Mouse Saliva Inhibits Transit of Influenza Virus to the Lower Respiratory Tract by Efficiently Blocking Influenza Virus Neuraminidase Activity.

We previously identified a novel inhibitor of influenza virus in mouse saliva that halts the progression of susceptible viruses from the upper to the lower respiratory tract of mice in vivo and neutralizes viral infectivity in MDCK cells. Here, we investigated the viral target of the salivary inhibitor by using reverse genetics to create hybrid viruses with some surface proteins derived from an inhibitor-sensitive strain and others from an inhibitor-resistant strain. These viruses demonstrated that the origin of the viral neuraminidase (NA), but not the hemagglutinin or matrix protein, was the determinant of susceptibility to the inhibitor. Comparison of the NA sequences of a panel of H3N2 viruses with differing sensitivities to the salivary inhibitor revealed that surface residues 368 to 370 (N2 numbering) outside the active site played a key role in resistance. Resistant viruses contained an EDS motif at this location, and mutation to either EES or KDS, found in highly susceptible strains, significantly increased in vitro susceptibility to the inhibitor and reduced the ability of the virus to progress to the lungs when the viral inoculum was initially confined to the upper respiratory tract. In the presence of saliva, viral strains with a susceptible NA could not be efficiently released from the surfaces of infected MDCK cells and had reduced enzymatic activity based on their ability to cleave substrate in vitro This work indicates that the mouse has evolved an innate inhibitor similar in function, though not in mechanism, to what humans have created synthetically as an antiviral drug for influenza virus.IMPORTANCE Despite widespread use of experimental pulmonary infection of the laboratory mouse to study influenza virus infection and pathogenesis, to our knowledge, mice do not naturally succumb to influenza. Here, we show that mice produce their own natural form of neuraminidase inhibitor in saliva that stops the virus from reaching the lungs, providing a possible mechanism through which the species may not experience severe influenza virus infection in the wild. We show that the murine salivary inhibitor targets the outer surface of the influenza virus neuraminidase, possibly occluding entry to the enzymatic site rather than binding within the active site like commercially available neuraminidase inhibitors. This knowledge sheds light on how the natural inhibitors of particular species combat infection.

Salivary Blockade Protects the Lower Respiratory Tract of Mice from Lethal Influenza Virus Infection.

It is possible to model the progression of influenza virus from the upper respiratory tract to the lower respiratory tract in the mouse using viral inoculum delivered in a restricted manner to the nose. In this model, infection with the A/Udorn/307/72 (Udorn) strain of virus results ultimately in high viral titers in both the trachea and lungs. In contrast, the A/Puerto Rico/8/34 (PR8) strain causes an infection that is almost entirely limited to the nasal passages. The factors that govern the progression of virus down the respiratory tract are not well understood. Here, we show that, while PR8 virus grows to high titers in the nose, an inhibitor present in the saliva blocks further progression of infection to the trachea and lungs and renders an otherwise lethal dose of virus completely asymptomatic. In vitro, the salivary inhibitor was capable of potent neutralization of PR8 virus and an additional 20 strains of type A virus and two type B strains that were tested. The exceptions were Udorn virus and the closely related H3N2 strains A/Port Chalmers/1/73 and A/Victoria/3/75. Characterization of the salivary inhibitor showed it to be independent of sialic acid and other carbohydrates for its function. This and other biochemical properties, together with its virus strain specificity and in vivo function, indicate that the mouse salivary inhibitor is a previously undescribed innate inhibitory molecule that may have evolved to provide pulmonary protection of the species from fatal influenza virus infection.IMPORTANCE Influenza A virus occasionally jumps from aquatic birds, its natural host, into mammals to cause outbreaks of varying severity, including pandemics in humans. Despite the laboratory mouse being used as a model to study influenza virus pathogenesis, natural outbreaks of influenza have not been reported in the species. Here, we shed light on one mechanism that might allow mice to be protected from influenza in the wild. We show that virus deposited in the mouse upper respiratory tract will not progress to the lower respiratory tract due to the presence of a potent inhibitor of the virus in saliva. Containing inhibitor-sensitive virus to the upper respiratory tract renders an otherwise lethal infection subclinical. This knowledge sheds light on how natural inhibitors may have evolved to improve survival in this species.

Predicting Disease Severity and Viral Spread of H5N1 Influenza Virus in Ferrets in the Context of Natural Exposure Routes.

UNLABELLED: Although avian H5N1 influenza virus has yet to develop the capacity for human-to-human spread, the severity of the rare cases of human infection has warranted intensive follow-up of potentially exposed individuals that may require antiviral prophylaxis. For countries where antiviral drugs are limited, the World Health Organization (WHO) has developed a risk categorization for different levels of exposure to environmental, poultry, or human sources of infection. While these take into account the infection source, they do not account for the likely mode of virus entry that the individual may have experienced from that source and how this could affect the disease outcome. Knowledge of the kinetics and spread of virus after natural routes of exposure may further inform the risk of infection, as well as the likely disease severity. Using the ferret model of H5N1 infection, we compared the commonly used but artificial inoculation method that saturates the total respiratory tract (TRT) with virus to upper respiratory tract (URT) and oral routes of delivery, those likely to be encountered by humans in nature. We show that there was no statistically significant difference in survival rate with the different routes of infection, but the disease characteristics were somewhat different. Following URT infection, viral spread to systemic organs was comparatively delayed and more focal than after TRT infection. By both routes, severe disease was associated with early viremia and central nervous system infection. After oral exposure to the virus, mild infections were common suggesting consumption of virus-contaminated liquids may be associated with seroconversion in the absence of severe disease. IMPORTANCE: Risks for human H5N1 infection include direct contact with infected birds and frequenting contaminated environments. We used H5N1 ferret infection models to show that breathing in the virus was more likely to produce clinical infection than swallowing contaminated liquid. We also showed that virus could spread from the respiratory tract to the brain, which was associated with end-stage disease, and very early viremia provided a marker for this. With upper respiratory tract exposure, infection of the brain was common but hard to detect, suggesting that human neurological infections might be typically undetected at autopsy. However, viral spread to systemic sites was slower after exposure to virus by this route than when virus was additionally delivered to the lungs, providing a better therapeutic window. In addition to exposure history, early parameters of infection, such as viremia, could help prioritize antiviral treatments for patients most at risk of succumbing to infection.

Rapid evolution of the PB1-F2 virulence protein expressed by human seasonal H3N2 influenza viruses reduces inflammatory responses to infection.

Influenza A virus (IAV) PB1-F2 protein has been linked to viral virulence. Strains of the H3N2 subtype historically express full-length PB1-F2 proteins but during the 2010-2011 influenza seasons, nearly half of the circulating H3N2 IAVs encoded truncated PB1-F2 protein. Using a panel of reverse engineered H3N2 IAVs differing only in the origin of the PB1 gene segment, we found that only the virus encoding the avian-derived 1968 PB1 gene matching the human pandemic strain enhanced cellular infiltrate into the alveolar spaces of infected mice. We linked this phenomenon to expression of full-length PB1-F2 protein encompassing critical "inflammatory" residues.

Influenza virus PB1 and neuraminidase gene segments can cosegregate during vaccine reassortment driven by interactions in the PB1 coding region.

UNLABELLED: Egg-grown influenza vaccine yields are maximized by infection with a seed virus produced by "classical reassortment" of a seasonal isolate with a highly egg-adapted strain. Seed viruses are selected based on a high-growth phenotype and the presence of the seasonal hemagglutinin (HA) and neuraminidase (NA) surface antigens. Retrospective analysis of H3N2 vaccine seed viruses indicated that, unlike other internal proteins that were predominantly derived from the high-growth parent A/Puerto Rico/8/34 (PR8), the polymerase subunit PB1 could be derived from either parent depending on the seasonal strain. We have recently shown that A/Udorn/307/72 (Udorn) models a seasonal isolate that yields reassortants bearing the seasonal PB1 gene. This is despite the fact that the reverse genetics-derived virus that includes Udorn PB1 with Udorn HA and NA on a PR8 background has inferior growth compared to the corresponding virus with PR8 PB1. Here we use competitive plasmid transfections to investigate the mechanisms driving selection of a less fit virus and show that the Udorn PB1 gene segment cosegregates with the Udorn NA gene segment. Analysis of chimeric PB1 genes revealed that the coselection of NA and PB1 segments was not directed through the previously identified packaging sequences but through interactions involving the internal coding region of the PB1 gene. This study identifies associations between viral genes that can direct selection in classical reassortment for vaccine production and which may also be of relevance to the gene constellations observed in past antigenic shift events where creation of a pandemic virus has involved reassortment. IMPORTANCE: Influenza vaccine must be produced and administered in a timely manner in order to provide protection during the winter season, and poor-growing vaccine seed viruses can compromise this process. To maximize vaccine yields, manufacturers create hybrid influenza viruses with gene segments encoding the surface antigens from a seasonal virus isolate, important for immunity, and others from a virus with high growth properties. This involves coinfection of cells with both parent viruses and selection of dominant progeny bearing the seasonal antigens. We show that this method of creating hybrid viruses does not necessarily select for the best yielding virus because preferential pairing of gene segments when progeny viruses are produced determines the genetic makeup of the hybrids. This not only has implications for how hybrid viruses are selected for vaccine production but also sheds light on what drives and limits hybrid gene combinations that arise in nature, leading to pandemics.

The source of the PB1 gene in influenza vaccine reassortants selectively alters the hemagglutinin content of the resulting seed virus.

The yields of egg-grown influenza vaccines are maximized by the production of a seed strain using a reassortment of the seasonal influenza virus isolate with a highly egg-adapted strain. The seed virus is selected based on high yields of viral hemagglutinin (HA) and expression of the surface antigens from the seasonal isolate. The remaining proteins are usually derived from the high-growth parent. However, a retrospective analysis of vaccine seeds revealed that the seasonal PB1 gene was selected in more than 50% of reassortment events. Using the model seasonal H3N2 virus A/Udorn/307/72 (Udorn) virus and the high-growth A/Puerto Rico/8/34 (PR8) virus, we assessed the influence of the source of the PB1 gene on virus growth and vaccine yield. Classical reassortment of these two strains led to the selection of viruses that predominantly had the Udorn PB1 gene. The presence of Udorn PB1 in the seed virus, however, did not result in higher yields of virus or HA compared to the yields in the corresponding seed virus with PR8 PB1. The 8-fold-fewer virions produced with the seed virus containing the Udorn PB1 were somewhat compensated for by a 4-fold increase in HA per virion. A higher HA/nucleoprotein (NP) ratio was found in past vaccine preparations when the seasonal PB1 was present, also indicative of a higher HA density in these vaccine viruses. As the HA viral RNA (vRNA) and mRNA levels in infected cells were similar, we propose that PB1 selectively alters the translation of viral mRNA. This study helps to explain the variability of vaccine seeds with respect to HA yield.

Neuraminidase-inhibiting antibody is a correlate of cross-protection against lethal H5N1 influenza virus in ferrets immunized with seasonal influenza vaccine.

In preparing for the threat of a pandemic of avian H5N1 influenza virus, we need to consider the significant delay (4 to 6 months) necessary to produce a strain-matched vaccine. As some degree of cross-reactivity between seasonal influenza vaccines and H5N1 virus has been reported, this was further explored in the ferret model to determine the targets of protective immunity. Ferrets were vaccinated with two intramuscular inoculations of trivalent inactivated split influenza vaccine or subcomponent vaccines, with and without adjuvant, and later challenged with a lethal dose of A/Vietnam/1203/2004 (H5N1) influenza virus. We confirmed that vaccination with seasonal influenza vaccine afforded partial protection against lethal H5N1 challenge and showed that use of either AlPO(4) or Iscomatrix adjuvant with the vaccine resulted in complete protection against disease and death. The protection was due exclusively to the H1N1 vaccine component, and although the hemagglutinin contributed to protection, the dominant protective response was targeted toward the neuraminidase (NA) and correlated with sialic acid cleavage-inhibiting antibody titers. Purified heterologous NA formulated with Iscomatrix adjuvant was also protective. These results suggest that adjuvanted seasonal trivalent vaccine could be used as an interim measure to decrease morbidity and mortality from H5N1 prior to the availability of a specific vaccine. The data also highlight that an inducer of cross-protective immunity is the NA, a protein whose levels are not normally monitored in vaccines and whose capacity to induce immunity in recipients is not normally assessed.

Influenza-induced inflammation drives pneumococcal otitis media.

Influenza A virus (IAV) predisposes individuals to secondary infections with the bacterium Streptococcus pneumoniae (the pneumococcus). Infections may manifest as pneumonia, sepsis, meningitis, or otitis media (OM). It remains controversial as to whether secondary pneumococcal disease is due to the induction of an aberrant immune response or IAV-induced immunosuppression. Moreover, as the majority of studies have been performed in the context of pneumococcal pneumonia, it remains unclear how far these findings can be extrapolated to other pneumococcal disease phenotypes such as OM. Here, we used an infant mouse model, human middle ear epithelial cells, and a series of reverse-engineered influenza viruses to investigate how IAV promotes bacterial OM. Our data suggest that the influenza virus HA facilitates disease by inducing a proinflammatory response in the middle ear cavity in a replication-dependent manner. Importantly, our findings suggest that it is the inflammatory response to IAV infection that mediates pneumococcal replication. This study thus provides the first evidence that inflammation drives pneumococcal replication in the middle ear cavity, which may have important implications for the treatment of pneumococcal OM.

A mouse model for the study of contact-dependent transmission of influenza A virus and the factors that govern transmissibility.

Influenza A virus transmission by direct contact is not well characterized. Here, we describe a mouse model for investigation of factors regulating contact-dependent transmission. Strains within the H3N2 but not H1N1 subtype of influenza virus were transmissible, and reverse-engineered viruses representing hybrids of these subtypes showed that the viral hemagglutinin is a determinant of the transmissible phenotype. Transmission to contact mice occurred within the first 6 to 54 h after cohousing with directly infected index mice, and the proportion of contacts infected within this period was reduced if the index mice had been preinfected with a heterologous subtype virus. A threshold level of virus present in the saliva of the index mice was identified, above which the likelihood of transmission was greatly increased. There was no correlation with transmission and viral loads in the nose or lung. This model could be useful for preclinical evaluation of antiviral and vaccine efficacy in combating contact-dependent transmission of influenza.

Mammalian infections with highly pathogenic avian influenza viruses renew concerns of pandemic potential.

There is unprecedented spread of highly pathogenic avian influenza A H5N1 viruses in bird species on five continents, and many reports of infections in mammals most likely resulting from consumption of infected birds. As H5N1 viruses infect more species, their geographical range increases and more viral variants are produced that could have new biological properties including adaptation to mammals and potentially to humans. This highlights the need to continually monitor and assess mammalian-origin H5N1 clade 2.3.4.4b viruses for the presence of mutations that could potentially increase their pandemic risk for humans. Fortunately, to date there have been a limited number of human cases, but infection of mammals increases the opportunity for the virus to acquire mutations that enhance efficient infection, replication, and spread in mammals, properties that have not been seen in these viruses in the past.

What Have We Learned by Resurrecting the 1918 Influenza Virus?

The 1918 Spanish influenza pandemic was one of the deadliest infectious disease events in recorded history, resulting in approximately 50-100 million deaths worldwide. The origins of the 1918 virus and the molecular basis for its exceptional virulence remained a mystery for much of the 20th century because the pandemic predated virologic techniques to isolate, passage, and store influenza viruses. In the late 1990s, overlapping fragments of influenza viral RNA preserved in the tissues of several 1918 victims were amplified and sequenced. The use of influenza reverse genetics then permitted scientists to reconstruct the 1918 virus entirely from cloned complementary DNA, leading to new insights into the origin of the virus and its pathogenicity. Here, we discuss some of the advances made by resurrection of the 1918 virus, including the rise of innovative molecular research, which is a topic in the dual use debate.

Role of the viral polymerase during adaptation of influenza A viruses to new hosts.

As a group, influenza-A viruses (IAV) infect a wide range of animal hosts, however, they are constrained to infecting selected host species by species-specific interactions between the host and virus, that are required for efficient replication of the viral RNA genome. When IAV cross the species barrier, they acquire mutations in the viral genome to enable interactions with the new host factors, or to compensate for their loss. The viral polymerase genes polymerase basic 1, polymerase basic 2, and polymerase-acidic are important sites of host adaptation. In this review, we discuss why the viral polymerase is so vital to the process of host adaptation, look at some of the known viral mutations, and host factors involved in adaptation, particularly of avian IAV to mammalian hosts.

Gene Segment Interactions Can Drive the Emergence of Dominant Yet Suboptimal Gene Constellations During Influenza Virus Reassortment.

A segmented genome enables influenza virus to undergo reassortment when two viruses infect the same cell. Although reassortment is involved in the creation of pandemic influenza strains and is routinely used to produce influenza vaccines, our understanding of the factors that drive the emergence of dominant gene constellations during this process is incomplete. Recently, we defined a spectrum of interactions between the gene segments of the A/Udorn/307/72 (H3N2) (Udorn) strain that occur within virus particles, a major interaction being between the NA and PB1 gene segments. In addition, we showed that the Udorn PB1 is preferentially incorporated into reassortant viruses that express the Udorn NA. Here we use an influenza vaccine seed production model where eggs are coinfected with Udorn and the high yielding A/Puerto Rico/8/34 (H1N1) (PR8) virus and track viral genotypes through the reassortment process under antibody selective pressure to determine the impact of Udorn NA-PB1 co-selection. We discovered that 86% of the reassortants contained the PB1 from the Udorn parent after the initial co-infection and this bias towards Udorn PB1 was maintained after two further passages. Included in these were certain gene constellations containing Udorn HA, NA, and PB1 that confered low replicative fitness yet rapidly became dominant at the expense of more fit progeny, even when co-infection ratios of the two viruses favoured PR8. Fitness was not compromised, however, in the corresponding reassortants that also contained Udorn NP. Of particular note is the observation that relatively unfit reassortants could still fulfil the role of vaccine seed candidates as they provided high haemagglutinin (HA) antigen yields through co-production of non-infectious particles and/or by more HA molecules per virion. Our data illustrate the dynamics and complexity of reassortment and highlight how major gene segment interactions formed during packaging, in addition to antibody pressure, initially restrict the reassortant viruses that are formed.

Inhibition of Influenza A Virus by Human Infant Saliva.

Innate antiviral factors in saliva play a role in protection against respiratory infections. We tested the anti-influenza virus activities of saliva samples taken from human infants, 1-12 months old, with no history of prior exposure to influenza. In contrast to the inhibitory activity we observed in mouse and ferret saliva, the activity of human infant saliva was complex, with both sialic acid-dependent and independent components, the proportion of which differed between individuals. Taken as a whole, we showed that the major anti-influenza activity of infant saliva is acquired over the first year of life and is associated with sialic acid-containing molecules. The activity of sialic acid-independent inhibitors was lower overall, more variable between individuals, and less dependent on age. The results show that the saliva of very young infants can provide a degree of protection against influenza, which may be critical in the absence of adaptive immunity.

Influenza Virus Neuraminidase Structure and Functions.

With the constant threat of emergence of a novel influenza virus pandemic, there must be continued evaluation of the molecular mechanisms that contribute to virulence. Although the influenza A virus surface glycoprotein neuraminidase (NA) has been studied mainly in the context of its role in viral release from cells, accumulating evidence suggests it plays an important, multifunctional role in virus infection and fitness. This review investigates the various structural features of NA, linking these with functional outcomes in viral replication. The contribution of evolving NA activity to viral attachment, entry and release of virions from infected cells, and maintenance of functional balance with the viral hemagglutinin are also discussed. Greater insight into the role of this important antiviral drug target is warranted.

The structure of the influenza A virus genome.

Influenza A viruses (IAVs) constitute a major threat to human health. The IAV genome consists of eight single-stranded viral RNA segments contained in separate viral ribonucleoprotein (vRNP) complexes that are packaged together into a single virus particle. The structure of viral RNA is believed to play a role in assembling the different vRNPs into budding virions1-8 and in directing reassortment between IAVs9. Reassortment between established human IAVs and IAVs harboured in the animal reservoir can lead to the emergence of pandemic influenza strains to which there is little pre-existing immunity in the human population10,11. While previous studies have revealed the overall organization of the proteins within vRNPs, characterization of viral RNA structure using conventional structural methods is hampered by limited resolution and an inability to resolve dynamic components12,13. Here, we employ multiple high-throughput sequencing approaches to generate a global high-resolution structure of the IAV genome. We show that different IAV genome segments acquire distinct RNA conformations and form both intra- and intersegment RNA interactions inside influenza virions. We use our detailed map of IAV genome structure to provide direct evidence for how intersegment RNA interactions drive vRNP cosegregation during reassortment between different IAV strains. The work presented here is a roadmap both for the development of improved vaccine strains and for the creation of a framework to 'risk assess' reassortment potential to better predict the emergence of new pandemic influenza strains.

Recombinant influenza virus expressing HIV-1 p24 capsid protein induces mucosal HIV-specific CD8 T-cell responses.

Influenza viruses are promising mucosal vaccine vectors for HIV but their use has been limited by difficulties in engineering the expression of large amounts of foreign protein. We developed recombinant influenza viruses incorporating the HIV-1 p24 gag capsid into the NS-segment of PR8 (H1N1) and X31 (H3N2) influenza viruses with the use of multiple 2A ribosomal skip sequences. Despite the insertion of a sizable HIV-1 gene into the influenza genome, recombinant viruses were readily rescued to high titers. Intracellular expression of p24 capsid was confirmed by in vitro infection assays. The recombinant influenza viruses were subsequently tested as mucosal vaccines in BALB/c mice. Recombinant viruses were attenuated and safe in immunized mice. Systemic and mucosal HIV-specific CD8 T-cell responses were elicited in mice that were immunized via intranasal route with a prime-boost regimen. Isolated HIV-specific CD8 T-cells displayed polyfunctional cytokine and degranulation profiles. Mice boosted via intravaginal route induced recall responses from the distal lung mucosa and developed heightened HIV-specific CD8 T-cell responses in the vaginal mucosa. These findings demonstrate the potential utility of recombinant influenza viruses as vaccines for mucosal immunity against HIV-1 infection.

Influenza NA and PB1 Gene Segments Interact during the Formation of Viral Progeny: Localization of the Binding Region within the PB1 Gene.

The influenza A virus genome comprises eight negative-sense viral RNAs (vRNAs) that form individual ribonucleoprotein (RNP) complexes. In order to incorporate a complete set of each of these vRNAs, the virus uses a selective packaging mechanism that facilitates co-packaging of specific gene segments but whose molecular basis is still not fully understood. Recently, we used a competitive transfection model where plasmids encoding the A/Puerto Rico/8/34 (PR8) and A/Udorn/307/72 (Udorn) PB1 gene segments were competed to show that the Udorn PB1 gene segment is preferentially co-packaged into progeny virions with the Udorn NA gene segment. Here we created chimeric PB1 genes combining both Udorn and PR8 PB1 sequences to further define the location within the Udorn PB1 gene that drives co-segregation of these genes and show that nucleotides 1776-2070 of the PB1 gene are crucial for preferential selection. In vitro assays examining specific interactions between Udorn NA vRNA and purified vRNAs transcribed from chimeric PB1 genes also supported the importance of this region in the PB1-NA interaction. Hence, this work identifies an association between viral genes that are co-selected during packaging. It also reveals a region potentially important in the RNP-RNP interactions within the supramolecular complex that is predicted to form prior to budding to allow one of each segment to be packaged in the viral progeny. Our study lays the foundation to understand the co-selection of specific genes, which may be critical to the emergence of new viruses with pandemic potential.

Serum amyloid P is a sialylated glycoprotein inhibitor of influenza A viruses.

Members of the pentraxin family, including PTX3 and serum amyloid P component (SAP), have been reported to play a role in innate host defence against a range of microbial pathogens, yet little is known regarding their antiviral activities. In this study, we demonstrate that human SAP binds to human influenza A virus (IAV) strains and mediates a range of antiviral activities, including inhibition of IAV-induced hemagglutination (HA), neutralization of virus infectivity and inhibition of the enzymatic activity of the viral neuraminidase (NA). Characterization of the anti-IAV activity of SAP after periodate or bacterial sialidase treatment demonstrated that α(2,6)-linked sialic acid residues on the glycosidic moiety of SAP are critical for recognition by the HA of susceptible IAV strains. Other proteins of the innate immune system, namely human surfactant protein A and porcine surfactant protein D, have been reported to express sialylated glycans which facilitate inhibition of particular IAV strains, yet the specific viral determinants for recognition of these inhibitors have not been defined. Herein, we have selected virus mutants in the presence of human SAP and identified specific residues in the receptor-binding pocket of the viral HA which are critical for recognition and therefore susceptibility to the antiviral activities of SAP. Given the widespread expression of α(2,6)-linked sialic acid in the human respiratory tract, we propose that SAP may act as an effective receptor mimic to limit IAV infection of airway epithelial cells.