Electrohydrodynamic Jet-Printed Ultrathin Polycaprolactone Scaffolds Mimicking Bruch’s Membrane for Retinal Pigment Epithelial Tissue Engineering

Age-related macular degeneration (AMD) is the leading cause of visual loss and affects millions of people worldwide. Dysfunction of the retinal pigment epithelium (RPE) is associated with the pathogenesis of AMD. The purpose of this work is to build and evaluate the performance of ultrathin scaffolds with an electrohydrodynamic jet (EHDJ) printing method for RPE cell culture. We printed two types of ultrathin (around 7 μm) polycaprolactone scaffolds with 20 μm and 50 μm pores, which possess mechanical properties resembling that of native human Bruch’s membrane and are biodegradable. Light microscopy and cell proliferation assay showed that adult human retinal pigment epithelial (ARPE-19) cells adhered and proliferated to form a monolayer on the scaffolds. The progress of culture matured on the scaffolds was demonstrated by immunofluorescence (actin, ZO-1, and Na+/K+-ATPase) and Western blot analysis of the respective proteins. The RPE cells cultured on EHDJ-printed scaffolds with 20 μm pores presented higher permeability, higher transepithelial potential difference, and higher expression level of Na+/K+-ATPase than those cultured on Transwell inserts. These findings suggest that the EHDJ printing can fabricate scaffolds that mimic Bruch’s membrane by promoting maturation of RPE cells to form a polarized and functional monolayered epithelium with potential as an in vitro model for studying retinal diseases and treatment methods.


Introduction
Age-related macular degeneration (AMD) is a significant cause of permanent visual loss and is estimated to bring a substantial global burden [1] . With the progression of AMD, retinal pigment epithelium (RPE), a pigmented and polarized monolayer tissue, gradually loses the ability to process the visual signals [2] . Scaffold-based International Journal of Bioprinting (2022)-Volume 8, Issue 3 RPE tissue engineering has been shown promise to build RPE models for discovering therapeutics agents and tissue transplant for patients with AMD [3][4][5] . In RPE tissue engineering, it is a great challenge to culture a mature and functional RPE monolayer on scaffolds that are biocompatible for transplant. At present, a commonly accepted approach for scaffold fabrication is to mimic the human Bruch's membrane, including the thickness and fibrous structures that could promote cell attachment and growth [3,6,7] . Bruch's membrane is a selective permeable extracellular matrix that supports RPE and has thickness of 2 -4 μm in healthy human eyes [5] . It is a great challenge in mimicking the complex functionalities and structures of Bruch's membrane in vitro.
Tremendous efforts have been put into developing new RPE culture systems from the perspectives of designing and fabricating culturing devices [4,[8][9][10] . One of the most important approaches is film casting, which can quickly produce porous membranes with thickness close to Bruch's membrane [2,5,10] . At present, this technique is widely used to build the commercialized permeable inserts, including Transwell ® and Millicell ® cell culture insert [11][12][13] . Permeable membranes of cell culture inserts were successfully transplanted into a non-human primate model [14] . However, those membranes share several drawbacks, including sheet-like morphology and undesirable mechanical properties. Another common approach to fabricate scaffolds for RPE is electrospinning, which can produce membranes with fibrous nature similar to that of Bruch's membrane with suitable thickness [3,6] . However, due to randomly oriented and highly packed structures, the capability of diffusion is greatly limited when the thickness of electrospun scaffolds is greater than 2 μm [10] . Conventional fabrication technologies for RPE cell culture can produce ultrathin structures. However, the internal order-less microstructures of those scaffolds have random pore distribution and poor interconnection and not possible to reproduce scaffolds. Three-dimensional (3D) printing technology can overcome this problem because it ensures the structure of the printed membranes highly orderly and repetitive at micrometer precision. Such high precision and reproducibility are critical not only for research purpose but also essential for commercial scale production of scaffolds.
Electrohydrodynamic jet (EHDJ) printing is a mature technology that features high precision and well-defined structure [15] . It can produce well-orientated micro-/nano-scale fibrous scaffolds with precise structure and shape control. EHDJ utilizes an electric field to pull fine fibers from the printing nozzle and control the fiber orientation with high-precision moving stage [16] . The scaffold structure can be designed by changing the stage moving path or optimizing key process parameters. The printed scaffolds have been successfully applied in 3D cell culture and tissue engineering [17][18][19][20] . However, this powerful technique has not been used to produce scaffold for RPE regeneration because the current EHDJ-printed scaffolds have large pore sizes (>50 μm), which are not suitable to hold the small RPE cells (tight size range within 8-12 μm) [21] . Due to Coulomb's effect and stage movement, it has been challenging to print scaffolds with smaller pore sizes (e.g., <50 μm) [18] . This limits the culturing of monolayer tissues, such as RPE on the surface of the porous scaffold. In a few studies recently, researchers have printed scaffolds with a pore size <50 μm. For example, He et al. successfully fabricated scaffolds with a fiber diameter of 200 nm and pore size of 10 μm [22] . This printing method was assisted with indium tin oxide glass, which made the scaffold hard to be removed from the glass. Nevertheless, those results demonstrate a great promise of obtaining scaffolds that satisfy the requirements of RPE regeneration by EHDJ printing. Herein, we reported our findings on developing the EHDJ printing method for ultrathin scaffolds with small pore sizes by optimizing EHDJ printing process parameters and applying the scaffolds for RPE monolayer culture.

Materials
Polycaprolactone (PCL) pellets with a molecular weight of 50,000 kDa were purchased from Perstorp Inc. (Capa™ 6500, Sweden). Glacial acetic acid (AcOH) in HPLC grade was purchased from Macklin Inc. (A801303, ≥99.9%, China). PCL ink with a 60% w/v concentration was prepared by mixing PCL pellets in AcOH and dissolving the mixture at 60°C for 3 h through a 100 W ultrasonication treatment. The solution was stored in an incubator for degassing at 26°C for 2 h. Subsequently, the ink was transferred to a syringe equipped with a stainless needle before printing. All the stainless nozzles used are 24-gauge with an approximate outer diameter of 0.57 mm and an inner diameter of 0.31 mm. Polyester-terephthalate (PET) Transwell ® with 0.4 μm pore membrane inserts were purchased from Corning Inc. (3470, USA).

EHDJ printing system
An EHDJ printing system developed in-house was used to conduct small pore size scaffold fabrication [16] . The EHDJ printing system ( Figure 1A) was built with an XYZ motion stage (Aerotech Inc., USA), a high-voltage power supply (DC voltage from 0 to 10 kV, Dongwen Inc., China), a single-channel syringe pump (NE-1000, New Era Pump System Inc., USA), and a digital microscope (B011, Supereyes, China). A silicon wafer (Ferrotec, Japan) was used as a scaffold collector attached to the stage fixed on the X-Y motion plane. The ambient parameters for scaffold fabrication were fixed at the temperature of 25 ± 1°C and relative humidity of 65 ± 10%.

Post-processing of printed RPE scaffolds
The printed scaffolds on silicon wafer were transferred into a vacuum drying chamber to discard the excess AcOH for 2 h. The scaffolds were detached from the silicon wafer, cut into round shapes with a diameter of 11 mm with a round shape cutter, and then attached to the empty cell culture inserts ( Figure 1B).

Morphological characterization
Fiber diameters, pore sizes, and fiber alignment of the scaffolds were examined and measured using an optical microscope (EZ4 HD, Leica, Germany). The results were expressed as mean ± standard deviations. Furthermore, the morphology was characterized using a scanning electron microscope (SEM, FEI Quanta 250, Thermo Fisher Scientific, USA). Samples were sectioned into the size of 5 mm × 5 mm and were mounted on a round metal stud using conductive copper tape for surface or cross-sectional imaging. Sputter coating was applied onto the sample with a current of 30 mA for 30 s to form a conductive layer. Subsequently, the samples were analyzed with the following settings: Beam energy at 10 kV, dwell time of 6 μs, a spot size of 4.0, and collection with a large field detector. Morphology information was obtained at 1000×, whereas the side-mounted samples were used for cross-sectional imaging and thickness measurements at 2000×. PET membranes were processed and evaluated using the same parameters and procedures, except the pore sizes. The porosity (Φ) was calculated by Where, ρ s and ρ 0 are the apparent densities of the scaffolds and biomaterials, respectively.

Mechanical testing
The initial part of the stress-strain curve was obtained using the Dynamic Mechanical Analysis (DMA, Q800, TA Instruments, USA) instrument at room temperature. The ultimate stress and strain were examined using a universal testing machine (HDB609B-S, Haida International Equipment Co., China) by stretching with an initial gauge length of 20.0 mm at speeds of 1 mm/min and 10 mm/min for the pre-loading and loading conditions, respectively. The scaffolds were cut into rectangular specimens (4 cm × 2 cm) and stretched along the wider side. The tensile stress (σ, MPa) is expressed as the ratio of the drawing force and the sum of the cross-sectional area perpendicular to the loading force direction. The strain (ε, %) is expressed as the ratio of the deformation and initial length. Young's modulus (E) is determined as the slope of the initial linear range of the stress and strain curve. The yield stress (σ Y ) and strain (ε Y ) were obtained using the modulus slope at a 2% strain offset. The ultimate tensile stress (σ b ) and strain (ε b ) were recorded at the breaking point when the sudden force drop was larger than 80%. The measurements were performed in triplicate.

Enzymatic degradation study
PCL scaffolds were weighed and placed in a 5 cm Petri dish. Lipase solution (5 mL, 10 mg/mL, Amano Lipase PS, from Burkholderiacepacia, 534641, Sigma-Aldrich, USA) was added to immerse the PCL scaffolds. Petri dishes were sealed by parafilm and placed in a thermal shaker (37°C, 10.47 rad/s) for a fixed time interval. Scaffolds were washed thoroughly by distilled water and vacuum dried for 24 hours to constant weight. Equation 2.2 was used to calculate the weight loss of scaffolds.
Weight loss (%) = 0 t 0 100 − × W W W (2.2) Where, W 0 is the initial weight of the sample and W t is the weight after different treatment times.

Water contact angle
The water contact angle was measured using a contact angle goniometer (SL200B, Kino) at 25 ± 1°C. Deionized water (4 μL) was deposited on the scaffolds, followed by stabilization of the droplet for 5 s, and then images were acquired and processed.

Cell culture
The immortalized adult human retinal pigment epithelial cell line  was obtained from the Cell Bank of Type Culture Collection of the Chinese Academy of Sciences. Before seeding to the scaffolds, the ARPE-19 cells were cultured in a 75 cm 2 cell culture flask with the Dulbecco's Modified Eagle Medium (DMEM)/ F12 (1:1) (HyClone, USA), supplemented with 10% fetal bovine serum (Serana, Germany) and 1% penicillinstreptomycin solution (Sigma-Aldrich, Germany) at 37°C with 5% carbon dioxide (CO 2 ).
The PCL scaffolds were soaked in 70% ethanol for 2 h and exposed to ultraviolet (UV) light for 2 h in a 24-well culture plate. The scaffolds were washed with phosphatebuffered saline (PBS, HyClone, USA) thrice before use. The cells were cultured to a confluency of 70 -80% in the culturing flasks and dissociated with trypsin-EDTA solution 0.25% (T4049, Sigma) to obtain a cell suspension. The cell concentrations, diameters, and viability were determined using an automated cell counter (TC20, Bio-Rad Laboratories, USA) and Trypan Blue solution (Beyotime, China). For the cell proliferation study, the cell suspension was diluted with a complete culture medium into a concentration of 1 × 10 6 cells/mL, and for the other studies, the concentration was 1 × 10 7 cells/mL.
The cell seeding procedure was divided into four steps ( Figure 1C). First, cell suspension (10 μL) was to the scaffolds and PET membranes and placed into the incubator for 3 h. Culture medium (90 μL) was added to the apical well and incubates for 12 h to allow the cells to attach to the scaffolds. Subsequently, culture medium (1 mL) was added to the basal well and incubates for 9 h. Cells that are not attached to scaffolds would pass the pores and sink to the bottom of the culture plate. Finally, the scaffolds with attached ARPE-19 cells were transferred to a new 24-well culture plate.

Cell proliferation
Proliferation and cytotoxicity of ARPE-19 cells were assessed using Cell Counting Kit-8 (CCK-8, Dojindo Molecular Technologies, Inc., Japan) assay. The cell growth was observed on 1, 3, 7, and 14 days after cell seeding. Cell culture media in each well were replaced with 100 μL growth media and 10 μL of CCK-8 reagent on the test day, followed by incubating at 37°C and 5% CO 2 for 1.5 h in the dark. The mixture (100 μL) was transferred to another 96-well plate and absorbance values were read at 450 nm using a microplate reader (Eon, BioTek, USA).

Permeability and diffusivity of scaffolds for in vitro RPE models
The permeabilities of growth substrates, with and without cells, were determined using two distinct molecular weights of fluorescein isothiocyanate (FITC)-dextran: 40 kDa and 500 kDa. Concentration standard curves in PBS were first generated for each molecular weight. Fluorescence intensities were measured in black 96-well plates with a multi-mode microplate reader (SpectraMax iD3, Molecular Devices, USA) at the excitation wavelength of 492 nm and emission wavelength of 518 nm.
The culture media were removed from both apical and basal wells and were washed with PBS. The basal wells were filled with 600 μL of PBS, and the apical side was filled with FITC-dextran solution (100 μL, 5 mg/mL dissolved in PBS). The permeation was monitored at regular intervals of 15,30,60,90, and 120 min using a 50 μL culture medium from the basal side. The following equation determined the apparent permeability coefficients P app (cm/s): Where, A is the surface area of the culture substrates (cm 2 ), C 0 is the concentration of a target molecule in the apical chamber (μg/cm 3 ), and dQ/dt is the linear slope of the molecular flux (μg/s).
For a better 3D visualization of cell nuclei and scaffold, spatial normalization is performed on the CLSM images to ensure the same resolution of each axis. Then, the scaffolds were manually rendered with white color (Figure 6D), and the visualization was generated using Mayavi2 with maximum intensity projection [23] .

Western blot analysis of proteins
ARPE-19 cells cultured on scaffolds for total protein extraction were washed with cold PBS before being lysed with RIPA lysis buffer at 4˚C for 30 minutes. Cell lysates were centrifuged, and the supernatant was collected. The protein concentrations were determined by the BCA protein assay kit (Beyotime, China). The same amount of proteins in each sample was mixed with a loading buffer and heated at 100°C for 5 min to denature proteins. The protein samples were dissolved in sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to PVDF membranes.
After treatment with 5% w/v blotting grade blocker, the membranes were incubated with corresponding primary antibodies: Na + /K + -ATPase and GAPDH (diluted at 1:1000 v/v in Tris Buffered Saline with Tween ® 20 (TBST), Beyotime, China). The corresponding signals were then captured using horseradish peroxidase (conjugated secondary anti-rabbit IgG antibodies, 1:5000 dilution v/v, Beyotime, China) on a chemiluminescence imaging system. Quantification of bands was performed using ImageJ software, and data were analyzed and compared. Western blot was performed in triplicates in three independent experiments.

Transepithelial potential (TEP) and transepithelial electrical resistance (TEER)
TEP and TEER across the ARPE-19 cells were measured using a Millicell-ERS-2 voltmeter (Millipore, USA) at 1, 2, and 3 weeks. The TEER value was calculated according to the following equation: is the resistance of the insert, and A is the membrane area (cm 2 ) of the insert.

Statistical analysis
All experiments were performed thrice, and values were expressed as the mean ± standard deviation, unless otherwise stated. Statistical analysis was performed by multiple unpaired t-tests with False Discovery Rate (FDR) approach (desired FDR = 1%) using GraphPad Prism (version 9.1.1, GraphPad Software, USA).

Results and discussion
3.1. EHDJ printing of ultrathin scaffolds with small pore size 3D printing inks suitable for EHDJ printing are usually made of insulating polymeric materials, which led to the charge accumulation in the printed fibers, cause Coulombic repulsion, and may result in the drift of fiber deposition [18,24] . Fabricating scaffolds with a small pore size (≤50 μm) demand the fiber line spacing being closer to the pore size. According to Coulomb's law, as the printed fibers are close to the jet (Figure 2A), Coulombic forces will significantly decrease the accuracy of the fiber deposition and alignment. To reduce such effects, we developed a manufacturing workflow to minimize the effects of Coulombic forces and allow printing of scaffolds with small pore sizes and well-aligned fibers.
It was suggested that the mechanical drawing force dominated by the stage speed is the determining factor on resolution, positioning, and alignment of the EHDJ-printed fibers [16] . Since the key parameters in the EHDJ printing process greatly impact the jet formation, several process parameters were kept constant during the optimization process of printing. The applied voltage was kept at 2 kV, the ink feeding rate was fixed at 0.6 μL/min, and the nozzle-to-substrate distance was set at 15 mm. According to the velocity feedback results (Figure 2A), the time required for stage complete traveling cycle of 80 mm was decreased from 598 ms to 348 ms by increasing the stage speed, which resulted in increased mechanical drawing force and decreased fiber diameter. In Figure 2C, the motion stage with a slower moving speed was printed more randomly deposited fibers, and the possible reason is that the fibers were repulsed or attracted by adjacent printed fibers. When stage speed was increased to 250 mm/s, fiber deposition and alignment accuracy were greatly improved with the desired average fiber spacing. By increasing the stage speed to 350 mm/s, the variations of the fiber spacing were significantly decreased as the fibers were precisely printed. With the assistance of mechanical drawing forces, we achieved the EHDJ printing of scaffolds with excellent fiber alignment and desired pore sizes.
Another critical factor that affects fiber deposition is the dwell time at the turning point. Since the designed scaffolds were printed in a zigzag moving path, the stage movement decelerates at each turning point (Figure 2B). At this moment, the solution accumulates and deposits on the substrate. Within a specific range, the feed rate decreases, the jet might be distorted and affect the fiber formation. With the increase of dwell time, the accumulated solution or fiber will change morphology at the turning point. Therefore, the effect of dwell time at the turning point should be minimized ( Figure 2C). Moreover, the increasing stage speed also reduced the effect of the accumulating solution at the deceleration period and turning point.

Comparison of the performance of RPE scaffolds and Transwell membrane
With the optimized printing parameters, we printed two types of monolayer PCL scaffolds with pore sizes of 20 μm (S20) and 50 μm (S50). We propose that EHDJprinted PCL scaffolds could mimic the mechanical properties of Bruch's membrane. To test this hypothesis, we first characterized the morphological and physical properties of the scaffolds. The commercialized permeable cell culture devices, particularly the PET Transwell with a membrane pore size of 0.4 μm, are the most used for in vitro RPE cultures and subsequent pathological investigations. Thus, membranes used in the Transwell plate were selected as the controls for comparison with the EHDJ-printed PCL scaffolds. From the top view of the commercial membrane (Figure 3A), the pores were randomly distributed, and pore density also varies from area to area, which is expected as the membrane was made using the track-etching process [12,13] . In contrast, the PCL scaffolds (Figure 3B and C) printed have well-aligned fibers in both X and Y directions and interconnected pores with equal sizes, and both scaffolds have similar fiber diameters (~20 μm). In Table 1, the commercial membrane has a pore density of 4.0 × 10 6 pores/cm 2 , and it is much higher than the 3D-printed PCL scaffolds, which are 6.25 × 10 4 pores/cm 2 and 2.04 × 10 4 pores/cm 2 for S20 and S50, respectively. However, the pore areas were drastically different. S50 has the largest pore area, which is around 2500 μm 2 , whereas the pore area of S20 is around 400 μm 2 . Both are much larger than that of the PET membranes (~0.13 μm 2 ). The commercial (C) Typical fiber morphology with same preset line spacing (50 μm) and different velocity (applied voltage = 1.8 kV, nozzle-tosubstrate distance = 1.5 mm, and solution feed rate = 1.5 μL/min), and corresponding frequency distribution of fiber spacing (n = 3); (D) morphology of printed fibers at turning points with different dwell times (scale bar = 400 μm).

A C D B
membrane has an extremely low porosity at around 1.5%. The porosity of the S50 PCL scaffold is around 55% and that of S20 is around 33%. Therefore, we expected the RPE cells to grow on the PET membrane and that the PCL scaffold would have significantly different mass exchange behavior with the culture media. The thickness of the membranes for RPE monolayer culture is another determining factor for suitability of implanting in vivo since there is limited subretinal space. The commercial PET membrane has a thickness of around 13 μm, measured by SEM. Due to the limitation of PET membrane fabrication, the actual values of membrane thickness may vary up to 60% of the nominal value (10 μm) [25] . In contrast, the printed scaffold has a uniform thickness of 7 μm, making it suitable for in vivo study and possible for implant purpose [2] .
It is also important for scaffolds to have similar biomechanical properties to Bruch's membrane to avoid incompatibility with surrounding tissues in vivo. In the human body, Young's modulus of Bruch's membrane ranges from 1.0 to 18.8 MPa [26] . In contrast, Young's modulus of the commercial PET membrane is about 10 times higher (~180 MPa) [27] , making it too stiff and susceptible to damage in the surrounding tissue if transplanted. The Young's modulus for S20 and S50 scaffolds are much lower, that is, 45.5 ± 5.3 MPa and 8.9 ± 3.2 MPa, respectively ( Table 2). In addition, the ultimate tensile stress, ultimate tensile strain, yield stress, and yield strain of the S20 and S50 were significantly different (Figure 4A and B). Since large pore scaffolds require lesser fibers, S50 scaffold is much easier to deform, whereas S20 can withstand higher stress. Both scaffolds have similar elastic properties compared with the actual Bruch's membrane. Young's moduli of S50 are within the range of Bruch's membrane and thus might be better suited for implant applications.
Moreover, both PCL scaffolds have a good degradation behavior under lipase treatment (Figure 4C), with ~40% of the scaffolds being degraded after 48 h of immersing with lipase solution in 37°C (calculated by Equation 2.2). In contrast, the commercial PET membrane is not biodegradable. Thus, the PCL scaffolds may be promising for culturing RPE monolayers, with potential application in transplantation.
Both PET (materials used to make commercial membrane) and PCL are hydrophobic. PET membrane exhibited a water contact angle of 62.28 ± 1.07° (Figure 4D), S20 scaffold has a higher water contact angle of 95.99 ± 1.72°, and the S50 has the highest water contact angle of 112.81 ± 1.66°. The connective tissues in vivo have higher hydrophilicity as they are made of proteins, but the hydrophobicity for PCL scaffolds is an essential factor as a substrate for much small monolayer cell RPE to grow on without falling through the pores (20 and 50 μm). The water on the hydrophobic scaffold can form a layer of water film and prevent cells from falling.

Proliferation and distribution of RPE cells on PCL scaffolds and membrane
The ARPE-19 cells were cultured over 70% confluency and inoculated to the scaffolds/membrane. ARPE-19 cells (94%) have cell sizes ranging from 9 to 20 μm with mean   (Figure S1A), comparable to RPE cells in the human macula (14 μm in average diameter and 12 μm in height [28] ). Around 1.0 × 10 4 RPE cells were seeded onto each culturing substrate giving a cell density of 3.0 × 10 4 cells/cm 2 . The PET membrane reached a 100% cell seeding efficiency ( Figure 5A) because RPE cells could not pass through the pores of the membrane (0.4 μm). The cell seeding efficiency on S20 could reach 92.87 ± 1.96% much higher than that on S50 (80.77 ± 5.82%), which is still satisfactory. We found that although the pore sizes of both scaffolds were much larger than the diameters of RPE cells, the cell suspension did not traverse the pores. This counterintuitive phenomenon may be due to the hydrophobicity of the scaffolds and the surface tension of cell suspension preventing the cells from leaking to the bottom of the culture plate. Thus, after 3 h of seeding, some cells can attach to the scaffolds and adhere to the fibers. Most cells did not traverse through scaffold pores to the basal wells of culture plate, despite being filled with culture media in the bottom well. After 3 h of cell attachment, several cells showed spindle-like and elongated morphologies on all the substrates ( Figure 5C).
Notably, most of the cell shape was round on the S50, possibly due to the inadequate support provided by the scaffolds for cell adherence. After 7 days of culture, RPE cells formed a complete monolayer tissue-like structure with even cellular distribution on PET membranes and S20 scaffolds, and the cell boundaries were hard to observe. Exceptionally, the RPE cells did not form a complete cell monolayer on the S50 scaffolds after 7 days of culturing ( Figure 5C).
The proliferation of RPE cells on the substrates was analyzed with a sensitive colorimetric assay ( Figure 5B). From day 1 to day 15, the cell amounts were similar on the PET membrane and the S20. As for the S50, the cell number was significantly lower than that of the S20 and PET membrane. On day 15, the cell numbers on different substrates were comparable because they reached confluence on day 7 and only had around 20% increase after day 7. Cell proliferation was similar on the PET membrane and S20 PCL scaffold from day 1 to 15.

Characterization of RPE maturation on substrates
The immunostaining results for actin, ZO-1, Na + /K + -ATPase, and the nuclei not only provide details about the maturation of RPE cells on the culturing substrate but also indicate whether in vitro RPE monolayer attains polarization ( Figure 6A). On day 7, the F-actin in the cells cultured on PET membrane and S20 PCL scaffold was relatively high. The distinct F-actin filaments were peripherally organized around the cells and appeared near the edges of the cells. However, the quantities of actin appeared lower on the cells cultured in S50, and F-actin was not observed on several nuclei. The formation of tight junctions is essential to the proper function of the RPE and its contribution to the maintenance of the outer blood-retinal barrier. ZO-1 is a well-recognized marker that indicates the presence of a tight junction complex in monolayered RPE cells (Figure 6B) [6] . The immunofluorescence analysis of the RPE monolayer on day 7 displayed cell junctional staining of ZO-1, supporting the presence of tight junctions. However, the S50 scaffold did not form proper junctional boundaries.
Cell polarity is a crucial characteristic of mature RPE cells related to the retina's unique basal and apical structures, which affect phagocytosis and material exchange [5,10] . Na + /K + -ATPase localization was investigated to determine the RPE cell polarity. Z-stack  confocal micrographs revealed that all the cells grown on three substrates had polarized cell expression of Na + / K + -ATPase ( Figure 6C). Na + /K + -ATPase functions to actively transport Na + and K + across the cell membrane to maintain a functional resting potential [5] . Compared to the PET membrane, the cells cultured on S20 and S50 had visibly more Na + /K + -ATPase located apically. When CLSM was used to visualize the nuclei, the same trend was observed, that is, S20 and PET had a monolayer, whereas S50 had multilayering cells.
Relative positions of RPE cells on culturing substrates are essential information revealing whether the scaffold can hold the RPE cells on the same level. Both scaffold and RPE were treated with lipophilic fluorescence dye and observed under CLSM. Stacks of images were processed into 3D images, as shown in Figure 6D. The images showed that the RPE could form a monolayer on the S20 scaffold. No collapse or cell nuclei were sinking into the pore, which could be explained by the presence of tight junctions in maintaining cell-cell contact and the monolayer integrity. In contrast, most cells cultured on S50 scaffolds were sinking into the pores and only a few still above the fibers, due to pore sizes are much larger than the cell sizes.

Ultrathin scaffolds with small pore size may help RPE cells to form a monolayer tissue
For an effective RPE model, the culturing substrate of RPE should mimic the permeability of the actual Bruch's membrane. As expected, EHDJ printed scaffolds had much higher permeability coefficients (P app ) than the PET membranes since they had much larger pore sizes. Especially, for the blank S50, the coefficient was close to infinity since both smaller probe molecules (40 kDa) and larger probe molecules (500 kDa) could permeate through the upper well to the bottom well immediately (Figure S2C and D). Blank S20 also had high permeability coefficients, which were around 17.14 ± 0.34 × 10 −6 cm / s for smaller molecules and 19.36 ± 0.94  1, 3, 7, and 15. (C) Optical images of cell growth on the culturing devices after 3 h and 7 days. Scale bar = 50 μm, *P < 0.05, **P < 0.001, ***P < 0.0001, ****P < 0.00001. A C B × 10 −6 cm / s for the larger molecules. The PET membrane also allowed the transport of probe molecules, which had P app ranging from 6.76 ± 1.48 × 10 −6 cm / s to 7.62 ± 0.65 × 10 −-6 cm / s. After 7 days of culture, the permeability coefficients for three substrates were analyzed again, and decreasing trends were observed due to the presence of cells (Figure 7A and B). An apparent decrease was observed for the PET membranes before and after RPE cell culture, which decreased by around 5.01 × 10 −6 cm / s. The same trend was also discovered on the EHDJ-printed scaffold, in which the ability for probe molecules to pass S20 and S50 had a remarkable decrease. For 40 kDa FITC-dextran, the permeability coefficient of the S20 dropped to 12.74 × 10 −6 cm / s ~ 16.88 × 10 −6 cm / s. For S50, P app was around 13.48 × 10 −6 cm / s after 7 days of culture. As a reference, the RPE-choroid complex in bovine eyes has the much lower P app for 40 kDa FITC-dextran at around 0.46 × 10 −7 cm / s [29] . Our results suggested that permeability can be adjusted by changing the preset pore size of EHDJ-printed scaffolds. Since RPE- Figure 7. Functionality comparison of monolayer ARPE-19 models. Average permeability coefficients of (A) 40 kDa and (B) 500 kDa FITC-dextran probes of S20, S50, and PET membrane without cell seeding and after 7 days of RPE seeding and culturing (n = 3). (C) TEER and (D) TEP change over 3 weeks (n = 3). (E) Western blot results revealed the significant upregulation of Na + /K + -ATPase in EHDJ-printed scaffold-based RPE models (n = 3). *P < 0.05, **P < 0.001, ***P < 0.0001, ****P < 0.00001. Bruch's membrane structure plays an essential role in the selectivity of nutrients and wastes in the outer bloodretinal barrier, thus it is important to mimic the actual permeability of the native membrane. The difference of cells and the stage of maturity of the monolayer of RPE may account for the large P app . Nonetheless, it is remarkable that all the culturing substrates had decreased permeability over time after RPE cell seeding, suggesting a steady gain of barrier function conferred by RPE cells on the scaffolds. Particularly, probe molecules pass freely through the pores of blank S50 scaffolds. After culturing, the monolayer cell formation is likely responsible for the drop of permeability.
The TEER level is another important indicator for the barrier function of cultured RPE monolayer [30] (Figure 7C). Before culturing, all substrates were measured with a culture medium for calculating background resistance ( Table 2). PET membrane has the most negligible porosity, which leads to the highest background TEER reading of 73.7 Ω·cm 2 . The TEERs of S20 and S50 were around 63.1 Ω·cm 2 and 50.2 Ω·cm 2 , respectively. The resistances measured of S20 and PET membrane were comparable throughout the entire duration and reached a plateau on day 14 with a TEER at around 47 Ω·cm 2 . Similar readings have been reported using immortalized RPE cells grown on Transwell membrane and electrospun scaffolds [11,31] . For RPE cultured on S50, the TEER reached the same level as the PET membrane and S20 at the end of the third culture week. This phenomenon suggested that cell sinking or collapsing in the pore could affect its formation of barrier function as proper monolayer was not developed. Nevertheless, it also could reach the same level of TEER after 3 weeks of culturing. In the human body, RPE together with retinal vascular endothelium forms the blood-retinal barrier, which is closely associated with photoreceptor outer segments. Thus, EHDJ-printed scaffolds are promising for in vitro culture of RPE to build tight junctions and finally present barrier function.
The barrier function and selectivity of RPE benefited from the cooperation of different proteins and their physiological functions, including Na + /K + -ATPase and tubulins. Na + /K + -ATPase can regulate sodium gradient formation for osmotic gradient and water diffusion [32,33] . Then, the tight junction complex proteins (such as ZO-1) at the boundaries of neighboring RPE cells will increase the resistance between the apical and the basal sides of the cells. Therefore, a high concentration of Na + will usually accumulate at the apical side of RPE and form a TEP difference. In our case, the TEP difference in S20 scaffold is significantly higher than that grown on PET membrane and S50 scaffolds ( Figure 7D). Thus, the expression of Na + /K + -ATPase in RPE cells cultured on S20 should be higher than RPE cultured on PET membrane and S50. As predicted, increased protein levels of Na + /K + -ATPase were quantified on the S20 scaffold ( Figure 7E). Studies have shown that Na + /K + -ATPase is significantly reduced in aged mice and that TEP plays a functional role in cell migration, division, polarization, and development [33] . These data are consistent with a conclusion that S20 scaffold may have the ability to provide a better growing environment for the RPE cells that manifest higher expression of Na + /K + -ATPase and increased TEP.

Conclusions
The first EHDJ-printed Bruch's membrane mimic was designed and made for integration into the current commercial freestanding structures. The PCL scaffolds showed similar biomimetic properties to Bruch's membrane, including native-like thicknesses, biomechanical properties, and permeability. It was determined that the S20 PCL scaffold had better performances in culturing a monolayer of ARPE-19 cells. Moreover, PCL scaffolds have good biodegradability and can mimic extracellular matrix environments in vivo; therefore, they are conducive to RPE maturation. Taken together, our results demonstrated that the EHDJ printing technique can fabricate scaffolds mimicking Bruch's membrane with high resolution and precision, and the printed scaffolds have great potential to help RPE cells form mature tissue with the desired functionality. Future work can explore the design and fabrication of scaffolds suitable for culturing embryonic stem cell-derived RPE for transplantation.

Funding
This work was financially supported by Key Program Special Fund in Xi'an JiaoTong-Liverpool University (XJTLU) under Grant KSF-E-37. This work was also supported by the National University of Singapore (Suzhou) Research Institute under an internal grant to the Center for Peak of Excellence on Biological Science and Food Engineering.