Development of High-Resolution Three-Dimensional-Printed Extracellular Matrix Scaffolds and Their Compatibility with Pluripotent Stem Cells and Early Retinal Cells

Abstract

Purpose:

Widely used approaches for retinal disease modeling and in vitro therapeutic testing can be augmented by using tissue-engineered scaffolds with a precise 3-dimensional structure. However, the materials currently used for these scaffolds are poorly matched to the biochemical and mechanical properties of the in vivo retina. Here, we create biopolymer-based scaffolds with a structure that is amenable to retinal tissue engineering and modeling.

Methods:

Optimal two-photon polymerization (TPP) settings, including laser power and scanning speed, are identified for 4 methacrylated biopolymer formulations: collagen, gelatin, hyaluronic acid (HA), and a 50/50 mixture of gelatin/HA, each with methylene blue as a photoinitiator. For select formulations, fabrication accuracy and swelling are determined and biocompatibility is evaluated by using human induced pluripotent stem cells and rat postnatal retinal cells.

Results:

TPP is feasible for each biopolymer formulation, but it is the most reliable for mixtures containing gelatin and the least reliable for HA alone. The mean size of microscaffold pores is within several microns of the intended value but the overall structure size is several times greater than the modeled volume. The addition of HA to gelatin scaffolds increases cell viability and promotes neuronal phenotype, including Tuj-1 expression and characteristic morphology.

Conclusion:

We successfully determined a useful range of TPP settings for 4 methacrylated biopolymer formulations. When crosslinked, these extracellular matrix-derived molecules support the growth and attachment of retinal cells. We anticipate that when combined with existing patient-specific approaches, this technique will enable more efficient and accurate retinal disease modeling and therapeutic testing in vitro than current techniques allow.

Introduction

Recent advances in genome editing^1,2 and three-dimensional (3D) retinal differentiation (ie, organoids or spheroids)^3–9 have taken in vitro retinal disease models closer to the in vivo state than ever before by enabling the study of genotype-to-phenotype relationship, and by introducing inherent intercellular interactions, respectively. However, induced pluripotent stem cell (iPSC)-derived retinal spheroids are often heterogeneous and their development can be inconsistent. Thus, monitoring or isolating distinct cell types or layers can be difficult with this approach, limiting access to valuable disease modeling information.

Further, retinal spheroids are far from optimal for testing candidate drug or gene therapies on patient-derived retinal cells. For example, iPSC-derived retinal spheroids are often varied in size, and since distinct layers are difficult to distinguish and the “system” is closed, subretinal or intravitreal administration of experimental therapeutics, measurement, or even estimation of their effect cannot be replicated in a realistic manner without compromising the spheroid.

As an alternative, if artificial photoreceptor cells could be created and organized in a distinct layer, they could also be combined with a retinal pigment epithelium monolayer and choroidal endothelium to form a complete, multilayered outer retinal unit. Further, inner retinal neurons could likely be grafted on top of the photoreceptor layer for a complete simulation of the in vivo retina. By addressing the complex in vitro organization of photoreceptors, the most critical cell type in the retina, such models would retain the valuable 3-dimensionality and intercellular interactions of retinal spheroids. Meanwhile, a larger-scale study of the in vitro retina and more effective monitoring of individual cell layers and their response to various drug or gene therapies would be attainable.

One of the major challenges remaining for realizing this level of sophistication in retinal modeling in vitro is achieving proper photoreceptor organization. We and others have developed several biomaterial-based strategies, primarily for retinal cell therapy, that may be applied to overcome this limitation. For example, microfabricated elastomers (nondegradable and degradable) can help organize and polarize photoreceptor precursor cells to form a uniform layer.¹⁰ Further, our group has demonstrated that two-photon polymerization (TPP), a form of high-resolution 3D printing, can also be used to create scaffolds (nondegradable) that organize retinal progenitor cells in a pattern that recapitulates the photoreceptor cell layer.¹¹ Recently, we also applied this fabrication technique to poly(caprolactone), a degradable polyester, and showed that the resulting scaffolds are well tolerated by retinal cells and in vivo retinal tissue.¹²

However, high-fidelity disease models should mimic the in vivo environment of the diseased cell type as closely as possible, including both biochemical and biomechanical cues. If cells are not exposed to the same stiffness signals that they would experience in their innate environment, they could respond to disease treatment differently than they would in the body.¹³ As such, the materials used to create the precise photoreceptor cell scaffolds described earlier have potentially suboptimal properties, in part because they do not recapitulate the low compressive modulus (10 kPa) of retinal tissue.¹⁴

Moreover, as synthetic polymers, the biochemical properties of these materials are far from a perfect match for the extracellular matrix (ECM) of the retina. In fact, one recent study that examined retinal progenitor cell migration and displacement on various ECM substrates suggests that incorporating these molecules into scaffolds could improve photoreceptor cell integration.¹⁵ Further, one group claims to have shown evidence that matching the culture substrate to composition of the interphotoreceptor matrix (IPM) enhances pluripotent stem cell differentiation and maturation toward photoreceptors,¹⁶ but this work has not yet been published in a peer-reviewed journal (only a conference abstract is available). Although naturally occurring polymers such as chitosan,¹⁷ hyaluronic acid (HA) and methylcellulose,^18,19 and even decellularized retinal ECM^20,21 have been used for retinal tissue engineering applications, the fabrication approaches used in these studies did not allow precise control of the scaffold structure at scales that are relevant to photoreceptor cell organization and alignment.

In this work, we used 3 modified ECM molecules (collagen, gelatin, and HA) to create scaffolds with structures that are appropriate for organizing human photoreceptor cells. The high-resolution 3D-printing technique we employed, TPP, can be used to overcome traditional fabrication limitations by restricting polymerization volume to only the focal point of a laser, thus creating individual polymer strands with a resolution up to 50 nm. In this case, we adapted ECM molecules to TPP by using methacrylate-modified versions of collagen, gelatin, and HA, and by mixing with a water-soluble photosensitive compound, methylene blue. Since collagen, gelatin, and HA are all naturally occurring polymers that, when crosslinked, have similar moduli to retinal tissue,^14,22–24 we anticipate that they will be more physiologically relevant substrates for retinal cells than conventional poly(styrene), polyesters, or poly(dimethylsiloxane) (PDMS).

For each methacrylated ECM molecule and for a mixture of methacrylated gelatin and HA, we identified the most efficient laser power and scanning speeds for TPP, created prototypic scaffolds, and determined the fidelity of the final structure to the original 3D model. Finally, we assessed the toxicity of each material using human iPSCs and postnatal rat retinal cells, and we examined retinal cell interactions with the biopolymer scaffolds. With the fundamental knowledge of this fabrication procedure, we and others are now poised to create structures that mimic the human photoreceptor layer in terms of structure, biochemistry, and stiffness. Not only will these “grafts” be useful for retinal cell therapy, but they may also be readily combined with other retinal cell types to effectively simulate the in vivo retina. This ability will enable more accurate retinal disease models and more rapid identification of promising gene and drug therapies for retinal disease.

Methods

Methacrylated biopolymer formulation

Methacrylated Type I Bovine Collagen (Advanced BioMatrix, Carlsbad, CA) was diluted to 8 mg/mL with 20 mM acetic acid and fully solubilized by continuous orbital shaking at 4°C for ∼36 h. After solubilization, 1 mL of collagen solution was mixed with 80 μL of the provided neutralization solution to achieve a pH of 7.0–7.4. Methacrylated Porcine Gelatin (Advanced BioMatrix) was warmed to 37°C; then, it was diluted with warm (60°C), sterile 1 × phosphate buffered saline (PBS; Invitrogen, ThermoFisher Scientific, Waltham, MA) to a concentration of 200 mg/mL. The mixture was fully solubilized by continuous orbital shaking at 37°C for 5–6 h. Methacrylated Recombinant HA (Advanced BioMatrix) was diluted with sterile 1 × PBS to a concentration of 30 mg/mL and fully solubilized by continuous orbital shaking at 4°C for ∼1 h. All solubilized methacrylated biopolymers were stored at 4°C until use.

Methylene blue (Alfa Aesar, Ward Hill, MA) was used as a photoinitiator at a concentration of 0.1 wt%, a concentration achieved by mixing 9 parts of solubilized biopolymer with 1 part of premixed methylene blue solution (1 wt% in 1 × PBS) and vortex mixing to homogeneity. A 50/50 (v/v) mixture of methacrylated gelatin and HA was also investigated. Collagen and HA solutions were kept on ice during experimentation, whereas gelatin and gelatin/HA solutions were kept at 37°C.

Two-photon polymerization

In general, 2-photon polymerized structures are created by rastering a laser in 3D paths, which are determined by input (eg, a 3D model) from the user. Crosslinking reactions are restricted to the focal point of the laser, the only instantaneous location where 2 photons may collide with a photoinitiator molecule simultaneously (within 1 fs of each other) to deliver sufficient energy to cleave the molecule, thereby triggering free radical polymerization. The user may control the rastering speed (a.k.a., scanning speed) and photonic output of the laser as a percentage of maximum (a.k.a., laser power) to modulate energy dose provided to the path and thus, control kinetics and final properties.

In this work, a Nanoscribe TPP system was employed (Nanoscribe Photonic Professional GT; Nanoscribe GmbH, Eggenstein-Leopoldshafen, Germany). The system is equipped with an 80-MHz femtosecond fiber laser with a wavelength of 780 ± 10 nm and a maximum power of 120 mW. The designated structures were created by using direct-laser-writing (DLW) mode with a 25 × objective (NA = 0.8).

Glass coverslips (CS-30R; Warner Instrument, Hamden, CT) that had been prefunctionalized with methacrylate groups, as previously described,¹¹ were used as substrates for TPP. Briefly, the coverslips were functionalized by exposure to an oxygen plasma (22.5 mL/min at 30 W radio frequency power for 3 min, Plasma Cleaner PCD-011 with PlasmaFlo; Harrick Plasma, Ithaca, NY), submerged in 0.02% 3-(trimethoxysilyl)propyl methacrylate (Sigma-Aldrich, St. Louis, MO) in hexanes (ThermoFisher Scientific) for a minimum of 2 h, rinsed in hexanes, rinsed with acetone, and allowed to dry at room temperature. Before each print, a functionalized coverslip was secured onto the sample holder and a droplet of oil (Immersol^™ 518 F; Carl Zeiss, Inc., Oberkochen, Germany) was placed on the center of the nonfunctionalized side.

A droplet of methacrylated biopolymer formulation with a photoinitiator was placed in the center of the functionalized side of the coverslip, the sample holder was inserted into the TPP system, and finally fabrication was instigated according to the methods described next.

Polymerization threshold and small-scale fidelity

Arrays of 3D stars, each created at a different scanning speed and laser power, were used to determine the polymerization threshold, which represents the minimum amount of energy required for crosslinking to occur. Briefly, a 0.9-μm tall 5-pointed star with a width of 9.5 μm (linear point-to-point distance) was created in AutoCAD 2015 (Autodesk, Inc., San Rafael, CA), sliced, and hatched at a distance of 0.1 μm by using DeScribe (version 2.2.1; Nanoscribe GmbH). Slicing distance refers to the vertical distance between lateral layers in the execution code, whereas hatching distance refers to the distance between individual lines within a lateral layer.

A simple program was created by using DeScribe that called the resulting linear commands, varying the laser power in increments of 2% from 2% to 100% for stars spaced 12 μm apart; then, this process was repeated at 24 distinct scanning speeds (rastering rates), from 6,000 μm/s to 40,000 μm/s (increments of 2,000 μm/s) and from 60,000 μm/s to 160,000 μm/s (increments of 20,000 μm/s) for a total of 1,200 unique combinations of laser power and scanning speed (Fig. 1).

FIG. 1.

Schematic of star array configuration. Polymerization outcomes were evaluated at a range of scanning speeds (left) by using an array of stars. At each scanning speed, the array consisted of stars produced by using 50 different laser powers: 2%–100% in increments of 2% (middle). At the fastest scanning speed that produced reliable structures, the width of each star was measured (right) to determine which laser power produced a structure closest to the model.

After printing, samples were thoroughly washed to remove any excess photoinitiator and unpolymerized material. Using a custom rinsing apparatus, samples were placed vertically in a small beaker of 1 × PBS without stirring, and they were allowed to soak at room temperature (all threshold samples, and collagen and HA fidelity samples) or at 37°C (gelatin and gelatin/HA fidelity samples). When short-term storage was required, samples were stored in a hydrated state in 1 × PBS until use. Finally, the arrays were imaged by using an EVOS FL microscope (ThermoFisher Scientific) with a far red (Cy5) filter for the threshold experiments.

For each scanning speed, the laser power at which a polymerized structure first appeared was deemed “first dot,” whereas the laser power at which the structure first had an appearance resembling a star, as determined qualitatively, was deemed “first star.”

For fidelity experiments, samples were imaged with a Leica TCS SPE DMi8 inverted confocal microscope system (Leica Microsystems, Wetzlar, Germany) by using autofluorescence of the structures in the Cy5 channel. A z-stack of the entire height of the sample was taken by using 1-μm increments. At each scanning speed that produced reliable arrays, point-to-point distance was measured in triplicate for all stars in the set by using ImageJ (version 1.52g; NIH) to identify the laser power at which stars best resembled the modeled width. 3D reconstructions were analyzed by using ImageJ volume viewer. Specifically, at each scanning speed, the height of the star at the fidelity point was measured by taking a cross-sectional image in the x–z plane and recording the height of the star.

Scaffold prototypes: fidelity and swelling

A 3D model for a 50-μm-tall, 180-μm-wide hexagonal scaffold with hexagonally spaced vertical pores (25 μm diameter, spaced 30 μm apart) and elliptical horizontal pores to permit diffusion (25 μm vertical axis and 15 μm horizontal axis for an average diameter of 20 μm) was created in AutoCAD as previously described.¹¹ This general scaffold was designed to approximate the structure of the retinal ECM and eventually facilitate photoreceptor cell packing while enabling structural integrity of the scaffold during fabrication and processing.

For each biopolymer, the scaffold 3D model was sliced and hatched at distances of 0.5 and 1.0 μm, respectively, and fabricated with appropriate scanning speed and laser power, as determined from the threshold and fidelity experiments. After printing was complete, each scaffold sample was transferred to a 35-mm Petri dish and placed at the bottom of a large recrystallization dish, which was filled with 1 × PBS and kept at ∼37°C. Gelatin samples were rinsed for a maximum of 10 min with gentle stirring whereas gelatin/HA samples were allowed to soak, with no stirring, for only ∼1 min, or until the photoinitiator had visibly dissipated from the sample. 3D images of each scaffold were collected by using confocal microscopy as described earlier.

Vertical and horizontal pore diameter and circularity were determined by using ImageJ. Briefly, for vertical pore images, the background was removed (rolling ball radius 50 pixels), the image was converted to 8-bit mode, and the automatic threshold feature was used to create a binary image. For horizontal pores, pore edges were manually traced by using the ellipse tool. Finally, the “analyze particles” feature was used to count and measure pores, including fitting to an ellipse.

For any given pore, the average of the major and minor ellipse axes was used as the pore diameter and circularity was calculated as the ratio of minor to major axes. In a second analysis, a similar image processing technique was applied, except the binary image was inverted and the analyze particles function was used to determine scaffold area at each slice. These were multiplied by the layer thickness (1 μm) and summed over the height of the scaffold to estimate total scaffold area. A volume ratio was then determined for each scaffold by dividing the measured volume by the modeled scaffold volume (4.11 × 10⁻⁴ mm³).

Single-photon polymerization

Broad-spectrum ultraviolet (UV) polymerization was used to create biopolymer samples that were large enough to assess biocompatibility of leachables. For each sample, a small volume (<100 μL) of each biopolymer/photoinitiator solution was placed on a clean glass slide and allowed to spread into a thin, homogenous layer. The slide was then placed under the UV lamp (OmniCure^® Series 2000 UV lamp with standard filter, 320–500 nm, equipped with 8-mm liquid light guide; Excelitas Technologies, Waltham, MA) at a distance of 50 mm for the minimum amount of time required to achieve polymerization (collagen: 180 s; gelatin: 45 s; gelatin/HA: 90 s; HA: 90 s). This experiment was performed in triplicate. After polymerization, each sample was carefully removed from the glass slide by using a razor blade, sterilized by submersion in 70% ethanol, and rinsed 3 times with sterile 1 × PBS.

iPSC culture and viability

Human iPSCs (episomally derived; Gibco, ThermoFisher Scientific) were cultured with Essential 8 (E8) Flex Medium (Gibco) supplemented with 10 ng/mL human recombinant fibroblast growth factor (FGF; Gibco) and 1 mg/mL Primocin (InvivoGen, San Diego, CA) on cell culture plates coated with recombinant human laminin-coated cell culture plates at 37°C and 5% CO₂ as previously described.²⁵ The medium was replenished every 24 h, and iPSCs were passaged at a ratio of 1:6 when ∼75% confluency was reached, typically 5–7 days after the previous passage.

Round specimens of single-photon polymerized polymer (5-mm diameter) were created by using a sterile biopsy punch and incubated in iPSC media (0.5 mL per sample, in a 24-well cell culture plate) at 37°C and 5% CO₂ overnight. After 24 h, each well of this “conditioned” media was transferred to a corresponding 24-well plate of iPSCs, cultured as described earlier. After an additional 24 h, the conditioned medium was aspirated and the cells were rinsed with HBSS. Fresh medium was then added along with 1 drop of the Live/Dead^® Viability/Cytotoxicity kit (Molecular Probes, Invitrogen, ThermoFisher Scientific) and incubated for 15 min before imaging on an EVOS FL microscope. Within each well, 3 locations were assessed by collecting 3 images at each: 1 with a DAPI filter (live cell nuclei), 1 with a GFP filter (dead cell nuclei), and 1 overlay images with both DAPI and GFP.

Viability was determined quantitatively by using image analysis (ImageJ) of the 2 fluorescent channels independently. Briefly, the background of each image was removed, then the image was converted to an 8-bit binary format by using the automatic threshold function, the image was inverted, and the watershed function was applied to distinguish individual nuclei. Finally, “analyze particles” were used to determine the total number of nuclei visible in each image, with a lower limit of 40 μm² so as to exclude background noise and non-nuclear debris. Viability was taken as the number of live cells (blue—green) divided by the total number of cells (blue).

Two-photon polymerized scaffold prototypes: biocompatibility

A short, rounded scaffold was constructed as described earlier for the purpose of evaluating biocompatibility. These scaffolds were 10 μm tall, 180 μm in diameter, with no horizontal cross-pores and identical vertical pores to the 50-μm-tall scaffolds. The geometry of the scaffold was changed from hexagonal to circular to reduce sharp edges and corners, which were anticipated to be unfavorable for cells. The model was sliced, hatched at distances of 0.5 and 0.15 μm, respectively, and printed by using the same scanning speed and laser power as the prototype scaffolds. An array of 32 of these short scaffolds were printed in a square-packed formation across a total, roughly circular area with an approximate diameter of 2 mm. After completing the scaffold washing process, as described earlier, the samples were submerged in 1 × PBS in a sterile 6-well plate for storage.

The samples were subsequently sterilized by first immersing in 35% ethanol for 1 h and then 70% ethanol for an additional hour. Glass cloning rings (6-mm outer diameter, 4-mm inner diameter; Sigma-Aldrich) were also sterilized with 70% ethanol for at least 1 h. After sterilization, cloning rings were adhered to the sample substrates by using sterile vacuum grease.

After mounting the cloning rings, each coverslip was placed in 1 well of a 6-well plate, 30 μL of 1 × HBSS with phenol red was added inside of the cloning ring on top of the scaffolds, and 3 mL of 1 × PBS was added to the surrounding well to prevent sample dehydration. The HBSS contains phenol red, which visibly contrasted with the colorless PBS; if the vacuum grease seal was incomplete, evidence of dilution in color was easily discernable and the seal was carefully reinforced. The HBSS was then gently aspirated, and a dilute solution of Matrigel [Corning, Corning, NY; 1 mL in 50 mL α-MEM (ThermoFisher)] was used to coat the scaffolds at room temperature for 30 min. The Matrigel solution was removed from the scaffolds immediately before cells were added.

Rat postnatal retinal cell isolation and culture

All animal experiments were conducted in accordance with the ARVO Statement for the Use of Animals and Ophthalmic and Vision Research and with the approval of the University of Iowa Animal Care and Use Committee. Retinal cells were harvested from 11 Sprague-Dawley littermates at postnatal day 1 (P1). Immediately postmortem, whole eyes were retrieved by using curved forceps. The retinae were subsequently dissected away and placed into 12 mL of 1 × HBSS in a 15-mL conical tube.

After centrifugation and supernatant removal, the retinal tissue was enzymatically disassociated by incubation in a 2-mL solution of 1:1 collagenase and 0.25% Trypsin/EDTA for 30 min at 37°C, with agitation (gentle manual shaking) every 5–10 min. Two mL biopsy media [10% fetal bovine serum (ThermoFisher) and 1% Primocin in α-MEM (ThermoFisher)] was then added to the solution to inactivate the trypsin, and the solution was triturated with a sterile glass pasture pipette. The solution was then filtered through a cell strainer (40 μm mesh; Corning) to separate the cells from any remaining ECM and viscous, extracellular DNA. The remaining solution was then centrifuged at 1,200 rpm for 5 min, and the cells were resuspended in biopsy media to a concentration of 1 million cells per mL. Excess cells were frozen in cryopreservation media (Synth-a-Freeze; ThermoFisher Scientific) and stored at −80°C.

Retinal cell viability and interactions with two-photon polymerized scaffolds

Day 1 postnatal rat retinal cells were retrieved from frozen storage, thawed, and resuspended in biopsy media. Cells were centrifuged for 5 min at 1,200 rpm to remove dimethyl sulfoxide (DMSO), resuspended in biopsy media, and seeded onto the scaffolds and a glass coverslip control that had been prepared for TPP (ie, functionalized to bind biopolymers) but did not carry any scaffolds and never came in contact with biopolymers at a concentration of 2 million cells/mL. Cells were cultured in biopsy medium, which was changed daily, and after 3 days, a live/dead assay was performed on the cells as described earlier. The medium was changed immediately before conducting the live/dead assay.

Immediately after isolation, postnatal rat retinal cells were plated at a concentration of 1 million cells/mL on biopolymer scaffold samples. Cells were cultured in biopsy medium, which was changed daily. Sample analysis was performed at 2 time points: 3 and 7 days. At the specified time point, the cloning ring was removed from the sample coverslip, the remaining solution was aspirated, and the vacuum grease was carefully removed with a cell scraper. Cells were fixed by using 1 mL of 4% paraformaldehyde (PFA) at 4°C for 30 min. The PFA was then removed, and samples were rinsed 3 times with 1 × PBS for 5 min. A hydrophobic pen (Research Products International) was used to demarcate the cell area and create a small confinement volume for antibody solution. Samples were allowed to react overnight at 4°C with Tuj-1 primary antibody (anti-Tuj-1, neuron-specific class III β-tubulin, raised in rabbit; Sigma-Aldrich) at a ratio of 1:1,000 in a 5% donkey serum with 3% bovine serum albumin, 3% sodium azide, and 0.5% Triton X in 1 × PBS.

The next day consisted of 5 washing steps before the addition of the secondary antibody. Each washing step comprised a high-volume wash in 40 mL of 0.1% Tween in 1 × PBS contained within an 8 × 12 cm plastic reservoir for 10–15 min each. Samples were then reacted with secondary antibody (AlexaFluor 568 donkey anti-rabbit; ThermoFisher Scientific) at 1:1,000 in the same serum solution described earlier at room temperature for at least 2 h. Another 2–3 washing steps were conducted and finally, samples were mounted by using mounting media with DAPI (Shandon AquaMount; ThermoFisher) and a 22 by 60 mm coverslip. Aluminum foil was used to cover the samples throughout the procedure to minimize photobleaching. Labeled sections were visualized by using confocal microscopy, as described earlier.

Statistical analyses

Before statistical analysis, pore geometry, scaffold volume, and viability datasets were each subjected to a D'Agostino and Pearson normality tests at a confidence interval of 95%. In each case, several of the datasets were not normally distributed. First, conventional Box-Cox transformations were applied with optimal delta (as identified using Minitab version 18.1). If this transformation did not result in a normally distributed dataset, the outliers in the original dataset were identified by using either the ROUT method with Q = 1% or Grubb's method at a 95% confidence interval (leachables viability data; 3 data points removed: 1 from each of 3 datasets).

Nonparametric methods were employed only if these adjustments did not result in a normally distributed dataset. One-sample t-tests or Wilcoxon signed-rank tests (nonparametric equivalent) were used to compare scaffold characteristics (pore diameter and circularity, scaffold volume) with expected values, 2-sample t-tests, or Mann-Whitney tests (nonparametric equivalents were used to compare 2 datasets, whereas 1-way analysis of variance followed by Tukey's multiple-comparisons tests were used to assess differences between more than 2 groups [non-parametric tests were not required for this analysis type]). All statistical tests were performed at a confidence interval of 95%.

Results

TPP settings were first established for each of 4 biopolymer formulations by iteratively varying scanning speed and laser power and assessing resulting polymerization outcomes (Table 1; Fig. 2). The minimum scanning speed required for TPP was highest for methacrylated collagen (26,000 μm/s) compared with other biopolymer formulations, which were all able to be crosslinked at the slowest scanning speed tested (6,000 μm/s) (Fig. 2A). For the collagen formulation in particular, the minimum laser power needed for polymerization decreased with faster scanning speed (Fig. 2A). The HA formulation printed the least consistently, but fabrication was feasible across a broad range of scanning speeds (Fig. 2C). Both gelatin and the gelatin/HA mixture polymerized consistently across a broad range of scanning speeds at relatively low laser powers: 10%–50% for gelatin alone and ∼10%–25% for gelatin/HA (Fig. 2B, D). However, at scanning speeds above 36,000 μm/s, some structures did form, then later conglomerated, and finally delaminated from the coverslip, preventing their characterization.

FIG. 2.

Two-photon polymerization threshold. The minimum laser power required to produce structures from 4 methacrylated biopolymer formulations: (A) collagen; (B) gelatin; (C) HA; and (D) 50/50 gelatin/HA. For any given scanning speed, “first dot” (light gray) represents the lowest laser power at which any polymerized structure was observed and “first star” (dark gray) represents the lowest laser power at which the structure qualitatively resembled a star. Each data point represents the mean, whereas error bars represent standard deviation. HA, hyaluronic acid.

Table 1.

Number of Replicates (n) Measured for Threshold Analysis

Scanning speed (μm/s)	Collagen	Gelatin	Gelatin/HA	HA
6,000		3	3	1
8,000		3	3	2
10,000		3	2	2
12,000		3	3	2
14,000		3	3	2^a
16,000		3^a	3	2^a
18,000		3	3	1
20,000		3	3	2
22,000		3	3	1
24,000		2	2	2
26,000	1	2	2	2
28,000	1	1	1	2
30,000	2	1		1
32,000	2^a	1		1
34,000	3^a	1
36,000	3	1
38,000	3
40,000	3

Unless otherwise indicated, the number of replicates for “first dot” and “first star” was the same (Fig. 2).

Indicates that “first star” data had 1 less measurement than this value.

HA, hyaluronic acid.

For all 4 biopolymer formulations, the use of scanning speeds greater than roughly 40,000 μm/s resulted in distorted, delaminated star structures (data not shown). Due to their inconsistent TPP, collagen and HA formulations were not included in some subsequent analyses.

For 2 of 3 replicates, the gelatin formulation reached its fidelity point (where the star width matched the width of the modeled star) at slow scanning speeds (6,000–18,000 μm/s) between 25% and 50% laser power (Fig. 3A). From 20,000 to 24,000 μm/s, none of the stars reached their fidelity point, whereas the fidelity point was between 25% and 60% laser power at fast scanning speeds (26,000–40,000 μm/s) for 2 of 3 samples (Fig. 3A). Conversely, HA/gelatin stars did not reach their fidelity point consistently until scanning speeds of 36,000 μm/s and above, with laser powers between 30% and 60%. Both biopolymer formulations printed consistently at the fastest feasible scanning speed, 40,000 μm/s. At this scanning speed, fidelity was reached at 42% laser power for gelatin and 36% laser power for the HA/gelatin mixture; these printing parameters were used to fabricate scaffolds for the subsequent experiments.

FIG. 3.

Two-photon polymerization small-scale fidelity. Quantitative values and representative star arrays are shown for 2 methacrylated biopolymer formulations: (A) gelatin and (B) 50/50 gelatin/HA. At each scanning speed, the lowest laser power at which stars first reached the modeled point-to-point distance was recorded as the fidelity point. If none of the stars in an array reached the fidelity point, an open shape was used to represent the laser power at which the star was closest to the desired value. The scale bar in each inset represents 10 μm.

After determining optimal printing parameters, prototype scaffolds composed of gelatin or gelatin/HA were created to better understand scale-up and the relationship between formulation and scaffold characteristics. Regardless of formulation, scaffold width and vertical pore size qualitatively appeared as expected at the bottom of each scaffold (Fig. 4A, B). Open, horizontal cross-pores were evident throughout the scaffold (Fig. 4A, B). However, as fabrication progressed, overall scaffold width increased and vertical pores became less circular in appearance (Fig. 4A, B, insets).

FIG. 4.

Two-photon polymerized biopolymer microscaffolds. 3D representations of (A) gelatin and (B) 50/50 gelatin/HA scaffolds as seen from the bottom, with sub-volumes depicting the fabrication of the scaffold in 5, 25-μm-tall segments. (C) Vertical pore diameter, (D) vertical pore circularity, (E) horizontal cross-pore diameter, and (F) horizontal cross-pore circularity of both scaffolds. In (C–F), the dashed lines represent the modeled or expected value. In (D, F), gray ellipses characterize the circularity of a given pore at intervals of 0.25. Box and whisker plots represent the median and upper and lower quartiles, with error bars representing minimum and maximum values. ***P < 0.01, ****P < 0.001. 3D, three-dimensional.

Gelatin/HA scaffolds contained small, punctate inhomogeneities throughout (Fig. 4B, inset). The gelatin scaffolds had an average vertical pore diameter close to the model, with a mean of 24.2 μm, whereas the gelatin/HA samples had an average vertical pore diameter more than 10% smaller than gelatin, at 21.5 μm (P < 0.0001, Fig. 4C and Table 2). Both formulations resulted in scaffolds with similar vertical pore circularity, with means near 0.75 (Fig. 4D and Table 2). Horizontal cross-pores also had average diameters close to the model: 22.7 μm for gelatin and 21.1 μm for gelatin/HA, which again was significantly lower than that for gelatin (7% difference, P < 0.001, Fig. 4E and Table 2). Regardless of formulation, the mean circularity of the horizontal cross-pores was ∼0.85 (Fig. 4F and Table 2).

Table 2.

Number of Replicates and P Values for Mann-Whitney t-Tests Comparing Vertical and Horizontal Pore Characteristics Between Gelatin and 50/50 Gelatin/Hyaluronic Acid 2-Photon Polymerized Scaffolds

	Replicates (n)		P value (Gel vs. Gel/HA)
	Gelatin	Gelatin/HA	Diameter	Circularity
Vertical pores	502	381	<0.0001	0.1817
Horizontal pores	108	70	<0.001	0.1031

Significant values (P < 0.05) are shown in bold.

Given the important role of hydration in the native ECM and the known tendency of gelatin and HA to act as hydrogels, we performed several analyses to characterize the swelling of 2-photon polymerized gelatin and gelatin/HA. For very small structures (stars), feature height exceeded the anticipated height several times over for all formulations (Fig. 5A) and all scanning speeds. This swelling was mostly dependent on biopolymer composition and independent of scanning speed. Gelatin exhibited the most swelling with an average height of 15 μm, roughly 15 times the modeled height. Gelatin/HA demonstrated the second highest amount of swelling with an average star height of 12 μm. Meanwhile, collagen and HA displayed the least amount of swelling, with average heights of 9 and 5 μm, respectively. Swelling was also observed in the scaffolds; the average height of gelatin and gelatin/HA scaffolds was around 100 μm (compared with the expected 50 μm), and expansion also occurred laterally (Fig. 4A, B). The mean volume ratio of the scaffolds was ∼3.1 for both biopolymers, with individual values ranging from 1.5 to almost 5 (Fig. 5B).

FIG. 5.

Biopolymer swelling. (A) The height of the fidelity point star at each scanning speed. The mean height value is plotted, and the error bars represent standard deviation. (B) The volume ratio is the ratio of the volume of the printed scaffold to the volume of the modeled scaffold. Each experimental replicate is plotted, and the error bars represent standard deviation. In (A, B), dashed lines represent the modeled or expected values.

To assess the cytotoxicity of the crosslinked biopolymers, we incubated samples from each formulation in cell culture media (“conditioned”) and exposed iPSCs to the resulting solution for 24 h. All 4 biopolymer formulations caused only minimal toxicity (viability was >90%) to human iPSCs (Fig. 6). Interestingly, the viability of the gelatin group was significantly less than the control (P < 0.05, Table 3); however, cells exposed to a combination of gelatin and HA closely resembled and even surpassed the average viability of the control group, and this was significantly higher than gelatin alone (P < 0.05, Table 3). Meanwhile, the average viability of cells exposed to conditioned media from HA alone was not significantly different than the control.

FIG. 6.

Compatibility of biopolymer leachables with pluripotent stem cells. (A) Relative viability of cells exposed to leachables from 4 UV-polymerized biopolymer formulations, compared with control (tissue culture plastic). Each data point (gray or red) represents an individual measurement whereas solid lines and error bars (black) represent the mean and standard deviation, respectively, for each group, with n ≥ 8, *P < 0.05. Data points shown in red are derived from the images in (B–F), which are meant to serve as representative images of the live/dead assay effectiveness for each formulation: (B) control; (C) collagen; (D) gelatin; (E) 50/50 gelatin/HA; and (F) HA. Blue and green fluorescence represents live and dead cells, respectively. Each scale bar represents 400 μm. UV, ultraviolet.

Table 3.

Adjusted P Values for t-Tests (Above Dotted Line) Comparing Average Viability with a Perfect Value (100%) and for Tukey's Post Hoc Analysis (Below Dotted Line) Comparing Mean Viability Across Different Biopolymers and the Control Group

Compared with	Formulation (n)
Compared with	Control (8)	Collagen (8)	Gelatin (8)	Gelatin/HA (9)	HA (9)
100%	0.0030	0.0078	0.0093	0.0603	0.0176
Control		0.8386	0.0238	0.9999	0.1236
Collagen	0.8386		0.2253	0.7396	0.6366
Gelatin	0.0238	0.2253		0.0125	0.9258
Gelatin/HA	0.9999	0.7396	0.0125		0.0744
HA	0.1236	0.6366	0.9258	0.0744

Significant values (P < 0.05) are shown in bold.

Retinal progenitor cells were isolated from postnatal rats and cultured on gelatin and gelatin/HA microscaffolds for up to 1 week. After 3 days, ∼88% of retinal cells were viable when plated on the control substrates (glass coverslip coated with Matrigel) (Fig. 7). In comparison, culturing retinal cells on 2-photon polymerized gelatin scaffolds significantly reduced viability (to 57%, P < 0.01, Fig. 7). However, incorporation of HA in the formulation mitigated cell death; retinal cells were 87% viable when cultured on gelatin/HA scaffolds, which was not significantly different from the control but was significantly higher than the viability of cells cultured on gelatin-alone scaffolds (P < 0.05, Fig. 7).

FIG. 7.

Compatibility of biopolymer microscaffolds cultured with retinal progenitor cells. Percent viability of primary postnatal rat retinal cells after 3 days of being cultured on 2-photon polymerized gelatin or 50/50 gelatin/HA scaffolds. Each data point represents an individual measurement whereas solid lines and error bars represent the mean and standard deviation, respectively, for each group, with n ≥ 8, *P < 0.05, and **P < 0.01.

Retinal cells adhered to both gelatin and gelatin/HA scaffolds and could also be found in and around the scaffolds (Fig. 8). Regardless of their substrate, a high proportion of these cells were neuronal, as expected for cells isolated from the postnatal retina. Qualitatively compared with gelatin scaffolds, the HA/gelatin scaffolds supported a greater number of cells, which appeared to also express more Tuj-1 and have more clearly distinguished neuronal morphology than gelatin scaffolds (Fig. 8D, G).

FIG. 8.

Retinal progenitor cell interactions with biopolymer microscaffolds. Immunofluorescent images of primary postnatal rat retinal cells after 3 days (A–D) or 7 days (E–H) of being cultured on 2-photon polymerized gelatin (A, E), 50/50 gelatin/HA (B, C, F, G) scaffolds, or a glass control with no scaffolds (D, H). Blue represents cell nuclei (DAPI) and red represents Tuj-1, a pan-neuronal marker. Scaffolds autofluoresce at both excitation wavelengths and thus, appear purple. Each 3D rendering (A, B, E, F) represents a 250-μm square with the stated thickness. Insets show 4 equal sub-volumes through the thickness of the z-stack at the area designated by the yellow dotted box. Higher magnification images (C, D, G, H) show neuronal cells with characteristic morphology interacting with 50/50 gelatin/HA scaffolds (C, G) and on the control surface (D, H). Scale bars represent 50 μm.

Discussion

As noted earlier, 2-photon polymerized collagen structures could not be fabricated consistently at scanning speeds below 26,000 μm/s. This behavior contrasts to the expected trend: that slowing scanning speed and increasing laser power will lead to a greater likelihood of polymerization. Based on our observations of collagen structures created below 26,000 μm/s, we hypothesize that this contradiction is due to over-polymerization at slow scanning speeds. In our experience, accumulating energy from prolonged exposure to the laser focal point can cause solvent evaporation in the laser path during presumed polymerization. This phenomenon includes the creation of vapor bubbles that diffract light and prevent TPP from proceeding.

Further, radiating heat can trigger initiation outside of the focal point, leading to imprecise polymerization and indistinguishable features. Indeed, the maximum recommended concentration of methacrylated collagen was lower than the other 2 methacrylated biopolymers (∼7.5 mg/mL compared with 30 and 200 mg/mL), so it may have been more prone to evaporative effects than its more concentrated counterparts. We recommend a future emphasis on comparing methacrylated biopolymers at consistent concentrations and on understanding the effect of biopolymer concentration on TPP behavior.

In general, rinsing the samples at the appropriate temperature for unreacted biopolymer solubility resulted in higher quality samples and thus more reliable analysis. Still, stars printed above 40,000 μm/s retained their distorted shapes (data not shown). This phenomenon has also been observed when using a variety of photopolymers, including acrylated polycaprolactone¹² and even commercially available resists optimized for the Nanoscribe system (unpublished data). In fact, this behavior can be explained by the short line distance used to fill small structures, combined with the time required for the system to change directions by 180° between lines. In other words, as the laser “turns around,” there is a mismatch of lines at the edges, creating the illusion that 2 identical structures are being formed. When printing is complete, the small structures simply appear distorted and frayed. For larger structures, these edge effects are not noticeable, and the small distortions are negligible in comparison to the whole.¹¹

Gelatin stars did not reach the fidelity point at scanning speeds between 20,000 and 24,000 μm/s regardless of laser power, which was an unexpected observation (Fig. 3). This phenomenon may be explained by the balance between over-polymerization, crosslinking density, and swelling. For example, at slow scanning speeds, high photon doses from the slower-moving laser can cause over-polymerization due to reactive light diffusion and heating, as described earlier. This behavior could inflate the size of the 2-photon polymerized structures such that they reach the fidelity point consistently in this range. Conversely, a faster-moving laser (at fast scanning speeds) provides a lower photon dose and theoretically, a lower crosslinking density than at slow scanning speeds.¹² The inverse relationship between crosslinking density and water uptake is well known; these structures are expected to expand as they absorb water during post-processing and thus also reach the fidelity point (and beyond) in this scanning speed range. On the other hand, we hypothesize that the scanning speed range in question (20,000–24,000 μm/s), where the fidelity point is not reached within the bounds of the experiment, represents a balanced region where neither over-polymerization nor swelling dominates the final structure characteristics. Fabricating large structures in this narrow region may produce more repeatable structures than at slower or faster scanning speeds, though this concept should be validated and explored in future studies.

Given that the gelatin/HA mixture did not reach the fidelity point until relatively fast scanning speeds (Fig. 3), this may indicate that adding HA to the formulation modulates both over-polymerization and swelling. Indeed, stars fabricated from HA alone were substantially (more than 50%) shorter than their gelatin counterparts at the same scanning speeds. Thus, combining the 2 biopolymers resulted in a consistent, ∼20% decrease in star height compared with gelatin alone (Fig. 5). However, scaffold volume ratio was analogous between the 2 groups; perhaps due to the influence of the larger slicing and hatching distances used for scaffolds compared with stars, which may have alleviated the formulation's influence on swelling.

The open structure of horizontal cross-pores in both gelatin and gelatin/HA scaffolds is encouraging (Fig. 4), since these channels are expected to enable cell–cell interactions within the scaffold and to facilitate free diffusion of nutrients or signaling molecules. Their retention in the final structure is an improvement over our own previous design and results in 2-photon polymerized poly(caprolactone) scaffolds.¹²

For both gelatin and gelatin/HA, scaffold width increased from bottom to top (Fig. 4). We hypothesize that this is due to photon absorption along the laser path length as light travels to the focal point. Based on the concentration of methylene blue in the solution and its molar extinction coefficient, only about one-third of the incident light intensity is expected to reach a focal point located 50 μm into the solution droplet (estimated by using the Beer-Lambert law). This light attenuation decreases perceived laser power and thus, is less likely to cause crosslinking, resulting in a decrease in crosslinking density. Thus, layers at the top of the scaffold likely swell more than their less-attenuated counterparts at the bottom of the scaffold. Given the logarithmic relationship between concentration and relative light transmission, even small decreases in photoinitiator concentration in future experiments would likely temper this behavior and absolve the scaffolds of this unusual phenomenon.

Abnormalities in vertical pore shape were also noted as fabrication progressed from bottom to top (Fig. 4). We hypothesize that the high-volume horizontal cross-pores caused a lack of structural support within these hydrogel scaffolds. Gravity thus caused the thin, internal scaffold pillars to lean slightly as they were forming and before completion of the horizontal cross-pore edge. This resulted in scaffolds with high fidelity to the model at the base with a misshapen structure on the top, which was more pronounced as the scaffold increased in height.

Gelatin scaffolds also had larger vertical and horizontal pore diameters than the gelatin/HA samples (Fig. 4). In the case of vertical pores, this corresponded to gelatin scaffolds having higher fidelity to the modeled pore size than gelatin/HA scaffolds, whereas the inverse was true for horizontal pores. The relationship between biopolymer formulation and pore diameter is an important consideration since ease of cell loading and cell density within the scaffold could be predicted or controlled by using pore size. Although the pore size differences observed here were statistically significant, likely due to a very high number of measurements and thus very powerful analysis, their functional meaning may not be substantial. For example, the 3D model may easily be adjusted to compensate for slight differences in pore size once the baseline fidelity is known.

However, gelatin/HA scaffolds were also not as repeatable as gelatin-alone scaffolds: Out of 9 total scaffolds attempted in this study, only 6 complete and 2 partial gelatin/HA scaffolds were achieved. Gelatin, on the other hand, consistently printed all scaffolds. Moreover, gelatin/HA scaffolds appeared to contain small pockets of inhomogenous material (Fig. 4). Given that this formulation was a 2-component system, we anticipate that this morphology was the result of microscale phase separation. Before polymerization, the gelatin/HA mixture appeared homogenous, and no phase separation was noted. However, the dynamic environment that occurs during photopolymerization has been known to induce phase separation and disruptions in anticipated polymer architecture.^26,27 These previous studies suggest that tuning polymerization rate could alleviate phase separation effects, but this theory should be investigated in future studies, particularly if additional components are added to the formulation.

The cell viability data we collected suggest that methacrylated gelatin and gelatin/HA biopolymer formulations are appropriately biocompatible for use in in vitro retinal tissue engineering. Interestingly, iPSCs and postnatal retinal cells were each more viable after being exposed to gelatin/HA biomaterials compared with gelatin alone (Figs. 6 and 7).

Further, despite efforts to rigorously wash the scaffolds after fabrication, we continually observed a higher amount of blue color in media conditioned with gelatin and in gelatin scaffolds than in their gelatin/HA counterparts, indicating that methylene blue may not have been completely rinsed from the system. However, since we suspect that gelatin samples had a lower crosslinked density than the other groups, diffusion of methylene blue out of the polymer would theoretically be less inhibited than for other formulations. Although these data support the argument that methylene blue is instigating cytotoxicity, this photoactive compound is also widely used for several therapeutic and diagnostic applications and is known to have relatively low systemic toxicity.²⁸ Regardless, a smaller photoinitiator concentration should be considered or a more rigorous washing protocol developed to minimize photoinitiator and unpolymerized material left in the scaffolds.

Postnatal retinal cells interacted uninhibited with 2-photon polymerized biopolymer microscaffolds; cells were observed on top of the scaffolds, surrounding the scaffolds, and inside of the scaffold pores (Fig. 8). These observations are an encouraging sign of these materials' biocompatibility. In addition, retinal cells continued to express Tuj-1 through 7 days in culture. This continued expression of a neuron-specific marker, which was particularly poignant in the gelatin/HA samples, suggests that the scaffolds enabled postnatal retinal cells to maintain their relatively sensitive fate.

One limitation of this study, and most similar studies, is that post-mitotic photoreceptor cells were not used to demonstrate compatibility. However, post-mitotic photoreceptor cells are very sensitive to isolation protocols and do not proliferate (by definition) after such a procedure, dampening their potential for use in retinal tissue modeling. Further, mature, post-mitotic photoreceptor cells are unlikely to be feasible for use in therapeutic applications for the same reasons and due to the high cost and long times required for their relatively low-yield maturation and isolation from culture.^2,4,29 Instead, retinal progenitor cells and photoreceptor precursor cells derived from pluripotent stem cells (either embryonic stem cells or iPSCs) are increasingly used for retina tissue modeling and cell replacement therapy.^2–9,29 Thus, maintenance of neural retinal fate is a critical characteristic of scaffolds used for retinal tissue-engineering materials, since these mitotic cell types can in some cases unintentionally revert to a more potent state or fail to differentiate into mature retinal phenotypes if not provided the correct mechanical or biochemical ques.

In fact, one recent study has shown that encapsulating pluripotent stem cell-derived optic vesicles in hydrogels affects retinal differentiation, which is dependent on the hydrogel composition,³⁰ suggesting that this will likely be the case for scaffolds as well. Further, disruption of key ECM molecules during in vivo development is known to cause structural anomalies and cellular dysfunction, further highlighting the important role that the matrix plays in retinal and photoreceptor differentiation and organization.³¹ However, the retinal ECM is also known to be dynamic during development,^31,32 complicating the identification of the ideal scaffold composition for developing retinal cells in vitro.

Although the biopolymers we investigated are all known components of the retinal ECM in some form (gelatin being generic hydrolyzed collagen),^21,31,33 none of them alone would adequately recapitulate the in vivo retinal ECM. Rather, this work demonstrates the advantages of creating a custom formulation for a particular application. For example, methacrylated HA could not easily form 2-photon polymerized structures on its own, whereas methacrylated gelatin formed very robust and predictable structures. Further, both HA alone and gelatin demonstrated sub-optimal biocompatibility with iPSCs and retinal cells. When combined, however, TPP was feasible, viability was not significantly different than the controls, and retinal neuronal fate seemed to be promoted.

Various collagens are known to play structural roles in the general retinal ECM,³⁴ whereas glycosaminoglycans such as HA and chondroitin sulfate are the major components of the IPM, which has very little, if any, collagenous matrix.^21,33,35 In addition, recent work suggests that HA, in particular, could contribute to regeneration of the retina in non-mammalian species (ie, axolotl),²⁸ Our findings further support the concept that HA plays a supportive and possibly pro-regenerative role for early retinal cells. This indicates that combined with robust biopolymer-based structural components when needed, HA's incorporation into scaffolds for retinal tissue engineering is beneficial and desirable.

In this study, we selected only commercially available methacrylated ECM biopolymers. However, many labs, including our own, have demonstrated successful (meth)acrylation of a wide range of biopolymers in academic labs,¹⁷ so 3D-printed scaffold composition need not be limited to combinations of only these 3 options. For example, adhesive glycoproteins, such as a few key laminin heterotrimers^32,36 and fibronectin,³³ as well as several proteoglycans, such as versican and brevican,³⁷ are also known to be present and important in the IPM. Thus, future work functionalizing these molecules with cross-linkable groups or targeting their incorporation in retinal tissue scaffolds may enhance model accuracy or even be used to reveal currently unknown disease phenotypes related to ECM mutations.

We anticipate that tailoring the material formulation to approximate the biochemical composition of the retinal ECM, particularly the IPM, will become best practice in the field of retinal tissue modeling. Here, we demonstrate that this concept is feasible and desirable; the 50/50 mixture of gelatin and HA in this study produced structures that were either comparable or superior to their homogeneous counterparts, in terms of both accuracy and biocompatibility. Thus, future studies can use the basic approach we have established here to identify appropriate settings for TPP for various mixtures and to continue to incorporate additional ECM molecules in the formulation to more closely replicate the IPM.

Further, quantifying the effect of crosslinking on the availability of cell binding sites can and should be pursued. If needed, alternative approaches should be pursued, such as molecular entrapment, homogenous surface coating, or, more precise, chemistry to ensure that sufficient integrin binding can occur. Moreover, fully validating the utility of high-resolution 3D-printed biopolymers using human retinal progenitor cells and photoreceptor precursors, as well as any other phenotypes that may be relevant for in vitro study, will be a critical step toward their implementation in retinal disease modeling and therapeutic testing.

Genome-editing and patient-specific approaches have already spurred rapid advances in understanding retinal disease. We expect that the tissue-engineering strategy we describe here will complement these existing strengths in the field by offering additional points of control for the in vitro model's microenvironment. In time, we anticipate that the enhanced accuracy of these in vitro retinal models will lead to better understanding and thus more effective treatment of a wide range of retinal diseases that are currently incurable.

Footnotes

Acknowledgments

The authors would like to acknowledge Mr. Yassine Filali for his technical assistance with some data analysis. The authors are also grateful for financial support from the Roy J. Carver Charitable Trust (Grant No. 18-5045), Research to Prevent Blindness, the International Retinal Research Foundation, and the Howard F. Ruby Endowment for Human Retinal Engineering.

Author Disclosure Statement

No competing financial interests exist.

References

Burnight

E.R.

, Gupta

, Wiley

L.A.

, Anfinson

K.R.

, Tran

, Triboulet

, Hoffmann

J.M.

, Klaahsen

D.L.

, Andorf

J.L.

, Jiao

, Sohn

E.H.

, Adur

M.K.

, Ross

J.W.

, Mullins

R.F.

, Daley

G.Q.

, Schlaeger

T.M.

, Stone

E.M.

, and Tucker

B.A.

Using CRISPR-Cas9 to generate gene-corrected autologous iPSCs for the treatment of inherited retinal degeneration. Mol. Ther. 25:1999–2013, 2017.

Burnight

E.R.

, Giacalone

J.C.

, Cooke

J.A.

, Thompson

J.R.

, Bohrer

L.R.

, Chirco

K.R.

, Drack

A.V.

, Fingert

J.H.

, Worthington

K.S.

, Wiley

L.A.

, Mullins

R.F.

, Stone

E.M.

, and Tucker

B.A.

CRISPR-Cas9 genome engineering: treating inherited retinal degeneration. Prog. Retin. Eye Res. 65:28–49, 2018.

Achberger

, Haderspeck

J.C.

, Kleger

, and Liebau

Stem cell-based retina models. Adv. Drug Deliv. Rev. 140:33–50, 2018.

Wiley

L.A.

, Burnight

E.R.

, DeLuca

A.P.

, Anfinson

K.R.

, Cranston

C.M.

, Kaalberg

E.E.

, Penticoff

J.A.

, Affatigato

L.M.

, Mullins

R.F.

, Stone

E.M.

, and Tucker

B.A.

cGMP production of patient-specific iPSCs and photoreceptor precursor cells to treat retinal degenerative blindness. Sci. Rep. 6:30742, 2016.

Sharma

T.P.

, Wiley

L.A.

, Whitmore

S.S.

, Anfinson

K.R.

, Cranston

C.M.

, Oppedal

D.J.

, Daggett

H.T.

, Mullins

R.F.

, Tucker

B.A.

, and Stone

E.M.

Patient-specific induced pluripotent stem cells to evaluate the pathophysiology of TRNT1-associated Retinitis pigmentosa. Stem Cell Res. 21:58–70, 2017.

Parfitt

D.A.

, Lane

, Ramsden

C.M.

, Carr

A.J.

, Munro

P.M.

, Jovanovic

, Schwarz

, Kanuga

, Muthiah

M.N.

, Hull

, Gallo

J.M.

, da Cruz

, Moore

A.T.

, Hardcastle

A.J.

, Coffey

P.J.

, and Cheetham

M.E.

Identification and correction of mechanisms underlying inherited blindness in human iPSC-derived optic cups. Cell Stem Cell. 18:769–781, 2016.

Capowski

E.E.

, Wright

L.S.

, Liang

, Phillips

M.J.

, Wallace

, Petelinsek

, Hagstrom

, Pinilla

, Borys

, Lien

, Min

J.H.

, Keles

, Thomson

J.A.

, and Gamm

D.M.

Regulation of WNT signaling by VSX2 during optic vesicle patterning in human induced pluripotent stem cells. Stem Cells. 34:2625–2634, 2016.

Bell

W.R.

, Meeker

A.K.

, Rizzo

, Rajpara

, Rosenthal

I.M.

, Flores Bellver

, Aparicio Domingo

, Zhong

, Barber

J.R.

, Joshu

C.E.

, Canto-Soler

M.V.

, Eberhart

C.G.

, and Heaphy

C.M.

A unique telomere DNA expansion phenotype in human retinal rod photoreceptors associated with aging and disease. Brain Pathol. 29:45–52, 2018.

Ueda

, Onishi

, Ito

S.I.

, Nakamura

, and Takahashi

Generation of three-dimensional retinal organoids expressing rhodopsin and S- and M-cone opsins from mouse stem cells. Biochem. Biophys. Res. Commun. 495:2595–2601, 2018.

10.

Jung

Y.H.

, Phillips

M.J.

, Lee

, Xie

, Ludwig

A.L.

, Chen

, Zheng

, Kim

T.J.

, Zhang

, Barney

, Min

, Barlow

, Gong

, Gamm

D.M.

, and Ma

3D microstructured scaffolds to support photoreceptor polarization and maturation. Adv. Mater. 30:e1803550, 2018.

11.

Worthington

K.S.

, Wiley

L.A.

, Kaalberg

E.E.

, Collins

M.M.

, Mullins

R.F.

, Stone

E.M.

, and Tucker

B.A.

Two-photon polymerization for production of human iPSC-derived retinal cell grafts. Acta Biomater. 55:385–395, 2017.

12.

Thompson

J.R.

, Worthington

K.S.

, Green

B.J.

, Mullin

N.K.

, Jiao

, Kaalberg

E.E.

, Wiley

L.A.

, Han

I.C.

, Russell

S.R.

, Sohn

E.H.

, Guymon

C.A.

, Mullins

R.F.

, Stone

E.M.

, and Tucker

B.A.

Two-photon polymerized poly(caprolactone) retinal cell delivery scaffolds and their systemic and retinal biocompatibility. Acta Biomater. 94:204–218, 2019.

13.

Discher

D.E.

, Janmey

, and Wang

Y.L.

Tissue cells feel and respond to the stiffness of their substrate. Science. 310:1139–1143, 2005.

14.

Worthington

K.S.

, Wiley

L.A.

, Bartlett

A.M.

, Stone

E.M.

, Mullins

R.F.

, Salem

A.K.

, Guymon

C.A.

, and Tucker

B.A.

Mechanical properties of murine and porcine ocular tissues in compression. Exp. Eye Res. 121:194–199, 2014.

15.

Thakur

, Mishra

, Pena

, Zhou

, Redenti

, Majeska

, and Vazquez

Collective adhesion and displacement of retinal progenitor cells upon extracellular matrix substrates of transplantable biomaterials. J. Tissue Eng. 9:2041731417751286, 2018.

16.

Tay

H.G.

, Cai

Z.H.

, Sun

, Guo

, Moreno-Moral

, Tan

W.S.

, and Tryggvason

Novel retina specific laminin isoforms recapitulate retinal interphotoreceptor matrix to generate human embryonic stem cell-derived photoreceptors. Invest. Ophthalmol. Vis. Sci. 59:3, 2018.

17.

Worthington

K.S.

, Green

B.J.

, Rethwisch

, Wiley

L.A.

, Tucker

B.A.

, Guymon

C.A.

, and Salem

A.K.

Neuronal differentiation of induced pluripotent stem cells on surfactant templated chitosan hydrogels. Biomacromolecules. 17:1684–1695, 2016.

18.

Ballios

B.G.

, Cooke

M.J.

, Donaldson

, Coles

B.L.

, Morshead

C.M.

, van der Kooy

, and Shoichet

M.S.

A hyaluronan-based injectable hydrogel improves the survival and integration of stem cell progeny following transplantation. Stem Cell Rep. 4:1031–1045, 2015.

19.

Ballios

B.G.

, Cooke

M.J.

, van der Kooy

, and Shoichet

M.S.

A hydrogel-based stem cell delivery system to treat retinal degenerative diseases. Biomaterials. 31:2555–2564, 2010.

20.

Baranov

, Michaelson

, Kundu

, Carrier

R.L.

, and Young

Interphotoreceptor matrix-poly(-caprolactone) composite scaffolds for human photoreceptor differentiation. J. Tissue Eng. 5:2041731414554139, 2014.

21.

Kundu

, Michaelson

, Talbot

, Baranov

, Young

M.J.

, and Carrier

R.L.

Decellularized retinal matrix: natural platforms for human retinal progenitor cell culture. Acta Biomater. 31:61–70, 2016.

22.

Ghodbane

S.A.

, and Dunn

M.G.

Physical and mechanical properties of cross-linked type I collagen scaffolds derived from bovine, porcine, and ovine tendons. J. Biomed. Mater. Res. A. 104:2685–2692, 2016.

23.

Ibrahim

, Kang

Q.K.

, and Ramamurthi

The impact of hyaluronic acid oligomer content on physical, mechanical, and biologic properties of divinyl sulfone-crosslinked hyaluronic acid hydrogels. J. Biomed. Mater. Res. A. 94:355–370, 2010.

24.

Eke

, Mangir

, Hasirci

, MacNeil

, and Hasirci

Development of a UV crosslinked biodegradable hydrogel containing adipose derived stem cells to promote vascularization for skin wounds and tissue engineering. Biomaterials. 129:188–198, 2017.

25.

Worthington

K.S.

, Do

A.V.

, Smith

, Tucker

B.A.

, and Salem

A.K.

Two-photon polymerization as a tool for studying 3D printed topography-induced stem cell fate. Macromol. Biosci. 19:e1800370, 2019.

26.

Forney

B.S.

, and Guymon

C.A.

Nanostructure evolution during photopolymerization in lyotropic liquid crystal templates. Macromolecules. 43:8502–8510, 2010.

27.

Hasa

, Scholte

J.P.

, Jessop

J.L.P.

, Stansbury

J.W.

, and Guymon

C.A.

Kinetically controlled photoinduced phase separation for hybrid radical/cationic systems. Macromolecules. 52:2975–2986, 2019.

28.

Kim

S.Y.

, Kundu

, Williams

, Yandulskaya

A.S.

, Monaghan

J.R.

, Carrier

R.L.

, and Linhardt

R.J.

Glycosaminoglycans compositional analysis of Urodele axolotl (Ambystoma mexicanum) and porcine retina. Glycoconj. J. 36:165–174, 2019.

29.

Wiley

L.A.

, Burnight

E.R.

, Songstad

A.E.

, Drack

A.V.

, Mullins

R.F.

, Stone

E.M.

, and Tucker

B.A.

Patient-specific induced pluripotent stem cells (iPSCs) for the study and treatment of retinal degenerative diseases. Prog. Retin. Eye Res. 44:15–35, 2015.

30.

Hunt

N.C.

, Hallam

, Karimi

, Mellough

C.B.

, Chen

, Steel

D.H.W.

, and Lako

3D culture of human pluripotent stem cells in RGD-alginate hydrogel improves retinal tissue development. Acta Biomater. 49:329–343, 2017.

31.

Taylor

, Arner

, Engelsberg

, and Ghosh

Scaffolding the retina: the interstitial extracellular matrix during rat retinal development. Int. J. Dev. Neurosci. 42:46–58, 2015.

32.

Bystrom

, Virtanen

, Rousselle

, Gullberg

, and Pedrosa-Domellof

Distribution of laminins in the developing human eye. Invest. Ophthalmol. Vis. Sci. 47:777–785, 2006.

33.

Ishikawa

, Sawada

, and Yoshitomi

Structure and function of the interphotoreceptor matrix surrounding retinal photoreceptor cells. Exp. Eye Res. 133:3–18, 2015.

34.

Peterson

P.E.

, Pow

C.S.

, Wilson

D.B.

, and Hendrickx

A.G.

Localisation of glycoproteins and glycosaminoglycans during early eye development in the macaque. J. Anat. 186:31–42, 1995.

35.

Clark

S.J.

, Keenan

T.D.

, Fielder

H.L.

, Collinson

L.J.

, Holley

R.J.

, Merry

C.L.

, van Kuppevelt

T.H.

, Day

A.J.

, and Bishop

P.N.

Mapping the differential distribution of glycosaminoglycans in the adult human retina, choroid, and sclera. Invest. Ophthalmol. Vis. Sci. 52:6511–6521, 2011.

36.

Libby

R.T.

, Champliaud

M.F.

, Claudepierre

, Xu

, Gibbons

E.P.

, Koch

, Burgeson

R.E.

, Hunter

D.D.

, and Brunken

W.J.

Laminin expression in adult and developing retinae: evidence of two novel CNS laminins. J. Neurosci. 20:6517–6528, 2000.

37.

Felemban

, Dorgau

, Hunt

N.C.

, Hallam

, Zerti

, Bauer

, Ding

, Collin

, Steel

, Krasnogor

, Al-Aama

, Lindsay

, Mellough

, and Lako

Extracellular matrix component expression in human pluripotent stem cell-derived retinal organoids recapitulates retinogenesis in vivo and reveals an important role for IMPG1 and CD44 in the development of photoreceptors and interphotoreceptor matrix. Acta Biomater. 74:207–221, 2018.