Small airway-on-a-chip enables analysis of human lung inflammation and drug responses in vitro

Here we describe the development of a human lung 'small airway-on-a-chip' containing a differentiated, mucociliary bronchiolar epithelium and an underlying microvascular endothelium that experiences fluid flow, which allows for analysis of organ-level lung pathophysiology in vitro. Exposure of the epithelium to interleukin-13 (IL-13) reconstituted the goblet cell hyperplasia, cytokine hypersecretion and decreased ciliary function of asthmatics. Small airway chips lined with epithelial cells from individuals with chronic obstructive pulmonary disease recapitulated features of the disease such as selective cytokine hypersecretion, increased neutrophil recruitment and clinical exacerbation by exposure to viral and bacterial infections. With this robust in vitro method for modeling human lung inflammatory disorders, it is possible to detect synergistic effects of lung endothelium and epithelium on cytokine secretion, identify new biomarkers of disease exacerbation and measure responses to anti-inflammatory compounds that inhibit cytokine-induced recruitment of circulating neutrophils under flow.

here we describe the development of a human lung 'small airway-on-a-chip' containing a differentiated, mucociliary bronchiolar epithelium and an underlying microvascular endothelium that experiences fluid flow, which allows for analysis of organ-level lung pathophysiology in vitro. exposure of the epithelium to interleukin-13 (il-13) reconstituted the goblet cell hyperplasia, cytokine hypersecretion and decreased ciliary function of asthmatics. small airway chips lined with epithelial cells from individuals with chronic obstructive pulmonary disease recapitulated features of the disease such as selective cytokine hypersecretion, increased neutrophil recruitment and clinical exacerbation by exposure to viral and bacterial infections. With this robust in vitro method for modeling human lung inflammatory disorders, it is possible to detect synergistic effects of lung endothelium and epithelium on cytokine secretion, identify new biomarkers of disease exacerbation and measure responses to anti-inflammatory compounds that inhibit cytokine-induced recruitment of circulating neutrophils under flow.
The development of new therapeutics for chronic respiratory diseases such as asthma and chronic obstructive pulmonary disease (COPD), which pose a huge public health burden 1 , has been hindered by the inability to study organ-level complexities of lung inflammation in vitro. Although hospitalization and mortality due to these diseases are often consequences of exacerbations triggered by pathogens 2,3 , there is currently no way to study these processes in human lung outside of the clinical setting. Animal models of asthma and COPD exist; however, their clinical relevance is questionable because the anatomy, immune systems and inflammatory responses of animal lungs differ greatly from those in humans [4][5][6] . For example, mucin-producing cells, which are central to the development of asthma, are less frequent in the respiratory trees of mice and rats than in those of humans 6 . Neutrophils, which increase in number dramatically in the lungs of patients with COPD and severe asthma [7][8][9] , constitute only 10-25% of circulating leukocytes in mice, whereas they represent 50-70% in humans 5 . Because many animal models do not accurately predict drug activities in humans, the pharmaceutical and biotechnology industries strive to replace or minimize the use of animal models for drug testing whenever possible 10 .
Airway inflammatory diseases have been modeled in vitro using cultures of primary or immortalized human epithelial cells, sometimes positioned at an air-liquid interface to induce epithelial differentiation 11 , or using cocultures of airway epithelium and tissue-resident immune cells (for example, macrophages or dendritic cells) 12 . However, lung inflammation is mediated by organ-level responses involving complex tissue-tissue interactions between the lung airway epithelium and the underlying microvascular endothelium that modulate immune reactions to respiratory pathogens and allergens [13][14][15] and alter the vascular cell adhesion molecular machinery that recruits circulating immune cells, such as neutrophils. This is important because neutrophil accumulation in the lung is associated with enhanced severity of airflow limitation in people with COPD 7 , and it has a critical role in severe asthma as well 8 . Unfortunately, it is not possible to study complex interactions among airway epithelium, endothelium and circulating neutrophils using existing in vitro lung models because most fail to recapitulate normal functional coupling between epithelium and endothelium, and none enables analysis of the recruitment of circulating immune cells under active fluid flow. The latter point is crucial because neutrophil adhesion to inflamed endothelium involves initial E-selectin-mediated rolling along the luminal surface of endothelium, which is followed by firm adhesion to ICAM-1 (ref. 16), and this dynamic shear stress-dependent response cannot be studied in a physiologically relevant way using static cell cultures.
Advances in microsystems engineering have recently made it possible to create biomimetic microfluidic cell-culture devices, known as organs-on-chips, that contain continuously perfused microchannels lined by living human cells and which recapitulate the multicellular architectures, tissue-tissue interfaces, physicochemical microenvironments and vascular perfusion of the body 17 , potentially offering new opportunities for disease (b) Diagram of the tissue-tissue interface that forms on-chip, showing differentiated airway epithelium (pink cells) cultured on top of a porous collagen-coated membrane at an airliquid interface in the upper channel and the endothelium below (orange cells) with flowing medium that feeds both tissue layers. (c) A 3D reconstruction showing fully differentiated, pseudostratified airway epithelium formed onchip by cultured hAECs (green, F-actin) with human pulmonary microvascular endothelial cells (red, F-actin) on the opposite side of the membrane. Blue denotes DAPI-stained nuclei; scale bar, 30 µm. (d) The differentiated human small airway epithelium exhibited continuous tight junctional connections onchip, as demonstrated by ZO1 staining (red). Scale bar, 20 µm. (e) The human (lung blood microvascular) endothelial monolayer formed on-chip also contained continuous adherens junctions between adjacent cells, as indicated by PECAM-1 staining (green). Scale bar, 20 µm. (f) Well-differentiated human airway epithelium formed on-chip using hAECs derived from healthy donors demonstrating the presence of high numbers of ciliated cells labeled for β-tubulin IV (green) and goblet cells stained with anti-MUC5AC (magenta). Scale bar, 20 µm. (g) Scanning electron micrograph of cilia (blue) on the apical surface of the differentiated airway epithelium formed on-chip (nonciliated cells are in brown). Scale bar, 10 µm. (h) Sequential frames of a video of the apical surface of differentiated epithelium recorded over 100 ms showing cilia beating at a frequency of ~10 Hz. A single cilium is highlighted in white in each frame; time stamps and the white arrow indicate the duration and direction of one forward and return stroke (supplementary Video 4 shows the full recording). Scale bar, 5 µm (applies to all images in panel). All images are representative of three to six independent experiments performed on cells from three to six different donors. modeling and drug-efficacy assessment 18 . We previously used this approach to successfully reconstitute the alveolar-capillary interface of the human lung air sac and associated inflammatory responses in vitro 19,20 . Here we adapted this approach to microengineer a human lung small airway-on-a-chip that supports full differentiation of a columnar, pseudostratified, mucociliary bronchiolar epithelium composed of cells isolated from healthy individuals or people with COPD and underlined by a functional microvascular endothelium. results on-chip reconstitution of a functional small airway To construct the human small airway-on-a-chip, we used soft lithography to create a microfluidic device made of poly(dimethylsiloxane) (PDMS) containing an upper channel with a height and width of 1 mm (similar to the radius of a human bronchiole) separated from a parallel lower microvascular channel (0.2 mm high × 1 mm wide) by a thin (10 µm), porous (0.4-µm pores), polyester membrane coated on both sides with type I collagen (Fig. 1a). We cultured primary human airway epithelial cells (hAECs) on top of the membrane until confluent with medium flowing (60 µl h −1 ) in both channels. To trigger the differentiation of lung airway epithelial cells, we removed the apical medium after 5 d and introduced air to create an air-liquid interface (ALI). We added retinoic acid (3 µg ml −1 ) to the medium flowing in the lower channel to prevent the development of a squamous phenotype. Three to five weeks later, we seeded primary human lung microvascular endothelial cells on the opposite side of the porous membrane and cultured them at the same flow rate until confluent to create a tissue-tissue interface (Fig. 1a,b).
Immunofluorescence confocal microscopic analysis showed that these culture conditions resulted in the formation of a pseudostratified, mucociliary airway epithelium on one side of an extracellular matrix-coated membrane and a planar microvascular endothelium on the opposite side ( Fig. 1c-f, Supplementary  Fig. 1a and Supplementary Videos 1 and 2). The epithelial cells were linked by a continuous band of ZO1-containing tight junctions along the lateral borders of their apical surfaces (Fig. 1d), whereas endothelial cells were joined by adherens junctions containing PECAM-1 (Fig. 1e). The epithelium formed from cells isolated from the small airway epithelium of lungs from both healthy humans (Fig. 1f,g and Supplementary Video 2) and people with COPD ( Supplementary Fig. 1a) contained many ciliated epithelial cells as well as mucus-producing goblet cells, club cells (Supplementary Fig. 1b) and basal cells with proportions strikingly similar to those found in normal human lung ( Table 1). Immunodetection of tight junctions was accompanied by the formation of a robust epithelial barrier that restricted the passage of fluorescent inulin (4 kDa) and dextrans (10 and 70 kDa) between the parallel channels during analysis  Fig. 1c). Electron microscopic analysis of the normal lung small airway-on-a-chip confirmed that the cilia on the apical surface of the polarized epithelium had the same structure (9+2 microtubule organization) and length (~6 µm) (Supplementary Fig. 1d) as healthy cilia found in living human lung in vivo 21,22 . High-speed microscopic imaging confirmed that these cilia beat actively at a frequency of 9-20 Hz ( Fig. 1h and Supplementary Video 3), which again is nearly the same as that observed in human airways 21,23 (Table 1). Moreover, when we measured mucociliary transport by introducing fluorescent polystyrene microbeads into the top channel, we observed rapid coordinated movement of the beads along the surface of the epithelium (Supplementary Video 4) as a result of active synchronized cilia beating; again, the particle velocity measured (50-100 µm s -1 ) was nearly identical to that observed in healthy human lung airway (Table 1). Thus, the human lung small airway-on-a-chip effectively recapitulated many of the structures and functions of healthy lung bronchioles and sustained them for weeks in vitro.
modeling asthma and lung inflammation on-chip IL-13 has a pivotal role in asthma, as it is necessary and sufficient to induce all features of allergic asthma in animal models of the disease 24 . IL-13 has a direct effect on human airway epithelium 25 and participates in airway inflammation, goblet cell hyperplasia and mucus hypersecretion, as well as subepithelial fibrosis and airway hyper-responsiveness 24 . When we treated the small airway-on-a-chip with IL-13 (100 ng ml −1 ) for 8 d, we observed a significant increase in the number of goblet cells (Fig. 2a,b), in line with other in vitro models 26 ; higher production and secretion of the inflammatory cytokines G-CSF and GM-CSF into the vascular channel, as measured in the fluid effluent at days 2 and 8, respectively (Fig. 2c); and a decrease in cilia beating frequency ( Fig. 2d) similar to that observed in the airway mucosa of individuals with asthma [27][28][29] .
As mucosal inflammation and its exacerbation by viral infections are major components of both asthma and COPD, and because it has been challenging to recapitulate these clinically relevant responses in vitro, we analyzed the effects of exposing the airway epithelium to the viral mimic polyinosinicpolycytidylic acid (poly(I:C)) (10 µg ml −1 ). Poly(I:C) is an analogue of the double-stranded RNA that is produced in infected cells during viral replication by respiratory viruses 30 . The synthetic nature of our engineered lung airway model enabled us to analyze both independent and collective responses of the lung epithelium and endothelium to poly(I:C) challenge, which is not possible in human subjects. Stimulation with poly(I:C) for 3 h triggered a proinflammatory response in the chips similar to that in human subjects with acute severe asthma exacerbations 8 , including large increases in the basal secretion of RANTES, IL-6 and interferon-γ-inducible protein 10 (IP-10) (Fig. 2e) that were threefold to sixfold higher when underlying endothelium was present relative to when the endothelium was absent. This effect seemed to be specific for these cytokines in that removal of the endothelium had no effect on the secretion  npg of GRO-α or IL-8 (Supplementary Fig. 2a). We found that this effect was synergistic, as stimulated endothelial cells alone secreted similar or lower levels of cytokines compared with the isolated epithelium (Fig. 2e). Notably, although previous studies have investigated airway epithelium homeostasis using cocultures of endothelium and lung basal cells 31,32 , inflammatory responses in fully differentiated human small airway epithelium have not been demonstrated previously in vitro. In addition, poly(I:C) induced upregulation of E-selectin and VCAM-1 expression in the underlying endothelium ( Supplementary  Fig. 2b); these proteins are crucial for the initial adhesion and rolling of neutrophils under flow in microvessels in inflamed living tissues 33 . In fact, this upregulation resulted in enhanced adhesion, rolling and total recruitment of circulating human neutrophils in the small airway chip when primary human neutrophils were flowed through the lower channel under physiological flow and shear stress (1 dyn cm −2 ) (Supplementary Video 5).
Thus, these findings reveal intrinsic cross-talk between human epithelium and endothelium during poly(I:C)-induced inflammation and corroborate past animal studies suggesting that the pulmonary endothelium contributes significantly to viral-induced inflammation by acting as a central regulator of the cytokine storm 15 . Also, when we compared responses to poly(I:C) stimulation in bronchiolar versus bronchial epithelial cells, we found that the two types of cells displayed nearly identical secretion of inflammatory cytokines (Supplementary Fig. 2c).

modeling coPd exacerbation on-chip
As pathogenic infections are a major cause of COPD exacerbation in patients, we next explored whether the lung small airwayon-a-chip could be used to model these inflammatory responses in vitro. We stimulated small airway chips lined by either normal or COPD epithelial cells with the viral mimic poly(I:C) or with lipopolysaccharide endotoxin (LPS), which is a bacterial wallderived component that stimulates cytokine production and has been widely used to simulate bacterial infections in vitro 34 . We analyzed secretion of the cytokine M-CSF because it promotes the differentiation and survival of macrophages, and we studied the production of IL-8 because it attracts neutrophils; these are two major immune cell types that accumulate in the airway tissue of individuals with COPD 35,36 . Stimulation with LPS and poly(I:C) respectively upregulated the secretion of IL-8 and M-CSF in the COPD chips, but these infection mimics did not produce any significant change in chips lined by healthy airway epithelial cells (Fig. 3). These results are consistent with findings from past studies showing that bronchial epithelial spheroids from subjects (a) Total area covered by goblet cells, effects on production of the cytokines G-CSF and GM-CSF, and cilia beating frequency after exposure of the epithelium to IL-13 with or without dexamethasone (Dex) or tofacitinib for 8 d (skewed cytokine data were log transformed before comparison; data represent mean and s.e.m. (compared to IL-13-treated condition) of cells from three healthy donors, with one or two biological replicates per donor; for goblet cell analysis, three to five representative areas per condition were used for quantification; n = 3-9). (b) Neutrophil-adhesion results compared with neutrophil-recruitment data obtained when similar studies were carried out in static Transwell cultures. Data represent mean and s.e.m. (chip versus Transwell) of cells from two to four different COPD donors, with one to four biological replicates per donor and three to ten independent fields of view per chip or Transwell; n = 15-59. Bud, budesonide; BRD4, bromodomain-containing protein 4. (c) Effects of (10 nM) budesonide and (500 nM) BRD4 inhibitor on expression of genes encoding the endothelial cell adhesion molecules E-selectin, VCAM-1 and ICAM-1, compared to the untreated (0.1% DMSO) group, as measured using real-time PCR; human pulmonary blood microvascular cells were studied (n.s., not significant; data represent mean and s.e.m. of cells from one of the donors studied in b; three or four biological replicates per condition; n = 3-4). Significance determined by unpaired Student's t-test; *P < 0.05, **P < 0.01, ***P < 0.001; n.s., not significant. npg Figure 5 | Therapeutic modulation of inflammatory cytokine and chemokine production in the COPD small airway chip. Analysis of the effects of budesonide (Bud) and BRD4 inhibitor on the secretion of IL-8, MCP-1, GRO-α, IL-6 and GM-CSF into the microvascular channel of chips lined by COPD epithelium cultured at the air-liquid interface for 3-5 weeks when challenged with poly(I:C) (unpaired Student's t-test; *P < 0.05, **P < 0.01, ***P < 0.001; data represent mean and s.e.m. (compared to untreated controls) of cells from three different COPD epithelial and two different lung blood microvascular endothelial cell donors, with one to four biological replicates per donor; n = 6-8).
with COPD exhibit greater LPS-induced IL-8 release than those from healthy controls 37 , and the enhanced levels of M-CSF that we detected might partly explain the accumulation of macrophages observed in the lungs of people with COPD 9 . In contrast, stimulation with poly(I:C) produced large increases in secretion of the cytokines IP-10 and RANTES in both healthy and COPD chips (Fig. 3). These results are consistent with the finding that serum IP-10 is an excellent clinical marker for acute viral exacerbation in COPD patients 2 . Our finding that M-CSF levels specifically increased only in poly(I:C)-treated small airway chips (Fig. 3) suggests that M-CSF could be a new biomarker for COPD exacerbations induced by respiratory viruses.

evaluation of therapeutic responses in small airway chips
Having established these microfluidic models of human asthmatic and COPD airways, we then asked whether this new in vitro method could be useful for drug discovery. As IL-13 signals through the JAK-STAT pathway, we investigated whether tofacitinib, a potent inhibitor of JAK1, -2 and -3 used for patients with rheumatoid arthritis 38 , could reverse the IL-13-induced phenotype. When we exposed normal airway epithelium on-chip to IL-13 to produce asthmatic changes in the epithelium and then treated with high doses of tofacitinib (1 and 10 µM) to inhibit JAK signaling, we observed suppressed goblet cell hyperplasia, decreased secretion of G-CSF and GM-CSF, and restoration of normal cilia beating frequency (Fig. 4a). In contrast, when similar chips exposed to IL-13 were treated with the corticosteroid dexamethasone, it was found to be ineffective, which is consistent with the clinical finding that although patients with moderate to severe asthma receive dexamethasone inhalation therapy, they often fail to respond to the treatment 39 . These results also validate recent reports suggesting that immunosuppressant JAK inhibitors may represent potential new therapeutics for allergic airway diseases 40 .
To test the breadth and robustness of our platform, we also explored whether we could use the human small airway-on-a-chip model of COPD exacerbation by microbial infection to identify pharmacological agents able to suppress the associated inflammation, specifically by inhibiting the recruitment of neutrophils circulating under flow. When we treated COPD epithelial cells that had been stimulated with poly(I:C) on-chip with the glucocorticoid drug budesonide, there was no significant change in neutrophil adhesion (Fig. 4b), which again is consistent with the drug's noted lack of activity in many COPD patients 41 . We then tested a new experimental anti-inflammatory drug, 2-methoxy-N-(3methyl-2-oxo-1,2-dihydroquinolin-6-yl) benzenesulfonamide 42 , which is a bromodomain-containing protein 4 (BRD4) inhibitor of NFκB signaling. We chose this drug because it belongs to an emerging class of new anti-inflammatory drugs that were recently shown to alleviate lung inflammation in an LPS-challenged mouse model 43 . When the BRD4 inhibitor was added to the inflamed human small airway chip, it suppressed neutrophil adhesion under flow by almost 75% (Fig. 4b). However, when we tested these same drugs on similarly differentiated airway epithelium maintained with an underlying endothelium in static Transwell cultures, we found that the inhibitory effect of the BRD4 inhibitor on neutrophil recruitment was reduced by one-third, whereas the results obtained with budesonide were similar (Fig. 4b). This differential response to the BRD4 inhibitor under static versus dynamic flow conditions suggests that its mechanism of action might involve preferential inhibition of early adhesion and rolling responses that are required for initiation of the neutrophilrecruitment cascade because they occur only under dynamic flow conditions 44 . In fact, our studies showed that treatment of inflamed endothelial cells with the BRD4 inhibitor resulted in a more than 50-fold reduction in expression levels of E-selectin and VCAM-1, as well as a 50% reduction in ICAM-1, whereas budesonide treatment did not significantly decrease the expression of either of these vascular cell adhesion molecules (Fig. 4c).
Consistent with its ability to suppress neutrophil adhesion, the BRD4 inhibitor also proved greatly superior to the corticosteroid in downregulating expression of the genes encoding IL-8, MCP-1, GRO-α and IL-6 in human lung microvascular endothelial cells (Supplementary Fig. 3), as well as suppressing secretion of these cytokines into the lower microvascular channel (Fig. 5). These are important findings because these cytokines are key chemoattractants for neutrophils in vivo. The BRD4 inhibitor also significantly suppressed the production of GM-CSF (Fig. 5), which is a neutrophil chemokine that enhances the functionality of neutrophils 45 , in addition to supporting bone marrow cell proliferation and the survival of granulocytic precursors. Although budesonide slightly decreased the expression of selected cytokine genes (Supplementary Fig. 3), it did not significantly suppress cytokine secretion (Fig. 5), which is the most critical determinant of the inflammatory response; these findings again mimic results from animal models of COPD and lung inflammation 43,46 . Thus, reconstitution of the COPD inflammatory phenotype on-chip not only enabled testing for the efficacy of new experimental therapeutics but also allowed dissection of the mechanism of drug action at the molecular level in a human organ context in vitro.

discussion
Although attempts have been made in the past to culture human airway epithelium in microfluidic devices along with endothelium and fibroblasts 47 , the small airway-on-a-chip described here achieved greater robustness and fidelity in mimicking the Because the organ-on-chip approach essentially represents 'synthetic biology' at the organ level, we also were able to independently control and vary virtually all system parameters, including the presence or absence of different cell types, vascular flow conditions and soluble factors, while simultaneously analyzing human organ-level responses in real time with molecular-scale resolution. This microengineered lung small airway-on-a-chip model also faithfully recapitulated in vivo organ-level responses to therapy, and thus it offers a powerful complement to animal models for both analyzing human pathophysiology and carrying out preclinical drug evaluations (more details can be found in the Supplementary Discussion).

methods
Methods and any associated references are available in the online version of the paper. online methods Device fabrication. We created the upper and lower layers of the microfluidic device (Fig. 1a) by casting PDMS pre-polymer on molds prepared using stereolithography (Fineline, USA). Curing (10:1 PDMS base to curing agent, wt/wt) was carried out overnight at 60 °C to produce devices containing two adjacent parallel microchannels (top channel, 1,000 µm wide × 1,000 µm high; bottom channel, 1,000 µm wide × 200 µm high; length of overlapping channels, 16.7 mm), which were used to form the airway lumen and the microvascular channel, respectively. The channels were separated by a thin (~10 µm) semiporous polyester membrane (0.4-µm pores) that was purchased from Maine Manufacturing (USA). The membrane was laser cut, plasmatreated using Plasma Etcher PE-100 (Plasma Etch, USA) with one cycle of 50 watts of oxygen gas (15 standard cm 3 min −1 ) for 15 s, sandwiched between the carefully aligned top and bottom channels, and treated with ultraviolet light in a UVO-Cleaner (Model 342, Jelight, USA) for 30 min before a collagen coating was applied.
Microfluidic cell culture. Primary hAECs obtained from commercial suppliers (Promocell, Germany; Lonza, USA; and Epithelix, Switzerland) were expanded in 75-cm 2 tissue culture flasks using airway epithelial growth medium (BEBM; catalog no. CC-3171) supplemented with growth factors and/or supplements (BulletKit Supplements; Lonza) until 70-80% confluent. We isolated the lung epithelial cells, on the basis of information provided by the vendors, by dissecting airways from whole lungs or lung biopsies, subjecting them to enzymatic digestion and subsequently culturing them in a selective growth medium. The age and sex of healthy and COPD donors are presented in Supplementary Table 1. All of the COPD subjects whose epithelial cells were studied in the experiment described in Figure 3 had a history of smoking. Cells were randomly checked with a mycoplasma biochemical detection kit (Lonza, USA) and were found to be negative. Before cell culture, the porous membrane of the device was coated on both sides with rat tail collagen type I (300 µg ml −1 ; Corning, USA) at 37 °C for 24 h. The hAECs were then trypsinized and seeded onto the collagen-coated porous membrane in the upper channel of the device at a concentration of 2 × 10 6 to 5 × 10 6 cells ml −1 (2 × 10 5 to 5 × 10 5 cells cm −2 , or 4 × 10 4 to 10 × 10 4 cells per chip) and allowed to attach under static conditions. Three hours later, the cell monolayer was washed with fresh medium, and the cultures were maintained in a submerged state until fully confluent (typically 4-5 d after seeding) under constant flow (60 µl h −1 ) using an IPC-N series peristaltic pump (Ismatec, Switzerland). The chips were cultured in an incubator containing 5% CO 2 and 16-18% O 2 at 85-95% humidity.
When cultures were confluent, the apical medium was removed and an ALI was generated to trigger differentiation. The hAECs were maintained at the ALI for 3-5 weeks with a constant flow of growth medium in the bottom channel, which was sufficient to support epithelial cell viability and function for weeks in culture. The apical surface of the epithelium was rinsed once weekly with growth medium to remove debris. To control for variability in our studies, we ensured that all cultures reached a similar level of morphological and functional differentiation before we carried out any experiments. In our initial characterization studies, we determined that the primary human epithelial cells formed a stable and terminally differentiated ciliated epithelium within 3-5 weeks after being placed under an ALI, as defined using noninvasive microscopic imaging to confirm the formation of a confluent epithelial layer expressing actively beating surface cilia, and chips that failed to reach this standard by 5 weeks were discarded. We did not observe any notable difference in the differentiation responses of cells from different donors in our studies. Mucociliary transport was visualized using 1-µm-diameter fluorescent microbeads (Life Technologies, USA) diluted in phosphate-buffered saline (PBS), injected in the upper channel of a fully differentiated small airway-on-a-chip and imaged using a high-speed Hamamastu ORCA-Flash 4.0 camera mounted on a Zeiss AxioObserver Z1 microscope. For permeability measurements we flowed inulin (4 kDa)-FITC, dextran (10 kDa)-Cascade blue or dextran (70 kDa)-Texas red diluted in growth medium (100 µg ml −1 ) at 60 µL h −1 in the top channel for 24 h and measured the fluorescence intensity of medium flowing out of the top and bottom channels in three individual small airway chips. The apparent permeability was calculated using the following formula: where P app is the apparent permeability, J is the molecular flux, A is the total area of diffusion, and ∆C is the average gradient (~1 because of the low flow rate). Once differentiation of the hAECs was accomplished, human lung microvascular endothelial cells (Lonza; purchased as 'lung blood microvascular' endothelial cells before the supplier switched to the 'lymphatic + vascular' mix) were used for analysis of neutrophil adhesion and rolling, drug-efficacy testing and experiments involving COPD chips. We performed flow cytometric analysis and confirmed that our endothelial cells were >98% pure blood microvascular endothelial cells free of cells expressing lymphatic antigens (lymphatic vessel endothelial receptor 1 and podoplanin). Human umbilical vein endothelial cells (Angioproteomie, USA, or Lonza) were used for synergistic cytokine secretion studies; both types of endothelial cells were seeded onto the lower side of the porous membrane through the bottom channel at a density of 2 × 10 7 cells ml −1 (2 × 10 5 cells cm −2 or 4 × 10 4 cells per chip). Endothelial cells were seeded in Lonza EBM-2 endothelial cell basal growth medium (catalog no. CC-3156) supplemented with EGM-2MV SingleQuot Kit growth factors and/or supplements under static conditions with the chip oriented upside down to allow cell adhesion to the undersurface of the porous membrane. Two hours later, flowing of growth medium (60 µL h −1 ) was resumed until cells were fully confluent (usually 3-6 d after seeding). Whe found no significant difference in the cell densities in the epithelial and endothelial layers when we compared epithelium cultured alone to that cultured in the presence of endothelium.
Immunofluorescence microscopy. For immunofluorescence staining, we washed cells by flowing PBS through the top and bottom channels, fixed them with 4% paraformaldehyde (Electron Microscopy Sciences, USA) for 15 min without flow, and washed them gently by flowing additional PBS before storing them at 4 °C until use. The cells cultured on-chip npg nAture methods were washed again with PBS, permeabilized with 0.2% Triton X-100 (Sigma) in PBS for 2 h and exposed to blocking buffer composed of PBS containing 1% BSA (Sigma) and 5% FBS (Life Technologies, USA) for 30 min at room temperature. Cultures were stained for ciliated cells (anti-β-tubulin IV, 1:100, clone ONS.1A6; Genetex, Taiwan), goblet cells (anti-Mucin5AC, clone H-160, Santa Cruz Biotechnology, USA), basal cells (anti-Cytokeratin 5, Abcam, USA), tight junctions (anti-ZO1, clone 1A12; Life Technologies, USA) or PECAM-1 (clone WM-58; eBioscience, USA). When cell-adhesion molecules were stained on endothelial cells, the cells were not permeabilized to visualize surface antigens. In all studies, antibodies diluted in blocking buffer were introduced in the channels and incubated for 1 h at room temperature or overnight at 4 °C, and cultures were washed with PBS three times each for 5 min at room temperature. Secondary antibodies (Life Technologies, USA) were incubated for 45 min at room temperature and washed three times with PBS. In some studies, TO-PRO-3 (Life Technologies, USA) was used to label nuclei after secondary antibody staining. Devices were then cut using a razor blade, and membranes were delicately lifted from the PDMS and mounted on microscopy slides in mounting medium (Vectashield; Vector Laboratories) with or without DAPI. Fluorescence imaging was carried out using confocal laserscanning microscopy (SP5 X MP DMI-6000, Germany). Image processing and three-dimensional Z-stack reconstruction were done using Imaris (Bitplane, Switzerland) and ImageJ software (downloaded from http://imagej.nih.gov/ij/). When two primary antibodies of the same species were used, the Zenon antibody labeling kit (Life Technologies, USA) was used to preconjugate the antibodies separately before addition to fixed cells. For IL-13 experiments, cultures were treated with IL-13 (Peprotech, USA), dexamethasone (Sigma-Aldrich, USA) or tofacitinib; fixed after 8 d of treatment; and stained for MUC5AC and DAPI. Tofacitinib was purchased commercially (CP-690550, Selleckchem, USA) and was not obtained from Pfizer, and the doses were optimized in-house. We quantified goblet cell hyperplasia by measuring the percentage of the projected area covered by goblet cells in three to five different fields for each condition.
Scanning and transmission electron microscopy. For electron microscopic analysis, the porous membranes supporting the cell cultures were excised from the devices with a razor blade, and cells were fixed with 2.5% glutaraldehyde (Electron Microscopy Sciences, USA) in 1% sodium cacodylate (Sigma) for 1 h at room temperature. Fixed cells were rinsed with 1% sodium cacodylate and post-fixed with 1% osmium tetroxide (Electron Microscopy Sciences, USA) in 1% sodium cacodylate for 90 min in a fume hood. Cells were dehydrated sequentially in ethanol gradients; rinsed in hexamethyldisilazane; air-dried overnight in a desiccator at room temperature; and then mounted on a conductive carbon support, coated with gold and imaged with a VEGA III scanning electron microscope (Tescan, Czech Republic).
For transmission electron microscopy, fixed cells were embedded in Taab 812 resin (Marivac Ltd., Nova Scotia, Canada) and incubated at 60 °C for 1 d. Samples were cut in 80-nm sections with a Leica ultracut microtome, picked up on 300-mesh formvar-carbon-coated copper grids, stained with 0.2% lead citrate, and viewed and imaged under a Philips Technai BioTwin Spirit electron microscope.
Analysis of ciliated, goblet and basal cells and of mucociliary transport. Cilia beating frequency was measured using a high-speed Hamamatsu ORCA-Flash 4.0 camera mounted on a Zeiss AxioObserver Z1 microscope. The epithelium was washed with warm PBS before analysis to avoid mucus accumulation and slowing of cilia beating. High-speed videos of beating cilia were recorded at approximately 200 frames per second and played at 10 frames per second for 1 s to allow manual counting. Four to five random areas of each chip were counted, and cilia beating frequencies were averaged for each condition. For IL-13 treatment experiments, cilia beating frequency was recorded and measured after 8 d of treatment (n = 3-4 chips). Epithelial cell types were quantified as previously described 50 . Briefly, for ciliated cells and basal cells, fully differentiated epithelia from three small airway chips were trypsinized, centrifuged (1,000g for 2 min), air-dried for 1 h at room temperature, fixed with 100% icecold acetone, rinsed with PBS, stained for β-tubulin IV (ciliated cell marker) or Cytokeratin 5 (basal cell marker) and counterstained with DAPI. Ciliated cells and basal cells were counted in six fields per chip, and percentages were calculated. For goblet cells, fully differentiated epithelia from four chips were fixed with 4% paraformaldehyde and stained for MUC5AC (goblet cell marker). We quantified goblet cells by measuring the percentage of the projected area covered by MUC5AC staining in five different fields for each condition. We evaluated mucociliary transport by measuring the displacement of 1-µm polystyrene fluorescent beads diluted in PBS and introduced in the top channel of the small airway chip for 1 s as shown in Supplementary Video 4.
Gene expression analysis. Total RNA was extracted from differentiated epithelial or endothelial cells in situ on-chip with the RNeasy mini kit (Qiagen, USA). The RNA was incubated with DNase I (Qiagen, USA) for 15 min at room temperature to remove residual contaminating genomic DNA, enzyme heat-inactivated at 65 °C for 5 min, and then reverse transcribed into cDNA using the SuperScript Reverse Transcriptase III kit (Invitrogen, USA). We primed cDNA synthesis by mixing 1 µl of oligodT (50 µM) and 1 µl of dNTP mix (10 mM) with up to 500 ng total RNA. Realtime PCR was carried out using a CFX-96 real-time PCR system (Bio-Rad, USA). Reactions contained 2 µl of cDNA, 10 µl of 2× Universal SYBR Green Supermix (Bio-Rad, USA), 3 µl of each forward and reverse primer (300 nM final concentration), and 2 µl of molecular biology-grade water, and results were quantified using the 2 −∆∆Ct method 51 . Primers were either designed using the online Primer3 application (http://frodo.wi.mit.edu/primer3/) or used from previous reports 52 , and sequences are presented in Supplementary Table 2. Analysis of chemokines and cytokines. The effluent of flowing medium was analyzed for a panel of cytokines and chemokines (IL-8, IP-10, RANTES, IL-6, M-CSF, G-CSF, GM-CSF and GRO-α) using custom Milliplex assay kits (Millipore, USA). Analyte concentrations were determined according to the manufacturer's instructions using a LuminexFlexMap 3D system coupled with Luminex XPONENT software (Luminex, USA). For endothelium depletion experiments, basal secretions were collected for each condition, and amounts of RANTES, IL-6 and IP-10 were measured at 24 h after poly(I:C) (InvivoGen, USA) treatment. For Toll-like receptor stimulation experiments, COPD and healthy npg epithelia were challenged with LPS (10 µg ml −1 ) or poly(I:C) (10 µg ml −1 ) for 1 h, and amounts of secreted IL-8, M-CSF, IP-10 and RANTES were measured 24 h after treatment. For COPD drug studies, in chips containing cocultures of differentiated COPD epithelium and microvascular pulmonary endothelium, the endothelial cells were treated with 10 nM budesonide (Sigma), 500 nM BRD4 inhibitor (provided by Pfizer) or 0.1% DMSO diluent (Sigma) under flow (60 µL h −1 ) through the vascular channel for 24 h before poly(I:C) (10 µg ml −1 ) was delivered into the airway channel for 6 h. The vascular effluents were then collected for cytokine and chemokine analysis.
Analysis of neutrophil adhesion. Neutrophils were isolated from fresh human blood by two-step gradient sedimentation. The peripheral blood mononuclear cells were removed from the polymorphonuclear cell population using Ficoll (Stem Cell Technologies, Canada) density centrifugation, and then the polymorphonuclear cells were isolated using a modified Percoll (Sigma) protocol 53 . Flow cytometry analysis for CD15 and CD16 expression confirmed that the purity of the neutrophil population was >93%. Isolated neutrophils were then live-stained for 30 min at 37 °C using CellTracker red or Hoechst (Life Technologies, USA), resuspended in RPMI medium containing 10% FBS (vol/vol) (Life Technologies, USA) and used within 3 h for recruitment and adhesion assays in chips. The chips were flipped upside down, and neutrophils (1 × 10 7 cells ml −1 ) were flowed (2.7 ml h −1 ; 1 dyn cm −2 ) through the microvascular channel of the device to mimic the physiological hemodynamic conditions in human postcapillary venules 33 . After 10 min, unbound neutrophils were washed away by flowing cell-free RPMI-10% FBS medium for 5 min, and micrographs of four or five random areas were taken for subsequent counting; we carried out quantification by counting attached neutrophils using ImageJ (http://imagej.nih.gov/ij/) and CellProfiler (http://www.cellprofiler.org) software. For experiments using Transwells, the inserts were placed upside down, and 50 µl of neutrophil suspension (1 × 10 7 cells ml −1 ) was added gently to endothelium on the undersurface of each insert and incubated for 10 min at 37 °C. The inserts were then reverted to the upright position in cell culture medium for 5 min to remove unbound neutrophils and subsequently imaged and analyzed as chips. Micrographs of three to ten random areas were taken to count adherent neutrophils.

Statistical analysis.
All results and error bars are presented as mean and s.e.m. Data were analyzed with an unpaired Student's t-test using Graphpad Prism (GraphPad Software Inc., San Diego, CA, USA) or Excel software (Microsoft, USA). Differences between groups were considered statistically significant when P < 0.05. Variability between biological replicates (for example, in Fig. 3) for a given donor was minimal (s.e.m. < 5% of mean).