Automating Vascular Biology: An End‐to‐End Automated Workflow for High‐Throughput Blood Vessel‐on‐a‐Chip Production and Multi‐Site Validation

ABSTRACT

There is a growing demand for automated organ‐on‐a‐chip platforms that are compatible with off‐the‐shelf robotic liquid‐handling systems and plate readers to improve reproducibility and scalable analysis. In this work, we present an end‐to‐end automated method for fabricating tubular blood vessel models at scale using a custom 384‐well open‐top platform (AngioPlate384), designed to support integration with liquid‐handling systems and large‐scale analysis. Our approach enables the generation of over 100 perfusable blood vessels fully embedded in hydrogel and supported by stromal cells (fibroblasts and pericytes), allowing both luminal and interstitial flow. Using this platform, we demonstrated that stromal co‐culture significantly enhances vascular barrier function, and results in an altered response to chemotherapeutics and to inflammatory stressors. This platform offers a robust and scalable approach to generating customizable blood vessel‐on‐a‐chip models for vascular biology studies, disease modeling, and preclinical testing. Its compatibility with automation and standardized workflows positions it as a powerful tool to accelerate the adoption of microphysiological systems in pharmaceutical research.

AngioPlate384 is a 384‐well open‐top platform that automates production of more than 100 miniaturized, perfusable blood vessels embedded in hydrogel and supported by stromal cells. Stromal‐endothelial co‐culture strengthens blood vessel barrier function and yields responses useful for translational planning. Scalable and automation‐ready, it suits drug screening and disease modeling.

Abbreviations

Terms

Abbreviation

bone morphogenetic protein 9

BMP‐9

bovine serum albumin

BSA

coefficients of variation

CV

dimethyl sulfoxide

DMSO

Dulbecco's Modified Eagle Medium

DMEM

Endothelial Cell Growth Medium 2

ECGM2

endothelial cells

EC

epidermal growth factor

EGF

extracellular matrix

ECM

fetal bovine serum

FBS

fibroblast growth factor 1

FGF‐1

fibroblast growth factor 2

FGF‐2

fibroblasts

FB

fluorescein Isothiocyanate

FITC

granulocyte‐colony stimulating factor

G‐CSF

granulocyte‐macrophage colony‐stimulating factor

GM‐CSF

half‐maximal effective concentration

EC₅₀

half‐maximal toxic concentration

TC₅₀

hepatic stellate cells

HSC

hepatocyte growth factor

HGF

human umbilical vein endothelial cells

HUVEC

interferon‐gamma

IFNγ

interleukin‐1 beta

IL‐1β

interleukin‐1 receptor antagonist

IL‐1RA

interleukin‐10

IL‐10

interleukin‐12 p70

IL‐12(p70)

interleukin‐12 subunit beta

IL‐12(p40)

interleukin‐2

IL‐2

interleukin‐4

IL‐4

interleukin‐8

IL‐8

lipopolysaccharide

LPS

maximum plasma concentrations

C _max

microphysiological system

MPS

monocyte chemoattractant protein‐1

MCP‐1

penicillin‐streptomycin

Pen‐Strep

Pericyte Growth Medium 2

PGM2

pericytes

PC

pharmacodynamic

PD

pharmacokinetic

PK

phosphate‐buffered saline

PBS

placental growth factor

PLGF

polyinosinic:polycytidylic acid

Poly(I:C)

quality control

QC

room temperature

RT

telomerase reverse transcriptase

TERT

tetramethylrhodamine isothiocyanate

TRITC

transforming growth factor‐beta

TGFβ

transmission electron microscopy

TEM

tumor necrosis factor‐alpha

TNFα

vascular endothelial growth factor‐A

VEGF‐A

vascular endothelial growth factor‐C

VEGF‐C

vascular endothelial growth factor‐D

VEGF‐D

1. Introduction

As the field of microphysiological systems continues to evolve, there is a growing need for platforms that are scalable, automation‐compatible, and reproducible, features essential for broader adoption in drug discovery and disease modeling [1]. Traditional closed‐channel microfluidic systems pose challenges in terms of accessibility, integration with liquid‐handling robotics, and standardization [2, 3, 4]. These limitations have prompted a shift toward open‐well formats, which offer simpler interfacing with commercially available robotic liquid‐handling systems, facilitate scalable sample collection, and support more flexible experimental workflows [5]. In parallel, gel‐based tissue models have gained attention for their ability to mimic native extracellular environments and offer the flexibility to construct complex 3D architectures [6]. So far, various strategies [7] for automation [8] and high‐throughput data acquisition [9, 10] have been adopted. But these approaches are often implemented in isolation. Here, we present AngioPlate384^TM integrated with an automation system, a workflow designed with the following four key aspects: (1) fully gel‐based, perfusable blood vessel models; (2) stromal co‐culture for enhanced physiological relevance; (3) a standardized 384‐well open‐top format enabling high‐throughput drug screening; and (4) end‐to‐end automation compatible with commercial liquid‐handling systems and plate readers. This integrated workflow reflects a broader goal of developing tissue models that are not only physiologically relevant but also robust and scalable for industrial and clinical translation.

Blood vessel models have emerged as a leading application of this new generation of tissue models and have provided new insights into drug transport and vascular disease progression [11, 12]. Yue and co‐workers developed a modular two‐layer microfluidic system that enabled precise control of interstitial flow and chamber geometry to study how shear stress gradients regulate angiogenesis and vessel density, achieving large‐scale perfusable networks with vertical anastomosis for drug screening [9]. Franca and colleagues revealed a mechanosensing role of perivascular cells in fibrotic extracellular matrix (ECM) environments, showing that perivascular cells, amplify endothelial dysfunction and inflammatory signaling via NOTCH3, and that silencing this pathway restores barrier integrity and normalizes IL‐8 secretion [13]. Rogers and co‐workers demonstrated that endothelial–pericyte (PC) co‐culture on opposing sides of a membrane stabilizes vascular barriers under inflammatory stress and reduces tumor necrosis factor‐alpha (TNFα)‐induced cytokine surges in a high‐throughput format compatible with multiple assays [14]. Ingber's group advanced systemic modeling by linking multiple vascularized organ chips with robotic fluid handling and real‐time imaging, enabling pharmacokinetic (PK) and pharmacodynamic (PD) studies and revealing how endothelial transport barriers shape drug distribution across interconnected organs [8]. Collectively, these innovations showcase how blood vessel models are evolving from static models to dynamic, multi‐cellular platforms that capture mechanobiology, inflammatory signaling, and pharmacological responses at scale.

While existing studies have provided important insights into vascular development, moving toward a more streamlined vascular discovery process, we developed an automated workflow that combines our previously established AngioPlate384 [15] with a robotic liquid‐handling system for scalable blood vessel model production and a plate reader for automated blood vessel barrier analysis. The tubular blood vessel model is fully gel‐based, membrane‐free, and enables luminal and interstitial flow through the surrounding gel matrix. Stromal cells such as PC and fibroblasts (FB) were incorporated to allow for the study of endothelial‐stromal interactions, offering insights into blood vessel stabilization and blood barrier integrity under both physiological and pathological conditions. The blood vessel model was further leveraged to study vascular injury by introducing various biochemical stimuli—including TNFα, transforming growth factor‐beta (TGFβ), interferon‐gamma (IFNγ), thrombin, polyinosinic:polycytidylic acid (Poly(I:C)), and lipopolysaccharide (LPS)—capturing cytokine signaling and endothelial responses under inflammatory, fibrotic, thrombotic, and infectious conditions. Furthermore, we developed a high‐throughput functional assay on our blood vessel‐on‐a‐chip platform to screen chemotherapeutic drugs.

To demonstrate the platform's adaptability, we expanded our cell sources beyond human umbilical vein endothelial cells (HUVEC) to include liver‐ and lung‐derived endothelial and stromal cells. This enabled the construction of organ‐specific blood vessel models and captured distinct responses to disease‐relevant stimuli such as fibrotic signaling. To further validate the reproducibility of our high‐throughput blood vessel‐on‐a‐chip system, teams from both academia (McMaster University, Canada) and industry (Merck & Co., Inc., Rahway, NJ, USA) independently replicated the setup. The industrial team used the same model of liquid‐handling system, obtained reagents from the same suppliers, and prepared the tissue setup according to the protocol provided by the academic team. Both groups achieved consistent blood vessel formation, with comparable permeability levels and biomarker expression, underscoring the robustness and translational potential of our approach.

2. Results

Building on our previously established AngioPlate384 platform [15], we developed an automated end‐to‐end workflow for blood vessel model production and analysis using a commercially available robotic liquid‐handling system and a plate reader. We streamlined the device fabrication by employing a customized 384‐well plate (Figure S1) with wells that have a maximum working volume of 120 µL. The plate features a patterned bottom sheet onto which sacrificial gelatin fibers were 3D‐printed. These fibers dissolve to create perfusable channels embedded in fibrin gel, which was cast using the robotic liquid‐handling system (Figure 1). Tissue barrier function, a key readout in our assay, was automated with the fluorescent reading function of a plate reader, avoiding the need for extensive image analysis. Vascular barrier maturation was achieved by co‐culturing HUVEC with stromal cells, specifically PC and/or FB. These two stromal cell types were selected based on prior studies demonstrating their individual roles in supporting vascular development and stability [14, 16, 17]. A total stromal cell concentration of 0.1 million cells/mL and 1:1 PC‐to‐FB ratio were chosen based on literature precedent [16, 18, 19, 20]. Stromal cells were then cultured from day 0 to day 4 in AngioPlate384 using medium supplemented with 10% fetal bovine serum (FBS) to provide additional matrix deposition prior to endothelial cell seeding (Figure 1f). On day 4, HUVEC were perfused into the hydrogel‐embedded channels, and the FBS concentration was reduced to 2% to limit stromal cell proliferation.

FIGURE 1.

Automated fabrication of high‐throughput blood vessels‐on‐a‐chip using AngioPlate384. (a) Each AngioPlate384 contains 128 tissue culture units for blood vessel construction. (b) Each unit includes an inlet well, a middle well, and an outlet well, connected by two microchannels: one linking the inlet and middle wells, and another connecting the middle and outlet wells. A sacrificial gelatin fiber spans all three wells, serving as the template for vascular lumen formation. Cross‐sectional view of a single AngioPlate384 unit assembled via (c) robotic liquid‐handling system. (d) Plates are incubated on a programmable rocker that alternates a 15° tilt every 5 min to enable hydrostatic pressure‐driven media perfusion. (e) Integrated plate reader for automated fluorescence‐based assay quantification and image acquisition. (f) Timeline and workflow for blood vessel‐on‐a‐chip assembly using the AngioPlate384 platform. On day 0, a hydrogel containing stromal cells is cast into the middle wells, and the sacrificial gelatin fiber is dissolved away to create a lumen. After 4 days of stromal cell maturation, EC are seeded into the resulting channel and cultures until day 13 at which point a perfusable blood vessel is formed. On day 14, treatment begins with the addition of drug‐containing media to the inlet and outlet wells, simulating intravenous or oral administration. For systemic exposure studies, treatment media can be added to all wells. Treatment continues until day 17, with daily media collection for cytokine analysis and replenishment with fresh media. On day 17, functional assays, such as permeability testing, and molecular analyses, including VE‐cadherin staining, may be performed to evaluate vascular barrier integrity. (g) The AngioPlate384 platform supports a wide range of applications, including the development of organ‐specific vascular models and the study of vascular injury and drug‐induced toxicity.

Analysis of cell nucleus orientation revealed that stromal cells promoted HUVEC alignment along the direction of flow (Figure S2a,c) [21], although they had minimal effect on cell elongation (Figure S2b). Confocal imaging further showed that co‐culture conditions resulted in higher expression of VE‐cadherin, a key component of endothelial adherens junctions, indicating enhanced endothelial junction integrity and blood vessel maturation (Figure 2a–d) [22]. A z‐stack‐derived cross‐sectional view confirmed the formation of an open lumen (Figure 2g), while transmission electron microscopy (TEM) imaging revealed stromal cells in close proximity (<1 µm) to endothelial cells (EC) (Figure 2j), closely mimicking native microvascular environments [23] and facilitating intercellular crosstalk [24]. These findings were further confirmed via 3D‐rendering of entire length of the confocal‐imaged blood vessel (Figure 2k–l).

FIGURE 2.

Establishment of mature blood vessels with stromal cells. (a–d) Representative images of blood vessels formed by different cell combinations on culture day 20: (a) HUVEC only, (b) HUVEC with FB, (c) HUVEC with PC, and (d) HUVEC with both PC and FB. Stromal cells, visualized via F‐actin staining, are embedded within the hydrogel and positioned adjacent to the blood vessels. Blood vessels are identified by VE‐cadherin and F‐actin staining, and all cell nuclei are counterstained with DAPI. (g) Higher‐magnification top and side view images of the region outlined by the white box in (d). (e,f,h,i) Quantification and visualization of dextran permeability using (e,f) TRITC‐labeled 65–85 kDa dextran and (h,i) FITC‐labeled 4 kDa dextran on culture day 11. Representative fluorescent images showing dextran diffusion after 1‐hour perfusion with the permeability medium. Apparent permeability (P _app) was assessed in four acellular or stromal‐only conditions (no cells, FB only, PC only, and PC + FB) and four vascularized conditions (blood vessels formed with: HUVEC‐only, HUVEC + FB, HUVEC + PC, and HUVEC + PC + FB). (*p < 0.05, ***p < 0.001, ****p < 0.0001, n = 5) (j) TEM imaging demonstrates co‐localization between EC and FB (top) and between EC and PC (bottom). (k) Confocal image of a blood vessel formed with HUVEC, PC, and FB, showing F‐actin staining of the cytoskeleton, CD31 expression in EC, and DAPI‐stained nuclei. (l) Magnified cross‐sectional view of the white‐boxed region in (k), highlighting interactions between stromal cells and the blood vessel structure. (m) Quantification of permeability, transport rate, and coefficient of variation (CV) for TRITC 65–85 kDa and FITC 4 kDa dextran across four AngioPlates384 on day 11. Three plates were prepared and tested at McMaster University, and one plate was tested independently by our industry partner, Merck & Co., Inc., Rahway, NJ, USA. (n _{per plate} = 45–78) (n,o) Representative images of blood vessels composed of HUVEC, PC, and FB, prepared manually (n) or via automated assembly (o). TRITC 65–85 kDa dextran (red) highlights the blood vessel lumen and confirms perfusability.

2.1. Different Co‐Cultures of Stromal Cells and HUVEC Exhibited Distinctive Cytokine Patterns

To assess how stromal cells influence endothelial maturation, we compared barrier function and cytokine profiles across monoculture and co‐culture conditions. To quantitatively assess barrier function, we conducted permeability assays on day 11 (Figure 2e,f,h,i) and day 18 (Figure S3) using 65–85 kDa tetramethylrhodamine isothiocyanate (TRITC)‐dextran and 4 kDa fluorescein isothiocyanate (FITC)‐dextran. Permeability was assessed using plate reader‐based fluorescence measurements after 60 min of continuous perfusion under rocking. This approach automated data collection and enabled scanning of over 100 tissues within minutes, compared to image acquisition, which can take up to an hour and is prone to variation caused by focal plane shifting. On day 11, all co‐culture conditions demonstrated significantly reduced permeability compared to HUVEC‐only controls, with the most pronounced improvements observed in blood vessels co‐cultured with PC or with both PC and FB. In these conditions, the dextran transport rate across the endothelium was reduced by more than half for both dextran sizes (Figure 2e,f,h,i). Importantly, stromal cells alone did not exhibit barrier‐forming properties, confirming the critical role of EC‐stromal interactions. By day 18, repeated permeability tests showed no significant changes within each group (Figure S3), suggesting that blood vessel barrier function remained stable for at least 18 days. The AngioPlate384 platform also demonstrated suitability for high‐throughput applications, with over 50 blood vessels successfully established on a single plate (Figure 2m–o). Blood vessel consistency across the plate was high, with coefficients of variation (CV) below 30% for 65–85 kDa dextran permeability, below 20% for 4 kDa dextran permeability (Figure 2m).

To investigate the molecular basis behind the observed differences in EC growth and barrier function (Figure 2e,f,h,i), we analyzed the cytokine profiles of three culture conditions: HUVEC alone, HUVEC co‐cultured with PC, and HUVEC co‐cultured with both PC and FB. The HUVEC + FB condition was excluded because permeability data showed only modest improvement in barrier function compared to HUVEC alone (Figure 2f,i). In contrast, HUVEC + PC and HUVEC + PC + FB demonstrated significant enhancements. Based on these observations, we focused on conditions that exhibited the most pronounced functional changes to better capture cytokine‐mediated mechanisms underlying barrier maturation. Cytokine levels were measured on day 7 (Figure 3a). Hierarchical clustering revealed that the cytokine signatures of the PC and PC + FB groups were more similar to each other than to the HUVEC‐only group, suggesting that stromal cell presence strongly influences endothelial signaling (Figure 3a). To further explore these differences, we identified cytokines that were significantly different in the stromal co‐culture groups relative to the HUVEC‐only (Figure 3b–i). In the HUVEC + PC group, we observed upregulation of key pro‐angiogenic cytokines including hepatocyte growth factor (HGF, 6.1 Log₂ fold change (FC)) [25], granulocyte‐colony stimulating factor (G‐CSF, 1.0 Log₂FC) [26], interleukin‐8 (IL‐8, 0.8 Log₂FC) [27], and vascular endothelial growth factor‐C (VEGF‐C, 0.6 Log₂FC) [28], along with downregulation of epidermal growth factor (EGF, −0.3 Log₂FC). Bone morphogenetic protein 9 (BMP‐9), a dual‐modulatory cytokine, was also upregulated (0.3 Log₂FC) [29] (Figure 3j–l). In the HUVEC + PC + FB group, similar trends were observed, with significant upregulation of HGF (6.2 Log₂FC), IL‐8 (0.8 Log₂FC), and VEGF‐C (0.4 Log₂FC), and downregulation of the vasoconstrictive endothelin‐1 [30] (−0.4 Log₂FC) (Figure 3m–o). Additional top upregulated cytokines in the HUVEC + PC group included fibroblast growth factor 1 (FGF‐1, 1.5 Log₂FC), vascular endothelial growth factor‐A (VEGF‐A, 1.1 Log₂FC), leptin (0.8 Log₂FC), vascular endothelial growth factor‐D (VEGF‐D, 0.6 Log₂FC), and angiopoietin‐2 (0.02 Log₂FC). For the HUVEC + PC + FB group, other top upregulated markers included VEGF‐A (1.1 Log₂FC), FGF‐1 (0.8 Log₂FC), endoglin (0.6 Log₂FC), G‐CSF (0.6 Log₂FC), leptin (0.4 Log₂FC), EGF(0.2 Log₂FC), Angiopoietin‐2 (0.1 Log₂FC), fibroblast growth factor 2 (FGF‐2, 0.1 Log₂FC), and placental growth factor (PLGF, 0.04 Log₂FC). Although not statistically significant, the most downregulated cytokine in the HUVEC + PC condition was endothelin‐1 (−0.7 Log₂FC), and in the HUVEC + PC + FB condition it was VEGF‐D (−0.4 Log₂FC) (Figure 3l,o).

FIGURE 3.

Cytokine analysis reveals the influence of stromal cells on vascular maturation. (a) The heat map displays differential expression of human angiogenesis cytokines and growth factors in blood vessels built with HUVEC alone, with HUVEC and PC, and with HUVEC, PC, and FB on day 11. Supernatants were obtained from n = 3 independent tissue culture units. (b–i) Cytokines with significantly different expression levels (HGF, IL‐8, VEGF‐C, G‐CSF, BMP‐9, Endoglin, EGF, and Endothelin‐1) across conditions: HUVEC‐only, HUVEC + PC, and HUVEC + PC + FB. (*p < 0.05, **p < 0.01) (j) Illustration of cytokines significantly modulated in the HUVEC + PC group compared to the HUVEC‐only group, including VEGF‐C, HGF, IL‐8, BMP‐9, G‐CSF, and EGF. (k) Volcano plot showing differentially expressed cytokines between HUVEC‐only and HUVEC + PC conditions. Significantly upregulated and downregulated cytokines (P < 0.05) are labeled. (l) Ranked expression of cytokines in the HUVEC + PC group, from most upregulated to most downregulated, relative to the HUVEC‐only group. (m) Illustration of cytokines significantly modulated in the HUVEC + PC + FB group compared to the HUVEC‐only group, including VEGF‐C, HGF, endothelin‐1, and IL‐8. (n) Volcano plot showing differentially expressed cytokines between the HUVEC‐only and HUVEC + PC + FB groups, with significantly upregulated and downregulated cytokines labeled (P < 0.05). (o) Ranked expression of cytokines in the HUVEC + PC + FB group, from most upregulated to most downregulated, relative to the HUVEC‐only group.

These results suggest that stromal cells contribute to the regulation of endothelial barrier function by enhancing the expression of pro‐angiogenic factors, such as HGF [25], and simultaneously downregulating molecules like endothelin‐1, which is a vasoconstrictor known to induce endothelial fibrosis and reactive oxygen species production [31]. Overall, our findings reveal distinct cytokine signatures associated with specific co‐culture configurations, supporting the role of stromal‐endothelial crosstalk in vascular function. Based on these results, we selected the HUVEC + PC + FB co‐culture condition for subsequent compound screening studies on AngioPlate384.

2.2. Modeling Pathophysiological Blood Vessels with Biological Compounds

We investigated the potential of AngioPlate384 to model a variety of vascular pathophysiological states blood vessels constructed with PC and FB. Blood vessels that met predefined quality control (QC) criteria on day 11 (P _app < 8 × 10⁻⁶ cm/s for 65–85 kDa dextran and P _app < 2.6 × 10⁻⁵ cm/s for 4 kDa dextran) were selected for downstream experimentation. From days 14 to 17, these blood vessels were exposed to a panel of biochemical compounds to induce distinct disease‐relevant phenotypes. Vascular responses, including changes in permeability and cytokine secretion profiles, were then assessed in a dose‐dependent manner (Figure 1f,g).

To model inflammation, we used IFNγ and TNFα, both of which are commonly used in vitro to induce vascular inflammation at concentrations ranging from 0.001 to 0.01 µg/mL for the former and 0.01 to 0.1 µg/mL for the latter [10, 32]. Their combination is known to exert synergistic inflammatory effects [33] in conditions such as atherosclerosis [34], sepsis [35], and autoimmune diseases [36, 37], including rheumatoid arthritis and multiple sclerosis [38]. Blood vessels were treated with escalating doses of IFNγ (0–1 µg/mL), TNFα (0–1 µg/mL), and their combinations, and were assessed on day 17, 72 h after treatment initiation. Permeability was measured across all conditions, and confocal imaging was performed for TNFα‐treated groups. Dose‐response curves were generated for TNFα treatment, and half‐maximal effective concentration (EC₅₀) values were calculated to determine the concentration at which half‐maximal permeability change occurred (Figure 4c,d). Following IFNγ treatment alone, no significant changes in vascular permeability were observed, except for a modest increase at 0.01 µg/mL for 65–85 kDa dextran (Figure 4a,b). Cytokine profiling under the highest IFNγ dose revealed a 3.4‐fold increase in monocyte chemoattractant protein‐1 (MCP‐1), a 2.2‐fold increase in interleukin‐4 (IL‐4), and a 0.2‐fold decrease in IL‐8 compared to the vehicle control (Figure 4e,f,m,o). The overall minimal change in permeability may therefore reflect a compensatory balance between these opposing signals. Additionally, other microenvironment‐specific factors could also be contributing to the observed stability in barrier function. In the TNFα‐only groups, significant permeability increases were observed at 0.01, 0.1, and 1 µg/mL for 65–85 kDa dextran, and at 0.01 and 1 µg/mL for 4 kDa dextran (Figure 4a,c). Cytokine analysis showed elevated levels of MCP‐1 (3.0‐fold), L‐13 (21.3‐fold) and granulocyte‐macrophage colony‐stimulating factor (GM‐CSF) (120.3‐fold) (Figure 4e–h). Confocal imaging revealed a dose‐dependent loss of VE‐cadherin signal, indicating junctional disruption and endothelial damage (Figure S4a). When TNFα was combined with 0.1 µg/mL IFNγ, barrier breakdown occurred at 0.1 µg/mL for 65–85 kDa dextran, and at 0.1 and 1 µg/mL for 4 kDa dextran (Figure 4a,d and Figure S4b). This condition led to substantial increases in MCP‐1 (2.9‐fold), interleukin‐13 (IL‐13, 17.9‐fold), GM‐CSF (355.1‐fold), interleukin‐12 subunit beta (IL‐12(p40), 14.0‐fold), interleukin‐1 receptor antagonist (IL‐1RA, 8.4‐fold), IL‐4 (2.6‐fold), and interleukin‐1 beta (IL‐1β, 6.6‐fold), (Figure 4e–i,k,m,n) when compared to the vehicle control. When comparing these values to the TNFα‐only group, both IL‐1β and IL‐1RA levels are elevated supporting a synergistic inflammatory effect consistent with previous findings [39].

FIGURE 4.

Development of blood vessels that mimic inflammation. (a) Representative fluorescent images showing diffusion of FITC‐labeled 4 kDa dextran and TRITC‐labeled 65–85 kDa dextran across blood vessels treated on days 14–17 with IFNγ alone (10⁻⁴–10⁰ µg/mL), TNFα alone (10⁻⁴–10⁰ µg/mL), or with a combination of TNFα and IFNγ (10⁻⁴–10⁰ and 0.1 µg/mL, respectively). (b) Apparent dextran permeability of IFNγ‐treated vessels as a function of dose (n = 23, *P < 0.05,). (c) Apparent dextran permeability of TNFα‐treated vessels treated across increasing doses (n = 17, *P < 0.05, **P < 0.01). (d) Apparent dextran permeability of TNFα‐ and IFNγ‐treated vessels across increasing TNFα doses (n = 16, *P < 0.05). (e) Heat map showing differential secretion of pro‐ and anti‐inflammatory cytokines in untreated vessels, IFNγ‐treated vessels (1 µg/mL), TNFα‐treated vessels (1 µg/mL), and vessels treated with both IFNγ and TNFα (0.1 and 1 µg/mL, respectively). (*P < 0.05, **P < 0.01, n = 9 for treated, n = 3 for untreated, pooled from days 14–17) (f–o) Quantification of differentially‐secreted cytokines (MCP‐1, IL‐13, GM‐CSF, IL‐12(p40), IFNγ, IL‐1RA, TNFα, IL‐4, IL‐1β, and IL‐8) across untreated controls, IFNγ‐treated vessels (1 µg/mL), TNFα‐treated vessels (1 µg/mL), and vessels treated with both IFNγ and TNFα (0.1 and 1 µg/mL, respectively). (*P < 0.05, **P < 0.01, n = 3 for each group, pooled from days 15–17).

We next investigated TGFβ, a cytokine known to drive fibrosis by promoting myofibroblast differentiation [40] and suppressing immune responses [41]. Given that TGFβ and IFNγ have been reported to reciprocally suppress each other's activities [42], we tested TGFβ across a concentration range (0–0.1 µg/mL) in standard medium, in combination with 0.1 µg/mL IFNγ. In standard media, TGFβ at 0.001 µg/mL significantly increased permeability for both dextran sizes (Figure 5b), but not at any other concentrations. Cytokine analysis following TGFβ treatment revealed reductions in IFNγ (0.5‐fold) and IL‐8 (0.6‐fold) (Figure 5j,n). Interestingly, at the same TGFβ concentration (0.001 µg/mL), combining it with IFNγ reduced permeability, suggesting blood vessel tightening (Figure 5a,c). Eight cytokines were significantly upregulated, including pro‐inflammatory interleukin‐12 p70 (IL‐12(p70), 25.1‐fold), interleukin‐2 (IL‐2, fivefold), IL‐1β (12.5‐fold) [43], IL‐13 (130.1‐fold), TNFα (1.8‐fold), MCP‐1 (2.9‐fold) and anti‐inflammatory markers [43, 44, 45] interleukin‐10 (IL‐10, 10.9‐fold), and IL‐4 (21.3‐fold),(Figure 5d–i,k–m). These results support a reciprocal regulatory interaction between TGFβ and IFNγ [42, 46], potentially mediated through Smad and JAK‐STAT signaling pathways [42, 47, 48]. These findings further establish the utility of AngioPlate384 for modeling both the anti‐inflammatory potential and barrier‐disruptive effects of TGFβ [42, 49].

FIGURE 5.

Development of blood vessels that mimic fibrotic conditions. (a) Representative fluorescent images showing diffusion of FITC‐labeled 4 kDa dextran and TRITC‐labeled 65–85 kDa dextran across vessels treated on days 14–17 with TGFβ alone (10⁻⁵–10⁻¹ µg/mL) or with a combination of TGFβ and IFNγ (10⁻⁵–10⁻¹ µg/mL and 0.1 µg/mL, respectively). (b) Apparent permeability of TGFβ‐treated vessels to 4 kDa and 65–85 kDa dextran as a function of TGFβ concentration (n = 22, *P < 0.05). (c) Apparent permeability of vessels treated with both TGFβ and IFNγ to 4 kDa and 65–85 kDa dextran across increasing TGFβ doses (n = 29, *P < 0.05). (d) Heat map showing differential secretion of pro‐ and anti‐inflammatory cytokines in untreated vessels, TGFβ‐treated vessels (0.1 µg/mL), and vessels treated with both TGFβ and IFNγ (both at 0.1 µg/mL). (*P < 0.05, ***P < 0.001, n = 6 for treated, n = 3 for untreated, pooled from days 15–17) (e–n) Quantification of differentially‐secreted cytokines (IL‐12(p70), IL‐2, IL‐1β, IL‐13, IL‐10, IFNγ, TNFα, IL‐4, MCP‐1, and IL‐8) across untreated controls, TGFβ‐treated vessels, and vessels treated with TGFβ and IFNγ. (*P < 0.05, ***P < 0.001, n = 3 for each group, pooled from days 15–17).

Next, we modeled thrombosis by increasing thrombin levels (0, 0.1, 0.5, and 1 U/mL) within the blood vessel lumen. Only the 1 U/mL thrombin group in heparin‐free conditions exhibited a significant increase in permeability for 65–85 kDa dextran (Figure S4e,f,i), supporting the role of thrombin in endothelial barrier breakdown when unopposed. To simulate viral infection, we applied Poly(I:C), a synthetic double‐stranded RNA analog [50], at concentrations ranging from 0 to 10 µg/mL. At the highest dose, blood vessels showed increased permeability for both dextran sizes (Figure S4g,i), with modest VE‐cadherin disruption on confocal imaging, less severe than in the TNFα group (Figure S4c). Cytokine analysis revealed a 3.4‐fold increase in MCP‐1 (Figure S4i), consistent with inflammatory activation. Finally, to model bacterial infection, we treated blood vessels with lipopolysaccharide (LPS, 0–10 µg/mL) [51]. However, no significant changes in permeability, VE‐cadherin expression, or cytokine secretion were observed (Figure S4d,h,i), suggesting the tested doses were insufficient to disrupt endothelial barrier function under the given conditions.

Collectively, these studies revealed distinct permeability and cytokine responses to TNFα, IFNγ, TGFβ, thrombin, Poly(I:C), and LPS. TNFα and IFNγ treatments induced marked barrier disruption and strong pro‐inflammatory cytokine secretion, while TGFβ triggered fibrotic signaling with reciprocal modulation by IFNγ. Thrombin and Poly(I:C) elicited thrombosis‐ and viral infection‐like responses, whereas LPS produced minimal changes under tested conditions, highlighting the platform's ability to capture both severe and subtle vascular injury signatures.

2.3. Screening Chemotherapy Drugs and Demonstrating Vascular Dose‐Dependent Responses

Given the clinical importance of vascular toxicity in chemotherapy [52, 53], we evaluated the impact of ten chemotherapy drugs on blood vessel integrity by screening them across physiologically relevant doses, testing five concentrations per drug ranging from 0 to 1 µm. Six of these drugs (bortezomib, vincristine, axitinib, imatinib, sorafenib, and tamoxifen) were tested at concentration ranges that span above and below their reported maximum plasma concentrations (C _max) in humans (Figure 6). This approach enables direct assessment of vascular toxicity at physiologically relevant doses. Concentration ranges were selected to reflect the biologically active forms of each drug, either as free plasma concentrations or active metabolites, providing a realistic approximation of in vivo cellular exposure. Because drugs are generally biologically active in their unbound form [54], we prepared the highest concentration from a 0.01 m stock solution of the drug in DMSO to achieve a final DMSO concentration of 0.01% in all treatments, minimizing any potential DMSO‐related effects on vascular function. The four drugs with C _max higher than the testing concentrations include paclitaxel, bleomycin, mitomycin, and amifostine (Figure 6d,f,g,i). The ten drugs were classified into two categories based on prior clinical and preclinical evidence [55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73]: vasculotoxic and non‐vasculotoxic. Following 72 h of exposure, vascular permeability was assessed using 65–85 kDa and 4 kDa dextran (Figure 6a). Drugs that caused permeability values exceeding the defined QC thresholds (P _app < 8 × 10⁻⁶ cm/s for 65–85 kDa dextran and P _app < 2.6 × 10⁻⁵ cm/s for 4 kDa dextran) were classified as vasculotoxic. Dose‐response curves were plotted for each drug (Figure 6b–k), from which half‐maximal toxic concentration (TC₅₀) can be derived to contextualize clinically relevant exposure and support interpretation of vasculotoxicity thresholds.

FIGURE 6.

Dose‐dependent responses to chemotherapy drugs assessed using the blood vessel assay on AngioPlate384. (a) Representative images showing dextran diffusion across a range of concentrations for bortezomib, vincristine, paclitaxel, axitinib, bleomycin, mitomycin, sorafenib, amifostine, imatinib, and tamoxifen. (b–k) Quantification of dextran transport rates across blood vessels treated with increasing doses of (b) bortezomib (n = 17), (c) vincristine (n = 14), (d) paclitaxel (n = 12), (e) axitinib (n = 11), (f) bleomycin (n = 9), (g) mitomycin (n = 11), (h) sorafenib (n = 13), (i) amifostine (n = 10), (j) imatinib (n = 13), and (k) tamoxifen (n = 13). The clinically relevant maximum plasma concentration (C _max) for each drug is indicated.

Among the tested compounds, several were expected to impair vascular function based on prior literature: bortezomib has been shown to induce apoptosis in EC [56, 57], vincristine can increase TNFα levels, leading to vascular dysfunction [55], paclitaxel disrupts microtubule function in EC [58, 59], and axitinib, an anti‐VEGF agent, is known to damage vasculature [60, 61]. We therefore anticipated that these four drugs would compromise blood vessel integrity on AngioPlate384. Other compounds, including bleomycin, mitomycin, and sorafenib, have been associated with vascular damage primarily through fibrotic or thrombotic mechanisms [59, 62, 63, 64, 65, 66, 67]. Bleomycin has been reported to induce extended cell‐cycle arrest, senescence, and mitotic cell death [74]. However, our use of the human telomerase reverse transcriptase (hTERT)‐immortalized endothelial cell line (HUVEC/TERT2) may not accurately reflect the true endothelial responses, as TERT expression can alter cell cycle progression [75]. In the case of mitomycin, which is known to induce thrombotic microangiopathy, the drug may require a longer exposure time to elicit changes in vascular tone [76]. A similar study by the Hughes group also reported minimal vascular disruption in their in vitro vascular network model, supporting the need to test longer durations and possibly higher doses [77]. Notably, the doses tested for both bleomycin and mitomycin in our study were below their reported C _max values (Figure 6f,g), due to limitations imposed by DMSO concentration in the treatment media. For sorafenib, a drug associated with cardiovascular toxicity, clinical data indicate that endothelial cell‐related adverse effects typically emerge after approximately six weeks of treatment in about 18.8% of patients [78]. In our model, we observed a trend toward vascular barrier disruption when using 4 kDa dextran, suggesting that longer treatment durations may be necessary to capture the chronic endothelial effects of sorafenib observed in clinical settings (Figure 6h). Amifostine, a known cytoprotective agent, has been reported to reduce vascular permeability and induce hypotension [68]. Imatinib, a tyrosine kinase inhibitor, has demonstrated anti‐angiogenic activity primarily by targeting PC rather than EC [70, 71]. Last, tamoxifen, a selective estrogen receptor modulator, has been shown to suppress angiogenesis while promoting endothelial repair [72, 73]. Accordingly, we expected minimal barrier disruption from these three agents (Figure 6i–k). The permeability assay outcomes aligned with our predictions: we observed 100% sensitivity in detecting the vascular‐disruptive effects of vincristine, bortezomib, paclitaxel, and axitinib, and 100% specificity in identifying bleomycin, mitomycin, sorafenib, amifostine, imatinib, and tamoxifen as non‐disruptive under our conditions (Figure 6b–k). These results validate the utility of AngioPlate384 for screening the vasculotoxic potential of chemotherapeutic agents in a high‐throughput, physiologically relevant model.

2.4. Organ‐Specific Blood Vessels Reveal Distinctive Patterns in Cytokines and Recovery

To investigate organ‐specific vascular responses, we replaced the stromal and EC in our system with organ‐matched sources to construct liver and lung blood vessels (Figure 7a,b,d,e). Control vessels composed of HUVEC, PC, and FB were compared to liver blood vessels (using liver EC and liver FB) and lung blood vessels (using lung EC and lung FB), and all constructs were evaluated using the same permeability assay. Permeability testing revealed comparable barrier function between the control and liver blood vessels, while lung blood vessels exhibited significantly tighter barriers (Figure 7f,g). Specifically, for 65–85 kDa dextran, lung blood vessels showed 0.65‐fold the permeability of both control and liver blood vessels (Figure 7f). For 4 kDa dextran, permeability was 0.78‐fold relative to control and 0.74‐fold relative to liver blood vessels (Figure 7g). These results align with established physiological differences: lung ECs typically form continuous, non‐fenestrated endothelium that restricts passage of large molecules [79], whereas liver sinusoidal ECs form fenestrated barriers that facilitate molecular exchange [80]. Confocal imaging also indicated lower VE‐cadherin expression in liver blood vessels compared to lung vessels, reflecting weaker adherens junctions in the liver endothelium (Figure 7a), consistent with prior studies [81, 82].

FIGURE 7.

Building organ‐specific blood vessels and modeling organ‐specific vascular responses. (a,b) Confocal images showing organ‐specific stromal cells (liver and lung) embedded within the hydrogel matrix, stained with F‐actin (red), and co‐localized with organ‐matched EC stained for VE‐cadherin (green) and F‐actin (red). Nuclei are counterstained with DAPI (blue). (d,e) TEM images were taken for both liver‐specific (d) and lung‐specific blood vessels (e), showing that liver FB and lung FB are located in the EC's vicinity. (f,g) Quantification and visualization of apparent dextran permeability to assess barrier function changes in blood vessels built with HUVEC, liver EC, and lung EC. The transportation rate of diffused TRITC‐labeled 65–85 kDa dextran (f) and FITC‐labeled 4 kDa dextran (g) dextran on day 11 is compared among three culture conditions: HUVEC with PC and FB (n = 38), liver EC with liver FB (n = 25), and lung EC with lung FB (n = 23). Fluorescent images of diffused dextran after 1 h perfusion of the permeability medium. (***P < 0.001, ****P < 0.0001) (c,h) Quantification and visualization of apparent dextran permeability to assess barrier function changes in blood vessels following TGFβ treatment and subsequent recovery. Permeability to TRITC‐labeled 65–85 kDa dextran (c) and FITC‐labeled 4 kDa dextran (h) is shown across three treatment conditions for each vessel type: HUVEC (vehicle: n = 12, TGFβ: n = 10, recovery: n = 10), liver EC (vehicle: n = 13, TGFβ: n = 7, recovery: n = 7), and lung EC (vehicle: n = 9, TGFβ: n = 6, recovery: n = 6). (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001) (i) Timeline for inducing pathophysiology by introducing TGFβ and allowing the blood vessels to recover from the diseased state. (j) Heat map showing differential expression of angiogenesis‐related human cytokines and growth factors in blood vessels composed of HUVEC, PC, and FB, compared to those constructed with liver EC and liver FB, or lung EC and lung FB on day 11. Supernatants were collected from n = 3 independent tissue culture units. (k) Volcano plot comparing cytokine expression between HUVEC and liver‐derived blood vessels, highlighting significantly upregulated and downregulated cytokines (P < 0.05). (l) Ranking of expression of cytokines in the liver blood vessels, ordered from most upregulated to most downregulated relative to HUVEC blood vessels. (m) Volcano plot comparing cytokine expression between HUVEC and lung‐derived blood vessels, highlighting significantly upregulated and downregulated cytokines (P < 0.05). (n) Ranking of expression of cytokines in the lung blood vessels, ordered from most upregulated to most downregulated relative to HUVEC blood vessels. (o–u) Measurements of differential concentrations of cytokines between the HUVEC blood vessels and the liver vessels: (o) leptin, (p) VEGF‐A, (q) IL‐8, (r) EGF, (s) HGF, (t) PLGF, (u) VEGF‐C. (*P < 0.05, **P < 0.01) (o) The measurement of differential concentrations of leptin between the HUVEC blood vessels and the lung vessels (*P = 0.0489).

To further elucidate the molecular drivers of these functional differences, we conducted cytokine profiling on day 7. Hierarchical clustering of cytokine expression showed that liver and lung blood vessels clustered together, distinct from the control group (Figure 7j). Compared to control blood vessels, liver blood vessels showed significant upregulation of leptin (0.7 Log₂FC), VEGF‐A (3.5 Log₂FC), IL‐8 (1.1 Log₂FC), and EGF (1.0 Log₂FC), alongside downregulation of HGF (–2.9 Log₂FC), PLGF (–1.7 Log₂FC), and VEGF‐C (–1.2 Log₂FC) (Figure 7k,l,o–u). Additional strongly downregulated, though not statistically significant, cytokines included follistatin (−2.9 Log₂FC), endoglin (−2.4 Log₂FC), angiopoietin‐2 (−2.3 Log₂FC), and G‐CSF (−1 Log₂FC) (Figure 7l). Lung blood vessels showed fewer significant changes compared to control, with leptin being the only cytokine significantly upregulated (0.6 Log₂FC) (Figure 7m–o). However, VEGF‐A (3.3 Log₂FC) and VEGF‐D (1.2 Log₂FC) also appeared among the top upregulated markers, and several cytokines—including G‐CSF (−2.0 Log₂FC), follistatin (−1.9 Log₂FC), angiopoietin‐2 (−1.4 Log₂FC), PLGF (−1.2 Log₂FC), and endoglin (−0.7 Log₂FC)—were downregulated (Figure 7n). The observed difference in HGF levels between liver and lung blood vessels reflects known organ‐specific biology—lung EC and FB actively secrete HGF to maintain barrier integrity [83], whereas liver EC produce only minimal HGF under healthy conditions [84], relying primarily on hepatic stellate cells (HSC) for angiocrine support [83, 85, 86, 87, 88, 89]. Conversely, IL‐8—a proinflammatory, proangiogenic cytokine known to increase endothelial permeability [90, 91, 92]—was the second‐most upregulated factor in the liver group. These opposing changes may contribute to the observed weakening of cellular junctions in liver blood vessels. This disparity, as recapitulated in our model, likely contributes to the tighter barrier phenotype observed in lung constructs compared to the higher permeability in liver constructs, underscoring its ability to capture organ‐specific vascular traits (Figure 7q,s).

To investigate how organ‐specific blood vessels respond to fibrotic signaling, we treated all blood vessel types with TGFβ—a known inducer of fibrosis [40, 41, 93, 94]—and tracked their recovery post‐treatment. Following QC screening, TGFβ was applied at 0.1 µg/mL from day 13 to 16, while control groups received vehicle. Permeability was assessed on day 16, and recovery was monitored over four subsequent days following TGFβ withdrawal (Figure 7i). On day 16, lung blood vessels exhibited the most pronounced increase in permeability with a 3.2‐fold increase for 65–85 kDa dextran and 2.7‐fold for 4 kDa dextran. Control blood vessels showed three and twofold increases, respectively, while liver blood vessels exhibited the smallest changes (1.54‐ and 1.75‐fold increases) (Figure 7c,h). These findings indicate that lung blood vessels are most sensitive to TGFβ‐induced barrier disruption, whereas liver blood vessels are relatively resistant within the AngioPlate384 system. Distinct differences were also observed in recovery. Lung blood vessels showed the greatest recovery, with permeability decreasing by 64.3% (65–85 kDa) and 42.7% (4 kDa) post‐treatment. The control group recovered by 59.5% and 26.2%, respectively, while liver blood vessels showed the lowest recovery (51.2% and 19.2%) (Figure 7c,h). These results suggest that AngioPlate384 can capture organ‐specific differences in vascular sensitivity to TGFβ and in their subsequent recovery trajectories. The data also highlight how underlying cellular and molecular characteristics—such as cytokine expression and adherens junction strength—shape the distinct responses of organ‐specific blood vessels to fibrotic cues.

3. Discussion

In this study, we present a scalable, high‐throughput platform for building complex 3D blood vessels‐on‐a‐chip using the 384‐well open‐top AngioPlate384 combined with an automated liquid‐handling system. This approach standardizes a previously developed vascularization method [15] by bioprinting 200 µm‐diameter sacrificial fibers directly into a 384‐well format. The resulting vascular networks are fully embedded in hydrogel and surrounded by stromal cells, offering a more physiologically relevant architecture. The open‐top design of the AngioPlate384 offers several advantages. It facilitates integration with automated liquid‐handling systems and simplifies large‐scale sample collection, enabling efficient workflows for high‐content screening. The open‐top, compartmentalized format also enables the simulation of various drug administration routes. For example, chemotherapeutic compounds can be added to only the inlet and outlet wells to mimic intravenous delivery, while local injections can be simulated by introducing compounds into the central wells. Additionally, apical versus basal compartments can be sampled independently to support PK/PD analyses, and this sample collection can be automated. Moreover, automation with AngioPlate384 is significantly simplified compared to traditional microfluidic platforms. Standard pipetting protocols can be directly applied to the 384‐well format without the need to align pipette tips into microchannels or ports, as is necessary with closed‐channel systems. In our protocol, we specifically avoid pipetting sub‐microliter volumes of reagents, making the system more robust and accessible to a wide range of users. A controlled shaking step during gel casting further enhances reproducibility across wells. Additionally, 3D printed structures are indirectly integrated into the plate design, allowing end users to create customizable tissue structures without needing access to a 3D printer. These workflow optimizations make AngioPlate384 particularly well‐suited for automated, large‐scale experimental pipelines.

In addition, the open‐top architecture supports basal‐side stimulation, a feature essential for mimicking systemic conditions such as inflammation, where biochemical signals act on both the apical and basal surfaces of the endothelium. We also demonstrated robust large‐scale data acquisition using AngioPlate384. Permeability measurements were fully automated using pre‐programmed protocols on a plate reader, and standardized assay conditions enabled consistent application across devices with minimal adjustments. Unlike permeability assays that rely on image analysis and are highly sensitive to focal plane variability, our plate reader‐based approach enables robust and rapid quantification of tissue barrier function with minimal user interference in the data collection and analysis. Each plate routinely yielded over 300 data points on permeability metrics; achieving a similar level of data acquisition with other platforms would require significantly more experimental cycles [8, 95].

Given the essential role of the vasculature in drug delivery, metabolism, and disease progression, particularly in cancer [96], there is a growing need for blood vessel models that are both physiologically relevant and scalable for compound screening. Our findings underscore the importance of stromal‐endothelial co‐culture. Incorporating PC and FB significantly enhanced blood vessel maturity and barrier integrity, as evidenced by reduced permeability and increased VE‐cadherin expression. This observation is consistent with previous reports highlighting the stabilizing role of stromal cells [14, 16, 17, 20]. The hydrogel‐embedded configuration facilitated close contact, often less than 1 µm, between stromal and EC, confirmed by TEM imaging. This proximity is essential for paracrine and juxtacrine signaling, which are known to support blood vessel formation and function [97, 98].

Incorporating stromal cells is critical for disease modeling. FB, for example, are primary mediators of TGFβ‐induced fibrosis [99], and we select organ‐specific FB to build organ‐specific fibrotic models. Our findings that lung blood vessels exhibit greater sensitivity to TGFβ‐induced barrier disruption compared to liver blood vessels are consistent with known organ‐specific mechanisms. The observed differences likely reflect organ‐specific FB phenotypes acting in concert with endothelial responses through microenvironment remodeling. In the lung, TGFβ directly increases endothelial permeability by promoting cytoskeletal contraction and junctional disassembly, facilitating rapid influx of plasma proteins and clotting factors [100]. In contrast, liver sinusoidal endothelial cells respond to TGFβ through a more gradual process of capillarization—loss of fenestrations and acquisition of a continuous capillary‐like phenotype—followed by partial endothelial‐to‐mesenchymal transition and, in some cases, myofibroblast‐like differentiation [101, 102]. This staged response likely explains the relative resistance of liver blood vessels to acute permeability changes in our model. Both endothelial and stromal cells are TGFβ‐responsive, and FB are primary mediators of fibrosis through ECM remodeling and cytokine secretion [103]. To further enhance physiological relevance for liver fibrosis modeling, future iterations of the platform could incorporate HSC, the major myofibroblast precursors in the liver, alongside endothelial and FB populations to better capture TGFβ‐driven fibrotic signaling and ECM remodeling. Beyond fibrosis, the platform could be adapted to model other organ‐specific diseases such as diabetic vasculopathy, neuroinflammation, or glomerulonephritis by modifying the stromal cell type and stimulation protocol.

While the AngioPlate384 platform already offers substantial versatility and physiological relevance, several areas still present exciting opportunities for further enhancement. The use of hTERT‐immortalized EC, such as TERT‐HUVEC, has enabled consistent performance and scalability, but these cells carry biological limitations, including elevated telomerase activity and potential for phenotypic drift [104]. A related limitation of our drug screening assay is that HUVEC/TERT2 cells may not fully replicate primary endothelial responses to certain agents, such as bleomycin, because TERT overexpression can alter cell cycle regulation and stress responses. To address this limitation, we have also successfully cultured primary liver and lung EC on the platform. Building on this, future work can explore the incorporation of induced pluripotent stem cell (iPSC)‐derived EC, including ‘reset’ vascular endothelial cells induced by the embryonic‐restricted ETS variant transcription factor 2 (EVT2) [105], which represent a promising avenue to enhance responsiveness to diverse microenvironments. Additionally, integrating smooth muscle cells and organ‐specific stromal cells such as podocytes (kidney) or astrocytes (brain) can further refine tissue‐specific blood vessel‐on‐a‐chip models, expanding the platform's relevance across a wider range of organ systems.

Flow dynamics within the current bi‐directional perfusion setup have proven sufficient to support endothelial alignment and tight junction formation, as evidenced by VE‐cadherin expression and reduced permeability. Looking ahead, the implementation of unidirectional flow patterns may be valuable for applications involving shear‐sensitive phenomena such as immune cell rolling and transmigration. Our recent work demonstrated that unidirectional, gravity‐driven perfusion on the UniPlate platform enables shear‐dependent behaviors such as immune cell rolling and transmigration [106]. Similar strategies could be adapted to AngioPlate384 by combining programmable tilt cycles with minor well‐plate design adjustments to promote sustained forward flow.

Matrix composition also offers a powerful lever for tuning the vascular microenvironment. While fibrin was strategically chosen for its pro‐vasculogenic properties, it may reflect an injury‐like ECM state. Exploring collagen‐based hydrogels as an alternative could provide a more homeostatic ECM mimic, particularly given collagen's prevalence in native interstitial tissue and its compatibility with the platform's thermal and automated fabrication workflow. An important consideration is that the ECM used in our organ‐specific blood vessel studies was not tailored to match organ‐specific composition. Fibrin was selected because of its ability to support robust blood vessel formation and its compatibility with the AngioPlate384 fabrication process. Future studies could explore organ‐matched ECM compositions or hybrid hydrogels incorporating collagen, laminin, or organ‐specific matrix proteins to better mimic physiological environments and refine organ‐specific blood vessel models.

To extend the platform's applicability to personalized medicine, patient‐derived cells offer tremendous potential. These cells could enable individualized modeling of drug responses and disease phenotypes. Although our current drug screening experiments were limited in scope due to resource constraints, Future studies could include additional donors to account for donor‐to‐donor and sex variability. The AngioPlate384 is also well positioned for integration with advanced omics technologies. Single‐cell RNA sequencing, mass spectrometry, and matrix profiling could provide high‐resolution insight into gene expression dynamics and ECM remodeling during vascular injury or drug exposure. Moreover, the platform's open‐top and standardized design is fully compatible with transendothelial electrical resistance (TEER) measurement systems as we have shown previously [107], facilitating real‐time electrophysiological assessment of barrier function.

4. Conclusion

The automated workflow that combines the AngioPlate384 platform with robotic liquid‐handling system and plate reader advances blood vessel modeling by uniting four key design approaches: a gel‐based, perfusable tissue architecture that eliminates membranes, stromal‐endothelial co‐culture to enhance physiological fidelity, a standardized 384‐well format for high‐throughput screening, and end‐to‐end automation for reproducibility and scalability. Together, these features enable robust modeling of vascular biology, including endothelial barrier maturation, cytokine‐mediated signaling, and organ‐specific vascular responses to injury. This versatility supports modeling across inflammatory, fibrotic, thrombotic, and infectious conditions, and demonstrated strong predictive accuracy in vascular drug screening. As microphysiological systems continue to evolve, the AngioPlate384 offers a compelling solution for bridging the gap between biological relevance and experimental scalability, paving the way for more predictive and personalized approaches in drug development, disease modeling, and precision medicine.

5. Materials and Methods

5.1. AngioPlate384 Fabrication and Procurement and Cell Culture

AngioPlate384 devices were fabricated according to previously published methods [15, 106, 108] and were also obtained commercially from OrganoBiotech Inc. (Cat# A002). Each AngioPlate384 consists of a customized 384‐well plate featuring an open‐top design with three interconnected wells per unit—an inlet, middle, and outlet well—linked by two microchannels. A sacrificial gelatin fiber spans all three wells and serves as the template for lumen formation. The bottom sheet of the plate is patterned with 3D‐printed fibers to support hydrogel casting, ensuring precise alignment and reproducibility. After fabrication, plates were sterilized, packaged in sealing bags, and supplied ready for hydrogel casting and cell seeding. All plates used in this study adhered to identical specifications to ensure reproducibility across experiments.

Original vials of human umbilical vein endothelial cells (HUVEC, EverCyte, CHT‐006‐0008), human primary lung fibroblasts (lung FB, ATCC, PCS‐201‐013) at passage 2 (P2), human pericytes from placenta (PC, PromoCell, C‐12980) at P2, human primary liver sinusoidal microvascular endothelial cells (liver EC, CellBiologics, H‐6017) at P3, human primary liver fibroblasts (liver FB, CellBiologics, H‐6019) at P3, and human primary lung microvascular endothelial cells (lung EC, CellBiologics, H‐6011) at P3 were obtained for cell culture. HUVEC, liver EC, lung EC, and liver FB were cultured in T75 flasks (CELLSTART, 82050‐856) coated with a 0.2% (w/v) gelatin solution (Sigma‐Aldrich, G9391) prepared in phosphate‐buffered saline (PBS, Gibco, 14190144) for 20 min at 37°C with 5% CO₂. Lung FB and PC were cultured in uncoated T75 flasks under the same environmental conditions. HUVEC were cultured in Endothelial Cell Growth Medium 2 (ECGM2, PromoCell, C‐22011) supplemented with 20 µg/mL G418 (Invivogen, ant‐gn‐1). Liver EC and lung EC were cultured in ECGM2 supplemented with an additional 3% FBS (Thermo Fisher Scientific, 12484028). Both lung FB and liver FB were cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco, 11 995 065) supplemented with 10% FBS, 1% HEPES (Gibco, 15 630 106), and 1% Penicillin‐Streptomycin (Pen‐Strep, Wisent Inc., 450‐201 EL). PC were cultured in Pericyte Growth Medium 2 (PGM2, PromoCell, C‐28041). All EC and FB were harvested using Trypsin‐EDTA (0.05%, Gibco, 25 300 054), while PC were harvested using Accutase (Gibco, A1110501). Working cell banks were established by expanding HUVEC to P4, and lung FB, PC, liver EC, liver FB, and lung EC to P6. HUVEC, lung FB, and PC were cryopreserved at a concentration of 7 × 10⁵ cells/mL in a freezing medium composed of 5% dimethyl sulfoxide (DMSO; Sigma‐Aldrich, D2650), 20% FBS, and 75% corresponding culture medium. Liver EC, liver FB, and lung EC were cryopreserved at 5 × 10⁵ cells/mL in a freezing medium containing 10% DMSO, 50% FBS, and 40% culture medium. Each cell suspension was aliquoted into cryogenic vials (VWR, 66 008‐751) at 1 mL per vial. For each round of experiments, fresh frozen vials were thawed from the respective working cell banks.

5.2. Hydrogel Casting

5.2.1. Manual Preparation

To generate human blood vessels the AngioPlate384 manually, stromal cells nearing confluence in T75 flasks were harvested on day 0, counted, and added into autoclaved 1.5 mL microcentrifuge tubes (VWR, 10025‐726) for centrifugation. The cell densities used for different experimental setups are detailed in Table S1. Following centrifugation, the supernatant was removed, and each cell pellet was resuspended in 125 µL of 10 mg/mL of Fibrinogen (Sigma‐Aldrich, F3879). Then, 25 µL of 7 U/mL Thrombin (Sigma‐Aldrich, T6884, stock solution at 10 U/mL dissolved in 0.1% w/v bovine serum albumin (BSA, Sigma‐Aldrich, A9418) in PBS, was added to the cell suspension. The mixture was pipetted up and down five times to ensure uniform mixing. Immediately after mixing, 25 µL of the suspension was dispensed into each of five user‐selected middle wells of the AngioPlate384, followed by a light tap on the plate to promote even distribution of the hydrogel within the wells. Because fibrin gelation occurs rapidly, pre‐gel solution for five wells was prepared at a time to minimize the interval between mixing and casting. This five‐well manual casting approach was optimized to ensure reproducibility and reduce variability, and the process was repeated in multiple rounds until all required wells were prepared.

5.2.2. Robotic Preparation

For robotic casting of human blood vessels on the AngioPlate384, stromal cells at near confluence in T75 flasks were harvested on day 0, counted, and distributed into four autoclaved 2 mL microcentrifuge tubes (VWR, 10025‐738) for centrifugation. After centrifugation, the supernatant was removed, and the cell pellets were resuspended in PBS. For blood vessels containing HUVEC, lung FB, and PC, a total stromal cell concentration of 0.1 million/mL was achieved. Depending on the number of AngioPlate384 columns to be prepared, four tubes containing 20 mg/mL Fibrinogen and four tubes containing 7 U/mL Thrombin were prepared and positioned at designated locations within the robotic handling system (Hamilton Company, NIMBUS4), as shown in Figure S5. The required volumes of Fibrinogen and Thrombin for different blood vessel quantities are listed in Table S2. A standard non‐treated 384‐well flat‐bottom plate (VWR, 732‐2906) was placed inside the system as a mixing reservoir, alongside the AngioPlate384, which was positioned on an integrated shaker. Once all the tubes and plates were securely positioned at designated locations (Figure S5), the system carried out automated hydrogel mixing and casting (Video S1) according to the programming flowchart provided in Figure S6.

Following casting, the AngioPlate384 was incubated on a level surface at room temperature for 30 min to allow crosslinking of Fibrinogen and Thrombin, forming a fibrin hydrogel with stromal cells suspended within the matrix. To remove the sacrificial gelatin fibers from the AngioPlate384, warm PBS was added to each well—100 µL to inlet and outlet wells, and 50 µL to middle wells—followed by incubation at 37°C with 5% CO₂ on a rocker (OrganoBiotech Inc., IFlowRocker) set to ±15°, alternating angle every 5 min. After 40 min, the PBS was aspirated and replaced with fresh PBS for an additional 20‐min incubation under the same conditions. The wells were then aspirated again. Finally, ECGM2 medium supplemented with 8% (v/v) FBS, 20 µg/mL aprotinin (Sigma‐Aldrich, 616370‐M) [109], and 1% Pen‐Strep was added to each well—90 µL to inlet and outlet wells, and 50 µL to middle wells. Stromal cells were cultured on the rocker at 37°C with 5% CO₂ from day 0 to day 4, with daily media changes to support maturation.

5.3. Blood Vessel Formation

On day 4, nearly confluent EC were harvested, counted, and transferred into a 50 mL Falcon tube for centrifugation. The resulting pellet was resuspended to a final concentration of 5 × 10⁵ cells/mL in ECGM2 supplemented with 20 µg/mL aprotinin and 1% Pen‐Strep. Culture media were removed from all wells, and 120 µL of the EC suspension was added to both the inlet and outlet wells of each channel. Equal volumes of EC suspension were added to the inlet and outlet wells to establish a hydrostatic pressure gradient, where the inlet and outlet wells maintained a higher liquid level than the middle well. This difference promoted cell delivery into the lumen from both ends. The approach was particularly effective in the AngioPlate384's open‐top, membrane‐free design, where the hydrogel's porosity supported interstitial flow and uniform cell distribution along the lumen. The plate was then incubated for 2 h at 37°C with 5% CO₂ on a leveled surface, allowing HUVEC to enter the hollow tubular structures within the hydrogel and adhere to the matrix. Following this 2‐hour incubation, 40 µL of ECGM2 supplemented with 20 µg/mL aprotinin and 1% Pen‐Strep was added to the middle wells. Co‐cultures of EC and stromal cells were maintained on the rocker at 37°C with 5% CO₂ to support blood vessel maturation. Media were refreshed daily by adding 90 µL to inlet and outlet wells and 50 µL to middle wells. The structural integrity and quality of the resulting blood vessels were evaluated starting from day 11 and onward for downstream applications.

5.4. Automated Permeability Assay

To evaluate the barrier function of the engineered blood vessels, a dextran solution was perfused through the blood vessel and molecular permeability was assessed via fluorescent imaging. The dextran solution was prepared using ECGM2 supplemented with 1 mg/mL of 65–85 kDa TRITC‐dextran (Sigma‐Aldrich, T1162) and 1 mg/mL of 4 kDa FITC‐dextran (Sigma‐Aldrich, 46944). A standard curve was generated by serial dilution of this dextran solution to final concentrations of 1, 0.1, 0.01, 0.001, 0.0001, and 0 mg/mL. For each concentration, 100 µL was added to three separate wells in a 384‐well plate for standard curve development. Each middle well was filled with 60 µL of ECGM2, while 90 µL of the dextran solution was added to both the inlet and outlet wells. Immediately after permeability assay media setup, the AngioPlate384 was placed into a plate reader (Agilent, BioTek Cytation 5) for fluorescence intensity measurement. Dual‐wavelength detection was configured with excitation/emission settings of 544/570 nm for TRITC and 491/516 nm for FITC. Fluorescence intensities of all sample wells were recorded at time points t = 0 and t = 60 min. During the 60‐min assay, the AngioPlate384 was incubated at 37°C with 5% CO₂ on the rocker to maintain perfusion.

Following the final fluorescence reading, fluorescent imaging was performed using the same plate reader equipped with TRITC and FITC filters and a 4× objective to obtain qualitative visual data. The dextran solution was then aspirated, and all wells were washed twice with ECGM2—90 µL for inlet and outlet wells and 50 µL for middle wells. For continued culture, ECGM2 supplemented with 20 µg/mL aprotinin, and 1% Pen‐Strep was added to all wells. Dextran transport from the blood vessel lumen into the surrounding hydrogel over the 60‐min interval was quantified by correlating fluorescence intensity to dextran mass using the standard curves. The transport rate (µg/h) was calculated based on the change in dextran mass between the t = 0 and t = 60‐min time points. Data analysis was conducted using Microsoft Excel, GraphPad Prism 10.2.1, and Fiji (ImageJ) version 2. Following each permeability assay and the final fluorescence reading, dextran‐containing media were completely aspirated, and two immediate washes were performed right after the assay to remove residual dextran. Daily media changes were then carried out for at least three consecutive days to ensure complete clearance from the hydrogel matrix before the next permeability assay.

To convert transport rates (dextran diffusion in µg/h) into apparent permeability (P _app), Equation (1) was used [110, 111]:

$$P_{\mathrm{app}}\left[\mathrm{cm}/\mathrm{s}\right]=\frac{dQ}{dt}\times \frac{1}{A\times C_{\mathrm{donor}}}$$ (1)

where dQ/dt is the transport rate (µg/h), A is the surface area of the blood vessel in contact with the hydrogel (1.06 × 10^{− 2} cm²), and C _donor is the initial dextran concentration (1 mg/mL). After simplification, the relationship between dQ/dt and P _app can be expressed as shown in Equation (2):

$$P_{\mathrm{app}}\left[\mathrm{cm}/\mathrm{s}\right]=\frac{dQ}{dt}\times 2.62\times {10}^{-5}\left[\frac{\mathrm{cm}\times \mathrm{h}}{\mu \mathrm{g}\times \mathrm{s}}\right]$$ (2)

This relationship allows for direct conversion of transport rate to P _app (cm/s), enabling comparison across devices and models. Using this method, we calculated average P _app values of 4.16 × 10^{− 6} cm/s for 65–85 kDa dextran and 2.38 × 10^{− 5} cm/s for 4 kDa dextran (Figure 2m), values that are approximately five to six times higher than those reported in vivo but consistent with other in vitro monolayer and 3D tubular models [112, 113].

5.5. Treatments

Three treatment base media were prepared using ECGM2 as the foundation. Treatment base medium I consisted of ECGM2 supplemented with 20 µg/mL aprotinin and 1% Pen‐Strep. Treatment base medium II included the same supplements as base I with the addition of 0.1 µg/mL IFNγ. Treatment base medium III was prepared using ECGM2 Kit (PromoCell, C‐22111) without heparin and supplemented with 20 µg/mL aprotinin and 1% Pen‐Strep. In treatment base medium I, various doses of inflammatory and fibrotic stimuli were prepared as follows: IFNγ and TNFα at 1, 0.1, 0.01, 0.001, 0.0001, and 0 µg/mL; TGFβ at 0.1, 0.01, 0.001, 0.0001, 0.00001, and 0 µg/mL; thrombin at 1, 0.5, and 0.1 U/mL; and Poly(I:C) and LPS at 10, 1, 0.1, 0.01, 0.001, and 0 µg/mL. In treatment base medium II, TNFα was tested at 1, 0.1, 0.01, 0.001, 0.0001, and 0 µg/mL, and TGFβ was tested at 0.1, 0.01, 0.001, 0.0001, 0.00001, and 0 µg/mL. In treatment base medium III, thrombin was tested at 1, 0.5, and 0.1 U/mL. Treatment media was added to all wells, with the exception of the thrombin group, where thrombin‐containing media was added only to the inlet and outlet wells to mimic localized thrombosis [114].

A panel of chemotherapy drugs obtained from the NIH (Table S3) was used to prepare drug treatment media using treatment base medium I. These drugs included bortezomib, vincristine, paclitaxel, axitinib, bleomycin, mitomycin, sorafenib, amifostine, imatinib, and tamoxifen, and were tested at concentrations of 1 × 10^{− 6}, 1 × 10^{− 7}, 1 × 10^{− 8}, 1 × 10^{− 9}, and 0 m. A vehicle control was prepared by adding 0.01% DMSO to treatment base medium I. Permeability tests were conducted on day 11 for QC. Treatments were applied from day 14 to day 17, with fresh media prepared on day 13 and changed daily. Cytokine samples were collected daily from day 15 to day 17, consisting of 80 µL from the inlet and outlet wells and 50 µL from the middle wells. During treatment, drug‐containing media was added only to the inlet and outlet wells, while drug‐free media was used in the middle wells to simulate PK/PD profiles of intravenous or oral drug administration.

To study organ‐specific vascular responses to TGFβ and their recovery after removal of TGFβ, doses of 0.1 and 0 µg/mL TGFβ were prepared in treatment base medium II. Permeability testing was conducted on day 10 as a QC measure. Treatments were administered from day 13 to day 16, with fresh media prepared on day 12 and changed daily. From day 16 to day 20, tissues previously exposed to TGFβ were cultured in treatment base medium II without TGFβ. Media was refreshed daily, with new preparations made on day 16 and day 18. In this set of experiments, treatment media was applied to all wells. A complete list of all compounds and drugs used in this study, including catalog numbers and stock concentrations, is provided in Table S3.

5.6. Cytokine Analysis

Culture media was collected at relevant time points—24, 48, and 72 h after treatment initiation—and stored at −80°C until analysis. For each sample, media collected across the different time points were pooled and transferred into 1.5 mL microcentrifuge tubes, followed by centrifugation at 1000 RPM for 10 min at 4°C. A volume of 75–100 µL of the resulting supernatant was carefully transferred into 0.65 mL microcentrifuge tubes and stored at −80°C until further analysis. Cytokine concentrations were measured by Eve Technologies using two multiplex assays. The Human Angiogenesis and Growth Factor 17‐Plex Discovery Assay was used to assess samples from different vessel culture conditions, while the Human Cytokine Proinflammatory Focused 15‐Plex Discovery Assay was used for all treatment groups. Data analysis, including quantification and clustering of cytokine expression profiles, was performed using GraphPad Prism version 10.2.1 and MATLAB R2023a.

5.7. Immunofluorescent Staining and TEM Imaging

Samples were washed twice with PBS and fixed in 4% paraformaldehyde (PFA, Electron Microscopy Sciences, EMS 15710‐S, diluted in PBS) for 1 h at room temperature (RT). Following fixation, samples were washed three times with PBS at 15‐min intervals and subsequently blocked with 10% FBS for 1 h at RT. Primary antibody staining was performed overnight at 4°C on a rocker using anti‐VE‐cadherin (Abcam, ab33168, 1:1000 dilution) or antiCD31 (Abcam, ab9498, 1:100 dilution), diluted in 2% FBS. After incubation, samples were washed with PBS over two consecutive nights on a rocker at 4°C, with PBS replaced once after the first night. Secondary and conjugated antibody staining was then performed overnight at 4°C on a rocker. Antibodies included Alexa Fluor 488 anti‐rabbit (Abcam, ab150077, 1:200 dilution), Phalloidin‐iFluor 594 Conjugate (Cayman Chemical, 20553, 1:1000 dilution), or DAPI (Sigma‐Aldrich, D9542, 1:1000 dilution), all diluted in 2% FBS. Following staining, samples were again washed with PBS over two nights on a rocker at 4°C, with one PBS change after the first night. Confocal imaging was conducted using a ZEISS 3i Marianas LightSheet microscope.

TEM was used to visualize the ultrastructure of the vessels and surrounding stromal cells. To retrieve samples from the AngioPlate384, the bottom sheet at the base of each well was cut along the well walls from the underside of the plate. Samples were carefully extracted using tweezers along with the adhesive backing. The isolated samples were then submitted to the Faculty of Health Sciences Electron Microscopy Facility at McMaster University for processing. TEM imaging of the processed sections was performed at the Canadian Centre for Electron Microscopy (CCEM).

5.8. Imaging Analysis

DAPI‐stained images of the blood vessels were analyzed in Fiji 2. After applying appropriate thresholding and performing automated measurements, quantitative data on nuclear orientation, circularity, and alignment were exported to MATLAB R2023a and GraphPad Prism for further analysis and visualization. Polar histograms and bar graphs were generated to represent these metrics. Nuclei oriented within 10° of the blood vessel centerlines were classified as “aligned” with the direction of flow.

5.9. Statistical Analysis

All data were presented as mean ± standard deviation (SD). Normality of the data was assessed using the Shapiro–Wilk test. If all groups passed the normality test, a Gaussian distribution was assumed and ANOVA parametric tests were applied; otherwise, nonparametric tests were used. Equality of variances was determined by calculating the ratio of the largest to the smallest SD among the groups. If this ratio was ≤4, equal variances were assumed; otherwise, unequal variances were considered. When the data was assumed to be normally distributed with equal variances, statistical significance was evaluated using ordinary one‐way ANOVA followed by Fisher's least significant difference post hoc test. If normality was assumed but variances were unequal, Brown–Forsythe and Welch ANOVA tests were applied, followed by Dunnett's T3 multiple comparisons test. For non‐Gaussian distributions, statistical significance was assessed using the Kruskal–Wallis test followed by Dunn's multiple comparisons test. Significance levels were denoted as follows: ns (not significant), * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001. Dose–response curves were fitted using nonlinear regression with a four‐parameter logistic model to calculate EC₅₀. All statistical analyses were performed using GraphPad Prism version 10.2.1.

Author Contributions

D.S.Y.L. and B.Z. designed the experiments. D.S.Y.L. performed the experiments and analyzed the data. H.M.H. assisted with compound screening. A.C., J.B., and N.A. assisted with data analysis. K.A.J. performed TEM imaging and provided technical support. S.R., F.Z., R.Y.C., N.K.S., and L.M.M. provided technical support. D.S.Y.L. and B.Z. interpreted the data. D.S.Y.L. wrote the manuscript. D.S.Y.L., S.K., and B.Z. edited the manuscript. M.D. and Y.H. supervised the work conducted at Merck & Co., Inc., Rahway, NJ, USA. B.Z. supervised all aspects of the work.

Conflicts of Interest

A PCT application on the technology has been filed by OrganoBiotech, Inc. D.S.Y.L, S.R., and B.Z. hold equity in the company. AngioPlate384^TM is commercialized by OrganoBiotech, Inc., D.S.Y.L., S.R., and B.Z. are co‐founders and hold equity in the company. B.Z. also received research support from Merck Sharp & Dohme LLC, a subsidiary of Merck & Co., Inc., Rahway, NJ, USA and OrganoBiotech, Inc. . D.S.Y.L., R.Y.C., N.K.S., L.M., M.D. and Y.H. are employees of Merck Sharp & Dohme LLC, a subsidiary of Merck & Co., Inc., Rahway, NJ, USA. The authors declare no other potential conflicts of interest related to the research, authorship, or publication of this article.

Supporting information

Supporting File 1: adhm70719‐sup‐0001‐SuppMat.docx.

Supporting File 2: adhm70719‐sup‐0002‐VideoS1.avi.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.