Bone marrow fibrocytes and endothelial-to-mesenchymal transition drive pathological ECM remodeling in proliferative diabetic retinopathy

Abstract

Proliferative diabetic retinopathy (PDR) is characterized by pathological neovascularization and fibrovascular membrane (FVM) formation, yet the cellular origins of fibroblasts within FVMs and vitreous collagen metabolism remain poorly understood. Vitreous humor and FVM specimens were collected from PDR patients and non-diabetic controls undergoing vitrectomy. Immunohistochemistry, enzyme-linked immunosorbent assay (ELISA), and enzymatic activity assays were performed. Double-positive staining for cluster of differentiation (CD)45/collagen type I (COL1) and CD45/S100 calcium-binding protein A4 (S100A4) identified bone marrow-derived fibrocytes, while CD31/COL1 co-expression indicated endothelial-to-mesenchymal transition (EndoMT). Fibronectin was significantly elevated in PDR vitreous (p < 0.01). Matrix metalloproteinase (MMP)-2, MMP-3, and MMP-9 activities were markedly increased in PDR (all P < 0.001), alongside elevated lysyl oxidase (LOX) activity (p < 0.001). Collagen IV detection rates were higher in PDR (63.2% vs. 15%, P < 0.01), and beta cross-linked C-telopeptide of type I collagen (β-CTX) levels were significantly increased (p < 0.001). Notably, total procollagen type I N-terminal propeptide (Total-PINP) was paradoxically elevated in PDR vitreous (p < 0.01), suggesting compensatory collagen synthesis during active neovascularization. These findings demonstrate that FVM fibroblasts originate from multiple sources including bone marrow-derived fibrocytes and EndoMT, and that PDR involves simultaneous MMP-mediated collagen degradation and LOX-mediated cross-linking, providing potential targets for anti-fibrotic therapies.

Introduction

Proliferative diabetic retinopathy (PDR) is defined by the formation of fibrovascular membranes (FVMs), a pathological process driven by abnormal extracellular matrix (ECM) deposition and remodeling. This process ultimately results in tractional retinal detachment and significant visual impairment [1, 2].

In PDR, FVMs display significant contractility and mechanical tension due to excessive accumulation of ECM, including collagen and fibronectin (FN1) [2, 3], along with the activation of myofibroblasts [4]. Immunohistochemical analyses have revealed increased levels of lysyl oxidase (LOX) in diabetic neovascular membranes, where it facilitates collagen cross-linking, thereby contributing to ECM stiffening and FVM contraction [5].The formation of FVM involves various cell types, including endothelial cells and circulating fibrocytes that differentiate into myofibroblasts, as well as activated fibroblasts that produce significant amounts of ECM; their sustained activation perpetuates the pathological fibrosis characteristic of PDR [6–9].

In FVMs, FN1 is upregulated with α-SMA-positive myofibroblasts and CD31-positive endothelial cells, indicating spatial interactions such as endothelial-to-mesenchymal transition (EndoMT) [3, 4, 10]. FN1 fibrils act as reservoirs for vascular endothelial growth factor (VEGF) and transforming growth factor-beta 1 (TGF-β1), thereby sustaining autocrine and paracrine signaling that promotes both neovascularization and fibrosis [11–13]. This provisional ECM facilitates the transition from vascular granulation tissue to contractile fibrotic scars [14].

ECM homeostasis in FVMs is further disrupted by the dysregulation of matrix metalloproteinases (MMPs), which is a hallmark of PDR progression [15, 16]. Collagenases, including MMP-1, -8, -13, and membrane type 1-MMP (MT1-MMP), cleave fibrous collagens, while gelatinases, specifically MMP-2 and MMP-9, which are predominant in PDR vitreous and FVMs, degrade denatured collagen and perivascular type IV collagen (CIV). This degradation promotes vascular leakage and retinal neovascularization. The activity of MMPs is counterbalanced by tissue inhibitors of metalloproteinases (TIMPs); excessive MMP activity leads to ECM degradation and vascular leakage, whereas the upregulation of TIMP-1 and TIMP-2 promotes fibrosis.The shared patterns of MMPs and TIMPs between PDR and proliferative vitreoretinopathy (PVR) suggest a common remodeling pathway, with perivascular deposition of MMP-9 specifically driving angiogenesis in diabetic retinopathy (DR) [17, 18]. Ultimately, pathological ECM degradation in FVMs generates traction on the retina, resulting in tractional retinal detachment [19, 20].

Under hyperglycemic conditions, the expression of LOX is upregulated in retinal endothelial cells, which enhances collagen cross-linking and thickens the basement membrane. This process increases vascular stiffness and FVM contractility. The resulting matrix hardening activates mechanosensitive NF-κB signaling and cell adhesion molecules, thereby promoting neutrophil adhesion. Additionally, soluble LOX directly activates neutrophils, leading to the release of elastase and superoxide. These enzymatic and pro-inflammatory effects of LOX represent a critical mechanism underlying early vascular injury in the diabetic retina [21–23].

Significant changes in collagen metabolism take place in the vitreous body in PDR, such as heightened cross-linking, glycosylation, degradation, and recombination. Nevertheless, the precise mechanisms driving this reprogramming are not fully elucidated [21, 24–27].

Procollagen type I N-propeptide (PINP) and the C-terminal telopeptide of type I collagen (β-CTX) are established markers of type I collagen synthesis and degradation, respectively, reflecting the dynamic equilibrium of collagen metabolism [28]. In the vitreous humor, PINP and β-CTX levels provide a quantitative basis for assessing ECM remodeling in FVMs [29–31].

Type I collagen (COL1), a key structural protein of the ECM, regulates cellular behavior through its interactions with integrins and discoidin domain receptors (DDRs). These interactions transduce physical and chemical signals via the PI3K/AKT and Wnt/β-catenin pathways [32, 33]. Under diabetic conditions, COL1 expression correlates with glucose parameters and modulates fibroblast function, thereby positioning it as a potential biomarker for diabetes [34]. However, the dynamics of COL1 in the vitreous of PDR remain unexplored.

Therefore, this study aims to trace the main cellular sources of these ECM components by detecting the levels of PINP, FN1, and β-CTX, as well as the activity of MMPs and LOX in the vitreous humor of PDR patients, and in combination with cell-derived markers (endothelial cell markers vs. fibroblast markers). Clarifying the cellular origin of ECM remodeling helps to understand the fundamental mechanism of FVM formation and provides a basis for distinguishing local ECM deposition from circulating cell-mediated fibrosis, ultimately offering new ideas for targeted intervention.

Materials and methods

This study received approval from the Ethics Committee of the NO.1 People's Hospital of Jiujiang City, Jiangxi Province (Approval number : JJSDYRMYY-YXLL-2025-112). Surgical specimens were managed in accordance with the Declaration of Helsinki. Informed consent was obtained from all patients prior to their inclusion in the study.

During vitrectomy, vitreous fluid samples and effusion boxes were collected from patients diagnosed with PDR between April 2025 and April 2026. Based on the Early Treatment Diabetic Retinopathy Study (ETDRS) criteria and were scheduled for transpars vitrectomy due to persistent vitreous hemorrhage, fibrovascular membrane hyperplasia, or tractional retinal detachment. The control group comprises patients without diabetes who underwent vitrectomy for idiopathic epiretinal membrane or macular hole. Exclusion criteria include a history of previous intraocular surgery (excluding cataract surgery), active uveitis, recent trauma, or systemic infection.At the beginning of vitrectomy, before starting the intravitreal infusion of balanced salt solution, undiluted vitreous fluid samples (0.5-1.0 mL) are aspirated under standardized conditions by the primary surgeon at the NO.1 People’s Hospital of Jiujiang City, Jiangxi Province. the Jiujiang City Key Laboratory of Cell Therapy by a member of the research team, where it was rapidly frozen using liquid nitrogen.All samples were stored at -80 °C. The precipitate and fibrous membrane tissue of the liquid in the effusion box after centrifugal treatment are used for cell culture.

Cell culture

The effluent from the glass body of the liquid collection cassette of a PDR patient was subjected to centrifugation at 3000 rpm for 15 min at room temperature. The resulting precipitate was then resuspended in a fibroblast serum-free basal medium (FibroLife, USA) [6]. The precipitated tissue and fibrous membrane tissue were cultured in a 12-well flat-bottom tissue culture plate, each well containing a volume of 1 mL, under conditions of 5% CO2 at 37 °C. The precipitated tissue cells were maintained in culture for a duration of 3 days, whereas the fibrous membrane tissue cells were cultured for a period ranging from 14 to 23 days. Subsequently, these cells were utilized for cellular immunofluorescence analysis.

Immunohistochemistry

For the immunofluorescence procedure, cultured cells were fixed using 4% paraformaldehyde at ambient temperature for a duration of 15 min, followed by three washes with phosphate-buffered saline. To prevent non-specific antibody binding, cells were incubated with 2% bovine serum albumin at room temperature for 1 h, after which they were washed with staining buffer. Subsequently, the slides were incubated overnight at 4 °C with the primary antibody diluted in staining buffer. The rabbit anti-human cluster of differentiation 45 (CD45) polyclonal antibody, diluted at 1:100 (Beyotime Biotechnology, China), was employed as a marker for hematopoietic cells. The rabbit anti-human cluster of differentiation 31 (CD31) polyclonal antibody, with a dilution of 1:200 (Beyotime Biotechnology, China), served as an endothelial cell marker. Additionally, the Mouse anti-human COL1 monoclonal antibody, diluted at 1:200 (Proteintech, China), was utilized as a marker for mesenchymal cells. The Mouse anti-human alpha-smooth muscle actin (α-SMA) monoclonal antibody, diluted at 1:1000 (Boster Biological Technology, China), was used, along with the mouse anti-human S100 calcium-binding protein A4 (S100A4) monoclonal antibody, diluted at 1:100 (Proteintech, China), which functioned as a fibroblast marker. The cells were washed three times with staining buffer, after which the secondary antibody was added, and the samples were incubated in the dark for 1 h. The secondary antibodies for CD45 and CD31 consisted of BeyoAb™ Alexa Fluor 594 (AF594)-labeled donkey anti-rabbit immunoglobulin G (IgG)(H + L), diluted at 1:200. The secondary antibodies targeting COL1, α-SMA, and S100A4 were fluorescein isothiocyanate (FITC)-conjugated AffiniPure goat anti-mouse IgG (H + L) (Beyotime Biotechnology, China), used at a dilution of 1:100 (BosterBiological Technology, China). Following staining, the cells were washed three times and fixed with 2-(4-Amidinophenyl)-6-indolecarbamidine dihydrochloride (DAPI; Beyotime Biotechnology, China). (Table 1)

Table 1.

List of primary and secondary antibodies used for immunofluorescence staining (IF)

Antibody	Source	Lot/cat. no.	Dilution
Primary antibodies
Rabbit anti-CD45	Beyotime biotechnology, China	AF7839	1:100
Rabbit anti-CD31	Beyotime biotechnology, China	Ar0099	1:200
Mouse anti-COL1	Proteintech, China	67288-1-Ig	1:200
Mouse anti-a-SMA	Boster biological Tec-hnology, China	BM002	1:1000
Mouse anti-S100A4	Proteintech, China	66489-1-lg	1:100
Secondary antibodies
Alexa594 anti-rabbit IgG	Beyotime Biotechnology, China	A551251128	1:200
FITC anti-mouse IgG	Boster Biological Tec-hnology, China	BA1101	1:100

Semi-quantitative immunofluorescence scoring was implemented due to unreliable absolute fluorescence intensity caused by limited cell yields and variable imaging conditions. A four-level ordinal scale was used: negative (0), weakly positive (1), moderately positive (2), and strongly positive (3). Two independent, blinded observers assessed images against a reference atlas, achieving a Cohen’s kappa of 0.85 for inter-observer agreement. Group comparisons employed non-parametric tests (Mann–Whitney U for two groups; Kruskal–Wallis with Dunn’s post-hoc for multiple groups). Results are shown as median (interquartile range) and strict positive rate (score ≥ 2), with effect size r calculated as |Z|/√N.

Enzyme-linked immunosorbent assay (ELISA)

The concentration levels of FN1, as well as the active expression levels of MMP-2, MMP-3, and MMP-9 in vitreous fluid, were quantified using ELISA with specific kits for human FN1, MMP-2, MMP-3, and MMP-9 (FN1 kit from Reed Biotech, China; MMPs kits from Perseebio, China). For the detection of FN1 concentration, the sample was diluted 50-fold, and 200 µL of the diluted vitreous fluid was utilized for the ELISA. In contrast, the vitreous fluid sample was diluted 5-fold prior to the detection of MMP-2, MMP-3, and MMP-9 activity, with 50 µL of the diluted vitreous fluid used for each ELISA. (Table 2)

Table 2.

List of reagents used for enzyme-linked immunosorbent assay (ELISA), colorimetric experiment, magnetic particle chemiluminescence immunoassay(MP-CLIA), cell culture

Reagents	Source	Lot/cat. no.
ELISA kit
Human FN1 ELISA Kit	Reed Biotech, China	RE1089H
Human MMP-2 ELISA kit	Perseebio, China	PE-0069H1,PE-0069H2
Human MMP-3 ELISA kit	Perseebio, China	PE-0108H1,PE-0108H2
Human MMP-9 ELISA kit	Perseebio, China	PE-0149H1,PE-0149H2
Amplite™ Fluorimetric LOX assay kit	AAT Bioquest, USA	15,255
Total Type I collagen amino-terminal elongated peptide assay Kit	Hotgen Biotech, China	25,100,128
β -collagen special sequence determination kit	Hotgen Biotech, China	25,100,127
CIV quantitative determination kit	Snibe, China	463,250,311
FibroLife™ serum-free cell Culture medium	FibroLife, USA	LL-0001;LM-0001;LS-1010

Colorimetric experiment

The active expression levels of LOX in vitreous fluid were quantified using a colorimetric assay with the human Amplite Fluorescence LOX Detection Kit(AAT Bioquest, USA). The vitreous samples were not diluted prior to analysis, and 25 µl of vitreous was utilized for each colorimetric experiment. (Table 2)

Magnetic particle chemiluminescence immunoassay(MP-CLIA)

The expression levels of PINP and β-CTX in vitreous fluid were quantified using MP-CLIA with specific kits designed for human PINP and β-CTX, (Hotgen, China). The vitreous samples were analyzed without prior dilution, utilizing 100 µl of vitreous fluid for each MP-CLIA measurement.

The expression levels of CIV were quantified in vitreous fluid using MP-CLIA kits designed for human CIV (Snibe, China). Prior to analysis, the vitreous samples were diluted fourfold, with 100 µl of the diluted vitreous utilized for each MP-CLIA. (Table 2)

Statistical analyses

Data distribution was evaluated using the Shapiro–Wilk test. Categorical variables were compared between groups utilizing the chi-square test or Fisher’s exact test, as appropriate. For continuous variables, the independent samples Mann–Whitney U test was employed for univariate group comparisons. To assess the independent effect of PDR on each biomarker while controlling for potential confounders, a general linear model (ANCOVA) was applied to Total-PINP, FN1, MMP-2, MMP-3, MMP-9, LOX, and β-CTX. Additionally, binary logistic regression was utilized for CIV, with sex, age, hypertension, creatinine, and urea included as covariates. Statistical analyses were conducted using SPSS version 20.0 (IBM Corp., Armonk, NY, USA). A two-tailed p value of < 0.05 was deemed statistically significant.

Results

Clinical and laboratory baseline characteristics of the study population

Vitreous samples were collected from 39 eyes of 36 patients with PDR (56.78 ± 1.50 years of age; 21 male and 15 female, 3 had both eyes affected) during initial pars plana vitrectomy. As a control, vitreous samples were collected from 39 eyes of 39 patients (57.23 ± 1.56 years of age; 31 male and 8 female) who underwent vitrectomy for epiretinal membrane (ERM) or macular hole (MH) or Rhegmatogenous retinal detachment (RRD). The two groups were well-matched in terms of age and sex; however, as anticipated, the PDR group demonstrated significantly elevated levels of fasting blood glucose, creatinine, and urea, along with a higher prevalence of hypertension (all p < 0.01), reflecting their diabetic condition. (Table 3)

Table 3.

Characteristics of study population

Characteristic	PDR (n = 39 eyes of 36 patients)	Ctrl (n = 39 eyes of 39 patients)	p- value
Age* (Mean ± SEM, y)	56.78 ± 1.50	57.23 ± 1.56	0.835
Gender† (male/female), n	21/15	31/8	0.078
FBG‡ (Mean ± SEM, mmol/L)	7.08 ± 0.37	5.23 ± 0.12	<0.001
Creatinine‡ (Mean ± SEM,µmol/L)	216.0 ± 36.8	69.62 ± 2.22	0.001
Urea‡ (Mean ± SEM, mmol/L)	11.02 ± 0.88	5.72 ± 0.23	<0.001
HTN† (yes/no), n	24/12	9/30	<0.001
CAD† (yes/no), n	1/35	1/38	>0.999
Personal history of cerebral infarction † (yes/no), n	3/33	0/39	0.106

This presents a comparison of baseline clinical and biochemical characteristics between the PDR and control groups. Continuous variables are expressed as mean ±standard error of mean(SEM)), while categorical variables are presented as counts (n). Statistical significance was defined as P-values < 0.05. The abbreviations used are as follows: PDR proliferative diabetic retinopathy, FBG fasting blood glucose, CAD coronary artery disease, and HTN hypertension. *Independent samples t-test. † Fisher’s Exact Test. ‡ Mann–Whitney U test

Cells in the liquid samples collected from the effusion box of PDR Patients

Immunocytochemical analysis was performed on waste liquid samples collected from the effusion box post-vitrectomy in patients with PDR to identify fibroblasts and investigate their origin. The cells in the samples exhibited polymorphic characteristics, with spindle-shaped and fusiform morphologies predominating. The phenotype of CD31⁺/COL1⁺⁺/α-SMA⁻^/+− was observed, suggesting that this group of cells might be in the EndoMT state, having the ability to produce collagen, but the myofibroblast characteristics were not obvious. The phenotype of CD45⁺/COL1⁺/S100A4⁺/α-SMA⁻ highly suggests that this group of cells belongs to fibrocytes - a type of cells differentiated from circulating monocytes/hematopoietic stem cells, which possess the characteristics of collagen synthesis and fibroblasts(Fig. 1A). These cells lack the contractile function characteristic of myofibroblasts (α-SMA⁻), yet they retain the capacity to synthesize collagen. In contrast, fibroblasts were not identified in the vitreous samples from Ctrl subjects (data not shown).

Fig. 1.

Immunostaining of cells in In the culture of the effusion box waste liquid and FVMs of patients with proliferative diabetic retinopathy. A Representative images demonstrate distinct phenotypic profiles of CD31⁺ and CD45⁺ cell populations. A CD31⁺ cells exhibit weak α-SMA expression (green) with minimal co-localization in the merged image. CD45⁺ cells show positive COL1 staining (green), with merged image indicating co-expression. CD45⁺ cells are negative for α-SMA. CD31⁺ cells display strong COL1 expression (green), with prominent co-localization observed in the merged image. B CD45⁺ cells exhibit S100A4 positivity (green), as confirmed by co-localization in the merged image was detected in FVM. 800X

Semi-quantitative analysis confirmed these phenotypic patterns (Fig. 2). CD31⁺ cells demonstrated robust COL1 expression (median score 3.0, IQR 3.0–3.0; strict positive rate 100%) alongside weak or absent α-SMA staining (median score 1.0, IQR 0.0–1.0; strict positive rate 0%). This expression profile was significantly greater than that observed in α-SMA⁺ cells (p < 0.001), indicating an EndoMT phenotype with collagen synthesis capability (Fig. 2A). CD45⁺ cells exhibited elevated COL1 and S100A4 expression (median score 2.0, IQR 2.0–2.0 and 82.4% strict positive, respectively) but were negative for α-SMA, aligning with characteristics of blood-derived fibroblasts (Fig. 2B). Notably, COL1⁺ cells displayed significantly higher immunostaining scores compared to α-SMA⁺ cells (p < 0.001, Fig. 2C), reflecting their distinct functional states.

Fig. 2.

Semi-quantitative analysis was conducted on immunofluorescence staining in primary cells obtained from patients diagnosed with PDR. A Examining CD31 expression in COL1⁺ (n = 32) and α-SMA⁺ (n = 6) cells. B Assessing CD45 expression in COL1⁺ (n = 21), α-SMA⁺ (n = 14), and S100A4⁺ (n = 17) cells. C Contrasting immunostaining scores between COL1⁺(n = 53) and α-SMA⁺(n = 20) cells. D Investigating CD31 and CD45 expression within α-SMA⁺ cells. The data are depicted as mean ± SD with individual data points superimposed. Statistical significance was evaluated using the Mann–Whitney U test A, C, D or Kruskal–Wallis test with Dunn’s post-hoc analysis B. ****p < 0.001; ns, not significant

Both cell groups lacked typical myofibroblast phenotypes, suggesting that intravitreal fibrosis may arise from multiple pathways of collagen deposition, including endothelial cell transdifferentiation and fibroblasts originating from inflammatory cells. Our application of immunocytochemistry to investigate the presence of fibroblasts in the FVM obtained from patients with PDR revealed CD45/S100A4 double-positive fibroblasts (Figure. 1B). These findings indicate that fibroblasts infiltrate the FVM of PDR patients and contribute to the tissue proliferation of the membrane.

CD45⁺ cells demonstrated significantly higher CD45 expression compared to α-SMA⁺ cells (p < 0.001). In contrast, no significant difference was found between COL1⁺ and S100A4⁺ cells (p = 0.741, Fig. 2B). Among α-SMA⁺ cells, CD31 and CD45 levels were similarly low and did not differ significantly (p = 0.232, Fig. 2D). This finding is consistent with a fully transdifferentiated myofibroblast phenotype that has lost both endothelial and immune characteristics.

Comparative analysis among groups

Biochemical analyses and comparisons of vitreous samples for FN1, Total-PINP, β-CTX, MMP-2, MMP-3, MMP-9, and CIV were conducted between the PDR group and the control group. The mean concentrations of FN1, Total-PINP, β-CTX, and the specific activities of MMP-2, MMP-3, and MMP-9 are summarized in Table 4.

Table 4.

Measured biomarkers in vitreous samples in both groups

Measured markers	Mean ± SEM (Mean-Rank)
	PDR	Control	p-value
MMP-2 (U/mL)	28.42 ± 0.85(n = 20)	12.85 ± 0.67(n = 20)	p < 0.0001
MMP-3 (U/mL)	355.78 ± 15.21(n = 20)	119.19 ± 6.34(n = 20)	p < 0.0001
MMP-9 (U/mL)	130.11 ± 5.78(n = 20)	37.91 ± 5.4(n = 20)	p < 0.0001
LOX (U/mL)	2.28 ± 0.09(n = 20)	0.81 ± 0.04 (n = 20)	p < 0.0001
FN1(ng/mL)	922.21 ± 246.13 (n = 19)	288.91 ± 59.52(n = 20)	p < 0.01
Total-PINP(ng/mL)	31.24 ± 9.96(n = 20)	5.80 ± 1.29(n = 20)	p < 0.01
β-CTX(ng/mL)	0.10 ± 0.01(n = 18)	0.01 ± 0.03(n = 20)	p < 0.0001
CIV(%)	66.67(n = 16)	18.75(n = 18)	p < 0.01

SEM standard error of mean, MMP matrix metalloproteinase, LOX lysyl oxidase, FN1 fibronectin 1, PINP procollagen type I N-terminal propeptide, β-CTX C-terminal telopeptide of type I collagen, CIV type IV collagen

Concentration of FN1 in vitreous specimens

Vitreous samples were collected from 19 eyes of 18 patients with PDR. For controls vitreous samples were collected from 20 eyes of 20 patients.

FN1 concentration was highest in the vitreous. FN1 levels in vitreous fluid were significantly elevated in PDR patients compared to controls (922.21 ± 246.13 ng/mL vs. 288.91 ± 59.52 ng/mL, respectively; p < 0.01).

Specific activity of MMP-2,MMP-3,MMP-9 in vitreous specimens

Vitreous samples were collected from 20 eyes of 19 patients with PDR. For controls vitreous samples were collected from 20 eyes of 20 patients.

The highest specific activity of MMP-2, MMP-3, and MMP-9 in the vitreous humor. MMP-2 levels in the vitreous humor were significantly elevated in PDR patients compared to controls (28.42 ± 0.85 U/mL vs. 12.85 ± 0.67 U/mL, respectively; P < 0.0001). Likewise, MMP-3 and MMP-9 levels were notably higher in PDR patients than in controls (MMP-3: 355.78 ± 15.21 U/mL vs. 119.19 ± 6.34 U/mL, respectively; P < 0.0001; MMP-9: 130.11 ± 5.78 U/mL vs. 37.91 ± 5.40 U/mL, respectively; p < 0.0001).

Specific activity of LOX in vitreous specimens

Vitreous samples were collected from 20 eyes of 20 patients with PDR. For controls, vitreous samples were collected from 20 eyes of 20 patients.

The vitreous exhibited the highest specific activity of LOX. In PDR patients, the levels of LOX in vitreous fluid were notably elevated compared to controls (2.28 ± 0.09 U/mL vs. 0.81 ± 0.04 U/mL, respectively; p < 0.0001).

Magnetic particle-chemiluminescence immunoassay(MP-CLIA)

Vitreous samples were collected from 20 eyes of 17 patients with PDR. For control, vitreous samples were collected from 20 eyes of 20 patients for Total-PINP concentration, Vitreous samples were collected from 18 eyes of 17 patients with PDR. For control, vitreous samples were collected from 20 eyes of 20 patients for β-CTX concentration.

The concentration of Total-PINP in vitreous fluid samples from PDR patients (31.24 ± 9.96 ng/ml) was significantly higher than in control samples (5.80 ± 1.29 ng/ml) (p < 0.05). Likewise, levels of β-CTX were significantly higher in PDR patients compared to controls (0.10 ± 0.01 ng/mL vs. 0.01 ± 0.03 ng/mL, respectively; p < 0.0001).

Vitreous samples were collected from 16 eyes of 15 patients with PDR. For control, vitreous samples were collected from 18 eyes of 18 patients for CIV concentration.

The CIV detection rate in the PDR group (66.67%, 12/16) was significantly higher than that in the control group (18.75%, 3/18) (p < 0.01, Fisher’s exact test), the odds ratio (OR) was 8.67 (95% CI: 1.76–42.61). Furthermore, the CIV concentration in vitreous fluid was significantly elevated in the PDR patients compared with controls (84.09 ± 138.70 ng/mL vs. 9.10 ± 26.46 ng/mL, respectively; p < 0.01).

Factors affecting biomarker levels in the whole sample

To assess whether the observed group differences were independent of baseline demographic and metabolic factors, we conducted an ANCOVA with Group (PDR vs. Control) as the primary independent variable and included sex, age, hypertension, creatinine, and urea as covariates (Table 5). Following adjustment, specific activity of MMP-2, MMP-3, MMP-9, and LOX levels remained significantly elevated in the PDR group (all P < 0.001), suggesting that these biomarkers are independently linked to PDR pathology. The β-CTX variable was natural log-transformed prior to ANCOVA due to its non-normal distribution (Shapiro–Wilk test); after adjustment, it also exhibited a significantly higher level in PDR (p = 0.004). Conversely, the group differences for PINP and FN1 were rendered non-significant (p = 0.808 and P = 0.759, respectively), which can be attributed to the strong independent effect of blood urea (PINP: p = 0.004; FN1: p = 0.027). Furthermore, female sex was associated with lower specific activity of MMP-9 levels (p = 0.008); however, the elevation of MMP-9 associated with PDR persisted even after controlling for this effect.

Table 5.

Factors affecting biomarker levels in the whole sample

Variables	Mean ± SEM (Mean Rank) or B regression coefficient ± SEM
	Total-PINP(n = 40, ng/mL)	FN1(n = 39, ng/mL)	MMP-2(n = 40 ,U/ml)	MMP-3(n = 40, U/ml)	MMP-9(n = 40, U/ml)	LOX(n = 40 ,U/ml)	CIV(n = 34, %)	β-CTX_ln(n = 38, ng/mL)
Age p-Value	-0.31 ± 0.67 0.652	-12.72 ± 19.21 0.513	-0.04 ± 0.07 0.593	-0.36 ± 1.04 0.733	0.07 ± 0.37 0.856	-0.01 ± 0.01 0.282	0.02 -0.06± 0.09 0.515	-0.01 ± 0.02 0.718
Gender p-value	-9.80 ± 11.11 0.384	-185.94 ± 280.39 0.512	1.16 ± 1.38 0.408	-3.68 ± 20.55 0.859	-20.96 ± 7.37 0.008	-0.07 ± 0.12 0.578	-1.39 ± 1.62 0.389	0.60 ± 0.48 0220
Creatinine(µmol/L) p-value	-0.06 ± 0.04 0.126	-1.10 ± 0.97 0.267	-0.00 ± 0.01 0.749	-0.11 ± 0.07 0.132	-0.03 ± 0.03 0.292	0.00 ± 0.00 0.985	0.04 ± 0.03 0.144	0.000 ± 0.000 0.486
Urea(mmol/L) p-value	4.83 ± 1.56 0.004	105.80 ± 45.77 0.027	0.03 ± 0.20 0.890	2.61 ± 2.93 0.380	1.20 ± 1.05 0.262	-0.01 ± 0.02 0.517	-0.34 ± 0.42 0.387	0.10 ± 0.07 0.120
HTN p-value	-9.35 ± 10.03 0.358	-8.79 ± 277.79 0.975	-0.81 ± 1.30 0.537	16.06 ± 19.30 0.411	4.19 ± 6.92 0.549	-0.08 ± 0.11 0.453	-0.42 ± 0.19 0.723	0.92 ± 0.42 0.035
Group (Control vs. PDR)	-3.30 ± 13.46 0.808	-114.00 ± 367.94 0.759	-16.39 ± 1.63 < 0.001	-234.16 ± 24.29 < 0.001	-75.23 ± 8.71 < 0.001	-1.54 ± 0.14 < 0.001	1.16 ± 2.19 0.596	−1.53 ± 0.48 0.004

SEM standard error of mean, MMP matrix metalloproteinase, LOX lysyl oxidase, FN1 fibronectin 1, PINP procollagen type I N-terminal propeptide, β-CTX C-terminal telopeptide of type I collagen, CIV type IV collagen.HTN (hypertension). β-CTX_ln was natural-log transformed

Factors affecting the levels of biomarkers in PDR patients

In a sub-analysis focused on patients with PDR (Table 6), urea emerged as an independent predictor of Total-PINP (p = 0.042), thereby confirming that the elevation of Total-PINP is influenced by systemic metabolic burden rather than mechanisms specific to PDR. Within this subgroup, urea did not demonstrate a significant association with FN1 (p = 0.154) .Additionally, female sex was linked to lower levels of MMP-9 in PDR patients (p = 0.044), which aligns with the findings from the whole-sample analysis.

Table 6.

Factors affecting the level of each biomarker separately in patients with PDR only

Variables	Total-mean ± SEM (Mean Rank) or B Regression coefficient ± SEM
	Total-PINP(n = 20, ng/mL)	FN1(n = 19, ng/mL)	MMP-2(n = 20, U/ml)	MMP-3(n = 20, U/ml)	MMP-9(n = 20, U/ml)	LOX(n = 20, U/ml)	CIV(n = 17,%)	β-CTX_ln(n = 38) (ng/mL)
Age p-value	-0.82 ± 1.27 0.530	-15.86 ± 41.98 0.712	0.07 ± 0.15 0.653	-0.78 ± 2.52 0.762	-0.42 ± 0.87 0.635	-0.01 ± 0.02 0.536	-0.08 ± 0.24 0.696	0.00 ± 0.00 0.994
Gender p-value	-18.66 ± 23.47 0.440	-408.98 ± 529.82 0.454	-1.16 ± 2.21 0.610	-19.73 ± 36.21 0.594	-27.61 ± 12.44 0.044	-0.19 ± 0.22 0.395	1.26 ± 3.23 0.696	0.60 ± 0.76 0.446
Creatinine(µmol/L) p-value	-0.07 ± 0.06 0.289	-1.08 ± 1.53 0.490	0.00 ± 0.01 0.712	-0.14 ± 0.10 0.207	-0.03 ± 0.04 0.337	0.00 ± 0.00 0.981	0.05 ± 0.05 0.305	0.00 ± 0.00 0.843
Urea(mmol/L) p-value	5.30 ± 2.36 0.042	122.53 ± 81.02 0.154	0.05 ± 0.26 0.851	3.99 ± 4.27 0.366	1.50 ± 1.47 0.324	-0.01 ± 0.03 0.744	-0.700.34 ± 1.00 0.482	0.11 ± 0.08 0.189
HTN p-value	-15.94 ± 21.86 0.478	-215.46 ± 538.70 0.696	-1.43 ± 2.17 0.521	45.55 ± 35.42 0.219	4.10 ± 12.17 0.741	-0.06 ± 0.22 0.772	-0.88 ± 2.24 0.695	0.12 ± 0.80 0.887

SEM standard error of mean, MMP matrix metalloproteinase, LOX lysyl oxidase, FN1 fibronectin 1, PINP procollagen type I N-terminal propeptide, β-CTX C-terminal telopeptide of type I collagen, CIV type IV collagen, HTN hypertension. β-CTX_ln was natural-log transformed

Discussion

PDR manifests with pathological neovascularization, vitreous hemorrhage, and the development of fibrovascular membranes FVMs, ultimately resulting in tractional retinal detachment and profound visual impairment. The pathophysiology of PDR encompasses intricate interactions involving cellular transdifferentiation, ECM restructuring, and angiogenic signaling. Comprehending the cellular sources of fibroblasts in FVMs and the underlying molecular pathways regulating collagen turnover in the vitreous humor is crucial for the advancement of precise therapeutic interventions.

Our findings revealed that the fibroblast-like cells isolated from pure vitreous humor, effusion fluids, and fibrous membranes were polymorphic, mostly spindle-shaped and fusiform, with limited cell numbers and poor expansion capacity. Notably, CD45/COL1 and CD45/S100A4 double-positive fibroblasts were detected in vitreous fluid samples and FVMs, suggesting a potential bone marrow origin of these fibroblasts. In addition, CD31/COL1 double-positive cells were also observed, whereas α-SMA was barely expressed in CD45/α-SMA double-positive cells, and CD31/α-SMA was weakly expressed or absent. Moreover, CD45/S100A4 double-positive cells were also identified in the fibrous membrane tissue. These observations suggest that fibroblast-like cells infiltrate into the vitreous humor and that endothelial cells are undergoing endothelial-to-mesenchymal transition EndoMT, which may be involved in the pathogenesis of PDR—particularly the formation of FVM—while simultaneously generating COL1.

Recent studies by Zou et al. have further elucidated the interplay among EndoMT, inflammation, and autophagy in PDR pathogenesis, identifying COL1A2 and CDH5 as key differentially expressed genes in EndoMT-related signaling pathways [35]. Furthermore, Nijim et al. provided a comprehensive review of EndoMT as a therapeutic target in DR, highlighting its contribution to blood-retinal barrier disruption, vascular leakage, and fibrosis [36].

Previous studies have reported that serum PINP is an independent protective factor for DR (OR = 0.915, p < 0.001), with PINP levels significantly lower in DR patients than in non-DR patients (29.17 vs. 39.03 ng/mL). In this study, we found the Total-PINP concentration was significantly higher in vitreous fluid samples from PDR patients than in those from control samples (31.24 ± 9.96 vs. 5.80 ± 1.29 ng/ml, p < 0.05). This result indicates that there are significant differences between systemic metabolism and the local intraocular microenvironment. The vitreous microenvironment has relatively independent ECM remodeling characteristics. The increase in vitreous PINP reflects the active collagen synthesis during the formation process of FVM. It is important to note that multivariate ANCOVA indicated the group difference for Total-PINP became non-significant after adjusting for urea (p = 0.808). In this analysis, urea emerged as a strong independent predictor (p = 0.004). These findings suggest that elevated vitreous Total-PINP primarily reflects the systemic burden of extracellular matrix turnover and impaired renal clearance in diabetic patients, rather than acting as a specific indicator of intraocular FVM activity.FN1 bind cell surfaces and various compounds including collagen, fibrin, heparin, DNA, and actin. Abu El-Asrar et al. demonstrated that myofibroblasts in PDR can originate from infiltrating fibrocytes of bone marrow origin as well as through EndoMT, suggesting multiple cellular sources for the contractile cells within fibrovascular membranes [7].This finding is consistent with our observation of CD45/S100A4 double-positive cells, supporting the notion that circulating fibrocytes contribute to the myofibroblast population in PDR. In this study, the FN1 level in the vitreous fluid samples of PDR patients was significantly higher than that in the control group. Therefore, FN1 may be involved not only in EndoMT but also in the formation of new blood vessels. The identification of both CD45/COL1-positive fibrocytes and CD31/COL1-positive endothelial-derived mesenchymal cells in our samples supports a dual-origin model for myofibroblast accumulation in FVMs, whereby circulating progenitors and resident endothelial cells converge to produce the collagen-rich matrix characteristic of PDR. In line with the findings for Total-PINP, FN1 also lost group significance following covariate adjustment (p = 0.759). In contrast, urea exhibited an independent effect (p = 0.027), further indicating that its elevation is primarily influenced by systemic diabetic burden rather than by PDR-specific fibrogenesis.

Unlike Total-PINP and FN1, which seem to be influenced by systemic factors, MMP-2, MMP-3, and MMP-9 maintained significantly elevated levels in PDR following ANCOVA adjustment for age, sex, hypertension, creatinine, and urea (all p < 0.001). This suggests that these proteases are indicative of authentic PDR-induced ECM restructuring, unaffected by metabolic variables.MMPs are the key enzyme family responsible for degrading the ECM, including collagen. In diseases such as diabetic retinopathy, the levels of MMP-3, MMP-9, and TIMP-1 in the vitreous body significantly increase and are associated with disease progression, indicating that the imbalance of MMP/TIMP can lead to abnormal collagen metabolism—the relatively enhanced activity of MMP can intensify collagen degradation and damage the structural integrity of the vitreous gel [17, 39–41]. Furthermore, the balance between MMPs and TIMPs is critical for maintaining ECM homeostasis; disruption of this balance in favor of MMP activity has been implicated in the pathological liquefaction of the vitreous gel and membrane contraction observed in PDR. In this study, the activities of MMP-2, MMP-3, and MMP-9 in the vitreous fluid samples of PDR patients were significantly increased compared with the control group (MMP-2: 28.42 ± 0.85 vs. 12.85 ± 0.67 U/mL; MMP-3: 355.78 ± 15.21 vs. 119.19 ± 6.34 U/mL; MMP-9: 130.11 ± 5.78 vs. 37.91 ± 5.4 U/mL). These findings suggest that abnormal collagen metabolism in the vitreous occurs in patients with PDR, leading to a disruption of vitreous gel integrity. Notably, while increased MMP activity degrades existing collagen, the simultaneous upregulation of collagen synthesis by infiltrating fibroblasts and EndoMT-derived mesenchymal cells may foster a dysregulated remodeling environment in which pathological fibrosis coexists with structural destabilization.

The core of vascular complications in diabetes is the excessive deposition and cross-linking of ECM [42]. LOX is a copper-dependent enzyme known primarily for its ability to cross-link collagen and elastin in the ECM, thereby stiffening it [23]. Some studies suggest that LOX is crucial for maintaining the structural integrity of the ECM by catalyzing the covalent cross-linking of collagen and elastin side chains [22, 43, 44]. In contrast to PINP and FN1, which were primarily driven by systemic metabolic burden, LOX activity remained significantly elevated in PDR after ANCOVA adjustment for sex, age, hypertension, creatinine, and urea (p < 0.001), indicating that LOX-mediated matrix cross-linking is an independent pathogenic event in PDR.Notably, in this study, LOX activity in vitreous fluid samples from PDR patients was higher than that in control patients (2.28 ± 0.09 vs. 0.81 ± 0.04 U/mL). Meanwhile, univariate analysis showed CIV detection was markedly higher in PDR patients (66.67% vs. 18.75%, p < 0.01), suggesting increased basement membrane collagen deposition in the vitreous, LOX may be activated to promote excessive cross-linking and hardening of collagen in the basement membrane, thereby damaging the blood-retinal barrier.

Studies have found that lower serum PINP and β-CTX levels are associated with a higher incidence of DR, and β-CTX is related to the severity of DR. We found that vitreous fluid β-CTX levels were also significantly elevated in PDR patients compared to controls (0.10 ± 0.01 vs. 0.01 ± 0.03 ng/mL; p < 0.001). Therefore, although overall bone metabolism tends to be characterized by low turnover in the diabetic state, there is active extracellular matrix remodeling in the local ocular environment, particularly in PDR. The increase of β-CTX—a degradation product of COL1—in the vitreous may reflect the formation and remodeling process of the neovascular membrane.

This study indicates that in patients with PDR, there are multiple sources of fibroblast-like cells (such as CD45/COL1 and CD45/S100A4 double-positive cells) in the vitreous body, accompanied by EndoMT of endothelial cells. These processes may be involved in the formation of the vitreous FVM. Meanwhile, the level of FN1 in the vitreous of PDR patients is significantly elevated, suggesting its involvement in EndoMT and neovascularization. The activities of MMP-2, MMP-3, and MMP-9 were significantly enhanced, indicating intensified collagen degradation and destruction of the vitreous gel structure. The increase in LOX activity and CIV levels suggests that LOX is activated under high-glucose conditions, leading to excessive cross-linking and hardening of the basement membrane and damage to the blood-retinal barrier. Furthermore, although serum PINP is decreased in DR (suggesting low bone turnover), the COL1 degradation product β-CTX in the vitreous body is significantly increased, reflecting active remodeling of the local extracellular matrix. In conclusion, the vitreous body in PDR is characterized by collagen metabolism disorder, abnormal matrix cross-linking, and cell transdifferentiation, which jointly promote FVM formation and disease progression.

From a clinical perspective, these findings have important implications for PDR management. Current anti-VEGF therapies, while effective in suppressing neovascularization, may not adequately address the fibrotic component of FVMs. Targeting the fibrotic cascade identified in this study, including myofibroblast accumulation, MMP-driven collagen degradation, and LOX-mediated matrix cross-linking, could provide complementary therapies to anti-VEGF treatments.Comparative proteomic and single-cell transcriptomic analyses of vitreous samples from PDR patients at different stages may further clarify the dynamic interplay between ECM synthesis and degradation in disease progression. Our study provides novel insights into the cellular and molecular mechanisms underlying FVM formation in PDR and lays the groundwork for developing targeted anti-fibrotic therapies.

Limitations of the study

The sample size of 20 per group is relatively small, which limits generalizability. Additionally, the cross-sectional design is unable to capture the temporal dynamics of collagen metabolism markers during PDR progression. Furthermore, the specific signaling pathways that connect hyperglycemia to EndoMT and fibrocyte recruitment have yet to be clarified. Lastly, the control group consisted of non-diabetic patients with other ocular pathologies, potentially introducing confounding factors.Furthermore, the evaluation of dual-positive cells in this research was conducted using immunofluorescence instead of flow cytometry due to the restricted yield and inadequate proliferative potential of primary cells derived from vitreous samples. Subsequent investigations with increased cell numbers or enhanced culture conditions could gain from utilizing Fluorescence-Activated Cell Sorting-based analysis to ascertain the exact proportion of individual cellular subsets.

Conclusions

This study demonstrates that FVM fibroblasts in PDR originate from multiple cellular sources, including bone marrow-derived fibrocytes and EndoMT. We further reveal that PDR involves simultaneous MMP-mediated collagen degradation and LOX-mediated cross-linking, with a paradoxical elevation of Total-PINP suggesting compensatory collagen synthesis during active neovascularization. These findings identify potential targets for anti-fibrotic therapies in PDR.

Author contributions

Conceptualization, T.T.X. (Ting-ting Xiao)and Z.P.Y.; writing—original draft preparation: T.T.X. (Ting-ting Xiao) and Z.P.Y; writing—review and editing: B.W.,T.T.X. (Ting-ting Xiao), Z.P.Y.; methodology, T.T.X. (Ting-ting Xiao); supervision, Z.P.Y., T.T.X. (Ting-ting Xiao). All authors have read and agreed to the published version of the manuscript.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

The study was carried out following the Declaration of Helsinki and received approval from the Ethics Committee of the First People’s Hospital of Jiujiang City, Jiangxi Province, China (Approval number : JJSDYRMYY-YXLL-2025-112; Approval date: April 1, 2025).

Informed consent

Informed consent was obtained from all subjects involved in the study.

Consent to participate

Informed consent was obtained from all individual participants included in the study.