Introduction

While the prevalence of diabetes mellitus (DM) has been rapidly growing over the past decades, diabetic retinopathy (DR) becomes one of the leading causes of blindness worldwide [1]. DR is characterized as the complication of retinal microvascular and blood-retina barrier (BRB) system that involves inflammation and neuronal injury [2].

Consisted of a monolayer of specialized epithelial cells located between the outer neural retina and choroid and forms the outer BRB, the retinal pigment epithelium (RPE) is essential in maintaining retinal homeostasis and easily impaired in the progress of DR, resulting in outer retinal abnormalities [3]. The RPE interacts with the photoreceptors to sustain visual functions with phagocytosis of photoreceptor outer segments, involvement of visual cycle, production of neurotrophins, and transportation of physiological materials (transporting glucose, fatty acids, water, and ions from blood to the outer retina; transporting metabolites produced by the outer retina to blood) [4, 5]. During DM, the impaired and malfunctioning RPE gives rise to worsen integration of the outer BRB, breakdown of cellular microenvironment, jeopardized photoreceptors, and decreased electrophysiology [3], proving that the RPE could be an important site of action in fighting against DR.

In order to clarify the pathological progress of RPE in DR, multiple studies focused on the molecular alterations of RPE cell lines under varying conditions of high glucose (HG) in vitro. Pathological molecular changes in inflammation, oxidative stress, mitophagy interruption, autophagy disturbance, genotoxic stress, ferritinophagy, ferroptosis, apoptosis, pyroptosis, and epithelial-mesenchymal transition have been reported to occur in the HG-treated RPE cells [6,7,8,9,10]. To the best of our knowledge, no systemic research has been done to profile the molecular variations in human primary RPE under HG. Moreover, most of the previous studies were based on infinite cell lines, where the RPE cells were cultured in HG for short period of time.

Under these circumstances, additional experiments should be conducted to overview the changing molecules of RPE cells in an environment resembling DM to realistically present the status of DR. Thus in this study, we investigated the variations of human primary RPE cells underwent prolonged treatment of HG. The cellular biology of HG-treated RPE cells was observed, while exploration on the whole transcriptome information was performed. Through providing a transcriptional atlas of human primary RPE cells treated with protracted HG, this research is meant to extend the understanding of RPE’s molecular mechanism in the pathological process of DR.

Methods

Human primary RPE cells

The bulbus oculi of 5 donors aged 20–40 years old were obtained from the Eye Bank of Guangdong Province, State Key laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University. The accidentally deceased donors did not have any ophthalmic or systemic diseases before death. Informed consents for use in scientific research were obtained from the donors or donors’ families by the Eye Bank of Guangdong Province. All the experiments were conducted in accordance with the Declaration of Helsinki. All the experiments were approved by the Institutional Review Board and the Medical Ethics Committee of State Key laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University (Approval NO. 2021KYPJ196). The anterior segments, vitreous bodies, and neural retinae of the bulbus oculi were removed to form eye cups. Later, the human primary RPE cells were isolated in the method referring to the former study [11]. The RPE cell type was verified with short tandem repeat analysis. The RPE cells were cultured in a culture medium consisted of 89% Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F-12; Gibco, Thermo Fisher Scientific, USA) culture medium with 10% fetal bovine serum (FBS; Gibco, Thermo Fisher Scientific, USA) and 1% penicillin/streptomycin (HyClone, GE Healthcare Life Science, USA), under a constant temperature condition at 37 ℃, in the 5% CO2-humidified atmosphere. All the RPE cells were passaged every 48 h when the cell density reached between 70 to 80%, and seeded in a density of 1 × 105 cells per cm2.

HG treatment

Human primary RPE cells of the third generation of were subjected to the protracted treatment of HG. The glucose concentration of the medium supplied for the treated group was adjusted to 35 mM. The glucose concentration in HG condition was set as 35 mM to double the glucose concentration in regular DMEM/F-12 culture medium where RPE cells are routinely cultured. Due to the mortality of human primary cells in culture, the whole passage number was usually limited under 6 to 7 generations. In order to better reflect the conditions of the RPE cells under HG but not decline in cell line, we kept the experimental loop within 6 passage generations: the human primary RPE cells were consistently cultured in the control (CTRL) condition or the HG condition for 3 generations (144 h/6 days) to protract the treatment before further examinations. Culture medium with mannitol of 25, 35, and 45 mM were applied to the human primary RPE cells to monitor the biological variations under different osmotic pressure.

Flow cytometry

MUSE Annexin V & Dead Cell Kit (Merck Millipore, Germany) was used for cell viability analysis. MUSE Oxidative Stress Kit (Merck Millipore, Germany) was used for cellular oxidative stress assay to detect reactive oxygen species (ROS) positive cells. Cell Cycle Detection Kit (BD Pharmingen, USA) was used for cell cycle assay. The RPE cells were separated and centrifuged after 3 generations of CTRL/HG treatment. The cells were then re-suspended and incubated with the reagents following the instructions of the kits. Later, the cell suspensions were detected on flow cytometer after thoroughly mixing.

Wound healing assay

The RPE cells were cultured in 12-well plates after 3 generations of CTRL/HG treatment. When the confluence of the RPE cells reached 80%, standard 200-µl pipette tip were used to create scratching wounds in the cell cohort. Photos were taken immediately after scratching (0 h), and 24/36 h later. The areas of wounds were measured using ImageJ software (National Institutes of Health, USA) and compared with that of the 0 h.

Cell activity

Cell Counting Kit 8 (CCK8; Dojindo, Japan) was utilized to detect the physiological activity of the RPE cells under CTRL or HG conditions. At the end of the third generation of treatment, the cells were seeded in 96 well plates at a density of 2 × 103 cells/well. The CCK8 assay was performed following the instruction. At the beginning and at the end of a 4 h incubation with the CCK8 reagents, the optical density at 450 nm wavelength (OD450) was measured and recorded.

Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) assay

After 3 generations of CTRL/HG treatment, cellular RNAs of the RPE cells were extracted by using Trizol reagent (Invitrogen, USA) and nucleic acid quality controlling was assessed. Subsequently, cDNA reverse transcription was performed by using the PrimeScript™ RT reagent Kit with gDNA Eraser (TaKaRa, Japan). SYBR Green Master Kit (Roche, Switzerland) was used to conducted the qPCR assay with the LightCycler 96 System (Roche, Switzerland). The endogenous control used was β-ACTIN (ACTB, actb). The primer sequences utilized are presented in Table 1.

Table 1 The primers used in this study
Full size table

Whole transcriptome

Total RNA was used for whole transcriptome. The quality of total RNA had reached the level of 28S/18S ≥ 1.5:1 and RIN ≥ 7, with initial quantities of 0.1–1 μg. Poly A-mRNA purification and fragmentation, 1st strand cDNA synthesis, 2nd strand cDNA synthesis, end repair and A-tailing and adapter ligation, PCR enrichment, library quality control, cluster generation, and sequencing by synthesis were gradually carried out as described before [12].

Differentially expressed genes (DEGs) screening

Based on the transcriptome dataset, we performed background calibration, normalization, and log2 transformation between the HG and CTRL groups. The results were analyzed using the Limma R package (version 3.40.6) [13, 14]. The data were firstly converted into log2 formation. Multiple linear regression were then performed, using the lmFit function. The significance was assessed in a Bayes framework implemented in the eBayes function with Fold change > 2 and P value < 0.05, for identifying the DEGs.

Weighted gene co-expression network analysis (WGCNA) and module gene selection

WGCNA was applied to explore the correlation of genes between the groups, based on the transcriptome dataset. The results were calculated and determined using the WGCNA R package (version 4.3.1) [15]. Based on the dataset, Median Absolute Deviation was calculated for each gene. Fifty percent of genes with the largest determined median absolute deviation was filtered to construct the DEGs matrix with the GoodSamplesGenes function. A scale-free co-expression network was then constructed. All pair-wise genes were processed with the Pearson’s correlation matrices and average linkage. A weighted adjacency matrix was constructed with the adjacency determined by the co-expression-similarity-derived threshold, with a power function A_mn =|C_mn|^β. The adjacency was transformed into a topological overlap matrix, with the β power of 9. The dissimilarity and gene ratio were also determined. Average linkage hierarchical clustering was conducted by using a dissimilarity measure based on the topological overlap matrix to group genes with similar expression profiles into different modules with a minimum size of 30 genes. The dissimilarity of the module eigen genes was calculated to further analyze the modules, with a cut-off point for the module dendrogram.

Functional enrichment and intersection analysis

To perform the functional enrichment analysis, the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways database, the WIKI pathway database, the Reactome, and the Gene Ontology (GO) system were utilized. The criteria was set a pValue < 0.05. Venn diagrams were used to analyze the intersections between the database-screened-out genes. The analyses were performed based on the differentially expressed genes, the intersection of DEGs and the genes of the most significant modules in WGCNA, and the intersection of the DEGs and the diabetic retinas’ genes from the GSE102485 dataset.

Protein–protein interaction

The networks of interaction between proteins was established by using the GENEMANIA database (http://morrislab.med.utoronto.ca) to explore the interactions between protein-coding genes screened out by the functional enrichment analysis.

Statistical analysis

All experiments were performed with with three replicates. Statistical analysis was performed with IBM SPSS Statistics software (IBM, USA). The data were presented as mean ± standard deviation (SD). Unpaired two tailed Student’s t-test or one way ANOVA with the Bonferroni’s post hoc test were used to determine the significances between groups. Significant P values are presented as * p < 0.05 vs. CTRL group, ** p < 0.01 vs. CTRL group, and *** p < 0.001 vs. CTRL group, while no significance is presented as ns. GraphPad Prism software (GraphPad Software Inc., USA) was used for graphical plotting.

Results

Protracted HG decreases cellular functions of human primary RPE cells

Cultured in the HG condition for three generations, the human primary RPE cells exhibited different extents of decreased cellular functions. The cell viability was significantly reduced by the protracted HG treatment (Fig. 1A, B). The osmotic pressure of 35 mM mannitol matches that of 35 mM glucose. To monitor the impact of osmotic pressure on the cells, we subjected the cells to culture medium with 25, 35, and 45 mM mannitol. After 3 generations, no change in cell viability was observed between the CTRL and mannitol-treated cells, indicating that the impact of osmotic pressure on this study is not significant (Fig. S1). Further experiment showed increased ROS induced by HG in the RPE cells (Fig. 1C, D). In addition, obvious G2/M arrest was observed in the prolonged HG-treated RPE cells (Fig. 1E, F). Wounds were scratched after HG treatment of three generations. The cells were still cultured in the same HG condition during the whole course of observation in which the photos were taken after 0, 12, 24, 36 h. It was found that the protracted HG had apparently impeded the migration capability of RPE cells (Fig. 1G, H). Decreased cell activity was also found in the HG group with CCK8 examination (Fig. 1I). These results indicates the cellular functions of human primary RPE cells hindered by protracted HG treatment, comparing with the CTRL cells.

Fig. 1
figure 1

Human primary RPE cells was impaired by protracted HG treatment. (A) Flow cytometry detected significantly decreased cell viability (B) in the protracted HG-treated RPE cells. (C) The ratio of ROS positive cells was higher (D) in the HG group comparing with the CTRL group. (E) Cell cycle assay showed a significant G2 phase arrest (F) in the RPE cells cultured under protracted HG condition. (G) Wound healing assay exhibited apparent decreased migration capability of the HG-treated RPE cells (H) in durations of 24 h and 36 h. (I) Declined cell activity was observed in the HG group by using CCK8 assay. Scale bar = 500 μm; CTRL = control, HG = high glucose; Histogram = mean ± SD, *p < 0.05, **p < 0.01, ***p < 0.001; ns = no significance

Full size image

Identification of DEGs and functional enrichment analysis

Based on the transcriptome dataset, 4125 DEGs were screened out between the protracted HG-treated and CTRL human primary RPE cells. The significant DEGs were defined with the thresholds of Fold-change > 2 and the cutoffs of P value < 0.05. Among these, 2768 genes were up-regulated and 1357 genes were down-regulated in the HG group (Fig. 2A). Enrichment analyses using the KEGG pathway, WIKI pathway, and Reactome databases were performed to explore the DEGs. The top 10 pathways in each of the functional enrichment were selected for further evaluation (Fig. 2B–D, Figs. S1, S3, and S4). The intersection of the enriched genes were identified (Fig. 3A, B). The differential expression of the 21 genes were exhibited in a heat map (Fig. 3C), among which 12 genes were up-regulated and 9 genes were down-regulated by the protracted HG. With RT-qPCR, transcriptional verification was performed on the intersected genes (Fig. 3D), where the genes were significantly regulated in consistent with the transcriptome.

Fig. 2
figure 2

Step 1: transcriptome analysis to collect transcriptional database of the RPE cells. (A) Volcano scatter plot presents the DEGs of the RPE cells with up-regulated and down-regulated gene expression. The up-regulated genes in the protracted HG-treated RPE cells were 2768, and the down-regulated genes were 1357. (B) Bubble plot shows the top 10 pathways enriched with the KEGG pathway database based on the DEGs. (C) Bubble plot shows the top 10 pathways in the enrichment with the WIKI pathway database. (D) Bubble plot shows the top 10 pathways found in the Reactome database, enriched with the DEGs.

Full size image
Fig. 3
figure 3

Step 1: identification of the key genes within DEGs. (A) Venn diagram shows a total of 21 genes were identified from the intersection of the top 10 pathways enriched by the KEGG, WIKI, and Reactome databases respectively. (B) Sankey diagram shows the affiliation of the 21 genes to the KEGG, WIKI, and Reactome databases. (C) Heat map shows the differential expression of the 21 genes in the transcriptome dataset. Among them, 12 genes were up-regulated by protracted HG. (D) Transcriptional verification of the 21 genes: With RT-qPCR, the genes were found to be significantly regulated and in consistent with the transcriptome dataset. CTRL = control, HG = high glucose; Symbols assigned to pathways: A1 = Ferroptosis, A2 = HIF-1 signaling pathway, A3 = Legionellosis, A4 = PI3K-Akt signaling pathway, A5 = Rheumatoid arthritis, A6 = Protein processing in endoplasmic reticulum, A7 = MAPK signaling pathway, B1 = FERROPTOSIS, B2 = PHOTODYNAMIC _ THERAPYINDUCED _ NFKB _ SURVIVAL _ SIGNALING, B3 = PARKINUBIQUITIN _ PROTEASOMAL _ SYSTEM _ PATHWAY, B4 = ANGIOGENESIS, B5 = PHOTODYNAMIC _ THERAPYINDUCED _ UNFOLDED _ PROTEIN _ RESPONSE, B6 = UNFOLDED _ PROTEIN _ RESPONSE, B7 = TRANSCRIPTIONAL _ CASCADE _ REGULATING _ ADIPOGENESIS, B8 = OLIGODENDROCYTE _ SPECIFICATION _ AND _ DIFFERENTIATION _ LEADING _ TO _ MYELIN _ COMPONENTS _ FOR _ CNS, B9 = MAMMARY _ GLAND _ DEVELOPMENT _ PATHWAY _ PREGNANCY _ AND _ LACTATION _ STAGE _ 3 _ OF _ 4, C1 = TRANSPORT _ OF _ SMALL _ MOLECULES, C2 = INTERLEUKIN _ 4 _ AND _ INTERLEUKIN _ 13 _ SIGNALING, C3 = HSF1 _ DEPENDENT _ TRANSACTIVATION, C4 = RESPONSE _ OF _ EIF2AK1 _ HRI _ TO _ HEME _ DEFICIENCY, C5 = UNFOLDED _ PROTEIN _ RESPONSE _ UPR, C6 = PERK _ REGULATES _ GENE _ EXPRESSION, C7 = ATF4 _ ACTIVATES _ GENES _ IN _ RESPONSE _ TO _ ENDOPLASMIC _ RETICULUM _ STRESS; Histogram = mean ± SD, *p < 0.05, **p < 0.01, ***p < 0.001; ns = no significance

Full size image

WGCNA with key module analysis

To explore the correlation network of gene expression, modularization was performed on the transcriptome dataset by utilizing WGCNA. Based on the average connectivity and scale independence, β value of 9 with an R2 of 0.85 was set as the soft threshold (Fig. 4A, B). A total of 28 gene co-expression modules were identified and compared (Fig. 4C, D). The generated gene co-expression modules were categorized according to the glucose treatment, and the verification was calculated along with significance (Fig. 5A). Among all, darkorange module (gene number = 8188, correlation coefficient = -0.99, p = 1.7 × 10−4) exhibited significant verification in the HG group (Fig. 5A). The correlations between module membership and gene significance of darkorange module was explored (Fig. 5B). While the correlation was significant, the r value was 0.97. Later, the intersection between the DEGs and darkorange was investigated (Fig. 5C). Owing to the reason that the sample size explored in this study was relatively small, we performed further pathway enrichments to reduce the impact of network noise generated from WGCNA. Therefore, the 1495 genes filtered by the intersection were enriched with the GO biological processes (Fig. 6A, Fig. S5), GO cellular components (Fig. 6B, Fig. S6), and GO molecular functions (Fig. 6C, Fig. S7) databases. The top 5 pathways were respectively filtered, where the enriched genes were intersected again to narrow down the range of targets (Fig. 7A). A combination of 12 genes were screened out (Fig. 7A, B). Among the 12 genes, 9 ones were up-regulated in the HG group, while the others were down-regulated (Fig. 7C).

Fig. 4
figure 4

Step 2: modularized gene expression analysis through WGCNA. (A, B) The soft threshold was selected as β = 9 according to the combined analysis of scale independence and average connectivity. (C) Heat map shows the 28 modules of eigengene adjacency. (D) Dendrogram shows the gene co-expression modules represented in different colors. CTRL = control, HG = high glucose

Full size image
Fig. 5
figure 5

Step 2: further analyzing the significant module from WGCNA. (A) The module-expression correlation is presented in a heat map. Numbers at the upper brackets represent the correlation coefficient, while those at lower brackets are p-value. HG treatment was significantly varied in the darkorange module. (B) The correlation between module membership and gene significance for the darkorange module is presented in dot plots. (C) A total of 1495 genes were detected with the intersection of DEGs and the genes in the darkorange module screened out with WGCNA, which is presented in a venn diagram

Full size image
Fig. 6
figure 6

Step 2: enrichments based on the results of WGCNA. Bubble plots present the top 5 pathways enriched from the genes in the darkorange module by utilizing databases of (A) the GO biological processes, (B) the GO cellular components, and (C) the GO molecular functions

Full size image
Fig. 7
figure 7

Step 2: identification of the crucial genes based of WGCNA. (A) Venn diagram shows the intersection of genes enriched out by the GO biological processes, GO cellular components, and GO molecular functions databases. A total of 12 genes were spotted. (B) The genes affiliated to different GO databases are presented in sankey diagram. (C) Heat map shows the regulating levels of the 12 genes screened out by the intersection. Among them, 9 genes were up-regulated by protracted HG, while there were 3 genes down-regulated. CTRL = control, HG = high glucose; Symbols assigned to pathways: a1 = RESPONSE _ TO _ TOPOLOGICALLY _ INCORRECT _ PROTEIN, a2 = CELLULAR _ RESPONSE _ TO _ TOPOLOGICALLY _ INCORRECT _ PROTEIN, a3 = RESPONSE _ TO _ ABIOTIC _ STIMULUS, a4 = RESPONSE _ TO _ ENDOPLASMIC _ RETICULUM _ STRESS, a5 = EXTERNAL _ ENCAPSULATING _ STRUCTURE _ ORGANIZATION, b1 = ENDOPLASMIC _ RETICULUM _ LUMEN, b2 = ENDOPLASMIC _ RETICULUM, b3 = ENDOPLASMIC _ RETICULUM _ CHAPERONE _ COMPLEX, b4 = EXTERNAL _ ENCAPSULATING _ STRUCTURE, c1 = MISFOLDED _ PROTEIN _ BINDING, c2 = RIBONUCLEOTIDE _ BINDING, c3 = UNFOLDED _ PROTEIN _ BINDING, c4 = PLATELET _ DERIVED _ GROWTH _ FACTOR _ RECEPTOR _ BINDING, c5 = C3HC4 _ TYPE _ RING _ FINGER _ DOMAIN _ BINDING

Full size image

Investigation crossing with the gene expression omnibus (GEO) database

In order to better clarify the importance of DEGs from the RPE cells, we conducted a crossed investigation using the GSE102485 dataset obtained from the GEO database. The GSE102485 dataset contains a set of transcriptome profiles collected from 30 samples of normal subjects and patients with proliferative DR [16]. In this study, we used the data of retinal tissues, excluding those of proliferative membrane, for the DEGs were screened out from cells but not secreted organism. The GSE102485 dataset was exploded with Student’s t-test, and 2669 genes were found to be significantly regulated upon fold change ≥ 2. The intersection of the datasets was found to be a combination of 96 genes (Fig. 8A). The top 10 pathways of the intersection enriched with the KEGG database were included (Fig. 8B). Among the enriched genes, 13 genes were up-regulated both in the HG-treated RPE cells and the diabetic retina from GSE102485, while there was none simultaneously down-regulated (Fig. 8C, D). Additionally, 1 gene was up-regulated in the cells but down-regulated in GSE102485, while there were 10 genes experienced down-regulation in the transcriptome dataset but up-regulation in GSE102485 (Fig. 8C).

Fig. 8
figure 8

Step 3: gene screening overlapping the GEO dataset. (A) Venn diagram shows the intersection of 96 genes detected between the DEGs and the GEO (GSE102485) dataset. (B) The top 10 pathways which were enriched based on the intersected genes with the KEGG pathway database is show in a pie chart. (C) Cross diagram shows the regulated genes screened out by the KEGG enrichment: 13 genes were simultaneously up-regulated in the transcriptome and GEO datasets GSE102485. None was simultaneously down-regulated. (D) The regulating levels of the 13 genes from the transcriptome dataset are presented in a heat map. CTRL = control, HG = high glucose

Full size image

Further exploration on the changing genes

Overlapping was performed based on the vitreous cytokines screened out by a former study to further explore the transcriptional significance of the current system [17]. Among all, 14 reported cytokines (angiopoietin 2, ANGPT2; C–C motif chemokine ligand 2, CCL2; C–C motif chemokine ligand 5, CCL5; cellular communication network factor 2, CTGF; interleukin 4, IL4; macrophage migration inhibitory factor, MIF; matrix metallopeptidase 1, MMP1; matrix metallopeptidase 3, MMP3; matrix metallopeptidase 7, MMP7; platelet derived growth factor subunit B, PDGFB; platelet derived growth factor D, PDGFD; transforming growth factor beta 1, TGFB1; transforming growth factor beta 3, TGFB3; vascular endothelial growth factor A, VEGFA) were detected by the transcriptome in this study, where VEGFA, CCL2, MMP3, PDGFB, and PDGFD were significantly up-regulated in the HG group (Fig. 9A). These genes were then analyzed with the CellTalker database, and the receptors in correspondence were retraced (Fig. 9B). The receptors were then backtracked in the transcriptome dataset, in which only kinase insert domain receptor (KDR) was found to be apparently up-regulated with fold change ≥ 2. According to the results above, it is indicated that VEGFA (which is the ligand of KDR) was the most significantly activated gene in RPE cells under prolonged exposure to high glucose.

Fig. 9
figure 9

Further analysis to screen out the key regulated genes in protracted HG-treated human primary RPE cells. (A) Volcano plot shows the hotspot genes detected in the present transcriptome dataset as target vitreous cytokines from DR patients reported in the former study [17]. (B) CellTalker database was applied to explore the ligand-receptor interaction to explore the receptors of VEGFA, CCL2, PDGFD, PDGFB, and MMP3. No significant receptor was found for MMP3 in the current transcriptome dataset. (C) Venn diagram shows the intersections of the 3 steps investigated previously. Only 2 genes, VEGF and ERN1, were detected as the common intersection. (D) The protein–protein interactions of the genes from all 3 steps were explored. The linked functions includes regulation of endothelial cell proliferation, regulation of vasculature development, and regulation of response to endoplasmic reticulum (ER) stress. The inner circle is consist of the genes screened out from all 3 steps (highlight VEGF and ERN1). The outer circle is consist of the related hub genes. Step 1 = transcriptome analysis to collect transcriptional database of the RPE cells, identification of the key genes within DEGs; Step 2 = modularized gene expression analysis through WGCNA, identification of the key genes based of WGCNA; Step 3 = gene screening via utilizing the GEO dataset

Full size image

The intersection of the genes acquired in the former steps (Result sections: “Identification of DEG and functional enrichment analysis”, “WGCNA with key module analysis”, and “Investigation crossing with the gene expression omnibus (GEO) database”) was also calculated, narrowing the DEGs to 2 (VEGFA and endoplasmic reticulum to nucleus signaling 1, ERN1) (Fig. 9C). The protein–protein interactions amongst all the genes filtered in the former 3 steps were explored, exhibiting a complex network of 32 genes identified by the transcriptome and 20 related genes (Fig. 9D, supplementary document 1 and 2). The 32 genes were briefly reviewed in supplementary document 3. The most significantly interacted functions were regulations of endothelial cell proliferation, vasculature development, and response to endoplasmic reticulum (ER) stress (Fig. 9D).

Discussion

Becoming the leading cause of blindness in modern society, DR is one of the arduous challenges in ophthalmology. DR is characterized as a microvascular complication in the ocular system, while retinal vasculature abnormalities and BRB breakdown are the significant pathology features [3]. As retinal microenvironment deteriorates, the damages worsen, ultimately resulting in irreversible visual impairment. While the RPE is major in maintaining retinal homeostasis, its importance in developing DR is gradually receiving attention in recent years. The RPE is one of the pathological sites during the process of DR: adverse alterations in the RPE were observed in diabetic rodents [18], while patients with DR could suffer abnormal autofluorescence and RPE thinning [19, 20]. HG and diabetic inflammation lead to increased oxidative stress, enhanced pro-inflammatory genes, disturbed mitophagy/autophagy, strengthened genotoxicity, activated endothelial-mesenchymal transition, undermined integrity of RPE tight junctions, resulting cell death [6,7,8,9,10, 21]. Different pathways participate in the hazards of HG on the RPE cells [3, 21,22,23]. Only a few studies worked on the RPE under prolonged HG. It was found that protracted HG caused oxidative damage to aRPE-19 cells that phagocytized oxidized rod outer segments [24]. Another study reported premature senescence induced by oxidative stress and autophagy in aRPE-19 cells under long-term HG [6]. However, there is little attention paid to the overall molecular variations in RPE under protracted HG exposure. Besides, although the injury in RPE is valued in former studies, the contribution of the RPE to DR remains to be further studied. Therefore, in this study, we systemically explored the transcriptional patterns of human primary RPE cells under protracted HG, aiming to provide better understanding and new insights for DR.

It was not commonly reported in previous studies that RPE cells were cultured and investigated under protracted HG. In vitro, the phagocytic ability of aRPE-19 cell line on rod outer segments was examined [24]. It was found that there was an accumulation of oxidative stress occurred on the aRPE-19 cells phagocytizing oxidized rod outer segments under a condition of 25 mH glucose for 7 to 9 days. Molecularly, these cell under oxidative stress experienced an increase in malondialdehyde, lipofuscins, BIP, cleaved caspase-8, but a decrease in LC3_18/20. Another study extended the HG condition to 2 month on aRPE-19 cells with glucose concentration of 75 mM [6]. This prolonged treatment of HG had induced retarded proliferative capability, oxidative stress, autophagic stress, ER stress, and genotoxicity in the cells. It was also reported that the senescence-associated proteins p53 and p21 were enhanced. These studies provided good evidence for exploring the characteristics and pathological mechanism of RPE under HG conditions. Nonetheless, the present study offered further and diverse information with human primary RPE cells and an overall pattern in transcriptional factors.

To investigate the changing spectrum of genetic transcription, we simulated the internal environment in vitro by extending the human primary RPE cells culturing under a HG condition. A series of bioinformatics analyses were later applied to develop the translational atlas. The RPE cells were consistently subjected to HG treatment of 3 generations, due to the mortality of primary cell line. The decreased viability accompanied with enhanced ROS indicated damage and oxidative stress induced by protracted HG in the RPE cells (Fig. 1A–D), which were in consistent with most of the former studies about DR [3, 21]. The G2 arrest and retarded migration capability found in the HG-treated RPE cells (Fig. 1E–H) together indicated a disrupted cellular proliferation and activity. Hence, it was evident that the prolonged exposure to HG produced cellular injuries.

Numerous factors and pathways have been identified as contributors to the development of DR pathogenesis [3, 25]. High throughput sequencing enables researchers to establish a better understanding of the overall molecular biological characteristics of different diseases [26, 27]. Instead of performing molecular experiments to detect the targeted pathways respectively, whole transcriptome analysis allows us to integrally explore the transcriptional map faster [28]. In a previous study, several genes encoding visual cycle proteins were found to be reduced in the retina of diabetic long evans rats, by using whole genome microarray [29]. Another study reported alterations of complement stimulation, AGE/RAGE activation, and inflammatory response in the RPE separated from early stage diabetic cynomolgus monkeys with RNA sequencing [30]. With sequencing, a group of former researchers found a total of 189 differentially expressed circRNAs in the retinal proliferative fibrovascular membranes of DR patients [9]. The RPE harvested from streptozotocin-induced early DR mice were subjected to RNA sequencing, spotting a bundle of 298 DEGs [31]. Nevertheless, no former transcriptome was performed to investigate human RPE cells under HG. To fill this vacancy, here in this study, we identified a set of 4125 DEGs in the protracted HG-treated human RPE cells, comparing with the control group (Fig. 2A). We included 10 pathways respectively in either the enrichment, in order to encompass candidates for further statistical analysis. After intersecting the enriched genes based on the KEGG pathway, WIKI pathway, and Reactome databases, the alteration in transcription were mainly concentrated into 21 genes (Figs. 2B–D and 3). Later, WGCNA, a data analyzing algorithm that identifies co-expressed gene sets with potential synergistic regulatory patterns through modularized processing [15], was performed as an auxiliary approach to funnel the pivotal information (Figs. 4 and 5). Noise could be generated by WGCNA when the sample size is small, causing difficulty in precise target acquisition. In order to include data with larger sample size, we made efforts to mine available databases. Unfortunately, it was unable to find reliable data sets with multiple samples of diabetic RPE. Therefore, we performed WGCNA on the local database. Data intersecting with DEGs assisted us to minimize bioinformatics noise and focus on the significantly changed genes. After conducting database enrichment and multiplicative intersectional analysis based on the WGCNA results, 12 genes of importance in HG were identified (Figs. 6 and 7). To further explore the changing molecules, the GEO-GSE102485 dataset was selected for its comprehensive collection of transcriptional information from human retina tissue [16]. In this step, 13 collinear genes were found in both the HG-treated RPE cells and GSE102485 dataset after KEGG enrichment (Fig. 8). Furthermore, we performed an interdisciplinary analysis with the vitreous cytokines reported in a former study [17] to additionally explore the transcriptome dataset, finding that VEGFA was the starred factor during HG (Fig. 9A, B). In the aggregate, 32 genes were distinguished by analyses of DEGs (Step 1), WGCNA (Step 2), and overlapping with the GSE102485 dataset (Step 3), where VEGFA and ERN1 were found to be the core factors (Fig. 9C, D).

The vascular endothelial growth factor (VEGF) system comprises a complex array of proteins that regulate physiological processes and promote pathological progression [32]. As a crucial component of the system, VEGFA plays a key role in the regulation of angiogenesis and becomes a prime target for neovascular ocular diseases such as DR, while intravitreous anti-VEGF drugs have gain great success in combating diabetic macular edema (DME) [33]. The enhanced expression of VEGFA identified through the bioinformatics analyses illustrated that it is an element of priority in the pathogenesis of human RPE under HG environment (Fig. 9). With VEGFA in proving, the protracted HG-treated RPE also presented an effective DR-simulating model in vitro. The other genes interacting with the VEGF system were also screened out in the RPE cells: KDR, also referred to as vascular endothelial growth factor receptor 2 (VEGFR2), serves as a receptor for the VEGF system and is crucial in controlling angiogenesis and vascular development [32]. Giving that inflammation is a major pathological feature of DR, intercellular adhesion molecule-1 (ICAM1) and interleukin-6 (IL6) promote a synergistic effect in leukocyte adhesion to vascular endothelium with VEGFA [34, 35], which could lead to retinal damages in HG condition. As one of the hypoxia-induced genes, PDGFB modulates neovascular formation in retina with VEGF [36], playing a part in angiogenesis [37]. Cyclin D1 (CCND1), a regulator in RPE proliferation [38], was reported to be significantly increased by VEGF in granulosa cells [39]. Fibroblast growth factor 2 (FGF2) is active in neovasculation and regulates the VEGF system by binding with fibroblast growth factor receptor 1 (FGFR1), where it promotes VEGFA/KDR but participates in preventing excessive angiogenesis with vascular endothelial growth factor B (VEGFB) [40]. Association was reported between the expression of VEGF and matrix metallopeptidase 3 (MMP3) [41, 42]. ATPase copper transporting alpha (ATP7A) maintains the integrity of KDR and promotes angiogenesis [43]. Solute carrier family 3 member 2 (SLC3A2) was found to be increased by glabridin, while VEGF experienced a reduction [44]. Associations were also found between factors including DNA damage inducible transcript 3 (DDIT3)/DNA damage inducible transcript 4 (DDIT4) and VEGFA/KDR [45, 46]. Apart from these, the cross-sectional analysis, combined with a previous study that identified intraocular biomarkers [17], further suggested that the RPE is one of the major contributors of vitreous VEGFA in DR (Fig. 9A, B). The significance of VEGFA emphasized the angiogenesic properties in RPE induced by excessive level of glucose.

ERN1 was the other gene ultimately screened out in the HG-treated RPE cells. As a sensor in the unfolded protein response (UPR), ERN1 plays an important part in intra-ER lumen signaling [47]. While ER stress promotes retinal inflammation and pathological damage [48, 49], ERN1 serves as a potential mediator in the development of retinopathy. A former report of abrogating angiogenesis by inhibiting ERN1 gave possible clues for its pathogenicity during DR [50]. In the current study, several genes linking to ER stress were also discovered to be significantly changed: The HSPA5 gene encodes the binding immunoglobulin protein (BiP) that is situated in the ER lumen and functions as an assistant during the ER-associated degeneration (ERAD), a constituent of the ER’s protein transport system and a controller of Ca2+ balance in the ER [51]. The involvement of HSPA5 during DR was proven in mice and human retinas [52], while its association with VEGFA was observed in extra-ocular tissues [53, 54]. In addition, DDIT3, DDIT4, DnaJ heat shock protein family (Hsp40) member A1 (DNAJA1), Hsp40 member B9 (DNAJB9), Hsp40 member B11 (DNAJB11), heat shock protein family A (Hsp70) member 1A (HSPA1A), Hsp70 member 1 like (HSPA1L), Hsp70 member 6 (HSPA6), Hsp70 member 8 (HSPA8), hypoxia up-regulated 1 (HYOU1), stromal cell derived factor 2 like 1 (SDF2L1), solute carrier family 39 member 14 (SLC39A14), X-box binding protein 1 (XBP1), are participants in ER stress [55,56,57,58,59,60,61,62,63,64]. These results revealed the priority of ER stress in HG-induced damages to human RPE.

Besides, there are no associations reported between diabetes-jeopardized RPE and the other genes with significant variations, including F-box protein 2 (FBXO2), microtubule associated protein 1 light chain 3 beta (MAP1LC3B), member RAS oncogene family (RAB39B), solute carrier family 40 member 1 (SLC40A1), solute carrier family 7 member 11 (SLC7A11), transferrin receptor (TFRC), voltage dependent anion channel 3 (VDAC3). These genes are mostly ferroptosis-related (Fig. 8B, Figs. S1 and S3). As more ferroptotic mechanisms become discovered in DR pathogenesis, the genes might throw light on future ophthalmic studies.

Based on different databases, pathways were enriched and raised in different steps in the present study. Several of these pathways has been investigated or discussed as mechanisms associated with DR: angiogenesis [65]; the PI3K-Akt, MAPK, and HIF-1 signaling pathway [66,67,68]; alcoholism [69]; ER stress that impacts on protein processing in endoplasmic reticulum and modulates UPR [49]; inflammation involving interleukin or inflammatory signaling that includes factors from the pathways of systemic lupus erythematosus or rheumatoid arthritis [70,71,72,73]; glycine, serine and threonine metabolism [74]; ferroptosis [75]; adipogenesis [76]; extracellular matrix organization [77]; platelet derived growth factor receptor binding [78, 79]; autophagy [80]; necroptosis and apoptosis [81]; fluid shear stress and atherosclerosis [82]; non-alcoholic fatty liver disease [83]; AGE-RAGE signaling pathway [84, 85]. The other pathways might share genes interact in DR so that they were enriched. Though there is no reports from publication, but this study provided an introduction for these molecular pathways for possible experiments in the future.

The findings of the present study serve to augment our understandings of RPE in DR. However, there are limitations in this study, for in vitro models could only partially reflect in vivo conditions. The period of HG treatment was also restricted by the mortality of primary cells. For better illustration, more investigations shall be carried out on animals or collecting clinical specimens, in order to deepen the knowledge about the important role of RPE in DR.

Conclusion

The exploration of the DEGs in this study demonstrated the molecular atlas of prolonged HG influences, which potentially revealed the pathological role of human RPE during DR. There were 2 genes associated with DR starred in this study. While VEGFA and ERN1 are priority to consider in the pathogenesis of HG-impacted cells, it is also crucial to take into account the other genes that were filtered (Fig. 9D). The current study systematically screened and listed the altered genes in human primary RPE cells under protracted HG: confirming factors investigated in former studies; providing novel DR-related factors that have been ignored in the past, for future researches on molecular biology. Among all enriched pathways, neovascular activity and ER stress showed major effects on the cells suffered protracted HG, giving rise to beneficial potentials for researches. Furthermore, the findings of this study highlight the significance of RPE in DR development, providing a possible model for disease countermeasures.