On-Chip Reconstitution of Uniformly Shear-Sensing 3D Matrix-embedded Multicellular Blood Microvessel

Abstract

Preclinical human-relevant modeling of organ-specific vasculature offers a unique opportunity to recreate pathophysiological intercellular, tissue-tissue, and cell-matrix interactions for a broad range of applications. Here, we present a reliable, and simply reproducible process for constructing user-controlled long rounded extracellular matrix (ECM)-embedded vascular microlumens on-chip for endothelization and co-culture with stromal cells obtained from human lung. We demonstrate the critical impact of microchannel cross-sectional geometry and length on uniform distribution and magnitude of vascular wall shear stress, which is key when emulating in vivo-observed blood flow biomechanics in health and disease. In addition, we provide an optimization protocol for multicellular culture and functional validation of the system. Moreover, we show the ability to finely tune rheology of the three-dimensional natural matrix surrounding the vascular microchannel to match pathophysiological stiffness. In summary, we provide the scientific community with a matrix-embedded microvasculature on-chip populated with all-primary human-derived pulmonary endothelial cells and fibroblasts to recapitulate and interrogate lung parenchymal biology, physiological responses, vascular biomechanics, and disease biogenesis in vitro. Such a mix-and-match synthetic platform can be feasibly adapted to study blood vessels, matrix, and ECM-embedded cells in other organs and be cellularized with additional stromal cells.

Graphical Abstract

This study presents design, development, and validation of a reliable and highly reproducible process for constructing user-controlled long rounded extracellular matrix (ECM)-embedded vascular microlumens on-chip for endothelization and co-culture with stromal cells. Organ-specific (lung in this case) primary human-derived endothelial cells, fibroblasts, and stiffness-adjustable hydrogels are used to recreate microvascular (MV) tissue architecture, physiological biology, vascular biomechanics, and disease biogenesis in vitro. Our method provides a state-of-the-art microphysiological system (referred to as ‘ECM MV-Chip’) that allows concurrent analysis of blood microvessels, perivascular matrix, and ECM-embedded stromal cells.

INTRODUCTION

Organs-on-Chips, commonly referred to as microphysiological systems (MPS), are biomimetic, microfluidic, cell culture devices that contain living tissue cells arranged to simulate tissue- and organ-level physiology, and have potentially emerged as powerful alternatives to animals and 2D culture models for preclinical research and trials^[1]. These devices contain continuously perfused hollow microchannels and/or chambers inhabited by living tissue cells arranged to simulate organ-level physiology^[1-2]. By recapitulating the multicellular architectures, tissue-tissue interfaces, chemical gradients, and mechanical cues, these devices produce levels of tissue and organ functionality not possible with conventional 2D or 3D culture systems. They also enable high-resolution, real-time imaging, and in vitro analysis of biochemical, genetic, and metabolic activities of living human cells in a functional tissue and organ context.

To reproduce diverse functions, pathophysiology, and dynamic cellular and biomechanical responses observed in blood vessels, extensive efforts have been made to emulate vasculature in Organs-on-Chips^[3]. These can be divided into (I) bottom-up efforts on creating microvascular networks within hydrogels^[3b-f], and (II) top-down efforts focused on developing well-defined perfusable vascular lumens on-chip^[3g-w]. Approach I relies on self-organization property of endothelial cells (EnCs), alone or in combination with other co-cultured cells, to form three-dimensional microvascular networks. While helpful in modeling vasculature, the network of blood vessels in this strategy either lacks perfusion or the vascular flow occurs but barrier function is not fully maintained throughout the established cellular network. Importantly, given the self-assembly nature of this method, exact control over vascular tube diameter, length, branching, or other geometrical features are often very limited. Approach II relies on seeding vascular EnCs in pre-formed hollow tubes embedded within a hydrogel or microfluidic lumens. Coating^[3g-j], stamping^[3k-m], casting^{[3n-r, 4]}, 3D-bioprinting^[3s-u], and viscous fingering^[5] are the most widely used methods in this approach. Despite their potential for vascularized tissue modeling, these methods have several drawbacks. For instance, for coating^[3g-j], the inner surfaces of a microfabricated channel are coated with extracellular matrix (ECM) protein(s) prior to EnC seeding. In the majority of cases, coating is performed on device lumens with rectangular or square cross-sections rather than in vivo-like fully circular/oval cross-sections. Importantly, the thin layer of ECM coating is not always maintained throughout the cell culture and, as such, EnCs are exposed to supra-physiologically stiff non-biological material (such as polydimethylsiloxane [PDMS], glass, or polymethyl methacrylate [PMMA]) used in device fabrication. Similarly, a stamping^[3k-m] approach typically generates square, rectangular, or trapezoidal cross-sectional vascular channels. Moreover, while it is possible to generate ECM-embedded vascular channels with this method, it is difficult to incorporate the vasculature with other components of the chip, such as a live airway lumen in a Lung-on-a-Chip for downstream applications. The casting and bioprinting methods, while offering the ability to generate rounded blood vessels in vitro, have notable challenges in their utility and broader adaptation. Casting^[3n-r], by which a lumenized endothelial compartment is formed within a hydrogel, often (1) utilizes fragile materials, such as PDMS rods to cast vascular channels, that have limited length and commonly require pre-treatment to render them hydrophobic, and (2) suffers medium-to-high failure rates on adaptation and reproduction. In addition, even when using more rigid materials for lumen formation, such as stainless-steel needles, there is a high likelihood of hydrogel pull-out on needle removal, and thus channel collapse and device failure. Bioprinting^[3s-u] is a more elaborate process and relies on removing sacrificial bio-inks (such as carbohydrate glass^[3s], pluronic acid^[3t], or wax^[3u]) cast as a 3D lattice from polymerized matrix surrounding them. While this is suitable for fabrication of relatively thick tissues and geometrically complex structures, the uniformity and smoothness of the vascular network is limited by the printing resolution as well as the deformation of the printed bio-ink. Additionally, 3D-bioprinting of vasculature is only possible with a limited pool of materials that can be used as sacrificial bio-ink; these must be easily dissolvable, either thermally or chemically, to be evacuated following scaffold ECM polymerization. Viscous fingering^[5] uses high pressure to flush out ECM hydrogel filled in a small channel to create vascular lumen. While it is practical to create rounded vascular channels, viscous fingering is only capable of creating channels surrounded by a thin ECM layer, ~100μm thick, comparable to that in the coating method. Moreover, it is challenging to change the channel geometries such as its length and diameter as it requires a whole adjustment on the optimal applied pressure. In addition, this method often suffers high level of variability (in precision of channel dimensions) between independent experiments and different users.

The complexity of these approaches, their inherent variabilities, and the discussed drawbacks have partly hampered their widespread adaptation and application. Thus, here we present a robust and reliable casting-based method that allows one to create cross-sectionally rounded ECM-embedded vascular microlumens of desired length in a controlled manner in an Organs-on-Chips device in vitro for endothelization and co-culture with stromal cells obtained from an organ of interest. Our approach offers high fidelity and enables reconstitution of simple, yet well-defined, uniform, and smooth vascular channels. Moreover, neither pre-treatment of the hollow tube-forming material nor matrix exposure to other reagents, such as aldehydes or pluronic acid, are needed. We describe a method for effective seeding and full coverage of all inner surfaces of vascular lumen on-chip with primary human lung microvascular endothelial cells (hLMVEnCs), along with an optimized protocol for co-culture with primary human lung fibroblasts (hLFs). We recreated physiologically relevant homeostatic ECM rheology using natural (instead of synthetic) polymers, with an ability to emulate diseased (e.g., fibrotic lung) matrix stiffness. We chose the lung as a representative organ; however, our approach can potentially be applied to any other organ of interest.

MATERIALS AND METHODS

SIMULATION OF FLUID FLOWS INSIDE VASCULAR CHANNELS ON-CHIP

We simulate the flow patterns inside a vascular channel using the Flow Simulation module of SolidWorks 2021 (SolidWorks Corp., Dassault Systems, Waltham, MA, USA). The simulated device consists of a vascular channel connected to an inlet and an outlet (Fig. 1 a). The fluid flow at the inlet is fully developed with a defined flow rate, while the pressure at the outlet’s outer end is set at atmospheric pressure simulating the actual conditions of vascular on-chip devices during operation. The wall shear stress $\tau$ and the flow rate of fluid in the channel $Q$ are related using

$$\tau =32\frac{\mu Q}{\pi d^3}$$ (1)

where $\mu$ is the viscosity of the fluid, and $d$ is the diameter of the channel assuming the channel is perfectly circular. In our simulation, the flow rate $Q$ is varied from 4.2 $\mu \mathrm{L}$ sec⁻¹ to 210 $\mu \mathrm{L}$ sec⁻¹ to achieve wall shear stress in the range of 0.2 dynes cm⁻² to 10 dynes cm⁻², which is the typical range of shear stress in various blood vessels including arteries, capillaries, and venules in vivo^[6] and in most in vitro studies^{[6b, 7]}.

Figure 1. Simulation of Fluid Flow in Blood Vessels On-Chips.

(a) Illustrations of the design of the chip used in our simulation studies. The channel has a length of 20 mm and a diameter of 650 μm; the diameter of the inlet port was 850 μm. (b) A representative colormap showing the distribution of flow velocity inside the simulated blood vessel when a flow rate of 42 μL sec⁻¹ was applied at the chip inlet. Depending on the fluid velocity, the flow was categorized into ‘developing flow’ or ‘fully developed flow’. (c) Plot showing the dependence of the center velocity (${\mathrm{V}}_{\mathrm{c}}$) versus the distance from inlet for various studied shear stress values ranging from 0.2 dynes cm⁻² to 10 dynes cm⁻²; this is the typical range of wall shear stress in in vivo^{[6, 15]} and in most of in vitro studies^{[3p, 7]}. The dashed line separating the developing regime (shaded) and fully developed regime (blank) was qualitatively drawn through the points where the center velocity approaches 95% of its steady state velocity in each case. (d) Colormaps showing the spatial distributions of velocity over cross-sections of the channels for three representative cross-sections: circular, square, and rectangular. Velocity near the channel’s wall was uniform in the case of the circular channel while it varied considerably (i.e., maximum at the midpoint of each side and minimum at every corner) for square and rectangular channels. (e) Plot showing the spatial distribution of the simulated wall shear stress over the perimeter of the channel for the three studied cross-sections. Wall shear stress was constant over the surface of a circular channel, while it significantly varied spatially when the square or rectangular channel was used. In all (b), (d), and (e), the results shown are for the representative case of 2 dynes cm⁻².

CELL CULTURE

Human lung fibroblasts (hLFs, Lonza, Cat. #CC-2512, Lot #19TL149590 [healthy, 70 years, male, Black]) were cultured in fibroblast growth media (FGM-2, Lonza, Cat. #CC-3132), hereafter referred to as Fibro Medium in a 75 cm² flask at 37°C and 5% CO2 in a humidity-regulated incubator. Human lung fibroblasts were not used beyond the passage 7 to ensure the phenotypes and functions of the cells^[3t]. On the day of the experiment, hLFs were dissolved into a single cell suspension using 0.05% trypsin-EDTA (Gibco, Cat. #25300054). Human lung microvascular endothelial cells (hLMVEnCs, PromoCell, Cat. #C-12281, Lot #480Z005 [healthy, 52 years, female, Caucasian]) were expanded in Endothelial cell growth medium 2 (MV2, PromoCell, Cat. #C-22121), hereon referred to as endothelial cell culture medium (EnC Medium). HLMVEnCs were not used beyond passage 5 to maintain good phenotypes and functions of the cells^[3t]. For harvesting, hLMVEnCs were dissolved using the detach kit consisting of 0.04% trypsin and 0.03% EDTA (PromoCell, Cat. #C-41220) according to the manufacturer’s instructions.

PREPARATION OF EXTRACELLULAR MATRIX HYDROGELS AND THEIR CASTING ON-CHIP

The representative ECM formulation in our experiments, adapted from the one used in previous work^[3t], consists of 7.5% gelatin, 15 mg mL⁻¹ fibrinogen, 2.5 mM calcium chloride, 1% transglutaminase (TG), and 4 U mL⁻¹ thrombin. To prepare the hydrogel, we first mixed the gelatin, fibrinogen, calcium chloride, TG, and phosphate buffered saline (PBS) together at the desired concentrations, and pre-incubated the mixture at 37°C for 45 mins to increase transparency. The PBS can be replaced by the fibroblast culture medium (Fibro Medium) to support cell viability in the case that stromal cells were embedded in the hydrogel. After preincubation, thrombin, and the fibroblast suspension (when fibroblasts are embedded in the ECM), were mixed into the solution and casted within the chip quickly (in minutes) to avoid pipette clogging by the hydrogel.

FABRICATION AND CELLULARIZATION OF THE FINAL OPTIMIZED ECM-MV CHIP

ECM MV-Chip fabrication was performed as following. First, we fabricated the PDMS top and bottom slabs using the standard soft lithography method^[8]. The chip halves were next bonded together using plasma-activated bonding technique. We then inserted two 18G stainless steel needles, with inner and outer diameters of 850 μm and 1.1 mm, respectively, into opposite sides of the PDMS chips to form inlet and outlet of the vascular channel (Fig. 2a).

Figure 2. Design and Construction of Microfluidic Long ECM-embedded Vascular Channel On-Chip.

(a) Schematic showing the design of our ECM MV-Chip, having the same dimensions as the simulated model in Fig. 1a. The enlarged figure shows the cross-section of a fully cellularized chip. Endothelial cells are expected to form a packed monolayer on the inner surfaces of the circular microchannel while fibroblasts would be embedded within the ECM, particularly in regions surrounding the vascular channel. (b) Schematics showing the main steps of our fabrication method: Molding, Casting, Polymerization, and Stabilization. (c) Plot showing the measured Young’s modulus of different ECM formulations used in our experiment. By varying either the gelatin’s processing temperature from 70°C to 90°C or the TG’s concentration from 1% wt. to 3% wt., we were able to vary the ECM’s stiffness from ~1 kPa to ~10 kPa, covering the typical range of various tissue’s stiffness. Statistical analysis was determined by unpaired two-tailed Student’s t-test at p-values: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***) (n = 4 hydrogel replicates per condition; 4 independent studies). (d) Top panel: An image combined from multiple phase-contrast snapshots showing a representative ECM MV-Chip after fabrication using the method in (b) using the ECM formulation of 70°C, 1% TG, 3.17 kPa stiffness. The channel has diameter of 650 μm and length of 40 mm. Bottom panel: left image: schematic showing the cross-section of the ECM MV-Chip’s design; middle and right images: bright-field snapshots showing two representative cross-sections of the fabricated microvascular channel after removing the glass capillary tube. The results show that the channel has smooth inner surface and uniform cross-sections along its entire length. Scale bars: 500 μm.

Before hydrogel casting, we treated the hydrogel chamber with 1 mg mL⁻¹ polydopamine (PDA) for 24 hrs to increase the adhesion of PDMS to hydrogel upon casting^[9]. A silicone tubing with an outer diameter of 650 μm with two stainless steel needles at its two ends was used for vascular channel formation during the gel casting (Fig. 2b, Molding). Following insertion of the silicone tubing into the chip through the vascular channel ports, we cast the fibroblast-laden (1 x 10⁶ cells mL⁻¹) ECM hydrogel solution into the chip through the casting ports connected to the ECM chamber (Fig. 2b, Molding and Casting). The casting was done with great care to prevent any air bubbles from being trapped inside the hydrogel. The chip was then incubated at 37°C for 1.5 hrs to allow the ECM to polymerize (Fig. 2b, Polymerization). Withdrawal of the silicone tubing after polymerization period was done in two steps: (1) one of the stainless-steel needles was removed from the silicone tubing from one end (Fig. 2b, blue arrow); and then (2) the silicone tubing was slowly pulled out of the chip from the other end (Fig. 2b, red arrow). We then supplied Fibro Medium to the channel and gently inserted a clean, sterilized glass capillary tube of 550 μm in diameter into the channel through the inlet while slowly rotating the glass capillary tube. The chip was incubated at 37°C overnight in a 5% CO₂ incubator while soaked in Fibro Medium (Fig. 2b, Stabilization). Next day, the glass capillary tube was gently removed from the chip and the chip was ready for vascularization. The cellularization process was done using our ‘four-stage seeding’ protocol as following. Prior to seeding, the chip was connected to a reservoir of Fibro Medium and kept at 37°C, 5% CO₂. In the first round of seeding, hLMVEnCs were harvested and resuspended in EnC Medium at density of 15 x 10⁶ cells mL⁻¹. We then pipetted 40 μl of the cell suspension into each chip via its inlet. The chips were then incubated inside a 37°C incubator for 30 mins. While waiting, fresh hLMVEnCs were harvested from another flask and resuspended in the EnC Medium at the same density (15 x 10⁶ cells mL⁻¹). In the second round of seeding, the chips were rotated 180 degrees before another 40 μl of the freshly harvested cell suspension was pipetted into each chip. The chips were again incubated for 30 mins at 37°C (while at 180-degree position). The third and four rounds of seeding were done similarly with rotation angles of 90 and 180 degrees, respectively, for cells to adhere to the two sides of the channels. After the four-stage seeding, the channels were flushed with fresh Fibro: EnC Medium supplemented with aprotinin of concentration 20 μg mL⁻¹ and incubated another two hours at 37°C. The chips were then connected to a peristaltic pump to induce a flow of 1 μL min⁻¹ until an endothelial monolayer was observed on-chip, typically after 24 hours.

ON-CHIP IMMUNOFLUORESCENT STAINING

Immunofluorescent staining and confocal microscopy were used to characterize the ECM MV-Chip. Cellularized chips were first washed with PBS (Gibco, Cat. #14190094) by perfusion for several minutes. Next, endothelial cells on-chip were fixed by perfusing 10% buffered formalin through the channel for 15 mins at 37°C. The ECM block then was cut out from the PDMS chip and bathed in 10% buffered formalin for another 2 hrs at room temperature (RT) for fixation of the embedded fibroblasts. The ECM block was subsequently washed in PBS overnight at 4°C. The next day, the ECM block was blocked for at least 2 hrs at RT using a blocking buffer containing 1% wt. BSA and 5% wt. FBS in PBS. The ECM block was then incubated for 24 hrs in blocking solution with the pre-conjugated primary antibody (Alexa Fluor 488-CD31, BioLegend, Cat. #102514). Washing of unbound primary antibody was done in PBS overnight at 4°C. Samples were counter-stained with Phalloidin (Abcam, Cat. #ab176752) and TO-PRO 3 Iodide (Invitrogen, Cat. #T3605) for 2 hrs then washed with PBS for 1 day before imaging.

ON-CHIP CELL VIABILITY ASSAY

The on-chip cell viability assay as shown in Fig. 4b was done as follows: Calcein AM (Invitrogen, Cat. #C1430) and propidium iodide (PI, BioLegend, Cat. #421301) were diluted in the culture media to the target concentrations of 1 μM. Before performing the live/dead staining, unbound cells were removed from the chip by perfusion of culture media through at a flow rate of 10 μL min⁻¹. The live/dead staining solution was applied by pipetting through the channel inlet. The chip was then incubated at 37°C for 15–30 mins before fresh media was perfused through the channel for a 5–10 mins wash. Processed samples were then imaged using confocal microscopy within 1 hr. after staining.

Figure 4. Optimized Multicellularization and Matrix Casting Process for Reconstitution of the 3D ECM-embedded Microvessel-On-a-Chip.

(a) Schematic showing the general steps of the protocol for co-culture of fibroblasts and endothelial cells on-chip. (b) Representative fluorescent image showing the viability via LIVE/DEAD staining of fibroblasts on Day 1 before the EnC seeding. The dashed lines indicate the perimeter of the vascular channel on-chip. (c, d) Representative fluorescent images showing multicellular components of our fully vascularized chip on Day 2. Endothelial monolayer (green) formed and fully covered the channel perimeter. Fibroblasts (blue) uniformly distributed in the ECM surrounding the channel. The images were constructed by superimposing multiple confocal scans at different positions across the channel’s diameter. (e) Cross-section of the vascular channel. The EnC monolayer appears as a thin green-blue layer fully covering the channel surface. In (c-e) green, blue, and red signals are respectively for CD31+, F-actin, and nuclei signals. Endothelial cells expressing both CD31 and F-actin appear in green and blue, while fibroblasts are only in blue. (f) A series of representative snapshots showing the diffusional dynamic of FITC-Dextran into the Fibro-ECM for the case of +Endo +Fibro during our experimental duration up to 30 mins. The vascular channel appears in green while the surrounding Fibro-ECM is in black. In (b-f), the scale bars represent 250 μm. (g) A plot showing the normalized fluorescent signal intensity (FSI) inside the ECM surrounding the microchannel versus the distance to microchannel perimeter for the experiment in (f). FSI values are the average of 4 independent chips and the error bars are SEM. The FSI appears to be stable in the Fibro-ECM evident by the overlap of the datasets at different times indicating that FITC-Dextran does not significantly diffuse into the ECM. (h) A plot showing the diffusion coefficients calculated from the FSI inside the ECM surrounding the vascular channel using Eq. (2). The column height shows the mean values; each data point is for an independent chip (n = 3 – 4 independent chips per condition). Statistical analysis was determined by unpaired two-tailed Student’s t-test at p-values: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***), and p < 0.0001 (****). In all figures from (b) to (h), the hydrogel formulation was 70°C, 1% TG having stiffness of 3.17 kPa.

IMAGE ACQUISITION AND ANALYSIS

In all our experiments, imaging was done with a confocal microscope (Leica DMi8) with objectives from 5X to 10X and laser sources 405 nm, 488 nm, 552 nm, and 638 nm depending on the fluorescent excitation wavelengths of the samples. The z-stack images obtained from the confocal microscope were then pre-processed by ImageJ (version 1.52) to enhance the image quality. Analysis of the desired parameters such as cell covered area, fibroblast density, or fluorescent intensity was done using customized MATLAB scripts, version 2019b.

RHEOLOGICAL CHARACTERIZATION

Gel’s rheology was measured at 37°C using a controlled stress rheometer (DHR-2, TA Instrument) with a 20-mm diameter flat-plate geometry and frequency sweep mode with a fixed strain (5%)^[10]. The measurements showed a typical frequency dependence of the shear storage and loss moduli (Supp. Fig. 1). The Young’s modulus $E$ of the hydrogel (e.g., in Fig. 2c) was calculated using $E=2(1+\nu )G^{\prime }$, where $\nu =0.5$ is the Poisson’s ratio of the gel and $G^{\prime }$ the shear storage modulus at frequency of 1 Hz ^[3t].

CHARACTERIZATION OF ADHESION OF ENDOTHELIAL CELLS ON FIBROBLASTS-ECM

In this experiment, we examined the adhesion of endothelial cells on Fibro-ECM varying the ratio of Fibro medium to Endo Medium in the culture media and the Endothelial Seeding Time defined as the duration between the day the Fibro-ECM was prepared, and the day endothelial cells were seeded on it. To do this, we performed two sets of experiments. Firstly, to examine the effect of the Fibro Medium, we varied the Endo Medium ratio (Fibro: EnC Medium ratio as 20:80, 40:60, 50:50, 60:40, and 80:20), while keeping the Endothelial Seeding Time fixed at Day 1. Secondly, to examine the effect of the Endothelial Seeding Time, we varied the seeding time from Day 0 to Day 4, while keeping a Fibro: EnC Medium ratio of 1:2.

The experimental procedure was as follows. Before the experiment, both hLFs and hLMVEnCs were expanded in appropriate media and culture conditions (see the Cell Culture Section). First, a hydrogel solution without thrombin was prepared having the PBS portion in the hydrogel mixture replaced by Fibro Medium and preincubated at 37°C as described above. hLFs were dissolved into a single cell suspension and gently mixed with the preincubated hydrogel solution at a density of 1 x 10⁶ cells mL⁻¹. Thrombin was added last to the solution and mixed thoroughly. The Fibro-ECM hydrogel was then casted onto a 24-well culture plate at 200 $\mu \mathrm{L}$ per well (i.e., 200,000 fibroblasts). The wells were polymerized at 37°C in a humidity-regulated incubator for 1.5 hrs After the curing time, 500 mL Fibro Medium containing 20 μg mL⁻¹ aprotinin (Sigma, Cat. #9087-70-1) was added to each well and further incubated until the seeding time of the endothelial cells.

On the day of endothelial seeding, hLMVEnCs were harvested and resuspended at a density of 20,000 cells mL⁻¹ in a culture medium of Fibro Medium and EnC Medium at ratios varying from 20:80 to 80:20 in the first set of the experiment or fixed at 1:2 in the second one. Aprotinin (20 μg mL⁻¹) was added to the media to prevent hydrogel degradation. We then replaced the media in the 24-well Fibro-ECM hydrogel well plate (prepared above) with the endothelial cell suspension 500 μL per well to obtain EnC seeding density of 5,000 cells cm⁻². The Fibro-ECM and endothelial cells were kept in the culture conditions for 24 hrs before evaluation. For evaluation, we stained the cells with CD31 antibody (BioLegend, Cat. #102514), F-actin biomarker (Phalloidin, Abcam, Cat. #ab176752), and nuclei biomarker (TO-PRO 3 Iodide, Invitrogen, Cat. #T3605), then imaged the cells using a confocal microscope (Leica, DMi8).

Endothelial and fibroblast cell densities were assessed by confocal microscopy using MATLAB. Briefly, the endothelial cell mean surface area was determined by measuring the area of every EnC clump in the CD31 fluorescent images. We then counted the number of EnCs in the clumps from the corresponding nuclei fluorescent images. If there were multiple EnCs in one clump, each cell took the average value of the area of that clump. To determine the fibroblast density in the Fibro-ECM, we first used a confocal microscope to scan a volume of the gel, i.e., from 50 μm to 90 μm, beneath the gel surface with a z-stack step of 2.5 μm. We then counted the total number of fibroblasts in this volume. The fibroblast density was determined by dividing the number of fibroblasts by the scanned volume.

DIFFUSIONAL PERMEABILITY TEST OF MICROVASCULATURE ON-CHIP

In this experiment, a camera-mounted on a wide-field microscope captured fluorescent images at 488 nm of live cells on the chip every 5 secs. Initially, a 25 μg mL⁻¹ FITC-conjugated 70-kDa dextran solution (FITC-Dextran, Sigma, Cat. #46945) in culture media was perfused through the live ECM MV-Chip at a rate of 15 μl min⁻¹ for 4 mins to replace the culture media inside the channel. Then, the flow rate of dextran-media solution was reduced to 1 μl min⁻¹ and the chip was imaged. The experiment was performed using a microscope incubator (OkoLab) at 37°C, 5% CO₂.

The fluorescent images were analyzed by a customized MATLAB script to calculate the fluorescence intensity at every point surrounding the channel over time. The diffusion coefficient P, a common parameter used to characterize the diffusional permeability of endothelial blood vessels, was calculated using the following equation^{[3t, 11]}

$$P=0.25\frac{I(t)-I(0)}{I(0)-Ib}\frac{d}{t},$$ (2)

where $\mathrm{l}(t)$ and $I(0)$ are respectively the fluorescent intensity at time $t$ and time $0$, $Ib$ is the background intensity, i.e., the fluorescent intensity of the image before introducing FITC-Dextran to the channel, and $d$ the diameter of the channel.

We performed the experiment for three different chip conditions (from 3 to 4 technical replicates per condition): (1) chips with fibroblasts in the ECM and EnC barrier (i.e., +Endo +Fibro); (2) chips with fibroblasts in the ECM and without EnC barrier (i.e., -Endo +Fibro); and (3) chips without either fibroblasts or EnCs (i.e., -Endo -Fibro). The analyzed results assessed the roles of fibroblasts and EnCs in preventing the diffusional permeability of FITC-Dextran to the ECM.

STATISTICAL ANALYSIS

Statistical data analysis was performed using MATLAB software program (version 2019b). Statistical significance was determined using unpaired two-tailed Student’s t-test (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). Sample size (n) and independent studies are as follows: Fig. 3c: n = 33–109 data points per condition, 3 independent studies; Fig. 3d: n = 3 data points per condition, 3 independent studies; Fig. 3e: 22–104 data points per condition, 3 independent studies; Fig. 4g: n = 4 replicates per condition, 1 independent study; and Fig. 4k: n = 3 – 4 data point per condition, 1 independent study. In all figures, the error bar, if presented, represents mean and standard error of the mean (SEM), i.e., mean ± SEM.

Figure 3. Characterization of Endothelial Cells Adhesion on to Fibroblast-laden ECM and Fibroblasts Abundance within ECM.

(a) Confocal immunostained micrographs showing endothelial cells spreading on Fibro-ECM surface when cultured at different Fibro: EnC Medium ratios ranging from 20:80 to 80:20 and fixed seeding time (Day 1). Endothelial (CD31+) cells appear in green while nuclei are shown in red. (b) Confocal images of the fibroblasts (F-actin+) are shown in blue at ~ 50 μm beneath the endothelial layer. (c) Images constructed from z-stacked confocal microscopy scans with steps of 2.5 μm showing the corresponding cross-sections of the Fibro-ECM layer of the experiment in (a). Endothelial cells (green signal) appear as a thin layer on the surface of the ECM while fibroblasts (blue signal) are embedded inside the ECM. In all (a), (b), and (c), from left to right, Fibro Medium percentage increases while EnC Medium percentage reduces; the scale bar represents 500 μm; the hydrogel formulation was 70°C, 1% TG having stiffness 3.17 kPa. (d) Plot showing the experimental data of endothelial cell adhesion – defined by mean covered surface area per EnC, when varying the Fibro: EnC Medium ratio. Each data point represents the mean cell surface area for a single EnC (n = 33–109 data points per condition; 3 independent studies). (e) Plot showing fibroblast abundance in the ECM when varying the Fibro: EnC Medium ratio. Each data point shows the total fibroblast count of one experimental replicate (n = 3 data points per condition; 3 independent studies). From (a) to (e), the seeding time was kept fixed at Day 1. (f) Plot showing the EnC adhesion when studying different endothelial seeding times. The Fibro: EnC Medium ratio in this experiment was 1:2. Each data point represents the mean cell surface area for a single EnC (n = 22–104 data points per condition; 3 independent studies). In (d), (e), and (f), the error bars show the SEM and statistical analyses were determined by unpaired two-tailed Student’s t-test at p-values: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***), and p < 0.0001 (****).

RESULTS

BOTH LENGTH AND CROSS-SECTIONAL GEOMETRY OF VASCULAR CHANNEL ARE IMPORTANT IN DETERMINING THE CHANNEL WALL SHEAR STRESS

To determine the importance of blood channel cross-sectional geometry and length, we simulated flow within a vascular microchannel with inlet and outlet conjunction regions of 850 μm diameter and a central cylindrical channel with diameter of 650 μm (Fig. 1a). The flow rate at the inlet was varied from 4.2 μL sec⁻¹ to 210 μL sec⁻¹ to achieve wall shear stress values of 0.2 dynes cm⁻² to 10 dynes cm⁻², which is similar to the physiological range of shear stress in human blood vessels ^[6]. A representative analysis is shown in Fig. 1b for a wall shear stress of 2 dynes cm⁻² and indicates that the flow velocity increases when entering the vascular channel due to the reduction in channel’s cross-sectional area compared to the inlet. The flow patterns inside the vascular channel could be divided into two major categories: (1) ‘developing flow’ – where the velocity is in transient state, and (2) ‘fully developed flow’ – where the velocity is at steady state. However, when the flows are developing, shear stress on the channel surface also changes according to the velocity, underscoring the need to ensure that the flow inside the channel is fully developed for applications requiring well-defined wall vascular shear stress values across the length of the blood lumen.

To illustrate the significance of flow considerations when designing a vascular microchannel Organ-on-a-Chip, we modeled velocity at the center of the channel (V_c) against distance from the channel inlet. We performed this analysis for multiple shear stress values (0.5–10 dynes cm⁻²). This approach allowed us to identify the minimum length of the channel needed to achieve fully developed flow. As shown in Fig. 1c, for any given channel geometry, (i) there is a minimum distance (shaded area) from the inlet that the fluid must travel to achieve expected wall shear stress, and (ii) this distance goes up considerably for larger shear values. For instance, for emulating post-capillary venule shear stress of 10 dynes cm^{−2 [6]}, we would require a channel length of > 8 mm so that the flow inside the vascular channel becomes fully developed. In other words, if one aims to mimic such a vascular shear stress but has a microfluidic channel shorter than 8 mm, it is unlikely they will be able to achieve the expected shear stress. Therefore, this observation reveals that when designing an Organs-on-Chips system containing a blood channel, the length of the vascular channel must be accurately calculated based on desired shear value, correcting for developing flow effect.

We went further and examined the effect of channel’s cross-sectional geometry on the spatial distribution of the wall shear stress (Fig. 1d, e). For this, we simulated vascular flow on three representative vascular microchannels that are circular, square, or rectangular on cross-section – the latter two are widely used designs^{[3j, 3q, 12]} – which have the same cross-sectional area (0.33 mm²), and all experience an input flow rate of 42 μL sec⁻¹ with basic tissue culture medium corresponding to wall shear stress of 2 dynes cm⁻². Reconstructing the colormap of the fully developed flow velocity distribution over the channels’ cross-sections, we observed that in all cases the flow had its highest velocity at the center of the channel and lowest at the channel wall, which is due to friction (Fig. 1d). However, we observed distinct differences across the three types of channel designs. In a device with circular cross-section, the velocity is distributed uniformly throughout the inner surface of the channel wall. On the other hand, in channels with square and rectangular cross-sections, the fluid velocity was highest at the midpoints of each side and lowest at the corners of the channels. Such non-uniform distribution of the wall velocity indicates that the wall shear stress, experienced by any endothelial cells coating the vascular wall, would also be non-uniform.

Next, we plotted the simulated wall shear stress $\tau$ at different locations on the channel surface represented by the azimuthal angle $\phi$ varying from 0 to $\pi$ (Fig. 1e). The wall shear stress of the circular channel was a constant – i.e., $\tau \approx 1.5\text{Pa}$ (15 dynes cm⁻²) across the channel surface. However, in channels with square and rectangular cross-sections, the wall shear stresses fluctuated periodically with lowest values at the channel corners and highest shear at the midpoints of each side, similar to flow velocity profile at the channel wall observed in Fig. 1d. The shear stress difference between the maximum and minimum values in these channels were as high as 2 Pa (1.5-fold the average wall shear stress), which reveals a substantial variation in the wall shear stress when utilizing a cross-sectionally square- or rectangular-shaped microchannel device. We note that there was a difference between the absolute values of the calculated wall shear stress using Eq. (1) and the actual simulated shear stress. This is because the calculated shear stress used the average velocity in the channel while the simulated shear stress used the simulated velocity near the channel wall. Nevertheless, the distribution patterns of the wall shear stress across the channel surfaces remain unchanged for the three types of channels we studied, implying a clear advantage for using circular microchannels in modeling blood vessels to ensure uniform distribution of shear by all endothelial cells.

RECONSTITUTION OF LONG, CROSS-SECTIONALLY CIRCULAR, EXTRACELLULAR MATRIX-EMBEDDED VASCULAR MICROCHANNELS ON-CHIP

As both the length and the cross-sectional geometry of the channel are important determinants for studying vascular wall shear stress (Fig. 1), we developed a simple, reliable, and highly reproducible method to construct mesoscale long (in centimeters), cross-sectionally rounded, and 3D microvasculature within a natural extracellular matrix channel on-chip (Fig. 2). This protocol yields an ECM-embedded Microvasculature-on-a-Chip (hereon referred to as ‘ECM MV-Chip’) that allows accurate recapitulation of pathophysiologically relevant mechanobiology within our blood vessels for Organs-on-Chips applications. The schematic of the ECM MV-Chip design is illustrated in Fig. 2a, top panel. The chip is supported by an outer PDMS case that is fabricated by standard soft lithography methods^[8]. It contains an enclosed rectangular chamber with dimensions of 20–40 mm × 4 mm × 2 mm (L×W×H), inside which the ECM is injected and cast through gel casting ports, and through the hydrogel block a vascular channel is formed at its center. The device dimensions are user-defined. Here we present a prototype developed in our laboratory, to illustrate methodology and model validation. The channel is connected to two needles with inner diameters of 850 μm and outer diameters of 1.1 mm, placed horizontally at two ends of the chips and functioning as the inlet and outlet of the vascular channel. Following successful chip fabrication, gel casting, and cellularization, one would have a biodevice (Fig. 2a, bottom panel) in which stromal cells such as fibroblasts are embedded within a natural ECM in the perivascular region – mimicking the tunica media and externa layers^[13] in vasculature or the interstitium between epithelia and endothelia in a given organ. At the center of the ECM block lies a uniformly circular vascular microchannel lined with barrier-forming endothelial cells to emulate tunica intima layer of blood vessels in human bodies.

To fabricate the vascular channel on-chip, we adapted the hydrogel casting method^{[3n, 3r]} with several major improvements. First, we treated the hydrogel chamber with 1 mg mL⁻¹ polydopamine (PDA) for 24 hrs to increase the adhesion of PDMS to hydrogel upon casting^[9]. A silicone tubing with outer diameter of 650 μm with two stainless steel needles at its two ends was used as replacement for a PDMS rod for vascular channel formation during the gel casting (Fig. 2b, Molding). Following insertion of the silicone tubing into the chip through the vascular channel ports, we cast the gelatin-encapsulating fibrin gel into the chip through the casting port connected to the ECM chamber (Fig. 2b, Molding and Casting). The casting was done with great care to prevent any air bubbles from being trapped inside the hydrogel. The chip was then incubated at 37°C for 1.5 hrs to allow the ECM to polymerize (Fig. 2b, Polymerization), followed by withdrawal of the silicone tubing in two steps: (1) one of the stainless-steel needles was removed from the silicone tubing from one end (Fig. 2b, blue arrow); and then (2) the silicone tubing was slowly pulled out of the chip from the other end (Fig. 2b, red arrow). As the silicone tubing is soft, this method of withdrawal will prevent damaging (e.g., collapse or pull-back) the vascular channel. We then inserted a sterilized glass capillary tube of 550 μm diameter into the channel and dipped the chip into PBS or Fibro Medium (in the case fibroblasts were embedded in the ECM) and incubated the chip overnight (Fig. 2b, Stabilization). This was necessary to further stabilize the channel geometry and increase the success rate of our method. Without this stabilization step, we found that due to the high swelling of the ECM, the channel would be more likely to deform or even collapse upon supplying culture medium into the lumen post-casting. During the stabilization step, replacing the silicone tubing with a clean glass capillary tube with smaller diameter is crucial; it not only minimizes any damage to the inner channel surfaces following the stabilization period, but also allows Fibro Medium to diffuse into the ECM through the gap between the capillary tube and the channel perimeter to support survival of the embedded fibroblasts. On the next day, the glass capillary tube was gently removed from the chip and the chip was ready for vascularization.

The ability to control and replicate pathophysiological matrix rheology is an important feature that is commonly required in vitro. In our studies, we adjusted the hydrogel stiffness by altering either the processing temperature of the gelatin or the concentration of transglutaminase (TG) in the gel formulation. By reducing the gelatin processing temperature from 90°C to 70°C and enhancing the TG concentration from 1% wt. to 3% wt., we were able to successfully vary the stiffness of the ECM hydrogel from ~1 kPa to ~10 kPa (Fig. 2c). We chose these stiffness values to emulate a wide range of tissue rheology from normal non-fibrotic to fibrotic in the lung^[14], for instance. In the subsequent experiments, as we chose lung microvasculature for an example, we used the hydrogel formulation of 70°C gelatin, 1% wt. TG, having a stiffness of 3.17 kPa, which reproduces the typical stiffness of normal lung tissue ranging from 0.7 kPa to 6 kPa^[14]. We demonstrated our method by fabricating channels with a diameter of 650 μm and length of 40 mm on-chip. After removing the glass capillary tube, the channel appeared to be uniform, smooth, and undamaged throughout its whole length (Fig. 2d). We also found that the cross-sectional area of the fabricated channel was slightly bigger than the design, a typical characteristic of casting technique (Fig. 2d, bottom panels). We note that similar results with high consistency in channel geometries were obtained in multiple independent experiments with other hydrogel formulations, confirming the success and the reproducibility of our method in fabricating long, ECM-embedded, rounded microvascular channels on-chip (Fig. 2d).

MULTICELLULAR CO-CULTURE OPTIMIZATION: HIGH PERCENTAGE OF ENDOTHELIAL MEDIA AND SHORT ENDOTHELIAL SEEDING TIME IMPROVE ADHESION OF ENDOTHELIAL CELLS TO FIBROBLAST-ECM

The ECM in all tissues and organs of our body not only provides cellular scaffolding for cellular constituents such as epithelium or endothelium that coat its surfaces, but also houses stromal/interstitial cells within the matrix itself. Thus, we set out to cellularize the ECM MV-Chip so that EnCs would line the inner surface of the vascular microchannel and form a tight barrier, and fibroblasts – as a representative stromal cell type and relevant to many pathologies, would be embedded in the gel. Successful adhesion of the endothelial cells to the fibroblast-laden ECM (hereafter referred to as Fibro-ECM) is crucial for robust and reliable execution of the vascularization process on-chip. Therefore, choosing lung as the organ of interest, we utilized primary human-derived normal (non-diseased) lung fibroblasts (hLFs) and lung microvascular endothelial cells (hLMVEnCs) to perform co-culture optimization. Two key parameters that we focused on were (i) the ratio of hLF: EnC growth culture media (Fibro Medium: EnC Medium), and (ii) the EnCs seeding time, which is defined as the duration of time following fibroblast-laden gel casting prior to seeding the EnCs on the Fibro-ECM surface.

Varying Fibro: EnC Medium ratio from 20:80 to 80:20 (Fig. 3) and performing confocal microscopy imaging on the immunostained cells, we observed that, as expected, lower EnC Medium % is qualitatively associated with lower EnC spread and adhesion to the Fibro-ECM, which was evident by a reduction in the CD31+ cell-covered areas on the Fibro-ECM surface (Fig. 3a, right panels). We also co-immunostained the cells in the gel with F-actin and nuclear dye and imaged them at the perivascular region (~ 50 μm below the endothelial cell layer) (Fig. 3b left panels, c) to mark the fibroblasts. We found lower Fibro Medium % is associated with lower F-actin and nuclear dye staining in the gel. To quantitatively characterize the abundance of endothelial cells and fibroblasts in these culture conditions, we analyzed the EnC-covered areas on the surface of Fibro-ECM and the density, i.e., abundance, of the fibroblasts within a given volume of the ECM, respectively (Fig. 3d, e). The results confirmed our qualitative observations in Fig. 3a, b. EnC covered areas were significantly reduced with the EnC Medium percentage. For instance, the mean covered area of EnCs decreased significantly from 2,370 $\mu m^2∕cell$ to 1,740 $\mu m^2∕cell$ and 1,250 $\mu m^2∕cell$ as the EnC Medium ratio dropped from 80% to 50% and ≤40%, respectively (Fig. 3d). We studied the 50% ratio for EnC Medium since it is a commonly applied approach (for 1:1 mixing of culture media for cellular co-cultures) by many investigators^{[3t, 7b]}. Additionally, we found that fibroblast density increases with the percentage of Fibro Medium (e.g., from 1,540 fibroblasts mm⁻³ to 2,823 fibroblasts mm⁻³) when Fibro Medium percentage changes from 20% to 80% (Fig. 3e). However, the increase in fibroblast density was not statistically significant comparing Fibro Medium percentage between from 20% to 50% and 60%. Interestingly, the lowest fibroblast density (1045 fibroblasts mm⁻³) was for 40% Fibro Medium.

In addition to optimization of culture media ratios, we also quantitatively characterized the effect of the EnC seeding time on the EnC adhesion by analyzing the EnC covered areas on the Fibro-ECM surface when varying the seeding time from 0 to 4 days post-Fibro-ECM gel casting (Fig. 3f). We observed that increasing the seeding time remarkably reduces the EnC covered areas which implies a drop in EnC adhesion. Particularly, EnC covered areas significantly dropped from 4,300 $\mu m^2∕cell$ at Day 0 to 3,100 $\mu m^2∕cell$ at Day 2 and subsequently to 2,700 $\mu m^2∕cell$ at Day 4. In other words, delaying the EnC seeding can negatively and significantly impact the endothelial cell adhesion (for vascularization).

ON-CHIP CO-CULTURE OF FIBROBLASTS AND MICROVASCULAR ENDOTHELIAL CELLS

The final, optimized protocol for generating ECM MV-Chip is illustrated in Fig. 4a. In preparation, the hLFs and hLMVEnCs were cultured (and expanded, if desired) in appropriate culture conditions (see Materials and Methods). On Day 0 (Fig. 4a, left panel), a Fibro-ECM hydrogel solution with density 1 × 10⁶ fibroblasts mL⁻¹ was cast into the hydrogel block of the device. The channel was then stabilized overnight in Fibro Medium (see Materials and Methods). On Day 1 (Fig. 4a, middle panel), we seeded the EnCs into the vascular lumen using our ‘four-stage seeding’ method we optimized for this process. Before EnC seeding, the glass capillary tube was gently removed from the chip, and the chip was connected to a reservoir of Fibro Medium and kept in 37°C, 5% CO₂ while waiting for EnC seeding. The hLMVEnCs were harvested and resuspended in the EnC Medium at density of 15 × 10⁶ cells mL⁻¹. To perform the four-stages seeding, we first pipetted 40 μl of the cell suspension into each chip via its inlet. The chips were quickly incubated in a 37°C incubator for 30 mins for the cells to adhere to the channel surface. While waiting, fresh hLMVEnCs were harvested from another flask and resuspended in the EnC Medium at the same density (15 × 10⁶ cells mL⁻¹) ready for the next seeding. In the second round of seeding, at the end of 30-min incubation, the chips were transferred to a biosafety cabinet and rotated 180 degrees. Another 40 μl of the freshly harvested cell suspension was pipetted into each chip. The second round of cell seeding also flushed out any non-adhered cells from the first seeding and replaced the media in the channel. The chips were then immediately incubated for 30 mins at 37°C (while at 180-degree position (i.e., upside down)) for the cells to adhere to the top surface of the channel. The third and fourth seeding rounds were done with rotation angles of 90 and 180 degrees, respectively, for the cells to attach to the two sides of the channel. After the four rounds of seeding, we gently pipetted fresh culture media containing Fibro: EnC Medium supplemented with freshly added aprotinin (which enables maintaining structural integrity of the gel in culture) and ascorbic acid (that supports EnC barrier formation) through the vascular microchannel to wash away non-adherent cells. The chip was then incubated at 37°C for another two hrs before being connected to a peristaltic pump for 1 μL min⁻¹ flow (a representative flow rate that can be adjusted by the user). After 24 hrs (Fig. 4a, right panel (Days)) under the flow, a monolayer of endothelial cells was formed on-chip for each tested device. The devices were then maintained in Fibro: EnC Medium without vascular endothelial growth factor (VEGF) under 1 μL min⁻¹ flow for subsequent assays.

To confirm the success of our approach, we first performed a viability test on-chip on Day 1 just before seeding the EnCs (see Materials and Methods). We visually observed that almost 100% of fibroblasts were alive with little, if any, dead cell signal (Fig. 4b), indicating that the stabilization period did not impact the viability of fibroblasts in the ECM. We then performed on-chip immunostaining and confocal microscopy imaging for chips on Day 2. As shown in Fig. 4c-e, and Supp. Fig. 2, the EnC monolayer (CD31+, green signals) entirely covered the vascular microchannel inner surfaces, and the fibroblasts (blue signals) uniformly distributed in the ECM, particularly in the areas surrounding the channel – that is the perivascular region. The 3D z-stack reconstructed confocal microscopic images of the vascular channel (Fig. 4e) revealed that the EnC monolayer was clearly present on the channel cross-section and has a thickness of ~ 30 μm. To confirm the robustness of our ECM MV-Chips, in an independent experiment, we maintained the fully cellularized devices for 96 hrs under 5 μL min⁻¹ flow and observed no visually distinguishable change in channel geometry as well as endothelial morphology (data not shown here).

Next, we went further to examine the EnC barrier function of our fully vascularized ECM MV-Chip by performing a permeability experiment and characterizing the diffusional dynamics of FITC-Dextran through the EnC monolayer (see Materials and Methods). We observed that the fluorescent signal intensity (FSI) of FITC-Dextran in the Fibro-ECM was stable over the experimental period of 30 mins (Fig. 4f), suggesting that the EnC barrier actively prevented diffusion of FITC-Dextran into the extracellular matrix. To confirm this observation, we acquired the FSI across the image (normalized against the signal intensity immediately at the channel perimeter) (Fig. 4g). We found that the normalized FSI in endothelialized Fibro-EMC does not change over the experimental time evident by the overlaps of the datasets of different time points. In contrast, when no EnC monolayer was present, FITC-Dextran significantly diffused into the ECM indicated by the upward shifts of the normalized FSI curves with time (see Supp. Fig. 3a, b). To quantitatively examine the diffusional dynamics of FITC-Dextran into the ECM, we calculated the diffusion coefficient P (see Materials and Methods) of three experimental conditions: (1) + Endo + Fibro (Fig. 4h, left column) – i.e., EMC MV-Chip with both EnCs and fibroblasts seeded, (2) - Endo + Fibro (Fig. 4h, middle column) – i.e., EM MV-Chip lacking EnCs, and (3) - Endo - Fibro (Fig. 4h, right column) – i.e., EM MV-Chip lacking EnCs and fibroblasts. We observed a distinct difference between the diffusion coefficients of the + Endo + Fibro and the other two cases. In particular, the diffusion coefficient was almost zero ($P\approx 1.9\times {10}^{-4}$ mm min⁻¹) for + Endo + Fibro chips while its value was significantly higher for the – Endo + Fibro and – Endo – Fibro chips ($P=(3.3\pm 0.8)\times {10}^{-3}$ mm min⁻¹ and $P=(5.7\pm 2.9)\times {10}^{-3}$ mm min⁻¹, respectively). Comparing the diffusion coefficient between + Endo + Fibro and – Endo – Fibro chips, we found a 20-fold increase in barrier function. Interestingly, fibroblasts alone also contributed to reducing the diffusional permeability of the channel. Comparing between the – Endo – Fibro and – Endo + Fibro chips, the diffusion coefficient drops almost 2 times. However, the change was not statistically significant (Fig. 4g). Nevertheless, this result implies that fibroblasts tend to grow better near the channel’s inner surface and partially form a fibroblast layer covering the channel surface. In summary, the result of the diffusional permeability test demonstrates the role of endothelial barrier in limiting the diffusion of small molecules across the channel and strongly confirms the success of our method for constructing the ECM MV-Chip.

DISCUSSION

Our results provide an important advance for design and development of Extracellular Matrix-embedded Multicellular Blood Microvessel-on-a-Chip, especially for studies on blood vessel mechanobiology. The findings from our simulation analysis on flow within the blood vessels on-chip revealed that both the length of the vascular channel and its cross-sectional geometry are crucial in Organs-on-Chips applications, especially with studies focusing on effects of shear stress on biological responses of tissues and organs. We demonstrated that shorter channels are not capable of stabilizing the fluid velocity inside the channel, which consequently causes nonuniform spatial distributions of the wall shear stress along the channel length. Moreover, we clearly demonstrate that using vascular channels with square or rectangular cross-sections are not appropriate to study the effect of fluid shear stress as the shear stress in such a channel is not uniformly distributed across the channel perimeter.^{[3j, 3q, 12]}.

We also provide a feasible, robust, and reproducible tissue engineering strategy with a high success rate that provides unique advantages compared with existing methods in forming centimeter-long, rounded ECM-embedded vascular lumens. First, we demonstrate that using silicone tubing, as replacement for PDMS rod, glass capillary tube, or metal rod which are commonly described in published literature, will ensure smooth and adherent-free detachment of the tubing from the surrounding ECM upon its withdrawal as hydrogel adhesion to the silicone tubing is lower than that of the PDA-treated PDMS chips. In addition, our approach of introducing a stabilization period allows the hydrogel to equilibrate its swelling process while the glass capillary tube inside the ECM helps maintain the channel's geometry. We found that without the stabilization period, the microvascular channels tend to severely deform or even collapse during endothelialization due to swelling of the hydrogel. As a result, combining silicone tubing and the channel stabilization period allows the formation of long, uniform, smooth, and curved vascular microchannels with a high success rate and reproducibility in both channel’s geometry and ECM properties. This is a crucial factor that contributes to the high success rate of the endothelization process. Notably, we were able to construct vascular channels as long as 40 mm that exhibited uniform cross-section along their length (Fig. 2d) with a 100% success rate. This would be very challenging, if not impossible, to reproduce with alternative microfabrication methods. Another advantage of our fabrication method is that it does not require any pre-treatment of the microvascular channel before cellularization. As a result, the inner surface of the vascular lumen post-casting is not exposed to other unwanted reagents such as aldehydes or pluronic acid (commonly used in some 3D-bioprinting approaches). This is also important to allow efficient endothelial cell attachment downstream as demonstrated by the success of our vascularization on-chip.

The biomaterial design of ECM also contributed to the success of our method. Gelatin-fibrinogen hydrogel has been extensively tested for constructing microvascular networks^{[3t, 7b]} via bioprinting and has shown excellent biocompatibility for endothelial cells and fibroblasts. To adapt this biomaterial mixture for casting microvascular channels in an enclosed chamber with very limited evaporation capacity as in our chips, we modified the hydrogel formulation by using thrombin of concentration 4 U mL⁻¹ (compared to 2 U mL⁻¹in literature ^[3t]). This is important as without the adjustment the curing time for the hydrogel in our chip would have been ~ 3 hrs which is detrimental for survival of the embedded fibroblasts during gel polymerization. Increasing thrombin concentration allowed us to build a rapid and stable gelation process, by reducing the curing time to 1.5 hrs which significantly helps with maintaining high viability of the embedded fibroblasts (Fig. 4b). Furthermore, our hydrogel formulations can be readily and feasibly be modified to adjust their rheology without affecting other properties of the gel. We demonstrated this by changing either the transglutaminase concentration or the processing temperature of gelatin. This allows users to design and fabricate vascular channels for various Organs-on-Chips applications requiring a wide range of ECM stiffnesses.

Our optimization studies on the Fibro: EnC Medium ratio and the EnC seeding time provide useful information for the development of a multicellular co-culture protocol not only for the platform presented here, but also for other related works of ECM-embedded Blood-Vessels-on-Chips. While existing literature^{[3n, 3t]} usually use an intuitive Fibro: EnC Medium ratio of 50:50, our results indicate that a better ratio for the co-culture in primary healthy human lung cells ranges from 20:80 to 40:60 to promote adhesion of endothelial cells to Fibro-ECM. Such media mixture not only will be favorable for the adhesion and spread of the EnCs, compared to the 50:50 media mixture ratio, but also it would not significantly hamper the growth of fibroblasts in the gel. As EnC adhesion is crucial very early during co-culture, a high percentage of EnC Medium in the culture media would substantially enhance the success rate of the on-chip vascularization process. Although these conditions need to be further validated using a larger donor pool, they can be a reasonable starting point. The dependence of EnC adhesion on seeding time was also important to optimize the co-culture protocol. For example, the overnight stabilization period allowed the channel to stabilize and tremendously increased the success rate of our method (to 100%), while minimizing the reduction in EnC adhesion with time. The reduction in EnC adhesion on Days 2 and 4 following Fibro-ECM casting can be attributed to the change in the surface properties of the ECM either caused by biological activities of the embedded fibroblasts or the degradation of the hydrogel over time. Furthermore, our newly developed ‘four-stage seeding’ approach is particularly useful for vascularizing microchannels on-chip. As EnCs lose their adhesion ability rapidly after seeding to the channel, seeding fresh cells on each surface of the channel will guarantee adequate adhesion as well as the uniformity of the EnC monolayer. In addition, our described method provides a detailed guideline for constructing any ECM-embedded microvascular on-chip and can be an appropriate platform to investigate the parenchyma-embedded vasculature in the Organs-on-Chips mimetics of other organs and tissues such as skin, kidney, and liver. Although adjustments in the media components or culturing time would be needed for vascularization of cells from other tissues, the overall method including the channel fabrication, and the co-culture protocol with high percentage of endothelial media, short endothelial seeding time and especially the four-stages seeding strategy can be a good starting point. Interestingly, our studies (Fig. 4e) indicate that the vascular cross-section becomes elliptic, although those of the sacrificed structure in the ECM casting process (i.e., the silicone tubing and glass capillary tube), were all circular. We speculate that the change in the channel cross-sectional geometry from circular (right after fabrication as in Fig. 2d) to elliptic is due to the swelling of the ECM inside the anisotropic PDMS chamber. In other words, as our ECM chamber had cross-sectional dimensions of 4 mm × 2 mm, given an isotropic swelling rate of the ECM hydrogel, the channel tends to favorably deform in the longer side of the chamber making it elliptic as observed. We anticipate that with an isotropic ECM chamber (e.g., 4 mm × 4 mm) the channel cross-section would remain circular throughout the experiment. However, in our experiment, we had to reduce the thickness of the ECM chamber to 2 mm for better imaging under a confocal microscope. Moreover, similar to the results in Fig. 2d, we found that the cross-sectional area of the fabricated vascular channel was slightly larger than that of the replacement due to the casting technique. To produce more accurate channel cross-section geometry, a smaller silicone tubing should be used. Nevertheless, we note that the small change in the cross-sectional geometry of the fabricated channel did not significantly affect the flows inside the channel as the channel maintains curved and smooth inner surface. The change also is unlikely to impact the vascularization process and the barrier function of the endothelial monolayer as shown in the permeability test. The full formation of EnC monolayer with the excellent EnC barrier function and the uniform distribution of live fibroblasts on-chip confirm the success of our co-culture protocol of EnCs and ECM-embedded fibroblasts in our ECM MV-Chip.

The major limitation of our study is the limited source of primary human cell donors (a healthy adult endothelial cell [hLMVEnC] donor and a healthy stromal cell [hLFs] donor). Additional studies are required to map inter-individual variability and impact of donor age, demographics, and co-morbidities on model establishment. However, this is beyond the scope of the present study, which focused on developing and optimizing the fabrication method of the ECM MV-Chip. We chose hLMVEnCs and hLFs from representative healthy donors for our studies and performed the experiments with multiple replicates and through independent timelines to confirm the feasibility, robustness, and reproducibility of fabricating the described model. Another drawback of our work is that only one chip design was tested. Although the representative configuration was shown to be useful and practical for biological applications, other channel geometries (e.g., microchannels with smaller diameters) need to be tested to further validate our method’s ability to generate a wide range of blood vessel types for on-chip applications. These limitations are beyond the scope of the existing work and will be addressed in our future studies. In particular, we plan to validate our model with cells from both healthy subjects and patients with fibrotic lung disorder. We also plan to validate model building using EnCs from other organs (e.g., liver, kidney) and vascular channels with different diameters. Moreover, we will perform further simulation analyses to provide a full picture of the dependence of channel geometry on the flow characteristics inside the microchannel. For instance, one intriguing question is how the inlet’s inner diameter and its relative position within the channel (e.g., eccentric or perpendicular) will affect the developing region of flow inside the channel. Answering such questions using simulations will help us expediently optimize the design of this and other devices.

In summary, here we present the design, development, and implementation of a novel and reliable methodology to construct multicellular microvasculature on-chip with extraordinary characteristics that cope with various physiological studies utilizing Organs-on-Chips technology. The ability to construct vasculature with organ-like cellular complexity and biocompatible geometry on-chip opens new avenues for utilizing Organs-on-Chips technology for studies of organ development, disease modeling, and regenerative medicine.