were examined across AY.39 sequences obtained from mink, humans, and WTD (Supplementary Fig. 9 and Supplementary Data 2). Mink sequences experienced a significant decline in C > T substitution frequency over time (R² = −0.40, p = 3 × 10⁻²⁰; Fig. 6b), in contrast to stable patterns in humans (R² = 0.004, p = 0.216). An upward trend in WTD is consistent with previous reports^13,14, but was not statistically significant in our dataset, likely due to limited sample size. Other substitutions also exhibit temporal patterns in mink, with the strongest effects observed for C > G (R²= −0.626) and T > A (R² = 0.578) (Supplementary Fig. 10; Supplementary Data 2). Mutation spectra differed significantly among mink, human, and deer in a principal components analysis (Supplementary Fig. 11). The State 2 deer sequences diverged from the mink clade over time, with the earliest deer sequence (December 4, 2022) clustering near the mink group along PC1 and deer sequences collected one month later (January 5, 2023) shifting toward the deer cluster, consistent with evolution from mink-like to deer-like (Supplementary Fig. 12). Permutation-based analyses confirmed significant differences between host groups (p < 0.001, Supplementary Data 2). As expected, AY.39 mink sequences associated with a single outbreak formed a more compact cluster (dispersion = 0.036) than AY.39 deer sequences sampled across multiple spillovers (dispersion = 0.066; Supplementary Fig. 13). Mutation spectrum divergence from the human baseline increased progressively over the course of the mink outbreak (R² = 0.504, p = 2.74 × 10⁻²⁸; Fig. 6c), demonstrating that the mink-specific spectral signature accumulated during circulation.

SARS-CoV-2 reinfections in Farm P mink

The housing of mink on Farm P mink in two separate barns (P1 and P2) allowed us to examine how the two waves of AY.39 progressed in the two buildings (Supplementary Fig. 12). Using conservative criteria (see Methods), 6/82 (7.3%) pens in barn P1 and 31/54 (57.4%) pens in barn P2 showed evidence of reinfection (see Source Data). On average, there was a span of ~44 days (range: 19–87 days, Fig. 5) between primary and secondary infections of identified reinfections in barn P1 and ~59 days for barn P2. Four pens in barn P2 had evidence of multiple rounds of reinfection (pens 30, 58,70 and 84). Eight pens in barn P2 had evidence of reinfection with a different clade, first with Clade B (with the NSP1:∆82–84 deletion) during wave 1 and then with Clade A (with the S:Y453F mink-specific adaptation) during wave 2 (Figs. 5a, 7). Reinfections over such a short period could have resulted from the evolution of the virus. Reinfection was highest in barn P2; the ∆82–84 deletion in a region of NSP1 is associated with recurring deletions in immunocompromised human patients with persistent infections^22,23. The clade A viruses that reinfected barn P2 animals had the S: Y453F mink-specific adaptation that increases affinity for mink and human ACE2 receptors and is linked to immune escape^24–26.

Fig. 7. Inter- and intra-clade reinfections in barn P2 pens.

a Stacked bar plots showing rRT-PCR test results for samples by collection date and clade designation for sequenced isolates (circles) collected from eight barn P2 pens identified as having evidence of reinfection. Each pen held the same 1 or 2 mink throughout the sample collection period. Only instances where all animals in the same pen were simultaneously infected by viruses from the same clade before or after at least two consecutive negative rRT-PCR test results between infections were considered. b Summary of reinfection dynamics in the eight pens. Mink cartoons represent sequenced samples, which are colored by clade: blue=clade A and orange=clade B. The latest clade B samples collected during the first infection period (i.e., prior to at least two consecutive negative rRT-PCR test results for all animals in the pen) are highlighted. For samples collected during secondary infections (i.e., after at least two consecutive negative rRT-PCR test results for all animals in the pen), the earliest dates of isolation are highlighted except for pen 12, where isolates simultaneously infecting both animals (providing evidence of reinfection) are highlighted instead. Periods between reinfections are shown, with positive rRT-PCR sample collections between reinfections shown in magenta and the days of consecutive negative rRT-PCR test results for all animals in the pen shown in grey. Source data are provided in a Source Data file.

Discussion

The voluntary participation of US mink farmers in active surveillance for SARS-CoV-2 revealed that mink continue to be susceptible hosts for human VOCs that have highly mutated away from the original SARS-CoV-2 virus that spilled over into US mink in 2020. Detecting numerous subclinical infections in mink highlights the need for active surveillance to accurately assess SARS-CoV-2 in animal populations, especially since the virus is often introduced by humans. SARS-CoV-2 outbreaks on the farms we studied were relatively short in duration, less than five months, but still provided time for the virus to evolve in new trajectories on mink farms, including mink-specific adaptations and diversification into multiple clades. Consistent with this, our analyses showed that SARS-CoV-2 evolved at approximately 1.5x the human rate in mink and exhibited strong host-specific mutational signatures that accumulated progressively during circulation, suggesting that the mink cellular environment actively reshapes the viral substitution landscape in ways that distinguish it from both human and deer-associated viruses. These signatures potentially reflect differences in the activity of host RNA editing enzymes, a phenomenon that has been described for SARS-CoV-2 and other RNA viruses^27–29. The complexity of viral evolution in mink was further amplified by the capacity for mink to be reinfected, which resulted in longer outbreaks with multiple peaks and provided extended opportunities for viral adaptation. However, the biological mechanisms underlying both these host-specific mutational processes and the reinfection dynamics in mink remain poorly understood, and experimental studies characterizing mink RNA editing machinery, viral-host interactions, and immune responses will be needed to fully interpret these findings. Finally, the detection in mink of a Delta variant that had ceased to be detected in humans represents the most surprising finding in this study, especially because a genetically similar virus with mink-adapted mutations was subsequently detected in local WTD. These mink-adapted mutations were not detected in the human population, and infected people remain the overwhelming source of SARS-CoV-2 detections. Together, these findings indicate that mink continue to be susceptible to SARS-CoV-2 viruses, even as the virus adapts to humans over time, and continued partnerships with mink farmers to monitor susceptibility and evolution is warranted given the frequency of reverse zoonosis.

Our study cannot resolve why human-origin AY.39 viruses were observed in farmed mink months after the sublineage ceased to be detected in humans. Our phylogenetic analysis provides no evidence that the AY.39 virus was at any point evolutionarily frozen in time, as would be expected in situations involving persistence in wastewater or another environment or infection by frozen meat fed to mink (e.g., raw venison). Rather, our results are consistent with AY.39 circulating and replicating for an extended period (4–8 months) in a natural host. One scenario is that AY.39 was transmitted at low levels in humans prior to its emergence in mink, which may have gone undetected due to undersampling of low-prevalence transmission chains and regional biases in genomic surveillance data. Immunocompromised humans could have sustained Delta and other variants after they went extinct in the general population, and AY.39 continued to be sporadically detected in humans for months after Omicron became dominant at the end of 2022. The likelihood of such a low-frequency variant infecting one of the few farm workers with occupational exposure to mink is low but cannot be ruled out.

Another possibility is that WTD or another non-human host served as an intermediary between humans and mink. Several studies in WTD document the persistence of SARS-CoV-2 lineages that are no longer in circulation among humans. Alpha, Delta, and Gamma variants were all detected in New York WTD during the 2020 and 2021 hunting seasons, long after the Alpha and Gamma variants were last detected in humans in New York³⁰. Similarly, a virus emerged in two mink farms in Poland in late 2022–early 2023 that was most closely related to a non-VOC lineage last detected in late 2020–early 2021³¹. One limitation of our study is that WTD samples were largely collected during hunting seasons, which means periods outside of a state’s hunting season are not well represented in the dataset. Movement of people and distance between farms have been identified as predictors of SARS-CoV-2 dispersal between mink farms in the Netherlands⁵, but the movements of animals, equipment, and workers between farms are typically not well documented, and the precise mode of transmission off farms remains unclear.

Interspecies transmission of SARS-CoV-2 between non-human hosts is rarely documented, either because it occurs infrequently or surveillance in non-human hosts is too low to capture it. Therefore, the discovery of genetically similar AY.39 viruses in farmed mink and WTD was surprising. To our knowledge, the only prior reported instance of interspecies transmission involving two non-human hosts is a report of SARS-CoV-2 transmitting from farmed mink to barn-dwelling feral cats in the Netherlands in 2020³². How SARS-CoV-2 viruses transmit from mink to WTD is less clear, but could involve escaped mink, which tested positive for SARS-CoV-2 near farms in British Columbia, Canada³³ and in Utah, USA³⁴. Dispersion of SARS-CoV-2 to outdoor air is limited in infected mink farms³⁵ and aerosolized transmission from mink to WTD is unlikely. Possible mink-to-deer SARS-CoV-2 transmission was hypothesized in a previous study by Pickering et al.¹⁴, where a highly divergent B.1.641 lineage found in WTD near the US-Canadian border in southwest Ontario, Canada, shared a common ancestor with viruses collected from a mink farm outbreak (and farm workers) located across the border in northern Michigan. The long branch lengths and large gaps in sampling make it difficult to ascertain whether spillover occurred directly from the mink farm to WTD, or whether the divergent WTD virus was sourced from an unsampled virus population circulating in humans or another host species. In contrast to our study, the Ontario WTD sequences identified by Pickering et al. did not have any known mink adaptations. Nor were the Ontario WTD sequences nested phylogenetically within a discrete mink clade in the way the WTD sequences in our study were nested within the mink-only AY.39 clade. While the Michigan mink outbreak is less clearly connected to the subsequent detection of the virus in WTD in Ontario, both studies expose major sampling gaps in animals that render it difficult to finely reconstruct multiple interspecies transmission events involving humans, mink, deer, and potentially other species.

Given the broad and largely unmonitored circulation of SARS-CoV-2 in wild deer populations, it remains plausible that deer could have served as the original reservoir in this instance, with subsequent spillover into farmed mink or shared transmission sources affecting both species independently. However, the limited genetic diversity of AY.39 in WTD across two different counties in December 2022–January 2023 does not support widespread or long-term transmission of AY.39 in the WTD population in State 2 that predates the mink outbreak. Nor does the presence of six amino acid changes in WTD that emerged in mink over the course of the farm outbreak that were not present in mink at the onset.

Our mutation spectrum analysis provided additional support for this interpretation: the four WTD sequences nested within the mink clade clustered near the mink spectral centroid when projected into PCA space, with the earliest sequence (December 4, 2022) showing the most mink-like spectrum and the three sequences collected one month later beginning to shift toward the human/deer spectral profile, consistent with recent mink ancestry followed by early-stage adaptation to the deer host environment. The most likely explanation is that these six mutations were present in WTD because mink were the source of the WTD outbreak, either directly or indirectly.

In conclusion, our study demonstrates the benefit of integrating multiple data streams from humans, domestic animals, and wildlife into a One Health framework so that sequences collected from different host species and locations can be compared and host jumps can be identified in all directions. Genomic data provide a valuable tool for understanding how pathogens transmit within farms and between livestock, farm workers, and wildlife, which can guide biosecurity practices and risk assessment. Our findings further suggest that mutation spectrum analysis represents a valuable complementary approach for characterizing viral evolution in farmed mink, and that future surveillance efforts should prioritize generating genomic datasets large enough to enable within-outbreak temporal analyses, a goal that is achievable through active farm-level surveillance programs such as those described here. Future studies should also include sampling of workers on mink farms, which could potentially shed more light on the source of uncommon or unexpected viruses, such as the AY.39 virus on Farm P. Currently, the absence of routine SARS-CoV-2 surveillance in non-human hosts means the US may not be picking up other cases of virus persistence, evolution, and inter-species transmission. Surveillance and studies should integrate the likelihood and magnitude of the risk of the disease on which they focus when designing which populations to study and how frequently and intensely to study them. Study of complex multi-host species disease systems such as SARS-CoV-2 requires building successful partnerships between government, academia, and agricultural communities.

Methods

Sample collection

All procedures were performed under the oversight of the Ohio State University Institutional Animal Care and Use Committee (approved protocol #2022A00000032). Eighteen mink farms from across the United States were included in this study. On each farm, staff selected 250 mink to longitudinally follow from July 2022 through harvest (November–December 2022). Each week, farm staff sampled the external nares of each mink using sterile swabs, which were subsequently placed in 1 mL viral transport media (BD UVT cat #220526, Franklin Lakes, New Jersey, USA). The samples were placed on ice packs and shipped to the laboratory, where, upon receipt, the samples were stored at -80 °C until diagnostic testing was performed. Nasal swab sampling resulted in the collection of 85,656 samples, of which 33,247 were tested after completion of the primary sample collection period based on serology data. In Farm P, additional samples from 50 and 250 new mink subjects were collected on April 1 and July 22, 2024, respectively, to assess whether the outbreak was active.

Every four weeks, farm staff collected blood from the mink in addition to the nasal swabs. Blood collection occurred by capillary puncture of a toenail. For each blood sample, 200 µl of blood was collected using a capillary blood collection system (Ram Scientific cat #07 7210, Austell, Georgia, USA). Blood samples were shipped to the lab on ice packs. On arrival, microcentrifuge tubes were centrifuged at 1200 x g for 10 minutes to separate serum from whole blood. Once centrifuged, serum samples were stored at −80 °C until diagnostic testing. Serum sampling resulted in the collection of 19,087 samples, of which 15,903 were tested after completion of the primary sample collection period. On April 1, 2024, an additional 50 mink serum samples were collected in Farm P to assess ongoing exposure to SARS-CoV-2.

Diagnostic testing

Potential SARS-CoV-2 exposure was investigated using the SARS-CoV-2 Surrogate Virus Neutralization Test (sVNT, GenScript L00847-A, Piscataway, New Jersey, USA) according to the manufacturer’s protocol. The assay was modified by replacing the wild-type RBD-HRP with an Omicron variant RBD-HRP (GenScript Z03730-500) and the corresponding Omicron neutralizing antibody standard (GenScript A02161-100). Serum samples from 12 farms with the presence of unvaccinated mink were tested for evidence of seroconversion. Based on the serology results, four farms (N, O, P, and S) with evidence of seroconversion were selected for investigation of SARS-CoV-2 infection via real-time reverse transcription polymerase chain reaction (rRT-PCR). This excludes ‘one gene’ positive samples, which were presumed positive and re-tested but were not confirmed as positive. More than half of all mink (392/750; 52.3%) tested positive at least once (Source Data and Supplementary Table 2). Most of the farms had some number of both vaccinated and unvaccinated mink, which created a level of uncertainty in the serology results as vaccinated mink would show up sera positive although they had not previously been exposed to SARS-CoV-2. rRT-PCR was also performed on samples collected every other week from the remaining eight farms with the presence of unvaccinated mink enrolled in the study; no infections were discovered. Farm S was epidemiologically linked to Farms N and O, and testing was initiated for 1470 nasal swabs collected between August 9 and September 13, 2022. Farm S had 4 mink test positive from the first two sample collection dates, followed by four consecutive weeks where all mink tested negative. No genetic sequences were obtained from Farm S.

Viral RNA was extracted using the Mag-Bind Viral DNA/RNA 96 kit (Omega Bio-tek, cat# M6246-03, Norcross, Georgia USA) per the manufacturer’s protocol. Extracted viral RNA was tested using the TaqPath COVID-19 Combo Kit (Applied Biosystems cat #A47814, Waltham, Massachusetts USA). A cycle threshold (Ct) value of 37 or below on at least two of the targets was considered a positive result.

Genomic sequencing

Original sample material for 526/760 (69.2%) PCR-positive nasal swab samples was sent to the National Veterinary Services Laboratories (NVSL) for whole-genome sequencing. Samples were selected for sequencing based on a cycle threshold (Ct) value of less than 34.0 on either two or three of the targeted genes. The Nextera XT DNA Sample Preparation Kit was used in accordance with manufacturer instructions to prepare cDNA libraries from viral RNA that was amplified via PCR. The 500-cycle MiSeq Reagent Kit v2 was used to perform sequencing. Sequences were assembled using IRMA v0.6.7 and DNAStar SeqMan NGen v14.0.1. A total of 462/526 (87.8%) assemblies were obtained from NVSL, of which 401 were selected based on a ≥ 80% genome coverage threshold. Pangolin v3.1.11³⁶ was used to assign variant lineages for all sequences. Variant annotation, validation, and quality assessment of the consensus sequences were performed using VADR³⁷. A total of 293 genomes that passed quality control were used in all downstream analyses and were deposited in GenBank. Accession numbers are available in Supplementary Table 3.

SARS-CoV-2 epidemiology in humans

Epidemiological curves of SARS-CoV-2 cases in humans were generated for each sampled US state using the number of daily reported COVID-19 cases in that state from January 1, 2021 to January 31, 2023 (source: Johns Hopkins University Center for Systems Science and Engineering (https://github.com/CSSEGISandData/COVID-19); retrieved through outbreak.info³⁸). Metadata for all human SARS-CoV-2 genome sequences from each state were downloaded from the Global Initiative on Sharing All Influenza Data (GISAID) database (https://gisaid.org/) to estimate the proportion of sequences belonging to different Pango lineages during each week of the epidemic. The datasets were filtered to only include submissions with complete collection dates and sufficient coverage to assign a Pango lineage, resulting in a State 1 dataset of 18,784 sequences and a State 2 dataset of 68,304 sequences. New daily cases were then binned by week and shaded by the estimated proportions of Pango lineages, which provided an approximation of the number of cases attributed to each Pango lineage in circulation. For simplicity, sublineages were consolidated into “Alpha”, “Delta”, “Omicron”, and “other” categories based on WHO variant classifications, except for sublineages of interest in State 1 (i.e., “Omicron (BA.4.1)”, “Omicron (BA.5.2.1)” and “Omicron (BF.1)”) and State 2 (i.e., “Delta (AY.39)”). Counts and proportions were calculated using R v4.4.0³⁹ and visualized using the ggplot2 R package⁴⁰.

Dataset curation

A global mink SARS-CoV-2 dataset was compiled by querying the GISAID database (accessed on July 3, 2024) for SARS-CoV-2 sequences collected from mink since the start of the pandemic. The search included all ‘Neovison vison’ (American mink), ‘Mustela lutreola’ (European mink), and ‘mink’ samples available in GISAID (n = 1505). The resulting mink dataset was filtered for quality by applying search filters to include only ‘complete’ genome sequences (i.e., with >29,000 nucleotides) while excluding sequences with low coverage (i.e., >5% unresolved nucleotides). The 1442 mink-associated sequences were used for subsequent phylogenetic analyses. A representative global human background dataset was generated by randomly selecting 5,000 sequences from the Nextstrain⁴¹ global SARS-CoV-2 build. The subsampling was limited to sequences from the full GenBank database that were collected from December 1, 2019, to the time when the last publicly available mink sample was collected on April 24, 2023. The randomly selected sequences were filtered to include only “complete” genome sequences with a good QC status assignment that were isolated from humans. The final dataset (n = 5,444) consisted of the 1442 global mink sequences, 3709 human background sequences, and the 293 US farmed mink sequences generated in this study.

Separate sublineage datasets of human Delta AY.39 and Omicron BF.1 genomic sequences were generated to assess the evolutionary relationships between human and farmed mink samples. Global sublineage datasets were compiled by querying the GISAID database (accessed on April 7, 2024) for complete SARS-CoV-2 sequences collected since the start of the pandemic. Search filters were applied to include “complete” genome sequences with complete collection dates and exclude sequences with low coverage. The initial background datasets included 50,853 AY.39 and 11,018 BF.1 sequences. Following multiple sequence alignment (see next section), sequences identified by Nextclade as having less than “good” overall QC status were removed. The datasets were subsampled further using a US-focused strategy that included one randomly selected genome for each collection date per US state and one randomly selected genome for each collection date per country outside of the US. All available non-human sequences were also included in each dataset. We identified and added the closest publicly available isolates to the AY.39 and BF.1 sequences generated in this study by placing farm sequences on the UShER (Ultrafast Sample placement on Existing tRee)⁴² pre-built phylogeny, as those additional isolates may help resolve evolutionary relationships in our phylogenetic analysis despite potential biases that might arise in our subsampling scheme. The final BF.1 dataset (n = 3070) consisted of 924 US human sequences, 2101 non-US human sequences, 1 non-human sequence (cat), and the 44 farm N samples. For the AY.39 dataset, four additional AY.39 sequences sampled from State 2 white-tailed deer (WTD) were sequenced and provided by the US Department of Agriculture (USDA) National Veterinary Services Laboratories (NVSL). Deer sequences were obtained from SARS-CoV-2-positive nasal swabs collected post-mortem from hunter-harvested deer; therefore, collection of these samples was exempt from animal use oversight. The final AY.39 dataset (n = 8069) included 7039 human sequences from the United States and US territories, 244 farm P samples from mink, 752 non-US human sequences, 34 non-human sequences (collected from the environment (n = 2), WTD (n = 28), dog (n = 1), lion (n = 2) and a tiger (n = 1)). Since only one Omicron BA.4.1 and four BA.5.2.1 viruses were sequenced from mink in our study, no further phylogenetic analyses were performed for these sublineages. The EPI_SET ids for GISAID sequences used across all datasets are listed in Supplementary Table 6.

Phylogenetic analyses

The curated datasets were aligned with NextClade v.3.3.1⁴³ using the Wuhan-Hu-1 genome (GenBank RefSeq accession: NC_045512) as a reference. Custom Python scripts were used to remove non-coding regions and mask sites that are known to be unreliable. Maximum-likelihood (ML) phylogenies were inferred for each dataset using IQ-Tree v2.47⁴² with a general-time reversible (GTR) model of nucleotide substitution with gamma-distributed (+G) rate variation among sites and 1000 ultra-fast bootstrap replicates, using the high-performance computational capabilities of the Biowulf Linux cluster at the National Institutes of Health (http://biowulf.nih.gov).

The temporal signal for the Delta AY.39 and Omicron BF.1 ML phylogenies was evaluated by plotting the root-to-tip distance for all sequences using the heuristic residual mean squared rooting approach in TempEst v1.5.3⁴⁴. Outlier sequences were identified based on significant deviations from the root-to-tip regression and removed from the dataset using TreeTime v0.10.1⁴⁵. Prior to performing Bayesian phylogenetic analyses, the curated datasets were further reduced using the monophyletic sampling approach available in smot v.0.17.4⁴⁶. The analyzed BF.1 dataset (n = 99) included 54 human, 1 cat and the 44 farm N sequences. For the AY.39 dataset, groups of identical mink sequences were also subsampled by keeping only one identical sample per collection date to reduce redundancy without significantly biasing the analyses. All AY.39 WTD sequences were kept in the dataset. The final AY.39 dataset (n = 593) consisted of 389 human, 28 WTD, and 176 Farm P mink AY.39 sequences. Time-scaled Bayesian phylogenetic analyses were performed using the Markov chain Monte Carlo (MCMC) method in BEAST v1.10.5⁴⁷, using GPUs available from the NIH Biowulf Linux cluster. A non-parametric Bayesian Skygrid⁴⁸ coalescent tree prior was specified with a GTR + G nucleotide substitution model with four gamma categories. A host-specific local clock (HSLC) model⁴⁹ was used to account for the differences in substitution rates between human, mink, and deer sequences. Monophyletic WTD clusters and singleton WTD viruses identified in the ML tree were constrained as groups to calculate the WTD evolutionary rates for the AY.39 viruses. The HSLC specification accommodated the nested relationship of the WTD transmission cluster within the farm P mink clade and therefore modeled the human-to-mink rate transition and the subsequent mink-to-WTD rate transition. The MCMC chain was run 3 or 4 times for each dataset for 72 to 100 million MCMC iterations using the BEAGLE 3⁵⁰ library to improve computational performance. Runs for the same dataset were combined using LogCombiner v1.10.4 while discarding 10% as burn-in. Convergence was determined by verifying that all parameters had ESS values ≥ 200 using Tracer v.1.7.2⁵¹. Maximum clade credibility trees (MCC) summarizing posterior tree distributions were generated using TreeAnnotator v1.10.4. Posterior distributions for the time to the most recent common ancestor for clades of interest were obtained from the posterior tree distributions using TreeStat v1.10.4. Phylogenies were visualized using the ggtree R package v3.13.1⁵².

Mutation analysis

Mutation calling and annotation for the mink farm isolates in this study were performed through NextClade v3.3.1 and the coronapp⁵³ COVID-19 genome annotator (http://giorgilab.unibo.it/coronannotator/). The prevalence of mutations among global human and mink SARS-CoV-2 isolates was obtained through GISAID after excluding sequences with low coverage.

Mutational spectrum analysis

Mutation spectrum analysis was performed on AY.39 sequences from mink (n = 176), humans (n = 389), and deer (n = 24) using the Nextclade mutation results. Given that all sequences belonged to PANGO lineage AY.39, viral lineage was eliminated as a potential confounding factor in host group comparisons. Sequences with fewer than 15 substitutions relative to Wuhan-Hu-1 (MN908947.3) were excluded from analysis. For each genome, the frequency of each of the 12 possible nucleotide substitution types was computed from the Nextclade “substitutions” field and normalized to proportions summing to one. The four deer samples phylogenetically linked to the mink outbreak were excluded from primary analyses and projected post-hoc into the ordination space to assess their spectral affinity relative to host group centroids.

We first assessed whether substitution frequencies changed systematically over time in human AY.39 sequences by regressing each of the 12 substitution type proportions (i.e., (C > T, C > A, C > G, T > C, T > A, T > G, G > A, G > T, G > C, A > G, A > T, A > C) against collection date. No significant temporal trend was detected for C > T substitutions (p = 0.216) and, while five of the twelve substitutions had a significant R², the maximum R² across all 12 substitution types was 0.058, indicating that temporal trends were weak within this lineage in the human reservoir (Supplementary Data 2). As a result, no temporal correction was applied to the subsequent ordination analysis.

To compare overall mutation spectra between host groups, we used permutation-based multivariate analysis of variance (adonis2(), vegan R package v2.7.3⁵⁴) with Bray-Curtis dissimilarity and 9,999 permutations. Although no significant temporal trends in substitution frequencies were detected within the human AY.39 sequences, mink sequences were collected substantially later than human and deer sequences (mink: August–November 2022; human: May 2021–January 2022; deer: November 2021–February 2022). To account for potential confounding from non-overlapping sampling periods, collection date was included as a covariate in the multivariate model; effect size is reported as R², representing the proportion of spectral variation explained by host group after accounting for collection date. Pairwise comparisons were Bonferroni-corrected. We also tested whether sequences within each host group were variable in their spectra using betadisper() with 9999 permutations, since high within-group variability can affect interpretation of between-group differences.

We performed principal component analysis (PCA) on raw substitution type proportions to visualize how host groups differ in spectrum space. As mentioned above, the four State 2 WTD sequences phylogenetically nested within the mink clade were excluded from the PCA and all other statistical analyses and were instead projected post-hoc into the PCA space using the loadings derived from the main analysis. To quantify how far each mink sequence had diverged from the human baseline in spectrum space, we computed the Euclidean distance of each mink sequence from the human AY.39 centroid using the first two principal components. We then assessed whether divergence from the human centroid increased with estimated mink circulation time using linear regression, with circulation time defined as days elapsed since the phylogenetically estimated date of outbreak origin (TMRCA: 2022-05-29).

Identification of farm P pen reinfections

Pen data were provided for mink housed in farm P. Since the same animals were repeatedly sampled during the study period, each pen—which contained 1 or 2 mink —was examined for evidence indicating that any of the animals was infected more than once (i.e., reinfected). However, while the same individuals were housed in the same pen during the entirety of the study, we cannot discard the possibility that samples might have been attributed to the wrong individual during sampling. Thus, to identify pens with evidence of reinfection, the following criteria were used: (a) two rounds of positive PCR results separated by at least two consecutive negative results were identified, and (b) in at least one round of infections, both animals were simultaneously positive (i.e., on the same date). A single round of infection was considered to have ended when all animals tested negative at the same time for two consecutive sample collections to account for possible false negatives.

Reporting summary

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

Author contributions

Conceptualization: J.K., M.N., A.S.B., C.H., H.H.H., J.E. Data curation: D.H., D.S.M., M.C.O., A.C.-B. Formal analysis: A.C.-B., M.I.N., P.L., M.C.O., N.N.C. Funding acquisition: J.K., M.N., A.S.B. Investigation: D.H., D.S.M., M.C.O., A.C.-B., N.N.C., M.L.K., J.C.C., A.S.B., C.H., H.H.H., J.E. Methodology: D.H., D.S.M., M.C.O., A.C.-B., N.N.C., M.L.K., J.C.C., A.S.B., M.I.N., P.L. Project administration: J.K., S.I.R., M.I.N., M.N., A.S.B. Resources: J.S.Y. Supervision: J.K., S.I.R., M.I.N., M.N., A.S.B., J.S.Y. Validation: A.C.-B., M.I.N., A.S.B., J.C.C., M.L.K. Visualization: A.C.-B., M.I.N., N.N.C., M.C.O. Writing – original draft: A.C.-B. Writing – review & editing: All authors.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was primarily supported by the United States Department of Agriculture, Animal and Plant Health Inspection Service under agreement AP22VSSP0000C061. Dr. Nelson’s work was supported by Intramural Research Program of the US National Library of Medicine at the NIH and the Centers of Excellence for Influenza Research and Response, National Institute of Allergy and Infectious Diseases, National Institutes of Health (NIH), Department of Health and Human Services, under contract 75N93021C00014. The contributions of the USDA and NIH author(s) are considered Works of the United States Government. The findings and conclusions presented in this paper are those of the author(s) and do not necessarily reflect the views of the USDA, NIH, or U.S. Department of Health and Human Services.

Data availability

All complete SARS-CoV-2 genomic consensus sequences from this study are deposited in the NCBI GenBank database, with accession codes PQ594496 to PQ594788; metadata for these sequences are available in Supplementary Table 3. Sequence accessions used in this study can be accessed in the supplementary material; the GenBank accessions for human background sequences in Fig. 3 are shared with the Source Data, and identifiers for GISAID EPI_SETS listing all GISAID accessions are found in Supplementary Table 6. Multiple sequence alignments are not provided due to the use of sequences under GISAID data-sharing restrictions. Phylogenies and BEAST XML configurations are available on GitHub (https://github.com/acrespo-virevol/sars-cov2-US-mink-surveillance) and Zenodo (10.5281/zenodo.20820255)⁵⁵. Source data for the figures are provided with this paper.

Code availability

The custom Python script for masking sequence alignments and code for the mutational spectrum analysis are openly available on GitHub (https://github.com/acrespo-virevol/sars-cov2-US-mink-surveillance) and Zenodo (10.5281/zenodo.20820255)⁵⁵.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75363-4.