Glial Cell–Derived Cytokines Attenuate the Breakdown of Vascular Integrity in Diabetic Retinopathy Free

The human glioblastoma cell line U373MG and endothelial cells cultured using the bovine brain microvascular endothelial cell system (bMVEC-B) (Cambrex, Baltimore, MD) were maintained in Dulbecco's modified Eagle's medium (Sigma, St. Louis, MO) supplemented with 10% fetal bovine serum (FBS) (Cell Culture Technologies, Lugano, Switzerland) or charcoal dextran–treated FBS (18), with 100 units/ml penicillin, and with 100 μg/ml streptomycin (Sigma). Primary culture of human astrocytes (Cambrex) was maintained in astrocyte growth medium (Cambrex). The 8th to 12th passages of bMVEC-B and 3rd to 4th passages of human astrocytes were used in this study. U373MG cells and human astrocytes were treated within a dose ranging from 10 nmol/l to 1 μmol/l ATRA or with 100 nmol/l ATRA, 100 nmol/l 9-cis retinoic acid (9cRA), and 10 nmol/l 4-[(5,6,7,8-tetrahydro-5,5,8,8-tetramethyl-2-naphthalenyl)carboxamido] benzoic acid (Am580). Confluent cultured endothelial cells were treated with 1 ng/ml GDNF recombinant protein (GDNFr-p) (PeproTech, Rocky Hill, NJ) or 100 nmol/l ATRA for up to 7 days, and RNA and protein samples were extracted from the cells treated for 7 days. For transfection of short interfering RNA (siRNA), glial cells were transfected with 150 nmol/l GDNF siRNA (Santa Cruz Biotechnologies, Santa Cruz, CA) by electroporation (capacitance extender 2.0 pulse/960 μPD; Bio-Rad, Hercules, CA). Unless otherwise specified, cells treated for 4 and 8 h were used for RNA and protein preparation, respectively.

Total RNA was extracted from cells using TRIzol Reagent (Invitrogen, Carisbad, CA). Then, RNA (1 μg) was reverse transcribed using poly-T oligonucleotide and M-MuLV reverse transcriptase (Invitrogen). For analysis of gene expression, the genes of interest were amplified from dilutions of cDNA with the TaqMan gene expression assay using ABI Prism 7000 (Applied Biosystems, Foster City, CA) or of PCR using specific sense and antisense primers for up to 40 cycles. We examined various cycling parameters for each PCR experiment to define optimal conditions for linearity to allow for semiquantitative analysis of signal intensity. To provide a qualitative control for reaction efficiency, PCRs were performed with primers coding for glyceraldehyde-3-phosphate dehydrogenase (GAPDH). (All primer sequences used in this study are available from the authors on request.) For the densitometric analysis, signals in RT-PCR analysis were quantitated using Scion Image 1.62 (Scion Corporation, Frederick, MD).

Endothelial cells were grown to confluence on transwell semipermeable membranes (pore size 0.4 μm; Costar, Cambridge, MA) in the presence or absence of 100 nmol/l ATRA or 1 ng/ml GDNFr-p in the lower chamber. In the coculture experiments, glial cells cultured in the lower chamber were treated with 100 nmol/l ATRA or 10 nmol/l Am580 for 8 h and cocultured with endothelial cells that were grown to confluence on transwell membranes in the upper chamber. Paracellular tracer flux was measured by applying [¹⁴C]inulin (MW 5 kDa) at 5 × 10⁵ dpm/well onto an endothelial monolayer in the apical compartment, and the samples were collected from the basolateral compartment in a time-dependent manner. Radioactivity of [¹⁴C] was counted using a scintillation counter (Beckman LS6500; Beckman Coulter, Fullerton, CA).

Confluent-cultured endothelial cells treated with GDNFr-p were immunolabeled with antibodies directed against claudin-5, occludin, and zonula occludens-1 (all from Invitrogen). Samples were then probed with corresponding secondary antibodies coupled with Alexa Fluor 594 (Invitrogen), and cells were examined using a laser-scanning confocal microscope (MRC 1024; Bio-Rad).

GDNF and VEGF proteins were quantitated by enzyme-linked immunosorbent assay (ELISA) using a GDNF Emax Immunoassay system (Promega, Madison, WI) or VEGF Immunoassay kit (Biosource, Camarillo, CA). In addition, GDNF and VEGF proteins were also examined by Western blot analysis. The protein concentration was determined with a bicinchoninic acid protein assay kit (Pierce, Rockford, IL). Twenty micrograms of each denatured protein was subjected to SDS-PAGE. The membrane was reacted with antibodies against GDNF, VEGF, and β-actin (all from Santa Cruz), and the immunoreactions were visualized using an enhanced chemiluminescent system (Amersham Bioscience, Buckinghamshire, U.K.).

Undiluted vitreous samples were obtained from 85 consecutive patients with diabetic retinopathy (65 patients) and macular holes (20 nondiabetic control subjects), who underwent vitrectomy (supplementary Table [available in an online appendix at http://dx.doi.org/10.2337/db06-1431]). Before the experiment, informed consent was obtained from each patient, and the experiment was approved by the institutional review board.

We found a putative human GDNF promoter fragment in the gene database (http://www.ncbi.nlm.nih.gov) based on the exon 1 sequence of GDNF cDNA (GeneBank accession no. NM_199234) and amplified it by genomic PCR using a downstream primer encompassing the transcription initiation codon 5′-AGCCACGACATCCCATAACT-3′ and forward primer 5′-TGCACTTTTCCTGTCTGC-3′, which corresponded to the sequence 1,793 bp (∼1.8 kb) upstream from the transcriptional initiation site. Since three p300 binding motifs were found in this region using a Web-based search for putative nuclear receptors responsible for GDNF expression (http://motif.genome.jp), we also amplified a 1.2-kb fragment that lacked all p300 binding sites using forward primer 5′-GGCACTGAGGTTTTTGCATT-3′ and reverse primer 5′-CTTCTCCACCACACGGTCTT-3′. After performing TA cloning of 1.8 kb and 1.2 kb GDNF promoter cDNAs using pCR 2.1 (Invitrogen), we cloned each cDNA into pGL-3Basic vector (Promega). HeLa cells (1 × 10⁵) were split for seeding into 24-well plates 24 h before transfection and transfected to each vector (4 μg) with 40 ng pRL-Tk (Promega) in 4 μl FuGENE 6 reagent (Roche Diagnostics, Basel, Switzerland). After 4 h treatment with 100 nmol/l ATRA or 10 nmol/l Am580, proteins were extracted using a passive lysis buffer (Promega) and were read using dual luciferase reagents (Promega) and an AB-2200 luminometer (Atto, Tokyo, Japan).

Glial cells treated with 100 nmol/l ATRA or 10 nmol/l Am580 for 4 h were subjected to chromatin immunoprecipitation (ChIP) assay. Aliquots from the cells were incubated with specific antibodies against p300 (Santa Cruz) and rotated overnight at 4°C. The immunocomplex was recovered by centrifugation and DNA extracted. DNA from the bulky lysate of the cells was directly amplified as a positive control, and a DNA template was isolated from the lysate without the immunoprecipitation step as a negative control.

C57BL/6 male mice (5 weeks old; CLEA, Tokyo, Japan) were housed in a specific pathogen-free facility and were maintained on standard mouse diet and water. To induce diabetes, mice were intraperitoneally injected with 40 mg/kg streptozotocin (Sigma) for 5 consecutive days. Control mice were intraperitoneally injected with a solvent alone (250 μl DMSO; Sigma). Six weeks after the verification of diabetes, mice were treated with 1.0 mg/kg ATRA every day, 3.75 mg/kg Am580 every other day, or DMSO every day for 1 week. To examine the leakage of retinal vessels, we injected 50 mg/kg fluorescein isothiocyanate (FITC)-dextran (4 kDa; Sigma) into mice via the vena cava, following which the mice were killed, and the bilateral eyes were enucleated 5 min after the FITC injection under general anesthesia. The aqueous phase of the sonicated right eye was examined to determine the FITC concentration (WALLAC 1420; Perkin Elmer, Wellesley, MA) and for quantification of GDNF and VEGF proteins. Left eyes were flat mounted, and the retinas were analyzed by laser-scanning confocal analysis. The FITC concentration in cardiac blood of each mouse was calculated to provide a quantitative control. All animal experiments were performed in accordance with the Guidelines for Animal Experiments of Sapporo Medical University School of Medicine.

All values are expressed as means and SDs of at least three independent experiments. Data were statistically analyzed with the Mann-Whitney U test for comparison of two groups and Fisher's test for comparison of three or more groups. Statistical significance was set at P < 0.05.

ATRA is a biologically active regulator for cell fate decision, having a broad range of functions involving cell differentiation, proliferation, and apoptosis in various cell types (15,16). Based on the recent observation that ATRA plays an important role in the induction of GDNF expression and responsiveness of glial cells (17), we examined whether ATRA could regulate the competence of glial cells to affect GDNF expression. We first found that ATRA upregulated GDNF mRNA expression in a time- and dose-dependent manner in U373MG cells (Fig. 1A and B), which express glial fibrillary acidic protein (GFAP), a specific differentiation marker of glial cells. ATRA clearly upregulated GDNF mRNA in a dose ranging from 10 nmol/l to 1 μmol/l ATRA, and the peak increase in the level of GDNF mRNA was achieved at 100 nmol/l (Fig. 1A and B). The induced level of GDNF mRNA was almost the same between 3 and 24 h after the treatment, indicating that U373MG required 3 h for full induction of GDNF mRNA (Fig. 1A). While GDNF expression was not changed in the medium supplemented with endogenous retinoic acid–depleted (i.e., charcoal-dextran treated) FBS, treatment with 100 nmol/l ATRA resulted in enhanced expression of GDNF (Fig. 1C), suggesting that the upregulation of GDNF required pharmacological doses of ATRA.

Our previous studies have demonstrated that GDNF is an important factor in regulating vascular permeability in vitro (12–14). Since it is possible that vascular permeability is determined by the balance between permeability-inducing and -inhibiting factors, we therefore defined the antipermeability activity as the expression level of VEGF subtracted from that of GDNF in a given setting, where the cells treated with a vehicle were used as a control and defined as 100% in PCR analysis. According to this concept, treatments with ATRA decreased VEGF mRNA expression and significantly increased the antipermeability activity (GDNF-VEGF) in U373MG cells (Fig. 1D).

The effects of endogenous retinoic acids are achieved primarily by two retinoic acid isomers (ATRA and 9cRA) and are mediated by two classes of nuclear receptors, RARs (α, β, and γ) and their heterodimeric counterparts retinoid-X receptors (RXRα, -β, and -γ) (16,19). We next examined GDNF expression by using early passages of human astrocytes in the presence of various synthetic retinoids to identify the retinoic acid nuclear receptor subtype(s) responsible for GDNF induction. Treatments with RAR pan-agonists (such as ATRA) and RARα agonists (such as Am580) (20,21) satisfactorily induced GDNF mRNA (Fig. 1E–G), whereas RXR pan-agonists (such as 9cRA and Hx630) did not show marked effects on GDNF when compared with RARα-specific agonists (data not shown). In addition, GFAP expression was significantly induced by the treatment with RARα agonists (Fig. 1E), indicating that RARα stimulants acted on glial cells as a differentiation inducer. Human astrocytes exclusively expressed RARα and -β (Fig. 1H), partially supporting the differential sensitivity of retinoic acids on astrocytes and the idea that RARα stimulants preferentially acted on glial cells to upregulate GDNF expression. This observation was consistent with the published result showing that RARα agonists were, at least in part, important regulators of GDNF expression (17). On the other hand, treatment with RARα stimulants downregulated VEGF expression (Fig. 1E) and were effective in increasing the antipermeability activities, being six times higher in the cells without retinoic acid treatment (Fig. 1I).

ATRA has been shown to trans-activate multiple signal transduction molecules, including p300/CREB (cyclic AMP response element) binding protein (CBP) signal transducer and activator of transcription-3 (STAT3), which eventually play important roles in glial cell differentiation (22–25). We thus examined the alteration in expression of genes belonging to the p300/CBP-STAT3 signal transduction pathway and found that treatments with ATRA and Am580 induced marked increases in the levels of p300/CBP, STAT3, smad1, Notch, Hes-1, and Hes-5 mRNA in glial cells (Fig. 2A). To explore the possible mechanism responsible for the retinoic acid–mediated GDNF upregulation, we isolated an ∼1.8-kb putative promoter fragment including the transcription initiation codon and ∼1.2-kb promoter fragment that lacked all p300 binding motifs and cloned these cDNAs into a luciferase reporter gene plasmid (Fig. 2B, left panel). Luciferase assay revealed that while ATRA and Am580 each exhibited enhanced activity for the full-length promoter, they were ineffective in the deletion mutant lacking the three p300 binding motifs (Fig. 2B, right panel). The percentages of decreases in luciferase activity were 67.0% in ATRA and 93.7% in Am580, compared with the cells transfected with the nondeletion construct (P < 0.05). In addition, ChIP assay demonstrated that p300 was selectively recruited to the GDNF promoter after the treatments with ATRA and Am580 (Fig. 2C). These observations suggested that the expression of GDNF was regulated through the recruitment of an RARα-driven trans-acting coactivator to the ∼1.8-kb 5′-flanking fragment of the GDNF promoter. Although we could not demonstrate a role for the p300/CBP-STAT3 signaling pathway in VEGF downregulation, we cannot rule out its involvement in this process.

Our previous study demonstrated that GDNF secreted from glial cells was a critical player in regulating the vascular permeability of the BRB (12–14). To examine whether the gene expression alterations in glial cells induced by RARα agonists modulated the barrier function of endothelial cells, we measured fluxes of radiolabeled molecules using [¹⁴C]inulin in a bMVEC-B. This system contained primary microvascular cells derived from bovine brain tissue for the study of microvascular endothelial cell tight junctions and the transport of molecules across the blood-brain barrier, having a histological architecture similar to the BRB (1,12–14). After verifying that the endothelial cells expressed GDNF receptor GFRα1 (Fig. 3A), we treated the endothelium with GDNFr-p and found that it significantly enhanced the barrier function (Fig. 3B). Considering the physiological relevance of our experiment to study the function of the BRB, which is a biological unit comprised of glial cells and capillary endothelium, we cultured glial cells in the lower chamber of a transwell and combined them with an endothelial monolayer that was grown to confluence in the upper chamber in a coculture experiment. Endothelial cells that were cocultured with glial cells pretreated with ATRA and Am580 showed a significant increase of barrier function (Fig. 3C, left panel). Although ATRA was reported to have the ability to enhance the endothelial barrier (26), ATRA alone did not have any effect on the endothelial cells (Fig. 3B), supporting our hypothesis that glial cell–derived cytokines affect the endothelial cells to alter the barrier function in a paracrine manner. Importantly, treatment with GDNF-specific siRNA, which effectively silenced the constitutively expressed GDNF in glial cells, completely inhibited the RARα-mediated enhancement of the endothelial barrier functions (Fig. 3C), strongly suggesting a direct mechanistic link between GDNF expression in glial cells and the regulation of the vascular permeability of the endothelial cells.

Consistent with our observations (Fig. 3B and C), the expression levels of claudin-5, an endothelial tight junction protein and a major determinant of vascular permeability, occludin, and zonula occludens-1 were increased (Fig. 3D), and they were concentrated along the cell-cell borders (Fig. 3E) after the treatment with GDNFr-p. This observation suggested that RARα stimulants regulated barrier function through modulation of expression of a number of tight junction–associated genes. There appeared to be a discrepancy between the results of the RT-PCR assay and immunohistochemistry; however, this might primarily have been due to the lack of commercially available bovine antibodies against tight junction proteins.

To further demonstrate the functional significance of glial GDNF, we next evaluated whether RARα stimulants inhibited the vascular leakage in murine diabetic retinopathy. After confirming diabetes in mice (Fig. 4A), we found that diabetic mice had significantly higher expression of both GDNF and VEGF (Fig. 4B and C). While VEGF protein showed significant decreases after the treatments with ATRA and Am580, these treatments did not change the protein levels of GDNF in whole lysates of eyes of diabetic animals (Fig. 4B and C). We cannot currently explain this observation. One possibility is that GDNF expression was induced by compensatory or reactive mechanisms in response to the upregulated VEGF in diabetic animals (5–8). Alternatively, a certain type of advanced glycation end product might have some as of yet unidentified effects on glial cells. However, these alterations in gene expression resulted in a significant increase of antipermeability activities, showing that the RARα stimulation induced apparently opposite trends in endothelial permeability (Fig. 4D). In addition, the altered phenotype mediated by RARα was sufficient for inhibiting vascular leakage to maintain vascular integrity in the retinal microenvironment (Fig. 4E and F). The changes in expression of GDNF and VEGF in diabetic mice were consistent with the results obtained for human vitreous fluid from patients with diabetic retinopathy, showing that the antipermeability activity was significantly decreased in diabetic patients (Fig. 5A and B).

Accumulated evidence has suggested that molecular processes involved in vascular growth and vascular hyperpermeability are based on the inappropriate regulation of VEGF (3,4). VEGF thus represents an important target for therapeutic intervention in diabetic retinopathy, and molecular pharmacology that directly inhibits the actions of VEGF has shown considerable promise but has not proven to be satisfactorily effective in blocking the pathology and development of microangiopathies (5–9). Since VEGF is important for maintaining endothelial physiology (27), VEGF suppression therapy may produce unidentified adverse effects on the neural retina. Therefore, a better understanding of the molecular events in vascular alterations is of clinical importance to identify more precise therapeutic targets for diabetic retinopathy.

Indeed, our previous studies have demonstrated that GDNF secreted from glial cells is a critical factor in regulating the vascular permeability of the BRB in a biological unit comprised of capillary endothelial cells and glial cells. In addition, the present study clearly demonstrated that RARα stimulants acted on glial cells, resulting in enhanced expression of GDNF and reciprocal downregulation of VEGF, causally limiting vascular permeability by modulating the tight junction function of the BRB-forming capillary endothelium. Furthermore, phenotypic transformation of glial cells mediated by RARα was sufficient for significant reductions of vascular leakage in diabetic retinopathy. Since the net balance of each participant such as VEGF (a permeability-promoting factor) and GDNF (a permeability-inhibiting factor) is important in diabetic retinopathy, it is acceptable to define the antipermeability activity as the expression level of VEGF subtracted from that of GDNF. Since GDNF is a 2.1 kDa protein and is not detectable in the serum in patients with diabetes (28), it is possible that GDNF is not able to permeate across the BRB. This explanation supports the possibility that the GDNF in the mouse eye originated from glial cells in the retinal microenvironment.

There was no apparent retinoic acid response element on the GDNF or VEGF promoter. However, RARα agonists could significantly alter the gene expression of GDNF and VEGF, and ATRA and Am580 trans-activated many signal transduction molecules, including p300/CBP and STAT3, which are known to play important roles in glial cell differentiation (22,23). In a search to find the putative nuclear factors responsible for GDNF expression, activator protein 1 and stimulatory protein-1 binding motifs were also found in both the GDNF and VEGF promoters, which are reported to be trans-activated by a certain type of retinoic acid (29). Therefore, it is feasible that retinoic acids favor multiple signal transduction components important for glial cell differentiation, associating with direct expression modulation in GDNF and VEGF mRNA at the level of the transcriptional machinery.

RARα stimulation of glial cells effectively inhibited the paracellular permeability of endothelial cells in a paracrine manner. We showed that RARα agonists regulated barrier function by modulating the expression of tight junction proteins such as claudin-5, occludin, and zonula occludens-1, although we cannot explain the mechanism by which GDNF induces the alterations in expression of tight junction–associated genes in endothelial cells. One possible explanation is based on evidence showing that GDNF affects a wide variety of signaling pathways, including those of phosphatidylinositol 3-kinase/Akt, extracellular signal–regulated kinase, p38 mitogen–activated protein kinase, and c-Jun NH₂-terminal kinase, which are known to be activated by rearrangement during transfection (RET). Ligation of GDNF to the GDNF-specific receptor GFRα1 leads to dimerization of RET tyrosine kinase (30–32); however, it was somewhat surprising that RET was not expressed in brain microvessels (33). Although further work is needed to address the detailed signal transduction pathway(s) in the endothelial cells in the event of GDNF stimulus, our observations provide important evidence that glial cell–derived cytokines such as GDNF causally limit vascular permeability by modulating the tight junction functions of capillary endothelial cells.

Our results have revealed the functional diversity of the retinal glia and that pharmacological modulation of glial cells serves as a critical barrier-protective mechanism against the development of diabetic retinopathy by regulating endothelial integrity. Given the functional roles of the neural retina in sensing of cellular derangement owing to various metabolic aspects of diabetes, retinoic acids protect against BRB breakdown by suppressing activated phenotypes of glial cells, which potentially provide intraretinal sources of cytokines and chemokines. Since the ocular outcome results from an imbalance of pro- and antipermeable factors, we believe that GDNF is a critical contributing factor for suppressing vascular hyperpermeability in the pathology of retinopathy. While more complex underlying mechanisms appear to be involved in the pathogenesis of diabetic retinopathy, we could successfully clarify that RARα stimulants preferentially inhibited vascular leakiness not only in vitro but also in vivo (Fig. 6). Future studies along this line will pave the way to understanding a wide variety of actions of glial cell–derived cytokines in limiting the progression of diabetic retinopathy. We believe that our results show the potential feasibility of glial cell–targeting therapies for diabetic retinopathy, and these discoveries and subsequent human clinical studies will improve the outcome of diabetes.

This study was supported by grants from the Ministry of Education, Culture, Sports, Science and Technology of Japan and the Japan Diabetes Foundation.

We thank Kim Barrymore for help with this manuscript, the Animal Care Facility of Sapporo Medical University, and Dr. Hiroyuki Kagechika (School of Biomedical Sciences, Tokyo Medical and Dental University, Tokyo, Japan) for the gift of various synthetic retinoids.