Detection of H5N1 in the air

During October to December of 2024, five dairy farms in the Central Valley of California identified as having HPAI H5N1 positive cows were enrolled in our initial air sampling studies (S1 Table and Fig 1A for general dairy farm layout). In the first phase of our study, we compared a variety of air sampling strategies during the active outbreak on dairy farms to define the optimal methodology that could be implemented in future sampling efforts. Aggregated milk from cows, referred to as bulk tank milk, was tested frequently for the presence of H5N1 viral RNA from late October 2024 until January 4, 2025 per federal recommendations [16]. The date that each farm tested positive and first observation of clinical signs are depicted in Fig 1B. H5N1 was detected in bulk tanks on farms prior to the observation of clinically ill cows. Ct values from bulk tank on the day of sampling performed in this study are shown in Fig 1C. Three different air sampling devices were tested in this initial phase (Fig 1D and 1E and Materials and methods): (1) open-face PTFE filter cassettes connected to an air sampling pump; (2) Sartorius MD8 Airport; and (3) AirPrep Cub 210. Open-face PTFE filter cassettes were worn on a backpack to model exposure to facility workers. A cone around the MD8 airport or open-face PTFE filter was used to minimize dilution of aerosols and droplets from the source (Fig 1D). On each farm, aerosols and droplets were collected from exhaled breath from individual cows or rows of cows, within milking parlors during milking, and within housing areas using a variety of these aerosol sampling devices. Seventy-one air samples were collected and analyzed for the presence of H5N1 viral RNA (Fig 1E and S2 Table). Six samples tested positive: five with the MD8 Airport sampler and one from an open face PTFE sampler worn during milking in the milking parlor (Fig 1E and S2 Table). Four positive samples were collected in the milking parlor and two, very weakly positive samples, were obtained from the exhaled breath of rows of 15–30 cows in head stanchions (Fig 1F). The portable nature of the MD8 allowed it collect air closer to the milking process (Fig 1D), which may account for the higher detection rate compared to the AirPrep Cub, which was placed stationary within the dairy parlor approximately 14 meters from the milking process. Based on these observations, further studies were conducted primarily using the MD8 Airport sampler on additional dairy farms.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Successful detection of H5N1 viral RNA in the air in milking parlors in the Central Valley of California using the MD8 Airport sampler.

(A) Top-down schematic of a representative dairy farm layout with multiple pens housed together under open-air structures, a separate hospital pen area, and a holding area where cows are moved prior to milking in the parlor. While the specific layout and structures could vary by farm, the housing pens were typically open-air structures with only a roof, allowing continuous exposure to sunlight and outside air. Holding pens were largely within the indoor dairy parlor structure, with some outdoor air penetrating into the space. The dairy parlor is predominantly an indoor space with air flow occurring through open windows with or without screens. Areas with asterisks denote spaces where air samples were collected. Arrows indicate the movement of cows from housing areas to holding pen, then to the milking parlor. Milk is aggregated into bulk tanks. ‘Sick’ cows, with clinical signs such as reduced appetite, fever, mastitis, or nasal discharge, were moved to a hospital pen and milked last into individual buckets; their milk was excluded from the bulk tank (for commercial supply) and entered the farm wastewater stream. Schematic created with BioRender.com (Campbell, A. (2026) https://BioRender.com/3gmitks). (B) Timeline of sampling during initial phase on dairy farms, including date the bulk tank tested positive, bulk tank status, onset of clinical signs in cows, and sampling days numbered 1–5. (C) Ct values for H5N1 viral RNA from bulk tank samples on each sampling day. (D) Photos depicting each type of air sampler. Left to right, open-face PTFE filters in use on the backpack while in a housing pen and milking parlor as indicated (yellow arrows); the MD8 Airport with a cone, to reduce dilution of the air, sampling the breath of a cow in a housing pen and during milking in the parlor; the AirPrep Cub (yellow circle) was placed either on the ground or was elevated to about six feet above the ground in milking parlors. (E) Summary of air sampling methods and H5N1 positive samples from the five dairy farms. (F) Ct values (average of duplicates) from the H5N1 positive air samples. Exhaled breath samples also collected using the MD8 Airport sampler.

https://doi.org/10.1371/journal.pbio.3003761.g001

Infectious aerosols in milking parlors

Based on the successful initial sampling phase, additional surveillance was performed on nine farms, with in-depth focus on three farms in southern California during their outbreaks in February and March 2025 (S1 and S4–S6 Tables and Fig 2A). On these farms, air samples from the milking parlor and exhaled breath of cows in housing pens were collected as previously described, using the MD8 Airport sampler fitted with a cone (Fig 1D). Sampling was conducted within days of each farm testing positive in the California state bulk tank testing program. Bulk tank milk samples from two of the three farms were collected on multiple days by research staff during the indicated time frame, highlighting the viral load of the herd during sampling (Fig 2B; farms EC, EG). The number of cows per herd was recorded for each farm (S1 Table), but the number of infected cows was unknown so sampling was targeted to milking of pens containing cows with known clinical signs.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Presence of infectious H5N1 virus in air samples from dairy farms in southern California.

(A) Timeline of sampling strategy for farms EB, EC, and EG. Light blue dots indicate milking parlor air samples, red dots indicate exhaled breath samples from cows, dark blue dots indicate water samples, and small gray dots indicate sampling dates. Detailed sampling information available in S4–S6 Tables. (B) H5N1 positivity in bulk tank milk samples over time from farms EC and EG as determined by viral RNA levels. (C) Viral genome concentration in the air (C_air) calculated as the genome copies (gc) per liter of air (L) from H5N1 positive air samples from the milking parlor (circles) or breath of cows (triangles) for farms EC (green, N = 18) or EG (purple, N = 4) over time. No template control samples, processed alongside surveillance samples, were run on every plate and were confirmed to have no amplification. (D) Infectious virus from all positive air samples (N = 22) from farms EC and EG was assessed by plaque assay. Colors are as indicated in (C), open circles denote samples that produced cytopathic effect and H5N1 positivity in TCID₅₀ assay but produced no plaques in a separate titration assay. Dotted black line indicates the limit of detection. For B–D, each dot is the average of technical duplicates.

https://doi.org/10.1371/journal.pbio.3003761.g002

H5N1 viral RNA was detected in 21 of 35 air samples (Fig 2C; S5 and S6 Tables). Samples are presented as genome copies per liter of air and range from 1 to 10,000. Exhaled breath was collected from rows of cows within the hospital pen on farm EG, and two of 10 samples tested weakly positive with 4–41 genome copies/L of air (C_air) (Fig 2C). For comparison, other studies of influenza A virus in air around livestock reported concentrations ranging from 7 to 440 gc/L in poultry barns or markets [17–21] and an average of 142–320 gc/L in swine barns [22,23]. Particle sizes generated in milking parlors were quantified using two methodologies, one to capture small aerosols (0.3–25 μm), and larger particles were measured by qualitative analysis of droplets collected on the MD8 gelatin filter (S1 Fig). These results reveal that both submicron aerosols and larger aerosols are likely produced during milking and could contain infectious virus. All positive samples collected on EC and EG were subject to viral titration, and infectious virus was detected in four samples with C_air values >1,000 collected with the MD8 Airport in a milking parlor (Fig 2D). Together, these results indicate that H5N1 viral RNA can be found in the air on farms and conclusively demonstrates that infectious H5N1 virus is present in the air during the milking process.

Detection of infectious H5N1 in farm wastewater sources

We next confirmed that both milking equipment and milk from sick cows on farms EC and EG harbored infectious H5N1 virus during the sampling period (S2 Fig). Since wastewater on farms could potentially be contaminated with H5N1 by these sources, we collected multiple water samples along the reclaimed water stream at each available point such as (1) water moving down the drain in the milking parlor, (2) at the sump pump adjacent to the dairy parlor, (3) fields where captured water is used, and (4) at manure lagoons where reclaimed water is stored (Fig 3A). H5N1 viral RNA was detected at each point of the waste stream, including in manure lagoons that are widely used by migratory birds and in fields with grazing cows (Fig 3B and S3–S6 Tables). Selected samples with an H5N1 RNA concentration exceeding 650 genome copies/mL were subject to viral titration, and two samples contained detectable infectious virus (Fig 3C). These results demonstrate that H5N1 is prevalent in reclaimed farm wastewater sites across dairy farms and may serve as another source of H5N1 spread between cows, to humans, and to peri-domestic animals.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Detection of H5N1 virus in multiple reclaimed wastewater sites in southern California and Central Valley farms.

(A) Top-down schematic of a representative dairy farm layout with farm wastewater sources depicted and on-site representative pictures included. Sources tested in the schematic match the colors in parts B and C. Schematic created with Biorender.com (Campbell, A. (2026) https://BioRender.com/3gmitks). (B) Water sources were screened for H5N1 viral RNA by digital droplet PCR and the data is plotted as a box plot. Each dot represents a single sample, solid horizontal lines represent median value, the lower edge is the first quartile, the upper edge is the third quartile, and max and min values are indicated with error bars. Details on location and date of each sample can be found in S3–S6 Tables. Briefly, milk cleanout (1) n = 7, sump pump (2) n = 7, field (3) n = 7, manure lagoon (4) n = 10. (C) Available samples with genome copies above 650 (indicated by dashed line, n = 6) were assessed for infectious H5N1 virus by either TCID₅₀ (circles) or plaque assay (square) and infectious titer from either assay is presented.

https://doi.org/10.1371/journal.pbio.3003761.g003

Presence of H5N1 viral variants in the air

Whole genome sequencing was performed on a subset of air samples and the reclaimed farm wastewater sample with highest genome copies from farm EC (S5 Table), and revealed all samples were within the H5N1 2.3.4.4b B3.13 subclade (Table 1). Variant analysis was conducted for each sample compared to a consensus sequence based on all samples on that farm. To improve the chances of successfully sequencing viral genomes from field samples, nucleic acids extracted from these samples were amplified using three different primer sets. Sequence coverage was variable across the genome and among samples (S3 Fig), and we report only variants that were present in all consensus sequences for a given sample that had coverage at a given locus with at least one consensus sequence derived from a coverage depth greater than 24 reads. Amino acid variants were observed in many samples compared to the farm consensus sequence. Air samples collected from the milking parlor during milking of the same pen of cows on subsequent days revealed variations in the HA, PB2, and PB1 segments (Table 1, compare EC36 to EC47 and EC60). Of note, HA mutations observed in EC36 identified a change at HA position 189 (193 H3 numbering), which is found within the 190-helix, a critical component of the receptor binding site [24,25]. While position 189 does not directly bind to α2,3-linked sialic acids [26,27], mutations at this site are associated with improved binding to human-like α2,6-linked sialic acid [28]. This observation suggests that viral variants with changes in the HA receptor binding site can emerge on farms. Variants were also observed between the air samples collected during milking of sick versus healthy cows on the same day (Table 1, compare EC47 to EC54 and EC60 to EC66). Sequences from an air sample collected in the milking parlor on March 4, 2025 closely matched the sequence observed in the water sample from the same farm on the same day (Tables 1 and S5, samples EC22 and EC27). Taken together, analysis of air and water samples highlight that there is genetic variation on a farm and more detailed analysis of sequences could decipher intra-herd transmission patterns.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Consensus variants identified in environmental samples from dairy farm EC.

https://doi.org/10.1371/journal.pbio.3003761.t001

Presence of H5N1 antibodies in subclinical cows

The lack of individual cattle testing prevents detection of subclinical infections. To explore whether H5N1 positivity could be predicted based on a subset of clinical signs, daily milk weight production data was used on Farm FB to bracket cows into three groups: (1) animals that suddenly went dry, likely from H5N1 infection, and were sent to slaughter for beef; (2) those that dropped milk weight and then recovered; and (3) animals without any drop in milk weight (Fig 4A). In April, only cows in groups 2 and 3 were present on farm FB. To assess whether H5N1 infection was observed in subclinical cows, the levels of IgG antibodies targeting H5N1 HA were assessed in expressed milk collected from four recovered cows and 10 animals without any drop in milk weight (Fig 4B). All four recovered animals had detectable H5 antibodies, and six of 10 subclinical cows had detectable H5 antibodies, albeit at a lower level than those observed in the recovered cows. These data suggest that subclinical cows have been exposed to sufficient virus antigen to produce a mucosal immune response within the mammary glands without manifestation of clinical disease.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Presence of anti-H5 HA antibodies in expressed milk from cows with subclinical disease.

(A) A biometric monitoring system on farm FB allowed for assessment of the daily milk weight produced by cows prior to H5 detection in the bulk tank and after. A subset of cows had a significant drop in milk weight that resulted in removal from the herd (red), a subset dropped milk weight and then recovered (black), and finally a subset of animals had less than a 10% milk weight drop (subclinical, blue). Each dot represents data from an individual animal. (B) In April 2025, expressed milk was collected from a subset of cows in the recovered and subclinical groups from part A. Anti-H5 HA IgG antibodies were assessed by ELISA and the endpoint titer is shown for 4 cows in the recovered (black group) and 10 cows in the subclinical group (blue).

https://doi.org/10.1371/journal.pbio.3003761.g004

H5N1 positivity in individual quarters is heterogenous within a given herd

Contaminated milking equipment is thought to be a primary mode of H5N1 cow-to-cow transmission [9] though definitive proof from natural or experimental infection scenarios is still lacking. There is a specific orientation of the milking equipment when placed on the cow udder such that each teat cup is almost always placed on the same teat or quarter of each cow. Thus, similar quarters might be predicted to be infected throughout a herd given these constraints if equipment was contaminated. To begin to examine the distribution of H5N1 infection in cow quarters, we initiated a longitudinal study on farm EC (Fig 5A). Twenty-three cows were screened for H5N1 positivity; of these, 14 cows were selected for longitudinal sampling, including three cows that screened negative. Milk samples from each individual quarter: left front (LF), right front (RF), left rear (LR), and right rear (RR) were collected by a dairy farm worker prior to milking daily over the course of seven days (Fig 5B).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Longitudinal sampling of individual quarters on cows from farm EC.

Farm EC tested positive on 2/24/2025. (A) Timeline of screening and longitudinal sampling (for 7 days, starting 3/20 or 3/21) of individual cows. (B) A worm-view schematic identifying the location of each quarter on the udder: right front [RF], right rear [RR], left rear [LR], and left front [LF]. Schematic created with BioRender.com (Campbell, A. (2026) https://BioRender.com/3gmitks). (C) Number of cows from the longitudinal study with the indicated number of quarters positive for H5N1 viral RNA. Quarters only positive on a single testing day were not included in C, as a single positive could be due to contamination during collection.

https://doi.org/10.1371/journal.pbio.3003761.g005

Analysis of individual cows over this timeframe revealed several important features of natural H5N1 infection progression. First, the number of H5N1 positive quarters per animal varied widely, with 5/12 cows having only one quarter positive and the remaining cows having 2–4 quarters positive (Figs 5C and 6). Second, the quarters that tested positive on the first day of sampling generally remained positive throughout the sampling period (Fig 6). This quarter-specific conservation of H5N1 positivity over time is consistent with observations from experimental infections where H5N1 virus remains within the infected quarters and does not spread to uninfected quarters [12,13,15]. Third, there was little variation in the viral RNA amount over the 7-day collection period. The absence of H5 viral RNA in specific quarters from infected animals is interesting and may represent quarters that could become infected later during the outbreak, although additional data is needed to confirm this.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Heterogeneity in H5N1 positive quarters among naturally infected cows.

H5N1 viral RNA was measured in milk from individual quarters of 14 cows; 12 PCR-positive cows are shown here, including one cow that had originally screened negative (#7814). Ct values averaged from technical duplicates are shown over time for each individual quarter (LF, LR, RR, RF). Yellow stars represent Ct values detected during screening of milk pooled from all 4 quarters of each cow. Cows within the dashed box indicate those that were treated for mastitis during sampling, such that their milk was not put into the bulk tank. Not all cows were available to be tested on all dates. Three cows were sent to slaughter during the sampling efforts: cows #7417, 7401, and 8264.

https://doi.org/10.1371/journal.pbio.3003761.g006

H5N1 in dairy cows was identified through investigation of mastitis of unknown origin [2]. Although mastitis (inflammation of the mammary gland that can cause abnormal consistency and discoloration of milk) is often associated with bacterial infections, it has been identified as a common clinical sign in H5N1-infected cows [29]. Since mastitis is usually identified in a single quarter within affected cows [30,31], it is likely that the same quarter would also be positive for H5N1. However, our screening results indicated that mastitis does not always overlap with H5N1 as seven cows were identified as H5N1 positive but lacking clinical signs of mastitis (Fig 7A). We also examined the distribution of mastitis by quarter in a set of cows on farm EC a few weeks after initiation of the H5N1 outbreak but prior to the longitudinal sampling. Within these cows, there was a predominance of mastitis in the RF quarter, with one animal having mastitis in two quarters (RF and RR) (Fig 7B). However, this pattern was not observed in the cows from our longitudinal study, where most cows had multiple quarters infected and there was a fairly even distribution of infection among the four quarters (Fig 7C). Although only a small number of cows were sampled, these results suggest that H5N1 positivity can occur in quarters that do not present with mastitis, further supporting the idea suggested by the antibody results (Fig 4) that nonclinical animals may be H5N1 positive.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. H5N1 infection does not always correlate with clinical mastitis.

(A) Cows screened for the longitudinal study described in Fig 5, were grouped by whether or not they were treated for mastitis. Each dot represents the Ct value of H5N1 RNA from pooled milk from all four quarters of an individual cow, categorized as either having had mastitis (red, N = 6) or being randomly selected (yellow, N = 18), and is the average of technical duplicates. Dotted line represents limit of detection. (B) Nine cows on farm EC were noted as having mastitis by milking parlor staff and the affected quarter was recorded by research personnel. These data were collected prior to screening of cows in part A and do not represent the same cows presented in part A or C. The asterisks denote that the cow with mastitis in quarter RR was also positive in RF and was the only cow to present with mastitis in 2 quarters on the farm. (C) The 12 cows from Fig 6 were assessed for H5N1 positivity in each quarter. Quarters only positive on a single testing day were not included in C, as a single positive could be due to contamination during collection.

https://doi.org/10.1371/journal.pbio.3003761.g007

Ethics statement

Work performed by Emory researchers was deemed to be exempt by the Emory University Institutional Animal Care and Use Committee. Work performed by Dr. Lombard’s team was deemed to be exempt by the Colorado State University Institutional Animal Care and Use Committee. All milk samples, including individual quarter and composite samples, were collected by farm workers during normal farm procedures and passed to research staff. Analysis of infectious viral titer was conducted in a BSL3 laboratory at Emory University under protocols approved by the Emory Institutional Biosafety Committee.

Dairy farm details

Initial sampling was conducted on five dairy farms in the Central Valley of California from October through December 2024; then subsequently on seven farms in southern California, and on two additional farms in the Central Valley from February to April 2025. Each of these dairy farms had confirmed H5N1 infections in cows as determined by screening of on-dairy bulk tank milk samples in accordance with the National Milk Testing Strategy Federal Order from December 2024. Bulk tank samples taken by our group were collected using either a clean metal dipper, a spigot in the tank, or aseptically with the needle-only QualiTru system into 50mL conical tubes, depending upon the setup of each dairy. The sizes and infrastructure of surveyed dairies varied and are detailed in S1 Table. Determinations of clinical signs and mastitis were made by dairy farm owners, workers, and/or veterinarians, depending on each dairy.

Air sampling devices and processing

Air samples were collected at sources including exhaled breath of individual cows and rows of cows in housing pens (15–30 cows per filter), within milking parlors during milking, and sites where wastewater is exposed to the environment such as open sump pumps, manure lagoons, and fields. The air samplers used in this study are described below. Details of sampling for each source on each farm can be found in S2–S6 Tables.

Surface swab collection and processing

The interior surfaces of milking liners as well as their exterior shells were sampled with pre-assembled swab kits (Becton Dickinson Ref. 220526). For milking unit shells, the flocked swab was unwrapped, pre-moistened by briefly dipping into the included tube of viral transport media (VTM), pressed and dragged over the surface, sealed into the tubes, and bagged for subsequent processing. Pre-assembled swab tubes (BD Ref. 220526) containing surface samples were processed using the same protocol at LVC and Emory University. Tubes were vortexed for 10 s with the flocked swab inside, then centrifuged for roughly 10 s at 1,000 rpm to collect all liquid at the bottom of the tube. An aliquot was then removed for RNA extraction, and the remaining sample was stored at 4 °C (LVC) or −20 °C (Emory).

Wastewater sample collection and processing

Wastewater samples were collected from multiple locations on each farm (see S3–S6 Tables). Whenever possible, wastewater from the beginning of the milk line cleanout was collected at the end of each milking shift. Samples were also collected from sump pumps, manure lagoons, and/or fields whenever possible. Wastewater samples were collected aseptically. When waste flows were within reach, the external sides of 50 mL falcon tubes or 250 mL and 1 L bottles were first sterilized with 10% bleach and then lowered into the water stream until tubes were filled. When waste flows were out of reach, a 25 ft pole with a bottle adapter (VWR) and 1 L bottles were used for sample collection. The base of the pole, bottle adapter, and a zip tie were sterilized by submerging them in 10% bleach for 5 min before sample collection. The outside of the bottle was also wiped with 10% bleach to sterilize and was then attached to the adapter with the zip tie. The sample was collected by dipping the bottle into the wastewater until full; full bottles were wiped with 10% bleach again and maintained at 4 °C from time of collection until processing at Emory University.

Quantifying H5N1 viral RNA

Infectious virus titration

Samples were tested for infectious virus within Emory BSL3 laboratory with approved protocols. Samples were titrated on confluent (MDCK) epithelial cells (ATCC) by plaque assay as previously described [40] or tissue culture infectious dose 50 (TCID₅₀) assay using the Reed and Muench method [41]. Any air samples positive for H5N1 viral RNA determined by RT-qPCR were assayed for presence of infectious virus as described above. All positive milk samples from cows in the longitudinal study were tested for infectivity, as well as positive milk samples collected from all four quarters of individual cows. All four positive surface swabs of milking unit shells as well as positive swabs of milking inflation liners were assayed for infectivity. Six positive wastewater samples were selected for infectivity testing with both TCID₅₀ and plaque assay based on having genome copy per milliliter concentrations above 650 and location of sample collection. All titration studies were done with infection media (1× Minimal Essential Media, Corning Ref. 15-010-CV; with 4mM L-Glutamine, Sigma Aldrich Ref. G7513; and 1:1,000 TPCK Trypsin, Worthington Biochemical Ref. LS003750) containing double the concentration of antibiotics and antimycotics (4× final concentration of Antibiotic-Antimycotic 100×, Gibco ref. 15240062) to reduce bacterial and fungal contaminants in the environmental samples. In many cases both assay types were used to determine infectious virus titration, since TCID₅₀ method allows for complementation of semi-infectious virus particles. The highest virus titer for a given sample is reported. In cases with ambiguous cytopathic effect, additional assays were performed to verify positivity, including agglutination of chicken red blood cells with cell supernatant and/or serial passage of 20 µL of the cell supernatant onto fresh MDCK cells to assess for presence of virus as compared to other environmental contaminants. In addition, supernatants from air samples that were positive were RNA extracted for sequencing efforts to identify viral variants.

Calculation of virus concentration per liter of air

All RT-qPCR plates processed at Emory University contained a dilution series of an in vitro transcribed RNA product of known copy number, allowing generation of a standard curve. This was used for absolute quantification of the genome copy numbers present in each reaction and per milliliter. The genome copy numbers per milliliter were then used to estimate the number of genome copies present per liter of air; the volume of air sampled, in liters: V_air = air flow rate (LPM) × sample duration (minutes). The virus concentration per liter of air was calculated as:

Estimation of aerosolized droplet sizes deposited on MD8 Airport filters

Photos were taken of 14 MD8 Airport gelatin filters immediately after air sampling. These photos were analyzed to determine the size of aerosolized droplets deposited on the filters using ImageJ. Photos were cropped to include only the filter surface area, then rescaled to 800 × 800 pixels to standardize and allow size conversion from pixels to millimeters since MD8 gelatin filters have 80 mm diameter. To detect spots, a Gaussian blur (σ = 2 pixels) was applied to reduce noise with the “Find Maxima” function to generate a mask from the detected maxima. The Regions of Interest (ROIs) were selected with a circularity filter applied (>0.5) to select only spherical spots and the ROIs were restored onto the original unmasked image for accurate pixel measurement. All detected ROIs were manually inspected to remove false positives and artifacts from maxima detection. The ROI measurement function was used to calculate the number of spots and their area (pixel²), and the data was exported as a.csv file for processing through a custom Python script (available at FigShare, https://doi.org/10.6084/m9.figshare.29627840) to convert spots from pixels² to mm² and calculate spot volumes.

Whole genome sequencing analysis

Whole genome sequencing was performed on nine air samples and one wastewater sample collected from farm EC (Table 1). For five of the air samples, a cell-passaged viral population was sequenced in addition to the original sample. For the wastewater sample, only a cell-passaged viral population was sequenced. Nucleic acids were extracted using the NucleoMag RNA kit (Machery-Nagel) with an epMotion 5075 (Eppendorf). The 8 genomic segments of IAV were amplified in a one-step multiplex RT-PCR [42]. For each specimen, extracted RNA, was amplified with each of three different primer sets: Uni12/Inf1, Uni12/Inf3, Uni13/Inf1 [42]; MBTuni-12, MBTuni-12.4, MBTuni-13 [43]; OptiF1, OptiF2, OptiR [44]. cDNA samples were prepared for sequencing using the Oxford Nanopore Technologies (ONT) platform. Unique barcodes were ligated to RT-PCR products from each sample using end repair and ligation kits (New England Biolabs) and then pooled. These pools were ligated to pore adaptors and sequenced with a GridION instrument. Sequence reads were aligned to the nearest IAV reference and sample consensus sequences were generated using IRMA [45]. Influenza virus clades were assigned based on hemagglutinin sequences using Nextclade [46]. The sample consensus sequences from each primer set were aligned using MUSCLE in MegAlign Pro (DNA Star). The overall consensus sequence from all samples sequenced in this study was used as a reference to identify variants in the individual sample consensus sequences from each primer set. Only mutations found in all of a sample’s consensus sequences that had read coverage at that genetic locus are reported. All reported mutations were identified at loci with 24-fold or greater coverage depth from at least one of the primer sets.

Longitudinal study of infected cows

Longitudinal sampling was conducted on dairy farm EC from 3/18/25 to 3/27/25. Cows were screened for H5N1 virus infection on 3/18–3/19 based on being in treatment for mastitis (as determined by dairy farm parlor workers) or were randomly selected during milking. Milk from cows on mastitis treatment doesn’t go into the bulk tank. A composite sample of milk was collected from all four quarters and shipped overnight at 4 °C to Emory University for RNA extraction and RT-qPCR as described above. Screening results informed the selection of cows for inclusion in the study group; selected cows were marked on right and left flanks to enable identification in pens and the milking parlor, and identifications were confirmed with cows’ ear tag numbers. Eleven H5-positive cows and three H5-negative cows were selected for the study group. One cow (7,814) was selected with a negative sample at screening, but on every subsequent study day tested positive for H5N1. Daily, from 3/20/25 to 3/27/25, a sample of milk from each quarter of the selected cows was collected into separate 50 mL conical tubes and stored at 4 °C until shipping and processing at Emory University for viral RNA extraction, RT-qPCR, and/or virus titering as detailed above. Some cows from this study were not available on all testing days. Sampling of the exhaled breath of all available cows on study on a single MD8 Airport gelatin filter was done on 3/19, 3/22, and 3/24. All air filter samples were stored at 4 °C until they could be processed at Emory University and stored at −20 °C prior to RNA extraction and RT-qPCR as described above.

Antibody detection in cows post-H5N1 outbreak