3D Printed Transwell Microfluidic Devices for Epithelial Cell Culture with Shear Stress

Abstract

In this paper, we describe how 3D printing can be used to fabricate a microfluidic-based transwell cell culture system with robust fluidic connections for long-term cell culture and recirculating flow. This approach consists of an electrospun collagen scaffold sandwiched between two laser-cut Teflon membranes that match the fluidic design. Madin-Darby canine kidney (MDCK) cells were cultured on the collagen scaffold to create an epithelial cell monolayer. Introduction of cells into the device was facilitated by a printed reservoir that could be closed after proper cell seeding with minimal effect of the flow profile over the cells. The resulting MDCK cell monolayer was exposed to continuous flow and transport through the cell layer and could be monitored by sampling from the basolateral channel network. COMSOL simulations and flow injection analysis were used to determine the effect of the reservoir geometry on the shear stress that cells experience. A variety of analytical tools were used to assess the effect of flow over the cells in this model. This includes confocal microscopy and potentiometric scanning ion conductance microscopy (to determine morphology and conductance), as well as transendothelial/epithelial electrical resistance (TEER) measurements and reverse transcription-quantitative polymerase chain reaction studies (for gene expression analysis). Finally, a drug transport study with the cell model was carried out using two drugs (caffeine and digoxin) to determine the apparent permeability of high and low permeability drugs, with results being similar to findings from in vivo studies as well as studies where MDCKs have been transfected to form more resistive barriers. This approach holds great promise for the creation of more in vivo-like, flow-based barrier models for transport studies.

1. Introduction

In drug transport studies, it is common to use static transwell devices where one cell type is grown on a synthetic membrane. A drug is usually added to the apical (top) side of the device and transport through the cell monolayer is monitored by sampling from the basolateral (bottom) side. While these simple, static approaches have been used to develop in vitro barrier models for the blood brain barrier, kidney, and placenta, , there are disadvantages. One is the manner in which the cells are grown on the membrane support. Cells in many biological systems are surrounded by a basement membrane, which is a thin sheet of an extracellular matrix (ECM). The ECM is a complex fibrous system that is made up of many macromolecules such as proteoglycans, glycosaminoglycans, collagen, fibronectin, elastin, and laminin and organized in a cell/tissue specific manner. , Among these macromolecules, collagen is the most abundant fibrous proteins found, accounting for ∼ 30% of the ECM. Using realistic ECMs for culturing cells is key to the development of tight junctions (TJs) between neighboring cells. These TJs attribute important biological functions including providing a barrier for selective permeability as well as maintaining the structural, compositional and functional polarity of cells. Formation of TJs also are important in helping to control paracellular transport of many molecules such as nutrients, drugs and toxins. Moreover, TJs play important roles in physiological homeostasis by regulating signal transduction. For the development of pharmaceuticals that target the central nervous system, the constrictive nature of TJs in the blood brain barrier (BBB) presents a drug design challenge where measuring transport across BBB mimics is critical. Consequently, developing in vitro platforms for studying and characterizing cellular TJs and barriers has become a fast-advancing research area.

Many transwell cell culture systems employ the use of permeable membranes such as polycarbonate (PC), poly­(ether sulfone) (PES) and polyester (PET). Studies have reported the use of these systems with synthetic permeable membranes for barrier models (including the BBB), drug development screening process across podocytes (to study the renal system), and cell migration and invasion studies. However, the microenvironment that cells experience on an ECM basement membrane plays a vital role in many cellular processes including cell proliferation and differentiation as well as regulating gene expression. ,, This has led to research on employing biologically derived materials to create the ECM in basement membranes. This includes materials such as collagen, silk, and laminin to create a better mimic of the in vivo microenvironment. −

Another disadvantage of static transwell devices is the absence of flow over the cell monolayer. In vivo, cells are exposed to flow and hemodynamic forces, with a shear stress over their apical surface. As a result, cells respond to this stimulus by adapting their behavior toward the mechanical cues arising from shear stress. , This phenomenon is known as mechano-transduction, a process that occurs when cells transform the biophysical (mechanical) stimuli into biochemical signaling within the cellular microenvironment. , Numerous studies have reported cells can alter their morphology, function and gene expression in response to shear stress. Endothelial cells at the BBB, epithelial cells in the intestine, , renal tubular epithelial cells in the kidney, , and corneal epithelial in the cornea all experience shear stress to some degree. Continuous flow over the cells helps to supply nutrients and remove waste; however, a comprehensive understanding of the significant impact of flow features over cells is an active research area. There is clearly a need to create flow-based transwell cell culture systems to investigate the effects of shear stress on these models and result in more in vivo-like barrier mimics.

There has been work on developing microfluidic approaches for transwell/transport studies. This includes work to study the impact of shear stress over cells by introducing flow into microfluidic devices. ,− In terms of epithelial-based fluidic barrier models, Kimura et al. fabricated a PDMS-based microfluidic device that has a bilayer microchannel and a polyethylene terephthalate porous membrane sandwiched between the two layers for culturing renal proximal tubule epithelial cells. The device was integrated into an existing automatic cell imaging system for drug efficacy and toxicity studies. In a separate work, Choudhury et al. devised a microfluidic device that has four primary components: a micropatterned PDMS block with channels, a polycarbonate porous membrane, a micropatterned intermediate layer, and a glass slide. The device was sealed with a combination of silicone sealant and plasma treatment. Human primary kidney epithelial cells, human ADPKD cystic cells, and mouse PKD2 knockout cells were cultured in the device under flow conditions, with the results providing insights into the interplay between pressure gradients and fluid transport in the kidney epithelium. In terms of the methods used to make microfluidic devices, 3D printing has recently been shown to provide many advantages over PDMS for fluidic studies. Important for this study, one advantage is the robust nature of the fluidic connections through the use of threaded ports. , Three-dimensional structures and flow paths can easily be printed, and the devices can be made in batch without the use of a clean room.

Here, we utilized 3D printing to develop a microfluidic-based transwell cell culture system that integrates flow over cells as well as the ability to easily sample from the basolateral side of the barrier, with robust fluidic connections for long-term (4–7 days) cell culture and recirculating flow. The new microfluidic-based transwell system consists of an electrospun collagen scaffold sandwiched between two laser-cut Teflon membranes that match the fluidic design. Madin-Darby canine kidney (MDCK) cells were cultured on the collagen scaffold to create an epithelial cell monolayer. This cell line, which has been widely used for developing kidney barrier models, was chosen due to its ability to form monolayers with tight junctions, high electrical resistance and low permeability. Cell introduction was facilitated by a printed reservoir that could be closed after proper cell seeding with minimal effect on the flow profile over the cells. The resulting MDCK cell monolayer was exposed to continuous flow and transport through the cell layer could be monitored by sampling from the basolateral channel network. The effect of the reservoir on the shear stress that cells experience in the device was investigated with COMSOL simulations and flow injection analysis. The influence of laminar flow at differing flow rates was characterized by using confocal microscopy and potentiometric scanning ion conductance microscopy (P-SICM) for cell morphology changes; impedance spectroscopy for transendothelial/epithelial electrical resistance (TEER) measurements, and reverse transcription-quantitative polymerase chain reaction (RT-qPCR) for gene expression analysis. Finally, a drug transport study with the kidney cell model was carried out using 2 drugs (caffeine and digoxin) and LC-UV to determine the apparent permeability (P_app) of high and low permeability drugs, with results similar to what has been seen with in vivo studies.

2. Experimental Section

2.1. Fabrication of Collagen Scaffold

All MDCK cells were cultured on a collagen scaffold made from electrospinning collagen I solution. Briefly, a homogeneous solution of 8% (w/v) collagen I (Elastin Products Co., Owensville, MO) in hexafluoroisopropanol (HFP, Millipore-Sigma, St. Louis, MO) was sprayed through a 15 cm long, 150 μm I.D. fused silica capillary from a screw thread glass vial onto a grounded aluminum foil sheet that was wrapped with lens paper. A high voltage of +15 kV was applied to the glass vial (via a platinum wire) as was 11 psi of helium. After 2 h of electrospinning, the collagen scaffolds were cross-linked with 1-ethyl-3-(3-(dimethylamino)­propyl) carbodiimide hydrochloride (EDC, 400 mM, Millipore-Sigma, St. Louis, MO) and N-hydroxysuccinimide (NHS, 400 mM, Millipore-Sigma, St. Louis, MO) in pure ethanol for 4 h, as previously described. To avoid exposing the scaffolds to the atmosphere and becoming denatured before cross-linking, the collagen scaffolds were handled in a glovebag (Cole Parmer, Vernon Hills, IL) filled with nitrogen gas. All the cross-linked collagen scaffolds were stored in a desiccator after rinsing with ethanol and vacuum drying for future use.

2.2. 3D Printed Devices Production and Assembly

All the devices used for this study were printed with a Stratasys J735 PolyJet printer (Eden Prairie, MN) using VeroClear resin. The device design was completed in SOLIDWORKS and file (.SLDPRT format) was uploaded to the printer. Upon the completion of printing all the device components, they were subsequently cleaned overnight in a Super DT3 CleanStation (PM Technologies, Inc., Osseo, MN) that contains a sodium hydroxide/sodium metasilicate solution to improve the biocompatibility of the devices. To ensure a smooth surface for sealing, all devices were polished with sandpaper using a commercial grinder-polisher (MetaServ 3000, Buehler, Lake Bluff, IL). The top component of the device was sonicated for approximately 8 h so that all the residue from the flow channel was removed. Before any cell work, printed parts were rinsed with DI water and sterilized with 70% ethanol.

As shown in Figure a, the device used for this study has two 3D printed components that, when assembled, form the top and bottom layers. Figure S1 contains key dimensions of the CAD device design. The channel dimensions given in the text and other figures were measured with optical microscopy or SEM. The top layer houses a flow channel that is 12 mm (length) x 1.10 mm (width) x 1.01 mm (depth), with printed threads for interfacing with commercially available flangeless fittings. The channel of the device was constructed in such a manner that the cell culture media can flow through from threaded inlet threads to the outlet and over the cell monolayer. This flow channel at the basement of the top layer is open after printing but when sealed against the bottom layer, it closes and allows the culture media to flow directly over the cells. This 3-dimensional fluidic channel was made such that support material could be easily removed. In the center of the top layer, an apical reservoir was printed so that the cell seeding onto the collagen scaffold is possible. After cells were added, this apical reservoir was left open for an hour so that cells were immobilized and attached to the scaffold under static conditions. After that period of time, the apical reservoir was closed with an M4 screw (M4–0.70, 10 mm in length, PEEK, Grainger, Lake Forest, IL) followed by the introduction of flow. Each screw was first added to an unassembled device to a point where the screw was flush with the top of the channel interface. The screw was then marked so that when it was used with an assembled device, the screw would not penetrate past the top of the channel. The M4 screw was polished flat before use (see Figure S2 for images of screw and device interface).

1.

3D printed microfluidic transwell device with integrated collagen scaffold. (a) Diagram showing CAD rendered design with an expanded view of they layers used in device. (b) Side view diagram of the assembled device. (c) Schematic of the experimental setup along with a picture of assembled device. Different flow rates can be introduced over the MDCK cells seeded on the collagen scaffold. Flow is directed into a Falcon tube (containing 5 mL of media), with a cap that has drilled holes for tubing and CO₂ exchange.

The complete device assembly involves sandwiching Teflon gaskets between two 3D printed layers as seen in Figure b. First, the electrospun collagen scaffold was sandwiched in between two laser-cut (275 μm thick) Teflon gaskets after mounting the collagen scaffold on large pore (30 μm) polycarbonate membrane (for mechanical support). The design of the Teflon gaskets (Grainger, Lake Forest, IL) was completed with SOLIDWORKS in such a manner that the top gasket was in compliance with the top layer of the device (having a replica of the open flow channel) and the bottom gasket aligned with the bottom layer of the device (having an opening to access the basolateral channel). The speed, power, and frequency parameters used for the laser cutter (Epilog Helix Laser, Colden, CO) were 100%, 50%, and 800 Hz, respectively. Then, the Teflon/collagen sandwich was placed in between the 3D printed layers. When all the assembled components were tightened with bolts, washers, and nuts (M4–0.70 × 25, Hillman, Cincinnati, OH), a leak-free device was achieved. Using a torque screwdriver, it was observed that applying 55 cN·m was sufficient to tighten all the components for assembling the flow device. The basolateral chamber was prefilled with 100 μL of cell culture media. The access port to this basolateral chamber was closed during the cell culture period to maintain fluid levels. After the cells were confluent (typically 4 days after postseeding), these access ports were made accessible for further characterization experiments such as flow studies, TEER measurements, and drug transport studies.

Male luer fittings (P-202X, IDEX Health & Science, Oak Harbor, WA) with Tygon tubing (0.02 in. ID and 0.06 in. OD, Cole-Parmer, Vernon Hills, IL) were threaded into the printed device. This Tygon tubing was further extended with PVC tubing (0.89 mm ID Orange/Orange, ISMATEC, Wertheim, Germany) to connect to a peristaltic pump (IPC multichannel pump, ISMATEC). The tubing was sterilized with 70% ethanol and prefilled with pre-warmed cell culture media before circulating the media. Figure c shows the setup for complete circulation of flow over a cell monolayer, with the peristaltic pump withdrawing media from a Falcon tube (containing 5 mL of media). In order to accommodate the inlet and outlet tubing as well as the exchange of carbon dioxide from the incubator, the cap of the Falcon tube had three drilled holes. Media in the Falcon tube was changed on a daily basis.

2.3. Cell Culture

Madin-Darby canine kidney cells (MDCK, Strain type: NBL-2, CCL-34, ATCC, Manassas, VA) between passages 10–20 were cultured and maintained in T-75 flasks (Corning, NY) using Eagle’s Minimum Essential Medium (EMEM, ATCC) supplemented with 10% Fetal Bovine Serum and 1% penicillin-streptomycin. Cells were subcultured every other day. Before seeding cells into the device, rinsing of the collagen scaffolds was required to eliminate any trace HFP and NHS. This was done by soaking the scaffolds in cell culture media for 15 min in the incubator, followed by rinsing with cell media. This step was repeated 4 times. Approximately 1.6 × 10⁵ cells were seeded into the assembled device via the apical reservoir onto the collagen scaffolds. The cells were allowed to attach to the scaffold for an hour before adding the M4 screw and introducing flow. After cell attachment and closing of the reservoir, different flow rates of cell culture media (0.015, 0.2, and 0.5 mL/min) were introduced through the device for at least 4 days to result in a confluent layer of MDCK cells.

2.4. Finite Element Methods Using COMSOL Simulation

All finite element methods (FEM) simulations were done with COMSOL software (version 6.2) and laminar flow physics. The purpose of these simulations was to estimate the shear stress that cells experience during proliferation at volumetric flow rates between 0.015 and 1.0 mL/min. The simulations consisted of two types of channel geometries, with and without a reservoir aperture. In both cases, the channel floor center consisted of a 50 × 50 array of hexagonally packed semispheres (5 μm radii; bottom 25% of all the cell layer height was removed to resemble cell–substrate attachment) to model a monolayer of cells. The average shear stress that the cell monolayer experiences (at each volumetric flow rate for both geometries) was estimated by determining the shear stress at the cell array’s most apical morphological point (see the SI for more information on these calculations).

2.5. Flow Injection Studies

In order to validate the COMSOL simulation and visualize the flow profile throughout the flow channel, two different devices were utilized, one with the cell culture reservoir and one without. Other than the reservoir, all other dimensions were identical. Using a syringe pump (70–2209, Harvard Apparatus, Holliston, MA), a discrete plug (introduced via an off-chip 4-port injector) of a 300 μM fluorescein solution was pumped into the devices at different flow rates (0.1 mL/min, 0.2 mL/min, 0.5 mL/min and 1.0 mL/min). All the injections were made using a 4-port injector with a 1 μL volume rotor (Valco Instruments, Houston, TX). The injected plug of analyte was detected 10 mm downstream from the reservoir location. Images and videos of fluorescent plugs were obtained using an inverted fluorescence microscope (Olympus BX60) fitted with a 10X objective (Olympus PLN2X, Japan). All images and videos were recorded with an AmScope (MU203-BI) camera and software, then processed and analyzed using ImageJ. For analysis and presentation purposes, all the data generated from ImageJ were then smoothed in PeakFit (San Jose, CA) with a Savitzky-Golay filter (0.5% window) to eliminate the background noise. The peak width at half height (W _1/2) was measured for each peak (n = 9 for each condition) using ChromPerfect software (Denville, NJ) and variance (σ²) was calculated from the relationship: W _1/2 = 2.354σ.

2.6. Immunofluorescence Imaging

Confocal imaging of MDKC cells to qualitatively evaluate the confluency and the direction of cell alignment as a function of flow rate was performed with a Leica SP8 confocal microscope. Cell samples were first rinsed with PBS (3X), followed by fixation with a 4% paraformaldehyde solution at 4 °C for 30 min. Cell permeabilization was then completed with a 0.1% Triton X-100 solution (Millipore-Sigma) by incubating the samples at room temperature for 15 min. Next, the samples were blocked with a 2% Bovine Serum Albumin solution (Millipore-Sigma) at room temperature for 45 min. The cells were then labeled with 5 μg/mL of ZO-1 primary monoclonal antibody conjugated with Alexa Flour 488 (MA3–39100-A488, Invitrogen, Rockford, IL) at 4 °C overnight. This stain was used to characterize the presence of ZO-1 proteins on the cell membrane. The cell samples were then stained at room temperature for 15 min each with 1:1000 (v/v) rhodamine phalloidin (R415, Invitrogen, Rockford, IL) for actin cytoskeleton visualization and 1:2000 (v/v) of 4’,6-diamidino-2-phenylindole (DAPI, 62248, ThermoFischer) for cell nuclei labeling. In between each staining step, the cell samples were rinsed three times with PBS. Post-image processing of all the microscopic images was accomplished with ImageJ. The inverse aspect ratio (IAR) was used to quantitate changes in cell morphology as a function of flow rate using the following equation:

$$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{s}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}(\mathrm{I}\mathrm{A}\mathrm{R})=\frac{\mathrm{length\; of\; the\; short\; axis}(w)}{\mathrm{length\; of\; the\; long\; axis}(l)}$$ 1

2.7. Scanning Ion Conductance Microscopy

All topography and local apparent conductance measurements were performed with a home-built scanning ion conductance microscope (SICM). Single barrel and theta barrel quartz capillaries (Sutter Inc., Novato, CA) were pulled with a P-2000 puller (Sutter Inc., Novato, CA) to produce nanopipettes for SICM measurements. Pipettes were filled with PBS (x1) at a pH of 7.2 prior to measurements. Collagen scaffolds were supported by a backing of a polycarbonate (PC) membrane (STERLITECH Corp., Auburn, WA), which were both adhered to a perfusion cell via 3 M tape masks with a 1 mm hole punched out from the center. The PC membrane had a pore size of 30 μm and a density of 1 × 10⁴ pores/cm².

Bare fiber measurements were done by submerging the membranes in culturing media to mimic the experimental conditions for cell culture. This media housed the reference Ag/AgCl for topographical measurements, along with the immersed (single barrel) pipet with a back inserted Ag/AgCl electrode. All resulting topographical maps were taken at a 2% threshold with a DC voltage of 0.2 V vs Ag/AgCl applied to the pipet electrode. Potentiometric-SICM (P-SICM) studies were performed with PBS (x1) prefilled theta pipet with back inserted Ag/AgCl electrodes, one electrode per barrel. MDCK cells were cultured with a flow rate of 0.2 mL/min for 4 days. After this period of time, the device was disassembled, and the collagen scaffold (with cells) was rinsed with PBS. The cells were fixed with a 4% paraformaldehyde solution at 4 °C for 30 min, followed by rinsing with PBS three times.

The sample was then placed in a perfusion cell by mounting the cells onto the center back side of a 3D printed Petri dish (40 mm diameter, 3 mm footed supports and an 8 mm diameter through hole; Formlabs Form2b SLA printer with BioMed Clear v1 resin) that was adhered with a 3 M tape mask with a 1 mm center hole. This device was placed on an optical transparent dish and both the Petri dish and the transparent dish were filled with PBS connecting the upper and lower chambers of the assembled perfusion cell where the sample bridged the connection between the chambers. The Ag/AgCl reference electrode and Pt counter mesh electrodes were always placed at the apical side (upper chamber of the perfusion cell), while another Ag/AgCl rod was placed on the basolateral side of the scaffold to serve as the working electrode to supply a transmembrane potential. A DC potential of 0.2 V vs Ag/AgCl was applied to the pipet electrode and a gated sawtooth AC waveform with a 200 mV_PP was applied to the working electrode at a frequency of 5 Hz. All data was processed via Gwydion imaging processing software.

2.8. Quantitative PCR Analysis for Gene Expression

To determine the number of cells present in the scaffold under different flow conditions (for normalization purposes), the cell containing collagen scaffolds were removed from the devices and incubated with 1 mL of 15 mg/mL collagenase solution (Elastin Products Co.) in a 37 °C water bath to dissolve the collagen. Then, a 10 μL aliquot of the resulting solution was added to a hemocytometer, and the cells were manually counted. The counting was repeated three times and the average number of cells were used to calculate the total number of cells on the scaffold.

To extract total RNA from the cells cultured on the scaffolds, the scaffolds were treated with lysis buffer. Total RNA was isolated using the Monarch Total RNA Miniprep Kit (New England Biolab, Ipswich, MA) according to the manufacturer’s instructions. For accurate comparison of gene expression between flow conditions, the RNA template amount was normalized based on the number of cells rather than the total template RNA in a sample. Briefly, we determined the total RNA yield and cell number from each scaffold and calculated the RNA content per cell. From the 0.015 mL/min flow rate (which exhibited the lowest amount of total RNA per cell), 30 ng of RNA was used as the reference input for real-time quantitative PCR (RT-qPCR), corresponding to RNA derived from approximately 1,094 cells. We then back-calculated the RNA amounts from the 0.2 mL/min and 0.5 mL/min flow rate conditions that represented the same number of cells (1,094) and used those amounts as template RNA for RT-qPCR to facilitate per-cell gene expression comparison across all different flow conditions. RT-qPCR was performed using a QuantStudio Real-Time PCR System (Applied Biosystems, Waltham, MA). The cycling conditions were 55 °C for 10 min, followed by 95 °C for 60 s, and 40–45 cycles of 95 °C for 10 s and 60 °C for 60 s, with the gene-specific primers as listed in Figure S3. The raw gene expression Cq values were normalized to canine GAPDH, which was the endogenous internal control. To further minimize variability, expression values were normalized to the average cell number, and results are presented as relative expression per cell.

2.9. TEER Measurements

A Gamry Reference 600 potentiostat (Warminster, PA) was used to measure the electrochemical impedance of the cell monolayer to determine the TEER values on day 4 of cell culture. An Ag/AgCl electrode was used as both reference and working electrodes. The Ag/AgCl electrode rod was fabricated by depositing chloride ions onto the silver wire after immersing the silver wire into an equimolar solution of 1 M hydrochloric acid and 1 M ferric chloride solution overnight. A platinum (Pt) wire was used as a counter electrode. To set up the three-electrode system, the working electrode (WE) was placed in the apical reservoir, the counter (CE) and reference (RE) electrodes were in the basolateral channel via the access port. The CE and RE electrodes were embedded inside flangeless fittings using Epoxy Gel (Permatex, Solon, OH) and then placed into the basolateral channel access ports. The parameters utilized for this study were: AC amplitude of 12 mV, a frequency range from 1 Hz to 100,000 Hz, and 10 points per decade. Using the Nyquist plot generated from the impedance measurement, the resistance of the cell monolayer was calculated as follows: ,

$${\mathit{R}}_{\mathrm{cell\; monolayer}}={\mathit{R}}_{\mathrm{substrate}+\mathrm{cell\; monolayer}}-{\mathit{R}}_{\mathrm{substrate}}$$ 2

The TEER value (Ω·cm²) was calculated by multiplying the resistance of cell monolayer by the active surface area of cells in the flow channel reservoir area (0.044 cm²).

2.10. LC-UV Studies for Drug Transport

Drug transport studies were performed on the fourth day of cell culture using caffeine (Alfa Aesar) and digoxin (TCI America, Portland, OR). Caffeine has been reported as a high permeability drug, whereas digoxin is commonly utilized as a low permeability drug for transport studies. , To perform an experiment, 500 μM of caffeine or 100 μM of digoxin solutions in transport buffer (HBSS) were perfused over the cell monolayer from the apical compartment. The media from the basolateral compartment was exchanged with pre-warmed HBSS. All of the drug transport experiments were executed in an incubator by performing replacement experiments under the “sink” conditions. Samples (10 μL) were taken from the basolateral compartment every 30 min for 2 h in the caffeine study and every 30 min for 3 h in the digoxin transport study.

All the collected samples were analyzed and quantified using high-performance liquid chromatography (HPLC, Shimazu) with a UV–vis detector. For the caffeine study, the analyte was separated on a Thermo Aquasil C18 (3 μm pore, 50 mm L x 2.1 mm ID) column and detected at the wavelength of 270 nm. For the digoxin study, an Armor C18 Analytical Column (5 μm pore, 50 mm L × 3.0 mm ID) column was used and digoxin was detected at 220 nm. For both caffeine and digoxin studies, acetonitrile with 0.1% formic acid (solvent A) and DI water with 0.1% formic acid (solvent B) were used as the mobile phase. In all analyses, 5 μL aliquots of the samples were injected into the separation column under binary gradient mode. The initial solvent concentration was kept at 5% aqueous acetonitrile for 1 min, followed by a linear increase of up to 95% aqueous acetonitrile over a period of 10 min, with this condition being held for 3 min before returning back to the initial concentration before the next injection.

Apparent permeability (P _app) values for the drug transport study were calculated using the following formula: ,−

$$P_{\mathrm{a}\mathrm{p}\mathrm{p}}=\frac{\frac{\mathrm{d}C}{\mathrm{d}t}\times V_{\mathrm{B}}}{A\times C_0}$$ 3

where dC/dt is the linear rate of change of concentration in the basolateral compartment with respect to time (μM/s), V _B is the total volume of the HBSS buffer in the basolateral chamber (100 μL), A is active surface area where the drug is transported from the apical to basolateral compartment (0.044 cm²), and C ₀ is the initial concentration of the drug perfused over the cell monolayer (500 μM for caffeine and 100 μM for digoxin, respectively).

3. Results and Discussion

3.1. Fabrication of Scaffolds and Devices

The main motivation of this work was to develop a robust transwell fluidic device that integrates biologically derived fibers for cell culture, a methodology to easily and reproducibly introduce cells into the device, a means to have flow (shear) over the cells, and utilize the device for transport studies. In order to create more realistic in vitro mimics of vivo barriers such as the kidney and BBB, it is essential to have flow (shear) over the cells so that the type of mechanotransduction stimuli that occurs in vivo is replicated. For the ECM replica, collagen scaffolds were employed in order to provide in vivo like matrix for biocompatibility, cell adhesion, proliferation and differentiation. The scaffolds used for this study were electrospun using 8% (w/v) collagen solution in HFP and cross-linked with an equimolar solution of EDC and NHS in such a way that the fiber dimensions were maintained to be approximately 500 nm in diameter with an overall scaffold thickness of 30 μm.

A PolyJet-based 3D printer with VeroClear resin was used to fabricate the flow device. The device was made up of two separate layers – the top layer for the flow channel and apical side of the cell layer and a bottom layer for a channel network representing the basolateral side of the cell layer. The top layer houses threaded ports to integrate flow from a peristaltic pump through the flow channel as well as access ports for the basolateral compartment. The two separate 3D printed layers were assembled with the collagen scaffold being sandwiched between laser-cut Teflon using bolts and nuts. The Teflon gaskets around the scaffold and between the layers enable a tightly sealed device that is leak-free after the assembly. The ability of 3D printing to produce threaded ports that can accept fittings made integration with tubing and off-chip peristaltic pumps facile and reproducible. After assembling the device, cell culture media could be introduced via a peristaltic pump. In these studies, three different flow rates were used: 0.015, 0.2, and 0.5 mL/min for cell culture studies. Cells could be cultured under flow conditions for up to 7-days without leakage issues (extended time periods were not investigated).

A key part of the device is the cell reservoir and how cells are introduced. In contrast to methods where cells are aspirated or flowed into the device, , this approach used a reservoir that was printed into the top layer so that cells can be pipetted and seeded onto the collagen scaffold under static conditions. After a period of time for cell attachment (typically 1 h), this reservoir could then be closed with a plastic M4 screw. To keep the constant laminar flow profile over the cells, it was important that this M4 screw was polished until a flat surface was achieved. When this screw closes, approximately 155 μm gap formed at the interface between the 3D printed threads of the cell reservoir and the screw (see Figure S2). For this work, cells were seeded on the collagen scaffold via the reservoir in the top layer. This step allowed the cells to settle and adhere onto the scaffold for an hour before the flow was introduced over the cells using the peristaltic pump (after closing the reservoir). A key advantage to this approach is that the cells can easily be accessed after the flow experiment is complete, in a manner that does not perturb the monolayer. As will be seen below, this enabled confocal imaging, gene expression and TEER studies in a manner where the effects of flow on the MDCK cells could be measured.

3.2. Flow Characterization Studies

The flow channel (12 mm × 1.10 mm × 1.01 mm) in the top layer of the 3D printed component was designed in such a way that the laminar flow of cell culture media would introduce shear stress over the cells after the cell reservoir is closed (all reported channel dimensions were measured with optical microscopy or SEM). The top layer was also designed to connect the apical flow channel to 3D printed threads from the sides of the device via a 3-dimensional flow path. To study the optimization of the flow profile through this designed flow channel and estimation of the shear stress the cells experience, finite element methods (FEM) using COMSOL was utilized. This current design included a reservoir aperture between the center of the channel ceiling and the bottom of the threaded plug that was measured to be 155 μm. FEM via COMSOL was used to solve the Navier–Stokes and continuity equations to determine the shear stress (see the SI for more information). These calculations were done for each of the following flow rates: 0.015, 0.1, 0.2, 0.5, and 1.0 mL/min, with resulting calculated shear stress values being 0.0096, 0.064, 0.13, 0.32, and 0.64 dyn/cm², respectively.

The effect of the reservoir aperture is represented in Figure , where Figure a is a schematic showing the key channel dimensions (including the 155 μm gap) and the direction of volumetric flow (0.5 mL/min for the data in this figure). Figure b shows the estimated profile of the fluid flow velocity at the channel center plane, down the length of the channel, parallel to flow direction from left to right. It shows that the 155 μm gap slightly perturbs the laminar flow, contributing to the 18% decrease in applied shear stress compared to the channel geometry without the gap. The effect of the reservoir aperture is further demonstrated in Figure S4a, which shows a 1D array of 30 velocity profile slices along the length of the channel, where the plane of each slice is orthogonal to the direction of flow. Figure S4b is a reference of unperturbed laminar flow in the channel while in Figure S4c the channel geometry becomes taller due to the reservoir gap. This causes a direct influence on the laminar flow, broadening and decreasing the flow intensity. However, as seen in Figure S4d, as the channel geometry returns to its original height past the reservoir, the flow original profile is restored, and a sharpening of the flow intensity is seen.

2.

(a) Schematic of the flow channel in order illustrate the rectangular shape, key dimensions (including the reservoir gap), and the flow direction. (b) COMSOL simulation with a 0.5 mL/min volumetric flow rate showing the fluid flow rate profile along the length of the channel as a 2D slice. Note that there is a monolayer of cells (50 × 50 array) at the center bottom wall of the channel in (b). (c) Overlaid results of variance (from flow injection experiments) and the inverse of estimated shear stress (from COMSOL simulations) as a function of the volumetric flow rate.

To experimentally determine the effect of the extra reservoir volume on flow over the cell monolayer, flow injection analysis (using fluorescein as the analyte) and fluorescence detection were utilized. Two different devices were used, one with a reservoir and one without. The detection point was the same on both devices, equivalent to 10 mm from the reservoir area exit. Discrete injections of fluorescein were made into each device (n = 9 for each condition). The resulting peaks were analyzed to determine the average peak width at half height (W _1/2) and peak variance (σ²). Comparing the peak variance of each device at various flow rates was used to determine the amount of band broadening (and therefore dilution and change in shear) that the cells experience. As shown in Figure c, as the flow rates increase, the difference in variance between the devices starts to narrow. At a flow rate of 0.2 mL/min, the reservoir increases the variance by 1.65 times, while at a flow rate of 0.5 mL/min there is only 1.34 times increase (Figure S5 contains representative data from the 0.5 mL/min comparison). At a flow rate of 1.0 mL/min, there is no statistical difference in the variances

Additionally, Figure c shows the inverse average shear stress results along apical points over a monolayer of cells as a function of the volumetric flow rate. This graph also has an overlay of the peak variance vs flow rate results from experimental studies. These studies show that the flow intensity profiles sharpen and increase as the flow rate increases, making the signal more uniform along the channel geometry including the reservoir area. This is reflected with the flow injection data, where the amount of band broadening (expressed as variance), decreases as flow rate increases, to the point where there is little to no difference between the two geometries at 0.5 mL/min or higher. This shows that increasing the volumetric flow rate in the reservoir device results in more heterogeneity of solutes under flow while increasing the average stress on the cell monolayer.

3.3. Changes in Cell Morphology As a Function of Flow Rate

There have been several reports of using MDCK cells in transport/barrier studies. These cells can form a cell monolayer when grown on membranes and produce high TEER values, indicating the presence of well-developed intercellular TJs. MDCK cells have been employed as a membrane screening tool as well as transport barrier model. It has been reported that this MDCK cell line has a characteristic of “cobblestone” morphology. In addition, the morphology of cells can be elongated and oriented along with the direction of the flow in response to the shear stress. ,

To study the cell morphology changes under the shear stress, three different flow rates were applied: 0.015, 0.2, and 0.5 mL/min. As mentioned beforehand, the calculated shear stress via COMSOL for these flow rates are 0.0096, 0.13, and 0.32 dyn/cm², respectively. In order to ascertain the effect of flow on cell morphology in this 3D printed flow-based transwell device, the direction of the flow being introduced over the cells was denoted during the cell culture. On the fourth day of culturing cells, the cell samples were taken out of the device and stained with ZO-1 antibodies (for ZO-1 protein), rhodamine phalloidin (for cytoskeleton staining) and DAPI (for labeling the cell nuclei). The previously reported static 3D printed device, and the current flow device were employed to study the cell morphology changes under no flow vs very slow flow (0.015 mL/min) conditions. As seen in Figures a and b, MDCK cells show the same morphology under no flow (0 dyn/cm²) and very slow flow (0.0096 dyn/cm²) depicting these conditions as “static like”. Under flow rates of 0.2 mL/min and 0.5 mL/min (0.13 dyn/cm² and 0.32 dyn/cm², respectively), MDCK epithelial cells respond to the shear stress by elongating and orienting themselves in accordance with the flow direction, as seen in Figures c and d. This change in cell morphology can be quantified by measuring the cell shape and elongation and calculating the inverse aspect ratio (IAR) using eq . A cell with a circular shape has an IAR = 1, while an elongated cell has an IAR < 1. As shown in Figure a, the IAR values for MDCK cells in this study clearly decrease as flow rates increase. These values show that MDCK cells respond to the shear stress by elongating and orientating their direction under flow rates of 0.2 and 0.5 mL/min. Since the 0.015 mL/min flow rate gave the same value as static conditions, this flow rate was used in subsequent studies to achieve static-like conditions.

3.

Confocal images of MDCK cells under different flow rates (0–0.5 mL/min). The epithelial cells respond by elongating and aligning to the direction of the flow. (a) Image of cells under no flow (static). (b) Cells cultured under a flow rate of 0.015 mL/min (similar morphology to cells cultured under static conditions). (c–d) Cells cultured under a flow rate of 0.2 and 0.5 mL/min, respectively. Cell nuclei are stained with DAPI (blue), actin in the cytoskeleton is stained with rhodamine phalloidin (red), and ZO-1 protein is labeled with ZO-1 antibodies and its conjugated dye (Alexa Flour488, green). Scale bar: 100 μm.

4.

Characterizing the response of MDCK cells to shear stress. (a) Plot of IAR vs flow rate shows the elongation of the cells in response to flow (n = 6). (b) Expression of genes that encode for tight junction proteins under different flow rates (n = 9). * indicates significantly different at 95% of confidence level.

Scaffolds used to culture cells in this study consisted of electrospun collagen fibers that serve as a more physiologically relevant substrate. A control experiment to collect the topographical maps of each substrate was done to estimate the surface roughness (see Figure S6). Cells were cultured onto a collagen fiber membrane and allowed to proliferate under a flow rate of 0.2 mL/min. The cells were then fixed to the substrate surface and measured with P-SICM. The purpose of this measurement was to observe cell morphology and spatially correlative chemical information along a section of the cell monolayer cultured under flow. As shown in Figure a, the topographical map displays cell junctions, some of which are easily discernible between a cell and the surrounding cell bodies, however, not all of them are apparent. Figure b, however, shows the spatially correlated local apparent conductance map, where there are larger local apparent conductance measurements over the cell junctions compared to those measured over the cell bodies. Moreover, cell junctions that were not apparent in the topography map are much more apparent in the local apparent conductance map.

5.

Topography (a) and local apparent conductance (b) maps of fixed MDCK cells, cultured onto an electrospun collagen fiber membrane taken at 250 nm/pixel resolution and 0.2 V_TM,PP at 5 Hz vs Ag/AgCl in PBS x1 buffer media.

3.4. Gene Expression Changes in Response to Shear Stress

In order to maintain physiological homeostasis, TJs play a vital role in biological systems. These TJs form barriers and act as fences and gates by restricting the transport of nutrients and toxins to the target organs such as kidneys, GI tract, and brain. TJ proteins can be classified as integral membrane proteins and peripheral membrane associated proteins. Integral membrane proteins are composed of claudins and occludin families, while peripheral membrane associated proteins are made up of zona occludens (ZO) families. In MDCK epithelial cells, it has been reported that ZO-1, claudin-1 and occludin proteins express more abundantly than many other tight junction proteins. Therefore, in this study we chose to perform expression experiments for genes that encode for these proteins, in order to determine the effect of flow on the development of TJs in MDCKs.

In order to quantitate the abundance of these three TJ proteins (ZO-1, claudin-1, and occludin) as a function of flow rate, the expression of genes that encode for these proteins was accomplished by using real-time quantitative PCR (qPCR). First, in order to get an accurate cell count for these studies, cells were cultured under conditions for 4 days, after which time the device was disassembled. The collagen scaffolds were digested with 1 mL of a 15 mg/mL collagenase solution, which liberated the cells for subsequent manual counting with a hemocytometer. The number of counted cells from the scaffolds exposed to low shear stress (37,200 ± 6,060) was significantly higher than those to higher shear stress (15,600 ± 3,050 and 6,670 ± 1670 for 0.13 dyn/cm² and 0.32 dyn/cm², respectively) as shown in Figure S7. This correlated with an increase in cell size in response to flow (Figure S7). This can be explained by the considerably elongated cells taking up more room in the flow channel, with fewer cells being required for confluence. Next, the expression changes of ZO-1, claudin-1, and occludin per cell across the three different flow rates were assessed using qPCR. Since the number of cells in each condition was flow rate dependent, we first quantified both the total RNA and cell count per scaffold, allowing us to calculate RNA per cell (Figure S7). According to the data, we observed that the 0.015 mL/min flow rate studies exhibited the lowest RNA per cell, and therefore, we used 30 ng of RNA for qPCR from that condition (corresponding to RNA derived from 1,094 cells). We then used RNA equivalent to 1,094 cells from the other flow studies to ensure the per-cell comparison of gene expression. All qPCR reactions were normalized to the expression of canine GAPDH, which served as the internal control. To confirm the reproducibility of the results, two separate peristaltic pumps with different flow rates (such as 0.015 mL/min vs 0.2 mL/min) were employed during the cell culture for this study to increase the reproducibility of the comparisons. The results, which are shown in Figure b, clearly show that normalized relative gene expression per cell for claudin-1 and occludin was at least 2-fold higher using the higher shear rates, as compared to the static condition. However, ZO-1 expression did not exhibit a significant increase as a function of flow rate. A consistent increase in total RNA per cell from cells that were exposed to higher flow rates was also observed (see Figure S7), indicating a likely enhancement of transcription. It is clear from these results that the mechanical stimuli from increased shear results in increased expression of claudin-1 and occludin, which are important tight junction proteins in MDCKs. These findings underscore the importance of considering both mechanical forces and cellular context in modeling epithelial physiology and reinforces the relevance of flow-based systems in studying barrier tissue development.

3.5. TEER Measurements

A common measure of cell confluence in transwell barrier studies is Trans-epithelial/endothelial Electrical Resistance (TEER). This noninvasive technique is commonly used each day of a transport study to asses cell confluency. With cells that form monolayers, TEER values increase over time, and when the values plateau, transport studies can be initiated as a confluent monolayer exists. To study the effect of flow over cells in the 3D printed device, as well as the integrity and functionality of the cell monolayer after long-term exposure to flow, impedance spectroscopy was used to determine TEER values. After the MDCK cell monolayer was seeded and exposed to flow for 4 days, the cell reservoir was opened (by removing the M4 screw) and impedance measurements were made with electrodes in the apical reservoir and the basolateral access reservoirs. A Nyquist plot was generated, and the TEER value was calculated with the consideration of the cell monolayer active surface area using eq . This was done for the 3 different flow conditions (static-like, 0.2 and 0.5 mL/min). As shown in Figure a, there was a significant increase in the TEER value when going from the low flow (static like) condition to higher shear rates. This correlates with the increased expression of tight junction marker data in the previous section, showing that the ability to have appreciable shear rates over the cells creates an increased barrier for passive transport.

6.

Characterization of cell monolayer integrity as a function of flow rate. (a) Plot of TEER measurements under different flow rates (n = 9). (b) Drug permeability results using caffeine (high-permeability drug) (n = 3) and digoxin (low-permeability drug) (n = 6). * indicates significantly different at 95% of the confidence level.

3.6. Drug Transport Study

Finally, we sought to demonstrate the ability of the 3D printed transwell system to be used in drug transport studies. After culturing cells for 4 days under various shear stresses, a drug transport study was performed in the incubator by first using caffeine, which is a high permeability drug. The cell culture media from the basolateral side of the cell monolayer from the 3D printed device was first replaced with pre-warmed HBSS buffer. Then 5 mL of 500 μM of a caffeine solution was introduced over the cells via the peristaltic pump (with the caffeine solution being added to the Falcon tube shown in Figure c). Samples (10 μL) were collected from the basolateral side of the cell monolayer via the access port every 30 min over a 2-h period. All of the samples were collected under “sink” conditions, where the concentration from the basolateral compartment never exceeds 10% of the concentration of analytes from the apical side. After collection, the amount of caffeine in each sample was determined using LC-UV. The apparent permeability (P_app) of the caffeine was calculated using eq . As shown in Figure b, the apparent permeability of caffeine was 28.5 ± 2.70 × 10^–6 cm/s (n = 3) for flow rates of 0.015 mL/min (static-like condition) and 16.5 ± 1.24 × 10^–6 cm/s (n = 3) for a flow rate of 0.2 mL/min. The apparent permeability of caffeine under flow conditions is smaller than in other studies using MDCK cells under static conditions (79.3 × 10^–6 cm/s) and compares well to previous studies where the apparent permeability of caffeine has been reported to be 11.1 × 10^–6 cm/s with in vivo studies (using rats). In addition, it has been reported that noncerebral cell lines such as MDCK can be transfected with a human multidrug resistance 1 (MDR-1) gene to prove that the transfected cell line exhibits in vivo-like restrictive paracellular pathway for screening of different pharmaceuticals. , Using the transfected MDCK cell line, the apparent permeability of caffeine has been reported as low as 20.2 × 10^–6 cm/s, which also compares well to the 0.2 mL/min flow rate findings of 16.5 × 10^–6 cm/s.

Using a similar procedure, another transport study was performed using digoxin, a commonly used low permeability drug. , A 100 μM solution of digoxin was perfused over the cell monolayer and 10 μL of samples from the basolateral side of the cell monolayer were collected every 30 min over a 3-h period. After analysis with LC-UV, the calculation for the apparent permeability (P_app) of the digoxin was completed using eq . The apparent permeability of digoxin was found to be 1.57 ± 0.143 × 10^–6 cm/s (n = 6) with a flow rate of 0.015 mL/min (static-like) and 0.230 ± 0.0175 × 10^–6 cm/s (n = 6) with the 0.2 mL/min flow rate. The apparent mobility under flow conditions was less than previous studies with unmodified MDCK with static conditions (2.37 × 10^–6 cm/s) and similar to an in vivo study using rats (0.10 cm/s x 10^–6 cm/s) and the previously mentioned transfected MDCK cell line (0.28 × 10^–6 cm/s⁴⁷). Again, these values compare favorably to the 0.2 mL/min findings with this nontransfected cell line. Overall, this data shows that with this flow device and flow rates, we are able to create MDCK barrier model that has permeability values that are similar to in vivo findings and also work with transfected cells that have been engineered to form more resistive barriers.

4. Conclusions

In this work, a 3D printed transwell microfluidic device was successfully designed and fabricated in a manner that epithelial (MDCK) cells could easily be immobilized and shear stress introduced over the cells for long-term cell culture studies. The approach enabled transport studies through an electrospun collagen ECM membrane. The flow profile was characterized with FEM via COMSOL as well as flow injection analysis. It was shown that cell morphology, ion conductance, TEER values and expression of genes that encode tight junction proteins are all affected by shear in the devices. A drug transport study using caffeine and digoxin showed that this approach led to a MDCK barrier model that has permeability values similar to those found with in vivo studies as well as when using MDCKs that have been transfected to form more resistive barriers. The observed increase in claudin and occludin expression aligns with enhanced epithelial barrier properties. This data is also well supported by the TEER and permeability data and suggests that fluid shear stress promotes tight junction maturation in a cell-intrinsic manner. It is clear from these studies that this robust 3D printed transwell cell culture system with flow over the cells and a biologically relevant ECM leads to more in vivo-like conditions for transport studies. The ability to easily access the cells after the flow experiment (by removing the screw) will allow future work to examine other proteins involved in junction formation as well as different quantitative approaches such as Western blots and metabolic profiling. Future work will also further develop this device to increase the shear rate that cells experience and transition to the use of endothelial cells in the presence of flowing red blood cells to result in a true BBB mimic.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmeasuresciau.5c00045.

• Figure S1, key dimensions (from CAD designs); Figure S2, micrographs of the flow channel/screw interface; Figure S3, table of primer sequences for the candidate genes studied using RT-qPCR; Figure S4, COMSOL simulations under 0.5 mL/min volumetric flow rate; Figure S5, representative flow data (fluorescence intensity over time) from flow injection analysis studies; Figure S6, topographical and local apparent conductance maps of electrospun collagen fiber membranes; Figure S7, amount of total RNA extracted per cell after detaching cells from the collagen scaffold by treating with collagenase and manual cell counting (PDF)

The authors declare no competing financial interest.