Quantitative Protein Expression in the Human Retinal Pigment Epithelium: Comparison Between Apical and Basolateral Plasma Membranes With Emphasis on Transporters

Transporter Protein Expression of Corneal Epithelium in Rabbit and Porcine: Evaluation of Models for Ocular Drug Transport Study

The transcorneal route is the main entry route for drugs to the intraocular parts, after topical administration. The outer surface, the corneal epithelium (CE), forms the rate-limiting barrier for drug permeability. Information about the role and protein expression of drug and amino acid transporter proteins in the CE is sparse and lacking. The aim of our study was to characterize transporter protein expression in rabbit and porcine CE to better understand potential drug and nutrient absorption after topical administration. Proteins, mainly Abc and Slc transporters, were characterized with quantitative targeted absolute proteomics and global untargeted proteomics methods. In the rabbit CE, 24 of 48 proteins were detected in the targeted approach, and 21 of these were quantified. In the porcine CE, 26 of 58 proteins were detected in the targeted approach, and 20 of these were quantified. Among these, 15 proteins were quantified in both animals: 4f2hc (Slc3a2), Aqp0, Asct1 (Slc1a4), Asct2 (Slc1a5), Glut1 (Slc2a1), Hmit (Slc2a13), Insr, Lat1 (Slc7a5), Mct1 (Slc16a1), Mct2 (Slc16a7), Mct4 (Slc16a3), Mrp 4 (Abcc4), Na+/K+-ATPase, Oatp3a1 (Slco3a1), and Snat2 (Slc38a2). Overall, the global proteomics results supported the targeted proteomics results. Organic anion transporting polypeptide Oatp3a1 was detected and quantified for the first time in both rabbit (1.4 ± 0.4 fmol/cm2) and porcine (11.1 ± 5.3 fmol/cm2) CE. High expression levels were observed for L-type amino acid transporter, Lat1, which was quantified with newly selected extracellular domain peptides in rabbit (48.9 ± 11.8 fmol/cm2) and porcine (37.6 ± 11.5 fmol/cm2) CE. The knowledge of transporter protein expression in ocular barriers is a key factor in the successful design of new ocular drugs, pharmacokinetic modeling, understanding ocular diseases, and the translation to human.

Increased sodium fluorescein transport by corticosteroids is inhibited by a LAT-1 specific inhibitor in retinal pigment epithelial cells in vitro

To investigate whether aldosterone (ALD) and hydrocortisone (HC) change the gene expression of SLC7A5, which encodes the large neutral amino acid transporter small subunit 1 (LAT1), and the transport activity of LAT1 in the retinal pigment epithelium (RPE) in vitro. ARPE-19 cells were grown to confluence. After withdrawing the serum, ALD or HC was added with several doses and incubated, and SLC7A5 gene expression was measured. The influx and efflux transport of sodium fluorescein (Na-F) were evaluated using the Transwell culture system. SLC7A5 gene expression was upregulated by ALD and downregulated by HC in a dose-dependent manner. Both ALD and HC significantly increased the influx and efflux Na-F transport of RPE cells at a dose that did not change the expression of SLC7A5. JPH203, a specific inhibitor of LAT1, significantly reduced accelerated Na-F transport. Both ALD and HC increased the gene expression of zonula occludin-1 (ZO-1) although they did not change the immunoreactivity of ZO-1 in RPE cells. LAT1 may play an important role in increasing Na-F transport associated with ALD and HC administration. A specific LAT1 inhibitor may effectively regulate the increased material transport of RPE induced by ALD and HC.

Dissimilar Effect of P-Glycoprotein and Breast Cancer Resistance Protein Inhibition on the Distribution of Erlotinib to the Retina and Brain in Humans and Mice

P-glycoprotein (P-gp) and breast cancer resistance protein (BCRP) are two ATP-binding cassette efflux transporters that are coexpressed at the human blood-brain barrier (BBB) and blood-retina barrier (BRB). While pharmacological inhibition of P-gp and/or BCRP results in increased brain distribution of dual P-gp/BCRP substrate drugs, such as the tyrosine kinase inhibitor erlotinib, the effect of P-gp and/or BCRP inhibition on the retinal distribution of such drugs has hardly been investigated. In this study, we used positron emission tomography (PET) imaging to assess the effect of transporter inhibition on the distribution of [11C]erlotinib to the human retina and brain. Twenty two healthy volunteers underwent two PET scans after intravenous (i.v.) injection of a microdose (<5 μg) of [11C]erlotinib, a baseline scan, and a second scan either with concurrent i.v. infusion of tariquidar to inhibit P-gp (n = 5) or after oral intake of single ascending doses of erlotinib (300 mg, 650 mg, or 1000 mg, n = 17) to saturate erlotinib transport. In addition, transport of [3H]erlotinib to the retina and brain was assessed in mice by in situ carotid perfusion under various drug transporter inhibition settings. In comparison to the baseline PET scan, coadministration of tariquidar or erlotinib led to a significant decrease of [11C]erlotinib total volume of distribution (VT) in the human retina by -25 ± 8% (p ≤ 0.05) and -41 ± 16% (p ≤ 0.001), respectively. In contrast, erlotinib intake led to a significant increase in [11C]erlotinib VT in the human brain (+20 ± 16%, p ≤ 0.001), while administration of tariquidar did not result in any significant changes. In situ carotid perfusion experiments showed that both P-gp and BCRP significantly limit the distribution of erlotinib to the mouse retina and brain but revealed a similar discordant effect at the mouse BRB and BBB following co-perfusion with tariquidar and erlotinib as in humans. Co-perfusion with prototypical inhibitors of solute carrier transporters did not reveal a significant contribution of organic cation transporters (e.g., OCTs and OCTNs) and organic anion-transporting polypeptides (e.g., OATP2B1) to the retinal and cerebral distribution of erlotinib. In conclusion, we observed a dissimilar effect after P-gp and/or BCRP inhibition on the retinal and cerebral distribution of [11C]erlotinib. The exact mechanism for this discrepancy remains unclear but may be related to the function of an unidentified erlotinib uptake carrier sensitive to tariquidar inhibition at the BRB. Our study highlights the great potential of PET to study drug distribution to the human retina and to assess the functional impact of membrane transporters on ocular drug distribution.

Selective drug delivery to the retinal cells: Biological barriers and avenues

Retinal drug delivery is a challenging, but important task, because most retinal diseases are still without any proper therapy. Drug delivery to the retina is hampered by the anatomical and physiological barriers resulting in minimal bioavailability after topical ocular and systemic administrations. Intravitreal injections are current method-of-choice in retinal delivery, but these injections show short duration of action for small molecules and low target bioavailability for many protein, gene based drugs and nanomedicines. State-of-art delivery systems are based on prolonged retention, controlled drug release and physical features (e.g. size and charge). However, drug delivery to the retina is not cell-specific and these approaches do not facilitate intracellular delivery of modern biological drugs (e.g. intracellular proteins, RNA based medicines, gene editing). In this focused review we highlight biological factors and mechanisms that form the basis for the selective retinal drug delivery systems in the future. Therefore, we are presenting current knowledge related to retinal membrane transporters, receptors and targeting ligands in relation to nanomedicines, conjugates, extracellular vesicles, and melanin binding. These issues are discussed in the light of retinal structure and cell types as well as future prospects in the field. Unlike in some other fields of targeted drug delivery (e.g. cancer research), selective delivery technologies have been rarely studied, even though cell targeted delivery may be even more feasible after local administration into the eye.

Characterizations of annexin A1-interacting proteins in apical membrane and cytosolic compartments of renal tubular epithelial cells

Annexin A1 (ANXA1) is a multifunctional calcium-binding protein that can bind to membrane phospholipids. Under high-calcium condition, ANXA1 expression increases on renal epithelial cell surface, leading to enhanced adhesion of calcium oxalate (CaOx) crystal (stone material) onto the cells. To regulate various cellular processes, ANXA1 interacts with many other intracellular protein partners. However, components of the ANXA1-interacting protein complex remain unclear. Herein, we characterized the interacting complexes of apical membrane (ApANXA1) and cytosolic (cyANXA1) forms of ANXA1 in apical membrane and cytosolic compartments, respectively, of renal epithelial cells under high-calcium condition using proteomic and bioinformatic approaches. After fractionation, the ApANXA1- and CyANXA1-interacting partners were identified by immunoprecipitation followed by nanoLC‑ESI‑Qq-TOF tandem mass spectrometry (IP-MS/MS). The ANXA1-interacting partners that were common in both apical membrane and cytosolic compartments and those unique in each compartment were then analyzed for their physico-chemical properties (molecular weight, isoelectric point, amino acid contents, instability index, aliphatic index, and grand average of hydropathicity), secondary structure (α-helix, β-turn, random coil, and extended strand), molecular functions, biological processes, reactome pathways and KEGG pathways. The data demonstrated that each set of these interacting proteins exhibited common and unique characteristics and properties. The knowledge from this study may lead to better understanding of the ApANXA1 and CyAXNA1 biochemistry and functions as well as the pathophysiology of CaOx kidney stone formation induced by high-calcium condition.

Comparison of barrier properties of outer blood-retinal barrier models - Human stem cell-based models as a novel tool for ocular drug discovery

The retinal pigment epithelial (RPE) cell monolayer forms the outer blood-retinal barrier and has a crucial role in ocular pharmacokinetics. Although several RPE cell models are available, there have been no systematic comparisons of their barrier properties with respect to drug permeability. We compared the barrier properties of RPE secondary cell lines (ARPE19, and ARPE19mel) and both primary (hfRPE) and stem-cell derived RPE (hESC-RPE) cells by investigating the permeability of nine drugs (aztreonam, ciprofloxacin, dexamethasone, fluconazole, ganciclovir, ketorolac, methotrexate, voriconazole, and quinidine) across cell monolayers. ARPE19, ARPE19mel, and hfRPE cells displayed a narrow P_app value range, with relatively high permeation rates (5.2-26 × 10^-6 cm/s). In contrast, hESC-RPE cells efficiently restricted the drug flux, and displayed even lower P_app values than those reported for bovine RPE-choroid, with the range of 0.4-32 cm^-6/s. Therefore, ARPE19, ARPE19mel, and hfRPE cells failed to form a tight barrier, whereas hESC-RPE cells restricted the drug flux to a similar extent as bovine RPE-choroid. Therefore, hESC-RPE cells are valuable tools in ocular drug discovery.

Ocular barriers as a double-edged sword: preventing and facilitating drug delivery to the retina

In recent decades, the growing of the aging population in the world brings increasingly heavy burden of vision-threatening retinal diseases. One of the biggest challenges in the treatment of retinal diseases is the effective drug delivery to the diseased area. Due to the existence of multiple anatomical and physiological barriers of the eye, commonly used oral drugs or topical eye drops cannot effectively reach the retinal lesions. Innovations in new drug formulations and delivery routes have been continuously applied to improve current drug delivery to the back of the eye. Unique ocular anatomical structures or physiological activities on these ocular barriers, in turn, can facilitate drug delivery to the retina if compatible formulations or delivery routes are properly designed or selected. This paper focuses on key barrier structures of the eye and summarizes advances of corresponding drug delivery means to the retina, including various local drug delivery routes by invasive approaches, as well as systemic eye drug delivery by non-invasive approaches.

Mechanisms of cellular retention of melanin bound drugs: Experiments and computational modeling

Melanin binding of drugs is known to increase drug concentrations and retention in pigmented eye tissues. Even though the correlation between melanin binding in vitro and exposure to pigmented eye in vivo has been shown, there is a discrepancy between rapid drug release from melanin particles in vitro and the long in vivo retention in the pigmented tissues. We investigated mechanisms and kinetics of pigment-related drug retention experimentally using isolated melanin particles from porcine retinal pigment epithelium and choroid, isolated porcine eye melanosomes, and re-pigmented ARPE-19 cells in a dynamic flow system. The experimental studies were supplemented with kinetic simulations. Affinity and capacity of levofloxacin, terazosin, papaverine, and timolol binding to melanin revealed K_d values of ≈ 50-150 μM and B_max ≈ 40-112 nmol.mg^-1. The drugs were released from melanin in <1 h (timolol) or in 6-12 h (other drugs). The drugs were released slower from the melanosomes than from melanin; the experimental differences ranged from 1.2-fold (papaverine) to 7.4-fold (timolol). Kinetic simulations supported the role of the melanosomal membrane in slowing down the release of melanin binders. In release studies from the pigmented ARPE-19 cells, drugs were released from the cellular melanin to the extracellular space in ≈ 1 day (timolol) and ≈ 11 days (levofloxacin), i.e., much slower than the release from melanin or melanosomes. Simulations of drug release from pigmented cells in the flow system matched the experimental data and enabled further sensitivity analyses. The simulations demonstrated a significant prolongation of drug retention in the cells as a function of decreasing drug permeability in the melanosomal membranes and increasing melanin content in the cells. Overall, we report the impact of cellular factors in prolonging drug retention and release from melanin-containing cells. These data and simulations will facilitate the design of melanin binding drugs with prolonged ocular actions.

Exploration of the Plausible Mechanism of Ethambutol Induced Ocular Toxicity by Using Proteomics Informed Physiologically Based Pharmacokinetic (PBPK) Modeling

PURPOSE

Ethambutol (EMB) is a first-line anti-tubercular drug that is known to cause optic neuropathy. The exact mechanism of its eye toxicity is unknown; however, proposition is metal chelating effect of both EMB and its metabolite 2,2'-(ethylenediamino)-dibutyric acid (EDBA). The latter is formed by sequential metabolism of EMB by alcohol dehydrogenases (ADHs) and aldehyde dehydrogenases (ALDHs). The purpose of this study was to predict the levels of drug and EDBA in the eye using physiologically based pharmacokinetic (PBPK) modeling.

METHODS

The PBPK model of EMB was developed using GastroPlus. The intrinsic hepatic clearance of ALDH, calculated by the model, was scaled down using proteomics data to estimate the rate of formation of EDBA in the eye. Additionally, the comparative permeability of EMB and EDBA was assessed by employing in silico and in vitro approaches. The rate of formation of EDBA in the eye and permeability data were then incorporated in a compartmental model to predict the ocular levels of EMB and EDBA.

RESULTS

The simulation results of compartmental model highlighted that there was an on-site formation of EDBA upon metabolism of EMB. Furthermore, in silico and in vitro studies revealed that EDBA possessed much lower permeability than EMB. These observations meant that once EDBA was formed in the eye, it was not permeated out and hence achieved higher ocular concentration.

CONCLUSION

The on-site formation of EDBA in the eye, its higher local concentration due to lower ocular clearance and its pre-known characteristic to chelate metal species better explains the ocular toxicity shown by EMB.

Recent advances in riboflavin transporter RFVT and its genetic disease

Riboflavin (vitamin B2) is essential for cellular growth and function. It is enzymatically converted to flavin mononucleotide (FMN) and flavin adenine dinucleotide (FAD), which participate in the metabolic oxidation-reduction reactions of carbohydrates, amino acids, and lipids. Human riboflavin transporters RFVT1, RFVT2, and RFVT3 have been identified and characterized since 2008. They are highly specific transporters of riboflavin. RFVT3 has functional characteristics different from those of RFVT1 and RFVT2. RFVT3 contributes to absorption in the small intestine, reabsorption in the kidney, and transport to the fetus in the placenta, while RFVT2 mediates the tissue distribution of riboflavin from the blood. Several mutations in the SLC52A2 gene encoding RFVT2 and the SLC52A3 gene encoding RFVT3 were found in patients with a rare neurological disorder known as Brown-Vialetto-Van Laere syndrome. These patients commonly present with bulbar palsy, hearing loss, muscle weakness, and respiratory symptoms in infancy or later in childhood. A decrease in plasma riboflavin levels has been observed in several cases. Recent studies on knockout mice and patient-derived cells have advanced the understanding of these mechanisms. Here, we summarize novel findings on RFVT1-3 and their genetic diseases and discuss their potential as therapeutic drugs.

Role of LINC01592 in TGF-β1-induced epithelial-mesenchymal transition of retinal pigment epithelial cells

Regulation of long-chain non-coding RNA01592 (LINC01592) in the process of transforming retinal pigment epithelial (RPE) cells into mesenchymal cells following induction by transforming growth factor beat 1 (TGF-β1) was investigated by interfering with LINC01592 expression in human RPE (hRPE) cells. LINC01592 expression in hRPE cells was significantly increased following treatment with 10 ng/mL TGF-β1 for 48 h. Expression of E-cadherin and Snail were decreased in hRPE cells following induction with TGF-β1 compared with the control group (P < 0.05). Following induction by TGF-β1, expression of E-cadherin, alpha-smooth muscle actin (α-SMA), and Snail were significantly lower in the LINC01592-knockdown group compared with the negative control group (P < 0.05). LINC01592 overexpression significantly enhanced the viability, proliferation, and migration of hRPE cells induced by TGF-β1 (P < 0.05). Following induction by TGF-β1, E-cadherin expression was significantly decreased and α-SMA and Snail expression were significantly increased in the LINC01592-overexpression group compared with the negative control group (P < 0.05). RPE cells induced by TGF-β1 exhibited epithelial-mesenchymal transition (EMT). Inhibiting LINC01592 expression could significantly reduce TGF-β1-induced EMT of hRPE cells. The regulatory effect of LINC01592 on EMT in hRPE cells induced by TGF-β1 provides a novel treatment for proliferative vitreoretinopathy.

Statins for the prevention of proliferative vitreoretinopathy: cellular responses in cultured cells and clinical statin concentrations in the vitreous

Proliferative vitreoretinopathy (PVR) with rhegmatogenous retinal detachment (RRD) is a complex inflammatory ocular disease. Statins are widely used cholesterol-lowering drugs with putative anti-inflammatory properties. In this study, we have explored their efficacy in controlling post-surgical PVR formation. Simvastatin (SIM), atorvastatin (ATV), or rosuvastatin (RSV) were added to cultures of human retinal pigment epithelial cells (ARPE-19) prior to exposure with the bacterial lipopolysaccharide (LPS), and the production of pro-inflammatory cytokines (IL-6, IL-8, MCP-1) was examined using an enzyme-linked immunosorbent assay. In addition, the concentrations of simvastatin, atorvastatin, rosuvastatin, and their metabolites were measured from the vitreal samples of 20 patients undergoing vitrectomy (16 of them receiving oral statin therapy) using an ultra-performance liquid chromatography-tandem mass spectrometer technique. All statins alleviated LPS-induced inflammation at 5 µM concentration in the ARPE-19 cell cultures. Statin levels in the vitreous samples ranged from 6 to 316 pg/mL (ca. 0.1-7 M^-10). Vitreal statin concentrations were similar to the typical steady-state unbound statin concentrations in plasma, indicating that only the unbound drug distributes from the blood circulation into the vitreous. Pharmacokinetic simulations of the intravitreal delivery of statins indicate that the measured clinical statin concentrations could be maintained with existing drug delivery technologies for months. Our results suggest that intravitreal statin therapy may have the potential in alleviating the risk of post-surgical PVR.

Using bioinformatic protein sequence similarity to investigate if SARS CoV-2 infection could cause an ocular autoimmune inflammatory reactions?

Although severe acute respiratory syndrome coronavirus 2 (SARS CoV-2) infection have emerged globally, findings related to ocular involvement and reported cases are quite limited. Immune reactions against viral infections are closely related to viral and host proteins sequence similarity. Molecular Mimicry has been described for many different viruses; sequence similarities of viral and human tissue proteins may trigger autoimmune reactions after viral infections due to similarities between viral and human structures. With this study, we aimed to investigate the protein sequence similarity of SARS CoV-2 with retinal proteins and retinal pigment epithelium (RPE) surface proteins. Retinal proteins involved in autoimmune retinopathy and retinal pigment epithelium surface transport proteins were analyzed in order to infer their structural similarity to surface glycoprotein (S), nucleocapsid phosphoprotein (N), membrane glycoprotein (M), envelope protein (E), ORF1ab polyprotein (orf1ab) proteins of SARS CoV-2. Protein similarity comparisons, 3D protein structure prediction, T cell epitopes-MHC binding prediction, B cell epitopes-MHC binding prediction and the evaluation of the antigenicity of peptides assessments were performed. The protein sequence analysis was made using the Pairwise Sequence Alignment and the LALIGN program. 3D protein structure estimates were made using Swiss Model with default settings and analyzed with TM-align web server. T-cell epitope identification was performed using the Immune Epitope Database and Analysis (IEDB) resource Tepitool. B cell epitopes based on sequence characteristics of the antigen was performed using amino acid scales and HMMs with the BepiPred 2.0 web server. The predicted peptides/epitopes in terms of antigenicity were examined using the default settings with the VaxiJen v2.0 server. Analyses showed that, there is a meaningful similarities between 6 retinal pigment epithelium surface transport proteins (MRP-4, MRP-5, RFC1, SNAT7, TAUT and MATE) and the SARS CoV-2 E protein. Immunoreactive epitopic sites of these proteins which are similar to protein E epitope can create an immune stimulation on T cytotoxic and T helper cells and 6 of these 9 epitopic sites are also vaxiJen. These result imply that autoimmune cross-reaction is likely between the studied RPE proteins and SARS CoV-2 E protein. The structure of SARS CoV-2, its proteins and immunologic reactions against these proteins remain largely unknown. Understanding the structure of SARS CoV-2 proteins and demonstration of similarity with human proteins are crucial to predict an autoimmune response associated with immunity against host proteins and its clinical manifestations as well as possible adverse effects of vaccination.

Expression of Pro-Angiogenic Markers Is Enhanced by Blue Light in Human RPE Cells

Inherited retinal dystrophies are characterized by photoreceptor death. Oxidative stress usually occurs, increasing vision loss, and oxidative damage is often reported in retinitis pigmentosa (RP). More than 300 genes have been reported as RP causing. In contrast, choroidal neovascularization (CNV) only occasionally develops in the late stages of RP. We herein study the regulation of RP causative genes that are likely linked to CNV onset under oxidative conditions. We studied how the endogenous adduct N-retinylidene-N-retinylethanolamine (A2E) affects the expression of angiogenic markers in human retinal pigment epithelium (H-RPE) cells and a possible correlation with RP-causing genes. H-RPE cells were exposed to A2E and blue light for 3 and 6h. By transcriptome analysis, genes differentially expressed between A2E-treated cells and untreated ones were detected. The quantification of differential gene expression was performed by the Limma R package. Enrichment pathway analysis by the FunRich tool and gene prioritization by ToppGene allowed us to identify dysregulated genes involved in angiogenesis and linked to RP development. Two RP causative genes, AHR and ROM1, can be associated with an increased risk of CNV development. Genetic analysis of RP patients affected by CNV will confirm this hypothesis.

The cell biology of the retinal pigment epithelium

The retinal pigment epithelium (RPE), a monolayer of post-mitotic polarized epithelial cells, strategically situated between the photoreceptors and the choroid, is the primary caretaker of photoreceptor health and function. Dysfunction of the RPE underlies many inherited and acquired diseases that cause permanent blindness. Decades of research have yielded valuable insight into the cell biology of the RPE. In recent years, new technologies such as live-cell imaging have resulted in major advancement in our understanding of areas such as the daily phagocytosis and clearance of photoreceptor outer segment tips, autophagy, endolysosome function, and the metabolic interplay between the RPE and photoreceptors. In this review, we aim to integrate these studies with an emphasis on appropriate models and techniques to investigate RPE cell biology and metabolism, and discuss how RPE cell biology informs our understanding of retinal disease.

Drug Flux Across RPE Cell Models: The Hunt for An Appropriate Outer Blood-Retinal Barrier Model for Use in Early Drug Discovery

The retinal pigment epithelial (RPE) cell monolayer forms the outer blood–retinal barrier and has a crucial role in ocular pharmacokinetics. Although several RPE cell models are available, there have been no systematic comparisons of their barrier properties with respect to drug permeability. We compared the barrier properties of several RPE secondary cell lines (ARPE19, ARPE19mel, and LEPI) and both primary (hfRPE) and stem-cell derived RPE (hESC-RPE) cells by investigating the permeability of nine drugs (aztreonam, ciprofloxacin, dexamethasone, fluconazole, ganciclovir, ketorolac, methotrexate, voriconazole, and quinidine) across cell monolayers. ARPE19, ARPE19mel, and hfRPE cells displayed a narrow P_app value range, with relatively high permeation rates (5.2–26 × 10⁻⁶ cm/s. In contrast, hESC-RPE and LEPI cells efficiently restricted the drug flux, and displayed even lower P_app values than those reported for bovine RPE-choroid, with the range of 0.4–32 cm⁻⁶/s (hESC-RPE cells) and 0.4–29 × 10⁻⁶ cm/s, (LEPI cells). Therefore, ARPE19, ARPE19mel, and hfRPE cells failed to form a tight barrier, whereas hESC-RPE and LEPI cells restricted the drug flux to a similar extent as bovine RPE-choroid. Therefore, LEPI and hESC-RPE cells are valuable tools in ocular drug discovery.