Endothelial Cells Derived From Patients With Diabetic Macular Edema Recapitulate Clinical Evaluations of Anti-VEGF Responsiveness Through the Neuronal Pentraxin 2 Pathway

Vascular endothelial growth factor (VEGF) inhibitors (anti-VEGFs) are used to treat diabetic macular edema (DME). Anti-VEGFs are given as an ocular injection and are required frequently and over the long-term. The work presented herein demonstrates that in only approximately a quarter of eyes does DME fully resolve during the 1st year of treatment with anti-VEGFs; many eyes do not respond adequately to this therapy. It shows also that in vitro readouts of induced pluripotent stem cell–derived endothelial cell (iPSC-EC) responses can be used to identify a patient’s response to anti-VEGFs; such identification has the potential to improve clinical decision-making about the use of alternative therapies for those who are unlikely to respond to anti-VEGFs. Furthermore, induced pluripotent stem cell (iPSC) technology may enable targeted drug development for DME that is not predominantly driven by VEGF. DME remains a leading cause of visual loss in people with diabetes (1). It results from the breakdown of the inner and outer blood-retinal barriers (BRBs), with subsequent leakage of blood, fluid, and lipids, and infiltration of inflammatory cells into the macula, which alters retinal architecture and function (2). Although the exact mechanisms that lead to DME are not completely understood, evidence suggests that VEGF plays a main role, often in unison with an array of proinflammatory cytokines (3). The amount of VEGF is elevated in the vitreous of patients with DME (4), and experimental studies have demonstrated that VEGF induces vascular leakage (by disrupting the integrity of endothelial cell [EC] junctional complexes) and a loss of barrier function (5).

Since 2010, numerous randomized controlled trials (RCTs) have demonstrated the benefit of VEGF-neutralizing antibodies to improve vision and reduce macular fluid in people with DME (6–10). Despite the proven efficacy of anti-VEGFs, however, it is now evident that not all patients with DME respond to them in a similar manner. To date, no consensus exists about a definition of “responder” and “nonresponder” with regard to anti-VEGFs (11–14), nor is there a robust predictor for who will and will not respond to this therapy. Anti-VEGFs are given as intravitreal injections, and in most patients they need to be administered monthly throughout the 1st year of treatment and at frequent intervals thereafter. Most patients with DME will require long-term treatment in order to maintain the benefits of the anti-VEGF: ∼50% of patients still require treatment 5 years after its initiation (15). An ability to predict a patient’s responsiveness to this therapy would be of great benefit, avoiding unnecessary risks and the inconvenience and distress caused by repeated injections, and allowing more timely intervention with alternative therapies (16). Such a personalized approach to anti-VEGF administration would avoid the loss of vision that may occur as a result of selecting or continuing an ineffective treatment, an unnecessary workload to health professionals, and costs for health services.

In this study, we evaluated the response to anti-VEGFs among a large cohort of patients with DME treated with anti-VEGFs. In only ∼28% of patients did the edema clear fully at any time during the 1st year of anti-VEGF treatment. Also, having previously developed a unique approach to generate iPSCs and differentiated endothelium from small volumes of blood (17), we designed a conjoined clinical and experimental study using iPSC microvascular endothelium derived from patients with diabetic retinopathy (DR) but without DME and from patients with DME who are classified as “full responders” or “nonresponders” to anti-VEGFs. Their iPSC-ECs were assessed for VEGF-induced permeability responses, phosphorylation pathways, and gene expression profiles. The clear differences between full responders and nonresponders may reflect a distinct pathophysiology in ECs from the patient groups and could not only offer useful clinical information but also provide scope for the development of new drugs.

Electronic records were retrospectively reviewed for all consecutive patients with newly diagnosed DME who had been examined at the Belfast Health and Social Care Trust (BHSCT). This study was part of an approved audit (audit approval number 5170) for which full ethical review was waived. A patient cohort was constructed. Eligibility criteria for this cohort included 1) receipt of the first dose of anti-VEGF between 1 March and 31 December 2014; 2) treatment exclusively with ranibizumab, as this was the first anti-VEGF agent approved for the treatment of DME in the U.K., and most patients at the BHSCT would have received this drug during the study period; 3) receipt of a minimum loading dose (injections over three consecutive months); and 4) a minimum follow-up of 12 months, with regular visits during that time (every 4–6 weeks).

Spectral-domain optical coherence tomography (SD-OCT) scans obtained throughout the whole 1st year of treatment were reviewed and graded by an expert clinician (N.L.) who was masked to clinical findings. The expert classified patients as “full responders” when the macula fully dried at any point during the 1st year or treatment (although the presence of small, isolated, or sparse cysts was allowed within this definition); as “partial responders” when the amount of fluid present at the macula was reduced but the fluid never fully dried during the 1st year of treatment; or as “nonresponders” when macular fluid remained unchanged upon qualitative evaluation of SD-OCT scans throughout the 1st year of treatment and when the central retinal thickness (CRT) changed ≤15% from the baseline thickness.

Differences were investigated between the aforementioned groups in demographics (age, sex), type of diabetes (type 1 vs. type 2), best-corrected visual acuity, number of injections received, and central retinal subfield thickness measured from SD-OCT scans. The χ² test was used in order to compare proportions (sex and type of diabetes) among anatomic response categories. Linear regression was fit by using generalized linear models, and SEs were adjusted for within-patient correlation by using a “sandwich” (Huber-White) variance estimator.

This prospective study was approved by the Office for Research Ethics Committees Northern Ireland under the title of “Stratifying Treatment for Diabetic Macular Oedema Using Induced Pluripotent Stem (iPS) Cells” (STREAMLINE) (reference 14/NI/1109; Integrated Research Application System project identifier 162254). Patients with DR were recruited from ophthalmic clinics at the BHSCT. Volunteers without diabetes were recruited as controls from among individuals accompanying patients to the BHSCT clinic and from Queen’s University Belfast.

Patients with diabetes and mild nonproliferative DR with no DME (n = 6), those with DME who were classified either as “full responders” (n = 6) or as “nonresponders” (n = 6) to anti-VEGF therapy (ranibizumab), and age-matched volunteers without diabetes (n = 6) were recruited into the study.

Information about the study was provided verbally and in writing to all potentially eligible participants. Informed consent was obtained from those who were willing and who consented to take part, before beginning the study procedures. Patients unable to provide informed consent for the study were excluded. The study adhered to the principles of ethical medical research as detailed in the Declaration of Helsinki.

On the day blood samples were collected, a detailed medical and ocular history was obtained and blood pressure was measured. A 20-mL sample of peripheral blood was obtained from all participants; the samples were collected by venepuncture in VACUETTE 4-mL K3 EDTA-coated tubes (item 454021; Greiner Bio-One International GmbH). Samples were processed in the laboratory by scientists masked to clinical findings and the origin of the samples (i.e., the scientists did not know from which donor group the samples came).

Mononuclear cells (MNCs) were isolated as previously described (17). Briefly, blood was separated by gradient by layering it on Histopaque solution (item 10771; Sigma-Aldrich) at a 1:1 ratio and then centrifuging it at 550g with the break off and at room temperature for 30 min. The MNCs’ buffy coat was collected with a soft plastic pipette and placed in a 15-mL tube, which was topped off with PBS and centrifuged at 300g at 4°C for 10 min. After three washes with PBS, the cells were resuspended in 1 mL of MNC medium, as previously described (17), and plated in MNC medium at a density of 4 million cells per milliliter. The cells were expanded for 7 days and then used for reprogramming.

iPSCs were generated from patients’ blood MNCs, as previously described (17). Up to 2 million MNCs were transfected with 5 μg of plasmid per million cells (a 5:1 ratio of pEB-C5 and pEB-Tg) by using a Human CD34+ Cell Nucleofector Kit (catalog no. VPA-1003; Lonza) and an Amaxa nucleofector (program T-016), according to the manufacturer’s protocol. pEB-C5 and pEB-Tg plasmids were a gift from Linzhao Cheng (plasmids 28213 and 28220; Addgene). On day 2 after transfection, the cells were seeded onto inactivated feeders (mouse embryonic fibroblasts from ATCC) in reprogramming medium (KnockOut DMEM/F-12 with MEM Non-Essential Amino Acids Solution and GlutaMAX [Thermo Fisher Scientific] supplemented with KnockOut Serum Replacement [20%] [Thermo Fisher Scientific], 0.1 mmol/L 2-mercaptoethanol [Thermo Fisher Scientific], and 10 ng/mL MACS human recombinant fibroblast growth factor 2 [Miltenyi Biotec]).

Reprogramming medium was changed every day and colonies began to appear from day 9. Colonies were individually expanded for the establishment of iPSC lines. For the maintenance of iPSC lines, reprogramming medium was changed every day and the iPSC colonies were subcultured once per week at a 1:6 ratio.

iPSCs were cultured under feeder-free conditions and seeded on Matrigel Growth Factor Reduced Basement Membrane Matrix (catalog no. 354230; Corning Life Sciences) in StemMACS iPS-Brew XF (130-104-368; Miltenyi Biotec). The next day, the medium was replaced with Neurobasal Medium (catalog no. 21103049; Thermo Fisher Scientific) with N-2 Supplement (catalog no. 17502001; Thermo Fisher Scientific) and B-27 Supplement (catalog no. 17504044; Thermo Fisher Scientific) supplemented with 25 ng/mL BMP4 Recombinant Human Protein (catalog no. PHC9531; Thermo Fisher Scientific) and 8 μmol/L CHIR99021 (SML1046; Sigma-Aldrich). The medium was replaced 72 h later with StemPro-34 SFM (catalog no. 10639011; Thermo Fisher Scientific) supplemented with 200 ng/mL VEGF Recombinant Human Protein (catalog no. PHC9391; Thermo Fisher Scientific) and 2 μmol/L forskolin (F6886; Sigma-Aldrich). This step was repeated after 24 h.

On day 6 of differentiation, magnetic activated cell sorting (MACS) of cells expressing the mature EC marker CD144 was performed by using anti–human CD144 (VE-Cadherin) MicroBeads (130-097-857; Miltenyi Biotec). The cells were seeded on plates coated with mouse collagen IV in EGM-2 10% FBS medium supplemented with 50 ng/mL VEGF and 10 μmol/L LY364947. Medium was replaced every other day and the cells were used for further analysis or were cryopreserved once they were confluent.

Relative mRNA expression of specific cell markers (and other genes of interest) was determined by using RNA extraction, RT-PCR, and quantitative RT-PCR, as previously described (18). Primers for CD235, CD71, OCT4, LIN28, NANOG, CD144, CD31, VEGFR2, and NPTX2 are listed in Supplementary Table 2. GAPDH, ATP5B, ACTB, and UBC were used as reference genes (all from Primer Design Human geNorm Kit [catalog code ge-SY-12; Integrated Sciences]). Sequences were not provided by the manufacturer.

Proteins of specific cell markers were detected by using Western blotting, as previously described (18). To ensure valid comparison between samples, cells were harvested at the same time point, confluence (70–80% for iPSCs, 80–100% for iPSC-ECs), or both. For the detection of phosphorylated proteins, Tris-buffered saline with Tween was used instead of PBS with Tween, and 5% BSA Tris-buffered saline with Tween was used as a blocking solution instead of 5% milk in PBS with Tween. Antibodies against TRA-1-60 (ab16288), OCT4 (ab19857), LIN28 (ab46020), NPTX2 (ab69858) (all from Abcam), CD144 (STJ96234; St John’s Laboratories), and Phospho-VEGFR2 (PA5-12598) and Phospho-eNOS (PA5-35879) (both from Thermo Fisher Scientific) were used. Antibodies against actin β (MAB8929; R&D Systems) or GAPDH (ab125247; Abcam) were used as the loading control. Western blotting was quantified by using ImageJ software, and samples were normalized to the loading control (actin β or GAPDH).

Immunofluorescence staining procedures have been detailed by Vilà-González et al. (17). Live staining of iPSCs was performed by using CDy1 Dye (14001; Active Motif), as indicated in the manufacturer’s instructions.

Migration capacity was studied through a wound healing or scratch assay. iPSC-ECs were seeded in wells of a 24-well plate at a density of 50,000 iPSC-ECs per well. Once the cells were confluent, the monolayer was scratched from one side of the well to the other by using a 200-μL pipette tip. Pictures of the wound were taken at the 0-, 4-, 8-, 12-, and 24-h time points. The migration capacity of the cells was calculated by measuring the change from baseline, in which the area of the wound (0 h after the scratch) minus the area of the wound at the various time points (4, 8, 12, and 24 h) equals the area the cells had covered during that time.

Proliferation of iPSC-ECs was assessed by using a CyQUANT NF Cell Proliferation Assay (C35006; Thermo Fisher Scientific), according to the manufacturer’s instructions. iPSC-ECs were seeded in a 96-well plate at a density of 5,000 cells per well and were allowed to adhere for at least 4 h. Cells were seeded in quadruplicate for each cell line and time point (8, 24, and 48 h). The values at the various time points were normalized to the values obtained at 8 h for each experiment.

Permeability response to VEGF in iPSC-ECs from different patients/donors was measured using xCELLigence RTCA DP (Real-Time Cell Analyzer–Dual Purpose; ACEA Biosciences, Inc.), which measures cell impedance in real time. A total of 20,000 cells in 200 μL medium (the medium used to maintain iPSC-ECs EGM-2 10% FBS media supplemented with 50 ng/mL VEGF and 10 mmol/L LY364947) was seeded in each well of a 16-well polyethylene terephthalate plate that had been precoated with mouse collagen IV. Four replicates were seeded for each condition. The cells reached confluence overnight. After seeding (24 h), the cells were starved for 2 h by replacing 150 μL of medium with endothelial basal medium plus HEPES at 25 mmol/L (15630106; Thermo Fisher Scientific). Next, 100 μL of the medium was replaced with starvation medium with 200, 100, or 50 ng/mL VEGF (for a final concentration of 100, 50, or 25 ng/mL). BSA (0.1%) in PBS was used as the vehicle control. Data were analyzed by using RTCA Data Analysis Software and Prism 5 software.

Quality control and RNA sequencing were analyzed at the Queen’s University Genomics Core Technology Unit. In brief, the cells were briefly washed with PBS and harvested by using QIAzol lysis buffer. Samples and the library were prepared as previously described (17). Libraries were sequenced on a NextSeq sequencing system (Illumina), and paired-end reads were mapped to the human reference genome (hg38)—allowing up to 2 mismatches and up to 10 hits per read—by using the CLC Genomics Workbench version 10.0.1 (https://www.qiagenbioinformatics.com). Reference sequences were annotated with genes and transcripts. Reference content was mapped to gene regions only, and the raw read counts were generated by HTSeq-Counts. To identify differentially expressed genes (DEGs) between RNA sequences of samples from responders and nonresponders, a differential expression analysis was performed by using DESeq2 in the R software package, followed by standard visualization by using EnhancedVolcano and Pheatmap in R. Comparative gene expression data were filtered according to the absolute log₂ fold change greater than +1 and an adjusted P value (q value) threshold of <0.05.

Lentiviral particles were produced by using MISSION shNT (sc-62166; Santa Cruz) or shNPTX2 plasmids (MISSION shNPTX2 (SHCLNG-NM_002523, TRCN0000373047; Sigma-Aldrich) according to a previously described protocol (18–20). The efficiency of the infection was 70–80%. Two hours before harvest, the cells were depleted of serum; they were treated with 50 ng/mL VEGF 5 and 20 min before harvest. Vehicle control cells (0.1% BSA in PBS), harvested at exactly 72 h, served as the control (0 min time point).

iPSC-ECs from nonresponders, which had been seeded in transwell plates, were treated with 1 mg/mL FITC-dextran 4 (46944; Sigma-Aldrich) between 3 and 24 h after transduction with shNPTX2 or shNT. Fluorescence was determined at 520-nm emission and 485-nm excitation, and relative percentage diffusion was calculated.

Data presented in the graphs were analyzed by using GraphPad Prism 5 software with a two-tailed Student t test for two groups or with pairwise comparisons or ANOVA; values are expressed as the mean ± SD. A P value <0.05, <0.01, or <0.001 was considered significant, as indicated.

The data sets and the patient-specific IPSC lines generated and analyzed during this study are available from the corresponding authors upon reasonable request.

A total of 140 eyes from 100 patients with DME were treated with ranibizumab; these eyes/patients formed the study cohort. Color fundus photographs of the right eye were obtained from an individual with no retinal disease (Fig. 1A, left) and from a patient with DME (Fig. 1A, right); hard exudation can be seen involving the macula and extending to the fovea (Fig. 1A, right, black arrow). A total of 172 eyes were treated during the study period; 32 were excluded because they did not meet the eligibility criteria for the study (<12 months of follow-up or use of other intravitreal therapy [i.e., aflibercept or steroids]).

The mean best-corrected visual acuity was 59.7 Early Treatment Diabetic Retinopathy Study letters (SD, 15.4 letters) at baseline and improved to 67.3 Early Treatment Diabetic Retinopathy Study letters (SD, 14.4 letters) at the 12-month follow-up, after a mean of 7.5 ranibizumab injections (SD, 2.04 injections). The anti-VEGF response observed is summarized in Supplementary Table 3, and the demographic and clinical characteristics of patients (eyes) included in the study are summarized in Supplementary Table 4. Examples of retinal disease in a full responder and in a nonresponder are shown in Fig. 1B. In full responders the macula dried after a median of 7 months (range, 1–12 months).

No differences were found among groups for any of the clinical parameters evaluated other than, as expected, central retinal subfield thickness and the number of intravitreal anti-VEGF injections received. Partial responders and nonresponders received more injections than did full responders. In 39 patients in the cohort, both eyes were eligible and thus were included in the study. Among these patients, concordance of the response between eyes (i.e., right eyes were in the same category of anti-VEGF response as left eyes) was observed in 28 (72%).

Demographics and characteristics of participants are detailed in Supplementary Table 1. iPSCs generated from donors with and from those without diabetes showed identical reprogramming and differentiating capacities. These iPSC lines were initially generated from three subjects from each cohort (patients with DR and no DME; patients with DR and DME that responded fully to anti-VEGF treatment [“full responders,” hereafter referred to as “responders”]; those who did not respond to this therapy [“nonresponders”]; and age-matched healthy volunteers). For each group, one iPSC line was selected, expanded, and characterized (Fig. 2) before being differentiated into ECs (iPSC-ECs) (Figs. 3 and 4). Established iPSC colonies were examined for pluripotency by using quantitative RT-PCR (Fig. 2A and B) and immunoblotting (Fig. 2C), which confirmed the high expression of pluripotency markers and the low expression of MNC markers at both the RNA and the protein level. iPSC colonies also stained positive for CDy1, Oct4, Lin28, and TRA-1-60 (Fig. 2D). Overall, iPSCs from the different groups were comparable in terms of expression of pluripotency markers and morphology, indicating the reprogrammable nature of each cohort’s samples.

Central to this study was the comparison of patients with DME; therefore, in order to avoid misidentification and to rule out possible cross-contamination of cell lines, short tandem repeat profiling of the iPSC lines was undertaken (Supplementary Table 5). When two cell lines shared less than 56% of the analyzed alleles, they could be considered to come from different individuals; the selected iPSC lines met these criteria (Supplementary Tables 5 and 6), confirming that they came from different donors. Moreover, none of the iPSCs matched any of the lines recorded by the European Collection of Authenticated Cell Cultures. This fact provided important assurance that the iPSCs used in this study were unique and patient-specific.

Upon iPSC differentiation to iPSC-ECs, flow cytometry demonstrated that before CD144 MACS selection, up to 57.2% of iPSC-ECs were CD144⁺ (Supplementary Fig. 1); up to 67.2% of cells were CD144⁺ in the samples from patients with DME (Supplementary Fig. 2). iPSC-ECs from patients with diabetes and volunteers without diabetes displayed a loss of expression of pluripotent markers such as OCT4, LIN28, and NANOG and upregulation of EC markers such as CD144, CD31, and VEGFR2 at the mRNA level (Figs. 3A and B and 4A and B), confirming differentiation toward the vascular lineage. Further characterization by immunoblotting (Figs. 3C and 4C) also confirmed that cells switched off expression of the pluripotency marker TRA-1-60 but highly expressed EC-specific markers such as CD144. There was no difference in EC gene expression and immunophenotype among iPSC-EC groups (Figs. 3 and 4), nor was there a difference, based on immunofluorescent staining, on adherens and tight junction integrity or VEGFR2 expression (Figs. 3D and 4D). These results clearly demonstrate that iPSC-ECs were successfully generated and expanded from all groups (donors without diabetes and patients with diabetes either with or without DME).

iPSC-ECs from donors without diabetes, patients with mild DR, and responders and nonresponders with DME displayed similar functional capacities. The migration capacity of iPSC-ECs from those without and those with diabetes, or from responders and nonresponders, was not significantly different when assessed by using a wound healing assay or scratch assay (Fig. 5A, B, D, and E). In a similar manner, no significant difference in their proliferation capacity was found (Fig. 5C and F). One of the main elements of DME is the excessive permeability of the normally “tight” retinal microvasculature. Therefore, EC permeability was tested in this study after stimulation with VEGF. Significant differences in VEGF-induced cell permeability were identified when cells from patients with diabetes were compared with cells from volunteers without diabetes (Fig. 6A and B) and when cells from responders with DME were compared with those from nonresponders with DME (Fig. 6C and D). In particular, iPSC-ECs from nonresponders had significantly higher cell impedance (represented as a cell index) shortly after VEGF was added (Fig. 6C and D). At 30 and even 90 min after exposure to VEGF, iPSC-ECs derived from the groups with DME showed decreased cell impedance (i.e., increased permeability), and none showed a recovery pattern like that observed in healthy volunteers (Fig. 6A and B). iPSC-ECs from nonresponders showed significantly lower impedance than was seen in responders (Fig. 6C and D). We validated these data by performing further permeability experiments using three more iPSC-ECs lines (a total of six for each group), which confirmed increased FITC-dextran leakage in nonresponders (Supplementary Fig. 4). The relative percentage increase in FITC-dextran diffusion in nonresponders validated the cell impedance data mentioned above. Importantly, we compared the barrier integrity of our iPSC-ECs with that of primary human retinal microvascular ECs (HRMECs) using the xCELLigence system. The results showed that the iPSC-ECs were able to reach a level of impedance (cell index; proportional to barrier integrity) that was comparable to that of the HRMECs (data not shown). In addition, the obtained cell index values were comparable to or higher than those obtained by other groups working with human brain microvascular ECs (21). We can therefore confirm that our cells’ barrier integrity matches that of central nervous system ECs.

Given that significant differences were identified in permeability responses of iPSC-ECs from responders and nonresponders with DME, we aimed to shed light on the mechanism by which this might happen. VEGF induction of EC permeability starts with the binding of VEGF to VEGFR2, which then is phosphorylated, triggering a complex signaling cascade that culminates in the opening of adherens and tight junctions between ECs. VEGFR2 phosphorylation was assessed by stimulating iPSC-ECs from responders and nonresponders with VEGF for 5 and 20 min. Untreated iPSC-ECs from both groups were used as controls (0 min of VEGF stimulation). Interestingly, Western blotting has shown phosphorylation of VEGFR2 that was higher at the 5-min time point in samples from nonresponders than in responders (Fig. 7A and B). As a readout for downstream VEGFR2 signaling, phosphorylation of endothelial nitric oxide synthase (eNOS), which has a role in the regulation of endothelial permeability, was evaluated; no significant difference was found between groups (Fig. 7A and B). These results indicate that ECs generated from nonresponders have an increased phosphorylation pattern from VEGFR2.

In order to obtain a more thorough analysis of the intrinsic differences between iPSC-ECs from responders and nonresponders, RNA sequencing was performed. This sequencing disclosed 13 DEGs (P <0.05) between responders and nonresponders (Fig. 8A and B). Further analysis revealed that these genes were involved in processes such as cell regeneration, repair, and survival (those showing higher expression in responders); and inflammation or susceptibility to diabetes (those showing higher expression in nonresponders). These 13 DEGs, of which 6 were upregulated in iPSC-ECs from responders (Fig. 8C) and 7 were upregulated in iPSC-ECs from nonresponders (Fig. 8D), could cluster biological replicates of similar samples together. Among these DEGs, neuronal pentraxin 2 (NPTX2) was particularly notable because it has been previously identified as a biomarker of edema in gliomas (22) and is also expressed in the retina (23). Strikingly, NPTX2 is also related to C-reactive protein, one of the inflammatory biomarkers involved in the pathogenesis of DR (24), which could mean that it contributes to inflammation in the microvasculature in nonresponders. The upregulation of NPTX2 was further assessed in iPSC-ECs from the different groups/cohorts by using quantitative RT-PCR (Fig. 8E). iPSC-ECs from nonresponders showed higher expression of NPTX2 than did those from responders. Because phosphorylated VEGFR2 was also higher in nonresponders after 5 min of exposure to VEGF (Fig. 7A), we hypothesized that a direct correlation may exist with NPTX2. As such, knockdown of NPTX2 for 72 h by lentiviral gene transfer in nonreponder iPSC-ECs showed significantly less phosphorylation of VEGFR2 at the 5-min time point than did the corresponding shNT control (Fig. 8F–H). In addition, given that NPTX2 is higher in nonresponders, VEGF levels might directly or indirectly regulate the level of NPTX2, as indicated in Fig. 8F. Further NPTX2 silencing experiments based on three additional patient-specific iPSC-EC lines from nonresponders showed a pattern of relative percentage decrease in diffusion of FITC-dextran in shNPTX2 (Supplementary Fig. 5). This could suggest that NPTX2 silencing might serve as a target for improving leakage in nonresponders.

This study shows that only a proportion of patients with DME achieve full resolution of their macular fluid during the 1st year of treatment with anti-VEGF therapy. After first characterizing the phenotype of patients with DME and their responsiveness to anti-VEGF therapy, we demonstrated functional differences in their iPSC-ECs in vitro, which is to our knowledge a novel discovery. Importantly, the ability to separate patients with DME into categories of “full responders” and “nonresponders” to anti-VEGFs in a laboratory setting creates a unique platform for deeper patient stratification and for the understanding of pathogenic mechanisms of disease and response to anti-VEGF treatment.

Previous studies used CRT as seen on SD-OCT scans to determine the response to anti-VEGF therapy. No consensus exists, however, on definitions of “responder” and “nonresponder.” Thus, Lee et al. (13) considered a response to have occurred when a CRT was reduced by >50 µm after three consecutive anti-VEGF injections, whereas Dabir et al. (14) determined a response when CRT was reduced by >10% after two consecutive injections. Pieramici et al. (11), using data from RISE and RIDE (A Study of Ranibizumab Injection in Subjects With Clinically Significant Macular Edema [ME] With Center Involvement Secondary to Diabetes Mellitus), proposed the definition of “immediate” and “delayed” responders based on whether a reduction in CRT of >10% or ≤10%, respectively, had occurred following three loading doses of anti-VEGF. Based on data from the DRCR.net protocol I, Dugel et al. (12) used the terms “strong early” and “limited early” responders for a reduction of ≥20% or <20%, respectively, of CRT on SD-OCT after the same loading doses. CRT as measured on SD-OCT scans, though helpful, may not be the best way to determine the response to anti-VEGF therapy given the great range of normal values among individuals, variability due to the instruments used in order to obtain CRT values, and the fact that DME may still be present in the context of a “normal” CRT value in patients in whom neurodegeneration may have occurred. We considered “full responders” to be those individuals who achieved a fully dry macula at any time during the 1st year of treatment. In our cohort, the macula of full responders dried after a median of 7 months (range, 1–12 months). Waiting longer than a year to determine a response to treatment and to consider switching to an alternative therapy may compromise functional recovery. In only a small percentage of patients in our study did the macula dry fully after a year of adequate treatment with an anti-VEGFs.

Although cost-effectiveness has been demonstrated for anti-VEGF therapies licensed for the treatment of DME and currently in use in clinical practice—ranibizumab [25] and aflibercept [26]—the data used to evaluate their clinical effectiveness and cost-effectiveness was obtained from RCTs. It is now acknowledged, however, that RCT results often are not reproducible in clinical practice (27). Thus, whether the cost-effectiveness of anti-VEGF therapies remains above appropriate thresholds in real-life clinical settings remains to be elucidated.

In this study we used iPSC technology to make an in vitro model of the retinal microvasculature of patients with DME. In most previous diabetes-related studies, the main source for iPSC generation was dermal fibroblasts (28). These cells are easily maintained and reprogrammed but must be obtained through a skin biopsy (29), a procedure that carries ethical challenges, especially for patients with diabetes, in whom healing can be delayed (30). In contrast, we demonstrated that iPSCs can be obtained successfully from blood MNCs by using nonintegrating episomal-based transfection and by avoiding the long-term incorporation of reprogramming factors into chromosomal DNA.

Patient-specific iPSCs were differentiated into vascular ECs, as they are critical components of the retinal neurovascular unit (NVU) and their dysfunction is intimately linked to a loss of integrity of the inner BRB. Primary ECs have been widely used to study endothelial permeability and molecular pathways in vitro (31,32), although they are often obtained from animals or from human donors after death, and the human disease–related information they can provide is limited. In contrast, iPSC-ECs generated in this study provided a human disease model that, importantly, is specifically linked to individual patients who have been clinically phenotyped. In addition, because iPSCs can retain information from the donor cells in terms of epigenetic memory (33), the subsequent differentiated cells may be predisposed to exhibit a donor-related epigenetic signature. One study recently reported that vascular progenitors generated from tankyrase inhibitor–regulated naive diabetic human iPSCs revascularized the ischemic retina, suggesting erasure of dysfunctional epigenetic donor-cell memory (34). This phenomenon also has been described in other cells such as cardiomyocytes, wherein treatment with large amounts of glucose induces “metabolic memory” (35).

iPSC-ECs, as generated herein, are not retina-specific. However, one of the key characteristics of retinal microvascular ECs is their low permeability, which is due to their tight barrier properties. Even if the iPSC-ECs used in this study are not retina-specific, their barrier properties, checked as cell impedance, are equivalent to those of HRMECs and human brain microvascular ECs (21). Furthermore, although ECs are key to the pathogenesis of DME, they are not the only cells responsible for the maintenance of the inner BRB; integrity of the NVU is also influenced by pericytes, Müller glia, neurons, and immune cells. In contrast, the outer BRB is formed by the retinal pigment epithelium, which is also dysfunctional in diabetes (36). Considering that iPSCs can differentiate into any type of cell in the body, studies are currently underway in our laboratory to generate retinal-specific cells such as retinal pigment epithelial cells and Müller glia in order to evaluate their role in DME in a patient-specific manner (37). Cocultures of iPSC-ECs with other iPSC-derived cells that form the NVU (38) would also provide a model that more closely resembles the living retina (39).

In this study, iPSC-ECs from donors without diabetes, patients with diabetes and mild DR, and responders and nonresponders with DME displayed similar functional capacities. The migration capacity of iPSC-ECs was not significantly different between these groups. In a similar manner, no significant difference was found in their proliferation capacity. In particular, when iPSC-ECs from responders and nonresponders were compared, no significant differences were found in their morphology, expression of EC markers, and most functional capacities, according to the tests performed. This confirms that iPSCs from patients with diabetes can be efficiently differentiated into functional ECs.

Because excessive EC permeability exists in DME, we assessed the barrier function capacity of iPSC-ECs and their permeability response to VEGF in vitro. When analyzing iPSC-ECs from responders and nonresponders, we found that these cells did not recover their initial EC barrier function even 90 min after stimulation by VEGF. Interestingly, our results showed that iPSC-ECs from nonresponders had a higher permeability response to VEGF than did cells from responders, suggesting that ECs from nonresponders are more sensitive to the effects of VEGF. This suggests that ECs from nonresponders may be activated even with very low levels of intraocular VEGF, which may remain despite the blockade achieved by anti-VEGFs. Because the VEGF-mediated increase of EC permeability is driven by a complex phosphorylation cascade (40), we assessed the phosphorylation status of VEGFR2, eNOS, and 46 other kinases (Fig. 7 and Supplementary Fig. 3) after VEGF stimulation in iPSC-ECs from responders and nonresponders. We focused on several VEGF-related pathways. Results showed that VEGFR2 was differentially phosphorylated between these groups, with nonresponders demonstrating greater phosphorylation of this receptor than responders. These results are in agreement with those from the permeability studies, reinforcing the hypothesis that nonresponders may be highly sensitive to the effects induced by VEGF. If the nonresponder receptors were more sensitive to VEGF and therefore had a lower threshold for VEGFR2 activation, even the low levels of VEGF that may persist after anti-VEGF blockade in vivo would be able to maintain vascular leakage through VEGFR2 signaling, with DME persisting despite treatment.

In order to deepen our understanding of the differences between these two groups, RNA sequencing of their iPSC-ECs was performed. A differential gene expression profile was observed between responders and nonresponders. A total of 13 DEGs were identified as significantly expressed when the adjusted P value was <0.05 and the absolute log₂ fold change was >1. One of the genes found to be upregulated in nonresponders was NPTX2, which has been identified as a potential biomarker of edema in gliomas (22) and seems to be related to C-reactive protein, one of the inflammatory biomarkers involved in the pathogenesis of DR (41). Thus higher expression of this protein in the endothelium of nonresponders could be driving edema through a VEGF parallel pathway, which could be an additional reason why these patients fail to respond to VEGF inhibition. Silencing NPTX2 in nonresponders diminished the effect of VEGF in the phosphorylation of VEGFR2, suggesting a possible effect in reducing inflammation-related events. Furthermore, NPTX2 silencing reduced the relative FITC-dextran leakage in nonresponders, showcasing that targeting this gene could play a role in correcting barrier integrity. Studies have shown that abnormal upregulation of NPTX2 expression was linked to the proliferation and metastasis of cancers such as clear-cell renal cell carcinoma through the activation of the Wnt/β-catenin signaling pathway. NPTX2 was shown to interact with frizzled class receptor 6 to induce β-catenin nuclear translocation, with an ensuing decrease in the expression of E-cadherin (42) and an increase in the expression of MYC, snail, N-cadherin, and cyclin D1, which induces VEGF production (43).

The strengths of this study include the identification of consecutive patients through an electronic database; the masking of those who assessed outcomes (i.e., the researcher who graded the OCT scans and, very importantly, the scientists undertaking the iPSC-EC experiments and analyzing their results, who were masked to the treatment response observed in the clinics); the evaluation of all OCT scans obtained throughout the entire 1st year of treatment in order to determine the treatment response in all patients included in the cohort; and the robust techniques used in order to obtain patient-specific iPSC and iPSC-ECs. One of the limitations is the small sample size (n = 6) used for the in vitro studies. However, because these patient-specific iPSC-ECs were generated from a very well-characterized clinical cohort, they revealed robust differences and allowed conclusions to be made. This study sets a foundation for identifying key pathophysiological differences between full responders and nonresponders to anti-VEGF treatment among patients with DME, which may have potential clinical relevance, and provides a scope for new drug development.

By generating iPSC-ECs from volunteers without diabetes, people with diabetes but no DME, and people with DME who were categorized as full responders or nonresponders to anti-VEGFs, we found increased cell permeability in those with diabetes after stimulation with VEGF in vitro—an observation that was exaggerated in nonresponders. Additionally, phosphorylation of key cellular proteins after exposure to VEGF, including VEGFR2, and gene expression profiles (e.g., of NPTX2) differed between full responders and nonresponders, allowing for the possible identification in the future of patients who will respond to treatment.

Acknowledgments. The authors thank Eimear Byrne (Wellcome-Wolfson Institute for Experimental Medicine, Queen’s University Belfast, UK) for her expert advice on the FITC-dextran experiments.

Funding. This work was supported by grants from the Medical Research Council, the Biotechnology and Biological Sciences Research Council, and the British Heart Foundation. The authors also acknowledge the support of the Belfast Association for the Blind.

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

Author Contributions. M.V.G., M.E., and S.K. conceived and designed the study; collected, assembled, analyzed, and interpreted data; and wrote the manuscript. H.N.-M. analyzed and interpreted RNA sequencing data. S.F., S.S., and G.V. collected and assembled data. D.J.G. provided study material and approved the final manuscript. A.W.S., N.L., and A.M. conceived and designed the study; collected, assembled, analyzed, and interpreted data; wrote the manuscript; obtained financial support; and approved the final manuscript. N.L. and A.M. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.