Mechanical control of innate immune responses against viral infection revealed in a human lung alveolus chip

Recapitulation of alveolar cell differentiation and alveolar-capillary interface formation

The human alveolus chip used in this study is a microfluidic device containing two parallel channels separated by an extracellular matrix (ECM)-coated porous membrane lined by primary human lung alveolar epithelium cells cultured under an ALI on its upper surface and primary human pulmonary microvascular endothelial cells on the lower surface, which are fed by continuous flow of culture medium through the lumen of the lower vascular channel (Fig. 1a)^10,12,13. The engineered alveolar-capillary interface is also exposed to cyclic mechanical deformations (5% cyclic strain at 0.25 Hz) to mimic physiological breathing motions at tidal volume¹⁴ and the normal respiratory rate of humans via application of cyclic suction to hollow side chambers within the flexible polydimethylsiloxane (PDMS) device (Fig. 1a).

Using a commercial source of primary human lung alveolar epithelial cells composed of 90% alveolar type II (ATII) cells (Supplementary Fig. 1a) and an optimized ECM coating that resembles the composition of basement membranes in vivo¹⁵ (200 µg/ml collagen IV and 15 µg/ml laminin), we found that the alveolus chips support differentiation of the ATII cells into to alveolar type I (ATI) cells (Fig. 1b and Supplementary Fig. 1b) resulting in a final ratio of ATII to ATI cells of 55:45 as demonstrated by immunostaining of ATII cell marker surfactant protein C (SPC) and the ATI cell marker RAGE (Fig. 1c). This is also accompanied by formation of a tight epithelial barrier (Supplementary Fig. 1c), which approximates that observed in vivo¹⁶. Confluent monolayers of interfaced alveolar epithelium and endothelium also display continuous cell-cell junctions stained apically with zonula occludens-1 (ZO-1) in epithelium and laterally with vascular endothelial-cadherin (VE-cadherin), respectively, by 14 days of culture (Fig. 1d), which are maintained for at least 5 weeks on-chip.

RNA-sequencing (RNA-seq) analysis of the alveolar epithelium showed that epithelial cells cultured on-chip exhibit extensive changes in mRNA expression over 14 days in culture (Fig. 1e) and display distinct transcriptome changes compared to the same cells maintained in conventional static Transwell cultures, as demonstrated by principal component analysis (PCA) and differential gene expression analysis (Supplementary Fig. 2a, b). Among the top differentially expressed genes (DEGs) are genes related to multicellular organismal homeostasis, respiratory tube development, and vasculature development, which is consistent with the establishment of epithelial-endothelial signaling crosstalk required for the maturation of the alveolar epithelium and the microvascular endothelium (Supplementary Fig. 2c and Supplementary Data 1 and 2) as we observed on-chip (Fig. 1).

Using short time-series expression miner (STEM)¹⁷, we identified five gene expression patterns that exhibited statistically significant changes over time when cultured on-chip in the presence of breathing-like motions (Fig. 1e and Supplementary Fig. 3a). Gene ontology (GO) analysis revealed that many of the upregulated genes are involved in biological processes relevant to differentiation, including anatomical structure development, cell adhesion, movement of cell or subcellular components, and ECM organization (Supplementary Fig. 3b). Examples include aquaporin 5 (AQP5) and podoplanin (PDPN) (Supplementary Fig. 4a) that are known markers for ATI cells¹⁸, confirming the differentiation from ATII to ATI cells that we observed by immunostaining (Fig. 1b and Supplementary Fig. 1b). Lipoprotein lipase (LPL) and surfactant protein D (SFTPD), two genes involved in surfactant production, also increased (Supplementary Fig. 4a), consistent with the previous finding that mechanical strain promotes enhanced surfactant secretion in vivo¹⁹ and on-chip⁴. The downregulated pathways (profiles 0 and 4, Supplementary Fig. 3b) are associated with cell junctions and ECM organization, which include genes such as CLDN11 (claudin 11), CDH5 (cadherin 5), and PLAT (plasminogen activator) (Supplementary Fig. 3c).

The upregulation of several ion channels and transporters (Supplementary Fig. 4b) suggests that alveolar fluid clearance, which is essential for maintenance of air-liquid interface, may be increased as well. Many genes belonging to defense responses also were upregulated from day 8 to 14 (Fig. 1f), including angiotensin converting enzyme 2 (ACE2), the receptor for SARS-CoV-2, which is also a key mediator of lung physiology and pathophysiology²⁰. Other important innate immune response genes include myxovirus resistance 1 (MX1), 2′−5′-oligoadenylate synthetase 1 (OAS1), DExD/H-Box Helicase 58 (DDX58), signal transducer and activator of transcription 1 (STAT1), which we confirmed using qPCR (Fig. 1g) and immunofluorescence microscopy (Fig. 1h). Importantly, this upregulation of genes related to alveolar development and defense response is also seen in human lung during the transition from fetus to birth (Supplementary Data 3)^21,22.

Significant genetic reprogramming was also observed in endothelial cells cultured in the presence of cyclic mechanical strain on-chip over 14 days in culture (Supplementary Fig. 5a, b). The Wnt signaling pathway is essential for cross-talk between endothelium and epithelium during lung development²³ and consistent with this, we found a number of Wnt ligands, including WNT3, WNT7A, WNT7B, WNT9A, and WNT10A, exhibit elevated expression in endothelial cells on-chip in the presence of breathing motions (Supplementary Fig. 5c). Together, our analysis reveals that human lung alveolus chips that are exposed to physiological breathing motions recapitulate perinatal maturation of the alveolar-capillary interface, cell-ECM interactions, and epithelial-endothelial crosstalk that are indispensable for the development, homeostasis, and regeneration of the lung alveolus²⁴.

Influenza A virus infection of the human lung alveolus chip

Influenza A virus entry is mediated by hemagglutinin binding to sialic acid receptor on host cells, with avian and mammalian viral strains preferentially binding to α−2,3- or α−2,6-linked sialylated glycans, respectively²⁵. Using glycan-specific lectins, we found that the human lung alveolar epithelial cells predominantly express α−2,3-linked sialic acid receptors on their apical surface on-chip (Fig. 2a). Consistent with this finding and results obtained in human lung tissues in vivo²⁶, we found that influenza A/HongKong/8/68 (HK/68; H3N2) and A/HongKong/156/1997 (HK/97; H5N1) viruses successfully infect the epithelium in the human alveolus chip, whereas the influenza A/WSN/33 (WSN; H1N1) virus does not, as detected by immunostaining of viral nuclear protein (NP) (Fig. 2b). The specificity of this response is further exemplified by the finding that H1N1 virus that preferentially infects large airways also infects human Lung Airway Chips lined by bronchial epithelium much more effectively than H3N2²⁷.

Consistent with the observation that influenza A virus mainly target ATII cells in the human lung²⁸, we found that H3N2 infection results in over 90% reduction of the surfactant gene expression (Fig. 2c). This is also consistent with recent observation that SARS-CoV-2 infection in the alveolus leads to suppression of the ATII program²⁹. Loss of ATII cells caused by H3N2 infection leads to increased lung permeability (Fig. 2d) as well as a 25% increase in apoptotic cells as indicated by Apopxin Green staining (Fig. 2e, f), while no necrosis was observed at 48 h after infection. The absence of necrosis-mediated cell death by H3N2 infection on-chip is consistent with reports from human biopsies indicating that this type of cell death is associated with more severe human and avian infections caused by H5N1 and H9N2 strains³⁰. We also observed a significant increase in cell proliferation by quantifying Ki67 staining (Fig. 2e, g), indicating the initiation of a repair program similar to that observed in vivo³¹. Similar injury and repair responses are also observed in the endothelial cells (Fig. 2e–g), supporting recent evidence that lung endothelium possesses substantial regenerative capacity after viral pneumonia³². In summary, our alveolus chip replicates both the viral and cellular tropisms and cellular phenotypical alterations that are known to be produced by influenza infection in the human distal lung, including loss of barrier, cell death, and regeneration.

Multi-level host responses to infection in the human lung alveolus chip

As cytokine production by lung epithelial and endothelial cells is a key feature of early host innate immune response to viral infections, we analyzed cytokines that are secreted into the basal vascular outflow using Luminex assays, analogous to measurement of plasma cytokine levels in vivo. These studies revealed that infection with influenza H3N2 virus induces significantly higher levels of interleukin 6 (IL-6), IL-8, interferon gamma-induced protein 10 (IP-10), tumor necrosis factor alpha (TNF), and granulocyte-macrophage colony-stimulating factor (GM-CSF) in the vascular outflows at 48 h post infection (hpi) compared to control uninfected chips (Fig. 3a). RNA-seq of the epithelial cells carried out at the same time point (48 hpi) revealed 496 upregulated genes and 538 downregulated genes (Fig. 3b). GO analysis indicated that the differentially expressed genes (DEGs) belong to biological pathways including cell division and DNA replication that may be associated with antiviral defense responses and activation of alveolar barrier repair in response to virus-induced injury (Fig. 3c and Supplementary Fig. 6a), which is consistent with the observation of increased Ki67⁺ proliferative cells (Fig. 2f). Some of these DEGs, such as CXCL10 (C-X-C Motif Chemokine Ligand 10, encodes IP-10) and IL6, were further confirmed by qPCR (Supplementary Fig. 6b).

A more detailed analysis of antiviral responses induced in the lung epithelium in the alveolus chip revealed a dominant type III interferon (IFN-III) response mediated by interferon lambda (IFNλ) (Supplementary Fig. 6c). Importantly, this finding differs from results obtained from analysis of human lung alveolar epithelial cells cultured in static 2D culture on rigid planar dishes where IFNβ-mediated type I interferon response prevails³³. However, our results replicate in vivo findings³⁴, which suggest that IFN-III mediates the front-line response to viral infection in mucosal tissues³⁵. In contrast to the epithelium (Fig. 3b), a more limited effect on the host transcriptome was observed in the endothelium (Fig. 3d). But enrichment analysis shows that genes involved in activation of immune responses and ion transport are also upregulated in these cells (Fig. 3e). Interestingly, however, this is likely through tissue-tissue signaling crosstalk as viral mRNA could not be detected within the endothelium.

Influenza virus infection also upregulates expression of endothelial adhesion molecules, thereby allowing the recruitment of leukocytes to the alveolus in vivo⁹. When fluorescently labeled primary human peripheral blood mononuclear cells (PBMCs) were flowed through the endothelium-lined vascular channel 24 hpi with influenza H3N2, we detected upregulation of ICAM1 and TNF in the endothelial cells by qPCR, indicating endothelial inflammation (Supplementary Fig. 6d). Indeed, this resulted in more than a 100-fold increase in the adhesion of the circulating PBMCs to the surface of the activated endothelium compared to control uninfected chips (Fig. 3f, g and Supplementary Fig. 6e). To determine which immune cell types are recruited, we isolated human CD14⁺ monocytes, CD3⁺ T cells, and CD19⁺ B cells, labeled each population with a different fluorescent cell tracker, mix them at a ratio of 6:3:1 (monocytes: T cells: B cells), and perfused the mixture through the vascular channel of the chip 24 h after infection (Supplementary Fig. 6f). Two hours after perfusion under static conditions, we found that all of the cell types were able to adhere to the endothelium or transmigrate up into the epithelium (Supplementary Fig. 6g and Fig. 3h). Interestingly, a higher percentage of B cells remained in the endothelium than T cells, which is in agreement with the timing of T cell activation preceding that of B cells. In addition, the recruitment of immune cells further increased the levels of cytokines, such as IL-8, IL-18, and S100 Calcium Binding Protein A8 and A9 (S100A8/9; also named as Calprotectin) (Supplementary Fig. 6h), consistent with our previous findings from the human Airway Chip³⁶. Thus, the human breathing alveolus chip faithfully replicates multi-level host innate immune responses of lung alveoli to influenza A virus infection.

Breathing-like mechanical deformations suppress viral infection on-chip

We next explored the role of cyclic respiratory motions in virus-induced respiratory infections by infecting alveolus chips with H3N2 virus on day 15 of culture and measuring viral loads in the presence or absence of physiological, breathing-like, mechanical deformations (5% strain, 0.25 Hz). Immunostaining for influenza virus nucleoprotein (NP) revealed significant suppression of viral infection in lung alveolar epithelial cells exposed to breathing motions 2 days following the introduction of virus on-chip compared to static chip controls (Fig. 4a). This inhibition by mechanical stimulation was further confirmed using qPCR and a plaque assay, which respectively show that the application of cyclic mechanical strain leads to a 50% reduction of viral mRNA in the epithelium (Fig. 4b) and ~80% reduction of viral titers in apical washes compared to static controls (Fig. 4c). This also was accompanied by a reduction in the production of the inflammatory cytokines, IL-6, RANTES (regulated on activation, normal T cell expressed and secreted), TRAIL (tumor necrosis factor-related apoptosis-inducing ligand or Apo 2 ligand), and IP-10, as measured in the vascular outflows from the chips (Fig. 4d). Moreover, similar inhibition of viral infection by cyclic mechanical strain was observed when the alveolus chips were infected with other respiratory viruses, including H5N1 influenza virus (Supplementary Fig. 7a) and the common cold OC43 coronavirus (Supplementary Fig. 7b).

Mechanical strain increases innate immunity in Lung Chips

To gain insight into how mechanical strain associated with physiological breathing motions might normally act to combat viral infection, we analyzed the RNA-seq analysis time course and noticed that exposure to continuous cyclic mechanical deformations resulted in increased expression of multiple IFN-related antiviral genes, including DDX58, MX1, OAS1, and STAT1, in both lung epithelial and endothelial cells from day 8 to day 14 of culture (Fig. 1f, g and Supplementary Fig. 5b). Indeed, many interferon-stimulated genes (ISGs) have higher expression in mechanically stimulated cells at day 14 on-chip but not in cells cultured in parallel in static Transwells (Supplementary Fig. 8a), and this was confirmed by qPCR analysis which showed increased expression of type I and type III interferons (Supplementary Fig. 8b).

To directly determine whether mechanical strain influences innate immunity pathway signaling, we cultured the alveolus chips in the presence or absence of cyclic mechanical strain from day 10 to 14 and performed RNA-seq analysis. Consistent with our earlier results, lung alveolar epithelial cells mechanically stimulated on-chip exhibit higher expression of ISGs and cytokines, such as MX1, MX2, IL6, CXCL10, CXCL5 (C-X-C motif chemokine ligand 5), IFI44L (interferon induced protein 44 like), and IFIH1 (interferon induced with helicase C domain 1), compared to static chip controls (Fig. 5a). Functional enrichment analysis of DEGs demonstrated that application of physiological cyclic strain activates pathways related to host defense response, while suppressing processes related to cell cycle and cell proliferation (Fig. 5b).

To further investigate the association between mechanical strain and innate immunity, we switched a subset of chips that had been exposed to 5% cyclic mechanical strain from day 10 to 14 to static conditions or to elevated mechanical strain (10%) condition on day 15 while continuing to mechanically stimulate the remaining chips at 5% for 2 additional days. The choice of 10% strain is based on our calculation^37,38,39 (Supplementary Data 4) that both low-tidal volume (6 ml/kg) and high-tidal volume ventilation (12 ml/kg) can lead to ~10% mechanical strain in the alveolus, which is often seen in patients with emphysema or lung edema due to increased lung compliance or surface tension⁴⁰. RNA-seq analysis of the epithelial cells on day 17 revealed that cessation of mechanical stimulation results in suppression of the innate immune response, exemplified by decreased expression of antiviral genes, such as IFIT1 (interferon-induced protein with tetratricopeptide repeats 1), IFIT2, OASL (2′−5′-Oligoadenylate Synthetase Like), and DDX58 (Fig. 5c and Supplementary Fig. 8c). On the other hand, elevating the level of mechanical stimulation to 10% strain results in modest upregulation of innate immune response genes (Fig. 5d, e and Supplementary Fig. 8d). In parallel, similar reduction in innate immune response pathway signaling is observed in endothelial cells when mechanical stimulation ceases (Fig. 5f and Supplementary Fig. 8e). However, increasing mechanical strain to 10% results in further upregulation of innate immune response genes (Fig. 5g, h and Supplementary Fig. 8f), including several inflammatory cytokines, such as CXCL3 (C-X-C motif chemokine ligand 3), CCL20 (C-C motif chemokine ligand 20), IL1A (interleukin-1 alpha), CXCL1 (C-X-C motif chemokine ligand 1), and CXCL5 (Fig. 5h). These results suggest that endothelial cells may be more sensitive to elevated mechanical strains than the epithelial cells. Together, these results demonstrate that cyclic mechanical forces similar to those experienced during breathing in the lung sustain higher levels of antiviral innate immunity in both lung alveolar epithelial cells and pulmonary microvascular endothelial cells compared cells cultured in the same chips under static conditions, which helps to explain why viral infection efficiency is reduced in the mechanically strained alveolus chips.

S100A7 mediates the effects of mechanical strain on lung innate immunity

We then leveraged this Organ Chip approach to explore the underlying mechanism responsible for these effects by examining our RNA-seq data sets. This analysis revealed that the S100 calcium-binding protein A7 (S100A7), which encodes a member of the S100 family of alarmins proteins and a ligand for RAGE, as one of the top genes differentially expressed in response to cyclic strain application in all experiments (Figs. 1f and 5a, c, e, and Supplementary Fig. 9a), and one that was not induced cells in static culture on-chip or in Transwells (Fig. 6a and Supplementary Fig. 9a). Chips cultured under physiological 5% strain from day 10 to 14 exhibit significantly higher mRNA levels of S100A7 in both epithelium and endothelium than when the cells were cultured under static conditions (Fig. 6b). This increase in S100A7 gene expression in response to mechanical stimulation is also associated with a significant increase in S100A7 protein secretion, as determined by quantifying its levels in apical washes from human alveolus chips using an enzyme-linked immunosorbent assay (ELISA) (Fig. 6c). Moreover, over-expression of S100A7 induces significant upregulation of the antiviral cytokines IFNβ1 and IFNλ1 in lung alveolar epithelial cells on-chip (Fig. 5d) as well as in conventional monoculture (Supplementary Fig. 9b) and in A549 lung cells, which similarly show both overexpression of S100A7 and reduced influenza H1N1 virus infection in a RAGE-dependent manner (Supplementary Fig. 9c, d). To directly examine the effect of S100A7 on gene transcriptional network, we performed RNA-seq analysis of alveolar epithelial cells on-chip after transfection of a plasmid expressing human S100A7. Differential gene expression analysis revealed that increased expression of S100A7 indeed upregulates many genes involved in the innate immune response, including IFNL1, CCL5 (C-C motif chemokine ligand 5), IFI6 (Interferon alpha-inducible protein 6), and IL6 (Fig. 6e, f). Therefore, S100A7 that is induced by breathing motions is able to drive the host innate immune response.

The mechanosensitive ion channel TRPV4 is upstream of S100A7

Cells can sense mechanical stimuli through diverse mechanosensitive molecules at the cell surface, including integrins, G protein coupled-receptors, growth factor receptors, and mechanosensitive ion channels (MSCs)⁴¹. The most rapid responses are mediated by MSCs that generate an ion flux in response to membrane deformation, and thereby regulate a wide variety of cellular processes, including innate immunity⁴². Mechanical strain has been reported to increase intracellular Ca²⁺ in lung alveolar epithelial cells and endothelial cells⁴³, and pharmacological or genetic modulation of calcium flux through the mechanosensitive TRPV4 ion channel was found to regulate pulmonary barrier function in the lung alveolus chip^5,12. Indeed, RNA-seq analysis confirmed that both cell types cultured in the lung alveolus chip express high level of TRPV4 as well as other MSCs, such as piezo type mechanosensitive ion channel component 1 (PIEZO1 and PIEZO2 (Supplementary Fig. 10). Importantly, when we perfused the TRPV4 inhibitor, GSK2193874, through the vascular channel of the alveolus chip while exposing the alveolar-capillary interface to 10% mechanical strain at 0.25 Hz for 48 h, the upregulation of S100A7 and CXCL10 was suppressed as measured by PCR in both epithelial cells (Fig. 6g) and endothelial cells (Fig. 6h), and this was accompanied by a reduction in the levels the inflammatory cytokines IL-6, IL-8, GM-CSF, and MCP-1 (Monocyte chemoattractant protein-1) in the vascular effluent (Fig. 6i).

To determine whether RAGE and TRPV4 mediate the effects of mechanical strain on innate immunity, we perfused the RAGE inhibitor azeliragon or the TRPV4 inhibitor GSK2193874 at non-toxic doses (Supplementary Fig. 11) through the vascular channel of the alveolus chip for 48 hours before infection with H3N2 while applying 5% mechanical strain (Fig. 7a). These studies confirmed that prophylactic inhibition of RAGE by pretreating the chips with either 20 or 100 nM azeliragon prevented the increase in cytokine production (Fig. 7b), although we observed that treatment with the higher dose resulted in a higher viral load (viral NP mRNA levels) when analyzed 48 h following infection (Fig. 7c). Interestingly, while pretreatment with the TRPV4 inhibitor also reduced cytokines, it decreased viral mRNA levels (Fig. 7b,c). This difference in viral load response suggests that TRPV4 is likely acting in multiple signaling pathways in parallel, and at least one other contributes to host response to viral infection via a distinct mechanism. In fact, prior work has shown that TRPV4 can influence viral infection by inducing nuclear translocation of DEAD-Box Helicase 3 X-Linked (DDX3X), a protein that promotes nuclear viral export and translation⁴⁴. Taken together, these results confirm that TRPV4 is one of the players that mediate the mechanotransduction process by which mechanical strain applied to the lung alveolar epithelium and endothelium results in activation of S100A7/RAGE pathway, and thereby drives an innate immune response.

RAGE inhibitors suppress viral infection-induced inflammatory responses

Interestingly, S100A7 gene expression and protein secretion are also upregulated in both epithelial cells and endothelial cells in response to infection of the alveolus chips with H3N2 influenza virus, and this is accompanied by upregulation of other S100 family members, including S100A8, S100A9, and S100A12 (Figs. 3b, d and 8a), which could further augment inflammation. As S100A7 and other members of the S100 family signal through RAGE⁴⁵, we next explored whether RAGE inhibitors can be used to suppress host inflammatory responses during viral infection in alveolus chips when administered the drugs in a therapeutic mode (2 hours after infection). One advantage of human Organ Chips is that they permit drug testing in a more clinically relevant pharmacological setting by enabling clinically relevant doses of drugs (e.g., maximum blood concentration or C_max) to be perfused through the vascular channel under flow as occurs in the vasculature of living organs⁴⁶ (Fig. 8b). When azeliragon, a RAGE inhibitor that demonstrated safety but not efficacy in Phase III clinical trials involving more than 2000 patients with Alzheimer Disease and diabetic nephropathy, was perfused at its C_max at 100 nM⁴⁷ through the vascular channel of human alveolus chips infected with H3N2 influenza virus while exposed to cyclic physiological (5%) strain, it significantly blocked induction of multiple cytokines, including IL-6, IL-8, IP-10 and RANTES (Fig. 8c), and similar effects are produced using a different RAGE inhibitor drug (FPS-ZM1) (Supplementary Fig. 12a). Interestingly, treatment with 100 nM azeliragon had no effect on viral load under these conditions (Fig. 8d), which is in contrast to when it was administered prophylactically (Fig. 7c). When we combined azeliragon with the potent ribonucleoside analog, molnupiravir, which demonstrates broad-spectrum antiviral activity against various RNA viruses, including influenza A (Fig. 8e), SARS-CoV-2, SARS-CoV, MERS-CoV, and Ebola^48,49, we found that the combination produced synergistic inhibition of inflammatory cytokine production following H3N2 infection (Fig. 8f and Supplementary Fig. 12b), whereas there was no added effect on viral infection (Fig. 8d and Supplementary Fig. 12c). Taken together, these data support the therapeutic potential for RAGE inhibitors, such as azeliragon, as suppressors of host-damaging inflammatory responses to viral infection when used alone or in combination with direct antiviral drugs.