Retinal vasculature-derived proteins serve as potential systemic biomarkers for diabetic retinopathy

Abstract

Purpose

Fibrovascular membrane (FVM), a pathological tissue causing retinal traction, represents an advanced stage of proliferative diabetic retinopathy (PDR). This study aimed to identify novel protein biomarkers associated with retinal microvascular damage leading to FVM formation and to evaluate the diagnostic potential of stratifin (SFN) and hornerin (HRNR) in patients with type 2 diabetes mellitus (T2DM) and diabetic retinopathy (DR).

Methods

Human retinal vasculature from non-diabetic donors and FVM tissues from PDR patients undergoing PPV were analyzed by one-dimensional liquid chromatography-mass spectrometry to identify differentially expressed proteins. Functional enrichment and protein interaction analyses were performed using STRING databases. Selected proteins’ messenger RNA expression was quantified by quantitative PCR (qPCR) in peripheral blood mononuclear cells from healthy, T2DM, and DR with nephropathy (DR with DN) patients. Serum levels of SFN and HRNR were measured by ELISA in T2DM, DR, and DR with DN cohorts. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves.

Results

A total of 44 FVM specific proteins were identified, notably SFN and HRNR linked to retinal angiogenesis. Serum HRNR levels were significantly decreased in DR with DN compared with T2DM, whereas SFN levels were significantly increased in DR versus T2DM. ROC analyses demonstrated acceptable accuracy for differentiating DR stages based on serum SFN and HRNR levels.

Conclusions

This is the first study to identify SFN and HRNR as systemic candidate biomarkers reflecting microvascular alterations in DR. Larger cohort studies are warranted to validate their clinical relevance for early detection and therapeutic targeting of DR.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Diabetic retinopathy (DR) results from microvascular damage in the retina due to diabetes, causing impaired blood-retinal barrier (BRB) integrity and leading to fibrovascular membrane formation in advanced proliferative DR (PDR).

• The interaction among retinal microvascular cells, pericytes, and endothelial cells is crucial for vessel stability and BRB maintenance.

• 14-3-3 protein isoforms and hornerin (HRNR) have recognized roles in cell-cycle regulation, apoptosis, angiogenesis, and stress responses, but their direct involvement in the pathogenesis of DR remains unclear.

WHAT THIS STUDY ADDS

• Identifies distinct proteomic level differences between non-diabetic retinal vasculature and PDR-associated fibrovascular membranes, which highlight the involvement of cells in angiogenic vasculature.

• Demonstrates that stratifin (SFN) and HRNR are significantly altered at mRNA and serum levels in DR with nephropathy, suggesting their contribution to retinal microvascular dysfunction.

• Establishes SFN and HRNR as potential systemic biomarkers with diagnostic value for distinguishing disease stages within the DR.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY

• Supports the development of serum-based biomarkers (SFN, HRNR) for early detection and staging of DR and DR with nephropathy.

• Opens avenues for targeting microvascular signaling pathways, particularly Phosphoinositide 3-kinase (PI3K)/Protein kinase B (PI3K/Akt) and platelet-derived growth factor receptor-β, in therapeutic strategies.

• Offers a proteomic reference that may guide future studies on retinal microvascular dysfunction in diabetes.

• Provides evidence that could inform clinical risk-stratification and screening policies for diabetic complications.

Introduction

Diabetic retinopathy (DR) is an ocular manifestation of type II diabetes mellitus (T2DM), causing microvascular damage in the retina leading to vision impairment. Based on a meta-analysis, the estimated global prevalence of DR, around 103 million people in 2020, may rise to 161 million in 2045.¹ According to a multicentric cross-sectional screening in India, 18% of participants among 42,000 were identified to have diabetes. The estimated prevalence was 12.5% for DR and 4% for vision-threatening DR, stating that 3 million people have vision-threatening DR.² In the retinal vasculature (RV), direct communication between the microvascular cells, namely, pericytes (PCs) and endothelial cells (ECs), is crucial for vessel stability and the maintenance of the blood-retinal barrier (BRB). During DR, PCs and ECs, which share a common basement membrane throughout vasculogenesis, tend to lose vessel integrity in the retina.³

Chronic hyperglycemia induces sclerosed arteries, leading to PC dropout in the capillaries. Increased venous pressure causes hemodynamic changes that damage ECs in DR. Subsequent thrombus formation raises venous pressure further, reducing retinal perfusion and resulting in ischemia, metabolic changes, microvascular cell death, and retinal angiogenesis. The proteins required to maintain BRB integrity and EC-PC homeostasis may be altered or lost during the late angiogenic stage of DR. In the retinal angiogenic process, the newly formed network of vessels does not meet the metabolic demands of the retina, causing vessel instability leading to fibrovascular membrane (FVM) formation at the interface of the retina between the inner limiting membrane and vitreous.⁴ The contraction of FVM tissue adhering to the vitreous often leads to vitreous hemorrhages and mild to severe retinal detachment,⁵ which leads to vision loss. The FVM tissue is excised during pars plana vitrectomy (PPV) surgery to treat tractional retinal detachment (TRD). And critically, FVM is enriched for vascular cells relative to whole retinal tissue, which makes it ideal for studying microvascular cells. The pathology underlying contraction of FVM due to microvascular proliferation, unstable attachment/migration of PCs and ECs, extracellular matrix deposition, and the associated proteomic changes in retinal angiogenesis remains unclear. Ideally, comparing non-diabetic retinas with mild, moderate, and severe non proliferative diabetic retinas, as well as PDR retinas from donor eyes, would provide a clearer understanding of the pathogenesis of retinal angiogenesis. However, scoring the stages of DR, accounting for postmortem changes in human donor eyes, and the unavailability of PDR donor eyes present significant study limitations. Additionally, animal models of DR have limitations in accurately recapitulating the disease pathology and selecting appropriate stages for study. In this study, we used FVM as a clinically relevant tissue to reliably uncover the proteomic changes involved in angiogenesis, focusing on microvascular cells. The lack of a suitable control is a challenge in using FVM as a study model for proliferative DR (PDR), as the epiretinal membrane from the macula may not be appropriate for comparison with peripheral FVM. Additionally, non-diabetic retina from donor eyes includes both vascular and non-vascular components, while FVM is exclusively vascular tissue. To address this, the healthy RV of the retina from non-diabetic human donor eyes, freed of nonvascular components, was compared with FVM to selectively enrich retinal microvascular cells while eliminating nonvascular neural components.

Comprehensive proteomic profiling of RV and FVM from PDR patients may provide insights into the severity of retinal angiogenesis. Therefore, the study aims to characterise the proteome of RV and FVM from PDR patients, to develop new biomarkers and therapeutic approaches for the prevention of vision loss associated with DR.

Research design and methods

Sample collection

Institutional Review Board approval was obtained from Sri Ramachandra Institute of Higher Education and Research, Chennai (IEC-NI/19/FEB/68/10; 22/04/2019). Non-diabetic donor eyes, FVM, and whole blood samples were collected from Sri Ramachandra Medical College & Research Institute, with informed consent in accordance with the Declaration of Helsinki. The study used non-diabetic donor eyes (n=6) from the eye bank, SRMC&RI, and FVM from PDR patients undergoing PPV (n=14) (online supplemental table S1). Whole blood from healthy, T2DM and DR subjects (n=5) was used for messenger RNA (mRNA) prevalidation (online supplemental table S4). The selected protein was further assessed in serum samples from T2DM (n=29), DR (n=29), and DR with nephropathy (n=28) (online supplemental table S5).

Inclusion and exclusion criteria

FVM/donor eyes

Baseline data for PDR patients included demographics, best-corrected visual acuity ≥20 and slit-lamp fundus evaluation. Exclusion criteria were uveitis, retinal vein occlusion, prior intraocular surgery except cataract, recent anti-vascular endothelial growth factor (VEGF) injection (<3 months), significant media opacity and combined rhegmatogenous TRD. Controls were age-matched non-diabetic donor eyes with no history of diabetes.

Whole blood samples

Patients on medications other than for diabetes were excluded. Demographic, clinical, and ophthalmic details were recorded, and routine lab tests, including glucose, cardiac markers, renal tests, and lipid profile, were assessed. The study included T2DM subjects without complications (HbA1c >6.4%, fasting glucose >140 mg/dL) and DR participants diagnosed using the International Clinical Diabetic Retinopathy severity scale. Exclusions included other systemic or ocular complications, intraocular infection, prior anti-VEGF (<3 months), and other eye diseases. DR patients with DN were identified by eGFR < 60 mL/min/1.73 m², albuminuria, renal ultrasonography, and ophthalmic evaluation.

Isolation of RV from the neural retina

The RVs from the neural retina of non-diabetic human donor eyes were isolated as previously reported with minor modifications.⁶ Four 8 mm peripheral retinal punches were taken, washed to remove vitreous or RPE cells and incubated in 3% trypsin (0.1 M Tris, pH 7.8) at 37°C. After 5 min, non-vascular cells disintegrated, and the vascular network was manually separated under a microscope. Among the four 8 mm punches of isolated RV, one portion was allocated for immunohistochemistry (IHC), while the remaining portions were used for mass spectrometry (MS) analysis based on total protein concentration.

Immunohistochemical analysis of microvascular cells

Before proteomics studies, the paraffin-embedded RV (n=3) and FVM (n=6) were subjected to IHC studies after antigen retrieval to evaluate the expression of microvascular cells using EC (CD-31, CD-34) and PC markers smooth muscle actin (SMA), platelet-derived growth factor receptPlatelet Derived Growth Factor Receptor-β (PDGFR-β), and desmin.⁷

Isolation of whole cell proteins from RV and FVM

The RVs were suspended in 1 mL of retinal lysis buffer and the whole cell proteins of RV (n=6) and FVM (n=6) were performed as reported earlier.⁸ The protein concentration of the supernatant was determined using the Bradford microplate assay method. The FVM sample size received during each study was a limiting factor, so the samples were pooled for further analysis.

One-dimensional gel electrophoresis

Proteins were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). Samples were adjusted to 50 µg with water, mixed with 4× lithium dodecyl sulfate (LDS) and 100 mM dithiothreitol, vortexed, and heated at 106°C for 10 min⁸ to denature the proteins. 30 µg of pooled RV and FVM (n=6) protein were loaded onto one-dimensional (1D) gels and run briefly at 20 mA for 5 min until reaching the resolving gel. Gels were stained with Coomassie Brilliant Blue R-250 for 30 min and destained overnight.

In-gel digestion

Excised gel bands were reduced with 10 mM dithiothreitol in 50 mM ammonium bicarbonate (ABC) buffer (1 hour, 56 °C), alkylated with 55 mM iodoacetamide and 50 mM ABC buffer (1 hour, RT), washed and dehydrated in 25 mM ABC in 50% (v/v) acetonitrile. Trypsin digestion was performed at 37 °C for 16 hours. Peptides were extracted twice with 70% acetonitrile/0.1% TFA, the eluates were pooled, dried and resuspended in 0.1% formic acid. Peptide yield was quantified by Nanodrop.⁹

1-D liquid chromatography-MS analysis

Proteomic analysis was conducted on an Eksigent nanoLC-425 system (Trap-Elute) with 10 µL of each sample (30 µg peptides) loaded onto a C18 trap column (Trajan ProteoCol C18P 3 µm 120 Å, 300 µm × 10 mm) and eluted through an analytical column (Thermo Scientific µPAC HPLC column, 50 cm NanoLC). Peptides were separated using a 90 min gradient at 0.3 nL/min and analysed on a SCIEX TripleTOF 5600 as mentioned earlier.¹⁰

Peptide and protein identification

Protein identification used ProteinPilot V.5.0.2 software (AB SCIEX) with the Paragon algorithm, applying a 1% false discovery rate (FDR) cut-off.¹¹ Data were searched against the Swiss-Prot human proteome and common repository of adventitious proteins (cRAP). Differentially abundant proteins were defined using paired Student’s t-test with <1% FDR, fold change ≥1.3 or ≤−1.3 (p<0.05). Exponentially Modified Protein Abundance Index (emPAI) values estimated relative protein abundance. The protein identification criteria are listed in online supplemental table S2.

Functional analysis

Protein interaction networks and functional enrichment were assessed using STRING V.12.0 and ShinyGO V.0.80 for Gene Ontology and Kyoto Encyclopedia of Genes and Genomes pathway analysis.¹²

mRNA expression studies

Total RNA was extracted from peripheral blood mononuclear cells using TRIzol and quantified with a Nanodrop. For quantitative PCR (qPCR), 100 ng RNA was reverse-transcribed to complementary DNA (cDNA), and mRNA levels of stratifin (SFN), 14-3-3 protein zeta (YWHAZ), 14-3-3 gamma (YWHAG), cofilin 1 (CFL1), calmodulin 3 (CALM3), hornerin (HRNR), and DNA-dependent protein kinase catalytic subunit (PRKDC) were measured using TB Green Premix Ex Taq II (RR820A, Takara Bio, Japan) with 10 ng cDNA. Reactions (20 µL) were run in duplicate on the QuantStudio V.5 Real-Time PCR system, and specificity was confirmed by melting curves. Primers (online supplemental table S3) were obtained from Bioserve. Cycling conditions included 95°C for 30 s, followed by 40 cycles of 95°C for 5 s and 60°C for 30 s, with a final melting curve step. Relative expression was calculated using 18S as the housekeeping gene.

Sandwich ELISA

The serum samples at the dilutions of 1:8 for HRNR and neat sample for SFN were used for HRNR (KBH2745, GENLISA, KRISHGEN Biosystems) and RBP-4SFN (KBH7014, GENLISA, KRISHGEN Biosystems) ELISA, respectively. Sandwich ELISA for SFN and HRNR was carried out as per the manufacturer’s instruction. Online supplemental table S6 includes the calibration curve range, limits of detection, intra-assay and inter-assay variability, sample dilution and quality-control procedures such as duplicate measurements and assay reproducibility checks

Sample size estimation

The serum sample size was calculated using ‘two-proportion hypothesis testing,' assuming that the effect size (protein level) may be 30% in T2DM and 66% in DR, or diabetic retinopathy with DN, with 80% power and a significance level of 0.05. This pilot study required 30 participants per group for validation.

Statistical analysis

The Kruskal-Wallis test was used to compare demographic and clinical variables, and one-way analysis of variance (ANOVA) with Dunnett’s post-test assessed differences among the three groups. ROC Area Under the Curve (AUC) values evaluated HRNR and SFN prognostic performance. Significance was set at p<0.05. Analyses were performed using GraphPad Prism (V.5; Graph-Pad Software, San Diego, California, USA) and Microsoft Excel.

Results

Clinical features of RV and FVM

RV was isolated from six non-diabetic human donor eyes (age 44–76 years, mean 57±13 years), and 14 FVM were obtained from patients undergoing PPV. Demographic details are provided in online supplemental table S1. All samples were collected according to inclusion/exclusion criteria with informed consent.

Characteristics of SMA, PDGFR-β, desmin, CD-31 and CD-34 in RV and FVM

Table 1 summarises patient characteristics and marker expression in RV (n=3) and FVM (n=6). SMA was strongly positive in both tissues, confirming smooth muscle cells and PCs. CD-34 showed mild positivity in RV and moderate levels in FVM, indicating ECs. Desmin was absent in RV but variably expressed in FVM, suggesting immature PCs in angiogenic membranes. PDGFR-β was highly expressed in all FVM samples but low in RV, consistent with angiogenic PCs. CD-31 was mildly to moderately positive in FVM and absent in RV, indicating angiogenic ECs. Figure 1 and onlinesupplemental figure 1AF show qualitative staining for PC markers (SMA, PDGFR-β, desmin) and EC markers (CD-31, CD-34), along with Ki67 confirming proliferation in FVM.

Table 1. Patient characteristics and quantitative expression of Ki67, SMA, desmin, PDGFR- β, CD-34 and in RV and FVM from retina of PDR.

S.no	Patient code	Age (years)	Gender	Diabetic duration (years)	Disease status	Ki67	Pericytes			Endothelial cells
							SMA	Desmin	PDGFR- β	CD-34	CD-31
1	RV-10	65	M	–	Non-diabetic	0	4	0	1+	1	0
2	RV-11	52	M	–		0	3	0	1+	1	0
3	RV-12	62	M	–		0	3	0	1+	2	0
4	FVM-02	49	M	14	Type II diabetes; PDR	NA	4	1+	4+	2	2+
5	FVM-03	32	M	3		NA	4	1	3+	3	2+
6	FVM-10	49	F	26		NA	3	2+	2+	2	1+
7	FVM-15	58	M	22		2+	4	3+	3+	4	2+
8	FVM-16	41	M	16		4+	3	3+	3+	2	2+
9	FVM-17	55	F	15		2+	4	3+	3+	4	2+

The scoring system is based on a scale of 0–4+ as previously described by Vorkauf et al, 1993. 4+: very high (75–100% positive cells); 3+: high (50–75% positive cells); 2+: moderate (25–50% positive cells); 1+: low (<25% positive cells); 0: negative (0 positive cells).

FVM, fibrovascular membrane; PDGFR, platelet derived growth factor receptor; PDR, proliferative diabetic retinopathy; RV, retinal vasculature; SMA, Smooth muscle actin.

Figure 1. Smooth muscle aMuscle Actin (SMA), desmin, platelet-derived growth factor recePlatelet Derived Growth Factor Receptor- β (PDGFR-β), CD-31, and CD-34 expression (representative images) in retinal vasculature (RV) and fibrovascular membrane (FVM). Desmin and PDGFR-β expression were more intense in FVM pericytes than RV; similarly, CD-31 expression was high in FVM endothelial cells. SMA and CD-34 positivity observed in both RV and FVM makes the near suitability of RV as a control for FVM. PDR, proliferative diabetic retinopathy.

1-D proteomic profiling of protein in RV and FVM samples

Semiquantitative proteomics was performed on pooled RV (n=6) and FVM (n=6) samples. The 1D SDS-PAGE profiles are shown in figure 2A, B, and nanodrop quantification yielded 3.07 µg peptides for RV and 2.57 µg for FVM. MS data were searched against the Swiss-Prot human proteome and cRAP, identifying 59 RV and 44 FVM proteins, with 46 shared as shown in figure 2C. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE^{13 14} partner repository with the dataset identifier PXD052211.

Figure 2. One-dimensional (1-D) proteomic profiling of retinal vasculature (RV) and fibrovascular membrane (FVM). (A) 1-D SDS PAGE profiling and (B) 1 cm short run profile of pooled samples of RV and FVM; (C) Venn diagram showing the number of unique and overlapping 1-D proteins of RV and FVM; (D) Functional network analysis using STRING V.12.0 online database, RV with 0.400 interaction score (D) and FVM proteins with 0.150 interaction score (E), the legend indicates the top 10 proteins associated with tissue specific to eye, blood and endocrine glands. (F) Heatmap showing the protein expression pattern of shared proteins based on Exponentially Modified Protein Abundance Index score. A1AT, alpha-1-antitrypsin; ACTB, beta-actin; ALBU, albumin; ANXA2, annexin A2; ANXA5, annexin A5; ANXA6, annexin A6; APOA1, apolipoprotein A1; CALM 3, calmodulin 3; CAH1, carbonic anhydrase 1; CAH2, carbonic anhydrase 2; CFL 1, cofilin 1; CO6A3, collagen alpha-3(VI) chain; CRABP1, cellular retinoic acid-binding protein 1; CRYAA, alpha-crystallin A chain; FIBA, fibrinogen alpha chain; FIBG, fibrinogen gamma chain; FN1, fibronectin 1; FRIL, ferritin light chain; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; GFAP, glial fibrillary acidic protein; GO, Gene Ontology; HRNR, hornerin; H37, histone H3.7; H4, histone H4; HBA, hemoglobin subunit alpha; HBB, hemoglobin subunit beta; HPT, haptoglobin; IGHG1, immunoglobulin heavy constant gamma 1; IGHG4, immunoglobulin heavy constant gamma 4; IGLC2, immunoglobulin lambda constant 2; ISM1, isthmin 1; kDa, kilodalton; KV228, immunoglobulin kappa variable 2-28; MYL6, myosin light chain 6; PM, protein marker; POSTN, periostin; PRDX2, peroxiredoxin 2; SAMP, serum amyloid P-component; SERPINF1, serpin family F member 1 (pigment epithelium-derived factor); TRF, serotransferrin; VIM, vimentin; VIME, vimentin; VTDB, vitamin D-binding protein; LC, liquid chromatography; MS, mass spectrometry; YWHAG; 14-3-3 gamma; YWHAZ, 14–3-3 protein zeta.

STRING analysis showed RV proteins enriched in glycolysis/gluconeogenesis, Alzheimer’s and HIF-1 pathways, with human YWHAZ, CALM3, and cofilin (CFL1/COF1) interacting with vimentin, actin, myosin light polypeptide six and GAPDH as shown in figure 2D. FVM proteins were enriched in complement/coagulation and cholesterol metabolism; HRNR, ISM-1, and YWHAG interacted with ANXA2, VIM, GFAP, PEDF and APOH (figure 2E).

emPAI comparison of shared proteins showed upregulation of carbonic anhydrase 1, α-crystallin A, peroxiredoxin-2 and periostin and downregulation of complement C3, vitamin D-binding protein, serotransferrin and HRNR (figure 2F). Based on functional enrichment, SFN, YWHAZ, YWHAG, CALM3, COF1, HRNR and PRKDC were selected for further validation. To identify key candidates, prevalidation qPCR analysis was performed, and the mRNA levels of SFN, YWHAG, CFL1 and HRNR were significantly reduced in DR with DN compared with T2DM and healthy controls (online supplemental figure 2), supporting their relevance for downstream validation.

Validation of serum HRNR and SFN

The demographic and clinical profiles of participants are shown in online supplemental table S5. Clinical parameters including eGFR, haemoglobin, fasting glucose, HbA1c, systolic pressure, serum urea and creatinine showed a progressive increase from T2DM to DR with DN.

Table 2 summarises serum HRNR and SFN levels. HRNR was significantly reduced in DR with DN compared with T2DM (p=0.001) (figure 3A), while SFN was significantly elevated in DR relative to T2DM (p=0.0011) (figure 3E). ROC analysis (figure 3B–D) showed that HRNR demonstrated acceptable diagnostic performance, with AUCs of 0.5803, 0.8639 and 0.7833 for T2DM versus DR, T2DM versus DR with DN and DR versus DR with DN, respectively. SFN showed AUCs of 0.7492, 0.5615 and 0.7144 (figure 3F–H). Table 3 presents AUC values, 95% CIs and significance.

Table 2: Levels. of serum HRNR and SFN in T2DM, DR and with DN.

Group	T2DM (n=29)	DR (n=29)	DR with DN (n=28)	P value
Serum HRNR (ng/mL)	10.2±7	8.8±6	3.9±2	0.001*
Serum SFN (µg/mL)	162±70	305±225	179±82	0.0011†

p<0.05, statistically significant unless specified (one way ANOVA followed by Dunnett’s Test *T2DM vs DR with DN; †T2DM vs DR).

DN, diabetic nephropathy; DR, diabetic retinopathy; HRNR, hornerin; SFN, stratifin.

Figure 3. Serum HRNR (A) and SFN (E) levels in T2DM, DR and DR with DN samples. Data shown are mean±SD (Dunnett’s multiple comparison test) *p=0.001, T2DM versus DR with DN; ** p=0.001, T2DM versus DR; ^∧p=0.0001, DR versus DR with DN, ^∧∧p=0.001, DR versus DR with DN. ROC curves for HRNR and SFN. Data of the serum samples are shown. ROC curves for HRNR were drawn to compare T2DM versus DR (B), T2DM versus DR with DN (C) and DR versus DR with DN (D). ROC curves for SFN were drawn to compare T2DM versus DR (F), T2DM versus DR with DN (G) and DR versus DR with DN (H). Values in parentheses indicate the area under the curve values with 95% CI. DN, diabetic nephropathy; DR, diabetic retinopathy; HRNR, hornerin; NS, not significant; ROC, receiver operating characteristic; SFN, stratifin; T2DM, type II diabetes mellitus.

Table 3. Hornerin and stratifin ROC analysis indicating AUC comparing T2DM, DR and with DN.

Datasets	Hornerin		Stratifin
	AUC (95% CI)	P value	AUC (95% CI)	P value
T2DM versus DR	0.58 (0.43 to 0.73)	NS	0.75 (0.62 to 0.89)	0.002
T2DM versus DR with DN	0.86 (0.77 to 0.96)	0.0002	0.56 (0.40 to 0.72)	NS
DR versus DR with DN	0.78 (0.67 to 0.90)	<0.0001	0.71 (0.57 to 0.86)	0.009

DN, diabetic nephropathy; DR, diabetic retinopathy; NS, not significant; ROC, receiver operating characteristic.

Discussion

The retinal microvasculature shows marked heterogeneity in PDR, particularly in TRD driven by ischaemic neovascularisation. Comparing FVM¹⁵ with healthy retinal vessels allowed us to explore proteomic changes associated with microvascular damage. The key aim of the study was to establish whether the RV could serve as a biologically relevant control for FVMs. The isolated RV is freed of non-vascular neural components while retaining the same fundamental vascular cell types PCs and ECs. This makes the RV directly comparable to FVM, which is also composed exclusively of vascular elements. Both RV and FVM contain PCs and ECs as part of their vascular architecture. However, in FVMs, these cells are present in a pathologically activated and recruited state, as evidenced by increased expression of PDGFR-β and desmin in PCs, and elevated CD31 in ECs, indicating active angiogenesis. In contrast, the RV represents a quiescent vascular state, providing a stable baseline for comparison. The presence of PCs and ECs in FVM,^{16 17} together with strong desmin, PDGFR-β and CD-31^{18 19} expression, indicates active angiogenic remodelling, while SMA and CD-34 positivity in both tissues reflects underlying healthy cell populations. Ki67 staining further supports the proliferative nature of FVM. These findings align with earlier reports²⁰ and reinforce the central role of angiogenic PCs and ECs in FVM formation. PDGF-B/PDGFR-β signalling is essential for retinal microvascular homeostasis, guiding EC-PC interactions that stabilise vessels.^{21 22} The higher PDGFR-β expression in FVM compared with RV suggests active recruitment of angiogenic PCs during PDR progression.²³ Similar PDGF-B-PDGFR-β-driven PC responses have been reported in neurovascular diseases such as Alzheimer’s,²⁴ underscoring its broader role in pathological angiogenesis.

1-D MS enabled high-resolution profiling of proteins involved in FVM formation, identifying 59 RV proteins, 44 FVM-specific proteins and 46 shared proteins. STRING analysis showed that FVM-unique proteins HRNR, ISM-1 and YWHAG interacted with ANXA2, VIM, GFAP and PEDF (SERPINF1), molecules associated with PC and EC function. Functional enrichment highlights pathways related to basement membrane dynamics. These processes are hallmarks of pathological angiogenesis and fibrosis in PDR. ANXA2 is known to mediate PC-EC adhesion, matrix remodelling and angiogenesis; its dysregulation contributes to early DR, and ANXA2 blockade in diabetic models prevents PC loss and pathological angiogenesis. Vimentin marks both ECs and PCs, while GFAP reflects glial activation within vessel walls. HRNR showed a high emPAI score in diabetic PC proteomes²⁵ and its predicted interaction with ANXA2 suggests a potential role in PC-associated pathology.

Among RV-specific proteins, CFL1, CALM3 and YWHAZ were major cytoskeletal regulators. Functional enrichment analysis indicates that these proteins are primarily involved in cytoskeletal organisation, smooth muscle contraction, focal adhesion, calcium-dependent protein binding and rho GTPase pathway. These pathways are characteristic of a stable, contractile and homeostatic vascular system, consistent with quiescent retinal vessels. CFL1 controls actin turnover and cell morphology.^{26 27} CALM3 influences Tie2 signalling, where impaired Ang1-Tie2 interactions destabilise vessels and reduce PC recruitment.²⁸ YWHAZ modulates contractility by regulating phosphorylation of CFL1 and MYL6; its downregulation alters ECM organisation and has been linked to glaucoma. YWHAZ also participates in glucose-stimulated insulin secretion²⁹ and is associated with enhanced VEGF and HIF-1α activity in glioma³⁰, underscoring its angiogenic relevance.

STRING analysis indicated that SFN, YWHAZ and YWHAG are enriched in PI3K-Akt signalling, a key angiogenic pathway.^{31 32} These isoforms contribute to cell-cycle control, apoptosis, signalling and energy regulation.³³ Prior findings showing PI3K-subunit targeting in FVM-derived ECs highlight this pathway as a potential therapeutic avenue in DR. Comparatively, while some core structural proteins such as ANXA2, VIM and GAPDH are shared between RV and FVM, their network context and functional associations differ significantly. In RV, they support vascular stability, whereas in FVM, they are integrated into pathways driving angiogenic activation and tissue remodelling. Overall, this network-based analysis provides functional validation of the proteomic findings, demonstrating a clear shift from vascular homeostasis in RV to pathological angiogenesis and fibrosis in FVM, thereby strengthening the rationale for biomarker selection.

Following the discovery phase, the key question in the validation stage was to determine whether the proteins identified at the retinal tissue level are also reflected systemically in DR to understand the translational relevance. So based on functional enrichment, selected proteins were prevalidated at the mRNA level, and significant candidates were further assessed systemically using serum. mRNA analysis showed marked downregulation of SFN and HRNR in DR with DN compared with diabetic controls. Serum validation confirmed these trends: HRNR levels were significantly reduced in DR with DN vs T2DM, while SFN levels were significantly elevated in DR compared with T2DM. ROC analysis showed that HRNR provided acceptable diagnostic performance, particularly for distinguishing T2DM from DR with DN, while SFN showed acceptable discrimination between T2DM and DR.

SFN has been reported to be involved in tumour angiogenesis.³⁴ It is also reported that the expression of matrix metalloproteinase-1, which is known to modify extracellular matrix, is stimulated by SFN-containing exosomes in fibroblasts, which further activates the p38/MAPK pathway.³⁵ This concept opens the possibility that SFN induces PCs for matrix remodelling which is important in retinal homeostasis and will likely be important in understanding the pathogenesis of DR. The lack or gain of SFN induces PCs/ECs, thereby reducing MMP production and resulting in a skewed retinal homeostasis towards FVM formation, which requires further investigation.

HRNR, a serine-rich protein, has been linked to neuronal intranuclear inclusion disease,³⁶ breast cancer progression³⁷ and regulation of tumour vascularity.³⁸ HRNR is a S100 fused-type protein³⁶ with 245 kDa that has been reported to be involved in non-VEGF-mediated angiogenesis³⁹ and inhibition of VEGF receptor 2 in HRNR knockdown mice models reduced tumour growth.³⁸ HRNR has been reported to be upregulated in apoptosis and/or necrosis,⁴⁰ while its expression has been noted in diverse tissues; there is a lack of information regarding its presence in human retinal sections and its role in diabetic eye disease. However, ROC curve analysis revealed that serum levels of SFN and HRNR proteins could serve as diagnostic biomarkers for DR. Molecular docking of SFN with PDGFR-β confirmed its robust interaction, stating its role in angiogenesis. To establish SFN and HRNR as potential biomarkers or therapeutic targets in the context of DR management, further studies with larger populations are needed.

Conclusion

The focus of this study was to find a novel biomarker linked to DR that primarily targets microvascular cells. HRNR and SFN were identified as candidate biomarkers for DR with DN and DR, respectively. Further studies are required to establish the proteins as reliable diagnostic markers or therapeutic drugs with larger scale samples.

Data availability statement

Data are available upon reasonable request.