Association and Age Heterogeneity of Systemic Metabolomic Signatures With Age-Related Cataract: A Mendelian Randomization and Heterogeneity Analysis Authors: Zhangluxi Liu, Ao Lu, Siyuan He, Yiying Wang, Guannan Su, Peizeng Yang Published: 2026 ABSTRACT PurposeTo evaluate systemic metabolomic signatures associated with age-related cataract and prioritize signals with age-stable or age-heterogeneous patterns.MethodsWe performed two-sample Mendelian randomization (MR) analyses of 249 Nightingale Health metabolomic traits against age-related cataract in FinnGen 12. Higher-confidence MR candidates were defined by inverse-variance weighted (IVW) false discovery rate (FDR) < 0.05, complete directional concordance across seven complementary MR and sensitivity analyses, and nominal support in at least five analyses. MR-prioritized candidates were then triangulated against UK Biobank incident cataract associations. Age-stratum heterogeneity was assessed using tertile-specific UK Biobank estimates and Cochran Q statistics.ResultsAmong 249 metabolites, 71 met the IVW FDR threshold, and 22 remained after higher-confidence MR filtering. Of these, 13 showed MR-concordant associations with incident cataract in the UK Biobank 500,000 summary layer. Age-stratum assessment prioritized eight metabolites, comprised of five age-stable fatty-acid/lipoprotein-related signals and three age-heterogeneous signals involving glycoprotein acetyls, phenylalanine, and total fatty acids.ConclusionsUsing an MR-guided triangulation framework, we prioritized eight systemic metabolomic signals associated with age-related cataract. These signals were comprised of an age-stable fatty-acid/lipoprotein axis and age-heterogeneous inflammatory, amino-acid, and fatty-acid traits, highlighting distinct systemic metabolic patterns that may inform future cataract biomarker and mechanistic studies.Translational RelevanceThis study provides a focused set of systemic metabolic candidates for future cataract biomarker validation and prospective risk-stratification research. The integration of MR prioritization, UK Biobank triangulation, and age-stratum heterogeneity assessment offers a transferable framework for studying metabolic signatures in age-related ocular disease. Introduction Age-related cataract is a major cause of visual impairment worldwide and remains the leading cause of blindness among adults 50 years and older.1–4 Although lens aging provides the proximate substrate for cataract formation, lens homeostasis is influenced by a broader systemic milieu accumulated over decades. Oxidative stress, antioxidant vulnerability,5,6 systemic inflammation and metabolic dysregulation,5 protein aggregation,7 and vascular endothelial dysfunction8 have all been implicated in lens opacification. These observations support the concept that cataract risk may reflect not only local lens senescence but also systemic biological states that affect redox balance, proteostasis, and metabolic homeostasis. Large-scale biobanks linked to quantitative metabolomics now provide an opportunity to evaluate this systemic dimension at population scale. Prior observational studies have associated serum lipid traits9 and cardiometabolic risk factors10 with lens opacities and cataract risk. Systemic inflammatory markers have also been linked to cataract, consistent with a contribution of broader inflammatory and vascular risk environments.8 More recent genome-wide association studies (GWAS)11,12 and Mendelian randomization (MR) analyses have extended the genetic evidence for cataract susceptibility and implicated behavioral, clinical, and metabolic pathways.13 However, most prior work has evaluated individual exposures or risk factors rather than systematically prioritizing metabolomic signatures supported across genetic and observational layers. Moreover, age may influence the detectability and interpretation of systemic metabolic signals: The aging lens undergoes cumulative oxidative and structural damage, whereas systemic lipid, inflammatory, and cardiometabolic profiles also change across later life. Prospective and longitudinal studies have suggested that metabolic-syndrome–related cataract risk may differ by age or over time.14–16 Whether cataract-associated metabolomic signals are stable across age strata or show age-stratum heterogeneity therefore remains unresolved. To address this question, we developed an MR-guided triangulation framework for prioritizing systemic metabolomic signals associated with age-related cataract. By integrating genetic evidence, UK Biobank observational associations, and age-stratified estimates, we distinguished signals that were consistent across age strata from those showing age-stratum heterogeneity. The resulting candidate set provides a basis for future biomarker validation and mechanistic studies of systemic metabolic contributions to cataract. Methods Study Design and Data Sources This summary-statistics study combined two-sample MR discovery with secondary analyses of UK Biobank Nightingale metabolite–disease association resources (Fig. 1, Supplementary Fig. S1).17–20 The MR exposure panel included 249 plasma nuclear magnetic resonance (NMR) metabolomic biomarkers measured by the Nightingale Health high-throughput platform, spanning lipoprotein subclass lipid measures, fatty acids and fatty-acid composition, amino acids, glycolysis-related metabolites, ketone bodies, and the inflammation marker glycoprotein acetyls.21 Genetic association estimates for these biomarkers were obtained from OpenGWAS-hosted UK Biobank Nightingale metabolite GWAS, with sample sizes up to 115,078 participants. The MR outcome dataset was FinnGen 12 age-related cataract, including 83,886 cases and 409,535 controls.22 MR-guided triangulation workflow. The study proceeded from 249 screened metabolites to 71 IVW/FDR-positive MR signals and then to 22 higher-confidence MR candidates requiring complete directional consistency across seven methods and nominal support in at least five methods. The primary incident UK Biobank summary analysis supported 13 candidates, which were separated into five stable-core and three age-heterogeneous metabolites. Additional UK Biobank analyses supported eight of eight, seven of eight, and seven of eight prioritized metabolites in the 500k prevalent, 120k incident, and 120k prevalent analyses, respectively. A subsequent sensitivity-tiering step integrated support across these analyses and the stringent MR screen. The observational triangulation layer used Nightingale metabolite–disease summary statistics from the UK Biobank 500,000 (500k) and 120,000 (120k) releases.17,18 These resources provided all-age and age-stratified associations between measured plasma biomarkers and cataract outcomes. As all analyses used deidentified, publicly available or approved-access summary statistics, no additional participant-level ethical approval was required.19,22 Mendelian Randomization Discovery For each metabolite, genetic instruments were selected at P < 1 × 10−6, clumped within a 500-kb window at r2 < 0.1, and filtered to remove weak instruments with F statistics < 10. Exposure and outcome alleles were harmonized with allele-frequency-aware handling of palindromic variants. Steiger filtering was applied to reduce the likelihood of reverse-direction instruments. The primary MR estimator was inverse-variance weighted (IVW) MR. To reduce dependence on a single estimator, IVW results were complemented by MR-Egger regression,23 weighted median,24 weighted mode,25 MR-PRESSO,26 radial MR,27 and MRConMix,28 in addition to IVW MR. Cochran Q statistics, MR-Egger intercept tests, MR-PRESSO global and outlier tests, radial outlier assessment, and leave-one-out analyses were used to evaluate heterogeneity, directional pleiotropy, and influential instruments. Notably, these diagnostics were interpreted jointly as robustness assessments, because NMR metabolomic traits, particularly lipid and lipoprotein measures, are polygenic and biologically correlated. A metabolite was classified as a higher-confidence MR candidate if it satisfied three criteria: IVW FDR < 0.05, complete directional concordance across the seven MR and sensitivity approaches, and nominal support (P < 0.05) in at least five of the seven approaches. As a stringent sensitivity analysis, the MR screen was repeated using a stricter genome-wide instrument-selection threshold (P < 5 × 10−8). UK Biobank Observational Triangulation The primary observational layer was the UK Biobank 500k incident cataract summary statistics from the Nightingale resource. Age-related cataract was defined using International Classification of Diseases, Tenth Revision (ICD-10) code H25, and sex-combined summary results were used. In the source resource, incident associations were estimated using Cox regression with age as the timescale and adjustment for sex and assessment center. Prevalent associations were estimated using logistic regression adjusted for age, sex, and assessment center. Biomarker associations were reported per 1-SD higher measured biomarker level after Nightingale spectrometer adjustment.17,18 An MR-prioritized metabolite was considered supported in the primary observational layer if the all-age UK Biobank 500k incident cataract association met FDR < 0.05 and showed the same direction of association as the MR estimate. Age-Stratum Heterogeneity and Priority Definitions Each UK Biobank Nightingale summary table included one all-age estimate and three age-stratified estimates. In the 500k release, age tertiles were 37 to 53, 54 to 61, and 62 to 73 years; in the 120k release, age tertiles were 39 to 53, 54 to 61, and 62 to 71 years. Among metabolites supported in the primary 500k incident analysis, age-stratum heterogeneity was assessed by comparing tertile-specific log hazard ratios (HRs) using Cochran Q tests, with FDR correction applied to the resulting heterogeneity P values. It should be noted that these analyses evaluated heterogeneity across age-stratum summary estimates rather than participant-level age-by-metabolite interactions. Two priority categories were defined using the primary UK Biobank 500k incident analysis. The stable-core category required all-age support, FDR-corrected support in all three age tertiles, MR-concordant direction for the all-age estimate and all three age-stratum estimates, and no FDR-significant age-stratum heterogeneity. The age-heterogeneous category required all-age support together with FDR-significant Cochran Q heterogeneity across the three age strata. The final prioritized metabolite reporting set was defined as the union of the stable-core and age-heterogeneous metabolites. The stable-core and age-heterogeneous categories were defined in the primary 500k incident analysis. Their consistency was then evaluated in sensitivity analyses using three complementary UK Biobank summary layers: the 500k prevalent, 120k incident, and 120k prevalent cataract analyses. Because the 120k release is nested within the 500k release, these analyses were used to assess consistency across ascertainment layers rather than as independent external replication. Together with the stringent-instrument MR analysis (P < 5 × 10−8), these sensitivity results were used to assign a descriptive prioritization tier to each prioritized metabolite. One point was assigned for support across all three additional UK Biobank layers, and one point was assigned for meeting the higher-confidence MR criteria under the stringent instrument threshold; metabolites with scores of 2, 1, and 0 were classified as Tier 3, Tier 2, and Tier 1, respectively. Statistical Reporting Benjamini–Hochberg FDR correction was applied separately to the IVW MR screen, all-age UK Biobank association tests, age-stratified association tests, and age-heterogeneity tests, because each test family addressed a distinct analytical question. MR estimates are reported as odds ratios per genetically proxied 1-SD higher metabolite level. UK Biobank incident and prevalent associations are reported as HRs and odds ratios, respectively, per 1-SD higher measured biomarker level. Analyses were performed in R 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria). The main R packages used were TwoSampleMR 0.7.0, ieugwasr 1.1.0.9000, data.table 1.18.2.1, dplyr 1.2.1, MRPRESSO 1.0, RadialMR 1.2.1, MendelianRandomization 0.10.0, and forestploter 1.1.3. Detailed MR results, diagnostic summaries, sensitivity analyses, and prioritized-metabolite data are provided in Supplementary Tables S1 to S6, and overall reporting followed STROBE-MR principles where applicable (Supplementary Table S7).29 Results MR Discovery Defined 22 Higher-Confidence Candidates Across 249 Nightingale metabolites, 71 IVW MR signals met the metabolome-wide FDR threshold (FDR < 0.05) (Fig. 1; Supplementary Tables S1, S2). These FDR-positive signals were predominantly lipoprotein related, especially very-low-density lipoprotein (VLDL) traits, with additional contributions from fatty-acid measures, glycoprotein acetyls, and phenylalanine. Applying the cross-estimator concordance and nominal-support criteria retained 22 higher-confidence MR candidates with directionally consistent evidence across IVW and the six complementary MR or sensitivity approaches (Fig. 1; Supplementary Tables S1–S3). UK Biobank Incident Triangulation Supported 13 MR-Prioritized Signals The 22 higher-confidence MR candidates were evaluated in the UK Biobank 500k incident cataract summary layer. Thirteen candidates met FDR < 0.05 and showed the same direction of association as the MR estimate (Fig. 2; Supplementary Table S6). These incident-supported signals included fatty-acid measures, glycoprotein acetyls, phenylalanine, and several VLDL-related lipid traits. This triangulation step narrowed the broader MR candidate set to directionally coherent metabolite–cataract associations suitable for age-stratum assessment. MR and primary incident associations for 13 incident-supported candidates. Forest plot shows MR odds ratios and primary incident UK Biobank 500k HRs for the 13 higher-confidence MR candidates that also met FDR < 0.05 with MR-concordant direction in the primary incident analysis. Rows are grouped as age-heterogeneous, stable core, or incident-supported-only according to the prioritization framework. Age-Stratum Analysis Prioritized Five Stable-Core and Three Age-Heterogeneous Metabolites Among the 13 incident-supported candidates, age-stratum assessment identified two complementary patterns (Table; Figs. 3, 4; Supplementary Table S6). Five metabolites met the stable-core definition: omega-6 fatty acids, polyunsaturated fatty acids, ratio of free cholesterol to total lipids in small high-density lipoprotein (HDL), total lipids in lipoprotein particles, and cholesterol in very small VLDL (Table; Fig. 3; Supplementary Table S6). Each showed all-age support, FDR-corrected support in all three age tertiles, MR-concordant direction across all all-age and age-stratified estimates, and no FDR-significant age-stratum heterogeneity. These stable-core metabolites formed a directionally consistent fatty-acid/lipoprotein-related component of the prioritized cataract-associated metabolomic signature. Table.Priority Metabolites Identified by Mendelian Randomization-Guided TriangulationMetaboliteBiochemical ClassMR ORMR FDR500k Incident HR500k Incident FDRPriority ClassHeterogeneity FDR500k Prevalent120k Incident120k PrevalentMR SensitivitySensitivity TierPolyunsaturated fatty acidsFatty acids0.9440.0170.935<0.001Stable core0.326YesYesYesYesTier 3Ratio of free cholesterol to total lipids in small HDLSmall HDL ratios0.9680.0240.936<0.001Stable core0.593YesYesYesYesTier 3Cholesterol in very small VLDLVery small VLDL (average diameter 31.3 nm)0.9590.0260.926<0.001Stable core0.812YesYesYesYesTier 3Total fatty acidsFatty acids0.9350.0050.9810.004Age-heterogeneous0.009YesYesYesYesTier 3Omega-6 fatty acidsFatty acids0.9450.0300.942<0.001Stable core0.328YesYesYesNoTier 2Total lipids in lipoprotein particlesTotal lipids0.9520.0480.931<0.001Stable core0.902YesYesYesNoTier 2PhenylalanineAromatic amino acids1.1320.0051.024<0.001Age-heterogeneous0.022YesNoYesYesTier 2Glycoprotein acetylsInflammation1.0530.0441.085<0.001Age-heterogeneous<0.001YesYesNoNoTier 1FDR, false discovery rate; HDL, high-density lipoprotein; HR, hazard ratio; MR, Mendelian randomization; OR, odds ratio; VLDL, very-low-density lipoprotein.MR OR denotes the odds ratio from the inverse-variance weighted Mendelian randomization estimate. Incident HR denotes the HR from the primary incident UK Biobank summary analysis. Sensitivity analyses denote the 500k prevalent analysis and the 120k incident and prevalent analyses from the UK Biobank Nightingale resource. MR sensitivity denotes full Stage 3 support under the P < 5 × 10−8 MR sensitivity analysis. Sensitivity tier is a descriptive support-score summary (Tier 3 = strongest), assigning one point for MR sensitivity support and one point for support in all three additional UK Biobank layers. The listed metabolites represent a prioritized reporting set; correlated lipid and lipoprotein traits should not be interpreted as independent biomarkers. Priority Metabolites Identified by Mendelian Randomization-Guided Triangulation FDR, false discovery rate; HDL, high-density lipoprotein; HR, hazard ratio; MR, Mendelian randomization; OR, odds ratio; VLDL, very-low-density lipoprotein. MR OR denotes the odds ratio from the inverse-variance weighted Mendelian randomization estimate. Incident HR denotes the HR from the primary incident UK Biobank summary analysis. Sensitivity analyses denote the 500k prevalent analysis and the 120k incident and prevalent analyses from the UK Biobank Nightingale resource. MR sensitivity denotes full Stage 3 support under the P < 5 × 10−8 MR sensitivity analysis. Sensitivity tier is a descriptive support-score summary (Tier 3 = strongest), assigning one point for MR sensitivity support and one point for support in all three additional UK Biobank layers. The listed metabolites represent a prioritized reporting set; correlated lipid and lipoprotein traits should not be interpreted as independent biomarkers. Cross-layer support for the prioritized metabolites. Annotated heatmap of effect estimates for the eight prioritized metabolites across the primary incident analysis and three additional UK Biobank summary analyses. Asterisks denote support defined as FDR < 0.05 with MR-concordant direction. Age-stratified HRs for metabolites with age-stratum heterogeneity. Tertile-specific HRs for the three metabolites with age-stratum heterogeneity identified in the primary incident analysis. Glycoprotein acetyls and phenylalanine show risk associations that weaken with age, whereas total fatty acids show an inverse association most apparent in older tertiles. Three additional metabolites met the age-heterogeneous definition: glycoprotein acetyls, phenylalanine, and total fatty acids (Table; Fig. 4; Supplementary Table S6). Glycoprotein acetyls showed risk-associated estimates that progressively attenuated across age tertiles, with HRs of 1.164, 1.123, and 1.053 from the youngest to oldest tertiles, respectively (heterogeneity FDR = 2.04 × 10−6). Phenylalanine showed a similar pattern, with corresponding HRs of 1.050, 1.050, and 1.011 (heterogeneity FDR = 0.022). Total fatty acids showed a protective association only in the oldest tertile (tertile 3 HR = 0.959; 95% confidence interval [CI], 0.944–0.974). Sensitivity Checks and Tiering of the Prioritized Metabolites The final prioritized set was comprised of five stable-core and three age-heterogeneous metabolites (Table; Fig. 3; Supplementary Figs. S2–S6; Supplementary Table S6). All eight were supported in the 500k prevalent cataract analysis. In the 120k incident analysis, seven of eight were supported; phenylalanine remained directionally concordant but did not meet the FDR threshold. In the 120k prevalent analysis, seven of eight were supported; glycoprotein acetyls remained directionally concordant but did not meet the FDR threshold. The stringent MR sensitivity analysis using genome-wide instrument selection (P < 5 × 10−8) further supported five prioritized metabolites: phenylalanine, total fatty acids, cholesterol in very small VLDL, ratio of free cholesterol to total lipids in small HDL, and polyunsaturated fatty acids. All eight prioritized metabolites retained directionally consistent IVW estimates under this stricter instrument threshold (Supplementary Tables S4, S5). Integrating support across the three additional UK Biobank layers and the stringent MR analysis yielded four Tier 3 metabolites, three Tier 2 metabolites, and one Tier 1 metabolite. Tier 3 included total fatty acids, cholesterol in very small VLDL, ratio of free cholesterol to total lipids in small HDL, and polyunsaturated fatty acids. Tier 2 included phenylalanine, total lipids in lipoprotein particles, and omega-6 fatty acids. Glycoprotein acetyls was classified as Tier 1. This tiering provided a descriptive hierarchy within the prioritized metabolite set, distinguishing the most consistently supported candidates across observational layers and MR instrument thresholds (Table). Discussion This MR-guided triangulation analysis prioritized eight systemic metabolomic signals associated with age-related cataract. The signals were comprised of two complementary components: a stable fatty-acid/lipoprotein-related component and a smaller age-heterogeneous component involving inflammation, aromatic amino-acid metabolism, and total fatty acids. The analytic workflow narrowed 249 Nightingale biomarkers to 22 higher-confidence MR candidates, 13 incident-supported candidates in UK Biobank, and finally eight prioritized metabolites after age-stratum assessment. The stable-core findings are consistent with eye-cohort evidence linking circulating lipid traits and cardiometabolic burden to lens opacity and cataract risk.9,10 They also align with prospective studies in which metabolic-syndrome components were associated with cataract incidence or extraction14–16 and with population-based evidence implicating low HDL, hyperglycemia, and adiposity-related metabolic syndrome in cataract.30 The stable-core metabolites centered on omega-6 and polyunsaturated fatty acids, free cholesterol content in small HDL, total lipids in lipoprotein particles, and cholesterol in very small VLDL. Recent genetic studies have identified cataract-associated loci and biological pathways,11,12 and MR studies have tested behavioral and clinical risk factors13 and adiposity-related liability.31 In this context, our findings prioritize circulating fatty-acid and lipoprotein traits as systemic correlates of age-related cataract. The age-heterogeneous metabolites added a further dimension to this systemic signature. Glycoprotein acetyls and phenylalanine showed risk-associated estimates that attenuated across older age strata, whereas total fatty acids showed near-null estimates in the younger and middle strata but an inverse estimate in the oldest stratum. This pattern is compatible with prospective data suggesting that metabolically patterned cataract risk may differ by age window.14–16 The glycoprotein acetyls finding is consistent with prior evidence linking inflammation to cataract8 and with studies positioning GlycA as a marker of chronic inflammatory burden.32,33 Phenylalanine has also been associated with cardiometabolic event risk34 and biological aging35 rather than with eye-specific pathology. These findings point to inflammatory, amino-acid, and fatty-acid–related systemic states whose associations with cataract may vary across later life. These systemic signals may also contribute to interpretation within established pathways of lens aging. Oxidative stress,6,36 endoplasmic-reticulum stress, impaired Nrf2-mediated cytoprotection,36 and protein misfolding are central to lens senescence and cataractogenesis.7 The fatty-acid/lipoprotein signals may reflect systemic redox and membrane-lipid homeostasis, whereas glycoprotein acetyls and phenylalanine may capture inflammatory and metabolic-stress states that shape lens vulnerability. Mechanistic follow-up should determine whether these plasma signals are reflected in aqueous-humor or lens metabolite profiles and whether they influence oxidative-stress, proteostasis, or cytoprotective phenotypes in lens models. The design has several methodological strengths. First, the study began with a metabolome-wide MR screen and applied cross-estimator concordance, pleiotropy diagnostics, outlier-aware sensitivity analyses, and leave-one-out assessment before observational follow-up. This reduced dependence on any single MR estimator and provided a structured basis for candidate prioritization. Second, the framework integrated genetic evidence with measured biomarker–disease associations in UK Biobank, allowing MR-prioritized signals to be evaluated against incident cataract associations. Third, age-stratum assessment separated signals with consistent support across age strata from those showing heterogeneity across age-stratum estimates. This was particularly important for the Nightingale platform, where many lipid and lipoprotein traits are highly correlated; the prioritized metabolites are therefore best interpreted as metabolic axes rather than as independent biomarkers. Finally, the use of summary statistics and explicit filtering criteria makes the analysis reproducible and readily extendable to other age-related ocular diseases. Several limitations should be considered. The study did not include tissue-level or functional experiments, so the findings prioritize metabolites for follow-up rather than establish lens-specific mechanisms. The UK Biobank analyses used portal-derived summary statistics, which prevented participant-level covariate refitting and formal modeling of age-by-metabolite interactions. Therefore, we could not additionally model medication use, diabetes status, smoking, diet, cataract surgery history, cataract subtype, cataract severity, or other participant-level clinical covariates. Consequently, the age-stratified results reflect macroscopic heterogeneity across life stages rather than participant-level estimates of age effect modification. The additional UK Biobank datasets also had partial sample overlap with the main 500k release and were therefore used as within-resource sensitivity analyses across different ascertainment layers. Finally, MR remains assumption dependent, and the predominantly European-ancestry source data may limit generalizability to other populations and earlier-onset cataract profiles. Translational Implication Although the current framework utilizes summary-level data for biomarker prioritization, it yields a highly refined candidate list that warrants targeted validation. Recent evidence linking metabolic syndrome,37 obesity,38,39 and bariatric surgery40 with cataract risk supports the broader view that cataract susceptibility may partly reflect modifiable systemic metabolic burden. Within this framework, the stable fatty-acid/lipoprotein signals may be useful for evaluation in late-life risk-prediction studies, whereas the inflammatory, aromatic-amino-acid, and total-fatty-acid signals warrant age-stratified validation. Prospective ocular cohorts and lens-biology experiments are needed to determine whether these systemic metabolic axes have clinical utility in cataract risk assessment or mechanistic relevance to lens aging. Conclusions Age-related cataract was associated with a stable systemic fatty-acid/lipoprotein signature and a distinct age-heterogeneous inflammatory and amino-acid-related pattern. The eight prioritized metabolites provide focused candidates for validation in prospective ocular cohorts and mechanistic studies of lens biology. These findings support age-stratified evaluation of systemic metabolic signatures in age-related cataract. Supplementary Material Supplement 1 Supplement 2