Specialized Retinal Endothelial Cells Modulate Blood-Retina Barrier in Diabetic Retinopathy

Diabetic retinopathy (DR) is the most common microvascular complication caused by diabetes mellitus (DM) and leads to vision loss and acquired blindness in patients with diabetes. DR can be divided into two stages based on the microvascular abnormalities of the retina. It ranges from nonproliferative DR (NPDR) to, ultimately, irreversible proliferative DR (PDR) which destroys the retina. It causes frequent retinal exudation and vitreous hemorrhage and contributes to vision loss. The pathology of NPDR is characterized by increased retinal inflammation, retinal vascular permeability, and blood-retinal barrier breakdown (1,2). Malignant NPDR results in uncontrolled proliferation and migration of retinal microvascular endothelial cells (ECs) and capillary formation, leading to retinal neovascularization in PDR (1,2). Currently, there are potential risks and complications associated with the existing therapeutic strategies in the clinic, all of which are aimed at the middle to late stage of DR (3). Thus, searching for early and effective diagnostic biomarkers and treatment targets for DR will reduce the blinding rate of patients with DR.

Retinal vascular cells mainly comprise pericytes and ECs. Pericyte loss is an early pathological feature of DR that is consistently present in the retinas of patients and animals with diabetes (4,5). Retinal ECs undergo a phenotypic switch from normal quiescent to apoptotic and active phenotype (6). Retinal EC dysfunction increases vascular permeability, macular edema, and angiogenesis. Thus, therapeutic intervention based on the regulation of the early stage of DR vascular pathologic disorder would provide a method for the prevention of and protection against diabetes-induced retinal vascular injury. The underlying molecular mechanisms associated with vascular dysfunction, especially endothelial dysfunction, are multifactorial, including advanced glycation end products and receptors, cytokines, chemokines, growth factors, oxidative stress, microRNA, etc. (7), but is still not clear so far.

EC function is closely related to sphingomyelin metabolism, which is essential to maintain EC homeostasis (8). Under pathological conditions, changes in sphingomyelin metabolites and its key enzymes, including sphingosine, ceramide (CER), sphingosine-1-phosphate (S1P), serine, sphingosine kinase (SPHK), CER kinase (CERK), and sphingomyelin-1-phosphate lyase (S1PL), play an essential role in EC function (8). CER is located at the core of the sphingomyelin metabolic pathway and is the center of sphingomyelin biosynthesis and catabolism (9). Inflammation and saturated fatty acids can induce the synthesis of CER (10). Previous studies have shown that the accumulation of CER was involved in the occurrence and development of obesity and diabetes (11,12). Reducing CER synthesis may be beneficial to the prevention or treatment of diabetes and its complications (13). CER can also induce oxidative stress, cell cycle arrest, apoptosis, and necrosis (14); however, the role of sphingomyelin metabolism in the blood-retinal barrier in DR remains unclear. Therefore, it is essential to study the changes in sphingomyelin metabolism in retinal ECs at the early stage of DR, explore the biomarkers of vascular dysfunction at the early stage of DR, and search for corresponding therapeutic targets for the early diagnosis and treatment of DR.

In this study, we performed droplet-based single-cell RNA sequencing (scRNA-seq) (10x Genomics) to explore the cellular composition of retinal ECs in type 2 diabetes at the transcriptomic level and identified a distinct EC population, EC-C4, that uniquely appeared in the early stage of DR. EC-C4 are highly expressed cell markers of ceramidase sphingolipid metabolism, including Acer2, Plpp1, and Sptlc2. Importantly, high ACER2 expression regulated EC barrier function and diabetes-induced retinal vascular integrity. Furthermore, we confirmed the protective role of ACER2 in diabetes-induced vascular leakage and neurovascular disease by intravitreal injection of recombinant ACER2 adeno-associated virus into diabetic mice. This study provides a new potential biological target for the early treatment of pathological vascular permeability in DR.

Detailed descriptions of experimental methods of the current study are provided in Supplementary Materials and Methods.

This study was approved by the Institutional Review Board of the Ethics Committee of the Tianjin Medical University Eye Hospital (Tianjin, China). Informed consent regarding the plasma samples was obtained from each patient following the requirements of the Ethics Committee of the Tianjin Medical University Eye Hospital. Details of the cohort are presented in Supplementary Table 10. Four groups of patients were included in this study, including NDM, patients with cataracts but without DM; NPDR-DME, patients with NPDR and diabetic macular edema (DME); NPDR-NDME, patients with NPDR but without DME; and PDR-A, patients with active PDR. Patients with other vitroretinal diseases or treated with preoperative intravitreal injections were excluded from the study.

All animal examinations were performed following the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Institutional Animal Care and Use Committee of Tianjin Medical University. Male homozygous db/db mice (BKS-Lepr^em2Cd479/Gpt) and db/m (BKS-Gpt) heterozygote mice from the same colony were purchased from GemPharmatech Co., Ltd. and housed in a pathogen-free environment with a 12-h light/dark cycle and free access to food and water.

The sample sizes were designed with adequate power based on the literature and our previous studies. The selected cohort sizes for all cell experiments were sufficient to give a power of 0.8 at an α of 0.05. The mice were randomly assigned to treatment groups. All the analyses were performed in a blinded manner. GraphPad Prism (version 8.0) was used for statistical analyses. All data are presented as mean ± SD.

The data and resources generated or analyzed in this study are available from the corresponding author upon reasonable request.

To explore the cellular composition of retinal ECs in type 2 diabetes at transcriptomic level, we performed droplet-based scRNA-seq (10x Genomics) on CD31 microbead-enriched ECs from control (db/m, n = 2) and type 2 diabetic (db/db, n = 2) mice at 20 weeks (Fig. 1A). Cells from all samples in each group were pooled and sequenced collectively, resulting in a retinal data set consisting of 18,376 cells (Supplementary Table 1) (see Methods). After bath correction, we used t-distributed Stochastic Neighbor Embedding (t-SNE) to partition cells, and classified the entire population into six major cell clusters (ECs, pericytes, photoreceptors, microglia, Müller cells, and neurons) based on their canonical gene expression signatures of Flt1, Pdgfrb, Rp1, Cx3cr1, Rlbp1, and Ttyh1 (Fig. 1B, Supplementary Fig. 1A–C, and Supplementary Table 2).

Further, to gain insight into the molecular differences between control and diabetic ECs, we annotated four EC subtypes and identified venous, arterial, and capillary ECs and EC-C4 using t-SNE. Cell subtypes were defined by established markers, such as arterial Alpl and Bmx, capillary Slc7a5 and Tfrc, and venous Nr2f2 and Bgn (Supplementary Fig. 1D and Supplementary Table 3) (15–18). Interestingly, the scRNA-seq data revealed a unique EC cluster that only exists in the diabetic retina; here, we call it EC-C4 (Fig. 1C and D). We next calculated the proportions of clustered cells to reveal cell-type heterogeneity between control and diabetic ECs. The db/db retina had a reduced proportion of venous (23.2% vs. 29.4%), arterial (5.5% vs. 7.6%), and capillary ECs (50.9% vs. 63.0%) compared with the db/m retina (Fig. 1E and Supplementary Table 3). EC-C4 was only present in db/db retina (20.3%) (Fig. 1E). Conventional cell functional markers were significantly more highly expressed in the corresponding EC subtypes compared with others (Fig. 1F and G), with markers Icam-1, Vcam-1, Lrg1, Lcn2, Acer2, and Plpp1 highly expressed in the EC-C4 subcluster (Fig. 1G and Supplementary Table 4). Gene ontology analysis showed that the vasculature development, hypoxia response, sprouting angiogenesis, and inflammatory response pathways were significantly upregulated in arterial ECs, venous ECs, and capillary ECs (Supplementary Fig. 1E and Supplementary Table 5) (adjusted P value < 0.05). EC-C4 was enriched in inflammation and cell migration pathways (Supplementary Fig. 1E) (adjusted P value < 0.05). Overall, our results showed that a distinct EC population appears in the diabetic retina.

The cross talk between vascular ECs and pericytes plays pivotal roles in microvascular stabilization, remodeling, and barrier function of the vessel wall (19). We identified diverse ligand-receptor interactions among the EC subclusters and pericytes (Supplementary Fig. 2A and B and Supplementary Table 6). EC-C4 interacted with pericytes via inflammatory and cell adhesion receptors Vcam-1 and Sema3c (Supplementary Fig. 2B) (adjusted P value < 0.05). Also, inflammation, leukocyte cell-cell adhesion, angiogenesis, EC migration, sphingolipid metabolism, and cytokine-mediated signaling pathways were significantly upregulated in EC-C4 (Fig. 2A and B and Supplementary Tables 7 and 8) (adjusted P value < 0.05). Among the EC subclusters, markers of cell adhesion, angiogenesis, and sphingolipid metabolism were uniquely expressed in EC-C4 (Fig. 2C and Supplementary Tables 7 and 8) (adjusted P value < 0.05). Sphingolipids, such as CER, sphingosine, and S1P, have roles in the regulation of cell adhesion, migration, inflammation, and angiogenesis (20). Ceramidases represent a family of sphingolipid-metabolizing enzymes, which catalyze the hydrolysis of CERs to generate sphingosine (21) (Supplementary Fig. 2C). Correspondingly, EC-C4 uniquely expressed the cell markers of ceramidases involved in sphingolipid metabolism, including Acer2, Plpp1, and Sptlc2 (Supplementary Fig. 2D and Supplementary Table 8) (adjusted P value < 0.05). The expression level of Acer2 was the highest among the ceramidases markers (Supplementary Fig. 2D and E and Supplementary Table 9). The changes in the expression of VCAM-1 and ACER2 were further validated by immunostaining in the retinal ECs of diabetic mice (Fig. 2D and E and Supplementary Fig. 2F and G). To assess the potential involvement of CER in DR, we determined the CER levels and their quantitative differences in plasma derived from the four groups of human donors (NDM, NPDR-DME, NPDR-NDME, and PDR-A) using highly specific liquid chromatography-tandem mass spectrometry. The results showed that, compared with NDM, the total CER levels in plasma were significantly elevated in NPDR-NDME/DME and further increased in active PDR (Fig. 2F).

To further explore whether the upregulation of ACER2 affects EC barrier function in DR, we first performed loss- and gain-of-function assays in vitro using ACER2 small interfering (siRNA) and adenovirus. The vascular endothelial adhesion molecule VE-cadherin is an important determinant of vascular endothelial barrier function and permeability. Phosphorylation of VE-cadherin at Y685 is relevant for the induction of vascular permeability, whereas phosphorylation at Y731 is involved in the leukocyte extravasation (22). ACER2 knockdown by siRNA further increased CER-induced VE-cadherin phosphorylation at Y685 and Y731 in both human retinal ECs (HRECs) and human umbilical vein ECs (HUVECs) (Fig. 3A and B and Supplementary Fig. 3A and B). Moreover, endothelial junctions were disorganized in CER-treated HRECs or HUVECs, which were further aggravated by ACER2 knockdown. ACER2 deficiency also resulted in a higher incidence of intercellular spaces between ECs compared with control (Fig. 3C and D and Supplementary Fig. 3C and D). We then determined junctional integrity and barrier function by performing transendothelial electrical resistance (TER). TER values were measured using EVOM2 to evaluate changes in the permeability of the HUVECs monolayers. The results showed that the decrease in TER values from baseline after CER treatment was further exacerbated by ACER2 knockdown (Supplementary Fig. 3E). We also detected the levels of sphingosine and S1P, products of CER catalyzed by ACER2 among the four groups. The data indicated that ACER2 deficiency blocked the generation of sphingosine and S1P, which may contribute to the breakdown of endothelial barrier function (Fig. 3E and Supplementary Fig. 3F).

Conversely, ACER2 overexpression by adenovirus (Ad-ACER2) decreased the Y685 and Y731 phosphorylation of VE-cadherin induced by CER treatment both in HRECs and in HUVECs (Fig. 4A and B and Supplementary Fig. 4A and B). Meanwhile, confluent HRECs or HUVECs showed continuous VE-cadherin staining after Ad-ACER2 infection. CER treatment increased endothelial monolayer leakiness, as reflected by a higher incidence of intercellular spaces during VE-cadherin staining, which was significantly rescued by ACER2 overexpression (Fig. 4C and D and Supplementary Fig. 4C and D). Furthermore, we observed that ACER2 increased the TER values, which were reduced by CER stimulation (Supplementary Fig. 4E). In addition, ACER2 overexpression promoted the generation of sphingosine and S1P (Fig. 4E and Supplementary Fig. 4F). Collectively, these results suggested that ACER2 can retard the barrier breakdown induced by CER.

In addition, ACER2 knockdown also increased vascular endothelial growth factor (VEGF)-mediated VE-cadherin phosphorylation at Y685 and Y731 in HRECs (Supplementary Fig. 5A). Inversely, ACER2 overexpression decreased the Y685 and Y731 phosphorylation of VE-cadherin induced by VEGF in HRECs (Supplementary Fig. 5B). Moreover, VEGF-induced tube formation and branching were inhibited in ACER2 knockdown cells, indicating a proangiogenic effect of ACER2 in ECs in angiogenesis, especially in response to VEGF stimulus (Supplementary Fig. 5C). However, the combination of ACER2 and Lucentis, an antiangiogenic ophthalmic agent targeted VEGF, synergistically inhibited VEGF-induced EC permeability (Supplementary Fig. 5D), providing a new therapeutic strategy for treating DR.

To test whether the regulation of ACER2 in vascular barrier function plays a protective role in DR, we generated a recombinant ACER2 adeno-associated virus (AAV-ACER2) and the corresponding control adeno-associated virus (AAV-Ctrl), which were delivered into the retinal vasculature in vivo by intravitreal injection. First, to verify the efficiency of intravitreal AAV injection, the significant upregulation of ACER2 in diabetic retinal tissue and Flag expression among groups was verified using immunofluorescence staining 2 weeks after injection (Fig. 5A). We also recorded the blood glucose, body weight, and blood lipid levels in all three groups (Supplementary Fig. 6A–C). Extravasation of the blood-retinal barrier–impermeable fluorescent tracer, Alexa Fluor 555 cadaverine, was observed in db/db mice, indicating compromised vascular integrity. However, AAV-ACER2 injection rescued diabetes-induced vessel leakiness (Fig. 5B). Furthermore, the fundus fluorescein angiography assay revealed that AAV-ACER2 administration alleviated retinal leakage caused by diabetes compared with AAV-Ctrl injection (Fig. 5C). Tight junction proteins, including ZO-1 and Claudin-5, are critical for determining the retinal vascular permeability and regulating vascular barrier function (23). Thus, we performed double staining for ZO-1 and Claudin-5 in db/m and db/db mice infected with AAV9-Ctrl or AAV9-ACER2 delivered by intravitreal injection. In the diabetic retina, the Claudin-5 and ZO-1 junction distribution was discontinuous. Injection of AAV9-ACER2 rescued junction destruction, suggesting that ACER2 administration protected against vascular leakage by preventing tight junction defects (Fig. 5D).

We next examined whether the overexpression of ACER2 affected glial function and inflammatory response. As illustrated in Fig. 6A, db/db mice displayed increased glial fibrillary acidic protein (GFAP) expression in astrocytes and the end feet of the Müller glia of the retinas, which was limited in db/m mice injected with AAV-Ctrl. However, AAV-ACER2 injection in db/db mice decreased the upregulation of GFAP in astrocytes and Müller glia compared with db/db mice treated with AAV-Ctrl (Fig. 6A and B). We also determined the CER levels and their quantitative differences in vitreous humor of db/m mice and db/db mice with AAV injection. The results showed that, compared with db/m mice with AAV-Ctrl injection, the total CER levels were significantly elevated in db/db mice delivered AAV-Ctrl, which was then reversed after ACER2 overexpression (Fig. 6C). In addition, diabetes-induced ICAM-1, VCAM-1, and TNFα expression was suppressed after AAV-ACER2 administration (Fig. 6D). Moreover, ACER2 overexpression decreased the numbers of TUNEL-positive cells, which were significantly increased in the retinas of db/db mice treated with AAV-Ctrl compared with the retinas of db/m mice (Supplementary Fig. 6D). Additionally, an electroretinogram assay was used to determine the electrical activity of retinal neurons and glia, and estimate the extent of retinal dysfunction (24). The results showed that db/db mice decreased amplitudes of both a- and b-waves in a light intensity–dependent manner compared with db/m mice. As expected, intravitreal injection of AAV-ACER2 significantly reduced the attenuation of a- and b-wave amplitudes compared with treatment with AAV-Ctrl (Fig. 6E). Subsequently, we examined acellular capillaries to assess the effect of ACER2 on DR vasculopathy (Fig. 6F and G). Notably, db/db mice treated with AAV-Ctrl showed significantly increased numbers of retinal acellular capillaries compared with db/m mice. In contrast, AAV-ACER2 injection significantly decreased numbers of retinal acellular capillaries compared with AAV-Ctrl (Fig. 6F). In summary, ACER2 overexpression retarded diabetic retinal inflammatory response and neurovascular disease.

The incidence of DR has increased dramatically. However, its underlying pathogenesis remains unclear. Retinal ECs are the main targets of diabetes-induced vascular injuries. Aberrant retinal EC function causes blood-retinal barrier breakdown and vascular dysfunction (25). In this study, we defined a unique EC-C4 subcluster that appeared only in the diabetic retina, using scRNA-seq analysis of retinal cells from control and diabetic mice. The characterization of genes as EC-specific markers revealed that ACER2, a ceramidase of the sphingolipid metabolic process, was highly expressed in diabetic retinas. Importantly, the upregulation of ACER2 modulated EC barrier function and retinal vascular integrity. We also verified the protective role of ACER2 in diabetes-induced vascular leakage and neurovascular disease, by intravitreal injection of recombinant ACER2 adeno-associated virus into diabetic mice. Our findings suggest that ACER2 is an effective therapeutic agent for treating pathological vascular permeability in DR.

The most significant finding of our study was the new EC-C4 subtype in the diabetic retina, which was linked to inflammatory response and EC migration. Retinal vessel integrity depends on the adhesion of EC to extracellular matrix and adjacent ECs (26,27). Endothelial dysfunction causes vascular barrier breakdown and inflammatory response that is closely associated with DR progression (28). Our scRNA-seq data identified several markers to distinguish the unique EC-C4 cells in the diabetic retina, among which cell adhesion molecules VCAM-1 and ICAM-1 typically exist at the basolateral surface of activated ECs and are highly expressed in the EC-C4 subcluster (29). Moreover, LRG1 and LCN2 have been implicated in inflammatory disorders and angiogenesis, they function as inflammatory markers, and they are upregulated in metabolic dysfunction (30,31). Another two makers of EC-C4, ACER2 and PLPP1, are effective targets in sphingolipid metabolism-associated inflammatory and cardiovascular diseases (32,33). Furthermore, differentially expressed gene enrichment analysis between EC-C4 and another three EC subcluster showed that the EC-C4 subcluster displayed functional enrichment in biological behaviors related to inflammatory response, leukocyte cell-cell adhesion, cytokine-mediated signaling pathway, angiogenesis, EC migration, and sphingolipid metabolic processes. This indicated that the unique EC-C4 subtype annotated in the diabetic retina may play an essential role in the early stages of DR.

A cell-cell interaction map of scRNA-seq analysis in our study showed that the new EC-C4 subtype may be responsible for cross talk with pericytes, thereby regulating retinal inflammation and cell adhesion. As the key component of retinal microvasculature, the normal pericyte-EC interplay is critical for the development and homeostasis of the retinal vasculature. ECs and pericytes communicate in adhesion plaques enriched with the extracellular matrix via gap junctions or other paracrine signaling factors (19,34). Tight adherens junctions between pericytes and endothelial, neuronal, and glial cells regulate the blood-retina barrier. Pericyte-EC interactions in pathological conditions disturbed by diabetes increase leukocyte adhesion, inflammatory cytokines, and chemokines, leading to blood-retina barrier breakdown and severe retinal vascular defects (35). Therefore, therapeutic interventions based on the regulation of pericyte-EC cross talk may protect against diabetes-induced retinal vascular injury. In our study, we identified all the altered ligand-receptor pairs between pericytes and the four different EC subtypes annotated in the scRNA-seq analysis. The interaction scores implied that the top ligands, such as VCAM-1 or SEMA3C, secreted by the new EC-C4 subtype were involved in cell adhesion and inflammation, indicating the important function of EC-C4 in communicating with pericytes in the retinal inflammatory response. Thus, we speculate that a further understanding of the cellular and molecular mechanisms of pericyte–EC-C4 interaction and perturbation may help in designing therapeutic interventions for preventing and treating retinal vascular disorders.

Our next main finding was that the highly expressed ceramidase ACER2 in the EC-C4 subcluster retarded barrier breakdown induced by CER in vitro and rescued diabetes-induced vessel leakiness in mouse models. Altered sphingolipid and glycosphingolipid metabolism causes several retinal diseases. Furthermore, increased sphingolipid composition may contribute to metabolic stress, resulting in retinal inflammation and neurodegeneration in DR (36). The conditions present in diabetes also promote sphingolipid production and the expression of enzymes involved in sphingolipid metabolism. Ceramidases catalyze the hydrolysis of CERs to sphingosine, which is further phosphorylated to form S1P. The S1P-S1P1 receptor signaling system plays an essential role in maintaining vascular integrity and regulates vascular permeability and inflammation in diabetes (37,–,39). The enriched ACER2 in EC-C4 subtypes may partially reduce CER content and contribute to the composition of S1P, which is an important regulator of cell adhesion and adherens junction assembly. Our metabolomics data also indicate a primary composition of the accumulating CER in the plasma of patients with diabetes, which may be a negative feedback regulation of ACER2 expression. However, the self-protective mechanism of ACER2 upregulation was not sufficient to reverse the massive damage induced by DR. We then injected AAV-ACER2 into mouse models, which significantly decreased the CER content. Moreover, the proangiogenic effect of ACER2 in VEGF-mediated angiogenesis may be attributed to the production of S1P, which has been shown to stimulate tube formation in ECs and to induce angiogenesis in vivo (40,41). The combination of ACER2 and Lucentis, an antiangiogenic ophthalmic agent targeted VEGF, synergistically inhibited VEGF-induced EC permeability, making up the proangiogenic effect of ACER2 and amplifying the protective role of ACER2 in EC barrier function and retinal vascular integrity. Thus, promoting the protective role of ACER2 is critical for the early treatment of pathological vascular permeability in DR by altering the ratio of CER to S1P. In addition, the composition of CERs and S1P in the plasma of patients may function as a clinical biomarker of the early stage of DR, and the combination of ACER2 and anti-VEGF drugs provides a new therapeutic strategy for treating DR.

Our study has several limitations. First, scRNA-seq was performed on diabetic ECs from db/db mice at 20 weeks of age, which reflects the early stage of DR. The different time points of db/db mice and the sample size used for scRNA-seq should be considered to fully understand the development and progression of diabetes. In addition, animal models and human diabetes exist differently; db/db mouse models only recapitulate part of the pathological complexity of patients with DR. Further studies such as the correlation analysis between ACER2 or the corresponding sphingolipid metabolism and patients with different stages of DR are necessary to translate this information to humans. Finally, the mechanism underlying ACER2 upregulation in diabetic ECs and the effect of ACER2 on retinal vascular barrier function requires further investigation.

Funding. This work was supported by grants from Tianjin Key Medical Discipline (Specialty) Construction Project (TJYXZDXK-037A), from National Natural Science Foundation of China (82171085, 81970392, 82370440) to X.L. and H.J., and from The Science & Technology Development Fund of Tianjin Education Commission for Higher Education (2021KJ234) and Open Project of Tianjin Key Laboratory of Retinal Functions and Diseases (2021tjswmq001) to X.Y.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.A., H.J., H.L., and X.L. designed the research; X.Y. and H.L. did most of the experiments; D.A. supervised and analyzed scRNA-seq; X.Y. did LC-MS/MS with help from Z.Z., W.Z., J.D., D.A., and H.J.; R.L., T.N., B.C., and Y.L. provided mouse model study support; H.J. and H.L. supervised the project; and X.Y. and H.L. wrote the manuscript. All authors read and approved the final manuscript. X.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.