The abundant expression of sialic acid receptors on human respiratory tract epithelia makes this tissue an attractive site of influenza A virus (IAV) replication, with respiratory disease being a frequent clinical presentation of infection. However, human ocular tissue (notably corneal and conjunctival epithelial cells) also bear sialic acid receptors and can support the replication of multiple IAV subtypes,¹ highlighting the need to study the capacity of IAVs to replicate in non-respiratory tissues and infect humans via non-respiratory routes. Ocular complications (primarily conjunctivitis) have been reported following human infection with seasonal, avian-origin, and swine-origin IAVs,² with or without known ocular exposure to the virus. Because ocular and respiratory systems are connected by the nasolacrimal duct, fluidics exchange between these sites occurs, and renders ocular tissues susceptible to both primary IAV infection following ocular exposure, and secondary infection following the establishment of a respiratory tract infection.¹ Furthermore, expired air from individuals with IAV infection could expose their ocular surface to infectious virus via convective flow.³ Although the properties governing IAV ocular tropism are not well understood, in-vitro studies support that ocular-tropic IAVs—notably those within the A(H7) subtype—elicit differential induction of cytokines and signalling pathways in human ocular cells compared with respiratory-tropic IAVs.^4,5

Outbreaks of clade 2.3.4.4b highly pathogenic avian influenza (HPAI) A(H5N1) virus in cattle, poultry, and other species in the USA since March, 2024 have resulted in multiple cases of human infection.⁶ Notably, many of these infections have been associated with conjunctivitis, with or without concurrent respiratory illness.^7,8 It is unclear whether virus isolates linked to outbreaks on dairy farms in 2024 possess an enhanced ability to use ocular tissue for virus entry or replication, or both, compared with other clade 2.3.4.4b isolates or other IAV subtypes, or whether the unique conditions associated with these outbreaks are facilitating conjunctival presentation in the absence of an ocular tropism.

Detection of higher viral RNA levels in a conjunctival specimen than in nasal specimens from the first confirmed human case from this outbreak⁷ could suggest that ocular tissues were initially exposed to the virus, that the virus was capable of enhanced replication in ocular tissues, or both. Reports of ocular disease and detection of viral RNA from ocular specimens of A(H5N1) virus-infected cats and cows on affected farms⁹ further supports the capacity of these viruses to replicate in mammalian ocular tissue.

In this study, we assessed the ability of A/Texas/37/2024 (Texas/37) and A/Michigan/90/2024 (MI/90) A(H5N1) viruses, isolated from conjunctival specimens from the first two confirmed infections in dairy farm workers in Texas and Michigan,^7,8 to replicate and elicit antiviral host responses in human corneal tissue constructs, compared with contemporary human and avian IAVs. Ferrets were used to assess pathogenicity, transmissibility, and tropism of Texas/37 virus (previously shown to cause fatal infection in serologically naive ferrets inoculated by the intranasal route with a high dose of the virus^10,11) following exposure of the respiratory tract or ocular surface.

We infected primary human nasal (EpiNasal) and corneal (EpiCorneal) tissue constructs with Texas/37, Neb/14, or NL/219 viruses. Although all viruses exhibited productive replication, no significant differences in total area under the curve were observed across all three viruses in either EpiNasal (figure 1A) or EpiCorneal (figure 1B) constructs (appendix p 4). EpiCorneal constructs infected with the full panel of viruses showed that Neb/14, NL/219, and NL/230 viruses all replicated faster (24 h titre) and to higher mean peak titres (4·2–4·7 log₁₀ PFU/mL) than a panel of 2.3.4.4b HPAI A(H5N1) viruses comprising Texas/37, MI/90, and Chile/25945 (3·2–3·8 log₁₀ PFU/mL; figure 1C). NL/219 and NL/230 viruses, and Texas/37 and MI/90 viruses, each differ between each other by 13 amino acids, including at amino acid position 627 in PB2. NL/219 and Texas/37 both have a lysine at position 627, an adaptive mutation associated with overcoming host range of avian influenza viruses.^7,8,12 Despite different disease profiles in humans, replication of the 2.3.4.4b A(H5N1) viruses Texas/37 and Chile/25945 was indistinguishable in EpiCorneal constructs (figure 1C).

All viruses exhibited productive replication (mean peak titre per virus 3·7–5·7 log₁₀ PFU/mL) in EpiCorneal constructs before collection of total RNA for inflammatory or antiviral response assessment, with no significant difference in mean titres among viruses of the same subtype. Only 13 genes were significantly upregulated or down-regulated by more than two-fold in virus-infected constructs compared with mock-infected constructs among all IAVs tested (all with fold regulation <10; figure 2A). All but one of these genes (in Chile/25945-infected constructs) were detected in constructs infected with either NL/230 or Neb/14 viruses; none of these were detected in constructs infected with Texas/37 virus (figure 2A). However, infection with MI/90 virus induced significantly higher levels (more than two-fold upregulation) of numerous Toll-like receptor (TLR) signalling, type 1 interferon (IFN) signalling, and IFN responsive genes than following infection with the Texas/37 virus (figure 2B).

In in-vivo experiments, ferrets that were serologically naive to IAV prior to inoculation exhibited clinical disease following inoculation with a high viral dose (10⁶ PFU) by the liquid ocular route, with elevated temperatures, weight loss, and presentation of clinical signs such as diarrhoea and nasal discharge detected in all animals (table). All inoculated animals reached humane endpoints 3 days after inoculation. Furthermore, similar to intranasally inoculated animals receiving the same viral dose,¹⁰ infectious virus was recovered from all tissues collected, including high-titre detection (>10⁴ PFU/ml) in the eyes and conjunctiva (figure 3A).

Ferrets inoculated by the aerosol inhalation route with a low dose of virus (15–16 PFU) displayed a delay in viral shedding in the upper respiratory tract (compared with ferrets receiving a high viral dose by the liquid ocular route; figure 4A, B), but they developed severe disease indistinguishable from that following the high-dose liquid ocular challenge described above, warranting humane euthanasia 3 days after inoculation (table). Post-mortem necropsy revealed elevated viral loads throughout the respiratory tract and numerous extrapulmonary sites (notably the liver, spleen, and intestine; figure 3B). Infectious virus was only infrequently detected in ocular (eyes and conjunctiva) and brain (cerebrum and olfactory bulb) tissues at low titre (<10² PFU/mL or PFU/g). Animals inoculated by the low-dose aerosol inhalation route did not transmit Texas/37 virus to contact animals in an RDT setting (table), probably due to a delay in productive viral replication in the upper respiratory tract resulting in a shortened interval of virus shedding into the environment before infected animals reached humane endpoints (figure 4B), compared with previous ferret experiments using a high challenge dose of virus.¹⁰ For this reason, subsequent transmission assessments were performed using a more permissive direct contact transmission model.

Ferrets exposed to aerosolised Texas/37 virus by the ocular-only route (52–132 PFU) became productively infected, resulting in clinical disease (table) and viral shedding in the upper respiratory tract (figure 4C), which was indistinguishable from animals inoculated by the respiratory inhalation route. All animals infected by this inoculation route reached euthanasia criteria 4 days after inoculation. The magnitude and distribution of viral detection in systemic tissues from these animals was similar to ferrets inoculated by the aerosol inhalation route (figure 3C), with ocular and brain tissues again exhibiting markedly reduced viral titres compared with respiratory tract specimens and other extrapulmonary sites. Using the DCT setting, contact animals who were cohoused with ferrets inoculated by the ocular-only aerosol route (n=3) all became infected, reaching humane endpoints 4–6 days post-contact (table, figure 4C).

Pre-immune ferrets rechallenged by the low-dose ocular-only route with Texas/37 virus (18–44 PFU) had fewer clinical signs of illness and showed reduced frequency and magnitude of infectious virus detection in nasal wash, conjunctival wash, and rectal swab specimens, compared with similarly exposed serologically naive animals (table, figure 4D). All pre-immune animals survived the challenge and seroconverted to Texas/37 virus. Serologically naive ferrets placed in direct contact with these animals did not become productively infected (table, figure 4D).

We found that A(H5N1) viruses isolated from the first two confirmed infections in dairy farm workers with conjunctivitis did not exhibit an enhanced capacity to replicate in human ocular tissue compared with an A(H5N1) isolate from a severe respiratory human infection or a A(H1N1) pdm09 virus. Ocular-only aerosol challenge with Texas/37 virus resulted in infection of ferrets with or without preexisting IAV immunity but caused severe and fatal illness only in serologically naive animals. This work suggests that 2.3.4.4b A(H5N1) viruses currently causing mammalian and poultry outbreaks in the USA do not possess a marked ability relative to other IAVs to infect ocular epithelia, but are nonetheless capable of mounting a productive, transmissible, and fatal infection in this mammalian model following low-dose ocular exposure.

Human corneal and conjunctival epithelial cells primarily express α2-3 linked sialic acid (ie, sialic acid linked to galactose via α2,3 glycosidic bonds), although α2-6 linked sialic (ie, sialic acid linked to galactose via α2,6 glycosidic bonds) can also be detected and are present in other tissues (including the nasolacrimal duct) in the ocular milieu. ^1,19 Despite strong binding of Texas/37 virus to α2,3-linked sialic acid,¹⁰ all A(H5N1) viruses replicated to lower overall peak titre than either A(H7N7) or A(H1N1)pdm09 viruses in EpiCorneal constructs. In agreement with previous work,⁵ relatively few genes deviated significantly from mock-infected EpiCorneal constructs when tested by PCR array. Furthermore, generally similar induction of innate antiviral responses among all A(H5N1) viruses examined 24 h after infection suggests that Texas/37 or MI/90 viruses do not possess a particular affinity to replicate or elicit robust host responses within ocular tissue compared with other IAVs, although additional investigation of later timepoints (eg, 48–72 h) is warranted. IAV infection of human ocular cells can lead to differential production of pro-inflammatory cytokines and chemokines.^4,5 Although changes in RNA transcript levels following IAV infection were generally within five-fold of mock-infected cells (figure 2A), further examination of the capacity of A(H5N1) viruses that have continued to emerge throughout 2024 to produce pro-inflammatory mediators in human corneal and conjunctival cells is still needed, as is a greater understanding of the host immune responses that govern IAV-associated conjunctivitis. The high titres observed with Neb/14 virus in EpiCorneal cells are in agreement with previous work supporting the capacity of some A(H1N1)pdm09 viruses to use ocular tissue to mount a productive infection in-vitro.^19,20 The eye represents an immune-privileged site, and caution should be applied in extrapolating results from this tissue to other anatomical locations. Ongoing assessments of host responses to clade 2.3.4.4b viruses in both human and bovine respiratory tract epithelial cells could reveal further insight into unique properties of these viruses compared with historical A(H5N1) strains.^21,22 Further investigation of the capacity of clade 2.3.4.4b A(H5N1) viruses to maintain infectivity during passage through human tear film is also warranted.¹⁹ Studies using different in-vitro and ex-vivo systems are needed to better emulate the complexity of both human corneal and conjunctival tissues.^4,11,20 Furthermore, including additional readouts (eg, immunohistochemistry) could help explore possible differences in viral infectivity during the first round of virus replication in these cells.¹⁹

Clade 2.3.4.4b HPAI A(H5N1) viruses isolated from both humans and animals since 2022 have shown a remarkable capacity to cause severe disease in mammals.^10,23–25 However, studies of these strains have rarely used inoculation doses of less than 10⁶ infectious units.²³ Although a high-dose liquid challenge can reveal the full capacity of an IAV to cause severe, systemic infection, low-dose aerosol-based challenge methods can better emulate infection progression without compromising virulence readouts.¹⁶ For these reasons, we conducted ferret ocular challenge studies using both liquid and aerosol inoculation procedures, and a range of inoculum doses. Although the sample size per group (n=3) in our study was small, the strongest conclusions drawn from pathotyping and transmission are informed by results that span multiple independent experiments.¹⁸ Of note, ferrets do not consistently present with macroscopic conjunctivitis following ocular exposure or infection with IAV, restricting our assessments to detection of infectious virus in ocular specimens. Unlike a high-dose ocular challenge, ferrets inoculated via inhalation or ocular-only aerosol-based challenges did not have high viral titres in the brain or olfactory bulb, suggesting that the virulent phenotype observed was not due to neurological invasion of the virus. In agreement with previous work showing that ocular inoculation does not select for mutations compared with intranasal routes,²⁶ sequencing of specimens from ferrets inoculated by the high-dose liquid ocular route revealed no consistent amino acid variants compared with ferrets inoculated by the respiratory inhalation route. Furthermore, despite detection of Texas/37 virus in the eyes, conjunctiva, and conjunctival wash specimens of ferrets inoculated with a high dose of virus,¹⁰ low-dose aerosol inoculation supports that this virus might not be detected at high titres in ocular tissue in ferrets once productive infection is established. Although additional research is warranted to better understand how Texas/37 virus so rapidly spreads systemically in serologically naive animals after inoculation, these data support the utility of aerosol inoculation approaches for these investigations. Abrogation of Texas/37 virus transmissibility by the airborne route in ferrets inoculated by low-dose aerosol inhalation (compared with those who received the high-dose challenge)¹⁰ further supports the utility of these methodologies to risk assessment activities.

The capacity of seasonal and zoonotic IAVs to cause productive infection in ferrets following ocular inoculation has been shown previously,² including fatal disease reported in ferrets following ocular challenge with the 2.3.4.4b A(H5N1) virus Chile/25945.²⁷ We extend these findings to show that, compared with serologically naive animals, ferrets with pre-existing heterosubtypic immunity to A(H1N1)pdm09 exhibit less severe disease following ocular virus exposure. Ferrets inoculated by the ocular route can release viral-laden aerosols into the environment after inoculation.²⁶ However, further study is needed to assess the effect that pre-existing immunity to IAV has on subsequent shedding of infectious virus into the environment, compared with serologically naive animals. Although our data, which are consistent with other work showing reduced virulence of a 2.3.4.4b A(H5N1) bovine isolate following intranasal inoculation of similarly pre-immune ferrets,^28,29 provide a potential explanation for why human cases have been generally mild, the complexity of innate and adaptive immune responses governing this phenotypic difference requires dedicated study.³⁰

This study has several limitations. For the in-vitro work, host responses were limited to PCR-based detection methods and did not include protein quantification of secreted cytokines. Aerosol exposure-based methods can result in variability in presented doses between animals that is not observed when using a liquid inoculum.¹⁶ For the in-vivo work, although multiple inoculation routes were assessed, each individual experiment had a small sample size. The reduced viral shedding by virus-challenged pre-immune ferrets further limited interpretation of the ways in which pre-existing immunity might modulate viral transmission to susceptible contact animals.

Despite the general absence of an ocular tropism observed for Texas/37 virus in ferrets in this study, there has been a high incidence of ocular complications among humans with confirmed infection. Therefore, additional study is needed to ascertain whether this high incidence is due to specific virological features of currently circulating 2.3.4.4b A(H5N1) viruses, or whether the unique conditions associated with human exposure activities during this outbreak (including milk splashing on the ocular surface,⁸ or self-inoculation of the ocular surface with hands exposed to contaminated milking equipment, or both²⁹) represent a contributing factor. Of note, aerosol-based experiments were conducted at elevated (approximately 65%) relative humidity to better emulate environmental conditions in dairy farms,³¹ but further experimentation (eg, assessing IAV infectivity in aerosols in milk) could be informative. Collectively, this work supports that the human ocular system represents a secondary mucosal surface susceptible to IAV exposure and infection, and highlights the importance of following recommendations for both ocular and respiratory protection when working with farm animals in locations with potential A(H5N1) detection.⁶