Serum cystatin C as a biomarker for diabetic retinopathy and its role in diabetic retinopathy-diabetic kidney disease comorbidity

Abstract

Background

To investigate whether serum cystatin C (CysC) can serve as a biomarker for diabetic retinopathy (DR), and explore the comorbidity between DR and diabetic kidney disease (DKD).

Methods

This study used a nested case-control design with Mendelian randomization (MR) analysis. A total of 128 type 2 diabetic patients were categorized into no DR (DM), non-proliferative DR (NPDR), and proliferative DR (PDR) groups with age- and sex- matched controls. Biochemical and renal biomarkers were assessed, including the absolute difference (abdiff) eGFR_abdiff (eGFR_CysC-eGFR_Scr) and relative difference (rediff) eGFR_rediff (eGFR_CysC/eGFR_Scr). Instrumental variables for renal function and DR subtypes were selected from the UK Biobank and FinnGen consortium. Genetic associations were primarily derived using the inverse-variance weighted approach, with sensitivity analyses and multivariable MR adjusting for confounders.

Results

Multivariable logistic analysis demonstrated that CysC levels were significantly associated with DM patients (OR=13.22, P=0.001) compared to healthy controls, and with NPDR patients (P=0.016) compared to DM patients, after adjusting for confounding variables; while eGFR_CysC was a protective factor for DM (P_{DM vs. Normal}=0.002); NPDR (P_{NPDR vs. DM}=0.025) and PDR (P_{PDR vs. NPDR}=0.013). Receiver operating characteristic (ROC) analysis confirmed the superior diagnostic efficacy of CysC and eGFR_CysC for both DM and NPDR, eGFR_rediff and urine Albumin-to-Creatinine Ratio (UACR) had superior predictive value for PDR compared to other renal biomarkers. Each standard deviation increases in serum CysC levels elevated the risks of DM (P=0.011), NPDR (P=0.00067), and PDR (P=0.042) by MR analysis.

Conclusions

The renal dysfunction biomarkers CysC and eGFR_CysC were identified as key biomarkers for NPDR, while eGFR_rediff and UACR were predictive for PDR. MR analyses confirmed CysC as a strong risk factor for DR susceptibility, particularly in the early stages.

Supplementary information

The online version contains supplementary material available at 10.1186/s12967-025-07552-6.

Introduction

Diabetic retinopathy (DR) and diabetic kidney disease (DKD) are among the most debilitating microvascular complications of diabetes mellitus (DM). DR and DKD share significant similarities in anatomy, developmental origin, genetics and pathology, suggesting common underlying disease mechanisms [1]. Clinical studies have also shown a strong correlation between DR severity and DKD progression [2, 3], indicating a potential interdependence between the two conditions. Given these shared characteristics, renal metabolic biomarkers hold promise as non-invasive tools for the early detection and diagnosis of DR.

Serum cystatin C (CysC), a low-molecular-weight cysteine protease inhibitor, has been closely linked to the early stage of DKD [4], demonstrating high sensitivity and specificity for detecting early glomerular filtration impairment and predicting adverse outcomes in chronic kidney disease (CKD) [5, 6]. According to the KDIGO Clinical Practice Guidelines [7], the primary diagnostic parameters for CKD include estimated glomerular filtration rate (eGFR) and urine albumin-to-creatinine ratio (ACR). The 2023 American Diabetes Association (ADA) guidelines further highlight that incorporating CysC with serum creatinine (Scr) for eGFR calculation significantly enhances the accuracy of renal function assessment [8]. Recent studies [9–12] have also shown that the absolute difference in eGFR_abdiff (eGFR_CysC - eGFR_Scr) and relative difference eGFR_rediff (eGFR_CysC/eGFR_Scr) may mitigate confounding effects from non-renal factors, such as high muscle mass, high-protein diets, inflammation, and cardiovascular diseases, thus improving the evaluation of diabetic microvascular complications [9]. Clinical studies have consistently demonstrated associations between ACR, eGFR, cystatin C, and urine microalbumin (UMA) with DR [13–15]. However, the precise relationships between these biomarkers and different stages of DR remain poorly understood.

Advancements in genome-wide association study (GWAS) data have enhanced the application of Mendelian randomization (MR) analysis. MR leverages the random allocation of genotypes during meiosis, using genetic variants as instrumental variables (IVs) to infer causal associations between exposures and disease outcomes, thereby effectively controlling for biases, including unmeasured or unknown confounders, that often affect clinical studies [16].

The novelty of our study lies in the investigation of serum CysC as a biomarker for DR and its potential role in the comorbidity between DR and DKD. The study provides new insights by using a nested case-control design combined with Mendelian randomization analysis, which strengthens the causal inference. We are the first to identify CysC and eGFR_CysC as key biomarkers for DM and NPDR, with CysC being strong association for DR susceptibility, particularly in its early stages. The study also highlights the importance of eGFR_rediff and UACR in predicting PDR, offering valuable insights into early detection and progression of diabetic complications. This approach, incorporating genetic analysis and multiple biomarkers, adds a novel dimension to understanding the complex relationship between DR and DKD (Fig. 1).

Fig. 1.

Flowchart of the study methodology. This study used a nested case-control design (a) with Mendelian randomization (MR) analysis (b). ACR urinary albumin-to-creatinine ratio, CKD chronic kidney disease, CysC cystatin C, eGFR estimated glomerular filtration rate (CKD-EPI 2021 equations using Scr, CysC, or both), eGFR_abdiff absolute eGFR difference (eGFR_CysC - eGFR_Scr), eGFR_rediff relative eGFR difference (eGFR_CysC/eGFR_Scr), GWAS genome-wide association study, IVW inverse variance weighted, MR Mendelian randomization, NC normal control, NPDR non-proliferative diabetic retinopathy, PDR proliferative diabetic retinopathy, ROC receiver operating characteristic, Scr serum creatinine, SNP single nucleotide polymorphism, T2DM type 2 diabetes mellitus, Ucr urinary creatinine, UK United Kingdom, UMA urinary microalbumin

Subjects, materials and methods

Nested case-control study

Ethics statement

The study protocol was approved by the Ethics Committee of Beijing Tongren Hospital, Capital Medical University (TRECKY2020-111), adhering to the Declaration of Helsinki, with written informed consent obtained from all participants.

Inclusion and exclusion criteria

This nested case-control study enrolled 173 participants between 2022 and 2025 at Beijing Tongren Hospital. The diagnostic criterion of Type 2 diabetes mellitus (T2DM) was according to the 2022 American Diabetes Association (ADA) Standards of Medical Care in Diabetes [17]. Based on the 2019 American Academy of Ophthalmology (AAO) Diabetic Retinopathy Clinical Guidelines [18], Patients with DM were classified to: (1) Diabetes group (DM group, n = 37); (2) Non-proliferative diabetic retinopathy group (NPDR group, n = 45); (3) Proliferative diabetic retinopathy group (PDR group, n = 46) groups. An age- and sex-matched normal group (n = 45) without diabetes or retinopathy was also recruited.

The inclusion criteria for patients were as following: (1) Type 2 diabetes mellitus; (2) Non-proliferative diabetic retinopathy; (3) Proliferative diabetic retinopathy; (4) Subjects exhibited good compliance, enabling completion of necessary ophthalmic and biochemical evaluations.

The exclusion criteria included: (1) Type 1 diabetes mellitus; (2) Non-diabetic retinal diseases, including Retinal vascular diseases, infectious/inflammatory retinopathies, degenerative diseases; (3) Intraocular surgery within the past 6 months; (4) Acute, chronic or end-stage kidney diseases; (5) Anterior segment opacity; (6) Severe systemic diseases limit the tolerance for eye examinations, including but not limited to acute myocardial infarction, unstable angina, severe heart failure, uncontrolled hypertension, recent cerebrovascular accident, severe respiratory failure, active tuberculosis, and severe hepatic failure etc.; (7) During the establishment of the cohort, patients with other diabetes-related complications such as cerebrovascular or cardiovascular diseases and neuropathy.

Data collection

Baseline demographic and clinical data were recorded for all participants, including age, sex, duration of diabetes, duration of hypertension (HBP), and medication usage. Fasting venous blood (5 mL) was collected to measure metabolic biomarkers using an automated biochemistry analyzer (Beckman AU5811; inter-assay CV 3–5%) including (1) the Glycemic parameters: Glycated hemoglobin (HbA1c) and fasting blood glucose (FBG); (2) the renal biomarkers: Serum cystatin C (CysC) and serum creatinine (Scr). Besides, midstream urine samples (10 mL) were collected to measure urinary creatinine (Ucr), urine albumin-to-creatinine ratio (ACR), and urinary microalbumin (UMA).

Considering that various factors, especially medications can influence serum levels of CysC, we further divided the participants into four groups based on their medication use: 1) no medication 2) anti -hypertensive (HBP) drugs, 3) anti-diabetic drugs and 4) both anti-HBP and anti-Diabetic drugs usage. Group comparisons revealed significant differences between the four groups (Supplementary Table 8). Consequently, we accounted for medication history as a potential confounder in the subsequent statistical analysis.

To minimize observer bias, a rigorous blinding procedure was implemented during cohort establishment. Each participant was assigned a clinical code for identification, ensuring independent clinical, laboratory, and analytical assessments and reducing the risk of observer bias.

Renal indicators

According to the 2025 ADA Standards of Care in Diabetes [19], ACR is classified as: (1) A1: ACR < 30 mg/g; (2) A2: ACR 30 - 300 mg/g; (3) A3: ACR > 300 mg/g. eGFR was calculated by the CKD-EPI 2021 equations based on Scr, serum CysC, and their combination [20]. The renal dysfunction stages were defined as following [19]: (1) G1: ≥ 90 mL/min/1.73 m²; (2) G2: 60 - 89 mL/min/1.73 m²; (3) G3a: 45 - 59 mL/min/1.73 m²; (4) G3b: 30 - 44 mL/min/1.73 m²; (5) G4: 15 - 29 mL/min/1.73 m²; (5) G5: < 15 mL/min/1.73 m². To mitigate non-renal confounding effects on eGFR, the absolute eGFR difference was defined as eGFR_abdiff = eGFR_CysC - eGFR_Scr and the relative eGFR difference was defined as eGFR_rediff = eGFR_CysC/eGFR_Scr [9–12].

Eye examinations

All participants underwent comprehensive eye assessment: (1) Best-corrected visual acuity (BCVA); (2) Non-contact tonometry (TX20, Canon Co., Ltd., Tokyo, Japan); (3) Slit-lamp biomicroscope (SL-IE, Topcon Co., Ltd., Tokyo, Japan); (4) Non-mydriatic fundus imaging (CR-1, Canon Co., Ltd.); (5) Swept-source optical coherence tomography (SS-OCT) and angiography (SS-OCTA): Performed using DRI OCT1 Atlantis (Topcon Co., Ltd.) and PLEX® Elite 9000 (Carl Zeiss Meditec Inc.), with macular-centered scans.

Sample size calculation

Sample size estimation was performed using PASS software (Version 2024; NCSS LLC, Kaysville, UT, USA) with two-sided tests set at α = 0.05 and 90% power. Based on preliminary data showing serum CysC levels of 0.68 ± 0.08 mg/L, 0.85 ± 0.14 mg/L, 1.01 ± 0.37 mg/L, and 1.23 ± 0.49 mg/L in normal, DM, NPDR, and PDR groups respectively, a minimum of 22 subjects per group was required. Accounting for biomarker variability, the final recruitment target was established at ≥ 35 participants per group. determined that a minimum of 22 subjects per group was required. Our final recruitment of ≥ 35 participants per group was designed to minimize individual variations and other potential influencing factors.

Statistical analysis

SPSS 25.0 (SPSS Inc., Chicago, IL, USA) was used. Normality and homogeneity of variance were assessed via Shapiro-Wilk and Levene tests. Group comparisons used ANOVA, Kruskal-Wallis, or chi-square tests. Corrections for multiple comparisons (Bonferroni) were applied to reduce the risk of type I error (Supplementary Table 5). Associations between variables and DR were evaluated using multivariable binary logistic regression analysis. The Receiver Operating Characteristic (ROC) analysis was adjusted for covariates including DM duration, HBP duration, DM medication, HBP medication, FBG and HbA1c. Firth’s logistic regression was used for sensitivity analysis (Supplementary Table 9). A two-tailed p < 0.05 was considered significant.

Mendelian randomization analysis

Mendelian randomization assumptions

This study strictly adheres to the three core assumptions of Mendelian randomization: (1) the selected IVs are significantly associated with renal function; (2) IVs are independent of confounders; (3) IVs affect DR only through renal function. Furthermore, this study strictly followed the Strengthening the Reporting of Observational Studies in Epidemiology for Mendelian Randomization (STROBE-MR) guidelines.

Data sources

GWAS data for Scr and serum CysC were derived from the UK Biobank, which enrolled 500,000 participants at baseline (2006–2010) and re-evaluated 20,000 participants in 2012–2013. Serum CysC levels (n = 389,834; mean = 0.91 mg/L, SD = 0.176) and Scr levels (n = 389,678; mean = 72.4 μmol/L, SD = 18.5) were analyzed. eGFR was calculated using CKD-EPI equations [20] based on Scr (eGFR_Scr) and CysC (eGFR_CysC). Genetic data for eGFR were obtained from a meta-analysis (n = 1,201,909) combining the CKD Genetics Consortium and UK Biobank GWAS [21].

To ensure the validity of MR analysis, we conducted comprehensive quality control of single-nucleotide polymorphisms (SNPs). First, only SNPs significantly associated with renal function at the genome-wide level (p < 5 × 10^{− 8}) were retained. Subsequently, linkage disequilibrium (LD) analysis with stringent parameters (R² < 0.001, clumping distance = 10,000 kb) was performed to filter SNPs. Additionally, ambiguous palindromic SNPs with intermediate allele frequencies were excluded. For unavailable original SNPs, proxy SNPs with high LD correlation (r² > 0.9) were identified and substituted using the LDlink database (https://ldlink.nci.nih.gov/). To mitigate weak instrument bias and ensure strong correlations between IVs and exposures, the proportion of variance explained by IVs was quantified by R² = 2 × EAF × (1 - EAF) × β². In addition, the F-statistics were computed [F = R² × (N − 2)/(1 - R²)] to assess IVs strength, and IVs with F > 10 were considered valid [22].

To reduce sample overlap, GWAS summary statistics for DM, DR, and its subtypes (defined by ICD-10 codes) were obtained from FinnGen Release 5 (https://r5.finngen.fi/), a Finnish cohort integrating genomic and national health registry data (n = 500,000). Covariates included sex, age, genotyping batch, and ten principal components (PCs) [23]. Case-control distributions were as follows: DM (29,193 cases vs. 182,573 controls), DR (14,584 cases vs. 202,082 controls), NPDR (455 cases vs. 204,208 controls), and PDR (8,681 cases vs. 204,208 controls). Disease classifications were assessed in the original studies and required no additional validation.

Statistical analysis

We employed the inverse-variance weighted (IVW) fixed-effects model as the primary method to assess genetic associations between IVs associated with renal function and DR risk. The IVW method demonstrates optimal statistical power when all IVs are valid and no horizontal pleiotropy exists [24]. To verify result robustness, we additionally utilized the weighted median (WM) analysis and MR-Egger regression analysis. The WM analysis provides reliable estimates even when some IVs lack statistical power, whereas MR-Egger regression primarily detects pleiotropy. A non-zero MR-Egger intercept with p > 0.05 indicates absence of pleiotropy. Notably, MR-Egger regression remains valid even if > 50% of SNPs exhibit pleiotropy. Furthermore, we applied Cochrane’s Q test to evaluate potential heterogeneity. When heterogeneity was detected (p < 0.05), the IVW random-effects model was adopted as the primary analytical approach [25]. To ensure genetic associations directionality, reverse MR analysis was conducted to exclude reverse causation. The results of the MR-PRESSO test were shown in Supplementary Table 6. Finally, multivariable MR (MVMR) analysis was performed to adjust for potential confounders (HBP and HbA1c). IV selection criteria for MVMR aligned with those of two-sample MR. The instrumental variables related to kidney function and DR, along with their R² and F-statistics, were shown in Supplementary Table 7. All analyses were conducted using the TwoSampleMR package (version 0.6.6) in R software (version 4.3.1).

Results

Nested case-control study

Baseline information and clinical characteristics

As shown in Table 1, this study included 173 participants aged 40 - 79 years, including 74 females and 99 males. The normal group (n = 45) had a mean age of 57.20 ± 10.82 years. The DM group (n = 37) had a mean age of 60.00 ± 10.43 years, while the NPDR group (n = 45) and PDR group (n = 46) showed mean ages of 59.49 ± 9.12 and 59.02 ± 9.14 years, respectively. HbA1c and FBG increased progressively with DR severity (P_HbA1c < 0.001, P_FBG < 0.001).

Table 1.

Baseline characteristics and clinical profiles of the normal, DM, NPDR and PDR groups

	Normal	DM	NPDR	PDR	H/χ2/F	P
Number	45	37	45	46	-	-
Age, years	57.20±10.82	60.00±10.43	59.49±9.12	59.02±9.14	0.65 a	0.582
Sex (Female/Male)	18/27	16/21	21/24	19/27	0.46 c	0.927
BMI, kg/m2	23.44 (22.13 - 26.37)	24.77 (23.89 - 26.35)	23.84 (22.96 - 25.81)	25.35 (22.94 - 27.68)	3.89b	0.273
Duration of DM, years	-	10.00 (5.00–17.50)	10.00 (8.00–18.00)	15.00 (8.75–20.00)	2.49b	0.289
Duration of HBP, years	-	0.00 (0–6.50)	3.00 (0–10.00)	2.00 (0–9.25)	5.58b	0.061
Medication use for DM (N/Y)	-	17/20	8/37	6/40	13.66c	0.001*
Medication use for HBP (N/Y)	-	28/9	23/22	22/24	7.483c	0.024*
HbA1c, %	5.60 (5.40–5.95)	6.80 (6.45–7.80)	7.40 (7.05–9.00)	8.60 (7.45–9.83)	82.39b	< 0.001*
FBG, mmol/L	5.51 (4.99–6.30)	7.70 (6.45–9.53)	8.50 (7.16–10.76)	8.10 (6.88–10.51)	18.91b	< 0.001*
Cystatin C, mg/L	0.67 (0.60–0.76)	0.84 (0.76–0.94)	0.91 (0.85–1.08)	1.16 (0.87–1.39)	89.73b	< 0.001*
Serum creatinine, μmoL/L	59.10 (55.50–67.20)	66.90 (56.55–73.75)	68.30 (55.20–81.90)	76.05 (62.80–97.20)	16.97b	< 0.001*
eGFRScr, mL/min/1.73 m2	103.57 (100.26–109.74)	99.69 (93.59–105.69)	96.21 (85.34–104.50)	91.19 (72.55–104.27)	24.91b	< 0.001*
eGFRCysC, mL/min/1.73 m2	111.24 (104.94–119.88)	96.64 (81.36–102.643)	87.04 (67.88–92.84)	62.97 (50.56–89.32)	91.52b	< 0.001*
eGFRScr+CysC, mL/min/1.73 m2	114.41 (109.78–120.09)	103.24 (91.75–108.99)	93.83 (79.58–104.26)	78.05 (61.69–101.61)	77.44b	< 0.001*
eGFRabdiff, mL/min/1.73 m2	7.07 (3.2–12.63)	−5.07 (−11.62–4.39)	−8.87 (−20.85–0.39)	−18.97 (−26.88–8.23)	78.428b	< 0.001*
eGFRrediff, mL/min/1.73 m2	1.07 (1.03–1.11)	0.95 (0.87–1.04)	0.91 (0.8–1)	0.77 (0.66–0.89)	79.278b	< 0.001*
Urine creatinine, μmol/L	12491 (8894–19630)	11884 (8721–14660)	11041 (6927–13997)	7611 (5448–12132)	15.19b	0.002*
ACR, mg/g Cr	6.67 (4.49–13.70)	13.14 (6.83–40.43)	23.19 (11.89–245.53)	132.80 (30.54–1475.36)	60.32b	< 0.001*
UMA, mg/dl	1.17 (0.72–2.08)	1.65 (0.89–3.94)	3.58 (1.26–27.70)	15.05 (1.89–146.50)	38.24b	< 0.001*

Values are presented as mean ± SD or median (interquartile range). *Statistically significant: P<0.05. According to the type of data and the data distribution, ^a one-way ANOVA analysis, ^b Kruskal-Walli’s analysis, ^c Chi-square test were applied. ACR, urinary albumin-to-creatinine ratio; CysC, cystatin C; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate (calculated using CKD-EPI 2021 equations based on Scr, CysC, or both Scr and CysC); eGFR_abdiff = eGFR_CysC - eGFR_Scr; eGFR_rediff = eGFR_CysC / eGFR_Scr; FBG, fasting blood glucose; HbA1c, glycated hemoglobin; HBP, high blood pressure; NPDR, non-proliferative diabetic retinopathy; PDR, proliferative diabetic retinopathy; Scr, serum creatinine; UMA, urinary microalbumin; BMI, body mass index; Y: Yes; N: No.

Among renal metabolic indicators, serum CysC and Scr levels increased with DR severity (both p < 0.001), while all eGFR indicators (eGFR_Scr, eGFR_CysC, eGFR_Scr+CysC, eGFR_abdiff, eGFR_rediff) declined (all p < 0.001). Similarly, urinary creatinine (p = 0.002), ACR (p < 0.001), and UMA (p < 0.001) increased with DR severity and exhibited significant differences among groups.

Cystatin C and eGFR_CysC as early biomarkers for DM and Early DR

As shown in Supplementary Table 1, when comparing the DM group with Normal controls after adjusting for confounders (age, sex, HbA1c, and FBG), multivariate binary regression analysis showed CysC was significantly associated with DM (OR = 13.22 per 0.1 mg/L, 95% CI = 2.78 - 62.89, p = 0.001). Scr was also significantly associated with DM (OR = 1.15, 95% CI = 1.04 - 1.26, p = 0.006). eGFR was significantly associated with DM (all p < 0.05).

As presented in Supplementary Table 2, in comparisons with the NPDR and DM groups after adjusting for confounders (DM duration, HBP duration, DM medication, HBP medication, HbA1c, and FBG), CysC was significantly associated with NPDR (OR = 1.56 per 0.1 mg/L, 95% CI = 1.11 - 2.31, p = 0.016). eGFR_CysC, and eGFR_Scr+CysC were significantly associated with NPDR (all p < 0.05).

As detailed in Supplementary Table 3, when combining NPDR and PDR into a DR group (vs. DM group) and adjusting for confounders, CysC (OR = 1.65 per 0.1 mg/L, 95% CI = 1.25 - 2.33, p = 0.002), Scr (OR = 1.03, 95% CI = 1.01 - 1.06, p = 0.033), and ACR (OR = 1.22 per100 mg/g Cr, 95% CI = 1.05 - 1.53, p = 0.036) were significantly associated with DR. eGFR_CysC, eGFR_Scr+CysC, eGFR_rediff and eGFR_abdiff were significantly associated with DR (all p < 0.05).

Diagnostic superiority of cystatin C and eGFR_CysC in DM and Early DR

The area under the curve (AUC) by ROC curve analysis for serum eGFR_CysC (AUC = 0. 969, 95% CI = 0.939 - 0.999, p < 0.001) and for CysC (AUC = 0.963, 95% CI = 0.917 - 0.999, p < 0.001) was higher than that of other indicators for DM (Fig. 2A). By calculating the Youden index, the cut-off values of CysC and eGFR_CysC were 0.745 mg/L (sensitivity: 83.78%, specificity: 71.11%) and 103.8 mL/min/1.73 m² (sensitivity: 78.38%, specificity: 84.44%), respectively.

Fig. 2.

ROC curve analysis of the renal biomarkers in normal, DM, NPDR, and PDR groups

ROC curve analysis was performed for Normal vs DM (a), DM vs NPDR (b), NPDR vs PDR (c), and DM vs DR (d), respectively.*Statistically significant: p < 0.05. ACR urinary albumin-to-creatinine ratio, AUC area under the curve, CysC cystatin C, DM diabetes mellitus, DR diabetic retinopathy (NPDR+PDR), eGFR estimated glomerular filtration rate (CKD-EPI 2021 equations using Scr, CysC, or both), eGFR_abdiff eGFR_CysC - eGFR_Scr, eGFR_rediff eGFR_CysC/eGFR_Scr, NPDR non-proliferative diabetic retinopathy, PDR proliferative diabetic retinopathy, ROC receiver operating characteristic, Scr serum creatinine, UMA urinary microalbumin

When compared with DM group (Fig. 2B), CysC in the NPDR group exhibited the largest AUC (AUC = 0.813, 95% CI = 0.722 - 0.905, p < 0.001), which was identical to that of eGFR_CysC (AUC = 0.813, 95% CI = 0.721 - 0.904, p < 0.001). The findings underscore the diagnostic significance of CysC and eGFR_CysC in the early-stage DR, demonstrating better discrimination between groups than other conventional renal biomarkers. By calculating the Youden index, the cut-off values of CysC and eGFR_CysC were 0.875 mg/L (sensitivity: 66.67%, specificity: 64.86%) and 89.60 mL/min/1.73 m² (sensitivity: 60.00%, specificity: 64.86%), respectively.

When analyzing the DR group (combined NPDR and PDR) versus the DM group (Fig. 2D), CysC showed the highest AUC (AUC = 0.850, 95% CI = 0.783 - 0.916, p < 0.001), followed closely by eGFR_CysC (AUC = 0.848, 95% CI = 0.782 - 0.915, p < 0.001). These results suggest that both eGFR_CysC and CysC are strongly associated with DR and serve as effective biomarkers for identifying DR. By calculating the Youden index, the cut-off values of CysC and eGFR_CysC were 0.895 mg/L (sensitivity: 65.56%, specificity: 70.27%) and 89.78 mL/min/1.73 m² (sensitivity: 68.13%, specificity: 64.86%), respectively.

Impact and predictive efficacy of eGFR_rediff and ACR in PDR

As shown in Supplementary Table 4, multivariate binary regression analysis identified that ACR (OR = 1.05 per 100 mg/g Cr, 95% CI = 1.005 - 1.102, p = 0.048) and UMA (OR = 1.04 per 10 mg/dl, 95% CI = 1.01 - 1.10, p = 0.048) were significantly associated with PDR in comparison with the NPDR group. Besides, eGFR_abdiff, eGFR_rediff, eGFR_CysC, and eGFR_Scr+CysC were also significantly associated with PDR (all p < 0.05).

ROC curve analysis (Fig. 2C) showed that several biomarkers had significant diagnostic ability for PDR. Among these, eGFR_rediff (AUC = 0.765, 95% CI = 0.667 - 0.863, p < 0.001), eGFR_CysC (AUC = 0.746, 95% CI = 0.645 - 0.847, p < 0.001) and ACR (AUC = 0.737, 95% CI = 0.634 - 0.847, p < 0.001) exhibited the largest AUCs, indicating their superior diagnostic performance for PDR compared to other renal indicators. By calculating the Youden index, the cut-off values of eGFR_rediff and ACR were 0.814 mL/min/1.73 m² (sensitivity: 65.22%, specificity: 66.67%) and 39.32 mg/g Cr (sensitivity: 71.74%, specificity: 60.00%), respectively.

Mendelian randomization analysis

Following quality control (removing LD-correlated and palindromic SNPs), 180 – 333 SNPs were analyzed. These SNPs exhibited F-statistics ranging from 24.9 to 14,399.9, confirming their strong validity as IVs. As shown in Fig. 3, the IVW method demonstrated that each SD increase in serum CysC levels significantly elevated the risks of DM (OR = 1.15, p = 0.011), DR (OR = 1.14, p = 0.0077), NPDR (OR = 2.00, p = 0.00067), and PDR (OR = 1.12, p = 0.042). MR-Egger and WM analyses showed consistent significant effects and genetic associations directions. Notably, no significant associations were observed between Scr or eGFR and DR. MR-Egger regression detected no evidence of horizontal pleiotropy, supporting the reliability of IVW-derived genetic associations. Significant heterogeneity was detected in multiple associations, prompting the use of random-effects models as the primary analytical approach to conservatively estimate genetic associations and generate reliable confidence intervals. Reverse MR analysis revealed no reverse causation (Fig. 3). In multivariable MR analysis adjusting for hypertension and HbA1c, CysC remained significantly associated with the above outcomes (Fig. 3).

Fig. 3.

Forest plot of mendelian randomization analyses for renal biomarkers in DM and DR cohorts. *statistically significant: p < 0.05. 95% ci 95% confidence interval, DM diabetes mellitus, DR diabetic retinopathy, eGFR estimated glomerular filtration rate (CKD-EPI 2021 equations using Scr or CysC), IVW-RE inverse-variance weighted random effects, NPDR non-proliferative diabetic retinopathy, or odds ratio, PDR proliferative diabetic retinopathy, SNPs single nucleotide polymorphisms, WMMR weighted median Mendelian randomization

Discussion

We identified elevated serum CysC was significantly associated with both DM and the early-stage DR by nest control study. Meanwhile, eGFR_CysC was also significantly associated with DR severity. In PDR patients, both eGFR_rediff, eGFR_CysC and ACR demonstrated superior predictive performance compared to other renal biomarkers. MR analysis further validated the genetic associations between CysC and DR severity. These findings reveal significant heterogeneity in the performance of renal biomarkers across different stages of DR, suggesting distinct molecular mechanisms underlying microvascular injury at different DR stages. Consequently, the combined evaluation of diverse renal biomarkers offers critical insights for the early screening, risk stratification, and severity monitoring of DR.

Cystatin C, a member of the cysteine protease inhibitor family, is eliminated exclusively by glomerular filtration. When renal filtration capacity declines, it accumulates in circulation and can be served as a sensitive indicator of early glomerular charge-selective barrier dysfunction [26]. Importantly, filtration abnormalities of CysC (13 kDa) precede urinary albumin leakage (66 kDa) [27, 28], enabling earlier detection of microvascular damage. This diagnostic advantage is further supported by elevated CysC levels in normoalbuminuric patients with normal eGFR, highlighting its sensitivity for incipient renal impairment [29]. Moreover, compared to Scr, CysC exhibits greater independence from non-renal influences (e.g., muscle mass, obesity, age, inflammation, chronic comorbidities) and enhanced detection stability, providing a more reliable reflection of renal impairment [12, 30].

Multiple studies have shown that elevated serum CysC levels are associated with DR progression [31]. In our cohort, eGFR_CysC also demonstrated superior clinical utility in DR. Our findings confirmed the clinical significance of eGFR_CysC in DR severity, with its diagnostic performance significantly surpassing eGFR_Scr. A meta-analysis [11] showed that eGFR_CysC has significant advantages over traditional eGFR_Scr in clinical practice, enabling more accurate detection of chronic kidney disease. It significantly improves patient reclassification, facilitates earlier intervention for high-risk individuals, and helps avoid overtreatment in low-risk patients. These advantages are particularly pronounced in populations with abnormal muscle mass, the elderly individuals, or diabetic patients, underscoring the central role of eGFR_CysC in chronic kidney disease management. Mechanistically, when glomerular basement membrane pore size decreases slightly, smaller-molecule creatinine still can pass through, making eGFR_CysC more sensitive to reflect the decline in filtration function [10]. Tsai et al. demonstrated that eGFR_CysC detects renal function decline in DM patients more sensitively than eGFR_Scr and shows stronger associations with microvascular complications including DR [32].

MR analysis further demonstrated significant genetic associations between CysC and DR severity, with reverse MR effectively excluding reverse causation. Importantly, multivariable regression analysis showed that CysC retained significant associations with DR severity, particularly in disease early-stage, aligning with our clinical observations. These genetic and clinical evidence collectively support CysC as both a mechanistic contributor and an effective biomarker for early DR detection. Although our MR analysis did not detect the evidence of reverse causation, the case-control design and potential measurement error in serum biomarker assessments may still introduce instability to the results. Future research should systematically evaluate the dynamic changes and clinical impacts of renal function biomarkers in DR progression through multicenter, large-sample prospective cohort studies.

Additionally, the diagnostic cut-off value of CysC or eGFR_CysC for DM and NPDR could be applied clinically for DM or DR screening. Based on our ROC curve analysis, for individuals without a history of DM, serum CysC levels ≥0.745 mg/L or eGFR_CysC < 103.8 mL/min/1.73 m² may serve as potential screening thresholds for early DM detection. Among patients with DM but without retinopathy, serum CysC levels ≥0.875 mg/L or eGFR_CysC < 89.60 mL/min/1.73 m² indicate a significantly higher risk of developing DR. Patients exceeding these thresholds could be prioritized for more frequent and comprehensive ophthalmological examinations, enabling earlier detection of DR and timely intervention before the onset of vision-threatening complications. Early identification could prompt clinicians to intensify glycemic, blood pressure, and lipid control for these individuals, potentially slowing the progression of both retinal and renal microvascular damage.

Based on these findings, we propose an updated renal–retinal axis (Fig. 4): Persistent hyperglycemia activates AGE–RAGE signaling and oxidative stress, leading to endothelial glycocalyx degradation and subsequent systemic endothelial dysfunction, which increases glomerular permeability and retinal microvascular leakage [33]. The resulting glomerular injury is reflected by a decline in eGFR and a rise in UACR. Albuminuria not only indicates glomerular damage but also reflects systemic endothelial inflammation, which is closely linked to the development and progression of DR [34, 35]. Furthermore, reduced eGFR implies impaired renal clearance and the accumulation of circulating pro-inflammatory and pro-angiogenic mediators, including vascular endothelial growth factor (VEGF) family members, that may exacerbate retinal barrier disruption [36, 37].

Fig. 4.

CysC in the pathogenesis of diabetic retinopathy: exploring the renal-retinal axis. CysC cystatin C, DM diabetes mellitus, NPDR non-proliferative diabetic retinopathy, PDR proliferative diabetic retinopathy, DR diabetic retinopathy, eGFR estimated glomerular filtration rate, IBRB inner blood-retinal barrier, OBRB outer blood-retinal barrier, RPE retinal pigment epithelium, VEGF vascular endothelial growth factor

Within this broader network, CysC may serve not only as a sensitive marker of early renal impairment, but also as an active molecular intermediary linking renal and retinal microvascular injury. Hyperglycemia-related glomerular impairment reduces eGFR and elevates circulating CysC levels. Elevated CysC actives VEGF signaling and inflammatory cascades [31]. Elevated VEGF expression level and inflammation are key factors of DR, contributing to disruption of both the inner and outer blood–retinal barriers. Furthermore, CysC deposition in retinal pigment epithelial (RPE) cells may directly impair RPE structure and function, promoting outer barrier breakdown. Through these mechanisms, CysC may acts as a multi-target mediator of inner and outer blood–retinal barrier disruption, thereby accelerating DR progression. Therefore, CysC plays an equally critical role within the renal–retinal axis.

We also found that both eGFR_abdiff and eGFR_rediff were significantly associated with DM and DR. In PDR, AUC of eGFR_rediff was higher than other renal indicators, underscoring its superior discriminative capacity. These findings align with He et al.‘s study [9], which not only confirmed the correlations between eGFR_abdiff/eGFR_rediff and DR, but also highlighted the superior predictive performance of eGFR_rediff for DR. The biological significance of eGFR_diff lies in its ability to account for confounding non-GFR factors. Serum creatinine is influenced by age, sex, muscle mass, protein intake, and physical activity, whereas cystatin C levels are affected by adiposity, systemic inflammation, thyroid disease, and steroid use [38]. The diagnostic advantage of eGFR_rediff therefore stems from its unique dual properties as a ratio indicator that integrates complementary information from CysC and Scr, captures pathophysiological changes mediated by non-GFR factors, and sensitively reflects cumulative renal injury from prolonged disease duration [9]. Emerging studies have further demonstrated the utility of eGFR_diff in various clinical contexts beyond diabetic complications, including hypertension [39], impaired muscle mass and frailty in older adults [40], and atrial fibrillation [41]. These findings highlight its potential as a broader indicator of kidney function, though further validation is needed before routine clinical implementation. The diagnostic advantage of eGFR_rediff due to its unique dual properties as a ratio indicator which integrates complementary information from CysC and Scr, captures pathophysiological changes mediated by non-GFR factors (e.g., inflammation and metabolic disorders), and sensitively reflects cumulative renal injury from prolonged disease duration [9].

Besides, ACR exhibited the third-highest predictive value after eGFR_rediff in PDR. This is supported by an 8-year cohort study which found that ACR > 30 mg/g significantly predicts PDR incidence [42]. Notably, ACR exhibited superior diagnostic performance compared to serum CysC in PDR. This disparity may be attributable to the confounding influence of proteinuria on CysC metabolism. Following glomerular filtration, most CysC undergoes proximal tubular reabsorption and catabolism, with its urinary excretion influenced by tubular functional integrity. Emerging evidence indicates that serum CysC correlates with urinary enzyme excretion, suggesting that tubular dysfunction alters CysC metabolism, and, proteinuria may competitively inhibit tubular reabsorption of cystatin C [43]. Supporting this hypothesis, Tkaczyk et al. observed increased urinary cystatin excretion in pediatric nephrotic syndrome patients with proteinuria [44]. Therefore, we hypothesize that elevated ACR in PDR may fluent tubular reabsorption of CysC, leading to reduced serum concentrations and thereby diminishing its diagnostic sensitivity in late-stage disease.

This study presents several novel contributions: First, we provide the first comprehensive evaluation of both conventional and emerging renal biomarkers, including the derived parameters eGFR_rediff and eGFR_abdiff, across different stages of DR. This systematic comparison offers a holistic view of their relative diagnostic performance. Second, we integrated data from a clinical cohort with evidence derived from MR analysis. This combined approach not only reinforces consistent clinical correlations but also provides genetic evidence for the observed associations. Furthermore, the identification of clinically applicable thresholds for CysC and eGFR_CysC improves the translational utility of these biomarkers for risk stratification in diabetic populations. Together, our findings advance the understanding of the relationship between renal impairment and diabetic retinopathy, offering important implications for early diagnosis and future therapeutic development.

This study has the following limitations: First, as a nested case-control study, future research should systematically evaluate the dynamic changes and impacts of renal function biomarkers in DR progression through multicenter, large-sample prospective cohort studies. We will address this aspect through larger-scale cohort studies in future research. Second, the genetic IVs used in MR analysis were derived from European population genomic databases, which may limit the cross-ethnic generalizability of the conclusions. Our study demonstrates relationships between renal biomarkers and DR in a Chinese cohort. However, our MR analysis relies on genetic data from European populations, which may introduce potential population stratification bias. Therefore, future large-scale GWAS in East Asian populations will be essential to validate this causal inference.

Conclusions

In summary, by analyzing clinical and metabolic characteristics of DR patients across different stages combined with MR analysis, this study systematically revealed the critical roles of CysC and eGFR_CysC in DM and early DR. Further analyses highlighted the superior predictive performance of eGFR_rediff and ACR in PDR. MR analysis confirmed that elevated serum CysC levels significantly increase genetic susceptibility to DR, particularly in early stages. Based on these findings, we propose a CysC-targeted primary prevention strategy and an ACR-monitoring-centered secondary prevention strategy, providing crucial evidence for precise DR staging and targeted interventions.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ACR

Albumin-to-creatinine ratio

ADA

American Diabetes Association

ANOVA

Analysis of variance

AUC

Area under the curve

BCVA

Best-corrected visual acuity

CI

Confidence interval

CKD

Chronic kidney disease

CKD-EPI

Chronic Kidney Disease Epidemiology Collaboration

CV

Coefficient of variation

CysC

Cystatin C

DKD

Diabetic kidney disease

DM

Diabetes mellitus

DR

Diabetic retinopathy

eGFR

Estimated glomerular filtration rate

FBG

Fasting blood glucose

GWAS

Genome-wide association study

HbA1c

Glycated hemoglobin

HBP

Hypertension

ICD-10

International Classification of Diseases, 10th Revision

IV

Instrumental variable

IVW

Inverse-variance weighted

LD

Linkage disequilibrium

MR

Mendelian randomization

MVMR

Multivariable Mendelian randomization

NPDR

Non-proliferative diabetic retinopathy

OR

Odds ratio

PDR

Proliferative diabetic retinopathy

PC

Principal component

ROC

Receiver operating characteristic

RPE

Retinal pigment epithelial

Scr

Serum creatinine

SD

Standard deviation

SNP

Single-nucleotide polymorphism

SS-OCT

Swept-source optical coherence tomography

SS-OCTA

Swept-source optical coherence tomography angiography

STROBE-MR

Strengthening the Reporting of Observational Studies in Epidemiology for Mendelian Randomization

T2DM

Type 2 diabetes mellitus

Ucr

Urinary creatinine

UMA

Urinary microalbumin

VEGF

Vascular endothelial growth factor

WM

Weighted median

Authors’ contributions

XSC wrote the original draft, developed the software, designed the methodology, performed formal analysis, curated data, and contributed to conceptualization. WCY wrote the original draft, developed the software, designed the methodology, performed formal analysis, and curated data. YYZ wrote the original draft and curated data. HYW curated data and contributed to conceptualization. LHP, YY, and XMG curated data. XYZ reviewed and edited the paper, created visualizations, performed validation, provided supervision, acquired resources, managed the project, acquired funding, and contributed to conceptualization.

Funding

This work was supported by the National Natural Science Foundation of China [82070988]; the Ministry of Science and Technology of the People’s Republic of China [2024YFE0100900]; and the National Health Commission of the People’s Republic of China [2023ZD0509002].

Data availability

The data from the nested case-control study are available from the corresponding author on reasonable request. The data for Mendelian randomization analyses were derived from public resources: UK Biobank (https://www.ukbio bank.ac.uk/), the CKD Genetics Consortium and UK Biobank GWAS meta-analysis (doi:10.1038/s41467-021–24491-0), and FinnGen Release 5 (https://r5.finngen.fi/).

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of Beijing Tongren Hospital, Capital Medical University (TRECKY2020-111), adhering to the Declaration of Helsinki, with written informed consent obtained from all participants.

Consent for publication

Not applicable.

Authors’ information

Not applicable.

Footnotes

Not applicable.

Competing interests

The authors declare no competing interests.