Ventilation‐induced epithelial injury drives biological onset of lung trauma in vitro and is mitigated with prophylactic anti‐inflammatory therapeutics

Abstract Mortality rates among patients suffering from acute respiratory failure remain perplexingly high despite the maintenance of blood oxygen homeostasis during ventilatory support. The biotrauma hypothesis advocates that mechanical forces from invasive ventilation trigger immunological mediators that spread systemically. Yet, how these forces elicit an immune response remains unclear. Here, a biomimetic in vitro three‐dimensional (3D) upper airways model allows to recapitulate lung injury and immune responses induced during invasive mechanical ventilation in neonates. Under such ventilatory support, flow‐induced stresses injure the bronchial epithelium of the intubated airways model and directly modulate epithelial cell inflammatory cytokine secretion associated with pulmonary injury. Fluorescence microscopy and biochemical analyses reveal site‐specific susceptibility to epithelial erosion in airways from jet‐flow impaction and are linked to increases in cell apoptosis and modulated secretions of cytokines IL‐6, ‐8, and ‐10. In an effort to mitigate the onset of biotrauma, prophylactic pharmacological treatment with Montelukast, a leukotriene receptor antagonist, reduces apoptosis and pro‐inflammatory signaling during invasive ventilation of the in vitro model. This 3D airway platform points to a previously overlooked origin of lung injury and showcases translational opportunities in preclinical pulmonary research toward protective therapies and improved protocols for patient care.

across the alveolar-capillary barrier into the wider systemic circulation. 3 In turn, inflammatory effects may spread and amplify throughout the body leading to multi-organ failure and eventually death. 3 Biotrauma was first proposed in the context of invasive mechanical ventilation; a life-supporting clinical intervention also recognized to concurrently cause or worsen lung morbidity. 2 Most recently, mechanical ventilation has gained increased scrutiny amidst the COVID-19 pandemic, [5][6][7] owing to alarmingly higher mortality rates among patients requiring respiratory support 8,9 ($50%-97%). The main physical mechanisms identified as contributing to pulmonary injury 10 include the overstretching during ventilation of lung tissue from over inflation (known as volu-or barotrauma) and the repeated opening and collapse of small airway units at excessively low volumes (i.e., atelectrauma). Various protective ventilation protocols have emerged to marginally improve patient outcomes and reduce mortality. 11 Yet eliminating injurious mechanical forces is elusive as underlined in clinical trials. 12 Recent efforts have begun exploring therapeutic opportunities (e.g. gene delivery) to mitigate lung injury in the deep alveolar regions. 13 Still, it remains widely unknown to what extent mechanical forces influence the downstream precursors leading to biotrauma. Indirect evidence from animal models 14,15 and clinical studies 11,16 has provided seminal support for the biotrauma hypothesis. However, a direct corroboration in vivo is challenging since lung-derived inflammatory biomarkers in patients are not readily collected in clinical settings in detecting the origins of cytokine release and trafficking. 17 In an effort toward bridging this gap, engineered in vitro lung models leveraging advances in tissue engineering and microfabrication have been increasingly utilized for advancing preclinical pulmonary research and explore lung injury. 18,19 For example, in vitro studies have characterized wounding in alveolar epithelial cells subjected to cyclic overstretching, including the release of inflammatory cytokines, impaired structural integrity via changes in tight junctions and plasma membrane breaks, and increased incidence of apoptosis. [20][21][22] Cellularized in vitro models have typically focused on the airway epithelial barrier of the deep acinar regions, 23 where mechanical stresses are dominated by tensile strains when the alveolar airway barrier expands and contracts cyclically. 24 In contrast, there has been little emphasis on the lungs' proximal regions (i.e., upper airways), where airflow dynamics are most prominent. Namely, the exposure of bronchial epithelial cells to respiratory flow-induced shear stresses has been proposed as a potential link between large ventilation pressures and morbidity and mortality 25 ; a situation strongly correlated in ventilated patients undergoing surgery in the absence of prior lung injury. 26 In support, computational fluid dynamics (CFD)-based in silico studies have found that upper airway flow phenomena and ensuing wall shear stresses (WSS) may contribute to ventilator-induced lung injury (VILI) via the biotraumatic pathway. [27][28][29] We have also recently underlined the plausible occurrence of such injury during invasive endotracheal intubation maneuvers. 30 Nevertheless, in silico studies are limited in addressing inflammatory cascades arising in conducting airways and have instead relied on extrapolation from hemodynamic studies with endothelial cells, limiting their clinical relevance. 31 In the present work, we explore for the first time the hypothesis of potential immunological mediators originating in the upper respiratory tract of preterm infants as a result of ventilatory flow-induced shear stresses. Preterm infants are particularly susceptible to VILI as a result of potentially extended periods of ventilatory support. 32 To this end, we developed a true-scale, three-dimensional (3D) bronchial epithelial airway in vitro model of the tracheobronchial tree subject to physiologically realistic ventilatory protocols in intubated pediatric populations that are prone to VILI. 32 We expose injurious effects of flow-induced WSS on the epithelial airway barrier populating the 3D airway lumen, leveraging phenotypical endpoints of epithelial structural integrity, cell apoptosis, and importantly the secretion of cytokines associated with inflammatory pathways. Our in vitro assays support the manifestation of shear flow-induced lung injury during mechanical ventilation, thus strengthening the biotrauma hypothesis.
Furthermore, our findings may help associate rare but acute adverse effects (e.g. necrotizing bronchitis, epithelial erosion, and loss of surface cilia) observed in clinical 33 and animal 34 studies on ventilation associated injury. To mitigate the initiation of such inflammatory cascades during ventilation, we demonstrate as a proof-of-concept the topical delivery of a widely used anti-inflammatory respiratory therapeutic as a prophylactic strategy for preventive action that supports opportunities in preclinical pulmonary research toward protective therapies. using an airway epithelial cell line permits experimental control and reproducibility as a preclinical benchmark by reducing variability in cell cultures arising from donor-to-donor differences 41 while maintaining key features of the bronchial epithelium in human lungs (e.g., tight junctions, mucus secretion, cytokine production and more).
Models were cultured under immersed conditions and tracked over 3-4 weeks until a fully confluent epithelial monolayer populated the entire 3D airway lumen ( Figure 1f). As recently shown in silico, jet-flow impaction in airways can be significant during invasive mechanical ventilation. 30 Hence, monitoring cell structural integrity is critical (see tight junction occluding protein staining in Figure S2), in particular, at the bifurcations (e.g., main carina in Figure 1f-h). Our development of more realistic 3D in vitro airway morphologies is supported by recent studies where Calu-3 cells (as well as other epithelial cells) cultured on curved membranes and inside lumen exhibit distinct characteristics from traditional two-dimensional monolayers, including cell density and shape, apoptotic ratios, and cross-sectional morphology. 42,43 Here, we specifically designed a planar bifurcating geometry to facilitate microscopy imaging by limiting the model's vertical dimension, whose 3D curvature of the inner lumen extends outside the depth of focus under higher magnifications ( Figure 1h).
The total volume enclosed within the model is 1.5 ml, efficiently and robustly removed or exchanged via plastic connector ports inserted into all outlets for model maintenance, allowing for the simple removal of medium and cell collection for further analysis such as ELISA and flow cytometry (see below).

| In vitro and in silico flow dynamics reveal focal shear stresses under invasive mechanical ventilation
The intricate flow patterns in the human respiratory airways are mainly driven by the interaction of time-dependent flow fields in the upper airway generations (starting in the trachea) with changing geometries, including changes in cross-sectional areas, wall curvatures, and carinal edges across the airway tree. 36 Figure 2f. Here, local WSS values exceed 1500 dyn cm À2 ; namely two orders of magnitude higher than levels found to impair epithelial permeability in human and mice bronchial epithelial cells. 44 We find lower (<300 dyn cm À2 ) WSS values concentrated in the two daughter bifurcations, an $80% attenuation relative to the first generation due to viscous energy dissipation and anticipated to occur in any symmetrically branching channel system. 45 We performed four additional simulations with smaller tidal volumes but identical flow rates (i.e., breathing frequency is increased according to the linear relationship Q ¼ f Â TV, where Q is the flow rate, f the frequency, and TV the tidal volume 30 ) and tracked the peak WSS values at the main carina. In Figure 2g, results are plotted as a F I G U R E 2 In vitro and in silico fluid dynamics analysis reveals a region of elevated shear stress concentrated at the tracheal carina during the inspiratory phase of intubated ventilation in a neonatal-sized airway model. (a) A sinusoidal flow profile mimicking oscillatory mechanical ventilation is imposed on the model via perfusion at the inlet (trachea). The first quarter of the cycle (see red arrow) marks the inspiratory phase's peak strength when air is pushed into the model. The flow is visualized and measured experimentally using tomographic particle image velocimetry (TPIV), demonstrated with a raw image (b) from one of four cameras in the TPIV setup (see Section 4 and Figures S3 and S4), showing illumination of 10 μm diameter fluorescent particles captured instantaneously while tracing the streamlines of the flow. Following image analysis (i.e., image pre-processing and TPIV algorithms), the three-dimensional transient flow is fully resolved. The flow field at peak inspiration is plotted in (c), with several orthogonal cut planes colored by the normalized velocity magnitude contour field and overlayed with velocity vectors. (d) An in silico, that is, computational fluid dynamics (CFD), solution is compared with the experimental data. During peak inspiration, the most dominant flow feature is captured by a synthetic jet exiting the endotracheal tube and impacting the first carina. A numerical solution allows for finer near-wall resolution and analysis of WSS, mapped by colored contours in (e) with a top view of the first bifurcation shown in the inset (f). Panels (b)-(f) feature a flow analysis of a representative low-frequency ventilation protocol most similar to conventional mechanical ventilation used in cellular in vitro experiments. In (g), maximum WSS levels measured at the first bifurcation are plotted as a function of normalized ventilation frequency, or α (i.e. Womersley number), for five ventilation protocols solved using CFD (see Figure S5) function of the Womersley number, a nondimensional frequency where D is the airway diameter and ν the kinematic viscosity. A reduction in maximum WSS values relative to α (varied between 2 and 5) indicates that lower tidal volumes at higher frequencies attenuate the flow jet's effects, implying a possible link to the acknowledged protective benefits of low tidal volume ventilation. 46 We note that negative pressure ventilation, 47 an alternative ventilatory support technique that drives gas flow into the lungs by reducing pressure around the patient's chest and thus more closely mimics normal physiological breathing, is fundamentally different from intubated, positive pressure driven ventilation described here and is thus anticipated to give rise to different WSS distributions that would resemble more closely normal breathing features.
We next compare these findings to previous numerical studies on flow-induced shear stress in human airways and find similarly reported patterns of local peaks at the bifurcating sites. 48 In an airway model of generations 3-5, elevated WSS ($5 dyn cm À2 ) were reported localized at the bifurcations, 49  were amplified more than twice in cases where ETT tubes were placed asymmetrically more proximal to the tracheal wall, indicating our relatively conservative approach for the ETT placement.  Before exposure to either ventilation or control conditions, models are imaged, and their culture medium collected for subsequent cytokines analysis, serving as a reference point for a priori conditions. (b) Next, the model is prepared for clinical ventilation by installing compliance adapters (latex balloons) at the four outlets (to simulate pressure conditions present in a full lung) and inserting a 3.0 mm ID uncuffed endotracheal tube (ETT) 2 cm above the first carina (marked B1), following clinical guidelines for proper insertion depth. The two daughter branches are marked B2L (left) and B2R (right) following anatomical orientation convention. Note that during ventilation, the ETT tip is placed symmetrically while here shown at a 90 rotation for clarity. (c) Models are ventilated inside the incubator with medical-grade air, supplied via a breathing circuit connected to a neonatal mechanical ventilator for 4 h using settings defined in Table 1. After the exposure, models are filled with medium and returned to the incubator for an additional 24-48 h before further analysis and injury characterization, including microscopy imaging. (d-f) Changes in the secretion of inflammatory cytokines following ventilation suggest a signaling pathway. Cytokines IL-6 (d), IL-8 (e), and IL-10 (f) were measured in culture medium collected from the models after 24 and 48 h periods before and following simulated ventilation, respectively. Measurements from repeated experiments are plotted as normalized fold change to highlight signaling changes resulting from the test exposure. In (d) and (e), we measured a significant reduction in pro-inflammatory IL-6 (75% at both 24 and 48 h) and IL-8 (50% at 24 h) relative to the control, while conversely in (f), we find that IL-10, an anti-inflammatory cytokine, is elevated >30% at the 24-h time point compared with the control. All values are normalized relative to their baseline, that is, divided by pre-exposure secretory levels. Cytokines were measured using ELISA. All graphs show mean (SD) values. *p < 0.05; ** p < 0.01; *** p < 0.001; ****p < 0.0001 cytokine secretory events were observed to be most prominent in the first 24 h following the exposure, while at 48 h effects were found diminished, indicating the transitory nature of the response to a single, 4-h ventilation exposure protocol. Notably, IL-10 is known to play a crucial role in regulating immune responses by limiting and ultimately terminating inflammatory events via the inhibition of cytokine synthesis. 61 One possible explanation for our findings is that at 24-h poststimulus, pro-inflammatory cytokine production is inhibited by IL-10 due to an anti-inflammatory response.

| Prophylactic leukotriene receptor antagonist reduces cell death and modulates the secretion of inflammatory cytokines
After establishing the presence of an inflammatory response to flowinduced shear stress in our model (Figure 4), we next sought a direct modulatory agent that could demonstrate the model's application for In Figure 5d-f, we report the effect of prophylactic Montelukast treatment on epithelial cell cytokine secretions. In the ventilated group, where we previously measured a reduction in proinflammatory IL-6 and IL-8 following ventilation exposure, we see a similar reduction (Figure 5a,b) although values of IL-6 are significantly higher (67%) in the pre-treated group relative to nontreated. We note that while IL-8 secretory levels were unchanged in the ventilated group following prophylactic treatment, a reduction of 50% was measured in the control (i.e., no ventilation exposure) group. Here, we may speculate that Montelukast acts to inhibit IL-8 in respiratory epithelial cells, as similarly reported in a previous review on the antiinflammatory activities of β2-agonisits. 63 For IL-10 (Figure 5f), we F I G U R E 5 A prophylactic leukotriene receptor antagonist (Montelukast) reduces cell apoptosis and modulates inflammatory cytokine secretion in bronchial epithelial airway models exposed to ventilation injury. Montelukast, an asthma anti-inflammatory medication commonly used to treat pulmonary asthma, was supplemented in the culture medium of models for 2 h before ventilation exposure. Flow cytometry was performed on cells collected from models 24 h following the ventilation/control and stained with Annexin-V and Propidium Iodide (PI), indicating % of the cell population (a) live, (b) apoptotic (c), or necrotic. Montelukast pretreatment significantly increased (55%) the number of live cells in the ventilated group by reducing apoptosis (75%) while showing no adverse impact on the control group's high live cell count. (c) An increase in necrotic cells (70%) was observed in the ventilated group while absent in control, with a negligible effect on %total of live cells due to low overall counts (maximum of 5% total population). (d-f) Cytokine secretions were measured in the culture medium collected from all models 24 h after the ventilation exposure. Cytokines IL-6 (d), IL-8 (e), and IL-10 (f) are plotted as fold change normalization to highlight changes in signaling resulting from the test exposure. (d) We report no significant difference in IL-6 secretion in the control groups, while an increase in IL-6 secretion (67%) was measured in the pre-treated ventilated group relative to nontreated. (e) In contrast to IL-6, IL-8 secretion was found reduced in the control group (50%) following prophylactic treatment, while we measured no difference due to treatment in the ventilated groups. Lastly, we measured the suppressory IL-10 secretion, with mean values unaltered in the control groups but reduced by 30% following prophylactic treatment with Montelukast. All cytokine secretion values are plotted as fold change, i.e., normalized relative to their baseline preexposure levels. Cytokines were measured using ELISA. All graphs show mean (SD) values. * p < 0.05; ** p < 0.01; *** p < 0.001; ****p < 0.0001 measure a reduction (30%) to near-baseline levels in the prophylactically treated ventilated models, a reversal relative to the nontreated exposure group. These results suggest that the inflammatory response elicited in our models by ventilation exposure is directly modulated by prophylactic treatment using Montelukast.

| DISCUSSION AND CONCLUSION
This study aimed to shed light on the physical pathways contributing to biotrauma; the type of VILI most poorly defined and understood. 4 Preterm infants are particularly susceptible to extended periods of ventilatory support; the more immature the newborn is at birth the longer the likely period of ventilation and thus the increased risk of respiratory distress severity. 64 Moreover, preterm infants are considered more susceptible to external stressors as their respiratory organs are not fully matured and functional. 64  We designed and constructed a 3D tracheobronchial airway model featuring a fully confluent cultured epithelial cell monolayer. We To isolate the effects of mechanical stress from intubated ventilation, we exposed our models to 4 h of clinical ventilation conditions in air; a duration that was short enough to avoid introducing interference owing to the removal of cells from maintenance conditions, that is, immersion in culture medium but sufficiently long to observe a clear cellular response. Furthermore, a 4-h exposure protocol follows previous studies investigating intubated ventilation in "one hit" in vivo 15,71 and in vitro models, 72 facilitating the interpretation of our results relative to the broader literature. In the clinical setting, patients can be supported by a ventilator for days or even weeks, likely amplifying the effects measured in our in vitro models following exposure to 4 h alone. In analyzing our models following exposure to ventilated conditions, we first explored changes to the epithelium layer's structural integrity. We found evidence of cell denudation localized primarily at the main carina, matching our flow analysis that identified concentrated levels of high shear stress at the same site, at maximum during the inhalation phase of simulated ventilation. The deleterious effects of a jet stream in the tracheobronchial region were first suspected in the context of high-frequency jet ventilation (HFJV); a strategy of delivering short pulses of pressurized gas directly into the upper airway through a custom-designed endotracheal lumen. Proposed initially to improve gas exchange and thereby reduce the severity of VILI in both infants 73 and adults, 74 HFJV is no longer recommended for use following a meta-analysis that failed to find an advantage in reducing mortality rates, 75  To further explore the cellular response in our model and the possible development of biotrauma following invasive ventilation, we measured cell death (apoptosis/necrosis assay) and cytokine secretions. Models were analyzed at 24 and 48 h following exposure, with viability in the ventilated group reduced by 60%, with more than half to apoptosis, while the controls retained over 95% live cells. Measurement of cytokines secreted in the culture medium revealed increased secretion of an anti-inflammatory mediator (IL-10) while simultaneously measuring reduced secretions of pro-inflammatory IL-6 and IL-8 linked to ventilation exposure. 46,61 Owing to its antiinflammatory effects, IL-10 has been proposed as a potential therapeutic after showing protective benefit in mice given nebulized IL-10 before injurious ventilation. 15 Mechanical ventilation is generally considered inflammation promoting, with different strategies eliciting different inflammatory responses measured by increased production of cytokines TNF-α, IL-1β, IL-6, and IL-8. 77 Interpreting our cytokine measurements could indicate the beginning of a healing process 56 within our model following a 4-h exposure period, though the role of mediators involved in respiratory inflammation and VILI remains controversial owing to inconsistent results. 15 (Figure 1d). More details can be found in a previous study using similar silicone phantom fabrication methods. 30 Following the complete dissolution of the last remaining printed material, the PDMS scaffold was left to dry in a well-ventilated space (i.e., chemical hood) for >72 h. PDMS material swells during immersed conditions in acetone, an effect reversed with sufficient drying time. Furthermore, we found that acetone's inadequate evaporation due to shorter drying periods disrupted later cell growth and prevented full monolayer confluence.

| Model coating
Twenty-four hours before cell seeding, the inner surface of the PDMS phantom models was coated with human plasma fibronectin (#354008) and collagen from bovine skin (#C4243, Sigma-Aldrich) following a coating study for optimal growing conditions of Calu-3 directly on PDMS surfaces (see Figure S1). To coax cells to grow directly on PDMS, we conducted a coating comparison study between cells growing in a 24-well plate. We took bright-field microscopy images of the wells 3-, 5-, and 7-days following cell seeding on three different coatings: 10% fetal bovine serum (FBS), 1% v/v collagen, and a combination of 1% v/v collagen, 1% v/v fibronectin in addition to no coating. The combined fibronectin and collagen coating was found to support full confluence consistently following 7 days and was subsequently used in all models.

| Numerical methods
The system of governing equations (i.e., conservation of mass and momentum) is solved numerically using a commercial software's FVM (ANSYS Fluent v19.2). The momentum equations are discretized using the second-order upwind scheme for velocity and second-order scheme for pressure, whereas, for coupling the velocity and pressure fields, the Coupled algorithm is applied along with a least-squares-based scheme for gradients. For numerical modeling purposes, the mouth opening is treated as a velocity inlet, and two bifurcation exits as pressure outlets. Cyclic flow conditions following a sinusoidal velocity profile were applied at the inlet as defined in Figure 2a, matching the flow rate ($9.5 ml/s) measured via tomographic PIV at the inlet to the phantom model. The doublebifurcating airway model, including the intubation tube, was meshed with tetrahedral cells using a commercial meshing software (ANSYS ICEM). Rigorous mesh convergence tests were first performed to select the optimal numerical setup ranging from 2.

| Biochemical and cytological analyses
Cytokine levels in sampled medium supernatants were measured using commercially available ELISA kits in a blinded fashion and by fol- Cells were subsequently centrifuged at 3.0 rpm for 3 min., the supernatant discarded and then resuspended in 500 μl of binding buffer, stained with 10 μl of AnnexinV-FITC and PI following manufacturer's instructions (Annexin-V Apoptosis Staining Detection Kit, ab14084, Abcam). Metabolic activity tests were performed using an alamarBlue cell viability assay (BUF012, Bio-rad). The absorbance of the incubated media containing 10% alamarBlue (2 h) was measured at 570 and 600 nm using a microplate reader (Synergy H1, BioTek).

| Prophylactic medication preparation
Montelukast sodium hydrate (#SML0101, Sigma-Aldrich) was diluted in DMSO to a stock concentration of 0.006 mg/ml and stored in À20 C. Before a prophylactic ventilation experiment, the stock working concentration was further diluted in a culture medium (MEMeagle) at 1:1000 before injection into the models (0.006 μg/ml).

| Statistical analyses
Data are presented as the mean with error bars showing the standard deviation (SD). Statistical analyses were performed using Prism 8.0 GraphPad software (GraphPad). Data were analyzed using a un/paired two-sided Student's t-test, One-Way analysis of variance (ANOVA) with Holm-Sidak's post-test, Two-Way ANOVA with Sidak's multiple comparisons test, or as indicated in figure legends. p-values <0.05 were considered statistically significant and are reported in figures using the following notation: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 (with * referring to the comparisons specified in each figure legend).

CONFLICT OF INTERESTS
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
The data that supports the findings of this study are available in the supplementary material of this article.