Triad of homocysteine, ROCK2 and vimentin as a convergent biomarker signature in pseudoexfoliation

Abstract

Background

Pseudoexfoliation (PEX) is a complex age-related disorder traditionally considered ocular in origin but increasingly recognized as a systemic condition involving extracellular matrix (ECM) dysregulation, oxidative stress, and cytoskeletal alterations. The disease typically presents as pseudoexfoliation syndrome (PEXS), with a subset of patients progressing to pseudoexfoliation glaucoma (PEXG), characterized by elevated intraocular pressure and optic nerve damage. While genetic susceptibility has been explored, dynamic molecular markers that reflect disease activity and progression remain underdeveloped.

Objective

This study investigates the plasma levels of three candidate biomolecules, homocysteine (Hcy), Rho-associated coiled-coil containing protein kinase 2 (ROCK2), and vimentin (VIM), in individuals with PEXS, PEXG, and age-matched controls, to evaluate their potential as systemic indicators of disease presence and progression.

Methods

A total of 72 participants (24 controls, 24 PEXS, 24 PEXG) were recruited from a single Indian cohort. Plasma Hcy was measured via fluorometric assay, and ROCK2/VIM by ELISA. Analyses included ANOVA, effect size estimation, logistic regression, LASSO modelling, ROC analysis, and unsupervised clustering. Marker interactions and stage associations were assessed through correlation and distribution analyses. Functional validation involved HLE-B3 cells treated with Hcy, followed by immunoblotting for ROCK2 and VIM. Bioinformatics (pathway enrichment, STITCH, TF prediction) identified upstream regulators and pathways.

Results

All three biomarkers were significantly elevated in PEXS and PEXG compared to controls (p < 0.001), with large effect sizes. Hcy and VIM levels progressively increased from PEXS to PEXG, while ROCK2 peaked in PEXS. The three-marker model combining Hcy, ROCK2, and VIM demonstrated a robust diagnostic profile, achieving a high sensitivity of 0.94 and specificity of 0.88, indicating a more balanced classification performance compared to individual or two-marker combinations. Clustering analyses revealed three molecular subgroups with moderate alignment to clinical staging (ARI = 0.32). Correlation and distribution analyses suggested stage-specific marker interactions. Bioinformatic findings present SP1 as a shared upstream regulator linking Hcy to ROCK2 and VIM expression, supported by in vitro findings.

Conclusion

Homocysteine, ROCK2, and vimentin form a clinically relevant biomarker panel with strong diagnostic and staging utility. The findings suggest a mechanistic axis driven by Hcy, possibly through SP1-mediated regulation, contributing to cytoskeletal and fibrotic alterations. This integrated pathway highlights potential molecular targets for pharmacological intervention and offers a foundation for future risk-based screening for pseudoexfoliation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07437-8.

Introduction

Systemic disorders are characterized by pathological processes that affect multiple organ systems through shared mechanisms. Systemic abnormalities can manifest in the eye, making ocular findings an important window into the broader state of systemic health [1]. While many ocular diseases are traditionally considered localized, emerging evidence suggests that some may originate from or be exacerbated by systemic dysfunctions. This perspective is particularly relevant in the case of pseudoexfoliation (PEX [MIM: 177650]), an age-related condition involving the abnormal accumulation of heterogeneous fibrillar extracellular material, known as pseudoexfoliative material (PEXM), on anterior segment structures such as the lens capsule (LC), iris, ciliary body, and trabecular meshwork [2, 3]. It is now increasingly recognized as a systemic disorder as PEXM has also been detected in extraocular tissues, including the skin, heart, lungs, liver, kidneys, and meninges in the PEX-affected individual [4–6].

The disease initially presents as pseudoexfoliation syndrome (PEXS), a non-glaucomatous stage, but in a subset of patients, it advances to pseudoexfoliation glaucoma (PEXG), an aggressive secondary open-angle glaucoma with elevated intraocular pressure (IOP) and progressive optic nerve damage, ultimately leading to irreversible vision loss [2].

PEXG is more difficult to manage than other forms of glaucoma due to its aggressive and unpredictable nature. There are currently no established medical or surgical therapies available to directly treat PEXS [7]. Consequently, partial clinical management is limited to monitoring and addressing secondary ocular complications arising from the disease. As there is no cure for pseudoexfoliation itself, treatment primarily focuses on IOP control, timely cataract extraction, and regular follow-up to prevent vision-threatening outcomes. However, therapeutic outcomes are often limited by characteristic features of PEXS, including zonular instability and poor pupillary dilation, which predispose to intraoperative complications and progressive glaucomatous damage, marked by progressive optic nerve degeneration due to high, unstable IOP and toxic deposition of pseudoexfoliative material, resulting in rapid thinning of the retinal nerve fibre layer and a significant risk of vision loss [8, 9]. Detection of PEXG relies on careful slit-lamp and gonioscopic examination, IOP monitoring, and optic nerve imaging [9]. Nevertheless, PEXG often remains undetected until advanced stages due to the absence of early clinical signs and reliable diagnostic tools, underscoring the need for biomarkers that enable early detection and risk stratification.

Initial efforts to identify risk factors mainly focused on genetic associations, aiming to find inherited variants that could help in early detection and prognosis. Several susceptibility loci have been identified in genes involved in fibrotic pathways and extracellular matrix (ECM) regulation, including Lysyl oxidase-like 1 (LOXL1), calcium voltage-gated channel subunit alpha 1 A (CACNA1A), contactin-associated protein-like 2 (CNTNAP2), clusterin (CLU), fibulin-5 (FBLN5), matrix metalloproteinases (MMPs), and their tissue inhibitors (TIMPs), among others. These genes are functionally connected to Transforming growth factor beta 1 (TGF-β) signaling and ECM remodeling, which are key to the development of PEXS. However, the clinical usefulness of these findings has been limited by population differences, modest predictive accuracy, and an incomplete understanding of how these genetic factors influence disease mechanisms [10–20].

Furthermore, genetic data alone fail to capture the dynamic biological processes involved in disease onset and progression. To bridge this gap, the focus has shifted toward circulatory biomarkers, including both proteomic and metabolomic profiles, that reflect the systemic footprint of PEX, evident in altered plasma, serum, and aqueous humor profiles. Multiple studies have reported elevated levels of inflammatory cytokines like Transforming growth factor beta 1 (TGF-β1), Interleukin-8 (IL-8), oxidative stress markers, and matrix-degrading enzymes in these biofluids. In addition, altered levels of several metabolites like amino acids, lipids, and oxidative stress markers have been reported, highlighting the presence of active circulating signals that mirror tissue-level dysfunctions central to PEX pathogenesis [20–24]. However, these studies have traditionally focused on isolated markers in separate studies, often within a localized ocular context rather than a systemic one. This reductionist approach overlooks the interconnected nature of pathophysiological processes. Without assessing network-level interactions, isolated biomarker evaluation often lacks the specificity needed to distinguish PEX from phenotypically similar conditions. To overcome these limitations, we hypothesize that evaluating a panel of functionally linked circulating biomarkers including both protein and metabolite, rather than isolated markers, can improve diagnostic specificity. Increasing evidence suggests that PEX pathology revolves around systematic impairment of proteostasis, dysregulated chaperone system, an imbalance in ECM maintenance, and a possible compromise of the cytoskeletal dynamics [17, 25–27]. Against this backdrop, specific molecular candidates such as homocysteine (Hcy), Vimentin (VIM), and Rho-associated coiled-coil containing protein kinase 2 (ROCK2) were considered as biologically plausible plasma biomarkers due to their known involvement in pathways relevant to PEX pathophysiology. Our previous studies have identified differential expression of ROCK2 and VIM in lens capsule tissues of PEX patients, both mediators of cytoskeletal dynamics [28, 29]. VIM, an intermediate filament protein involved in cytoskeletal integrity, cellular stress responses, and protein turnover, has also been upregulated in plasma and AH. Beyond its structural role, VIM is now recognized as a dynamic mediator of cellular homeostasis, including autophagic processes and inflammatory signaling, both of which are perturbed in PEX [30–32].

Meanwhile, ROCK2, a key effector in the RhoA signaling pathway, regulates cytoskeletal organization, ECM remodelling, and cell contractility. Its dysregulation has been linked to increased aqueous outflow resistance and glaucomatous optic nerve damage. ROCK2 has also been implicated in IOP regulation, ciliary muscle contractility, and neurotoxicity, with inhibitors shown to reduce IOP and counter Aβ-induced neurotoxicity [33–36].

Hcy, a sulfur-containing amino acid, is a well-established marker of systemic metabolic stress and vascular dysfunction. It is an established mediator of oxidative damage, alters cytoskeletal dynamics, and modulates Rho/ROCK signaling mechanisms that are relevant to endothelial injury and overlap with those observed in PEX-affected ocular tissues. Several studies have reported elevated Hcy concentrations in plasma and aqueous humor, suggesting a link between systemic metabolic dysregulation and local ocular tissue pathology. Taken together, the vascular, metabolic, and cytoskeletal effects of Hcy position it as a biologically plausible and mechanistically relevant circulating biomarker [37–39].

Hcy, ROCK2, and VIM form a biologically relevant marker panel that may reflect interconnected processes. Although these markers have been studied individually, their combined evaluation as circulating indicators remains largely unexplored [28, 29, 40–43]. Examining them together in the same individual may help reveal coordinated molecular disturbances and improve differentiation between disease states. We hypothesize that joint profiling of Hcy, ROCK2, and VIM in plasma, a minimally invasive source, can enhance diagnostic precision and provide insight into disease progression. This study assesses their individual and combined expression patterns across clinically stratified groups to evaluate their potential as a composite biomarker signature.

Materials and methods

Study subjects’ selection, recruitment, and sample collection

This study was approved by the Institutional Biosafety and Human Ethics Committee of the National Institute of Science Education and Research (NISER), Bhubaneswar, and adhered to the tenets of the Declaration of Helsinki. Participants were recruited from Sri Sri Borda Hospital, Bhubaneswar, India, in collaboration with Dr. Pranjya Paramita Mohanty. The criteria for subject selection followed our previously published protocol [25]. All participants underwent a detailed ocular examination, including slit-lamp biomicroscopy, ocular biometry, Goldmann applanation tonometry, + 90D fundus evaluation, and four-mirror gonioscopy. Cataract patients aged above 40 years with clinically evident pseudoexfoliative material over the lens capsule and pupillary ruff, untreated intraocular pressure (IOP) less than 21 mmHg, and no visual field defects were classified as having PEXS. Patients with untreated IOP above 21 mmHg, glaucomatous optic nerve head damage, and repeatable visual field defects consistent with disc damage were categorized as having PEXG. Patients with retinal or corneal pathology that could interfere with reliable visual field testing were excluded from the PEXG group. Age-matched cataract patients without clinical signs of PEXS or PEXG, with untreated IOP less than 21 mmHg, normal optic discs, and normal visual fields were included as controls. Participants with systemic diseases such as diabetes were excluded. All subjects were newly diagnosed at presentation and had not received prior IOP-lowering therapy. Following diagnosis, they were promptly started on appropriate medications, and those requiring further intervention underwent glaucoma surgery. Age- and sex-matched participants were selected from this well-characterized cohort for the study. A total of 72 participants were enrolled, including 24 each in the control, PEXS, and PEXG groups. Informed written consent was obtained from all participants prior to their recruitment. Peripheral blood was collected in EDTA-coated vacutainers and centrifuged at 3000 rpm for 10 min at 4 °C to separate plasma, which was aliquoted and stored at − 80 °C until use. Age- and sex-matched plasma samples were used for all downstream assays.

Enzyme-linked immunosorbent assay (ELISA)

Plasma levels of VIM and ROCK2 were measured using commercial ELISA kits (Human Vimentin ELISA Kit, CSB-E08982h; and Human ROCK2 ELISA Kit, CSB-E14993h; both from CUSABIO) according to the manufacturer’s protocols. For VIM quantification, plasma samples were diluted 1:1000 using the supplied dilution buffer, while ROCK2 levels were measured using undiluted plasma due to its low abundance. Samples were added to 96-well plates pre-coated with capture antibody and incubated at 37 °C for 2 h. After washing to remove unbound proteins, wells were incubated with biotin-conjugated detection antibody for 1 h, followed by horseradish peroxidase (HRP)-conjugated avidin for another hour at 37 °C. Colour development was achieved by adding substrate solution and incubating for 20 min, after which the reaction was stopped using 2 N sulfuric acid. Absorbance was measured at 450 nm with a reference wavelength of 540 nm using a Varioskan Flash Multimode Reader (Thermo Scientific).

Quantification of plasma homocysteine levels

Total plasma homocysteine levels were measured using a fluorescence-based Homocysteine Assay Kit (ab228559, Abcam, USA), following the manufacturer’s instructions. Plasma samples were thawed on ice and centrifuged at 12,000 × g for 10 min at 4 °C to remove precipitates. For each reaction, 10 µL of clarified plasma was used. The assay uses an enzymatic system that reduces disulfide forms of homocysteine to free homocysteine, which is then cleaved by a homocysteine-selective enzyme to generate a reactive intermediate. This intermediate reacts with a fluorogenic probe to produce a stable fluorophore with excitation/emission maxima at 658/708 nm. The reactions were carried out in 96-well plates compatible with fluorescence detection, and the fluorescence intensity was recorded using the Varioskan Flash Multimode Reader (Thermo Scientific). A standard curve was generated using serial dilutions of the provided homocysteine disulfide standard, and concentrations in the plasma samples were interpolated from this curve.

Cell culture

Human lens epithelial cells (HLE-B3; ATCC CRL-11421, Virginia, USA), derived from infant human lens epithelium and immortalized using an adenovirus 12-Simian Virus 40 (Ad12SV40) hybrid, were cultured in Dulbecco’s Modified Eagle Medium: Nutrient Mixture F-12 (DMEM/F12; 16000044, Invitrogen, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS; Invitrogen) and 1% penicillin-streptomycin (A001, HiMedia, India). Cells were maintained at 37 °C in a humidified incubator with 5% CO₂. Cells were treated with L-homocysteine (Sigma-Aldrich, Cat# 69453) at final concentrations of 500 µM or 1000 µM for 48 h prior to sample collection for downstream analyses. Sterile water was used as the vehicle control.

Immunoblotting

Immunoblotting was performed to evaluate the expression of VIM and ROCK2 in HLE-B3 cells. Protein lysates were separated on 12% SDS-polyacrylamide gels and transferred to PVDF membranes (IPVH00010, Millipore-Merck). Membranes were incubated overnight at 4 °C with primary antibodies against vimentin (ab45939, Abcam; 1:500), ROCK2 (sc-398519, Santa Cruz Biotechnology; 1:200), and GAPDH (ABM22C5, Abgenex, India; 1:250) as a loading control. Following incubation with appropriate HRP-conjugated secondary antibodies (goat anti-rabbit IgG and goat anti-mouse IgG; GeNei, India; 1:5000), signal detection was performed using Clarity Max™ Western ECL substrate (Bio-Rad, Hercules, CA, USA) according to the manufacturer’s instructions. Chemiluminescent signals were captured using the ChemiDoc™ MP Imaging System (Bio-Rad) and quantified by densitometry using Image Lab software (version 5.2.1, Bio-Rad).

Statistical analysis

Age-sex matched samples were taken for the experiments. No data was missing for the participants. Statistical analyses were performed using GraphPad Prism v9.5, SPSS (version 23.0; SPSS, Chicago, IL)., and R v4.2.2. Normality and homogeneity of variance were assessed using the Shapiro–Wilk and Levene’s tests, respectively. Parametric data were analyzed using one-way ANOVA with Sidak’s post hoc test; nonparametric data used Kruskal–Wallis with Dunn’s correction. Two-group comparisons employed unpaired t-tests and Mann–Whitney U tests for parametric and non-parametric data, respectively. Correlations were assessed using Spearman coefficients. Diagnostic performance was evaluated using logistic regression and ROC curve analysis, reporting AUC, sensitivity, specificity, and cut-off values. DeLong’s test compared AUCs. Model calibration was assessed via the Hosmer–Lemeshow test and bootstrap-based calibration curves. LASSO regression with repeated 5-fold cross-validation and 1,000 bootstraps was used to prevent overfitting. Risk stratification was based on predicted probabilities, and group distributions were tested using Chi-square analysis. Decision curve analysis (DCA) evaluated clinical utility, and k-means clustering with PCA was used to identify biomarker-based subgroups. A p-value of < 0.05 was considered statistically significant for all tests and indicated by asterisks in the figures.

Results

Demographic and clinical data

A total of 72 participants (24 controls, 24 PEXS, and 24 PEXG samples) were included in the study. The study groups were age- and gender-matched to minimize confounding effects and enhance the reliability of biomarker identification. The mean age of the control group was 67.6 ± 9.2 years, while the mean age of the PEXS group was 69.3 ± 9.5 years, and the PEXG group had a mean age of 68.0 ± 8.6 years. The mean age for the PEX (PEXS + PEXG) group is 68.7 ± 9.0 years. Statistical analysis revealed no significant age differences between the groups (p = 0.78). Gender distribution was balanced across the groups, with a male-to-female ratio of 14:10, showing no significant differences between groups (Supplementary Table 1). The number of females visiting the hospital was relatively very low, as elderly women are less likely to receive cataract surgery compared to men in low-resource countries/ localities.

Elevated plasma levels of homocysteine, ROCK2, and VIM in pseudoexfoliation

To investigate systemic alterations associated with PEX, concentrations of Hcy, ROCK2, and VIM in plasma were quantified across the control, PEXS, and PEXG groups. Elevated Hcy levels have been previously reported in PEX within select populations; its status in the East Indian population has not been characterized. All three markers were measured concurrently from the same set of individuals to enable paired comparisons and minimize inter-individual variability. In the present study, plasma Hcy concentrations were quantified to evaluate group-wise differences. A significant difference in plasma Hcy levels was observed among the study groups, as determined by the Kruskal–Wallis test (H = 52.13, p < 0.0001). Post hoc Dunn’s test revealed significantly elevated Hcy levels in the PEX group (35.47 ± 17.67 µmol/L, n = 48) compared to controls (8.65 ± 5.44 µmol/L, n = 24; p < 0.0001). Subgroup analysis showed the highest concentrations in PEXG (38.22 ± 19.17 µmol/L), followed by PEXS (32.72 ± 15.96 µmol/L) (Fig. 1A). Dunn’s test further confirmed significantly increased Hcy levels in both PEXS and PEXG compared to controls (p < 0.0001), with no significant difference between the two disease subgroups (p > 0.99).

Fig. 1.

Plasma levels of homocysteine, ROCK2, and vimentin in pseudoexfoliation syndrome and glaucoma. Column scatter plots display quantified plasma concentrations of (A) Hcy, (B) ROCK2, and (C) VIM which were significantly higher in the PEX (PEXS + PEXG) group (n = 48), as well as in both PEXS and PEXG subgroups (n = 24 each), compared to controls (n = 24). Kruskal–Wallis tests revealed significant group-wise differences for all markers (p < 0.001 for Hcy and VIM, p = 0.001 for ROCK2). Dunn’s post hoc test confirmed elevated levels in PEXS and PEXG vs. controls (p < 0.01), with no significant differences between the two PEX subgroups. Data are presented as mean ± SD

To evaluate the systemic expression of ROCK2 and VIM, plasma levels were quantified using ELISA. A significant difference in plasma ROCK2 levels was observed among the study groups, as determined by the Kruskal–Wallis test (H = 15.13, p = 0.0017). Post hoc Dunn’s test indicated significantly elevated ROCK2 levels in the PEX group (455.89 ± 216.91 pg/mL, n = 48) compared to controls (277.47 ± 116.66 pg/mL, n = 24; p = 0.0012). Subgroup analysis revealed the highest levels in PEXS (469.94 ± 263.75 pg/mL), followed by PEXG (441.83 ± 161.78 pg/mL) (Fig. 1B). Dunn’s test further confirmed significantly increased ROCK2 levels in both PEXS (p = 0.0090) and PEXG (p = 0.0045) relative to controls, with no significant difference between the two disease subgroups (p > 0.99). Given our previous findings implicating VIM as a potential biomarker for PEX, plasma VIM levels were re-evaluated to validate this observation and assess its association with ROCK2 and Hcy. VIM levels mirrored the trend observed for ROCK2. As elevated VIM levels had previously been reported in pseudoexfoliation, plasma levels were also re-evaluated to confirm findings and explore their relationship with other proteins. Plasma VIM levels also differed significantly across groups (Kruskal–Wallis test, H = 46.63, p < 0.0001). Dunn’s post hoc analysis showed significantly higher VIM concentrations in the PEX group (651.90 ± 179.20 ng/mL, n = 48) compared to controls (408.04 ± 72.90 ng/mL, n = 24; p < 0.0001). Among subgroups, VIM levels were highest in PEXG (694.27 ± 188.78 ng/mL), followed by PEXS (609.53 ± 161.97 ng/mL) (Fig. 1C). Dunn’s test confirmed significantly elevated VIM levels in both PEXS and PEXG versus controls (p < 0.0001), with no significant difference between the subgroups (p = 0.39). All values are expressed as mean ± SD unless otherwise stated.

Further, Normality testing (Shapiro–Wilk test) done prior to statistical tests revealed that the concentration of Hcy and ROCK2 levels were normally distributed in both the control and PEXG groups (p > 0.05), but not in the PEXS group (p = 0.0022 and 0.0053, respectively). In contrast, VIM exhibited a normal distribution in the control and PEXS groups (p = 0.23 and 0.72), but showed significant deviation from normality in the PEXG group (p < 0.0001). These observations were consistent with Q–Q plot analyses, demonstrating deviations from expected linearity (Fig. 2A-C). These distribution differences suggest underlying biological heterogeneity between disease stages, as observed in increased variability for VIM in advanced PEXG, and more heterogeneous patterns for Hcy and ROCK2 in the earlier PEXS group.

Fig. 2.

Distribution analysis and effect sizes of plasma biomarkers across study groups. (A–C) Q–Q plots display the distribution of plasma Hcy, ROCK2, and VIM levels in Control, PEXS, and PEXG groups relative to a theoretical normal distribution. Deviations from the reference line (y = x) indicate non-normality. (D) Violin plots show biomarker distribution and effect sizes across group comparisons: Control vs. PEX group, Control vs. PEXS, Control vs. PEXG, and PEXS vs. PEXG. Boxplots indicate medians and interquartile ranges. Cohen’s d values quantify effect sizes. Hcy and VIM showed large effects and consistent upregulation across disease groups, while ROCK2 showed moderate elevation. Minimal separation was observed between PEXS and PEXG for all markers

Further, effect size analysis revealed that Hcy showed consistently large differences across all control–disease comparisons (Cohen’s d = − 1.81 to − 2.10; η² >0.42), indicating strong discriminatory potential. VIM also demonstrated large effects, especially in PEXG (d = − 2.00), supporting its role in disease progression. ROCK2 showed moderate to large effects (d ≈ − 0.94 to − 1.17) but lower variance explained (η² ≈ 0.17–0.26), suggesting early but less dynamic changes. PEXS vs. PEXG comparisons yielded small effect sizes (d < |0.48|), indicating minimal stage-specific differences (Fig. 2D).

Diagnostic performance of individual plasma biomarkers in differentiating PEXS and PEXG from controls

To evaluate the diagnostic utility of individual plasma biomarkers, ROC curve analyses were performed for Hcy, ROCK2, and VIM (Fig. 3). The diagnostic performance of each biomolecule was assessed separately for PEXS, PEXG, and the PEX group against the control group. Hcy consistently demonstrated excellent diagnostic performance across all comparisons. The area under the curve (AUC) for distinguishing all PEX cases from controls was 0.98 (95% CI: 0.95–1.00), with an optimal cut-off value of > 15.09 µmol/L (Fig. 3A). In subgroup analyses, Hcy yielded an AUC of 0.98 (95% CI: 0.95–1.00) with a cut-off of > 15.19 µmol/L for distinguishing PEXS from controls, and an AUC of 0.97 (95% CI: 0.92–1.00) with a cut-off of > 21.40 µmol/L for differentiating PEXG (Fig. 3B and C). These results indicate that Hcy levels increase with disease severity, as reflected in the higher cut-off for PEXG. In contrast, ROCK2 displayed moderate discriminatory performance, with AUCs of 0.77 (95% CI: 0.66–0.88), 0.74 (95% CI: 0.60–0.88), and 0.80 (95% CI: 0.68–0.93) for the PEX, PEXS, and PEXG comparisons, respectively. The corresponding cut-off values for ROCK2 were > 375.10 pg/mL, > 378.50 pg/mL, and > 353.10 pg/mL (Fig. 3D and E, and 3F). VIM also showed high diagnostic utility as Hcy, particularly in the PEXG group, where it achieved an AUC of 0.99 (95% CI: 0.97–1.00) with a cut-off of > 486.30 pg/mL. For the PEX group and PEXS group, VIM produced AUCs of 0.94 (95% CI: 0.88–0.99) and 0.89 (95% CI: 0.79–0.99), with corresponding cut-offs of > 484.60 ng/mL and > 467.30 ng/mL, respectively (Fig. 3E and F). Compared to Hcy, VIM’s performance was similar in PEXG but slightly lower in earlier disease stages (Fig. 3G, H and I). Among the three biomarkers evaluated, Hcy yielded the highest AUCs across all comparisons, indicating strong potential as a standalone biomarker for PEX-related conditions. VIM followed closely, particularly in PEXG, while ROCK2 consistently demonstrated the lowest performance. While the AUCs for Hcy and VIM were similar in specific comparisons, particularly in PEXG, overlapping confidence intervals suggest that the observed differences in performance may not reach statistical significance. While Hcy exhibited the highest overall diagnostic accuracy, its optimal cut-off values increased with disease progression, suggesting a stage-dependent upregulation. In contrast, VIM demonstrated relatively stable thresholds across stages, and ROCK2 showed only modest shifts, potentially reflecting distinct biological functions or expression dynamics throughout disease progression.

Fig. 3.

Receiver operating characteristic curves (ROC) of plasma homocysteine, ROCK2, and Vimentin for distinguishing PEXS and PEXG from control subjects. (A-C) Hcy showed excellent diagnostic performance with AUCs of 0.98 (95% CI: 0.95–1.00) for Control vs. PEX (PEXS + PEXG), 0.98 (95% CI: 0.95–1.00) for Control vs. PEXS, and 0.97 (95% CI: 0.92–1.00) for Control vs. PEXG. Optimal cut-off values were > 15.09 µmol/L, > 15.19 µmol/L, and > 21.40 µmol/L, respectively. (D-F) ROC analysis of ROCK2 yielded AUCs of 0.77 (95% CI: 0.66–0.88), 0.74 (95% CI: 0.60–0.88), and 0.80 (95% CI: 0.68–0.93) for Control vs. PEX (PEXS + PEXG), PEXS, and PEXG, with corresponding cut-offs of > 375.1 pg/mL, > 378.5 pg/mL, and > 353.1 pg/mL. (G-I) VIM showed AUCs of 0.94 (95% CI: 0.88–0.99), 0.89 (95% CI: 0.79–0.99), and 0.99 (95% CI: 0.97–1.00) for Control vs. PEX (PEXS + PEXG), PEXS, and PEXG, with optimal cut-offs of > 484.60 pg/mL, > 467.30 pg/mL, and > 486.30 pg/mL, respectively

Correlation patterns of homocysteine, ROCK2, and vimentin

To assess potential relationships between the expression levels of Hcy, ROCK2, and VIM across the study population, a global correlation analysis was performed by combining data from the control, PEXS, and PEXG groups. The results revealed a strong positive correlation between Hcy and VIM (r = 0.76, p < 0.001), indicating a significant association between elevated Hcy levels and increased VIM expression. Similarly, Hcy levels showed a strong correlation with ROCK2 (r = 0.71, p < 0.001), while the strongest correlation was observed between ROCK2 and VIM (r = 0.79, p < 0.001) (Fig. 4A). These findings suggest that the expression levels of the metabolite and the two proteins are tightly interrelated across the clinical spectrum, potentially reflecting coordinated regulation or involvement in common biological pathways relevant to pseudoexfoliation pathogenesis.

Fig. 4.

Correlation analyses between homocysteine, ROCK2, and Vimentin. (A) Global correlation matrix for all the study subjects shows strong positive correlation between expression of Hcy and ROCK2 (ρ = 0.62), ROCK2 and VIM (ρ = 0.69), VIM and Hcy (ρ = 0.79). (B) Stratified correlation analysis showing preservation of associations within individual groups with varying strength. (C) Dot plot displaying Spearman correlation coefficients (ρ) for each biomarker pair: ROCK2 and VIM, ROCK2 and Hcy, and VIM and Hcy across the three study groups. Strong positive correlations were observed among all pairs in PEXS and PEXG, while correlations in the control group were weak or absent. Notably, the correlation between ROCK2 and VIM increases in PEXS but declines in PEXG. In contrast, the strength of association between Hcy and both ROCK2 and VIM increases progressively from PEXS to PEXG. The dashed line at ρ = 0 indicates no correlation

Further, stratified correlation analysis revealed that these associations were preserved within individual groups, though with varying strengths. In the control group, weak correlations were observed, whereas in the PEXS group, stronger correlations emerged, particularly between Hcy and VIM (r = 0.72) and between ROCK2 and VIM (r = 0.75). Notably, in the PEXG group, the correlations remained strong or even intensified (e.g., ROCK2 vs. VIM: r = 0.82), indicating a potentially heightened co-regulation of these biomarkers in more advanced disease states (Fig. 4B). The strength of association between Hcy and both ROCK2 and VIM increases progressively from PEXS to PEXG (Fig. 4C). These results suggest that the interrelationship among Hcy, ROCK2, and VIM may become more pronounced with disease progression, reflecting shared molecular mechanisms contributing to PEX pathology.

Validation of the interrelationship between homocysteine, ROCK2, and vimentin in human lens epithelial cells (HLEB-3)

To determine whether these observed correlations reflect causal regulation, we assessed whether Hcy acts upstream by directly modulating the expression of ROCK2 and VIM using HLE-B3 as an in vitro model. Given the strong positive correlations observed between plasma Hcy, ROCK2, and VIM in patients with PEXS and PEXG, we hypothesized that Hcy may act as an upstream regulator influencing their expression. To test effect of chronic Hcy levels, HLE-B3 cells were treated with increasing concentrations of Hcy (500 µM and 1000 µM) for 48 h, and protein levels were analyzed (Fig. 5A and B). Exposure to Hcy resulted in a concentration-dependent increase in the expression of both ROCK2 and VIM compared to vehicle controls. Densitometric quantification normalized to GAPDH confirmed the dose-dependent induction of both proteins (Fig. 5C and D). ROCK2 expression increased by approximately 2.40-fold at 500 µM Hcy (p = 0.26) and reached ~ 3.12-fold at 1000 µM, which was statistically significant (p = 0.03). Similarly, VIM levels rose by ~ 1.70-fold at 500 µM Hcy (p = 0.24) and were further upregulated to ~ 3.26-fold at 1000 µM with strong statistical significance (p = 0.003). Data represent the mean ± SD from five independent biological replicates (n = 5).

Fig. 5.

Effect of elevated homocysteine on ROCK2 and VIM in HLE-B3 cells. (A-B) Representative immunoblots showing ROCK2 and VIM protein levels in HLE-B3 cells treated with 0 (Vehicle Control), 500 µM, or 1000 µM Hcy for 48 h. GAPDH was used as a loading control. (C–D) Quantitative densitometric analysis of ROCK2 (C) and VIM (D) normalized to GAPDH. Data are presented as mean ± SD from five independent biological replicates (n = 5). Treatment with Hcy led to a concentration-dependent increase in expression of both markers. Statistical analysis was performed using the Mann–Whitney U test. Significance is indicated as follows: p < 0.05 (*), and p < 0.01 (**)

These findings provide mechanistic support for the clinical correlation data, demonstrating that elevated Hcy can upregulate VIM and ROCK2 expression in lens epithelial cells. This suggests a potential causal link and highlights Hcy as a molecular mediator capable of influencing cytoskeletal remodelling and signaling pathways relevant to PEX pathogenesis.

Diagnostic performance of biomarker combinations

Following the establishment of homocysteine’s mechanistic role in regulating ROCK2 and VIM, their combined diagnostic performance was evaluated using multivariate logistic regression and ROC curve analysis. The performance of all two- and three-biomarker combinations was assessed based on AUC, sensitivity, specificity, and Youden’s Index. Among the two-biomarker models, the combination of Hcy and VIM demonstrated the highest discriminative performance, achieving an AUC of 0.98, a sensitivity of 0.92, and a cut-off threshold of 0.80 (Fig. 6A-D; Table 1). This was closely followed by the Hcy and ROCK2 model (AUC = 0.97, sensitivity = 0.94, cut-off = 0.48) and the ROCK2 and VIM pair (AUC = 0.94, sensitivity = 0.92, cut-off = 0.47). Notably, while these models showed strong performance, the lower thresholds observed in the ROCK2–VIM and Hcy–ROCK2 combinations were associated with reduced specificity, suggesting a potential trade-off between sensitivity and specificity in these cases.

Fig. 6.

Diagnostic performance of two-marker and three-marker biomarker models. Receiver operating characteristic (ROC) curves comparing the classification performance of different biomarker combinations for diagnostic accuracy. (A) Two-marker model with Hcy and VIM achieved an AUC of 0.98 (95% CI: 0.95–1.00), a cut-off of 0.80, sensitivity of 0.91, and specificity of 0.83. (B) Two-marker model with Hcy and ROCK2 showed an AUC of 0.97 (95% CI: 0.94–1.00), with a cut-off of 0.48, sensitivity of 0.94, and specificity of 0.83. (C) Two-marker model with VIM and ROCK2 resulted in an AUC of 0.94 (95% CI: 0.88–0.99), a cut-off of 0.47, sensitivity of 0.88, and specificity of 0.80. (D) Three-marker model combining Hcy, ROCK2, and VIM yielded comparable diagnostic performance, with an AUC of 0.98 (95% CI: 0.96–1.00), but with a balanced cut-off score of 0.77, sensitivity of 0.94, and specificity of 0.88

Table 1.

Predictive measures of combined biomarker models

Biomarker Pairs	AUC	Sensitivity	Specificity	Optimal Cut-Off score	Accuracy	Youlden’s Index
Hcy and VIM	0.98	0.92	0.83	0.80	0.89	0.89
Hcy and ROCK2	0.97	0.94	0.83	0.48	0.90	0.90
ROCK2 and VIM	0.94	0.92	0.79	0.47	0.88	0.88
Hcy, ROCK2 and VIM	0.98	0.94	0.88	0.77	0.92	0.92

* Note: The cut-off value is set to 0.5

The three-biomarker combination of Hcy, ROCK2, and VIM demonstrated the highest overall predictive performance among all tested models. This model yielded an AUC of 0.98, indicating excellent discriminative ability. It achieved a sensitivity of 0.94, specificity of 0.88, and Youden’s Index of 0.92, surpassing the performance of all two-marker combinations (Fig. 6A-D; Table 1). The optimal decision threshold for the three-marker model was 0.77, slightly lower than that of the top two-marker model (Hcy–VIM, threshold = 0.80), but associated with higher overall accuracy.

Although the three-marker model demonstrated comparable AUC as Hcy-VIM, the overlap in confidence intervals with those of the two-marker models suggested that the improvement in performance may not be statistically significant based on AUC alone. To formally assess this, we applied DeLong’s test for pairwise comparison of ROC curves (Supplementary Table 2). The combined model exhibited a significantly higher AUC compared to the single ROCK2 model (Z = 4.0, p < 0.0001) and showed a trend toward significance relative to the VIM model (Z = 1.8, p = 0.071). However, no significant difference was observed when compared to the Hcy model (Z = 0.63, p = 0.53), indicating that Hcy may serve as the most robust individual predictor among the three biomarkers.

To estimate the probability of PEX based on biomarker profiles, a multivariate logistic regression model was constructed incorporating plasma levels of Hcy, ROCK2, and VIM. The outcome variable was disease status (1 = PEXS/PEXG, 0 = control). The final fitted equation was based on Logit(p) = β₀ + β₁·(Hcy) + β₂·(ROCK2) + β₃·(VIM) Where p represents the predicted probability of disease, each β coefficient reflects the log-odds change in disease probability associated with a one-unit increase in the respective biomarker, controlling for the others. For our model, the fitted equation is:

All three biomarkers were statistically significant contributors to the model (p < 0.05).

Multicollinearity among the predictor variables was assessed using the Variance Inflation Factor (VIF). All three biomarkers exhibited VIF values well below the accepted threshold of 5 (Hcy: 1.16, ROCK2: 1.40, VIM: 1.25), indicating no significant multicollinearity and supporting the stability of the regression model.

To evaluate the diagnostic potential of the three candidate biomarkers, we constructed individual logistic regression models and a penalized regression model using LASSO (Fig. 7A-D). Overall, the AUC values obtained from LASSO models were marginally lower but highly consistent with the raw models across all group comparisons. Cross-validation and bootstrap resampling were employed to estimate model performance and control for overfitting. ROC analysis demonstrated strong discrimination between study groups, particularly in the control vs. PEXG and control vs. PEX comparisons.

Fig. 7.

Receiver Operating Characteristic (ROC) curves for individual and combined biomarker models distinguishing study groups. (A) Control vs. PEX (PEXS + PEXG), (B) Control vs. PEXS, (C) Control vs. PEXG, (D) PEXS vs. PEXG. Each panel shows ROC curves for models based on individual biomarkers: Homocysteine (blue), ROCK2 (orange), Vimentin (green), and a combined LASSO logistic regression model (red). LASSO models were trained using repeated 5-fold cross-validation, and optimism-corrected AUCs were estimated using 1000 bootstrap replicates. AUC values are displayed within each panel

Risk prediction and stratification based on the multi-marker logistic regression model

To assess the discriminatory predictive performance of the three-marker logistic regression model, individual risk scores were calculated for all 72 subjects. Based on predefined probability thresholds, participants were stratified into three risk categories: low risk (P < 0.33), moderate risk (0.33 ≤ P < 0.66), and high risk (P ≥ 0.66). Risk score density plots revealed clear separation across groups. Controls showed a left-skewed distribution concentrated in the low-risk range, while PEXS and PEXG groups shifted toward higher predicted probabilities. Most PEXG cases clustered near a score of 1.0, indicating strong model confidence, whereas as expected PEXS showed a broader distribution with more moderate-risk cases. (Fig. 8A-D). Among all participants, 22 (30.6%) were categorized as low risk, 7 (9.7%) as moderate risk, and 43 (59.7%) as high risk. In the control group (n = 24), the majority (87.5%) were classified as low risk, with only one individual (4.2%) assigned to the high-risk group. Conversely, among PEXS cases (n = 24), 87.5% were in the high-risk category, with only one subject (4.2%) falling into the low-risk range. A similar trend was observed in the PEXG group (n = 24), where 87.5% were classified as high risk and none as low risk. When PEXS and PEXG were combined into a single disease group (PEX, n = 48), 87.5% of patients were in the high-risk category, underscoring the robustness of the model in distinguishing affected individuals in this cohort (Table 2).

Fig. 8.

Risk score distribution across stratified disease groups. Density plots showing the distribution of predicted risk scores for different binary logistic regression models. (A) Control vs. PEX (PEXS + PEXG): Controls (blue) show low predicted risk, while PEX (PEXS + PEXG) cases (salmon) are skewed toward high-risk probabilities. (B) Control vs. PEXS: Controls (blue) are concentrated at low risk, whereas PEXS cases (green) show high predicted probabilities. (C) Same comparison as B but with adjusted density estimation, showing sharp peaks due to extreme predicted scores. (D) PEXS vs. PEXG: Moderate overlap is observed between PEXS (green) and PEXG (lavender), indicating intermediate separation of risk profiles

Table 2.

Distribution of subjects across predicted risk categories based on logistic regression model

Risk Group	Control (n = 24)	PEXS (n = 24)	PEXG (n = 24)	PEX (PEXS + PEXG)(n = 48)
Low Risk	21 (87.5%)	1 (4.2%)	0 (0.0%)	1 (2.1%)
Moderate Risk	2 (8.3%)	2 (8.3%)	3 (12.5%)	5 (10.4%)
High Risk	1 (4.2%)	21 (87.5%)	21 (87.5%)	42 (87.5%)

Chi-square analysis revealed a statistically significant association between risk group classification and disease status in all pairwise comparisons. A strong association was observed between Control vs. PEX (χ² = 52.70, df = 2, p = 3.61 × 10⁻¹²), Control vs. PEXS (χ² = 37.34, df = 2, p = 7.82 × 10⁻⁹), and Control vs. PEXG (χ² = 34.38, df = 2, p = 3.42 × 10⁻⁸), confirming the model’s ability to effectively separate affected individuals from controls. Additionally, a significant difference in risk category distribution between PEXS and PEXG was observed (χ² = 7.34, df = 2, p = 0.03), suggesting that the model may also capture underlying differences between early and advanced disease stages. These findings further support the discriminatory power and clinical utility of the multi-marker risk model for disease stratification. Chi-square analysis showed a significant association between predicted risk categories and disease groups. Compared to controls, both PEXS (χ² = 37.33, p < 0.0001) and PEXG (χ² = 34.38, p < 0.0001) had significantly higher proportions in the high-risk group. The combined PEX patient group also showed a strong association with risk category (χ² = 52.70, p < 0.0001). Notably, a significant difference was observed between PEXS and PEXG (χ² = 7.33, p = 0.03), indicating distinct risk distributions across disease stages. The mean predicted probability was significantly higher in cases than in controls (0.75 ± 0.12 vs. 0.22 ± 0.14, respectively; p < 0.001), further supporting the model’s ability to distinguish affected individuals. These findings underscore the model’s ability to stratify risk effectively.

Model calibration was assessed using the Hosmer–Lemeshow goodness-of-fit test, which revealed no significant difference between observed and predicted probabilities (χ² = 1.72, df = 8, p = 0.99), indicating excellent fit of the logistic regression model to the observed data (Fig. 9A).

Fig. 9.

Calibration and clinical utility of the combined biomarker prediction model. (A) Calibration plot for the logistic regression model combining Hcy, ROCK2, and VIM. The x-axis shows predicted disease probability, and the y-axis indicates observed frequency. The navy-blue curve represents the bias-corrected calibration (1000 bootstrap resamples); the diagonal grey dashed line indicates perfect prediction. The model demonstrated good calibration, with a mean absolute error of 0.04 and a mean squared error of 0.003. (B) Decision Curve Analysis (DCA) assessing the net clinical benefit of the combined biomarker model. The standardized net benefit is plotted against a range of threshold probabilities. The blue line denotes the biomarker model; the red dashed and gray lines represent the “Treat All” and “Treat None” strategies, respectively. The model shows superior net benefit across the clinically relevant threshold range (0.1–0.8), supporting its potential for individualized, risk-based decision-making

Further, calibration analysis using 1000 bootstrap resamples showed a good agreement between the predicted and observed probabilities. The mean absolute error was 0.04, with a mean squared error of 0.003, and the 90th percentile of absolute errors was 0.08. These results support the high reliability and predictive consistency of the biomarker panel-based logistic regression model.

To assess the clinical utility of the biomarker-based prediction model incorporating Hcy, ROCK2, and VIM, we performed a decision curve analysis. The DCA plot (Fig. 9B) compares the standardized net benefit of the biomarker model across a range of threshold probabilities (0–1) against the default strategies of “treat all” and “treat none.” The biomarker model demonstrated consistently higher net benefit across a clinically relevant threshold range of approximately 0.10 to 0.80, indicating improved decision-making utility over both extremes. This suggests that using the composite risk score derived from Hcy, ROCK2, and VIM may aid in identifying high-risk individuals more effectively than indiscriminate or no treatment approaches. The maximum net benefit was observed in the lower to mid-threshold range, which may reflect the model’s strength in early risk discrimination.

To explore whether the expression profiles of the biomarkers HCY, ROCK2, and VIM could stratify individuals into biologically meaningful groups, we performed k-means clustering on z-score–normalized expression data. Based on convergence across three independent validation methods (the elbow plot, silhouette width, and gap statistic), we determined that k = 3 clusters best captured the structure of the dataset. The Adjusted Rand Index (ARI) comparing unsupervised clusters to clinical labels (Control, PEXS, PEXG) was 0.32, indicating moderate alignment between molecular profiles and disease classification.

Principal component analysis (PCA) revealed that the first two components explained 91.5% of the total variance (PC1: 80.4%, PC2: 11.1%). PC1 reflected overall biomarker burden, while PC2 captured more subtle, marker-specific variation. The PCA plot with 95% confidence ellipses (Fig. 10A) showed clear separation of three k-means clusters, indicating distinct molecular phenotypes. Cluster 2, positioned at the high end of PC1, was enriched for PEXG cases and characterized by high biomarker expression. Cluster 3, located at the low end of PC1, consisted mostly of control subjects with low expression profiles. Cluster 1 occupied the intermediate space with mixed diagnoses and expression levels.

Fig. 10.

k-Means (k = 3) clustering of subjects based on biomarker profiles and mean biomarker expression based on z-score. (A) PCA plot showing three clusters derived from k-means clustering of standardized Hcy, ROCK2, and VIM values. Clusters show differential enrichment of disease states, with Cluster 3 predominantly composed of controls, Cluster 1 enriched in PEXS, and Cluster 2 enriched in PEXG. Subjects are coloured by disease status and shaped by cluster label. The ellipses represent the 95% confidence boundaries for each cluster. (B) Mean biomarker expression (z-score ± SEM) across three unsupervised clusters. Cluster 2 shows coordinated upregulation of Hcy, ROCK2, and VIM, consistent with advanced disease (PEXG). Cluster 3 exhibits downregulated expression typical of controls. Cluster 1 shows mild elevations, suggesting early disease activity (PEXS)

To further characterize the clusters, we examined mean biomarker z-scores (Fig. 10B). Cluster 2 showed coordinated upregulation of HCY, ROCK2, and VIM (z-scores > + 1.5), indicating a high-activity molecular profile consistent with advanced disease. Cluster 3 showed uniformly reduced expression (z-scores < 0), aligning with a biologically stable control group. Cluster 1 showed modest biomarker increases, potentially representing early or subclinical PEXS.

These findings support the concept that molecular clustering based on biomarker expression aligns with clinical severity, and more importantly, reveals biological transitions not captured by standard diagnostics. This data-driven stratification may facilitate early detection of high-risk individuals and offer a molecular framework for personalized disease staging in PEXS.

Systems-level bioinformatic analysis reveals converging pathways in PEX pathogenesis

To understand functional significance of the candidate biomarkers, systems-level bioinformatic analysis was performed, integrating pathway enrichment, network modelling, and prediction of transcriptional regulation, aimed at uncovering upstream regulators and downstream biological processes. Joint pathway enrichment analysis via MetaboAnalyst 6.0 revealed several significantly enriched pathways (p < 0.05) with high impact scores, collectively implicating key biological processes such as oxidative stress, vascular remodelling, inflammation, apoptosis, and cytoskeletal regulation. Specifically, enrichment of NOD-like receptor and chemokine signaling pointed to chronic inflammation. Focal adhesion, vascular smooth muscle contraction, and Wnt signaling implicated tissue remodelling and endothelial dysfunction; while cysteine/methionine metabolism and folate pathways reflected Hcy-driven redox imbalance. Sphingolipid and cAMP/cGMP signaling pathways indicated stress adaptation and apoptosis. These pathways were visualized in a bubble plot (Fig. 11A), mapping statistical significance against pathway impact, and further grouped into five major functional modules via unsupervised clustering: [1] methylation/redox metabolism [2], cytoskeletal dynamics and adhesion [3], fibrosis and immune signaling [4], vascular regulation, and [5] lipid signaling (Fig. 11B). The dominant enrichment of redox-related pathways supports a mechanistic link between elevated Hcy and oxidative stress-mediated cytoskeletal and fibrotic changes.

Fig. 11.

Integrated pathway enrichment analysis of homocysteine, VIM, and ROCK2. (A) Bubble plot showing significantly enriched pathways based on –log₁₀(p-value) and pathway impact. Bubble size reflects the number of matched features; colour indicates functional categories. Key pathways highlight redox imbalance, cytoskeletal disruption, and vascular dysfunction. (B) Clustered bar plot grouping enriched pathways (p < 0.05) into functional categories. The most enriched categories were methylation/redox metabolism, cytoskeletal remodelling, and immune signaling, suggesting a central role for metabolic and mechanical stress in PEX pathogenesis

To contextualize these findings within molecular interaction networks, a STITCH-based analysis was performed using Hcy and 42 prioritized genes (Supplementary Table 3). This input set included key candidates such as VIM and ROCK2, functional interactors from STRING, literature-curated PEX-associated genes, and enzymes involved in Hcy metabolism, revealing distinct modules: a cytoskeletal/adhesion cluster centred on ROCK2, VIM, and RHOA; a redox/epigenetic module involving MTHFR, CBS, AHCY, and DNMTs; a fibrosis-related cluster connecting TGFB1, ZEB1/2, and CTNNB1; and a stress-response module comprising CASP3, HSPA1A, and UBB (Fig. 12). These modules reinforce the convergence of metabolic dysregulation, epigenetic disturbance, and mechanical stress in PEX.

Fig. 12.

STITCH network of PEX-associated genes and homocysteine metabolism. The network includes VIM, ROCK2, their interactors, and homocysteine-metabolizing enzymes. Nodes represent proteins/metabolites; edges indicate functional associations. Distinct modules include cytoskeletal/adhesion (VIM, ROCK1/2, RHOA), methylation/redox (MTHFR, CBS, DNMTs), TGF-β/WNT signaling (TGFB1, ZEB1/2, CTNNB1), and stress response (CASP3, HSPA1A, UBB), illustrating converging pathways in PEX pathogenesis

To further define upstream regulatory mechanisms, transcription factor (TF) analysis using TRRUST and ChEA3 identified SP1 as a top-scoring TF, predicted to regulate multiple PEX-associated genes, including VIM, ROCK2, TGFB1, and DNMT1. Additional regulators such as TP53 and SALL4 were also predicted, indicating potential regulatory redundancy (Fig. 13A). In silico analysis using the UCSC Genome Browser and JASPAR (motif MA0079.3) revealed multiple SP1 binding sites within the proximal promoter regions (− 1500 bp upstream of the TSS) of both VIM and ROCK2. All predicted sites were located upstream of the TSS, with several showing high relative scores (> 0.9), suggesting strong binding potential. These findings indicate that SP1 may transcriptionally regulate both genes through promoter interactions (Figure 13B and C, Supplementary Table 4).

Fig. 13.

SP1 as a key transcriptional regulator in Pseudoexfoliation. (A) Transcription factor enrichment using TRRUST and ChEA3 identified SP1 as a top-scoring regulator of PEX-associated genes (B–C). In silico prediction of putative SP1 binding sites performed within the proximal promoter regions (− 1500 bp upstream of the TSS) of VIM and ROCK2 using the UCSC Genome Browser and JASPAR (motif MA0079.3)

Discussion

In this study, we evaluated the diagnostic and mechanistic potential of three plasma biomarkers: Hcy, ROCK2, and VIM in an Indian cohort. All three markers were significantly elevated in PEXS and PEXG compared to controls, with Hcy and VIM levels progressively increasing from PEXS to PEXG, suggesting their association with disease severity. Conversely, ROCK2 expression peaked in PEXS and declined in PEXG, potentially reflecting the onset of active remodelling in early disease that diminishes or becomes dysregulated in later stages. Importantly, Hcy has been independently shown to be elevated in both plasma and aqueous humor in other PEX cohorts, reinforcing its systemic and ocular relevance. Moreover, our previous work reported increased ROCK2 and VIM expression in lens capsule tissue of PEX patients, further supporting systemic dysregulation [28, 29, 40–43].

These biomarker dynamics align with their known biological roles: Hcy as a mediator of redox stress and endothelial dysfunction; ROCK2 as a key effector in fibrotic and ECM pathways; and VIM as a marker of cytoskeletal remodelling and cellular stress [31, 33, 44].

From a diagnostic perspective, Hcy emerged as the strongest individual classifier, while VIM showed high specificity for PEXG. Although ROCK2’s standalone performance was moderate, its inclusion in multi-marker models enhanced diagnostic resolution. The two-marker model combining Hcy and VIM yielded the highest AUC (0.98), while the three-marker model including Hcy, ROCK2, and VIM achieved comparable AUC (0.98), but demonstrated better balance in sensitivity (0.94) and specificity (0.88), along with the highest Youden’s index. This highlights the value of combining metabolically and structurally distinct markers to achieve greater diagnostic accuracy. Unsupervised clustering based on biomarker expression patterns identified three molecular subgroups. One cluster, enriched in PEXG cases, showed co-elevation of all markers and likely represents a high-risk phenotype. A second cluster was dominated by controls, and a third cluster showed intermediate biomarker elevations that may reflect early-stage, borderline, or subclinical disease. The ARI of 0.32 indicates moderate concordance between unsupervised biomarker-based clustering and clinical group labels, suggesting that while biomarker expression partially aligns with clinical staging, emphasizing that more biomarker profiles with clinical assessment could enhance early detection and risk classification in PEX.

Correlation analysis revealed that marker associations, weak in controls, became strongly positive in PEXS and PEXG, particularly between Hcy-VIM and Hcy-ROCK2, supporting a model where Hcy drives downstream cytoskeletal and fibrotic signaling. The weakening of the ROCK2-VIM correlation in PEXG may reflect cytoskeletal disruption in advanced disease. Additionally, non-normal distributions of Hcy and ROCK2 in PEXS suggest early-stage heterogeneity, while VIM’s skewed distribution in PEXG points to increased cytoskeletal instability in PEX.

Given that Hcy emerged as a consistently elevated biomarker in our cohort, we focused our mechanistic investigation on its potential role as a key upstream disruptor in PEX. Elevated Hcy is a well-documented driver of oxidative stress, epigenetic remodelling, and fibrotic signaling in several age-related and ocular pathologies. In PEX, increased Hcy levels correlate with oxidative damage markers such as lipid peroxidation, protein carbonylation, and 8-OHdG.

Our systems-level analyses reinforced this link. Pathway enrichment showed that differentially expressed genes in PEX are involved in redox/methylation balance, cytoskeletal organization, ECM remodelling, and immune signaling. Importantly, transcriptional network modelling identified SP1 as a shared upstream regulator of VIM, ROCK2, and TGFB1-suggesting a redox-sensitive transcriptional mechanism connecting elevated Hcy to fibrotic and cytoskeletal gene expression.

One major downstream consequence of elevated Hcy is epigenetic dysregulation, particularly involving DNA methyltransferases (DNMTs). In PEX, DNMT1 is downregulated while DNMT3A is upregulated, corresponding to gene-specific methylation changes. Promoter hypomethylation of CLU, likely due to reduced DNMT1 levels, facilitates enhanced SP1 binding and increased CLU transcription. In contrast, exonic hypermethylation of HSP70 correlates with elevated DNMT3A expression, indicating selective methylation remodelling of stress-responsive genes [26, 45].

This imbalance in DNMT expression may result from Hcy-induced depletion of S-adenosylmethionine (SAM), the essential methyl donor for DNMT activity, thereby altering the global and locus-specific methylation landscape [46].

Critically, Hcy may exert its epigenetic influence via modulation of SP1, a redox-sensitive transcription factor. Prior studies have shown that Hcy can downregulate DNMT1 through SP1-dependent mechanisms, and that SP1 activity is highly sensitive to oxidative stress. The promoters of key PEX-associated genes, including DNMT1, VIM, and ROCK2, contain multiple GC-rich SP1 binding motifs [47–50]. It is possible that in the redox-imbalanced environment of PEX, Hcy may impair SP1 function, leading to reduced DNMT1 expression and altered epigenetic regulation. Moreover, SP1 may act as a dual regulator—either shielding promoters from DNA methylation by occupying GC-rich regions, or alternatively recruiting DNMT1 to specific sites depending on cellular context. These context-dependent roles position SP1 as a critical node linking oxidative stress to epigenetic dysregulation in PEX,

In parallel, Hcy has been shown to stimulate TGF-β1 expression via SP1/NF-κB pathways. TGF-β is a central mediator of fibrosis and EMT, processes that are also prominent in PEX pathology [51]. Notably, TGF-β signaling can itself regulate DNMT1 and DNMT3A expression, establishing a feedback loop between metabolic, inflammatory, and epigenetic signals [52, 53]. These interactions may underlie the observed upregulation of fibrotic genes such as VIM and ROCK2 in PEX, both of which harbour SP1 binding motifs in their proximal promoters. While direct evidence of SP1 expression in PEX is currently lacking, our in-silico predictions suggest its potential role as a master regulator that links metabolic stress to cytoskeletal remodelling and ECM deposition.

Based on our findings, we propose a model (Fig. 14) wherein elevated Hcy levels initiate a cascade of oxidative and epigenetic disruptions, mediated through SP1 dysfunction, leading to selective gene expression changes that promote fibrotic remodelling (e.g., increased VIM and ROCK2) in PEX. This hypothesis, while requiring experimental validation, provides the first-of-its-kind unifying framework connecting Hcy metabolism, redox imbalance, transcriptional regulation, and epigenetic mis-programming in the pathogenesis of PEX.

Fig. 14.

Proposed pathophysiological model linking systemic homocysteine elevation to ocular remodelling in pseudoexfoliation disease

With promising diagnostic and mechanistic associations established, the next challenge lies in ensuring these biomarkers can be reliably translated into clinical use. While all three biomarkers are quantifiable by standard ELISA, several factors must be considered for clinical translation. The level of biomarkers may be sensitive to sample handling, processing delays, and storage conditions. However, we attempted to minimize these variations as much as possible by standardizing sample collection procedures, fixing the time of sample collection, extracting plasma immediately after collection, preventing hemolysis, processing all samples within a consistent time frame, and storing them under controlled temperature conditions until analysis. Therefore, pre-analytical and biological variability must be rigorously controlled in clinical settings.

Despite promising insights, several limitations merit consideration. The sample size was modest, and recruitment was geographically restricted to a single region in India. While effect sizes were large, population homogeneity may have inflated group separation and introduced a risk of overfitting. While we used internal validation through bootstrapping, the absence of an external validation cohort limits generalizability.

Although previous studies have proposed various candidate biomarkers for PEX, such as TGF-β1 and IL-8, most have shown limited and inconsistent clinical utility. TGF-β1, despite being elevated in several biological fluids, offers only modest diagnostic performance and lacks specificity for disease severity [18]. Additionally, many of its fibrotic and cytoskeletal effects occur downstream of the Rho/ROCK pathway and are functionally captured by ROCK2 and VIM. The immunoassays used to measure TGF-β1 often fail to distinguish between its latent and active forms, further limiting interpretability. As such, its inclusion in a diagnostic panel would likely be redundant and add little diagnostic value.

Our biomarker panel was deliberately constructed to represent three distinct yet converging aspects of PEX pathophysiology. Hcy captures systemic metabolic and oxidative stress, and may trigger endothelial dysfunction. ROCK2 reflects ECM remodelling and fibrotic signaling, operating downstream of multiple stress pathways. VIM, a structural intermediate filament, marks cytoskeletal remodelling and cellular stress. By focusing on functionally proximal, plasma-detectable markers, this panel enhances both diagnostic precision and translational potential. While Hcy and VIM are not specific to PEX and may be elevated in other systemic conditions, the inclusion of ROCK2 improves both diagnostic accuracy and disease specificity.

Although Hcy likely acts as an upstream regulator, its elevation may also reflect broader systemic dysfunction. Deficiencies in folate and vitamin B12, both common in aging PEX populations, can contribute to Hcy elevation [40, 54]. However, PEX-associated oxidative stress or ECM disruption might also impair Hcy metabolism, suggesting a bidirectional relationship. Thus, Hcy may serve as both a driver and a sensor of systemic and local dysregulation, positioned at the intersection of metabolic, epigenetic, and vascular pathways. Hcy is a well-established risk factor for cardiovascular diseases, and its elevation in PEX patients likely reflects a shared biochemical and pathogenic nexus between ocular and systemic vascular dysfunction. Several studies have documented an increased incidence of cardiovascular disorders in individuals with PEX, suggesting that PEX may function as an independent cardiovascular risk factor or emerge from a common systemic pathology [55]. Although these associations require validation in larger populations, they underscore the need for integrated ocular–systemic biomarker assessment combined with parallel cardiovascular evaluation in future studies.

With the groundwork laid for precision diagnosis, our future studies will involve larger, multi-centre, and ethnically diverse cohorts to evaluate the robustness and generalizability of the proposed biomarker panel. We also aim to expand the panel by incorporating additional markers and integrating multi-omics approaches. Longitudinal studies in at-risk individuals, such as those with a family history of PEX will help elucidate biomarker dynamics during preclinical stages and provide insights into early disease trajectories. Experimental modulation of SP1 and its downstream targets (ROCK2, VIM, TGF-β1) may help confirm causal relationships. Importantly, exploring whether targeting this biomarker axis delays progression from PEXS to PEXG may offer translational value.

Conclusion

Our findings support the existence of a coordinated biomarker axis linking systemic metabolic stress, fibrotic remodelling, and cytoskeletal alterations across the spectrum of pseudoexfoliation disease. This study identifies and mechanistically validates a three-marker plasma panel that reflects distinct yet converging pathophysiological processes in pseudoexfoliation syndrome and glaucoma. The integration of molecular profiling, statistical modelling, and systems biology supports a disease framework centred on homocysteine-driven cytoskeletal and fibrotic dysregulation, with translational potential for early diagnosis, risk stratification, and therapeutic targeting. With further validation, this approach may advance precision medicine in PEX-related disorders.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AH

Aqueous Humor

AUC

Area Under the Curve

CBS

Cystathionine β-Synthase

CLU

Clusterin

CNTNAP2

Contactin Associated Protein-Like 2

DNMTs

DNA Methyltransferases

ECM

Extracellular Matrix

ELISA

Enzyme-Linked Immunosorbent Assay

EMT

Epithelial-Mesenchymal Transition

FBLN5

Fibulin-5

GAPDH

Glyceraldehyde 3-phosphate dehydrogenase

Hcy

Homocysteine

HLE-B3

Human Lens Epithelial Cell Line B3

IL-8

Interleukin-8

IOP

Intraocular Pressure

LC

Lens Capsule

LOXL1

Lysyl Oxidase-Like 1

LASSO

Least Absolute Shrinkage and Selection Operator

MMPs

Matrix Metalloproteinases

PCA

Principal Component Analysis

PEX

Pseudoexfoliation (PEXS + PEXG)

PEXS

Pseudoexfoliation Syndrome

PEXG

Pseudoexfoliation Glaucoma

RHOA

Ras Homolog Family Member A

ROC

Receiver Operating Characteristic

ROCK2

Rho-associated Coiled-coil Containing Protein Kinase 2

SAM

S-Adenosyl Methionine

SP1

Specificity Protein 1

TGF-β1

Transforming Growth Factor Beta 1

TIMPs

Tissue Inhibitors of Metalloproteinases

TRRUST

Transcriptional Regulatory Relationships Unraveled by Sentence-based Text mining

VIM

Vimentin

Authors’ contributions

LS performed the experiments, analyzed the data, and prepared the manuscript; PPM supervised data collection. DPA conceptualized and supervised the study, acquired the funding, and reviewed and edited the manuscript.

Funding

Open access funding provided by Department of Atomic Energy. This work was supported by the extramural research grants [grant number CRG/2019/002705 and SPG/2022/000325] from the Science and Engineering Research Board, India, and an intramural grant [grant number RIN-4002-SBS] from the National Institute of Science Education and Research (NISER), an autonomous organization under Department of Atomic Energy, Government of India.

Data availability

Data related to this paper may be requested from the corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Biosafety and Human Ethics Committee of the National Institute of Science Education and Research and adhered to the tenets of the Declaration of Helsinki. Written informed consent was obtained from all the participants in the study.

Consent for publication

Not applicable.

Competing interests

The authors have declared no conflicts of interest.