Human Vascular Wall Microfluidic Model for Preclinical Evaluation of Drug-Induced Vascular Injury

Abstract

Drug-induced vascular injury (DIVI) in preclinical animal models often leads to candidate compound termination during drug development. DIVI has not been documented in human clinical trials with drugs that cause DIVI in preclinical animals. A robust human preclinical assay for DIVI is needed as an early vascular injury screen. A human vascular wall microfluidic tissue chip was developed with a human umbilical vein endothelial cell (HUVEC)—umbilical artery smooth muscle cell (vascular smooth muscle cell, VSMC) bilayer matured under physiological shear stress. Optimized temporal flow profiles produced HUVEC-VSMC bilayers with quiescent endothelial cell (EC) monolayers, EC tight junctions, and contractile VSMC morphology. Dose–response testing (3–30 μM concentration) was conducted with minoxidil and tadalafil vasodilators. Both drugs have demonstrated preclinical DIVI but lack clinical evidence. The permeability of severely damaged engineered bilayers (30 μM tadalafil) was 4.1 times that of the untreated controls. Immunohistochemical protein assays revealed contrasting perspectives on tadalafil and minoxidil-induced damage. Tadalafil impacted the endothelial monolayer with minor injury to the contractile VSMCs, whereas minoxidil demonstrated minor EC barrier injury but damaged VSMCs and activated ECs in a dose–response manner. This proof-of-concept human vascular wall bilayer model of DIVI is a critical step toward developing a preclinical human screening assay for drug development.

Impact statement

More than 90% of drug candidates fail during clinical trials due to human efficacy and toxicity concerns. Preclinical studies rely heavily on animal models, although animal toxicity and drug metabolism responses often differ from humans. During the drug development process, perfused in vitro human tissue chips could model the clinical drug response and potential toxicity of candidate compounds. Our long-term objective is to develop a human vascular wall tissue chip to screen for drug-induced vascular injury. Its application could ultimately reduce drug development delays and costs, and improve patient safety.

Introduction

Drug-induced vascular injury (DIVI) can occur during preclinical animal testing, impacting ∼2.5% of drug candidates.¹ DIVI is characterized by vascular endothelial and smooth muscle damage and inflammation and is recognized across drug classes.² Lesions rapidly develop in vascular beds, specifically the arteriole networks of the rat mesentery, dog coronary, and nonhuman primate coronary and testis. The observation of DIVI during preclinical testing delays or terminates the drug development process.

Clinical relevance of DIVI is not clear. After minoxidil treatment, canine coronary artery and rat mesenteric arterioles incurred vascular lesions, but similar injury was not identified at autopsy in minoxidil-treated patients.³ To detect DIVI, pharmaceutical safety toxicology consortiums are developing universal biomarker panels.^2,4–6 Clinical validation continues across drug classes against preclinical animal biomarkers and histopathology. An in vitro physiological vascular assay of human DIVI would help decipher these interrelationships and bridge animal studies and human clinical trials during drug development.

An in vitro model of DIVI should emulate the native physiological environment of a human arteriole and its response to DIVI-inducing compounds. This model should mimic an arteriole wall with a confluent monolayer of endothelial cells (ECs) and a tunica media with 1–3 layers of vascular smooth muscle cells (VSMCs). It should mature (<2 weeks) under physiological shear stress (∼1.5 Pa) to achieve EC quiescence, effective intercellular EC junctions and barrier function, contractile VSMCs, and functional EC-VSMC signaling.

Many two-dimensional (2D) and three-dimensional (3D) vascular models have been engineered. Several 3D models have demonstrated effective EC barrier function but were matured under low shear stress or lacked VSMCs.^7–11 Many 2D vascular wall EC-VSMC bilayer models have been cultured under arteriole physiological shear stress.^12–14 These 2D models often lack confluent EC monolayers and membrane-segregated junctional proteins, focusing primarily on EC-VSMC interactions. Our previous pilot study engineered a vascular wall bilayer model of DIVI with rat aortic ECs and VSMCs, matured under physiological shear stress.¹⁵ Treatment with known rat DIVI compounds induced vascular injury (VI), revealed by RBC extravasation. Damage was not confirmed, and an RBC extravasation assay presents a quantitative challenge.

In this proof-of-concept study, we developed a human physiological vascular-wall bilayer microfluidic model to characterize the impact of acute DIVI. Human umbilical vein endothelial cells (HUVECs or ECs) and human umbilical artery smooth muscle cells (HUASMCs or VSMCs) were used as model human vascular cells. Vascular wall bilayers were engineered under physiological shear stress in less than 10 days culture. In an acute dose–response study, the bilayers were challenged with the vasodilators minoxidil and tadalafil, and a permeability assay screened VI. We assessed the impact of drug treatment on endothelial barrier function, EC activation, and cellular integrity through the expression of EC and VSMC markers.

Methods

Materials

Vendor details are provided in Supplementary Table S1 for the equipment, cells, media, drugs, and reagents and Supplementary Table S2 for the device.

Microfluidic device design and assembly

The microfluidic device (Fig. 1a) housed a 200 × 200 μm EC channel (layer 2) and a 400 × 400 μm VSMC channel (layer 5). A track-etched membrane (TEM) (layer 3) separated the parallel channels. The TEM was not bonded within the device but was reversibly compression-sealed by the clamping force between top and bottom plates (layers 1, 5) (Supplementary Fig. S1). An impermeable mask (layer 4) with a 12-mm-long rectangular hole limited communication between channels to a rectangular central region. The top plate housed the VSMC and EC channel inlets and outlets (Fig. 1b). Fujifilm supplied all devices.

FIG. 1.

Microfluidic DIVI device in a closed bioreactor system. (a) The EC (layer 2, PDMS) and VSMC (layer 5, COC) channels were separated by a removable polycarbonate TEM (layer 3, 5 μm or 0.4 μm pores), which was reversibly sealed by the clamping force between the top and bottom COC plates (layers 1 and 5). The cellularized TEM membrane was easily removed from the device for post-run cell analysis. A slit in the polypropylene mask (layer 4) allowed interchannel communication in a central 12 mm zone. (b) The assembled device with EC (red) and VSMC (blue) channels. Cross-section depicts EC (red) and VSMC (green) bilayer on TEM. Layer numbers defined in (a). (c) The EC (red) and VSMC (blue) media recirculation flow paths from vented holding reservoirs through the device. (d) A fixture housed the device and its bioreactor components. Bubbles were flushed through ports 1 and 2 during startup. VSMCs and ECs were seeded through ports 3 and 4, respectively. (e) After VSMC seeding, the device was inverted to allow cell attachment to the TEM bottom. (f) The protocol timeline relative to EC seeding day. COC, cyclic olefin copolymer; DIVI, drug-induced vascular injury; EC, endothelial cell; PDMS, polydimethylsiloxane; TEM, track-etched membrane; VSMC, vascular smooth muscle cell. Color images are available online.

Recirculation system and fixturing

Independent flow paths were established through the VSMC and EC channels (Fig. 1c). A microperistaltic pump recirculated cell-specific medium in a closed loop from a vented medium reservoir through the device channel. A custom 3D-printed fixture securely held all components for a single device (Fig. 1d and Supplementary Fig. S2), allowing undisturbed in-fixture cell imaging and device inversion after VSMC seeding (Fig. 1e). Tilting the vented reservoirs (15° from horizontal) minimized flow path height variations. All parts that contacted medium were sterilized with ethylene oxide gas.

Porous membrane preparation

The 5-μm-pore TEMs qualified the engineered bilayer and characterized the endothelial barrier permeability. Drug dose–response studies utilized 0.4-μm-pore TEMs to eliminate transmembrane cell migration. Polyvinylpyrrolidone-coated TEMs were soaked in methanol to remove the wetting agent, and then oxygen-plasma treated and immediately spray-coated with collagen (0.5 mg mL⁻¹ in 70% ethanol). Dry collagen-coated TEMs were assembled into devices or stored in a dust- and pathogen-free environment until use.

Device seeding and culture under physiological shear stress

Pooled HUVEC sources were purchased prelabeled to express red fluorescent protein (RFP) or labeled with mCherry fluorescent protein using standard lentivirus protocols,¹⁶ achieving >90% labeling efficiency. Constitutive fluorescent protein expression facilitated live-EC imaging within microfluidic devices. Two unlabeled HUASMC sources were utilized. Cells (passages 2–4) showed no cell source differences. ECs and VSMCs were cultured in EC and VSMC growth media, respectively, with 1% pen-strep.

A device batch (8) was seeded and cultured, following the protocol timeline (Fig. 1f). A programmable controller (16-channel) controlled microperistaltic pump flow (8 EC, 8 VSMC). On the VSMC seeding day (day-1, relative to EC seeding), human plasma fibronectin (3% in water) conditioned EC channels. After 2 h, EC and VSMC media flushed respective tubing lines and device channels.

VSMC channels were seeded at low cell density. Each pump pulled cell suspension from the VSMC outlet port (Figure 1d, port 3) into its device (45 μL, 5 × 10⁵ cells mL⁻¹, 35 μL min⁻¹). Devices were immediately inverted (Fig. 1e) to allow cell attachment to TEM bottoms. After an 8-h static hold, medium was recirculated through each VSMC channel (0.7 μL min⁻¹).

On day 0, VSMC medium flow was stopped, devices were reinverted, and EC channels were seeded at high cell density. Each pump pulled cell suspension from the EC outlet port (Fig. 1d, port 4) into its device (23 μL, 6 × 0⁶ cells mL⁻¹, 35 μL min⁻¹). After a 3-h static hold, programmed flow through each EC channel was initiated (0.7 μL min⁻¹, 8.8 mPa shear stress) and maintained until day 3 when epifluorescence consistently demonstrated confluent ECs. The flow rate was ramped, achieving 120 μL min⁻¹ (1.5 Pa physiological shear stress) by day 8, and was held for 24 h before drug testing. EC medium (10 mL per device) was replaced every 4 days. Devices were imaged by epifluorescence daily or at medium change.

Drug testing

The engineered bilayers were challenged with minoxidil or tadalafil. Before each test, a 10 mM minoxidil stock solution was prepared in sterile distilled water and immediately diluted in medium (3, 10, and 30 μM). Tadalafil has negligible solubility in water; a 50 mM stock solution was prepared in dimethyl sulfoxide (DMSO), stored at −20°C, and diluted at use (3, 10, and 30 μM) in medium (0.02% DMSO). The 30 μM tadalafil solution did not fully dissolve and was used only for permeability testing. On day 9 post-EC seeding, 10 mL prepared drug solution replaced EC medium for each device, and the flow rate was maintained at 120 μL min⁻¹ for 24 h. Controls were medium with 0.02% DMSO or medium only.

Permeability testing

The permeabilities of the vascular bilayers were assessed in medium-only (n = 7) and 30 μM tadalafil-treated (n = 3) devices. A syringe pump charged the EC channel with fluorescein solution (0.3 μM in medium), which was statically held. The VSMC channel was slowly flushed with continuous medium flow through syringe pump to maintain near-zero dye concentration. An epifluorescent inverted microscope acquired images every 15 s for ∼45 min along the EC channel length using the microscope tiling function (Supplementary Fig. S3a).

Image brightness linearly correlated with dye concentration. The EC channel dye concentration decayed exponentially. Using nonlinear regression, an exponential function was fit to the intensity data at 200 bins along the EC channel length (Supplementary Data S1) to yield the decay half-life (t_1/2, s) (Eq. S3) and the permeability as a function of location; P = h_EC * ln(2)/t_1/2 where P = permeability (cm s⁻¹) and h_EC = EC channel height (cm) (Eq. S4). The half-lives of bins within the central 8-mm zone were averaged to estimate the device t_1/2. The half-lives for multiple devices were averaged to assess the overall permeability for a treatment condition.

Immunohistochemical evaluation of DIVI

Immunohistochemical (IHC) evaluated the acute DIVI responses of engineered bilayers in tadalafil (3 and 10 μM) and minoxidil (3, 10, and 30 μM)-treated devices, with medium-only and 0.02% DMSO-treated devices as controls (n = 1–3 devices per condition). All bilayers were fixed in situ with 4% paraformaldehyde (20 min). After removal, the medium-only TEMs were left whole. The 0.02% DMSO and drug-treated TEMs were sectioned into thirds to individually screen three EC activation markers within each sample (Supplementary Fig. S3b).

Immunohistochemistry was performed using standard protocols. Supplementary Table S3 provides protocol details and antibody and probe specifics. Whole membranes were treated with anti-zonula occludens-1 (ZO-1, TJP1) and phalloidin. Membrane sections were labeled as

• Inlet: anti-intercellular adhesion molecule 1 (ICAM1) and phalloidin;

• Mid: antivascular cell adhesion protein 1 (VCAM1) and phalloidin; and

• Outlet: antivascular endothelial-cadherin (VE-cadherin, CDH5) and anti-E-selectin (SELE).

Cell nuclei were stained with DAPI.

Stained sections were imaged by confocal or epifluorescent microscopy. During an imaging session, settings were fixed for each color channel. Images were processed using Image J/Fiji with left-to-right flow orientation. If necessary, brightness/contrast was similarly adjusted in all images acquired during a single microscope session. The EC inverse aspect ratio (minor axis length [major axis length]⁻¹) was estimated using ImageJ from ZO-1-labeled images (>30 cells).

Statistical analysis

All numerical results are presented as mean ± standard error. Statistical analysis of two independent experiments was assessed using a one-way analysis of variance test where p < 0.05 represents statistical significance.

Experiment

Engineering a physiological 2D vascular wall bilayer

We engineered a 2D human EC-VSMC bilayer in a microfluidic device with two parallel channels separated by a TEM (Fig. 1a). The EC channel size mimicked arteriole dimensions. Communication between the EC and VSMC channels occurred only in a central 12 mm zone where the flow was visually well-behaved. Continued maintenance of VSMC health indicated sufficient diffusion of nutrients and oxygen from the EC channel. Several EC medium flow profiles were tested (Supplementary Fig. S4). The flow profile was selected that produced quiescent, confluent ECs: 0.7 μL min⁻¹ hold until EC confluence (day 3), 5 days linear ramp to physiological flow (120 μL min⁻¹), and 24 h hold.

The engineered bilayer demonstrated morphological characteristics of a physiological arteriole wall: quiescent ECs with membrane-segregated ZO-1, cortical f-actin rings, and minimal stress fibers (Fig. 2a). Spindle-shaped VSMCs, 1–2 cell layers thick, possessed contractile morphology (Fig. 2b). Minimal VSMC migration occurred across the TEM.

FIG. 2.

Physiological 2D vascular wall bilayer. (a) The quiescent EC monolayer exhibited ZO-1 (green) membrane segregation with cortical f-actin rims and minimal stress fibers (phalloidin, red); DAPI-stained nuclei are magenta and blue at low and high magnification, respectively. Occasional VSMCs were observed (white arrows) with a 5-μm-pore TEM. Boxed area magnified in the bottom image. Yellow dots are aberrant staining of TEM pores. (b) VSMCs (phalloidin, green) possessed spindle-like shapes with contractile phenotype and aligned with EC medium flow. Scale bars: (a) top, (b) 100 μm; (a) bottom 50 μm. 2D, two-dimensional; DAPI; ZO-1. Color images are available online.

Fluorescein diffusion assay characterizes permeability along EC channel

A real-time diffusion-based permeability assay assessed the barrier functions of medium-only control and heavily damaged (30 μM tadalafil) bilayers. Tadalafil at 30 μM exceeded its solubility limit in 0.02% DMSO. An exponential decay function modeled the fluorescein intensity decay data along the EC channel, and the model well fit the data within the central open zone (correlation coefficient = 0.999) (Supplementary Fig. S5). The t_1/2 and permeability profiles were estimated from these data (Supplementary Data S1). The average permeability of the engineered bilayers treated with 30 μM tadalafil (n = 2) (1.20 × 10⁻⁴ ± 7.07 × 10⁻⁷ cm s⁻¹) was 4.1 times that of the medium-only control (n = 2) (2.95 × 10⁻⁵ ± 1.51 × 10⁻⁶ cm s⁻¹) (Fig. 3a and Supplementary Table S4).

FIG. 3.

Endothelial barrier function characterized by fluorescein permeability. (a) The average permeability of medium controls was significantly less than after 30 μM tadalafil treatment (**p < 0.00001). (b) Fluorescein intensity images as a function of device-specific t_1/2 multiples after tadalafil (device T-1) and medium-only (device T-7) treatments. Intensity decay was uniform within the open zone (between dashed vertical lines). (c) Confluent ECs in the medium-only control demonstrated continuous VE-cadherin (green) segregation in a linear pattern at paracellular clefts. Cell-sized intercellular gaps (arrowheads) with little VE-cadherin membrane segregation and stitch-like VE-cadherin patterns were observed in the tadalafil-treated device. Scale bars: (b) 500 μm, (c) 100 μm. Color images are available online.

Permeability profiles revealed clear distinctions between tadalafil-treated and medium-only devices, confirmed by EC barrier morphologies. The intensity images depicted the signal intensity decay at device-specific t_1/2 multiples (Fig. 3b) (half-lives: 116 ± 7.8 s for tadalafil [device T-1] and 486 ± 11.3 s medium-only [device T-7]). The tadalafil-treated profile showed high uniform permeability within the open area and lower permeability in the impermeable area. VE-cadherin expression revealed considerable EC barrier injury with cell-sized intercellular gaps and minimal VE-cadherin membrane segregation (Fig. 3c). In contrast, the medium-only EC barrier demonstrated lower uniform permeability within the open area. Continuous VE-cadherin segregation with minimal EC interjunctional holes confirmed a confluent EC barrier.

Our permeability assay also characterized device quality before drug testing (Supplementary Data S1). Localized high permeability areas identified processing defects that were difficult to detect during culture (Supplementary Fig. S6). Permeability profiles quantified EC barrier damage before and after drug testing.

Drug treatments impact endothelial junctional integrity and cell viability

DIVI was visualized by IHC in response to acute drug treatment. Mature engineered vascular wall bilayers were exposed to tadalafil (3 and 10 μM in 0.02% DMSO), minoxidil (3, 10, and 30 μM), or 0.02% DMSO. The DMSO concentration was maintained at 0.02% as higher concentrations (0.1%) altered EC junctional architecture (Supplementary Fig. S7). At 0.02% DMSO, endothelial junctional integrity was little affected; VE-cadherin segregated to the membrane with continuous patterning and few interjunctional gaps (Fig. 4a). Cortical f-actin bands were observed, and thin stress fibers aligned in the flow direction (Fig. 4b).

FIG. 4.

Effect of drug exposure on endothelial VE-cadherin and f-actin expression and VSMC phenotype. (a) Continuous VE-cadherin (green) patterning suggested effective endothelial barrier function. Cell boundaries lacking VE-cadherin expression identified significant intercellular gaps (white arrowheads), whereas noncontinuous VE-cadherin patterning pinpointed fine gaps. JAILs (yellow arrows) denoted ongoing junctional remodeling. (b) Actin filaments (phalloidin, green) connecting neighboring cells pinpointed fine intercellular gaps (yellow arrowheads) and larger gaps (white arrowheads). Increased stress fiber thickness was detected with drug treatment. Partial stress fiber disassembly (yellow arrow) was observed at higher drug concentrations. Complete actin filament depolymerization and shape-distorted nuclei (DAPI, blue) identified apoptotic cells (white arrows). (c) Sparse VSMCs monolayers (phalloidin, green) retained a contractile phenotype in the DMSO control and at low drug concentrations. Synthetic VSMC phenotype observed with high concentration minoxidil. Noncontinuous f-actin filaments and rounded, shrunken, or vacuolated cells identified VSMC damage. Boxed areas are magnified in the corresponding bottom images. Scale bars: 50 μm (low magnification), 20 μm (high magnification). DMSO, dimethyl sulfoxide; JAILs, junction-associated intermittent lamellipodia. Color images are available online.

Minoxidil treatment induced minor endothelial barrier damage. At 3 μM, VE-cadherin junctional profiles were mostly continuous with minor intercellular junctional perturbations and some noncontinuous stitch-like patterns. At 10 and 30 μM, interrupted VE-cadherin distributions identified fine interjunctional holes (<5 μm). Junction-associated intermittent lamellipodia (JAILs) were detected, indicative of junctional remodeling.

In contrast, tadalafil treatment produced considerable EC junctional damage. Large intercellular holes (>10 μm) were observed, denoted by substantial gaps in peripheral VE-cadherin patterning. These holes appeared to increase in size and number with increasing tadalafil concentration. VE-cadherin patterning, where present, was interrupted without JAILs, confirming junctional damage.

Actin cytoskeleton alterations depicted EC injury. Stress fiber number and thickness visually increased, and prominent actin cortical rings decreased across all drug doses (Fig. 4b). Actin filament networks connecting neighboring cells identified fine aligned intercellular gaps, similar to defects pinpointed by punctate VE-cadherin expression. Minor intercellular gaps were observed with minoxidil treatment and larger holes with tadalafil treatment.

Stress fibers disassembled or were severed mid-fiber at EC leading edges (downstream relative to flow direction) in engineered bilayers treated with high drug concentrations (10 μM tadalafil, 10 and 30 μM minoxidil). Apoptotic cells with little cytoplasm, minimal stress fibers, and irregular nuclei were observed post-10 μM tadalafil treatment.

Drug treatments alter vascular smooth muscle morphology and impact cell integrity

F-actin expression characterized VSMC morphology. After drug testing, sparse 1–2 cell layers were observed with VSMCs aligned in the direction of EC medium flow (Fig. 4c). In the DMSO control and all tadalafil-treated samples, actin fibers spanned the cell lengths, and the cells possessed contractile cell morphology. Minor damage was noted after 10 μM tadalafil treatment as perpendicular nicks along the cell lengths.

Minoxidil-treated VSMCs incurred considerable injury. Post-3 μM minoxidil treatment, VSMCs retained contractile morphology but bore perpendicular nicks along the cell lengths. The cells acquired a synthetic morphology with blebs that lacked organized actin fibers at 10 μM minoxidil and appeared shriveled with bent morphologies at 30 μM minoxidil.

ICAM1 expression characterizes EC activation after drug treatments

We examined the acute inflammatory response to drug treatments through EC adhesion molecule expression. VCAM1 and E-selectin were not sensitive indicators of EC activation (Supplementary Fig. S8), but ICAM1 expression varied across treatment conditions. Low-concentration drug treatment (3 μM minoxidil or tadalafil) and the DMSO control minimally activated ECs. Under epifluorescence, asymmetric ICAM1 expression occurred, with minimal ICAM1 expression in the cell bulk and greater leading-edge expression (Fig. 5a). Under confocal microscopy, ECs demonstrated either imperceptible expression or highly asymmetric ICAM1 distribution with a leading-edge bias (Fig. 5b).

FIG. 5.

EC activation after drug testing. (a) In the DMSO control and at low drug concentrations, asymmetric intracellular ICAM1 expression (purple) was observed by epifluorescent imaging. At higher drug concentrations, homogenous intracellular ICAM1 expression was found with significant cell-to-cell variability. (b) Intracellular and intercellular ICAM1 variability was observed at low drug concentrations by confocal imaging (ICAM1 [red], phalloidin [green], and DAPI [blue]). Higher-level ICAM1 expression with less intercellular variability occurred at higher drug concentrations. Scale bars: (a) 100 μm, (b) 50 μm. ICAM1, intercellular adhesion molecule 1. Color images are available online.

Physiological shear stress shifts the maximum intracellular height and local shear stress to the leading edge of the cell.¹⁷ The ICAM1 gene reporter region contains a shear stress-response element, so asymmetric ICAM1 expression develops.¹⁸ Tadalafil treatment induced a moderate ICAM1 dose response. At 10 μM tadalafil, expression variability continued across the culture but with higher overall ICAM1 expression compared with 3 μM tadalafil. Minoxidil treatment visually induced greater and more uniform ICAM1 expression than tadalafil treatment. Nearly all ECs expressed ICAM1 (10 and 30 μM minoxidil), and expression levels increased with increasing minoxidil concentration.

Discussion

In this proof-of-concept study, a human 2D vascular wall bilayer model of DIVI was engineered in 10 days in a modular microfluidic system. We replicated an arteriole wall architecture with HUVECs and HUASMCs cultured on opposite sides of porous membranes under physiological shear stress. Engineered bilayers were acutely challenged with the vasodilators minoxidil and tadalafil across 3–30 μM concentrations. A permeability assay demonstrated significantly increased bilayer permeability with high tadalafil concentration relative to a medium-only control. Spatial protein expression assays showed distinct differences in EC and VSMC injury after minoxidil and tadalafil treatments.

A portable microfluidic system was engineered with media recirculated using novel microperistaltic pumps, the first use reported in tissue chip systems. All system components were housed in a compact fixture compatible with live-cell imaging. Recirculation drastically reduced medium use,¹⁵ which supported cost-effective long-term culture at physiological flow rates. Other vascular tissue chips utilized standard peristaltic pumps, limiting in-run cell culture assessment.⁸ Organ-on-chip solid organ models employed on-chip recirculation pumps, operating at substantially lower flow rates.^19–22

A physiological vascular bilayer formed using a flow profile with a 3-days hold at low flow and a 5-days linear ramp to physiological shear stress. Our previous rat-based study used a syringe pump with an exponential 8-days flow profile.¹⁵ Rat ECs were more elongated and aligned with flow than our HUVECs because of species/tissue source dissimilarities and syringe pump-microperistaltic pump flow pulsation differences.²³

This physiological vascular bilayer served as the starting point for our drug-treatment studies. Unidirectional laminar flow and high shear stress activate mechanosensory signaling networks that support EC quiescence and survival, VSMC contractility, and vascular integrity.²⁴ We demonstrated quiescent confluent ECs in medium-only controls. ZO-1 cell membrane segregation visually indicated endothelial barrier optimization; it regulates actomyosin activity that links VE-cadherin to the force cytoskeleton to ensure maximum junctional tension and optimal barrier integrity.²⁵

Cortical f-actin rings, minimal intercellular holes, and low levels of flow-aligned stress fibers were also observed. These nonactivated ECs expressed low levels of pro-inflammatory EC adhesion molecules,²⁶ consistent with ECs exposed to physiological unidirectional flow.²⁷ VSMCs possessed a physiological contractile morphology. Permeability studies confirmed EC barrier function, with a demonstrated average permeability similar to the lowest-reported EC-monolayer permeability after physiological shear stress exposure.⁹

Drug selection was challenging as compounds known to cause human DIVI are not readily available. Minoxidil was chosen to directly compare with our previous DIVI rat-based testing. Minoxidil activates ATP-sensitive potassium channels in VSMCs to effect vasodilation. Tadalafil, a phosphodiesterase-5 inhibitor (PDE5i), was selected as rodent and canine vascular pathologies have occurred. Tadalafil blocks cGMP degradation by PDE5 in VSMCs, enhancing the effects of endothelium-dependent nitric oxide (NO)-mediated vasodilation. In canines, coronary arterial lesions developed at 5–16 times the maximum recommended clinical tadalafil dose (MRHD).²⁸ We used 3–30 μM tadalafil and minoxidil concentration, with 3 μM approximating the MRHD plasma concentration.

A fluorescein permeability screen characterized endothelial barrier damage after drug treatment. Our novel method estimated decreasing fluorescein concentration by fluorescent intensity decay rather than the customary direct concentration measurement of diffused fluorescein. Our estimated permeabilities were comparable with traditional methods as our medium-only permeability (2.95 × 10⁻⁵ cm s⁻¹) was similar to 2.28 × 10⁻⁵ cm s⁻¹ for HUVECs perfused under physiological shear and permeability tested with 3 kDa dextran.⁹

An average permeability ratio of 4.1 was observed between heavily damaged bilayers (tadalafil, 30 μM) and defect-free control bilayers. Undissolved drug in the 30 μM tadalafil solution likely caused considerable EC damage. Simulating a heavily damaged bilayer with a bare porous membrane, our measured 4.1 permeability ratio exceeded the 2.5 ratio reported for bare porous membranes versus control EC monolayers (2.4–2.5).^29,30 Our engineered bilayers matured for longer times under physiological shear (8 days vs. 3–4 days), which may have improved barrier function and decreased permeability.³¹ The reported studies lacked VSMCs, which led to reduced transport across the vascular bilayer.¹⁵ Healthy VSMCs may have limited diffusion across our control engineered bilayer.

After 30 μM tadalafil treatment, VSMCs were likely heavily damaged, which would have increased the damaged-to-control permeability ratio. Also, HUVEC monolayers treated with TNF-α (2.5–500 ng mL⁻¹) experienced ∼2.2 times (1.8–2.5) greater permeability than untreated controls.^30,32 TNF-α treatment decreases VE-cadherin expression, increasing endothelial barrier permeability and intercellular gap formation.³³ Based on these comparisons, 2–4 permeability ratios are likely for moderate and severe damage, respectively. This narrow range suggests that this assay could act as a “go/no-go” screen but not identify subtle differences between drug treatments.

To gain VI insight, we examined the spatial expression of proteins involved in EC junction formation, maintenance, and activation and VSMC morphology. The combined expression of VE-cadherin and f-actin characterized EC junctional remodeling.³⁴ VE-cadherin organizes intercellular junctions; its phosphorylation, triggered by drugs and other factors, mediates VE-cadherin disassembly. VE-cadherin internalization and stress fiber formation lead to cell retraction, intercellular gap formation, and increased vascular permeability.³⁵ Conversely, JAILs represent positive junctional remodeling; these structures form new VE-cadherin adhesion sites, which support intercellular gap closure and re-establish effective barrier function.³⁶

IHC assays provided contrasting perspectives on VI caused by tadalafil and minoxidil treatments. Tadalafil damaged the endothelial monolayer with minor impact on contractile VSMCs. VE-cadherin disassembly and paracellular gap formation occurred in a dose–response manner, with EC apoptosis observed at the highest tadalafil concentration. Only modest EC activation and VSMC injury occurred.

These post-tadalafil treatment observations suggest high NO levels induced injury. Sildenafil, a PDE5i similar to tadalafil, enhances endothelial NO production by PDE5 inhibition and eNOS activation through a P13K-dependent pathway.³⁷ HUVECs also express significant PDE5 levels³⁸; NO production decreased VE-cadherin expression and increased permeability in cultured HUVECs.³⁹ High NO levels attenuated ICAM1 expression in HUVECs⁴⁰ and promoted apoptosis.⁴¹ Future studies will track NO concentration post-drug treatment.

Animal studies with minoxidil found endothelial damage with major paracellular gaps,^42–44 unlike our observations with human cells. Endothelial injury was primarily attributed to drug-induced systemic hemodynamic fluctuations and not direct endothelium interaction. Increased endothelial shear stress led to leukocyte and platelet adhesion, suggesting EC activation with elevated ICAM1 expression. RBC hemorrhage into the medial layer compressed and necrosed the VSMCs.² Humans lack exaggerated pharmacological response to minoxidil, with VI undetected upon autopsy.³ The roles remain unclear of animal and human pharmacokinetic and metabolic differences on VI.

Physiological vascular models allow independent examination of these critical factors. Our model characterized the impact of minoxidil treatment without exaggerated hemodynamic fluctuations. We found heavily damaged VSMCs with minor endothelial injury. The EC barrier appeared largely intact with minimal VE-cadherin disassembly and few paracellular gaps. The noncontinuous stitch-like VE-cadherin patterns and JAILs suggested ongoing junctional remodeling.^34,45 Endothelial stress fiber disassembly was observed, requiring further investigation. However, stress fiber disassembly occurs during junctional remodeling and apoptosis initiation.^46,47 Furthermore, EC activation through ICAM1 expression developed in a dose–response manner.

These minoxidil-treatment data contrasted with our previous rat-based findings. Significant RBC extravasation was observed after 30 μM minoxidil treatment; endothelial barrier damage was not visually confirmed.¹⁵ Both models lacked hemodynamic fluctuations, suggesting dissimilar human and rat biological responses. Furthermore, the active metabolite minoxidil sulfate mainly mediates vasodilation,⁴⁸ although high minoxidil concentrations contribute in vitro.⁴⁹ We will verify the injury profiles after minoxidil sulfate and minoxidil treatments in our system in future studies. Nonetheless, our engineered vascular wall bilayer model can separately discern EC and VSMC injury.

Future testing will address the device and assay limitations. Quantitative gene and protein expression assays for EC and VSMC injury will complement the spatial protein expression assays. Results from rapid-screening permeability and transendothelial electrical resistance assays will be correlated with quantitative gene and protein expression. To minimize lipophilic drug adsorption, plastic materials, polysulfone or cyclic olefin copolymer,^19,20 will replace polydimethylsiloxane (PDMS). Primary and iPS-derived vascular cells from relevant organ-specific vessels^50,51 will be employed to model adult human vascular beds. Systemic changes induced by DIVI compounds will be incorporated into the microfluidic model. Finally, we will verify our rat-based vascular wall bilayer model against the in vivo rat mesentery arteriole model across known DIVI and non-DIVI compounds.

Conclusion

This proof-of-concept study demonstrates the promise of a human 2D vascular wall bilayer microfluidic model for screening DIVI. Results demonstrated relatively minor but reproducible and significant permeability differences between control and severely damaged engineered vascular wall bilayers. Spatial protein expression assays discerned EC and VSMC damage differences after treatment with vasodilators with different mechanisms of action. Future studies will incorporate hemodynamic disturbances induced by DIVI compound treatment to clarify the roles of hemodynamic changes versus biological effects.

Authors' Contributions

E.E., N.E., O.M., H.M., D.H., M.N., K.T., K.O., T.O., O.R., and E.E. assisted with experimental protocol development. K.O., T.O., and E.E. played substantial roles in the design and manufacture of functional microfluidic devices. E.E. and N.E. carried out all laboratory work. C.S., C.N., and E.E. participated in data analysis and interpretation. C.S. and C.N. conceived of and designed the study. C.S. drafted the article with critical revisions from C.N., E.E., and N.E. C.S. coordinated the study. All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Disclosure Statement

Fujifilm partially funded this study and supplied devices but had no role in the study design; data collection, analysis, or interpretation; or article preparation.

Funding Information

This study was supported by Fujifilm and the Massachusetts General Hospital, Department of Surgery.

Supplementary Material

Supplementary Data S1

Supplementary Figure S1

Supplementary Figure S2

Supplementary Figure S3

Supplementary Figure S4

Supplementary Figure S5

Supplementary Figure S6

Supplementary Figure S7

Supplementary Figure S8

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3

Supplementary Table S4

Supplementary: Qualification of flow ramp

Supplementary: Screening endothelial barrier function before drug testing

Supplementary: Permeability model to quantify EC barrier function