A genome-wide association study identifies Asian-specific genetic susceptibility for epigenetic age acceleration

Abstract

Genome-wide association studies (GWASs) have been performed on multiple epigenetic clocks to understand the genetic underpinnings of epigenetic age acceleration (EAA). To explore ethnic differences and individual variability in Asian populations, we conducted a GWAS of EAA using multiple epigenetic clocks in the Korean population. Genome-wide single-nucleotide polymorphism (SNP) and methylation data were obtained from 1962 Korean individuals. Five epigenetic clocks (Hannum, DNAm PhenoAge, GrimAge, Zhang2019, and Horvath-Skin&Blood) were considered. GWASs, as well as gene-based, pathway enrichment, quantitative real-time polymerase chain reaction, and quantitative trait loci analyses, were performed to investigate the biological mechanisms. Pleiotropy analyses and drug repurposing were used to explore clinical implications. Six SNPs were independently associated with EAAs (P < 5 × 10⁻⁸). The DNAm PhenoAge-EAA was associated with ASPA and SPATA22 (P < 2.66 × 10⁻⁶), involved in Canavan disease and apoptosis, and the Zhang2019 was associated with ENO2, GNB3, USP5, TPI1, and CDCA3 (P < 2.66 × 10⁻⁶), which are implicated in monosaccharide catabolism. Particularly, the expression levels of SPATA22, ENO2, USP5, and TPI1 increased in senescent cells. EAA-associated genes showed strong affinities with GW0742, a peroxisome proliferator-activated receptor-β/δ (PPAR-β/δ) agonist. In addition, the Zhang2019-EAA-associated SNP showed pleiotropy in liver function biomarkers, including iron and bilirubin. We identified six Asian-specific EAA-associated genetic variants, including five novel SNPs. The novel SNPs, which are rare in European populations, could implicate ancestry-related genetic differences in epigenetic aging. Furthermore, the genetic implications of the DNAm PhenoAge-EAA and Zhang2019-EAA on apoptosis and liver function may contribute to a better understanding of EAA in East Asian populations.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-025-01885-2.

Introduction

Globally, people live longer, and the population is aging. The United Nations Population Division estimates that 900 million people worldwide will be aged 60 years or older by 2050 [1]. As the number of elderly individuals increases, extensive research on the aging process and age-related diseases is required. One factor that hinders aging research is that, while chronological aging occurs at a consistent rate for everyone, the risk of age-related diseases and mortality varies [2], as aging is a highly multifaceted process involving physical, physiological, and biological changes within the body [3, 4]. Although aging cannot be fully explained by a single metric [5], there have been a lot of efforts to identify and quantify the epigenetic aspects of aging. Accordingly, a diversity of epigenetic clocks designed to estimate epigenetic age (EA) has been developed based on multiple DNA methylation cytosine-phosphate-guanine (CpG) sites [6] influenced by genetic and/or environmental factors (Supplemental material) [7]. The first-generation epigenetic clocks, such as the Horvath-pan tissue [8], Hannum [9], Horvath-Skin & Blood [10], and Zhang2019 [11], were designed to predict chronological age. In contrast, the second-generation epigenetic clocks, such as DNAm PhenoAge [12] and GrimAge [13], reflect individual health conditions better by considering the risk factors of morbidity and mortality. Additionally, EA acceleration (EAA), the disparity between EA and chronological age, can be estimated by subtracting chronological ages from EAs (also referred to as delta age), or as the residuals from a linear model by regressing EAs on chronological ages [14]. Given that EAA accounts for chronological age and focuses on inter-individual variability, EAA can be influenced more strongly by genetic factors than EA [14, 15].

Epigenetic clocks differ in their association with health outcomes [16], suggesting that they might reflect different biological aspects of human aging [3]. According to previous studies on the association between epigenetic clocks and aging-related traits [3], certain epigenetic clocks have shared associations with aging-related traits while others have not [17]. This may be due to the varying amount and composition of markers comprising epigenetic clocks. For example, the Horvath-pan tissue clock captures cell-intrinsic changes in DNA methylation and reflects the innate aging process, whereas the Hanum clock is influenced by blood cell proportions and captures immunosenescence [3]. Even though the correlations between different epigenetic clocks and EA tend to be relatively strong (r = 0.88–0.90), there could be weak correlations with different EAAs (r = 0.11–0.57) [18]. Therefore, different epigenetic clocks can be considered independent biomarkers, and it is important to consider multiple epigenetic clocks for aging research.

Several genome-wide association studies (GWASs) on EAA have been conducted to explore epigenetic aging. The Horvath-intrinsic EAA (IEAA) has been shown to be genetically associated with metabolism (NHLRC1, TPMT, and KDM1B), apoptosis (FAIM), and autophagy (TERT) [3, 19], while Hannum-extrinsic EAA (EEAA) has been primarily associated with the immune pathway (TRIM46 and MUC1) [3]. Furthermore, Dong et al. highlighted the association between genetic variants within SELP and Horvath-EAA in childhood cancer survivors [18]. A novel variant associated with GrimAge-EAA and IBA57 was identified in a GWAS in the Taiwanese population [20]. Previous GWASs on EAA included multi-ethnic participants of European, African, and Hispanic ancestry [2, 3, 19]; however, only one study examined an Asian (Taiwanese) population [20]. Asians tend to live longer than other ethnic groups, which may be due to different genetic factors attributable to the heterogeneity of allelic architecture and/or linkage disequilibrium (LD) across ancestral populations [21]. Further genetic analyses of Asian populations may identify and validate novel and Asian-specific susceptibilities and inter-individual genetic variability of EAA.

In this study, we estimated EA and EAAs in 1962 middle-aged and elderly Korean individuals using eight epigenetic clocks (Horvath-pan tissue, Hannum, DNAm PhenoAge, GrimAge, GrimAge2, Zhang2019, Horvath-Skin & Blood, and DNAmFitAge). A total of five epigenetic clocks were retained for further analysis following the exclusion of unreliable epigenetic clocks due to missing CpG probes or low correlations with chronological age (r < 0.6). A GWAS was conducted to understand the genetic susceptibility and inter-individual variability of EAAs. To explore the potential underlying mechanisms involved in DNA methylation-related biological aging, further functional analyses were conducted based on the identified genetic susceptibilities of EAAs, including quantitative trait loci (QTL), gene-based pathway enrichment, and quantitative real-time polymerase chain reaction (qRT-PCR) analyses. In addition, we conducted a pleiotropy analysis to explore the genetic correlation between EAA-susceptible single-nucleotide polymorphisms (SNPs) and various phenotypes, including diseases and lifestyle traits, providing epidemiological and clinical evidence for the implications of these genetic findings on EAA susceptibility. To explore candidate drugs for epigenetic aging, deep-learning-based drug repurposing analyses were also conducted on the EAA-associated genes identified in this study. The workflow of this study is illustrated in Fig. 1.

Fig. 1.

Overall workflow of the study. 1 This study estimated EA and EAA using multiple epigenetic clocks in 1962 Korean individuals. 2 GWAS was conducted on each EAA to understand the genetic susceptibility and inter-individual variability of EAAs. 3 The secondary SNP-based functional analyses included colocalization and pleiotropy analysis. 4 Gene-based and pathway-enrichment analyses were performed to explore EAA-associated biological pathways. Abbreviations: Korean Genome and Epidemiology Study (KoGES), genome-wide association study (GWAS), single-nucleotide polymorphism (SNP), expression quantitative trait loci (eQTL), methylation quantitative trait loci (meQTL), epigenetic age (EA), epigenetic age acceleration (EAA)

Methods

Study population and data collection

This study was conducted using bioresources from the National Biobank of Korea and Korea Disease Control and Prevention Agency, Republic of Korea (NBK-2023–043). In this study, middle-aged and older adults, aged 40 or more years old, were recruited from two population-based KoGES cohorts [22]: Health Examinee Study (HEXA) and Ansan/Ansung study. Among the 59,155 HEXA and 5942 Ansan/Ansung study participants, those without DNA methylation data (N_HEXA = 58,333; N_{Ansan/Ansung study} = 4017) or genetic data (N_HEXA = 462; N_{Ansan/Ansung study} = 323) were excluded. A total of 1962 participants (N_HEXA = 360, N_{Ansan/Ansung study} = 1602) with acceptable-quality genetic and DNA methylation data were selected for this study (Figure S1).

Demographic and epidemiological data were collected using self-reported questionnaires. Anthropometric data were measured during physical examination, and clinical data were collected through laboratory experiments conducted in hospitals across Korea [22]. We collected data on sex, chronological age, body mass index (BMI), alcohol consumption, smoking status, regular physical activity, college graduation, and history of chronic diseases (hypertension, diabetes, arthritis, allergic diseases, hyperlipidemia, gout, gastritis, and ulcers). Chronic diseases were selected for analysis if more than 5% of participants reported having been diagnosed. As DNA methylation levels Change over time, the time of data collection was matched with the time of blood sampling for DNA methylation level measurement. Among the available epidemiological and clinical variables, those with an observed frequency exceeding 20 were selected for pleiotropy analysis. In total, 80 phenotypes (e.g., serum iron, C-reactive protein, and platelets) were selected for pleiotropy analysis (Data S1-2). The study protocol was approved by the Korean National Institutes of Health and the Institutional Review Board (IRB: CBNU-202405-HR-0120).

Genotyping and imputation

For genotyping, the Korea Biobank Array (KCHIP) platform detected approximately 800,000 (800 K) SNPs in each sample [23]. Quality control was performed to exclude low-quality samples and SNPs from further analyses. The exclusion criteria for low-quality samples were as follows: (1) call rate < 97%, (2) excessive singletons, (3) sex discrepancy, (4) cryptic second-degree relatives, and (5) withdrawal and blind replicates [22]. In addition, SNPs with Hardy–Weinberg equilibrium P < 1 × 10⁻⁶ or call rate < 0.95 were excluded, resulting in approximately 460 K SNPs remaining [22]. Imputation was performed using IMPUTE4 software based on the 1 K genome project phase 3 and the Korean reference genome [23, 24]. In total, 8000 K SNPs were imputed after considering the following exclusion criteria: (1) information content metric < 0.8 and (2) minor allele frequency (MAF) < 0.01 [23]. Consequently, 7,975,575 SNPs were included in this study.

DNA methylation and epigenetic clocks

Genome-wide methylation data were generated from the blood samples of the participants using the Infinium Methylaion450K Beadchip (450 K array) and Infinium Methylation EPIC Beadchip (EPIC array). Based on the cohort and array, we collected and analyzed raw DNA methylation intensity data using the R package ChAMP [25]. Quality control procedures were conducted on the DNA methylation data, excluding probes with the following criteria: (1) detection P-value > 0.01, (2) bead count < 3 in at least 5% of samples, (3) located on a non-CpG site or sex chromosome, (4) located near East Asian-specific SNPs identified by Nordlund et al. [26], and (5) aligned to multiple locations. DNA methylation levels were described as β-values and normalized by beta-mixture quantile method. Consequently, 732,046 CpG sites for the HEXA EPIC array, 724,792 for the Ansan/Ansung study EPIC, and 412,970 for the 450 K array were used in the current analysis.

Eight different types of EAs (Horvath-pan tissue, Hannum, DNAm PhenoAge, GrimAge, GrimAge2, Zhang2019, Horvath-Skin & Blood, and DNAmFitAge) were estimated using an online DNA Methylation Age Calculator (https://dnamage.clockfoundation.org/) [8]. The normalized DNA methylation beta-value files and annotation files that contained sample IDs, tissue type detecting DNA methylation, chronological age, and sex of the subject were submitted. GrimAge2 and DNAmFitAge were excluded from further analysis due to incomplete coverage of the required CpG probes in our DNA methylation arrays, with 21–22.5% of probes missing for GrimAge2 and 17.2–20.8% for DNAmFitAge. The EAAs were calculated as residuals from a linear model by regressing the estimated EAs on chronological ages. Pearson’s correlation analysis was used to measure the correlation between EAs and chronological age as well as between EAs and EAAs. Horvath-pan tissue-EA showed a weak correlation with chronological age (r < 0.6) and was excluded from further analyses (Figure S2) [27].

Statistical analysis of demographic and epidemiological variables

We investigated the association between EAAs and potential confounding factors for each EAA. Student’s t-test was used to compare EAAs between the two groups of binary variables, whereas Pearson’s correlation analysis was used for continuous variables. Among the demographic and epidemiological variables, chronological age (years) and BMI (kg/m²) were analyzed as continuous variables. In contrast, the cohort (HEXA and Ansan/Ansung studies), DNA methylation array (EPIC and 450 K), sex (male and female), alcohol consumption (never and ever), smoking status (never and ever), regular physical activity (no and yes), college graduation (no and yes), and history of chronic diseases (no and yes) were analyzed as binary variables (Table S1). Among the above potential confounding variables, statistically significant differences (P < 0.05) in more than three of the five EAAs were extracted. Accordingly, cohort, DNA methylation array, sex, regular physical activity, alcohol consumption, smoking status, college graduation, and history of gastritis or ulcer were adjusted for subsequent GWASs. All statistical analyses were performed using R software (version 4.3.3) [28].

GWAS on EAAs and replication of previously identified SNPs

For the EAA GWAS, a multivariable linear regression model was used with adjustments for the covariates described above. A GWAS was conducted using PLINK (version 1.9) on each EAA calculated from Hannum, Horvath-Skin & Blood, DNAm PhenoAge, GrimAge, and Zhang2019, but not Horvath-pan tissue because of the low correlation between EA and chronological age (r = 0.45) [29]. We considered the EAA susceptibility SNPs that met the genome-wide significance threshold (P < 5 × 10⁻⁸) and grouped them into independent loci based on distance (± 1000 kb) and LD structure (r² < 0.01). Using LocusZoom [30], regional plots were generated to visualize the LD relationship and annotate the EAA susceptibility genes.

Based on our GWAS results, the replication of the EAA susceptibility SNPs reported in previous studies was evaluated. A summary of previous studies was obtained from McCartney et al. (2021) and the GWAS Catalog on 10/03/2024 [2, 31]. In this study, the replication analysis included three EAAs (Hannum, DNAm PhenoAge, and GrimAge) because there have been no GWASs on Horvath-Skin & Blood and Zhang2019. To date, 24 SNPs have been associated with Hannum-EAAs, 24 with DNAm PhenoAge-EAAs, and 27 with GrimAge-EAAs. In the replication study, SNPs on the sex chromosome or those with P > 5 × 10⁻⁸ were excluded. Genetic variants that were absent from our GWAS were excluded. For SNPs reported in several studies, the one with the smallest P-value was selected. Consequently, 12 susceptibility SNPs for Hannum-EAA, 12 for DNAm PhenoAge-EAA, and 9 for GrimAge-EAA were selected for the current replication analysis. We considered SNPs to be significantly replicated when they met the following criteria: (1) P < 0.05, and (2) the effect directions were consistent in previous and current studies. The detailed process is shown in Figure S3.

Functional annotation and QTL analysis

To functionally annotate the identified EAA susceptibility SNPs (P < 5 × 10⁻⁸), we performed the Annotate variation (ANNOVAR) enrichment test using the FUMA software [32]. Based on the variant position, the SNPs were classified as exonic, intergenic, splicing site, 5′/3′-UTR, or upstream/downstream of genes. Furthermore, expression quantitative trait loci (eQTL) and methylation quantitative trait loci (meQTL) analyses were performed to explore the genes and CpG sites related to EAA susceptibility SNPs. eQTL analysis was conducted based on cis-eQTL association between SNPs and genes, within a 1 Mb window. eQTL association data were obtained from Genotype-Tissue Expression (GTEx) version 8 [33], and the significance level was FDR-corrected (FDR < 0.05). For meQTL analysis, CpG sites associated with EAA susceptibility SNPs were obtained from the Genetics of DNA Methylation Consortium (GoDMC) database [34], with a significance threshold of P < 5 × 10⁻⁸.

Gene-based, gene-set, and pathway enrichment analysis

Gene-based and gene-set analyses were conducted using multi-marker genomic annotation (MAGMA) [35]. In the gene-based analysis, individual SNP statistics for protein-coding genes were combined to obtain gene-based statistics. The significance level of the gene-based analysis tested on 18,805 genes was determined using Bonferroni correction (P < 2.66 × 10⁻⁶ = 0.05/18,805). Gene-set analysis was conducted to determine which gene sets representing biological pathways, cellular components, or molecular functions were enriched in the EAA-associated genes identified from gene-based analysis. A total of 17,009 gene sets obtained from the molecular signature database (MsigDB v7.0) were analyzed, and the significance level was determined with Bonferroni correction (P < 2.94 × 10⁻⁶ = 0.05/17,009) [36].

To confirm whether the identified EAA-associated genes were differentially expressed in senescent cells, HEK293T cells were cultured in DMEM supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. All cells were incubated at 37 °C in a 5% CO₂ atmosphere. To induce senescence, HEK293T cells were treated with 100 µM or 300 µM H₂O₂ for 72 h. Total RNA was extracted using the eCube Tissue RNA Mini Kit (Philekorea), according to the manufacturer’s protocol. Subsequently, 1 μg of RNA was reverse-transcribed using the AccuPower® CycleScript RT PreMix (Bioneer). The differential expression of EAA-associated genes in senescent cells was analyzed using qRT-PCR with SYBR Green Master Mix, and the expression levels were normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as an internal control. Relative expression levels were calculated using the comparative ΔΔCt method.

To facilitate the biological interpretation of EAA-associated genes, pathway enrichment analysis was performed using the ClueGO plug-in (version 2.5.10) in Cytoscape [37]. Based on various resources (e.g., Gene Ontology [38], Kyoto Encyclopedia of Genes and Genome Pathways [39], Reactome [40], and Wikipathways [41]), the EAA-associated genes obtained from the gene-based analysis were investigated for their biological pathways. Pathways containing two or more EAA-associated genes were selected, and their significance was calculated using a two-sided hypergeometric test with Bonferroni correction (P_Bonferroni < 0.05). To estimate the degree of Linkage between selected pathways, kappa statistics were calculated, and pathways with kappa statistics over 0.4 were assigned to the same group.

Pleiotropy analysis

To investigate the genetic relationships between EAA and various health-related traits, pleiotropy analyses were conducted based on the identified EAA susceptibility SNPs. A multivariable Linear regression model was adapted for 59 anthropometric and clinical variables (e.g., plasma platelets, red blood cells, and cholesterol). Additionally, 21 self-reported histories of various common disease diagnoses were analyzed as binary variables using a multivariable logistic regression model. Each model was adjusted for the same covariates used in the EAA GWAS (cohort, DNA methylation array, sex, regular physical activity, smoking, alcohol consumption, college, and history of gastritis or ulcers). Pleiotropy analysis was performed using the R package PheWAS [42].

Deep learning-based drug repurposing analyses

To identify potential drug candidates for epigenetic aging, we used DeepPurpose [43], a deep learning-based Python library for predicting drug-target interactions (DPI). DeepPurpose employs an encoder-decoder architecture to estimate the DPI affinities of a given target. In this study, we explored drugs with strong affinities for EAA-associated genes identified by gene-based analysis. The degree of affinity was prioritized by the binding score, with small values indicating strong affinities. According to the default settings of DeepPurpose, drug candidates with less than 30 binding scores with the given target were considered significant.

Results

Characteristics of the study population

The characteristics of the study participants (N = 1962) are summarized in Table S2. Participants were recruited from the Ansan/Ansung study (N = 1602, 81.7%) and the HEXA (N = 360, 19.3%). Genome-wide DNA methylation levels were determined using EPIC (N = 1682, 85.9%) and 450 K array (N = 280, 14.1%). Among the participants, 55.96% were male, 55.61% had consumed alcohol before, 45.41% had smoked tobacco before, 39.19% engaged in regular physical activity, 19.41% graduated from college or university, and 0.2% (cancer), 24.67% (gastritis or ulcer), and 25.99% (hypertension) had a reported history of chronic diseases. The average (± SD) of chronological age and BMI were 57.85 years (± 8.68) and 24.40 kg/m² (± 3.26), respectively.

Of the sixteen characteristic variables described above, eight were statistically significant (P < 0.05) for more than three out of five EAAs and were incorporated into the regression model for the GWAS analysis (cohort, DNA methylation array, sex, regular physical activity, smoking, alcohol consumption, college, and history of gastritis or ulcer; Table S1).

Epigenetic age and epigenetic age acceleration

We measured the correlation between six EA estimates and chronological age. The five EAs (Horvath-Skin & Blood, Hannum, DNAm PhenoAge, GrimAge, and Zhang2019) had moderate Pearson’s correlations with chronological age (r = 0.66–0.93, Figure S2). The strongest correlation was observed in the Zhang2019 (r = 0.93). As the Horvath-pan tissue clock had a relatively weak correlation with chronological age (r = 0.45), it was excluded from further analyses. Based on pairwise correlations between the five EAs, Pearson’s coefficients ranged from 0.62 to 0.88, indicating moderate correlations (Fig. 2A). However, the pair-wise correlations between five EAAs were relatively weak (r =  − 0.01 to 0.66, Fig. 2B). Horvath-skin and blood EAA and Hannum-EAAs showed the weakest correlation (r =  − 0.01), while Hannum-EAAs and Zhang2019-EAAs had the strongest correlation (r = 0.66).

Fig. 2.

Pairwise correlations of epigenetic age and epigenetic age acceleration. Scatter plots and Pearson’s coefficient of five epigenetic age (a) and epigenetic age acceleration (b). Pearson’s coefficients are statistically significant, with P < 0.001 denoted as ***. Each dot represents a single study participant

GWAS on EAAs

GWASs were conducted on the EAAs calculated from the five selected epigenetic clocks. The genomic inflation factors ranged from 1.00 to 1.02, indicating little systemic inflation (Figure S4). In total, six SNPs were independently genome-wide significant with respect to EAAs (P < 5 × 10⁻⁸, Table 1, Fig. 3A). Specifically, the DNAm PhenoAge-EAA GWAS identified two significant SNPs, including rs3829957 mapped to ASPA and SPATA22 (β(T) =  − 0.838, P = 2.89 × 10⁻⁸) and rs74333895 mapped to CDH4 (β(T) =  − 1.971, P = 4.63 × 10⁻⁸). Furthermore, 75 marginally significant SNPs (P < 1 × 10⁻⁵) were found at the same genomic locus as rs3829957 with a strong LD (r² > 0.8; Figure S5). The GrimAge-EAA GWAS identified three independently significant SNPs on chromosome 1, including rs76713297 mapped to RP3-340N1.6 (β(A) = 3.751, P = 1.66 × 10⁻²¹), rs72969023 mapped to LINC01141 (β(T) = 2.216, P = 1.45 × 10⁻⁹), and rs74072052 located on intergenic region (β(A) =  − 1.092, P = 3.87 × 10⁻⁸). The Zhang2019-EAA-associated SNP was rs3741922 mapped to ENO2 (β(A) =  − 0.602, P = 3.52 × 10⁻⁹), which has 16 other genome-wide significant SNPs (P < 5 × 10⁻⁸) and 36 marginally significant SNPs (P < 1 × 10⁻⁵) in the same locus with strong LD (r² > 0.8, Figure S5). The estimated heritability (standard error) was 0.25 (0.19) for DNAm PhenoAge-EAA and 0.35 (0.19) for Zhang2019-EAA whereas close to 0 for GrimAge-EAA. There was no genome-wide significant signal in our GWASs for the Horvath-Skin & Blood EAA and Hannum-EAA (Figure S6).

Table 1.

SNPs associated with epigenetic age accelerations

SNP	CHR	BP	Nearest genesa	Functional annotation	EA	RA	EAF (KoGES GWAS)	EAF (1 KG EUR)	BETA	SE	P	Previously reported (r2 > 0.5)	Epigenetic clock
rs3829957	17	3,378,876	ASPA, SPATA22	5′ UTR	T	C	0.40	0.19	− 0.838	0.15	2.89E − 08	Yes	DNAm PhenoAge
rs74333895	20	60,159,282	CDH4	Intronic	T	G	0.04	0.00	− 1.971	0.359	4.63E − 08	Novel	DNAm PhenoAge
rs76713297	1	20,573,260	RP3-340N1.6	Intergenic	A	G	0.02	0.00	3.751	0.389	1.66E − 21	Novel	GrimAge
rs72969023	1	20,757,779	LINC01141	Intronic	T	C	0.02	0.01	2.216	0.365	1.45E − 09	Novel	GrimAge
rs74072052	1	55,392,936	Intergenic	Intergenic	A	G	0.08	0.00	− 1.092	0.198	3.87E − 08	Novel	GrimAge
rs3741922	12	7,032,254	ENO2	3′ UTR	A	G	0.18	0.00	− 0.602	0.102	3.52E − 09	Novel	Zhang2019

^aNearest gene is located within ± 10 kb of the corresponding SNP

Single nucleotide polymorphism (SNP), chromosome (CHR), base pair (BP), untranslated region (UTR), effect allele (EA), reference allele (RE), effect allele frequency (EAF), Korean Genome and Epidemiology Study (KoGES), genome-wide association study (GWAS), European (EUR), standard error (SE)

Fig. 3.

Results of GWAS and genome-wide gene-based analyses of EAA. Manhattan plot of GWAS (a) and genome-wide gene-based analyses (b) of EAAs with significant signals in the current study. Red dotted lines indicate the genome-wide significance level (P < 5 × 10⁻⁸) and Bonferroni significance level (P < 2.66 × 10⁻⁶) in (a) and (b), respectively. There was no genome-wide significant signal in GWASs for the Horvath-Skin & Blood EAA and Hannum-EAA. Abbreviations: genome-wide association study (GWAS), epigenetic age acceleration (EAA)

Previously identified SNPs

In total, 33 previously reported EAA susceptibility SNPs were evaluated for their significance in our GWAS (Table S3). Among these, 10 SNPs (30.3%) were replicated in our GWAS, with concordant effect directions and nominal significance thresholds (P < 0.05). According to the ancestry in which the GWAS was conducted, SNPs identified in East Asia were replicated in 33.3% (1/3), Europe in 36.4% (8/22), and Africa in 14.3% (1/7; Table S3). We examined 12 Hannum-EAA susceptibility SNPs, and four (rs28366133, rs111731678, rs4838595, and rs3893495) were replicated (33.3%). Similarly, five (rs678553, rs75407001, rs1142345, rs3829957, and rs116853700) of the 12 SNPs (41.7%) and one (rs9386796) of the nine SNPs (11.1%) were replicated for DNAm PhenoAge-EAA and GrimAge-EAA, respectively. In our GWAS, DNAm PhenoAge-EAA-susceptibility SNPs showed the best results (41.67%). In particular, rs3829957, identified in a European population by McCartney et.al, was the most significant SNP identified in our Asian population-based DNAm PhenoAge-EAA GWAS.

QTL analyses

The identified SNPs were analyzed using eQTL and meQTL to determine their relationships with gene expression. The eQTL analyses identified four pairs of SNP-gene associations (false discovery rate (FDR) < 0.05, Table S4). Specifically, rs3829957(T), one of the DNAm PhenoAge-EAA-associated SNPs, was associated with the expression of ASPA across seven different tissues including the left ventricle (β =  − 0.442, FDR = 9.47 × 10⁻¹⁹), tibial artery (β =  − 0.21, FDR = 3.69 × 10⁻¹³), and atrial appendage (β =  − 0.252, FDR = 9.64 × 10⁻⁹). Furthermore, this SNP was involved in the expressions of ORE1E1 (testis tissue, β = 0.824, FDR = 2.83 × 10⁻²³) and SHPK (amygdala tissue, β =  − 0.478, FDR = 2.96 × 10⁻¹⁴), which are located nearby ASPA. The meQTL analysis identified significant pairs of rs3829957(T) and 24 CpG sites (P < 5 × 10⁻⁸) including cg07863524 mapped to OR3A4P (β = 0.527, P < 9.12 × 10⁻³⁰²), cg16505927 mapped to SPATA22 (β = 0.423, P < 9.12 × 10⁻³⁰²), and cg02228185 mapped to ASPA (β = 0.664, P < 9.12 × 10⁻³⁰², Table S5). Notable, cg02228185 is one of the 513 CpG sites used for DNAm PhenoAge calculation. Considering the identified CpG sites were located within ASPA and nearby genes (OR1A1, OR1D4, OR3A2, OR3A4P, OR3A3, OR1E2, and SPATA22), this may strengthen the eQTL analysis results, which indicate an association between rs3829957 and ASPA expression.

Gene-based and experimental analyses

To investigate the biological pathways underlying epigenetic aging, gene-based and pathway enrichment analyses were performed using the results of the EAA GWAS. For the MAGMA gene-based analysis, 18,805 protein-coding genes were evaluated for their association with EAAs. Consequently, the DNAm PhenoAge-EAA-associated genes were SPATA22 and ASPA, whereas Zhang2019-EAA was associated with five genes, ENO2, USP5, GNB3, TPI1, and CDCA3 (P < 2.66 × 10⁻⁶ = 0.05/18,805, Fig. 3B). Seven EAA-associated genes were evaluated for expression in H₂O₂-induced senescent cells. After treatment with 300 μM H₂O₂, the expression levels of SPATA22, ENO2, TPI1, and USP5 significantly increased (P < 0.01) compared to control and cell groups treated with 100 μM H₂O₂ (Figure S7).

Gene-set and pathway enrichment analyses

The MAGMA gene-set enrichment analysis revealed that seven gene sets (negative regulation of intrinsic apoptosis signaling pathway by p53 class mediator, negative regulation of signal transduction by p53 class mediator, regulation of intrinsic apoptotic signaling pathway in response to DNA damage by p53 class mediator, regulation of apoptotic signaling pathway, regulation of intrinsic apoptotic signaling pathway by p53 class mediator, regulation of intrinsic apoptotic signaling pathway, and negative regulation of intrinsic apoptotic signaling pathway) involved in apoptotic signaling pathways were significantly associated with DNAm PhenoAge-EAA (P < 2.94 × 10⁻⁶ = 0.05/17,009, Fig. 4A). Additionally, in the ClueGO pathway enrichment analysis, five Zhang2019-EAA-associated genes were found to be involved in ten biological pathways, represented by the monosaccharide catabolism pathway (Fig. 4B).

Fig. 4.

Results of gene-set and pathway enrichment analysis. Gene set enrichment analysis of DNAm PhenoAge-EAAs (a) and pathway enrichment analysis of Zhang2019-EAA-associated genes (b). The red dotted line in a indicates the Bonferroni significance line (P < 2.94 × 10⁻⁶), and seven apoptosis-related pathways met the threshold. A total of ten Zhang2019-EAA enriched pathways in b were related to each other and the highlighted one, monosaccharide Catabolic process, represents the remaining 9 pathways

Pleiotropy analyses

Based on the six EAA susceptibility SNPs, pleiotropy analyses were performed on 80 health-related traits, including 59 anthropometric and clinical measures and 21 self-reported disease diagnoses (Data S1-2). Consequently, significant associations were observed between rs3741922(A), which is a Zhang2019-EAA susceptibility SNP, and plasma iron (β =  − 0.015, FDR = 0.03), total bilirubin (β =  − 0.1, FDR = 0.04), direct bilirubin (β =  − 0.032, FDR = 0.05), and indirect bilirubin (β =  − 0.072, FDR = 0.04, Fig. 5).

Fig. 5.

Significant associations based on pleiotropy analysis of rs3741922. Forest plot of pleiotropy analysis between the Zhang2019-EAA susceptibility SNP rs3741922 and 80 health-related traits. The statistical values for the four traits that showed significance at FDR < 0.05 were presented. There was no significant pair between the remaining five EAA susceptibility SNPs and 80 health-related traits. Abbreviations: standard error (SE), false discovery rate (FDR), epigenetic age acceleration (EAA), single nucleotide polymorphism (SNP)

Deep learning-based drug repurposing analyses

Based on the results of a genome-wide gene-based association study, DNAm PhenoAge-EAA-associated genes (ASPA and SPATA22) and Zhang2019-EAA-associated genes (GNB3, ENO2, USP5, TPI1, and CDCA3) were evaluated for their affinity to the drugs (Table S6-7). Notably, GW0742, a peroxisome proliferator-activated receptor-β/δ (PPAR-β/δ) agonist, showed the strongest affinity (binding score 1.09–10.95) with all seven targets evaluated in this study.

Discussion

As a surrogate for epigenetic aging, we calculated EAAs and identified genetic factors in 1962 East Asian (Korean) individuals. Among the five types of epigenetic clocks (Horvath-Skin & Blood, Hannum, DNAm PhenoAge, GrimAge, and Zhang2019), the EA calculated from Zhang2019 showed a remarkable correlation with chronological age (r = 0.93). Considering that East Asian individuals were included in the algorithm development, Zhang2019 could reflect genetic differences in ancestry and best represent the epigenetic age of this population. GWASs were conducted for each EAA, and independently significant signals (P < 5 × 10⁻⁸) were observed for DNAm PhenoAge-EAA (rs3829957 mapped to ASPA and SPATA22; rs74333895 mapped to CDH4), GrimAge-EAA (rs76713297 mapped to RP3-340N1.6; rs72969023 mapped to LINC01141; rs74072052 located in the intergenic region), and Zhang2019-EAA (rs3741922 mapped to ENO2).

While five of the identified six SNPs were novel (r² < 0.5), the remaining rs3829957, previously reported to be associated with DNAm PhenoAge-EAA in European ancestry [2], indicated trans-ethnic validation by the current GWAS in Asian ancestry. Additionally, the replication rates of SNPs identified in the East Asian and European populations were comparable at 33.3% (1/3) and 36.4% (8/22), respectively. This suggests that a portion of genetic factors associated with EAA may be shared across ethnicities, although a systematic analysis in East Asian populations has not been sufficiently conducted. In contrast, the newly identified SNPs in this study implicated the genetic heterogeneity between ethnicities, showing effect allele frequencies between 0.00 and 0.01 in European populations while ranging from 0.02 to 0.18 in our study population [24]. Despite the need for replication in other East Asian populations, these markers may be East Asian-specific genetic markers for EAAs. Therefore, the heterogeneous genetic architecture between ancestral populations could result in differences in allele frequencies and ancestry-specific genetic markers. Thus, the genetic basis of epigenetic aging can be explained by a partially shared but heterogeneous genetic architecture across ancestral populations.

A DNAm PhenoAge-EAA-associated SNP, rs3829957, is an intron variant of ASPA and SPATA22 and was the most significant DNAm PhenoAge-EAA susceptibility variant in the current study. The regional plot pinpointed ASPA and SPATA22 genes showing nearby SNPs with strong signals for EAA and LD with rs3829957. Based on QTL analyses, rs3829957 was associated with ASPA expression, and the methylation levels of CpG sites mapped to ASPA and SPATA22, including cg02228185 used for DNAm PhenoAge calculation. This indicates that rs3829957 may be a regulatory genetic biomarker of DNAm PhenoAge-EAA via ASPA and SPATA22, both of which are significantly associated with DNAm PhenoAge-EAA in genome-wide gene-based analyses. Subsequent RT-qPCR analysis revealed increased SPATA22 expression in senescent cells. Furthermore, genome-wide gene-based analysis revealed that these two genes were significantly associated with DNAm PhenoAge-EAA. ASPA encodes an enzyme called aspartoacylase that promotes deacetylation of N-acetylaspartic acid (NAA) to produce acetate and l-aspartate [44]. Although its precise function remains unknown, this catabolic reaction appears to be important for myelin lipid formation [44]. According to previous studies, mutant ASPA leads to an abnormal elevation of NAA levels, causing Canavan disease, a fatal neurodegenerative disorder [45]. In addition, NAA levels have been found to be abnormally low in patients with multiple sclerosis and Alzheimer’s disease [46, 47]. SPATA22 encodes a protein called spermatogenesis-associated 22 (SPATA22) that plays a key role during meiosis. A previous study found that mice with mutant Spata22 were infertile due to meiotic prophase arrest [48]. According to the MAGMA gene set analysis, the gene sets involved in the apoptotic signaling pathway were significantly associated with DNAm PhenoAge-EAA. In a previous study that performed global gene expression profiling in patients with Canavan disease with severe ASPA mutations, genes involved in apoptosis and inflammation were significantly dysregulated (e.g., SERPINAB7, PDCD11, and CASP4) [49]. This suggests that ASPA plays a direct or indirect role in the apoptotic pathway. Considering that cumulative evidence suggests that the deregulation of apoptosis is associated with aging [50], DNAm PhenoAge-EAA susceptibility SNPs seem to reflect the genetic factors of epigenetic aging.

Another highlighted EAA susceptibility SNP was rs3741922, and nearby SNPs within the same genomic region with high LD also showed strong signals for EAA. A Zhang2019-EAA-associated SNP, rs3741922, is located on the 3′ untranslated region of ENO2 and is the most significant Zhang2019-EAA susceptibility SNP in the current study. ENO2 encodes an enzyme called γ-enolase, which Catalyzes the conversion of 2-phosphoglycerate into phosphoenolpyruvate in the glycolytic pathway, eventually leading to a gain of adenosine triphosphate [51]. Previous studies found that γ-enolase levels are associated with neurological disease [52] and several types of cancer [53–55]. In the current study, genome-wide gene-based analysis identified Zhang2019-EAA-associated genes as ENO2, USP5, GNB3, TPI1, and CDCA3, with ENO2 being the most significant. In particular, based on the qRT-PCR results, ENO2, USP5, and TPI1 expressions were increased in senescent cells. Subsequent pathway enrichment analysis revealed that Zhang2019-EAA-associated genes are primarily involved in the monosaccharide catabolic process, which involves breaking down monosaccharides to produce energy. USP5 encodes an enzyme called ubiquitin-specific peptidase 5 (USP5), which disassembles polyubiquitin chains on degraded proteins into ubiquitin monomers. A recent study reported that the knockdown of USP5 inhibited aerobic glycolysis in triple-negative breast cancer cells [56]. Aerobic glycolysis is observed in cancer cells, increasing glucose uptake and facilitating high rate of energy production [57]. It was also suggested that USP5 is associated with other cancer types, including non-small cell lung cancer [58] and hepatocellular carcinoma [59]. GNB3 encodes an enzyme called G-protein β3 (GNB3), a subunit of G-protein which mediates diverse cellular functions such as differentiation, sensation, growth, and homeostasis [60]. GNB3 is associated with chronic diseases such as obesity, type 2 diabetes, glucose intolerance, and metabolic syndrome [60, 61]. TPI1 encodes triosephosphate isomerase 1 (TPI1), which regulates the interconversion between glyceraldehyde 3-phosphate and dihydroxyacetone phosphate during glycolysis and glucogenesis. A previous study highlighted that extracellular vesicles with reduced TPI1 levels enhanced aerobic glycolysis in recipient hepatocarcinogenic cells [62]. CDCA3 encodes an enzyme called cell division cycle associated 3 (CDCA3) that participates in the cell cycle and endocytosis [63]. CDCA3 has been associated with the growth and proliferation of various types of cancers, including hepatocellular carcinoma [64], non-small cell lung cancer [65], and prostate cancer [66]. Based on previous studies, six Zhang2019-EAA-associated genes have been reported to participate in glycolysis or to be associated with various cancer types. Considering that glycolysis occurs primarily in the liver and that aging is associated with gradual alterations in hepatic function and structure [67], Zhang2019-EAA and liver function may be genetically related.

According to the pleiotropy analysis in the current study, rs3741922 was associated with four biochemical markers: total bilirubin, indirect bilirubin, direct bilirubin, and plasma iron. Bilirubin and plasma iron levels are elevated during the catabolism of heme-containing proteins, such as hemoglobin [68, 69]. The initial bilirubin is water-insoluble and is referred to as indirect bilirubin. Once the indirect bilirubin bound to albumin is absorbed by the liver, water-soluble direct bilirubin is produced. Numerous factors contribute to increased bilirubin levels [68]. During excess hemolysis, bilirubin production can exceed the liver uptake capacity, leading to elevated serum indirect bilirubin concentrations [68]. Additionally, both direct and indirect bilirubin levels can be elevated in hepatocellular diseases, such as viral, alcoholic, and toxic hepatitis [68]. Bilirubin levels have been used as a diagnostic biomarker for liver diseases, such as acute hepatitis, liver cirrhosis, and autoimmune hepatitis [68, 70, 71]. The Zhang2019-EAA susceptibility SNP rs3741922 was found to be associated with lower levels of Zhang2019-EAA and bilirubin. It is possible that this locus is related to epigenetic aging, as explained by liver function. In support of this hypothesis, a genome-wide gene-based analysis identified that Zhang2019-associated genes primarily involve in monosaccharide catabolism mainly in the liver. The potential mechanisms of epigenetic aging based on the associations and implications identified in this study are described in Figure S8.

The drug repurposing analyses identified GW0742, a PPAR-β/δ agonist, as having the strongest affinity with both DNAm PhenoAge-EAA-associated genes (ASAP and SPATA22) and Zhang2019-EAA-associated genes (ENO2, USP5, GNB3, TPI1, and CDCA3). PPAR-β/δ is a nuclear receptor that plays a crucial role in cholesterol release regulation and is associated with metabolic syndrome and diabetes mellitus type 2 [72]. In a previous study, GW0742 improved glucose homeostasis in diabetic rats by PPAR-δ activation [73]. This could explain the high affinity between GW0742 and the Zhang2019-EAA-associated genes, which are primarily involved in monosaccharide catabolism. Considering the high relevance between apoptosis and metabolic pathways [74], GW0742 may also regulate cell fate through the DNAm PhenoAge-EAA genes, which are associated with the apoptosis pathway.

Despite these findings, this study has several limitations. First, it had a relatively small number of participants compared with previous EAA GWAS [2, 3, 19]. However, Wan-Yu Lin (2021) reported meaningful EAA-susceptibility SNPs (rs117530284, rs189669793, and rs3893495) in 2309 Taiwanese individuals [20]. This sample size was comparable to that of the current study (N = 1962); therefore, our GWAS may have moderate statistical power. Second, although several covariates were considered to adjust for confounding factors, residual bias may have remained because aging is a multifaceted and complex process affected by various factors.

Conclusions

This study is one of the leading EAA GWAS conducted on East Asian populations and provides ancestral diversity in epigenetic aging research. In addition, this study reports for the first time the genetic variants associated with Zhang2019-EAA. In the current study, we performed GWASs on five types of EAA in 1962 Korean individuals and identified six loci associated with the DNAm PhenoAge-EAA, GrimAge-EAA, and Zhang2019-EAA. Five of the SNPs identified have not been reported previously and are rare in European populations (0.0–0.01), suggesting that they are ancestral-specific genetic loci in epigenetic aging. Subsequent analyses, including gene-based, QTL, and pleiotropy analyses, highlighted rs3829957 and rs3741922 as being involved in apoptosis and liver function, respectively. By identifying genetic factors associated with different EAA, this study contributes to a better understanding of differential epigenetic aging in East Asian populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

K.O. and N.S. designed the study and wrote the manuscript; K.O. and D.Y. performed all analyses; K.O., D.Y., M.Y., S.Y., and N.S. interpreted the data; K.O., M.Y., S.Y., D.Y., and N.S. assisted with the data collection; B.K., S.H., and H.P. conducted the experiments; D.Y., M.Y., S.Y., C.L., B.K., S.H., Q.D., Z.W., H.P., and N.S. critically revised the manuscript; All authors contributed to data interpretation and manuscript revision and approved the final manuscript for publication.

Funding

This study was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean Government (MSIT) (No. RS-2024–00358322, RS-2025-02273102). This research was supported by the Bio & Medical Technology Development Program of the National Research Foundation (NRF), funded by the Korean government (MSIT) (No. RS-2024–00440787).

Data Availability

The demographic, DNA methylome, and SNP genotype data used in this study were provided by the National Biobank of Korea (NBK) and the Center for Disease Control and Prevention, Republic of Korea. These data can be requested from the website (https://biobank.nih.go.kr), and it is strictly prohibited to disclose it without permission.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Korean National Institutes of Health and the Institutional Review Board (IRB: CBNU-202405-HR-0120). Informed consent was exempted because the data used in this study were provided by the National Biobank of Korea (NBK) and the Center for Disease Control and Prevention, Republic of Korea.

Consent for publication

All author consent to publication in Geroscience.

Conflict of interest

The authors declare no competing interests.