High-Throughput Human Primary Cell-Based Airway Model for Evaluating Influenza, Coronavirus, or other Respiratory Viruses in vitro

Influenza and other respiratory viruses represent a significant threat to public health, national security, and the world economy, and can lead to the emergence of global pandemics such as the current COVID-19 crisis. One of the greatest barriers to the development of effective therapeutic agents to treat influenza, coronaviruses, and many other infections of the respiratory tract is the absence of a robust preclinical model. Preclinical studies currently rely on high-throughput, low-fidelity in vitro screening with cell lines and/or low-throughput animal models that often provide a poor correlation to human clinical responses. Here, we introduce a human primary airway epithelial cell-based model integrated into a high-throughput platform where tissues are cultured at an air-liquid interface (PREDICT96-ALI). We present results on the application of this platform to influenza and coronavirus infections, providing multiple readouts capable of evaluating viral infection kinetics and potentially the efficacy of therapeutic agents in an in vitro system. Several strains of influenza A virus are shown to successfully infect the human primary cell-based airway tissue cultured at an air-liquid interface (ALI), and as a proof-of-concept, the effect of the antiviral oseltamivir on one strain of Influenza A is evaluated. Human coronaviruses NL63 (HCoV-NL63) and SARS-CoV-2 enter host cells via ACE2 and utilize the protease TMPRSS2 for protein priming, and we confirm expression of both in our ALI model. We also demonstrate coronavirus infection in this system with HCoV-NL63, observing sufficient viral propagation over 96 hours post-infection to indicate successful infection of the primary cell-based model. This new capability has the potential to address a gap in the rapid assessment of therapeutic efficacy of various small molecules and antiviral agents against influenza and other respiratory viruses including coronaviruses.


Introduction
The specter of a pandemic respiratory virus such as influenza or the current outbreak of coronavirus represents one of the greatest threats to human health [1,2]. For influenza, the development of efficacious, long-lasting, and seasonally-consistent vaccinations has been challenged by the occurrence of both antigenic drift and shift [3]. Currently, four antivirals are in wide use for treatment of influenza, three neuraminidase inhibitors and more recently, a cap-dependent endonuclease inhibitor, baloxivir marboxil [4]. These treatments are generally effective, but genetic changes to endemic viruses present challenges similar to those encountered during the development of vaccines [5]. These challenges are exacerbated by the long developmental timelines typically associated with manufacturing, scale up, and regulatory approval for novel antivirals or vaccinations. The current COVID-19 pandemic of the novel coronavirus has spurred aggressive vaccine development efforts, but even with this accelerated timeline, the current wave of infection is projected to yield significant morbidity and mortality [6]. Therefore, a critical element of medical treatment strategy is the application of repurposed FDA-approved antiviral medications originally developed for other diseases [7].
Several preclinical research tools are available to model respiratory viruses, but due to the complexity of the human lung and requirements to accurately model species-specific viral infections, these models provide a poor correlation as a basis for guiding clinical strategy [8]. For example, a standard approach to investigate influenza A viruses (IAV) typically begins with the inoculation of cell lines in conventional culture well plates, such as Madin-Darby canine kidney (MDCK) or Calu-3 cells [9,10]. While these cells infect easily and provide a rapid early assessment of the potential efficacy of a therapeutic for IAV, they lack the critical mechanisms and signaling pathways that healthy human airway tissues provide in response to viral infection. Therefore, although these systems are easy to use, they are not often accurate predictors of human clinical efficacy [11]. For coronaviruses, Calu-3 or Vero E6 cells are often used for similar reasons, but again differ by species, cell source, disease state, and innate antiviral signaling pathways [12,13] compared to normal, healthy human conducting airway and alveolar epithelium [14], thus providing an insufficient model. Unfortunately, complications also exist within animal models of IAV, not only in mice, rats, and guinea pigs, but even in more human-relevant options for evaluation such as ferrets and macaques [15,16]. These models often require species-adapted viral strains, and fail to recapitulate features of human innate antiviral responses and clinical symptoms of infection [17,18].
Similar challenges are faced for coronavirus infection studies, where the low-throughput and limited availability of appropriate animal models, combined with concerns about their predictive power, are forcing early and often high-risk testing in humans. While many experimental studies are underway to identify drugs with antiviral activity against SARS-CoV-2 [19], the discovery pipeline would benefit significantly from rapid and accurate preclinical models based on human cell culture.
The challenges with preclinical models and tools have spurred the development of cell culture platforms comprising human primary cells in a microfluidic environment, commonly known as microphysiological systems (MPS) or organs-on-chips [20]. Over the past two decades, these devices have evolved from early toxicology platforms to today's wide range of technologies including multi-organ systems [21,22] and a variety of organ-specific models including airway or alveolar tissue systems [23,24]. While the integration of human primary cells or stem cell-derived populations in a physiologically-relevant microenvironment represents a tremendous advance over the use of cell lines in principle, in practice the application of these systems in pharmacological research and development has been severely gated by several factors. These limitations include low-throughput and complex operation, a lack of relevant metrics for system assessment, the use of research-grade materials and components, the lack of critical components of tissue or organ structure required to replicate infection, and most critically, an absence of confidence in the in vitro-in vivo correlation (IVIVC) for these technologies. Our group has developed systems that offer higher throughput, compatibility with existing pharmaceutical laboratory infrastructure, convenient readouts for quality control and physiological monitoring, and that are demonstrating the potential as platforms for preclinical studies across many disease areas and application domains [25,26].
In this manuscript, we build on our previous work and the work of others to develop an in vitro model with an ALI capability sufficient to support physiologically relevant epithelial differentiation such as that in gut or kidney models [25,26]. For functional proximal airway or alveolar models, human primary cells should be cultured at an ALI representing a suitable analogue for the respiratory barrier tissues that mimics mucus formation, mucociliary flow, and a range of physiologically relevant responses [23,24] tissues cultured at an ALI, and viral infection kinetics are monitored through a combination of readouts including quantitation of viral load. As a proof-of-concept, viral infection kinetics are investigated for the A/WSN/33 H1N1 strain of IAV in response to dosing with oseltamivir, demonstrating how this system might be applied to evaluate the efficacy of antiviral therapeutic compounds. We also confirm expression of the receptor ACE2 and the serine protease TMPRSS2 in these tissues, and we demonstrate viral propagation and infection kinetics of HCoV-NL63, which, like SARS-CoV and SARS-CoV-2, utilizes ACE2 as its target receptor. These results demonstrate that the PREDICT96-ALI platform can be used to model IAV and coronavirus infections and could potentially be used to screen the efficacy of candidate therapeutics in a clinically relevant manner.

Microfluidic Platform
The PREDICT96 organ-on-chip platform consists of a microfluidic culture plate with 96 individual devices and a perfusion system driven by 192 microfluidic pumps integrated into the plate lid [25,26]. Here the platform is adapted to support culture of barrier tissues, such as airway tissues, by establishment and maintenance of an ALI.
Each device in the PREDICT96-ALI culture plate was formed from a 2x2 array within a standard 384-well plate, with the wall between the top two wells removed to create a 3-well cluster. This configuration is shown in Figure 1a. A slotted hard plastic microwell (0.864 mm height, 1 mm width, 2.5 mm length) was positioned in the center of the large well, with a microfluidic channel (0.25 mm height, 1 mm width) aligned underneath that was linked to the two remaining wells in each device. The channel was capped from below with a thin optically clear layer. A microporous membrane was used to separate the top microwell from the bottom microchannel to permit the establishment of an ALI.

Integrated Micropumps
As described in our previous work [25], in order to establish recirculating flow in the microchannels, we incorporate a self-priming micropump array into the lid that serves as the fluidic interface with the culture plate via stainless steel tubes. The

Culture, Cryopreservation, and PREDICT96-ALI Seeding and Differentiation of NHBEs
Normal Human Bronchial Epithelial Cells (NHBEs) were purchased from Lifeline Cell Technology and Lonza (Table 1). To maintain stocks of each NHBE donor, cells were thawed and plated at 2500 cells/cm 2 on 804G media-coated tissue culture flasks (804G media made as described elsewhere [27]), cultured in Bronchialife (Lifeline Cell Technology), and passaged using Accutase (Sigma). After two passages, NHBEs were cryopreserved using 65% FBS (Thermo Fisher), 25% Bronchialife (Lifeline Cell Technology), and 10% DMSO (Sigma). Prior to seeding, PREDICT96-ALI plates were sterilized overnight with ethylene oxide gas followed by one week of outgassing in a vacuum chamber. PREDICT96-ALI plates were subsequently treated with plasma for 120 seconds and washed briefly with 70% ethanol, rinsed three times in distilled water, and coated overnight in 804G media. To seed PREDICT96-ALI plates, NHBEs were thawed, counted, re-suspended in complete small airway epithelial cell growth media (SAGM; Lonza), 100 U/mL penicillinstreptomycin (Thermo Fisher), 5 M ROCKi (Tocris), 1 M A-83-01 (Tocris), 0.2 M DMH-1 (Tocris), 0.5 M CHIR99021 (Tocris) as described elsewhere [28] (hereafter referred to as SAGM + 4i), and plated at 10,000 cells per device in a 3 L volume directly onto the membrane. After 48 hours (h) of growth in SAGM + 4i media, differentiation was initiated using fresh HBTEC-ALI media (Lifeline Cell Technology) plus 100 U/mL penicillin/streptomycin. After 48 h of submerged differentiation, the ALI was initiated by aspirating media from the apical surface of the tissue in the top chamber, while 60 L of fresh HBTEC-ALI or custom-ALI media was added to the bottom chamber. Pumping was initiated at 1 L /min in the bottom channel, and the media in the bottom channel was changed daily thereafter. Tissues were matured over the course of 3-5 weeks in ALI culture prior to viral infection experiments. To remove accumulated mucus from the apical surface of the maturing tissue, 100 µL of 1x Hank's Balanced Salt Solution (HBSS) was added to the apical surface of each tissue and incubated on the tissues for 1 h at 37 ⁰C with rocking, subsequently followed by an additional 5 min wash with 100 µL of 1x HBSS at room temperature every 7 days (d). Mucus washings were pooled, collected and stored at -80⁰C until processed. Tissue maturity and quality control was scored by a combination of metrics including barrier function, percent ciliated cells, ciliary beat, mucus secretion, and global tissue morphology, and this score was used to determine if individual devices of PREDICT96-ALI airway tissue were suitable for downstream experimentation and viral infection.
Downstream processing of tissues following viral inoculation was generally conducted at 4-6 weeks of ALI and included tissue apical washes, basal media collection, measurement of barrier function, harvest of tissue for RNA extraction, and fixation for immunofluorescence imaging.

Immunofluorescence and confocal imaging
To prepare PREDICT96-ALI airway tissues for immunofluorescence staining, tissues were fixed in situ using 4% paraformaldehyde (Sigma) for 15 min at room temperature rocking and washed three times with 1x phosphate buffered saline (PBS). Tissues were subsequently permeabilized with 0.3% Triton-100 (Sigma) for 30 min and blocked with 3% normal goat serum (NGS, Thermo Fisher) containing 0.1% Tween-20 (Sigma) for 60 min at room temperature. Primary antibodies for basal cells (CK5, Thermo Fisher), goblet cells (Muc5AC, Thermo), ciliated cells (acetylated-tubulin, Abcam and beta-tubulin, Sigma), cellular proliferation (Ki67, Abcam), and influenza A nucleoprotein (IAV-NP, Abcam) were diluted 1:100 in 3% NGS including 0.1% Tween-20 and incubated overnight at 4°C rocking. Devices were washed twice with PBS followed by one rinse with 3% NGS plus 0.1% Tween-20 for 5 min each with rocking and secondary antibodies Alexa Fluor 488 goat anti-mouse IgG (Thermo Fisher) and Alexa Fluor 555 goat anti-rabbit IgG (Thermo Fisher) were diluted 1:300 and incubated for 18 h at 4 ⁰C with rocking. Secondary antibodies were removed by washing three times with PBS containing 0.1% Tween-20, and tissues were incubated with rocking at room temperature for 30 min with Hoechst 33342 nuclear stain (Thermo Fisher, 10 mg/mL stock) at a 1:800 dilution and Phalloidin-iFluor 647 Reagent (Abcam) at 1:1000. After 30 min, the PREDICT96-ALI tissues were washed at least three times for 5 min with PBS prior to imaging. Stained PREDICT96-ALI devices were imaged using a Zeiss LSM700 laser scanning confocal microscope and Zen Black software. Tile scans and z-stacks of the tissue at 20x and 40x magnification were acquired. and titered by viral plaque assay using MDCK cells as previously reported [29]. Viral infection experiments using PREDICT96-ALI tissue used A/WSN/33 H1N1 at passage 1, A/California/04/09 H1N1 at passage 3, and A/Hong Kong/8/68 H3N2 at passage 2.

Viral Stocks, Propagation, Titer and Anti-Viral Compound
Human coronavirus strain (HCoV) NL63 was purchased from ATCC and propagated two times in Lilly Laboratories Cell-Monkey Kidney 2 (LLC-MK2) using a method adapted from Herzog [30]. In brief, HCoV- Viral plaque assay using LLC-MK2 cells were used to titer propagated HCoV-NL63. A 10-fold serial dilution was performed using the propagated virus, followed by 1 h of viral binding (100 µL of appropriate viral dilution per well) with rocking at 34 °C on confluent LLC-MK2 cells. After unbound virus was removed with one wash of phosphate-buffered saline (PBS), the cells were overlaid with agar (0.3%) in EMEM supplemented with 0.1% BSA and 100 U/mL penicillin-streptomycin. After the agar solidified at room temperature, the plates were incubated for 6 d at 34 °C until diffuse CPE was observed. The gel overlay was removed and the wells washed once with PBS. Cells were fixed using 4% paraformaldehyde (Sigma) over night at 4°C and stained with 0.1% crystal violet (Sigma) in 4% ethanol (Sigma) for 30 min at room temperature. Stained cells were washed in PBS and imaged, and plaque counts were determined using images analyzed using a customized MATLAB script to determine the plaque forming units (PFU) per mL.
Oseltamivir carboxylate (F. Hoffmann-La Roche Ltd., Basel, Switzerland) was used in anti-viral screens performed on PREDICT96-ALI tissue. Oseltamivir was diluted to 1 µM in complete HBTEC-ALI or custom-ALI media and applied to the bottom channel of PREDICT-ALI airway tissues subject to anti-viral evaluation following the standard apical wash of the tissue and 2 h prior to IAV-inoculation. Upon and after IAVinoculation of the apical side of the PREDICT96-ALI tissue, 1 µM oseltamivir was maintained in the basal media for the duration of the study. Basal media changes including oseltamivir occurred every 24 h.

Infection of PREDICT96-ALI Tissues with Viral Strains
To prepare PREDICT96-ALI tissues for influenza and coronavirus infection, a mucus wash using 1x HBSS as Comparative cycle threshold (Ct) values were determined using the method described by Schmittgen and Livak [31] using normalization to the housekeeping gene GAPDH.

Statistical analysis
Data are presented as mean ± standard deviation and were analyzed using GraphPad Prism version 8.3.1 for Windows (GraphPad Software). Statistical significance was determined using two-way analysis of variance (ANOVA) with Tukey's or Sidak's post hoc test for multiple comparisons, where appropriate. A pvalue lower than 0.05 was considered statistically significant and is indicated in figures as follows: * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001, and **** P ≤ 0.0001.

Ethics statement
No ethics approval was required for this work.

Creation of a high-throughput ALI platform for dynamic culture of human airway tissue: PREDICT96-ALI
The adaptation of the PREDICT96 platform [26] for the ALI model allows for the ability to inoculate cultured tissues in each channel apically or basally, to sample the basal media at regular intervals, and to examine the tissues at the completion of each study, as shown in

PREDICT96-ALI recapitulates healthy human airway structure and composition
First, the PREDICT96-ALI system was used to establish the healthy baseline airway model, prepared as described in the Methods section, and morphology and cell populations were evaluated using high resolution confocal microscopy and immunohistochemistry. An illustration depicts the development of a healthy airway model within PREDICT96-ALI (Figure 2a). The experimental timeline and setup for the system is outlined in Figure 2b. Fluorescent staining of ciliated cells, goblet cells, and basal cells (Figure   2c), as well as a cross-section of the pseudostratified epithelial layer (Figure 2d) are shown for a typical airway tissue within a single device of PREDICT96-ALI. Our observations for airway tissues in the PREDICT96-ALI platform regarding the relative populations of ciliated cells, basal cells, and goblet cells are consistent with our previously reported observations in earlier generations of our airway models [23].

PREDICT96-ALI airway model supports infection with influenza A virus
Next, we examined the effect of inoculation of airway ALI cultures with various strains of IAV, including A/WSN/33 H1N1, A/California/04/09 H1N1, and A/Hong Kong/8/68 H3N2. In

Influenza A virus replication is reduced by oseltamivir in PREDICT96-ALI airway model
To determine if the PREDICT96-ALI airway model can be used to evaluate the efficacy of potential antiviral therapeutics, we investigated the effect of the antiviral agent oseltamivir (Tamiflu) -the most commonly used clinical anti-influenza therapy -for its ability to reduce viral load in IAV-inoculated PREDICT96-ALI airway tissue. The active form of oseltamivir, oseltamivir carboxylate, was introduced 2 h before viral inoculation and maintained in the bottom channel of the PREDICT96-ALI airway tissues during the course of infection with A/WSN/33 H1N1 virus at various MOIs. Oseltamivir (1 μM) significantly reduced influenza replication up to 48 h p.i. (Figure 5) relative to control tissue devices for the A/WSN/33 H1N1 strain at MOI = 1 and prevented virus-induced disruption to barrier function and epithelial tight junction formation (data not shown). Measurements for viral copies made in the PREDICT96-ALI airway model reflect those made in clinical settings among patients exposed to influenza and treated with oseltamivir [32,33,34,35], suggesting that the PREDICT96-ALI airway model can serve as a preclinical tool to evaluate potential therapies for combating respiratory infections of the human airway.

ACE2 and TMPRSS2 transcript expression in PREDICT96-ALI airway model
In preparation for testing the viral respiratory infection model with coronaviruses, we probed for transcript expression of cell surface proteins ACE2 and TMPRSS2 known to play central roles in facilitating viral infection by SARS-CoV-2 [36]. Both HCoV-NL63 and SARS-CoV-2 gain entry into host cells via the ACE2 receptor, while TMPRSS2 facilitates SARS-CoV infection via cleavage mechanisms, and thus its presence is an important aspect of the model. Data for ACE2 and TMPRSS2 transcripts for cell populations derived from two different donors are plotted on a semi-logarithmic scale in Figure 6. Data from each donor evaluated suggest a lack of statistical difference in transcript expression in ACE2 and TMPRSS2 between donors. The consistency between these key cellular factors among different donors is encouraging, as it suggests that the platform is capable of supporting proliferation and maturation of tissues derived from numerous donors. Thus, evaluation of many donors can be accommodated in the PREDICT96-ALI system both as a healthy model and in the presence of pathogenic challenge.

PREDICT96-ALI airway model supports infection with human coronavirus NL63
We have also investigated viral infection with HCoV-NL63 in the PREDICT96-ALI culture system with primary NHBEs, as shown in Figures 7a and 7b for two different donor cell populations. In each case, viral copies in the supernatant were measured with qRT-PCR over periods extending to 96 h p.i. and across a range of MOIs. We observed a clear increase in viral load over these time periods for both donors, indicating the presence of viral propagation in the airway tissues. Viral copies rose sharply from 0 to 48 h for each MOI tested, while from 48 to 96 h the rate of viral propagation decreased and differed between the two donors.

Discussion
In this work, we leverage the PREDICT96-ALI platform to create a high-throughput airway infection model A critically important need in the field of respiratory virus research is the establishment of robust in vitro disease models that provide the precision, validation, throughput, and utility necessary for routine operation in drug development laboratories. Until recently, in vitro models suitable for the study of respiratory viral infections have been either high-throughput but simplistic 2D culture plate-based systems utilizing cell lines, or low-throughput and complex organ-on-chip models that are difficult to accommodate in the workflow of many laboratories. The PREDICT96 platform has been reported previously, and represents the first high-throughput organ-on-a-chip platform with integrated precision flow control in a standard ANSI SLAS (American National Standards Institute / Society for Laboratory Automation and Screening) well plate format [25]. Here, we deployed the airway tissue model successfully

Author Interests
The authors are all employees of Draper or the University of Massachusetts Medical School and declare no conflicts.        from PREDICT96-ALI airway tissue. PREDICT96-ALI airway tissue from Donor A and B were matured to 4 weeks ALI and shown to have a non-significant (ns) difference in ACE2 and TMPRSS2 gene transcript levels when compared to one another. Comparative CT values were determined using the method described by Schmittgen and Livak [31] and used to determine the relative quantification of gene expression using GAPDH as a reference gene. N = 3-4 tissue replicates per donor.