Genetic and epigenetic underpinnings of biological aging: a multi-omics study integrating Mendelian randomization, spatial transcriptomics, and drug target discovery Authors: Chun Zhang, Jingqi Zhang Published: 2026 ABSTRACT AbstractInflammaging represents a hallmark of biological aging, yet the causal inflammatory mediators driving multi-dimensional epigenetic aging and their effector genes remain poorly characterized at the genetic level. We developed a four-tier analytical framework integrating causal screening, multi-omics effector gene mapping, spatial transcriptomics, and drug target evaluation. Two-sample Mendelian randomization (MR) of 91 circulating inflammatory proteins against six aging phenotypes identified IL-12B, IFNG, and IL-2 as the most robust pro-aging mediators with consistent effects across independent outcomes. Using multi-omics summary-based MR (SMR) as the core analytical engine, we integrated four-layer whole-blood molecular QTL resources eQTL (eQTLGen, n = 31,684), sQTL (GTEx, n = 755), pQTL (INTERVAL + SCALLOP, n = 34,232), and mQTL (McRae et al., n = 1,980) — with GWAS summary statistics for four epigenetic age acceleration measures. At a stringent threshold (P_SMR < 1×10⁻¹²), seven high-confidence effector genes were identified: NHLRC1, TPMT, SELP, and RIPPLY3 for IEAA; ZNF373A and PLDN for HannumAA; and EDARADD for PhenoAA. The chromosome 6p21 NHLRC1–TPMT locus, overwhelmingly driven by methylation QTL signals (−log₁₀P = 26.06), emerged as the dominant genetic node of epigenetic aging. Spatial projection via gsMap onto a mouse E16.5 embryo atlas (121,767 cells) revealed preferential enrichment in smooth muscle and lung, with EDARADD showing marked specificity in mucosal epithelium. Cross-database drug target mining classified TPMT and SELP as repurposable known targets and NHLRC1 as a high-priority novel druggable candidate. This study provides multi-omics convergent causal evidence for inflammation-driven epigenetic aging and delivers genetically anchored targets for precision anti-aging intervention.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s13148-026-02184-z.Keywords: Epigenetic aging, Mendelian randomization, Inflammaging, Spatial transcriptomics, Drug target Introduction Population aging has emerged as a major global public health challenge, with aging itself representing a central risk factor for cardiovascular disease, neurodegenerative disorders, metabolic syndrome, and multiple malignancies [1, 2]. Yet the magnitude of inter-individual variation in biological aging rate is substantial even among individuals of identical chronological age, indicating that genetic, epigenetic, and environmental factors rather than age itself are primary determinants of this heterogeneity [3–5], underscoring the need for precise quantitative tools to capture such variation. Epigenetic clocks have provided a transformative solution: algorithm-based models trained on genome-wide DNA methylation patterns, including the Hannum clock [6], Horvath clock [7], PhenoAge [8], and GrimAge [9], enable high-fidelity estimation of biological age and have demonstrated superior predictive power over chronological age for all-cause mortality, chronic disease incidence, and functional decline across multiple independent cohorts. Epigenetic age acceleration, defined as the degree to which biological age exceeds chronological age, has consequently become a central intermediate phenotype for investigating the biological mechanisms of aging [10]. Chronic low-grade inflammation is among the most reproducible molecular hallmarks of aging, a phenomenon collectively termed inflammaging [11, 12]. With advancing age, circulating levels of pro-inflammatory cytokines including IL-6, TNF, and CXCL10 rise progressively, accelerating biological aging through the promotion of cellular senescence, telomere attrition, and epigenetic drift [13]. Epidemiological studies have established positive associations between inflammatory burden and epigenetic age acceleration at the population level, yet such designs are inherently susceptible to confounding and reverse causation [14, 15]. Mendelian randomization (MR) offers a means to overcome this limitation, and prior studies have applied this approach to evaluate the causal effects of specific inflammatory proteins on cardiovascular and metabolic disease endpoints [16], with some work also exploring associations between inflammatory markers and individual epigenetic clocks [17, 18]. However, these studies have examined a limited range of aging phenotypes, making it impossible to determine whether individual inflammatory proteins exert uniform causal effects across multiple dimensions of biological aging or whether their effects are specific to particular aging measures, leaving the causal genetic landscape of inflammation-driven epigenetic aging incomplete. Tracing causal inflammatory mediators to specific effector genes represents an essential step from genetic association toward mechanistic understanding. Summary-based Mendelian randomization (SMR) enables systematic evaluation of whether genome-wide association study (GWAS) signals are mediated through changes in gene expression, splicing, protein abundance, or DNA methylation by jointly analyzing GWAS and molecular quantitative trait locus (QTL) data [19]. However, existing applications of SMR to aging phenotypes have largely been confined to a single molecular layer, failing to leverage cross-layer convergence of QTL evidence to strengthen effector gene prioritization. Moreover, these studies have typically stopped at statistical association, without systematic extension toward spatial tissue localization of candidate genes or evaluation of their therapeutic tractability. To address these limitations, we built upon MR-based causal screening and deployed multi-omics SMR, systematically integrating four-layer large-scale whole-blood QTL resources: eQTL (eQTLGen [20], n = 31,684), sQTL (GTEx [21], n = 755), pQTL (INTERVAL [22] + SCALLOP [23], n = 34,232), and mQTL (McRae et al. [24], n = 1,980) with GWAS summary statistics for four epigenetic age acceleration measures, leveraging cross-layer evidence convergence to enhance the confidence of effector gene prioritization. Candidate genes were further spatially resolved by projecting genetic signals onto a mouse E16.5 embryo spatial transcriptomic atlas comprising 121,767 single cells via the gsMap framework [25], delineating tissue-specific enrichment patterns within anatomical coordinates. High-confidence candidates were then evaluated for druggability across five major pharmacological databases, directly linking genetic causal inference to translational intervention potential. Together, this framework provides an end-to-end analytical path from genetic signal to effector gene, from spatial localization to drug target, for inflammation-driven epigenetic aging. Materials and methods Mendelian randomization analysis of inflammatory cytokines on aging biomarkers Exposure data Genetic instruments for 91 circulating inflammatory proteins were derived from a genome-wide protein quantitative trait loci (pQTL) study by Zhao et al. [26], involving 14,824 European-ancestry participants. Instrumental variables (IVs) were selected based on the following criteria: (i) SNPs with genome-wide sub-threshold significance (P < 5 × 10⁻⁶) were retained, while those previously associated with aging-related outcomes were excluded to minimize horizontal pleiotropy; (ii) linkage disequilibrium clumping was applied using a 10,000 kb window and r² < 0.001; (iii) SNPs with minor allele frequency < 0.01 were removed; and (iv) instrument strength was evaluated using F-statistics (F = R² × (N − 2) / (1 − R²), where R² = 2 × EAF × (1 − EAF) × β²), with only instruments having F > 10 retained. After applying all selection criteria, the number of SNPs retained per inflammatory protein ranged from 3 to 24 across the 91 proteins. Following harmonization with each outcome, the total numbers of SNP-outcome pairs retained were 1,365 for HannumAA, 1,364 for IEAA, 1,367 for PhenoAA, 974 for GrimAA, 1,372 for the frailty index, and 1,526 for telomere length. Outcome data Four epigenetic age acceleration measures were derived from the large-scale GWAS by McCartney et al. [27], encompassing Hannum age acceleration (HannumAA; n = 34,449), which reflects immune cell composition-dependent systemic aging; intrinsic epigenetic age acceleration (IEAA; n = 34,461), which captures cell-intrinsic aging processes independent of blood cell composition; PhenoAge acceleration (PhenoAA; n = 34,463), a second-generation clock trained on clinical biomarkers with strong predictive capacity for multi-system functional decline; and GrimAge acceleration (GrimAA; n = 34,467), which integrates plasma protein markers and lifestyle factors and currently demonstrates the strongest predictive performance for all-cause mortality and cardiovascular events among available epigenetic clocks [9]. By spanning distinct biological dimensions of epigenetic aging, these four measures collectively constitute a comprehensive phenotypic spectrum for the outcome analyses. Frailty index data were obtained from Atkins et al. (2021), which reports GWAS summary statistics for a continuous deficit-accumulation frailty index derived from 49 self-reported health deficits in 175,226 European-ancestry participants [28] (n = 175,226), and telomere length data were derived from a UK Biobank-based meta-analysis (ieu-b-4879; n = 472,174) [29]. All outcome datasets were restricted to individuals of European ancestry to ensure population consistency with the exposure data. Statistical analysis Two-sample MR analyses were performed using the TwoSampleMR R package (version 4.2). The inverse variance weighted (IVW) method served as the primary approach. Sensitivity analyses included the weighted median method (valid when ≥ 50% of instrumental weight derives from valid instruments), MR-Egger regression (allowing directional pleiotropy assessment via intercept), and simple/weighted mode methods. Heterogeneity was assessed using Cochran’s Q statistic (P < 0.05 indicating significance). Directional pleiotropy was evaluated via the MR-Egger intercept test, with significant results (P < 0.05) interpreted cautiously. Reverse MR analysis was additionally performed to evaluate potential reverse causation. The six aging phenotypes (HannumAA, IEAA, PhenoAA, GrimAA, frailty index, and telomere length) were treated as exposures, and the 91 inflammatory cytokines as outcomes. Instrumental variables for each aging phenotype were selected from their respective GWAS summary statistics using the same criteria as the forward analysis (genome-wide significance P < 5 × 10⁻⁸; LD clumping r² < 0.001; F > 10). Cytokine outcome data were obtained from the EBI GWAS Catalog. The same MR methods (IVW, weighted median, MR-Egger) and sensitivity analyses (Cochran’s Q, MR-Egger intercept) were applied. Functional annotation and enrichment analysis using FUMA To characterize the biological functions and tissue-specific expression patterns of the inflammatory proteins identified through MR analysis as causally associated with the six aging-related outcomes, we performed functional annotation using the FUMA platform (https://fuma.ctglab.nl/) [30]. The candidate genes encoding the MR-significant cytokines were submitted to the GENE2FUNC module for comprehensive functional enrichment analysis. Tissue-specific expression analysis Expression enrichment was assessed across 30 general tissue categories using the GTEx v8 database. The MHC region genes were excluded to reduce potential immunogenetic confounding. Statistical significance was determined using the Benjamini-Hochberg method with a false discovery rate (FDR) threshold of 0.05. Particular attention was given to enrichment patterns in immune-related and hematopoietic tissues, given the inflammatory nature of the candidate proteins, as well as in tissues previously implicated in aging processes. Gene Ontology and pathway enrichment analysis Functional enrichment analyses were conducted using the Molecular Signatures Database (MSigDB). Gene Ontology (GO) enrichment was performed for two categories: Cellular Component (CC) and Molecular Function (MF). Pathway enrichment analysis was carried out using two complementary databases: the Kyoto Encyclopedia of Genes and Genomes (KEGG) and the Reactome Pathway Database. This dual-pathway approach was adopted to maximize pathway coverage, as KEGG provides curated metabolic and signaling pathway annotations while Reactome offers detailed mechanistic representations of biological processes. All enrichment analyses employed hypergeometric tests with Benjamini-Hochberg FDR correction (FDR < 0.05). Multi-omics Summary-based Mendelian Randomization (SMR) analysis Overview To prioritize putative causal genes underlying the genetic associations with aging biomarkers, we performed multi-omics Summary-based Mendelian Randomization (SMR) analyses [19]. SMR leverages SNPs as instrumental variables to test whether the effect of a SNP on a trait is mediated through its effect on an intermediate molecular phenotype, thereby integrating GWAS summary statistics with molecular quantitative trait locus (QTL) data to infer causal relationships between molecular exposures and complex traits. QTL data sources The SMR analyses were conducted by integrating whole-blood QTL resources from multiple large-scale cohorts spanning four molecular layers. Expression QTLs (eQTLs) were derived from the eQTLGen consortium meta-analysis (n = 31,684) [20], which provides comprehensive genome-wide cis-eQTL summary statistics for blood-expressed genes. Splicing QTLs (sQTLs) were obtained from the Genotype-Tissue Expression (GTEx) project (n = 755) [21], capturing genetic effects on RNA splicing variation. Protein QTLs (pQTLs) were sourced from two complementary resources: the INTERVAL study (n = 3,301) [22] and the SCALLOP consortium (n = 30,931) [23], which together provide extensive coverage of circulating protein abundance variation. Methylation QTLs (mQTLs) were obtained from McRae et al. (n = 1,980) [24], representing genetic effects on DNA methylation levels at CpG sites across the genome. All QTL datasets comprised genome-wide associations between SNPs and their corresponding molecular phenotypes. GWAS summary statistics for aging phenotypes GWAS summary statistics for the four epigenetic age acceleration measures (HannumAA, IEAA, PhenoAA, and GrimAA) were as described in Sect. 2.1. SMR and HEIDI analysis The SMR analysis was run comprehensively against all molecular phenotypes (gene expression, splicing, protein abundance, and DNA methylation) available within the genome-wide QTL datasets. The minor allele frequency (MAF) threshold for instrumental variables was set at 0.01. For each identified association, we reported the SMR effect size (b_SMR) and the corresponding P-value (P_SMR). The significance threshold was defined based on Bonferroni correction accounting for the number of probes tested within each molecular layer. To distinguish genuine causal or pleiotropic associations from confounding due to linkage disequilibrium (LD), we applied the Heterogeneity in Dependent Instruments (HEIDI) test. The HEIDI test evaluates whether the observed association pattern across multiple SNPs at a given locus is consistent with a single causal variant model. Associations with P_HEIDI < 0.05 were considered to exhibit significant heterogeneity, suggesting that the observed signal might arise from LD between two distinct causal variants rather than a shared causal mechanism, and such signals were excluded from further causal interpretation. Multi-omics SMR Manhattan plot visualization For visualization, Manhattan-style plots were constructed for each age acceleration measure. The maximum − log₁₀(P_SMR) across all four omics layers (eSMR, sSMR, pSMR, and mSMR) was assigned as the representative value for each gene. Three significance thresholds were displayed: highly significant (P < 1 × 10⁻¹²), significant (P < 1 × 10⁻¹⁰), and suggestive (P < 1 × 10⁻⁸). Gene labels were annotated for loci reaching − log₁₀P ≥ 7, with text positions optimized using the adjustText algorithm. All plots were generated using Python (v3.10) with matplotlib. Drug target mining for SMR highly significant genes To explore the translational potential of the prioritized causal genes, we performed systematic drug target mining for the seven genes reaching the highly significant threshold (P_SMR < 1 × 10⁻¹²) in the multi-omics SMR analysis: NHLRC1, TPMT, SELP, RIPPLY3, ZNF373A, PLDN (BLOC1S6), and EDARADD. Drug-gene interactions were comprehensively queried across five pharmacogenomics and drug-target databases. The Drug-Gene Interaction Database (DGIdb v4.0; https://www.dgidb.org/) was used as the primary resource to identify known drug-gene interactions, providing interaction scores that reflect the strength and confidence of each association based on curated literature, clinical annotations, and pathway data. The Therapeutic Target Database (TTD; http://db.idrblab.net/ttd/) was queried to retrieve additional drug candidates, clinical development stages, and therapeutic indications. The Open Targets Platform (https://platform.opentargets.org/) was used to cross-reference genetic evidence with known drug mechanisms and tractability assessments. DrugBank (v5.1; https://go.drugbank.com/) was consulted for detailed pharmacological information including drug classification, mechanism of action, and regulatory approval status. ChEMBL (https://www.ebi.ac.uk/chembl/) was searched for bioactivity data and preclinical compound information for targets lacking approved therapeutics. For each gene, we assessed: (i) whether the encoded protein is a known drug target with approved or investigational therapeutics in any of the queried databases; (ii) the clinical development stage of associated compounds (approved, Phase I–III, or investigational); (iii) the mechanism of action; and (iv) the primary therapeutic indication. Based on the aggregated query results, genes were classified into two druggability tiers: “Known Target” for genes with at least one drug-gene interaction documented across any of the five databases, and “Novel Target” for genes with no existing drug-gene interaction records in any database, representing potential opportunities for future pharmacological development. Drug pipeline visualization and druggability classification were generated using Python (v3.10) with matplotlib. Disease and phenotype enrichment analysis of SMR highly significant genes To investigate the disease relevance of the seven SMR highly significant genes (P_SMR < 1 × 10⁻¹²), we performed disease and phenotype enrichment analysis using the Enrichr platform (https://maayanlab.cloud/Enrichr/) [31]. The seven candidate genes (NHLRC1, TPMT, SELP, RIPPLY3, ZNF373A, PLDN, and EDARADD) were submitted as a gene set to the Enrichr web server. Enrichment was assessed against the GWAS Catalog 2025 gene-disease association library within the Diseases/Drugs module. This library integrates curated gene-trait associations from the NHGRI-EBI GWAS Catalog, enabling systematic evaluation of whether the candidate gene set is over-represented among genes previously implicated in specific diseases or phenotypes through genome-wide association studies. Statistical significance was determined using Fisher’s exact test, with multiple testing correction applied via the Benjamini-Hochberg method. Enrichment results were reported as odds ratio, combined score (computed as the log of the p-value multiplied by the z-score of the deviation from the expected rank), and adjusted p-value. Spatial mapping of genetically associated loci using gsMap To investigate the spatial expression patterns and tissue-specific distributions of genes prioritized through multi-omics SMR analysis, we employed the gsMap [25] (genetically informed spatial mapping of cells for complex traits) framework to project genetic associations of aging biomarkers onto spatial transcriptomic data. Spatial transcriptomic profiles were obtained from the Mouse Organogenesis Spatiotemporal Transcriptomic Atlas (MOSTA) [32], specifically the E16.5 mouse embryo sagittal section (E16.5_E1S1.MOSTA), which contains 121,767 spatially resolved cells across 25 annotated tissue types. Spatial coordinates and cell-type annotations were extracted from the original h5ad files. For the genetic component, summary statistics for Hannum age acceleration (HannumAA; GCST90014289; ) were reformatted to meet gsMap requirements. Cross-species integration was performed by mapping human candidate genes identified in the SMR analysis to their mouse orthologs using the human-mouse reference database provided by gsMap. Only genes with identifiable mouse orthologs were retained for downstream spatial analysis. The spatial linkage disequilibrium score regression (spatial-LDSC) method was integrated to estimate the genetic correlation between trait-associated variants and spatially resolved gene expression profiles. For each individual cell, the effect size (beta), standard error (SE), Z-score, and association P-value were calculated. To evaluate associations at the cell-type level, we applied the Cauchy combination test, which aggregates P-values from all cells within a specific tissue annotation while accounting for underlying correlation structures. The direction of association was assessed using the mean Z-score. Spatial distribution visualizations were generated using the matplotlib and scanpy packages in Python (v3.10). Statistical significance was defined as PCauchy < 0.05. Cell types with PCauchy < 0.10 are reported as nominally suggestive and those with 0.10 ≤ PCauchy < 0.15 as weak trend (Fig. 1). Schematic overview of the study design and analytical workflow Outcome Bidirectional Mendelian randomization analysis of inflammatory cytokines and aging biomarkers Two-sample MR analyses identified multiple causal associations between circulating inflammatory proteins and aging-related phenotypes across six outcomes (IVW method, P < 0.05; Fig. 2A–G). Cross-phenotype analysis revealed that IL-12B, IFNG, and IL-2 consistently promoted biological aging across independent measures, identifying these cytokines as putative central mediators linking chronic inflammation to accelerated aging. Mendelian randomization analysis of circulating inflammatory cytokines on aging biomarkers. A–F Volcano plots showing inverse variance weighted (IVW) Mendelian randomization estimates for the effects of 91 circulating inflammatory proteins on six aging-related outcomes: PhenoAge acceleration (PhenoAA; A), GrimAge acceleration (GrimAA; B), Hannum age acceleration (HannumAA; C), intrinsic epigenetic age acceleration (IEAA; D), frailty index (E), and telomere length (F). G Forest plots summarizing the significant IVW Mendelian randomization associations (P < 0.05). Abbreviations: IVW, inverse variance weighted; PhenoAA, PhenoAge acceleration; GrimAA, GrimAge acceleration; HannumAA, Hannum age acceleration; IEAA, intrinsic epigenetic age acceleration IL-12B exhibited the broadest cross-phenotype pro-aging profile, with significant associations across PhenoAge acceleration (β = 0.307, P = 0.006), GrimAge acceleration (β = 0.239, P = 0.026), and telomere shortening (β = −0.032, P = 0.009), establishing it as the most robust inflammatory mediator of biological aging in this analysis. IFNG showed concordant pro-aging effects across two epigenetic clocks, accelerating both GrimAge (β = 0.427, P = 0.036) and Hannum epigenetic age (β = 0.358, P = 0.045). IL-2 similarly drove acceleration of both PhenoAge (β = 0.442, P = 0.019) and Hannum age (β = 0.301, P = 0.036), further implicating T-cell activation signaling in epigenetic aging (Fig. 2G). Among phenotype-specific associations, PhenoAge acceleration showed the broadest inflammatory landscape (n = 11 cytokines; Fig. 2A), with Oncostatin M exhibiting the largest effect size (β = 0.620, 95% CI: 0.243–0.998, P = 1.29 × 10⁻³), suggesting that JAK-STAT pathway activation directly contributes to epigenetic age acceleration. For GrimAge acceleration, IL-17 C emerged as the strongest risk factor (β = 0.355, P = 0.010), while FLT3LG was the sole protective cytokine (β = −0.414, P = 0.025; Fig. 2B), pointing to a potential role of hematopoietic progenitor signaling in counteracting inflammation-driven aging. Telomere length analysis provided structural evidence for inflammation-driven cellular aging (Fig. 2F). CXCL10 promoted both telomere shortening (β = −0.049, P = 0.004) and frailty (β = 0.026, P = 0.010), suggesting it may mediate multi-dimensional aging phenotypes through accelerated telomere attrition. IL-1 A exerted the strongest telomere-shortening effect (β = −0.022, P = 0.002), consistent with its established role in activating the NLRP3 inflammasome and DNA damage responses. Associations with the frailty index (Fig. 2E) were nominally significant but with notably smaller effect sizes, likely reflecting the polygenic and multifactorial nature of this composite phenotype. Sensitivity analyses for the 34 headline associations are presented in Supplementary Table S5. Most associations showed no significant heterogeneity or directional pleiotropy, supporting the robustness of the primary IVW estimates. Three associations, namely telomere length with CXCL10, telomere length with IL-12B, and GrimAA with FLT3LG, exhibited significant heterogeneity, suggesting potential violations of the homogeneity assumption that warrant cautious interpretation. Significant directional pleiotropy was detected for frailty index with TNF-β and telomere length with TNF-β, indicating that these associations may be subject to horizontal pleiotropic effects and should be interpreted with caution. Reverse MR analysis, treating the six aging phenotypes as exposures and 91 cytokines as outcomes, revealed widespread bidirectional relationships. The number of FDR-significant cytokine associations varied substantially across aging exposures, ranging from 19 out of 91 for IEAA to 65 out of 91 for HannumAA, with GrimAA (48/91), telomere length (64/91), and frailty index (62/91) also showing broad effects. These findings indicate that epigenetic aging and related phenotypes exert pervasive feedback effects on the inflammatory proteome, consistent with a mutually reinforcing inflammaging cycle. Detailed reverse MR results are provided in Supplementary Figures S1–S7. Functional annotation and enrichment analysis of MR-significant inflammatory cytokines Tissue-specific expression analysis (GTEx v8) showed that MR-significant cytokine-encoding genes were most enriched in lung, spleen, and whole blood, predominantly driven by up-regulated DEGs (Fig. 3A), consistent with systemic immune activation as a primary driver of inflammaging. Pathway enrichment converged on cytokine–cytokine receptor interaction, JAK-STAT signaling, and interleukin signaling across both KEGG and Reactome databases (Fig. 3B), collectively implicating the senescence-associated secretory phenotype (SASP) axis as the mechanistic link between chronic inflammation and accelerated biological aging. Functional enrichment and tissue-specific expression analysis of MR-significant inflammatory cytokines. A Tissue-specific expression enrichment (GTEx v8) of MR-significant genes across 30 tissues, plotted as − log₁₀(P-value). B Pathway enrichment analysis of MR-significant genes using Reactome (top) and KEGG (bottom) databases. C Gene Ontology (GO) Molecular Function (MF) enrichment analysis of MR-significant genes. D Gene Ontology (GO) Cellular Component (CC) enrichment analysis of MR-significant genes. Abbreviations: MR, Mendelian randomization; GTEx, Genotype-Tissue Expression; KEGG, Kyoto Encyclopedia of Genes and Genomes; GO, Gene Ontology; MF, Molecular Function; CC, Cellular Component At the molecular level, GO enrichment confirmed that these proteins function primarily as secreted extracellular ligands, with MF terms dominated by cytokine activity and receptor binding (Fig. 3C), and CC terms restricted to the cell surface and external plasma membrane (Fig. 3D). Together, these findings delineate a coherent functional profile in which the causal inflammatory mediators of aging operate through surface receptor-driven transcriptional programs rather than intracellular mechanisms. Multi-omics SMR analysis identifies genes associated with epigenetic age acceleration Multi-omics SMR analysis, integrating expression, splicing, protein, and methylation QTL data, identified candidate causal genes across four epigenetic age acceleration measures (Fig. 4). IEAA yielded the largest number of associations (44 genes, 22 loci), followed by HannumAA (18 genes, 10 loci) and GrimAA (18 genes, 10 loci), while PhenoAA showed the fewest signals (14 genes, 10 loci). Multi-omics SMR analysis identifies genes associated with epigenetic age acceleration measures. A–D Manhattan plots of multi-omics summary-based Mendelian randomization (SMR) results for four epigenetic age acceleration outcomes: GrimAge acceleration (GrimAA; A), Hannum age acceleration (HannumAA; B), intrinsic epigenetic age acceleration (IEAA; C), and PhenoAge acceleration (PhenoAA; D). The y-axis represents − log₁₀(P_SMR), and the x-axis represents chromosome position. Three significance thresholds are indicated: highly significant (P < 1 × 10⁻¹², red), significant (P < 1 × 10⁻¹⁰, pink), and suggestive (P < 1 × 10⁻⁸, orange). Abbreviations: SMR, summary-based Mendelian randomization; GrimAA, GrimAge acceleration; HannumAA, Hannum age acceleration; IEAA, intrinsic epigenetic age acceleration; PhenoAA, PhenoAge acceleration The chromosome 6p21 locus anchored by NHLRC1 and TPMT emerged as the most consistently implicated region, reaching genome-wide significance in both IEAA and PhenoAA, and representing the strongest genetic link to epigenetic aging identified in this analysis. Applying a stringent threshold of P_SMR < 1 × 10⁻¹², seven high-confidence candidate genes were prioritized: NHLRC1, TPMT, SELP, and RIPPLY3 from IEAA; ZNF373A and PLDN from HannumAA; and EDARADD from PhenoAA collectively forming the core gene set for downstream functional annotation and drug target mining. At the chromosome 6p21 locus, NHLRC1 attained the highest SMR significance across all outcomes (IEAA: −log₁₀PSMR = 26.06; PhenoAA: 14.13), driven primarily by methylation QTL evidence, implicating epigenetic regulation of this E3 ubiquitin ligase as a key molecular mechanism underlying intrinsic epigenetic aging. TPMT at the same locus showed convergent multi-omics support across expression, splicing, and methylation layers (eSMR = 18.47, sSMR = 21.72 in IEAA; 14.11 in PhenoAA), strengthening the causal inference and suggesting that both proteasomal and metabolic pathways at this locus jointly contribute to epigenetic age acceleration (Fig. 4C and D). Beyond the 6p21 locus, SELP (IEAA: −log₁₀PSMR = 14.63) and RIPPLY3 (− log₁₀PSMR = 14.55) on chromosome 1 reached highly significant thresholds in IEAA, both driven by methylation QTL signals (Fig. 4C). For HannumAA, ZNF373A and PLDN on chromosome 10 were the lead signals, with ZNF373A supported by convergent evidence across methylation and splicing layers (Fig. 4B). EDARADD on chromosome 1 reached highly significant threshold specifically in PhenoAA (− log₁₀PSMR = 12.07; Fig. 4D), consistent with its epithelial expression pattern identified in the spatial mapping analysis. Two cross-cutting features warrant emphasis. First, methylation QTL-driven signals (mSMR) contributed disproportionately across all outcomes, suggesting that DNA methylation — itself the molecular substrate of epigenetic clocks mediates a substantial share of the genetic effects on epigenetic aging. Second, genes with convergent evidence across multiple QTL layers (TPMT: eSMR + sSMR; ZNF373A: eSMR + sSMR + mSMR) consistently ranked among the top associations, underscoring multi-omics convergence as a reliable criterion for prioritizing high-confidence causal candidates. Drug target mining of SMR highly significant genes Among the seven genes reaching PSMR < 1 × 10⁻¹², cross-database querying across five resources (DGIdb, TTD, Open Targets, DrugBank, and ChEMBL) classified three as established drug targets and four as novel targets (Fig. 5A). Druggability and disease relevance of SMR-prioritized genes. A Bar plot showing the maximum −log₁₀(P_SMR) for the seven prioritized genes (NHLRC1, TPMT, SELP, RIPPLY3, ZNF373A, PLDN, EDARADD). Red bars indicate known drug targets; blue bars indicate novel targets. B Summary of drug–gene interactions for the prioritized genes, including approved and investigational drugs, their mechanisms of action, and data sources. C Donut plot illustrating druggability classification (known vs. novel targets) and bar plot of top disease/phenotype enrichment results for the seven genes. D Table of top 10 enriched phenotypes from disease enrichment analysis, including P-values, adjusted P-values, odds ratios, and combined scores. Abbreviations: SMR, summary-based Mendelian randomization; NHLRC1, NHL repeat containing E3 ubiquitin protein ligase 1; TPMT, thiopurine S-methyltransferase; SELP, selectin P; RIPPLY3, ripply transcriptional repressor 3; ZNF373A, zinc finger protein 373A; PLDN, pallidin; EDARADD, EDAR associated death domain. Among established targets, TPMT (max − log₁₀P = 21.72 in IEAA) possessed the most extensively characterized drug interaction profile (Fig. 5B). Three thiopurine immunosuppressants mercaptopurine, thioguanine, and azathioprine are directly metabolized by TPMT, with pharmacogenomic testing already integrated into clinical practice, suggesting that the TPMT metabolic axis represents a tractable entry point for epigenetic aging intervention. SELP (max − log₁₀P = 14.63 in IEAA) harbored the most diverse clinical pipeline, with the approved anti-thrombotic crizanlizumab and multiple investigational anti-inflammatory agents targeting this gene; its genetic association with epigenetic aging provides a clear translational rationale for repurposing. Among novel targets, NHLRC1 carried the highest SMR significance (− log₁₀P = 26.06) and encodes an E3 ubiquitin ligase involved in protein quality control and autophagy — both central mechanisms of aging biology — with Open Targets assigning it medium-to-high druggability. Although no direct pharmacological tools exist for EDARADD (− log₁₀P = 12.07), the EDA-A1 receptor agonist EDI200, targeting the same signaling pathway, is currently in Phase II clinical trials, offering an indirect intervention strategy. RIPPLY3 and ZNF373A currently lack pharmacological tools and await further structural characterization. Disease enrichment analysis confirms aging and vascular relevance Enrichr analysis against the GWAS Catalog 2025 library revealed significant enrichment of the seven SMR highly significant genes across multiple disease and phenotype categories (Fig. 5C & D). The most significant enrichment was observed for intrinsic epigenetic age acceleration (adjusted P = 0.018, odds ratio = 555.19, combined score = 3,338.15), providing independent validation that the identified gene set is enriched among GWAS-implicated loci for the same aging phenotype examined in our SMR analysis. Total ventricular volume ranked second (adjusted P = 0.018, OR = 475.86), followed by L-selectin levels (adjusted P = 0.026, OR = 208.09), consistent with the identification of SELP as a causal gene for IEAA. Hippocampal atrophy (adjusted P = 0.026, OR = 114.74) and late-onset Alzheimer’s disease (adjusted P = 0.030, OR = 73.88) were also significantly enriched, suggesting potential links between the prioritized aging genes and neurodegenerative processes. Additional significant enrichments included urate levels with BMI interaction (adjusted P = 0.026), activated partial thromboplastin time (adjusted P = 0.026), and autoimmune thyroid disease (adjusted P = 0.041), reflecting the involvement of these genes in metabolic, coagulation, and autoimmune pathways. Optic disc size reached nominal significance (adjusted P = 0.047). Birth weight showed suggestive enrichment (adjusted P = 0.109) but did not reach significance after correction. These enrichment results converge on three biological themes: (i) epigenetic aging and neurodegeneration, (ii) vascular and coagulation biology, and (iii) immune and metabolic regulation, consistent with the multi-omics SMR and MR findings linking inflammatory pathways to biological aging. Spatial mapping of genetically associated loci using gsMap Spatial visualization of HannumAA-associated variants (Fig. 6B) revealed a heterogeneous distribution across the tissue section, with the strongest signals (− log₁₀ P > 3.5) concentrated in peripheral connective and epithelial regions, while the central nervous system displayed comparatively weaker associations. Spatial mapping of HannumAA genetic associations using gsMap. A Spatial distribution of 25 annotated cell types in the E16.5 mouse embryo sagittal section (MOSTA dataset). B Spatial distribution of − log₁₀(P-values) for HannumAA genetic associations across the mouse embryo. C Scatter plot of spatial-LDSC effect size (beta) versus Z-score, colored by − log₁₀(P-value). D Spatial expression patterns of five prioritized target genes (NHLRC1, TPMT, SELP, RIPPLY3, EDARADD) in the mouse embryo. E Heatmap of target gene expression across different cell types, with color intensity representing mean expression level. Abbreviations: HannumAA, Hannum age acceleration; gsMap, genetically informed spatial mapping; MOSTA, Mouse Organogenesis Spatiotemporal Transcriptomic Atlas; LDSC, linkage disequilibrium score regression; NHLRC1, NHL repeat containing E3 ubiquitin protein ligase 1; TPMT, thiopurine S-methyltransferase; SELP, selectin P; RIPPLY3, ripply transcriptional repressor 3; EDARADD, EDAR associated death domain In the spatial-LDSC analysis, a strong positive correlation between beta and Z-score was observed across cells (Fig. 6C), indicating directional consistency of genetic effects. Cauchy combination testing (Table 1) identified smooth muscle, lung, and choroid plexus as nominally suggestive (PCauchy < 0.10) with positive enrichment, while liver, GI tract, and kidney showed weak positive trends (0.10 ≤ PCauchy < 0.15); none of these associations reached statistical significance. Conversely, dorsal root ganglion, spinal cord, and brain exhibited negative associations, likely reflecting an inherent divergence between the transcriptional programs of rapidly developing neural tissues at E16.5 and the genetic architecture of epigenetic aging. Table 1Cell-type-level genetic enrichment for HannumAA using spatial-LDSC and Cauchy combination testCell TypeP_CauchyMean Z-scoreMean BetaDirectionInterpretationSmooth muscle0.0680.974.09 × 10⁻⁹PositiveNominally suggestiveLung0.0751.003.74 × 10⁻⁹PositiveNominally suggestiveChoroid plexus0.0840.762.86 × 10⁻⁹PositiveNominally suggestiveLiver0.1020.653.63 × 10⁻⁹PositiveWeak trendGI tract0.1180.702.76 × 10⁻⁹PositiveWeak trendKidney0.1300.672.50 × 10⁻⁹PositiveWeak trendOnly cell types with P_Cauchy < 0.15 are shown. None of the associations reached statistical significance (P < 0.05). Nominally suggestive: P < 0.10; Weak trend: 0.10 ≤ P < 0.15 Cell-type-level genetic enrichment for HannumAA using spatial-LDSC and Cauchy combination test Only cell types with P_Cauchy < 0.15 are shown. None of the associations reached statistical significance (P < 0.05). Nominally suggestive: P < 0.10; Weak trend: 0.10 ≤ P < 0.15 Spatial expression analysis of the five candidate genes (Fig. 6D and E) revealed distinct tissue-enrichment profiles. EDARADD showed the most pronounced specificity in mucosal epithelium, consistent with the epithelial role of the EDA signaling pathway; RIPPLY3 exhibited focal high expression in lung, aligning with the positive LDSC enrichment in that tissue; TPMT was broadly expressed across cell types; NHLRC1 was enriched in neural tissues; and SELP showed uniformly low expression throughout. Discussion By deploying a four-tier analytical framework integrating two-sample MR causal screening, multi-omics SMR effector gene mapping, gsMap spatial transcriptomic projection, and cross-database drug target evaluation, this study systematically delineated the causal genetic architecture of inflammation-driven epigenetic aging and traced genetic signals to spatially tissue-specific effector genes and druggable targets. Causal pro-aging effects of inflammatory cytokines MR analysis established IL-12B, IFNG, and IL-2 as the most consistently significant pro-aging mediators across independent epigenetic aging measures, corroborating the directionality of prior observational evidence while providing genetic causal support. IL-12B showed significant associations with PhenoAge acceleration, GrimAge acceleration, and telomere shortening, suggesting that its pro-aging influence may span both epigenetic and structural dimensions of biological aging. An independent line of evidence from Tian et al. identified IL-12B as a key mediating inflammatory factor linking the gut microbial genus Prevotella 7 to GrimAge acceleration [17], further substantiating its upstream role in systemic inflammaging. Sustained pathological expression of IL-12B (p40) chronically engages the JAK-STAT4 signaling axis, driving NK cells and effector T cells into a state of persistent hyperactivation; this sustained immune overactivation not only promotes chronic high-level IFN-γ release [33] but also amplifies tissue oxidative stress and DNA damage, establishing it as a central upstream driver of the senescence-associated secretory phenotype (SASP) [34–36]. The clinical availability of ustekinumab, a monoclonal antibody targeting IL-12B (p40) that suppresses NK/T cell hyperactivation and is approved for psoriasis and Crohn’s disease, provides direct evidence for the therapeutic tractability of this target. The effects of IFNG and IL-2 were relatively concentrated on the Hannum and GrimAge clocks the former reflecting extrinsic immune cell composition-dependent aging and the latter integrating mortality-associated molecular markers — jointly implicating T cell-mediated adaptive immune activation as a key upstream driver of inflammaging. The capacity of IFN-γ and IL-2 to cooperatively induce epigenetic reprogramming provides a mechanistic link between these cytokines and clock-based aging readouts [37]. Notably, JAK-STAT signaling emerged as the central downstream pathway shared across all three cytokines and ranked among the most significantly enriched pathways in functional annotation. This converges with existing intervention evidence: ruxolitinib has been shown to potently suppress the SASP and reverse aging-related physical decline in vivo [38], positioning JAK inhibitors as a cornerstone class of senomorphics [39]; targeting this pathway also attenuates tau pathology and cognitive decline driven by senescent glia in the central nervous system [40]. These genetic findings thus provide upstream causal support for the rationale of JAK-STAT-directed interventions in epigenetic aging, suggesting that inflammatory activation of this pathway represents a causal contributor to, rather than a mere correlate of, accelerated biological aging. The chromosome 6p21 NHLRC1–TPMT locus as the dominant genetic node Multi-omics SMR analysis identified the chromosome 6p21 NHLRC1–TPMT locus as the strongest genetic node of epigenetic aging, with signals reaching extreme significance in both IEAA and PhenoAA and driven predominantly by methylation QTL evidence. The predominance of mQTL signals indicates that the influence of this locus on epigenetic aging is mediated primarily through DNA methylation regulation rather than changes in gene expression levels alone, directly consistent with the molecular substrate of epigenetic clocks. NHLRC1 encodes the E3 ubiquitin ligase malin, which participates in ubiquitin-mediated protein degradation and autophagy regulation both recognized as central mechanisms of aging biology and dysregulation of protein quality control systems is considered a key driver of cellular senescence and tissue functional decline [41, 42]. TPMT received convergent support across expression, splicing, and methylation QTL layers at the same locus, indicating that its effects involve multiple levels of transcriptional regulation and RNA processing. TPMT encodes thiopurine S-methyltransferase, a key enzyme in purine metabolism [43]; purine metabolic dysregulation is closely associated with oxidative stress and mitochondrial dysfunction, both classical mechanisms of aging biology [44], providing a metabolically grounded rationale for TPMT involvement in epigenetic aging regulation. The convergence of such strong cross-phenotype genetic signals from two functionally distinct genes at the same locus suggests the existence of a shared regulatory mechanism at 6p21 influencing multiple aging-related molecular pathways, warranting focused functional investigation. Biological characterization and spatial tissue localization of effector genes SELP encodes P-selectin, a key mediator of leukocyte-endothelial adhesion with central roles in inflammatory recruitment and thrombosis [45], and its strong genetic association with IEAA implicates the vascular inflammation–endothelial function axis in the causal gene network of epigenetic aging. The significant enrichment for activated partial thromboplastin time in disease enrichment analysis further supports this vascular biology theme. EDARADD, as a critical adaptor protein in the EDA signaling pathway [46], showed specific association with PhenoAA — the clock with the strongest predictive capacity for multi-system functional decline and chronic disease risk suggesting that epithelial developmental and homeostatic pathways may contribute to phenotypic aging through effects on multi-system functional integration. Spatial analysis via gsMap revealed marked specificity of EDARADD enrichment in mucosal epithelium (Fig. 6E), consistent with the established role of EDA signaling in epithelial tissue development and homeostasis, providing tissue-contextual interpretation for its genetic effect on PhenoAA. The focal high expression of RIPPLY3 in lung tissue was corroborated by positive spatial LDSC enrichment in that tissue, suggesting a potential role for the lung as an aging-sensitive organ in the genetic regulation of IEAA, although the underlying mechanism remains to be elucidated. The marginally significant positive enrichment of smooth muscle in Cauchy combination testing is consistent with the established role of vascular smooth muscle cells in inflammaging and arteriosclerosis, offering a vascular biological perspective on the tissue origin of these genetic signals. Translational implications of drug target mining The druggability of TPMT is well established, with three thiopurine drugs in routine clinical use and pharmacogenomic testing integrated into standard practice. The independent genetic causal evidence provided here from an epigenetic aging perspective suggests that monitoring TPMT activity may carry clinical relevance for assessing epigenetic aging risk, and offers a genetic reinterpretation of the aging-related effects of thiopurine drugs. SELP has a diversified drug pipeline, with the approved antithrombotic crizanlizumab targeting P-selectin and several investigational anti-inflammatory agents directed at this target [47]; the genetic causal evidence from this study extends SELP from a vascular disease target to a candidate for epigenetic aging intervention, providing genetic support for exploring potential aging indications. NHLRC1, the gene with the highest SMR significance, currently lacks direct pharmacological tools, but the ubiquitin-proteasome system and autophagy pathways in which its encoded protein participates both have active drug development pipelines [48], and the medium-to-high druggability rating from Open Targets indicates structural tractability; development of targeted tools against NHLRC1, or indirect modulation of its pathway via autophagy activators, may represent a new direction for epigenetic aging intervention. The EDA pathway receptor agonist ER-004, targeting the same signaling axis as EDARADD, is currently in Phase II clinical trials, providing an existing clinical development platform for indirect intervention in this pathway [49]. Limitations Both MR and SMR analyses were based on GWAS summary statistics from European-ancestry populations, and the generalizability of the identified genetic effects to other ancestries requires independent validation; population-specific linkage disequilibrium structures may also influence instrumental variable selection and the accuracy of SMR inference. The mQTL dataset used in SMR analysis (n = 1,980) was relatively limited in sample size; although methylation QTL signals contributed most prominently in this study, the smaller reference panel may have constrained statistical power, potentially leaving some true effector genes undetected. The spatial mapping analysis relied on a mouse E16.5 embryo atlas, and projecting adult human aging-related genetic signals onto a developmental model introduces dual heterogeneity across species and developmental stage; the tissue enrichment patterns revealed here require further validation against human adult tissue transcriptomic data. Drug target evaluation was based on literature integration from existing databases, and the druggability assessments for candidate genes await empirical support from in vitro functional studies and animal model experiments. Finally, the MR framework assumes the absence of horizontal pleiotropy between inflammatory proteins and aging phenotypes; although this study systematically evaluated this assumption through multiple sensitivity analyses, definitively excluding this possibility remains challenging against a complex pleiotropic genetic background. Although potential sample overlap between the exposure pQTL cohort and the McCartney et al. GWAS cannot be fully excluded, any resulting bias in two-sample MR acts toward the null rather than toward the confounded estimate, and would be minimal given that all retained instruments exceeded the F-statistic threshold of 10 in our analysis. Conclusion By deploying multi-omics SMR integrated with four-layer large-scale whole-blood QTL resources as the core inferential engine, this study systematically delineated the effector gene network underlying inflammation-driven epigenetic aging and extended genetic findings toward translational application through spatial transcriptomic localization and drug target evaluation. The chromosome 6p21 NHLRC1–TPMT locus showed the strongest cross-phenotype genetic signal identified in this analysis, with TPMT, SELP, and NHLRC1 representing genetically prioritized candidates for future targeted intervention studies. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Supplementary Material 6. Supplementary Material 7. Acknowledgements The authors extend their gratitude to all researchers who have generously provided the GWAS summary statistics that supported this study. Authors’ contributions **Chun Zhang: ** Conceptualization, Investigation, Formal analysis, Data curation, Validation, Visualization, Writing - original draft, Writing - review & editing.**Jingqi Zhang: ** Conceptualization, Supervision, Writing - review & editing.Chun Zhang and Jingqi Zhang contributed equally to this work. Funding This work did not receive any funding. Data availability All analysis code is available on GitHub at: https://github.com/zhngjingqi-design/MR_Aging_Multiomics. Full SMR summary statistics are publicly available at Zenodo: https://zenodo.org/records/20108780 (https://doi.org/10.5281/zenodo.20108780). Declarations Ethics approval and consent to participate The present study only utilized publicly available summary-level data derived from genome-wide association studies, which did not involve individual participant data. Therefore, separate ethics approval is not required. The ethical approvals associated with the original GWAS datasets are available in the respective primary publications. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Chun Zhang and Jingqi Zhang contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Supplementary Material 6. Supplementary Material 7. Data Availability Statement All analysis code is available on GitHub at: https://github.com/zhngjingqi-design/MR_Aging_Multiomics. Full SMR summary statistics are publicly available at Zenodo: https://zenodo.org/records/20108780 (https://doi.org/10.5281/zenodo.20108780).