Abstract Obesity and abdominal obesity contribute to significant metabolic health risks through distinct pathophysiological mechanisms. We conducted an epigenome-wide association study to identify differential DNA methylation patterns associated with body mass index (BMI)-defined and abdominal obesity and explore their relationships with dietary intake among Korean adults (n = 1,526). We identified 23 and 1931 DMPs associated with abdominal obesity and severe BMI-defined obesity, respectively, with four CpG sites common to both phenotypes. The most significant associations were at cg10323433 and cg10501210 in serotonin receptor 2 A (HTR2A) for abdominal obesity (Δβ = −0.023) and BMI-defined obesity (Δβ = −0.021), respectively. Most DMPs (> 75%) exhibited hypomethylation in obesity with progressive changes correlating with obesity severity. Hierarchical clustering revealed distinct dietary associations: WHR-related DMPs correlated with traditional fermented foods, whereas BMI-related DMPs showed stronger associations with fruit consumption. Sites with hypomethylation in obesity consistently demonstrated positive correlations with fat intake but negative correlations with carbohydrates. The distinct associations between methylation patterns and dietary components suggest that different foods may influence epigenetic modifications that are specific to overall adiposity or fat distribution, providing potential targets for nutritional interventions to modify obesity-related epigenetic signatures. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-025-13868-6. Keywords: DNA methylation, Obesity, Waist-to-hip ratio, Dietary patterns, Epigenetics Subject terms: Epigenetics, Epigenomics, Obesity, Nutrition Introduction Obesity, defined by elevated BMI, and abdominal obesity, characterized by increased waist circumference (WC) or waist-to-hip ratio (WHR), are major public health concerns associated with increased risks of cardiovascular diseases, diabetes mellitus, and metabolic disorders^[32]1,[33]2. In South Korea, obesity is defined as BMI ≥ 25 kg/m², while abdominal obesity is defined as WC ≥ 90 cm for men and ≥ 85 cm for women^[34]3,[35]4. WHR has been recognized as a superior indicator for predicting metabolic risks compared to BMI or WC alone, as demonstrated in studies showing its strong association with diabetes prevalence and insulin secretion phases^[36]5,[37]6. Recent epigenome-wide association studies (EWAS) have identified DNA methylation patterns associated with adiposity measures. For instance, a study conducted within the RODAM project aimed to identify specific DNA methylation sites associated with BMI and WC among Ghanaians. This study identified differentially methylated positions (DMPs) and regions (DMRs) associated with these adiposity measures, highlighting the epigenetic mechanisms that may contribute to obesity and its distribution in the body^[38]7 emphasizing the significance of epigenetic factors in comprehending the intricacy of adiposity beyond solely genetic predispositions. In addition, the Korean Genome and Epidemiology Study (KoGES) has been crucial in investigating the genetic and environmental factors underlying various chronic diseases in Koreans^[39]8. Although GWAS utilizing KoGES data have shed light on numerous health conditions, EWAS focusing on obesity and abdominal obesity has been significantly lacking in this population. EWAS provide insights into how environmental and lifestyle factors can impact gene expression through epigenetic changes, such as DNA methylation, shedding light on the intricate mechanisms of obesity^[40]9. A notable EWAS, using KoGES data, identified 48 CpGs linked to obesity in Korean individuals by examining blood and fat tissue samples. However, this study focused solely on BMI-defined obesity and did not examine abdominal obesity phenotypes or dietary correlations. Moreover, the relatively smaller sample size (n = 902) may have limited the detection of obesity-related epigenetic signatures. The lack of dietary integration in previous obesity EWAS represents a significant gap, given the established role of nutrition in modulating DNA methylation patterns. A significant finding was a CpG site on the CPA3 promoter, which was associated with enhanced expression in obesity and potential implications for T2DM and asthma^[41]10. The lack of EWAS on obesity in the KoGES data indicates a missed opportunity to investigate the epigenetic complexities of these conditions in Koreans. Addressing this gap could reveal new epigenetic markers and pathways influencing obesity, offering new targets for intervention and enhancing our understanding of the roles of genetics, epigenetics, and environmental factors in the development of obesity. The association between DNA methylation levels and food intake involves a complex interplay influenced by various dietary components and patterns. This relationship has been studied, highlighting the impact of specific food groups and nutrients on DNA methylation. For instance, ultra-processed food consumption in children has been linked to suggestive changes in methylation at specific CpG sites, although these changes were not significant at the false discovery rate (FDR) level, indicating a potential but not definitive impact on health outcomes, such as carcinomas and pathways related to thyroid hormones and liver function^[42]11. Similarly, dietary patterns, including nutrient intake, such as vitamin B12, folate, and choline, modulate DNA methylation, which plays a critical role in gene regulation and the development of metabolic disorders, such as obesity and T2DM^[43]12. The consumption of industrialized foods has been associated with methylation changes in the NR3C1 gene, linked to stress responses and metabolic alterations^[44]13. Furthermore, specific food groups, such as nuts, seeds, and dairy products, have been associated with methylation changes in genes relevant to metabolic health^[45]14. Moreover, global DNA methylation levels have been associated with micronutrient intake, such as zinc and vitamin B3, indicating their role as cofactors in methylation pathways^[46]15. We conducted an EWAS to investigate the DNA methylation patterns associated with obesity and abdominal obesity using data from the KoGES. We also aimed to identify specific DMPs related to various adiposity measures, including BMI and WHR, in Korean adults. Furthermore, we explored the relationship between these methylation markers and dietary patterns by analyzing food frequency questionnaire data to understand how specific food intake may influence the epigenetic modifications associated with obesity phenotypes. By integrating methylation data with comprehensive anthropometric measurements and detailed dietary information, we could provide insights into the epigenetic mechanisms underlying obesity in the Korean population and highlight potential epigenetic biomarkers that could serve as targets for intervention strategies addressing obesity-related health complications. Methods Data source The clinical, epidemiological, and DNA methylation array datasets were obtained from the Ansung–Ansan cohort of the KoGES, facilitated by the Korea Center for Disease Control and Prevention^[47]8. The participants’ age range was 40–69 years. For this study, we utilized DNA methylation data from the Infinium Methylation 850k (HM850k or EPIC) array, specifically sourced from the fourth follow-up cohort (2009–2010), including 1,526 samples with 865,918 CpG probes. Dietary intake data were obtained from the same participants during the second follow-up survey (2005–2006). We used a semiquantitative food frequency questionnaire (Semi-FFQ) to assess the daily consumption of 106 food items and intake of macro- and micronutrients. Ethical compliance All methods were performed in accordance with the relevant guidelines and regulations, including the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of the Korea Food Research Institute (Approval No.: 2022-01-002-001). Written informed consent was obtained from all participants. Study design and participant classification This study was designed to investigate distinct DNA methylation patterns associated with obesity phenotypes using data from the Ansung–Ansan cohort of the KoGES. We included 1,526 participants and categorized them based on BMI and abdominal obesity status (Fig. [48]1). Fig. 1. [49]Fig. 1 [50]Open in a new tab Workflow for DNA methylation analysis in relation to obesity phenotypes. The study initially had 1,528 participants, of which two were excluded (one due to quality control issues and one due to missing smoking data). The remaining 1,526 participants’ DNA methylation data were normalized using the BMIQ method, followed by batch effect correction using the ComBat algorithm and removal of cell-type heterogeneity using the champ.refbase function. Subsequently, the cohort was stratified into two parallel analysis streams: obesity groups based on BMI classification (lean, overweight, obesity (stage I), severe obesity) and abdominal obesity groups based on waist-to-hip ratio (WHR) (lean; abdominal obesity). Covariates were adjusted separately for each stream, with the BMI analysis adjusting for age, sex, smoking habit, and WHR, whereas the WHR analysis adjusted for age, sex, smoking habit, and BMI. Differentially methylated positions (DMPs) were identified using thresholds of FDR < 0.05 and absolute β-value difference of ≥ 0.02 and ≥ 0.01 for the BMI and WHR groups, respectively. Post-hoc analyses included correlation analyses between BMI and WHR, with methylation levels of the top 20 DMPs, and with food or nutrient intake. BMI-based classifications followed the Korean Society for the Study of Obesity guidelines: underweight (< 18.5 kg/m²), normal weight (18.5–22.9 kg/m²), overweight (23–24.9 kg/m²), class I obesity (25–29.9 kg/m²), class II obesity (30–34.9 kg/m²), and class III obesity (≥ 35 kg/m²). For statistical analysis, participants with class II (n = 62) and class III obesity (n = 3) were combined into a “severe obesity” group because the number of individuals in class III was too small for meaningful analysis. This grouping was performed to ensure sufficient statistical power. The remaining BMI categories were maintained as defined above. Abdominal obesity was defined according to sex-specific WC thresholds: ≥ 90 and ≥ 85 cm for men and women, respectively. WHR of ≥ 0.90 and ≥ 0.85 for men and women, respectively, was indicative of central adiposity. Based on these criteria, 1,089 and 437 participants were classified as having abdominal obesity and being lean, respectively. We excluded individuals with incomplete anthropometric measurements or those with undetermined obesity status. After collecting DNA methylation data, we performed comprehensive quality control, normalization using BMIQ method, and batch effect correction using ComBat as detailed in the methods below. We adjusted for multiple covariates including age, sex, smoking status, WHR (for BMI-based analysis), and BMI (for abdominal obesity analysis) to control for potential confounding factors. Subsequently, we identified DMPs associated with different obesity phenotypes using stringent statistical criteria. The top DMPs were analyzed for correlation with anthropometric measurements and metabolic parameters using Spearman correlation (p < 0.05). The top DMPs were defined as the most statistically significant positions ranked by their adjusted p-values (FDR), with the top 20 DMPs (lowest FDR values) selected for subsequent correlation analyses with anthropometric measurements and dietary variables. DNA methylation data processing and quality control The HM850k DNA methylation data for a cohort of 1,526 participants were imported using the “champ.load” function from the “ChAMP” R package (ver. 2.26.0)^[51]16. They were categorized based on their obesity status, with 501, 418, 542, and 65 individuals classified as normal weight, overweight, class I obese, and severely obese, respectively. Additionally, 1,089 participants met the criteria for abdominal obesity, whereas 437 were classified as lean. Following data import, a comprehensive filtration process (Supplementary Fig. [52]S1) resulted in the selection of 717,924 CpG probes from the initial pool of 865,918 CpG probes. To address the technical discrepancies between the type I and II probes on the array, we utilized the beta-mixture quantile normalization (BMIQ) method^[53]17 using the “champ.norm” function. Batch effects arising from different slides were detected using the “champ.SVD” function and subsequently corrected using the “champ.runcombat” function of the ChAMP package. To account for potential confounding effects of blood cell composition differences between obesity groups, we estimated cell-type proportions using the Houseman algorithm implemented in the “champ.refbase” function of the ChAMP package. This reference-based deconvolution method uses methylation profiles from sorted blood cell populations to estimate proportions of six major blood cell types: CD8 + T cells, CD4 + T cells, natural killer cells, B cells, monocytes, and granulocytes from whole blood DNA methylation data. The estimated cell-type proportions were subsequently included as additional covariates in all differential methylation analyses to control for potential confounding effects of cellular heterogeneity. For the BMI-based obesity analysis, we adjusted for age, sex, smoking habit, WHR, and estimated cell-type proportions as potential confounding factors. For the abdominal obesity analysis based on WC thresholds, we adjusted for age, sex, smoking habit, BMI, and estimated cell-type proportions as covariates. These adjustments were performed using the “limma” R package^[54]18. Moreover, these preprocessing steps ensured the reliability and comparability of the methylation data across the obesity phenotypes for subsequent differential methylation analyses. Differential methylation analysis and post-hoc correlation studies The DMPs were identified using the “champ.dmp” function within the ChAMP package. We established two parallel analytical pathways focusing on different obesity phenotypes: BMI-based and abdominal obesity groups. For the BMI-based obesity analysis, the participants were categorized as lean, overweight, obese (stage I), or severely obese. Covariates, including sex, smoking habit, and WHR were adjusted in the analytical model. DMPs were identified based on two criteria: FDR of < 0.05 and an absolute β-value difference (|Δβ-value|) of ≥ 0.02. For the abdominal obesity analysis, they were classified as either lean or having abdominal obesity based on WC thresholds. The analytical model was adjusted for age, sex, smoking habit, and BMI as covariates. DMPs for this comparison were identified using an FDR of < 0.05 and |Δβ-value| of ≥ 0.01 as thresholds. The genomic distribution and functional annotation of identified DMPs were analyzed using the built-in annotation features of the ChAMP package, which utilizes the Illumina HumanMethylationEPIC BeadChip annotation data based on the human reference genome hg19 (GRCh37). DMPs were classified according to their genomic location relative to CpG islands: CpG islands (regions with > 50% GC content, > 200 bp length, and observed/expected CpG ratio > 0.6), shores (regions within 2 kb of CpG islands), shelves (regions 2–4 kb from CpG islands), and open sea (regions > 4 kb from CpG islands). Additionally, DMPs were annotated based on their position relative to genes: promoter regions (transcription start site [TSS] to TSS-1500), gene body, and intergenic regions. The functional genomic context was further characterized by annotating DMPs to specific gene features including 5’ untranslated regions (5’UTR), first exons, gene bodies, and 3’ untranslated regions (3’UTR). After DMP identification from both analytical approaches, we conducted post-hoc analyses, including correlation analysis between BMI or WHR and methylation levels of the top 20 DMPs and correlation analysis between BMI or WHR and food or nutrient intake. For the BMI-based obesity analysis, differential methylation was assessed using linear regression models defined as: Inline graphic where β[ij] denotes the methylation level at CpG site i for participant j, µ represents the overall mean, α[i] is the obesity category effect, X[ijk] denotes the covariates (age, sex, smoking habit, WHR), and ε[ij] is the residual error. Our covariate selection was based on established causal frameworks in obesity research, prioritizing true confounders while avoiding potential mediators that could mask direct obesity-methylation associations. We included core demographic variables (age, sex), lifestyle factors (smoking), and relevant anthropometric measures (WHR for BMI analysis, BMI for WHR analysis). Our primary model balances adequate confounder control with preservation of direct obesity effects and optimal statistical efficiency. To evaluate robustness, we performed sensitivity analyses using alternative covariate combinations, which confirmed that our most significant findings (core DMPs) remained consistent across different adjustment strategies. Dietary assessment and correlation analysis We collected dietary intake data using a validated Semi-FFQ during the second follow-up survey (2005–2006) of the Ansung–Ansan cohort. This Semi-FFQ was specifically developed for Korean populations and has been validated for use in genome epidemiologic studies, demonstrating good reproducibility and validity for assessing dietary intake in Korean adults^[55]19,[56]20. The Semi-FFQ assessed the usual intake of 106 food items over the previous year, capturing information on consumption frequency and portion size. The daily intake of each food item was calculated by multiplying the reported frequency by the standard portion size. Nutrient intake, including macro- (carbohydrates, proteins, and fats) and micronutrients (vitamins and minerals), was estimated based on the Korean Food Composition Table. For the dietary intake analysis, we excluded participants with extreme energy intake values to reduce potential bias from misreporting. After removing outliers (values falling below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR of daily energy intake, where Q1 and Q3 are the first and third quartiles, respectively), 1,382 of 1,526 participants remained for the diet-methylation correlation analyses. All food item and nutrient intake values were standardized per 1,000 kcal of energy intake to account for differences in total energy consumption among participants, allowing for better comparison of dietary composition irrespective of total caloric intake. To evaluate the relationship between DNA methylation patterns and dietary intake, we conducted correlation analyses between methylation levels at the identified DMPs and various food items and nutrients. Spearman’s rank correlation was used to account for potential nonlinear relationships and nonnormal distributions in the dietary data. We calculated the correlation coefficients (r) and p-values for each correlation analysis, with statistical significance determined at p < 0.05. FDR correction was applied to control for multiple testing. Spearman’s rank correlation coefficient (ρ) was calculated according to the following formula, where d[i] represents the difference between ranks for each paired observation and n is the sample size: Inline graphic Hierarchical clustering analysis was performed to identify patterns of correlations between methylation at obesity-associated CpG sites and food or nutrient intake. We conducted this analysis separately for WHR- and BMI-associated DMPs to identify potential differences in the diet-methylation relationships between these two obesity phenotypes. The clustering was visualized using heatmaps generated with the “pheatmap” R package, with the color intensity representing strength and direction of correlations. For the food item correlation analysis, we focused on the major food groups commonly consumed in the Korean diet, including animal-based (pork belly, beef, fish), plant-based (vegetables, fruits, grains), and fermented foods (kimchi, fermented soybean soup). For the nutrient correlation analysis, we examined macronutrients (carbohydrates, proteins, and fats), fatty acid subtypes, fiber, cholesterol, and essential micronutrients (vitamins and minerals). These analyses aimed to identify specific dietary components that may influence the DNA methylation patterns associated with different obesity phenotypes. Statistical analyses All statistical analyses were performed using R software (version 4.2.1). Descriptive statistics are presented as mean ± standard deviation for continuous variables and frequencies (percentages) for categorical variables. Participant characteristics were compared between obesity groups using analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables, with post-hoc comparisons performed using Tukey’s honest significant difference test. For DNA methylation analysis, differential methylation was assessed using linear models implemented in the limma package through the ChAMP pipeline. The analytical models were adjusted for relevant covariates: age, sex, smoking habit, and WHR for BMI-based obesity analysis, and age, sex, smoking habit, and BMI for abdominal obesity analysis. Multiple testing correction was performed using the Benjamini-Hochberg false discovery rate (FDR) method. Correlation analyses between methylation levels at identified DMPs and anthropometric measurements (BMI, WHR) were performed using Spearman’s rank correlation to account for potential non-linear relationships. Similarly, correlations between methylation levels and dietary intake variables were assessed using Spearman’s correlation, with all dietary variables standardized per 1,000 kcal of energy intake. Statistical significance for correlations was determined at p < 0.05, with FDR correction applied for multiple testing. Results Clinical and biochemical parameters based on WHR categories We analyzed 1,526 participants, with 1,089 (71.4%) classified as having abdominal obesity based on the WHR criteria. Participants with abdominal obesity were significantly older (median, 61 vs. 53 years, p < 0.001) and showed different sex distribution (p = 0.003). Complete demographic and clinical characteristics are presented in Table [57]1. Table 1. Comparison of clinical characteristics and biochemical parameters between participants with and without abdominal obesity based on waist-to-hip ratio (WHR). Overall N = 1,526^1 Lean N = 437^1 Abdominal obesity N = 1,089^1 p-value^2 Sex 0.003 Male 807 (53%) 257 (59%) 550 (51%) Female 719 (47%) 180 (41%) 539 (49%) Age 58 (52, 68) 53 (50, 58) 61 (54, 69) < 0.001 Smoking habit 0.036 No 894 (59%) 237 (54%) 657 (60%) Yes (past smoker) 342 (22%) 116 (27%) 226 (21%) Yes (current smoker) 290 (19%) 84 (19%) 206 (19%) Waist-hip ratio (all) 0.92 (0.87, 0.98) 0.84 (0.81, 0.87) 0.95 (0.92, 1.00) < 0.001 Waist-hip ratio (male) 0.93 (0.89, 0.98) 0.87 (0.85, 0.89) 0.96 (0.93, 0.99) < 0.001 Waist-hip ratio (female) 0.92 (0.85, 0.98) 0.81 (0.78, 0.83) 0.94 (0.90, 1.00) < 0.001 BMI (kg/m²) 24.20 (22.30, 26.20) 22.70 (21.10, 24.40) 24.80 (23.00, 26.70) < 0.001 HbA1c (%) 5.60 (5.30, 6.50) 5.40 (5.20, 5.60) 5.60 (5.40, 6.70) < 0.001 Fasting Glucose (mg/dL) 94 (88, 115) 90 (86, 96) 96 (90, 123) < 0.001 BUN (mg/dL) ^3 15.2 (12.7, 18.0) 14.6 (12.3, 16.9) 15.5 (12.9, 18.4) < 0.001 Creatinine (mg/dL) 0.92 (0.83, 1.05) 0.94 (0.84, 1.06) 0.91 (0.82, 1.04) 0.030 AST (SGOT) (IU/L) 24 (20, 29) 23 (20, 28) 24 (20, 29) 0.2 ALT (SGPT) (IU/L) 20 (16, 28) 19 (15, 25) 21 (17, 29) < 0.001 Total Cholesterol (mg/dL) 188 (167, 213) 192 (168, 215) 187 (166, 212) 0.053 HDL-Cholesterol (mg/dL) 42 (35, 49) 45 (38, 53) 40 (34, 47) < 0.001 Triglyceride (mg/dL) 120 (86, 172) 100 (73, 136) 130 (93, 185) < 0.001 hs-CRP (mg/L) 0.67 (0.35, 1.38) 0.47 (0.26, 1.03) 0.78 (0.40, 1.54) < 0.001 White Blood Cell count (Thous/uL) 5.30 (4.50, 6.40) 5.00 (4.30, 5.90) 5.50 (4.60, 6.60) < 0.001 Red Blood Cell count (Mil/uL) 4.43 (4.12, 4.75) 4.49 (4.18, 4.79) 4.41 (4.10, 4.74) 0.009 Hemoglobin (g/dL) 13.70 (12.70, 14.80) 13.90 (12.90, 15.00) 13.60 (12.70, 14.60) < 0.001 Hematocrit (%) 41.1 (38.3, 44.2) 41.6 (38.7, 44.6) 40.9 (38.2, 43.9) 0.002 Platelet count (Thous/uL) 254 (217, 294) 249 (214, 282) 255 (219, 300) 0.003 Homocysteine (µmol/L) 13.0 (11.0, 15.8) 12.5 (10.6, 15.3) 13.2 (11.1, 16.0) 0.003 Obesity (BMI) < 0.001 Lean 501 (33%) 232 (53%) 269 (25%) Obese (Severe) 65 (4.3%) 1 (0.2%) 64 (5.9%) Obese (Stage I) 542 (36%) 81 (19%) 461 (42%) Overweight 418 (27%) 123 (28%) 295 (27%) [58]Open in a new tab ^1n (%); Median (Q1, Q3) ^2Pearson’s Chi-squared test; Wilcoxon rank sum test ³BUN, blood urea nitrogen (a biochemical marker reflecting kidney function and protein metabolism). As expected, anthropometric measurements showed marked differences between groups. The median WHR was substantially higher in the abdominal obesity group (0.95 vs. 0.84, p < 0.001), with consistent patterns in both sexes. BMI was also significantly higher in participants with abdominal obesity (24.80 vs. 22.70 kg/m², p < 0.001). Notably, only 25% of participants with abdominal obesity were classified as lean by BMI criteria compared with 53% in the non-abdominal obesity group (p < 0.001). Participants with abdominal obesity demonstrated significantly worse metabolic profiles. Glycemic markers were elevated, including HbA1c (5.60% vs. 5.40%, p < 0.001) and fasting glucose (96 vs. 90 mg/dL, p < 0.001). Renal function showed higher BUN (p < 0.001) but slightly lower creatinine levels (p = 0.030). Hepatic function revealed significantly higher ALT levels (p < 0.001), while AST showed no difference. The lipid profile displayed an atherogenic pattern in the abdominal obesity group, with significantly lower HDL-cholesterol (40 vs. 45 mg/dL, p < 0.001) and higher triglycerides (130 vs. 100 mg/dL, p < 0.001). Total cholesterol showed a non-significant trend toward lower levels (p = 0.053). Inflammatory markers were consistently elevated in the abdominal obesity group, including hs-CRP (0.78 vs. 0.47 mg/L, p < 0.001) and white blood cell count (p < 0.001). Hematological parameters showed lower red blood cell indices, hemoglobin, and hematocrit (all p < 0.01), while platelet count was higher (p = 0.003). Homocysteine levels were also elevated in the abdominal obesity group (p = 0.003). Clinical and biochemical parameters based on BMI categories Participants were stratified into four BMI categories: lean (n = 501, 32.8%), overweight (n = 418, 27.4%), obese stage I (n = 542, 35.5%), and severely obese (n = 65, 4.3%). Sex distribution varied significantly across categories (p = 0.001), with women comprising 71% of the severely obese group compared to 45–47% in other categories. Age did not differ significantly between groups (p = 0.13). Complete characteristics are presented in Table [59]2. Table 2. Comparison of clinical characteristics and biochemical parameters between participants with and without abdominal obesity based on body-mass index (BMI). Overall N = 1,526^1 Lean N = 501^1 Overweight N = 418^1 Obese (Stage I) N = 542^1 Obese (Severe) N = 65^1 p-value^2 Sex 0.001 Male 807 (53%) 264 (53%) 224 (54%) 300 (55%) 19 (29%) Female 719 (47%) 237 (47%) 194 (46%) 242 (45%) 46 (71%) Age 58 (52, 68) 59 (52, 69) 56 (52, 67) 58 (52, 67) 59 (53, 67) 0.13 Smoking habit 0.012 No 894 (59%) 300 (60%) 241 (58%) 311 (57%) 42 (65%) Yes (past smoker) 342 (22%) 93 (19%) 92 (22%) 138 (25%) 19 (29%) Yes (current smoker) 290 (19%) 108 (22%) 85 (20%) 93 (17%) 4 (6.2%) Waist-hip ratio (all) 0.92 (0.87, 0.98) 0.89 (0.84, 0.93) 0.92 (0.87, 0.98) 0.95 (0.91, 1.00) 0.99 (0.93, 1.02) < 0.001 Waist-hip ratio (male) 0.93 (0.89, 0.98) 0.90 (0.86, 0.94) 0.93 (0.89, 0.98) 0.96 (0.92, 1.00) 0.99 (0.95, 1.01) < 0.001 Waist-hip ratio (female) 0.92 (0.85, 0.98) 0.87 (0.82, 0.93) 0.92 (0.84, 0.98) 0.94 (0.89, 1.00) 1.00 (0.92, 1.03) < 0.001 BMI (kg/m²) 24.20 (22.30, 26.20) 21.50 (20.50, 22.30) 23.90 (23.40, 24.50) 26.50 (25.60, 27.70) 31.20 (30.70, 32.60) < 0.001 HbA1c (%) 5.60 (5.30, 6.50) 5.50 (5.30, 5.60) 5.50 (5.30, 6.30) 5.80 (5.40, 6.80) 6.60 (5.60, 7.00) < 0.001 Fasting Glucose (mg/dL) 94 (88, 115) 90 (86, 98) 93 (88, 110) 98 (91, 125) 106 (94, 124) < 0.001 BUN (mg/dL) ^3 15.2 (12.7, 18.0) 15.2 (12.6, 17.7) 14.8 (12.4, 17.6) 15.4 (13.2, 18.7) 16.0 (13.3, 17.8) 0.017 Creatinine (mg/dL) 0.92 (0.83, 1.05) 0.90 (0.82, 1.01) 0.93 (0.83, 1.06) 0.95 (0.84, 1.07) 0.87 (0.82, 1.04) < 0.001 AST (SGOT) (IU/L) 24 (20, 29) 23 (20, 27) 23 (20, 28) 24 (21, 29) 25 (22, 34) 0.010 ALT (SGPT) (IU/L) 20 (16, 28) 18 (15, 24) 20 (16, 26) 23 (18, 32) 25 (19, 36) < 0.001 Total Cholesterol (mg/dL) 188 (167, 213) 187 (166, 209) 190 (168, 215) 188 (167, 214) 191 (171, 218) 0.7 HDL-Cholesterol (mg/dL) 42 (35, 49) 45 (38, 53) 41 (35, 47) 39 (34, 46) 40 (34, 47) < 0.001 Triglyceride (mg/dL) 120 (86, 172) 102 (74, 140) 124 (88, 179) 135 (98, 188) 155 (97, 204) < 0.001 hs-CRP (mg/L) 0.67 (0.35, 1.38) 0.45 (0.26, 1.03) 0.63 (0.35, 1.24) 0.87 (0.47, 1.59) 1.53 (0.95, 2.42) < 0.001 White Blood Cell count (Thous/uL) 5.30 (4.50, 6.40) 5.00 (4.20, 6.10) 5.30 (4.50, 6.30) 5.50 (4.70, 6.50) 6.00 (5.10, 7.10) < 0.001 Red Blood Cell count (Mil/uL) 4.43 (4.12, 4.75) 4.33 (4.04, 4.63) 4.45 (4.19, 4.78) 4.52 (4.20, 4.84) 4.38 (4.14, 4.75) < 0.001 Hemoglobin (g/dL) 13.70 (12.70, 14.80) 13.30 (12.50, 14.50) 13.80 (12.80, 14.80) 13.90 (12.80, 14.90) 13.40 (12.60, 14.70) < 0.001 Hematocrit (%) 41.1 (38.3, 44.2) 40.4 (37.8, 43.5) 41.4 (38.6, 44.3) 41.5 (38.6, 44.6) 40.1 (37.8, 43.3) 0.001 Platelet count (Thous/uL) 254 (217, 294) 251 (209, 290) 258 (224, 296) 253 (217, 295) 269 (227, 311) 0.070 Homocysteine (µmol/L) 13.0 (11.0, 15.8) 12.9 (10.7, 15.8) 13.2 (11.1, 16.1) 13.0 (11.1, 15.8) 12.9 (11.0, 16.1) 0.4 Abdominal obesity < 0.001 Lean 437 (29%) 232 (46%) 123 (29%) 81 (15%) 1 (1.5%) Abdominal obesity 1,089 (71%) 269 (54%) 295 (71%) 461 (85%) 64 (98%) [60]Open in a new tab ^1n (%); Median (Q1, Q3) ^2Pearson’s Chi-squared test; Kruskal-Wallis rank sum test ³BUN, blood urea nitrogen (a biochemical marker reflecting kidney function and protein metabolism). WHR increased progressively across BMI categories from 0.89 in the lean group to 0.99 in the severely obese group (p < 0.001), with consistent patterns in both sexes. The prevalence of abdominal obesity dramatically increased from 54% in the lean group to 98% in the severely obese group (p < 0.001). Metabolic parameters worsened progressively with increasing BMI. Glycemic markers showed clear dose-response relationships, with HbA1c increasing from 5.50% in the lean group to 6.60% in the severely obese group (p < 0.001), and fasting glucose increasing from 90 to 106 mg/dL (p < 0.001). Hepatic enzymes demonstrated significant elevation with higher BMI, with both AST (p = 0.010) and ALT (p < 0.001) showing progressive increases. Lipid parameters revealed marked differences across BMI categories. HDL-cholesterol decreased from 45 mg/dL in the lean group to 39–40 mg/dL in obese groups (p < 0.001), while triglycerides progressively increased from 102 to 155 mg/dL (p < 0.001). Total cholesterol levels did not differ significantly (p = 0.7). Inflammatory markers showed strong BMI associations, with hs-CRP increasing dramatically from 0.45 mg/L in the lean group to 1.53 mg/L in the severely obese group (p < 0.001). White blood cell count also increased progressively with BMI (p < 0.001). Hematological parameters varied significantly, with red blood cell count, hemoglobin, and hematocrit peaking in the obese stage I group (all p ≤ 0.001). Homocysteine levels showed no significant differences across BMI categories (p = 0.4). Identification of DMPs associated with abdominal obesity EWAS identified multiple DMPs that were significantly associated with abdominal obesity. Among the 23 DMPs that met our stringent criteria of absolute delta-beta value ≥ 0.02 and FDR < 0.05, those ranked by statistical significance are presented in Table [61]3. Among these, cg10323433 in the promoter region (TSS1500) of the serotonin receptor 2 A (HTR2A) on chromosome 13 showed the strongest association (Δβ = −0.023, p = 3.20 × 10^−57, FDR = 3.64 × 10^−56). This hypomethylation pattern suggests potential regulatory implications in the serotonin signaling pathways relevant to obesity. Table 3. DMPs associated with abdominal obesity. Rank CpG Chromosome Location Gene Feature CpG island △β p-value Fdr 1 cg10323433 13 47,471,562 HTR2A TSS1500 opensea −0.023132 3.20E-57 3.64E-56 2 cg06032337 6 29,648,468 IGR opensea 0.020440 7.41E-55 7.26E-54 3 cg10501210 1 207,997,020 IGR opensea −0.025675 2.77E-54 2.62E-53 4 cg13039251 5 32,018,601 PDZD2 Body opensea −0.023529 1.57E-53 1.42E-52 5 cg15830864 1 44,877,749 RNF220 5’UTR shelf −0.022073 4.55E-47 2.90E-46 6 cg14193434 13 67,802,884 PCDH9 5’UTR shore −0.021310 3.85E-46 2.33E-45 7 cg00740914 2 66,652,111 IGR shore −0.021945 1.65E-44 9.27E-44 8 cg27286120 13 67,802,706 PCDH9 5’UTR shore −0.022547 9.68E-43 5.00E-42 9 cg16781885 4 172,973,467 GALNTL6 Body opensea −0.021510 7.72E-40 3.51E-39 10 cg16867657 6 11,044,877 ELOVL2 TSS1500 island 0.020245 9.09E-40 4.13E-39 11 cg05825244 20 2,730,488 EBF4 Body island 0.020960 9.42E-40 4.28E-39 12 cg22943590 2 66,648,797 IGR shelf −0.025578 2.29E-39 1.02E-38 13 cg08124030 3 149,095,283 TM4SF1 1stExon opensea −0.021363 6.79E-39 2.97E-38 14 cg04875128 15 31,775,895 OTUD7A Body island 0.021373 4.51E-36 1.76E-35 15 cg02228185 17 3,379,567 ASPA 1stExon opensea −0.021499 2.43E-35 9.20E-35 16 cg11807280 2 66,654,644 IGR shore −0.023939 5.17E-35 1.94E-34 17 cg13001142 6 147,528,521 STXBP5 Body shelf −0.022820 6.11E-35 2.28E-34 18 cg23379566 3 149,094,892 TM4SF1 Body opensea −0.020765 1.04E-34 3.86E-34 19 cg16714760 4 155,386,347 DCHS2 Body opensea −0.021586 4.97E-29 1.49E-28 20 cg18826637 2 145,116,633 IGR opensea −0.021172 1.99E-27 5.66E-27 21 cg16440561 2 220,312,854 SPEG Body island 0.020019 1.98E-13 3.57E-13 22 cg10661558 21 15,443,159 IGR island 0.020591 6.66E-04 8.42E-04 23 cg09516963 12 68,042,445 DYRK2 TSS200 island 0.023340 1.45E-03 1.80E-03 [62]Open in a new tab The second most significant DMP (cg06032337) was located in an intergenic region on chromosome 6 and exhibited hypermethylation (Δβ = 0.020, p = 7.41 × 10^−55, FDR = 7.26 × 10^−54). Another notable hypomethylated position was cg10501210 on chromosome 1 (Δβ = −0.026, p = 2.77 × 10^−54, FDR = 2.62 × 10^−53), also in an intergenic region. Several genes with potential functional relevance to metabolic processes were identified among the top DMPs, including PDZ domain containing 2 (PDZD2) (cg13039251), ring finger protein 220 (RNF220) (cg15830864), and protocadherin 9 (PCDH9) (cg14193434 and cg27286120). Interestingly, PCDH9 was represented by two distinct DMPs (ranked 6th and 8th) in the 5′UTR region and exhibiting consistent hypomethylation patterns (Δβ = −0.021 and − 0.023, respectively), indicating a robust association with abdominal obesity (Fig. [63]2A). Fig. 2. [64]Fig. 2 [65]Open in a new tab Methylation patterns of the top differentially methylated positions (DMPs) associated with obesity phenotypes. (A) Box plots showing the methylation levels (β-values) of the top 10 DMPs associated with abdominal obesity. Comparisons are shown between individuals classified as lean and those with abdominal obesity (determined by WHR). Each plot displays the CpG site ID and its corresponding delta-beta value. Most DMPs (9/10) showed hypomethylation in abdominal obesity, with cg06032337 being the only site showing hypermethylation. (B) Box plots illustrating methylation levels (β-values) of the top 10 DMPs associated with BMI-defined obesity categories. For each CpG site, methylation patterns were compared across four BMI categories: lean, overweight, stage I obesity, and severe obesity. A progressive change in methylation was observed with increasing obesity severity, with 8/10 and 2/10 sites (cg15229836 and cg26237810) showing hypomethylation and hypermethylation, respectively. Analysis of the genomic distribution of these DMPs revealed that 45% (9/20), 15% (3/20), and 30% (6/20) were located in gene bodies, promoter regions (TSS1500 or 1stExon), and intergenic regions, respectively. Regarding the CpG context, most DMPs (55%, 11/20) were located in “opensea” regions distant from the CpG islands, whereas 25% (5/20) were in “shore” regions adjacent to the CpG islands. The magnitude of methylation differences (Δβ) ranged from − 0.026 to 0.021, with most sites (85%, 17/20) showing hypomethylation in individuals with abdominal obesity. All identified DMPs were highly significant, with FDR-corrected p-values ranging from 3.64 × 10^−56 to 5.66 × 10^−27, indicating robust epigenetic associations with abdominal obesity. Identification of DMPs associated with BMI-related obesity EWAS identified significant DMPs associated with BMI-related obesity across different weight categories. After applying our strict criteria of absolute delta-beta-value ≥ 0.1 and FDR < 0.05, we identified 0, 51, and 1931 DMPs for the lean vs. overweight, lean vs. obese (stage I), and lean vs. severely obese comparisons, respectively. Table [66]4 presents the top 20 DMPs ranked by statistical significance, derived from the average ranks of probes in the lean vs. obese (stage I) and lean vs. severely obese comparisons, highlighting the most consistently significant methylation changes across the obesity spectrum. Table 4. Top 20 DMPs associated with BMI-related obesity. Rank CpG Chromosome Location Gene Feature CpG island Lean to overweight Lean to obese (Stage I) Lean to obese (Severe) △β p-value Fdr Rank △β p-value Fdr Rank △β p-value Fdr Rank 1 cg10501210 1 207,997,020 IGR opensea −0.006895 2.06E-13 9.07E-11 - −0.012667 1.08E-46 2.30E-44 2 −0.021296 1.16E-28 1.47E-26 1 2 cg10323433 13 47,471,562 HTR2A TSS1500 opensea −0.005930 1.79E-12 1.30E-10 - −0.011156 5.97E-45 5.66E-43 3 −0.019079 1.75E-28 1.98E-26 2 3 cg07888917 2 12,108,682 IGR opensea −0.005573 8.90E-13 1.01E-10 - −0.010570 7.13E-47 1.70E-44 1 −0.015439 1.79E-22 2.57E-21 9 4 cg15570860 11 8,986,840 TMEM9B TSS1500 shore 0.008981 6.19E-12 2.53E-10 - 0.017119 7.58E-44 5.08E-42 5 0.027949 1.20E-25 3.80E-24 6 5 cg09408571 1 101,003,634 GPR88 TSS200 shore −0.006117 4.46E-12 2.09E-10 - −0.011655 3.42E-44 2.53E-42 4 −0.018413 1.50E-24 3.51E-23 8 5 cg15229836 2 239,553,380 IGR opensea 0.005740 2.49E-12 1.52E-10 - 0.010513 9.61E-43 4.76E-41 7 0.017446 9.11E-26 3.01E-24 5 7 cg26237810 1 200,669,214 IGR opensea 0.006094 1.06E-11 3.54E-10 - 0.011657 6.01E-43 3.13E-41 6 0.016933 1.46E-20 1.47E-19 11 7 cg07412545 1 101,003,924 GPR88 1stExon shore −0.007352 2.57E-11 6.43E-10 - −0.014066 1.44E-41 5.46E-40 10 −0.023299 3.62E-25 1.00E-23 7 7 cg06223162 1 101,003,688 GPR88 TSS200 shore −0.007690 2.53E-11 6.37E-10 - −0.014453 4.73E-40 1.32E-38 14 −0.025993 3.25E-28 3.10E-26 3 10 cg11839163 7 134,617,869 CALD1 Body opensea −0.005797 1.40E-10 2.25E-09 - −0.010901 1.33E-37 2.52E-36 16 −0.020135 1.45E-27 9.75E-26 4 11 cg13139335 8 96,614,915 C8orf37-AS1 Body opensea 0.006408 8.35E-11 1.52E-09 - 0.012387 3.54E-40 1.01E-38 13 0.019309 5.85E-22 7.59E-21 10 12 cg07965995 17 811,389 NXN Body opensea 0.005589 9.81E-11 1.72E-09 - 0.010877 2.63E-40 7.72E-39 11 0.015977 8.81E-20 7.81E-19 13 13 cg01791648 3 101,232,718 SENP7 TSS1500 shore −0.005547 9.66E-12 3.33E-10 - −0.010287 2.93E-42 1.29E-40 9 −0.013725 3.44E-17 2.10E-16 16 14 cg08124030 3 149,095,283 TM4SF1 1stExon opensea −0.006060 1.25E-11 3.95E-10 - −0.011402 1.62E-42 7.59E-41 8 −0.014513 5.29E-16 2.77E-15 19 15 cg11218872 3 193,988,737 IGR shore −0.005648 5.80E-11 1.16E-09 - −0.010818 3.41E-40 9.81E-39 12 −0.014571 5.22E-17 3.11E-16 17 16 cg23719650 3 193,988,507 IGR shore −0.005483 9.00E-11 1.61E-09 - −0.010540 1.43E-39 3.70E-38 15 −0.014407 3.08E-17 1.89E-16 15 17 cg16932827 3 193,988,639 IGR shore −0.005406 1.01E-09 1.07E-08 - −0.010480 7.18E-36 1.07E-34 17 −0.013163 1.34E-13 5.29E-13 23 18 cg00440468 1 19,110,768 IGR island −0.007374 4.39E-09 3.59E-08 - −0.013658 9.88E-31 8.31E-30 20 −0.019034 4.01E-14 1.68E-13 22 19 cg10725542 3 149,094,653 TM4SF1 Body opensea −0.005401 3.88E-09 3.24E-08 - −0.010011 1.46E-32 1.49E-31 18 −0.011192 6.06E-10 1.66E-09 28 20 cg23379566 3 149,094,892 TM4SF1 Body opensea −0.006141 7.88E-09 5.87E-08 - −0.011261 6.21E-31 5.32E-30 19 −0.012961 6.18E-10 1.69E-09 29 [67]Open in a new tab The most significant DMP was cg10501210 in an intergenic region on chromosome 1, exhibiting progressive hypomethylation with increasing BMI category (Δβ = −0.006895 for lean to overweight, Δβ = −0.012667 for lean to obese stage I, and Δβ = −0.021296 for lean to severely obese; p = 1.16 × 10^−28, FDR = 1.47 × 10^−26 in severe obesity). This progressive methylation pattern across BMI categories indicates a dose–response relationship between methylation levels and obesity severity (Fig. [68]2B). The second-ranked DMP, cg10323433, consistently showed hypomethylation with increasing obesity severity (Δβ = −0.005930, − 0.011156, and − 0.019079 across the three categories, respectively). Notably, this CpG site was also identified as the top-ranking DMP associated with abdominal obesity. Our analysis revealed several genes with multiple DMPs. The G protein-coupled receptor 88 (GPR88) harbored three DMPs among the top 10 (cg09408571, cg07412545, and cg06223162), all located in the promoter regions (TSS200 or 1stExon) and consistently showed hypomethylation patterns across the obesity categories. Similarly, Transmembrane 4-L Six Family Member 1 (TM4SF1) contained three DMPs (cg08124030, cg10725542, and cg23379566) among the top 20. Interestingly, three consecutive DMPs (cg11218872, cg23719650, and cg16932827) in an intergenic region on chromosome 3 showed coordinated hypomethylation patterns, indicating a potential regulatory hotspot associated with obesity. Analysis of genomic distribution revealed that 35% (7/20), 20% (4/20), and 45% (9/20) of DMPs were located in gene promoter regions (TSS1500, TSS200, or 1stExon), gene bodies, and intergenic regions, respectively. Regarding the CpG context, 50% (10/20) were in “opensea” regions distant from the CpG islands, 45% (9/20) in “shore” regions, and only 5% (1/20) were directly within the CpG islands. The magnitude of methylation differences was consistently greater in severely obese individuals than in overweight or class I obese individuals (severe obesity group: Δβ values, from − 0.026 to 0.028). Most DMPs (75%, 15/20) exhibited hypomethylation with increasing BMI, indicating a predominant pattern of decreased methylation associated with obesity progression. Pathway enrichment analysis of obesity-associated DMPs To understand the biological significance of obesity-associated methylation changes, we performed pathway enrichment analysis using the 1,931 DMPs identified in the lean vs. severely obese comparison. Gene annotation identified 936 unique genes, of which 836 genes (89.3%) were successfully mapped for pathway analysis. KEGG pathway enrichment analysis revealed four significantly enriched pathways (FDR < 0.05) directly relevant to obesity pathophysiology (Supplementary Figure S2). The most significantly enriched pathway was “Glutamatergic synapse” (p.adjust = 7.33 × 10⁻⁴, 16 genes), which is crucial for hypothalamic appetite regulation and energy homeostasis. “Neuroactive ligand-receptor interaction” (p.adjust = 2.16 × 10⁻³, 30 genes) was the second most significant pathway, encompassing key hormonal signaling components including leptin, ghrelin, and insulin receptors. Additionally, “Calcium signaling pathway” (p.adjust = 2.14 × 10⁻², 21 genes) and “Neuroactive ligand signaling” (p.adjust = 4.32 × 10⁻², 17 genes) were significantly enriched, reflecting calcium-mediated metabolic processes and neuroendocrine regulation in obesity development. These findings indicate that obesity-associated DNA methylation changes primarily target central nervous system pathways involved in appetite control, energy balance, and metabolic regulation, providing molecular insights into the epigenetic mechanisms underlying obesity pathogenesis. Correlation between the methylation levels of the top DMPs and BMI or WHR Our comprehensive analysis of DNA methylation patterns in relation to obesity metrics revealed significant associations between methylation levels at specific CpG sites and both WHR and BMI (Fig. [69]3). Fig. 3. [70]Fig. 3 [71]Open in a new tab Correlation of DNA methylation levels at the top CpG sites with obesity measures. (A) Scatter plots showing the relationship between the WHR and DNA methylation levels (beta value) at the top 10 WHR-associated CpG sites. Each panel represents a different CpG site (labeled at top), with individual data points colored by the correlation direction (blue for negative, red for positive). The regression lines show the linear relationship between WHR and methylation level. The strongest negative correlation was observed at cg10501210 (r = − 0.521, p = 7.84 × 10 − ^107), whereas the strongest positive correlation was found at cg16867657 (r = 0.459, p = 1.81 × 10^−80). The correlation coefficients (r) and p-values are shown in each panel. Most sites (8 of 10) demonstrated negative correlations with WHR, indicating that increased abdominal obesity is predominantly associated with hypomethylation at these loci. (B) Scatter plots depicting the relationship between BMI and DNA methylation levels at the top 10 BMI-associated CpG sites. Similar to the WHR-associated sites, most (7 of 10) of the top BMI-associated CpG sites exhibited negative correlations, with correlation coefficients ranging from − 0.452 (cg10501210, p = 7.25 × 10^−78) to 0.438 (cg15570860, p = 1.88 × 10^−72). The most significant correlations were observed at cg10501210 and cg10323433, also significantly associated with WHR, indicating their potential role in regulating genes involved in general adiposity and fat distribution. The color scale on the right indicates the direction and strength of the correlation, with blue and red representing negative and positive correlations, respectively. All identified correlations remained significant after FDR correction for multiple testing. This figure demonstrates the robust relationship between DNA methylation at specific CpG sites and obesity measures in the KoGES cohort. Figure [72]3A depicts the relationship between WHR and DNA methylation levels for the top 10 CpG sites that showed substantial correlations, with coefficients ranging from − 0.512 (cg10323433, p = 7.60 × 10^−103) to 0.459 (cg16867657, p = 1.81 × 10^−80). Notably, 8 of the 10 top sites were negatively correlated with WHR, indicating that increased abdominal obesity is predominantly associated with hypomethylation at these loci. The strongest negative correlation was observed at cg10501210 (r = − 0.521, p = 7.84 × 10^−107), whereas the most significant positive correlation was found at cg16867657 (r = 0.459, p = 1.81 × 10^−80). Figure [73]3B shows the relationship between BMI and DNA methylation levels for the top 10 BMI-associated CpG sites. Seven of 10 top BMI-associated CpG sites exhibited negative correlations, with correlation coefficients ranging from − 0.452 (cg10501210, p = 7.25 × 10^−78) to 0.438 (cg15570860, p = 1.88 × 10^−72). The strongest negative and strongest positive associations were observed at cg10501210 (r = − 0.452, p = 7.25 × 10^−78) and cg15570860 (r = 0.438, p = 1.88 × 10^−72), respectively. Interestingly, we identified cg10501210 and cg10323433 as overlapping CpG sites significantly correlated with WHR and BMI, indicating their potential role in regulating genes involved in general and abdominal obesity. All identified correlations remained significant after FDR correction for multiple testing, with q-values ranging from 1.57 × 10^−105 to 6.09 × 10[−77] for the top associations. Supplementary Tables S1 and S2 show 20 and 213 CpG sites significantly correlated with WHR (|r| > 0.275, FDR < 1 × 10^−28) and BMI (|r| > 0.4, FDR < 7 × 10^−59), respectively. The predominance of negative correlations in both analyses indicates that obesity is generally associated with widespread DNA hypomethylation at these regulatory sites, which may contribute to altered gene expression patterns in obesity-related pathways. When comparing our correlation analysis with the differential methylation analysis (Tables [74]3, 4 and 5), we observed consistencies and notable differences. The top DMPs for abdominal obesity in Table [75]3 largely overlapped with the top WHR-correlated CpG sites in our correlation analysis, with cg10323433, cg06032337, cg10501210, and cg16867657 appearing prominently in both analyses. Similarly, the top BMI-associated DMPs in Table [76]4 considerably overlapped with our BMI correlation results, particularly for cg10501210, cg10323433, and cg15570860. However, the correlation analysis revealed stronger statistical associations than the differential methylation approach, with substantially lower p-values (e.g., 7.84 × 10^−107 vs. 3.20 × 10^−57 for cg10323433 in WHR analysis). Additionally, the correlation analysis identified several CpG sites not captured among the top hits in the differential methylation approach, indicating that linear correlation may detect more subtle but consistent methylation changes across the obesity spectrum rather than just differences between categorical groups. Based on these correlation patterns, we next examined which CpG sites were specifically or commonly associated with each obesity measure to identify distinct and shared epigenetic signatures. CpG sites specifically and commonly associated with WHR and BMI Building upon the correlation analysis results, we conducted a comparative analysis to distinguish between epigenetic markers specifically associated with abdominal obesity (WHR) and overall obesity (BMI). Supplementary Table S3 shows distinct sets of CpG sites uniquely associated with either WHR or BMI, as well as common sites associated with both measures. Among the top DMPs, 16 CpG sites were specifically associated with abdominal obesity (WHR), including cg06032337, cg13039251, cg15830864, and cg14193434, which were predominantly located in gene bodies and intergenic regions, indicating potential regulatory mechanisms specific to fat distribution patterns rather than overall adiposity. Similarly, 16 CpG sites were uniquely associated with BMI-related obesity, including cg07888917, cg15570860, cg09408571, and cg15229836, with several located in the promoter regions of metabolic regulatory genes, particularly multiple sites in the GPR88 gene (cg09408571, cg07412545, and cg06223162), which encodes a G protein-coupled receptor implicated in energy homeostasis. Notably, four CpG sites were significantly associated with both WHR and BMI: cg10323433, cg10501210, cg08124030, and cg23379566. Among these, cg10323433 showed the strongest correlation with WHR (r = − 0.512) and cg10501210 with BMI (r = − 0.452), as demonstrated in our correlation analysis. This indicates a potential common epigenetic mechanism involving serotonin signaling that may influence overall adiposity and fat distribution. The second common marker, cg10501210, exhibited strong negative correlations with both measures and was identified as the top-ranked DMP for BMI-related severe obesity. The remaining two common DMPs (cg08124030 and cg23379566) were located in the TM4SF1 gene, encoding a cell surface antigen and implicated in cellular migration and tumor progression. The identification of shared and distinct DMPs indicates that although certain epigenetic mechanisms may be common to overall and abdominal obesity, specific epigenetic signatures distinguish these phenotypes. These findings highlight the complex epigenetic architecture underlying different obesity manifestations and indicate that targeted interventions might need to consider the specific obesity phenotype. Intercorrelations among obesity-associated CpG sites To assess the degree of coordination among the identified obesity-related epigenetic markers, we conducted comprehensive pairwise correlation analysis between all CpG sites within each phenotype. Among the 43 BMI-associated CpG sites, 630 of 903 possible pairs (69.8%) showed strong correlations (|r| ≥ 0.5), with 327 pairs (36.2%) exhibiting extremely strong correlations (|r| ≥ 0.8). Similarly, WHR-associated CpG sites demonstrated even higher intercorrelation, with 202 of 253 pairs (79.8%) showing strong correlations and 118 pairs (46.6%) showing extremely strong correlations. Hierarchical clustering analysis based on correlation patterns identified three distinct subgroups of co-correlated CpG sites for each obesity phenotype (Supplementary Figure S3). These correlation-based subgroups suggest coordinated epigenetic regulation within functional modules, indicating that obesity-related DNA methylation changes occur in a coordinated rather than random pattern. The higher proportion of strong correlations among WHR-associated CpG sites compared to BMI-associated sites suggests that abdominal obesity involves more tightly coordinated epigenetic mechanisms, potentially reflecting the specific metabolic pathways involved in central fat distribution. Association between methylation levels of CpG sites and food intake in relation to WHR Our analysis revealed significant associations between DNA methylation and dietary patterns in relation to abdominal obesity. From the correlation data between CpG sites and food items, we identified numerous significant correlations between DNA methylation and various food item consumption (Supplementary Table S4). In addition, Supplementary Table S5 reveals strong correlations between CpG methylation and specific nutrient intakes, particularly fat and carbohydrates. The strongest correlations were observed between CpG methylation and pork belly intake, with multiple CpG sites showing significant associations. The most prominent correlations were found at cg15830864 (r = 0.2616, p = 4.57E − 23), cg27286120 (r = 0.2545, p = 7.21E − 22), cg14193434 (r = 0.2540, p = 8.66E − 22), and cg11807280 (r = 0.2506, p = 3.07E − 21), whereas cg05825244 showed a significant negative correlation (r = − 0.2529, p = 1.32E − 21). Notably, Table [77]3 presents that most of these CpG sites (cg15830864, cg27286120, cg14193434, and cg11807280) showed significant hypomethylation in abdominal obesity (Δβ ranging from − 0.021 to − 0.024), whereas cg05825244 showed hypermethylation (Δβ = 0.021), consistent with their correlations with food items. Other mushroom consumption was significantly associated with methylation at multiple CpG sites, including cg08124030 (r = 0.2340, p = 1.21E − 18), cg16781885 (r = 0.2264, p = 1.59E − 17), and cg23379566 (r = 0.2201, p = 1.29E − 16), whereas cg16867657 (r = − 0.2226, p = 5.54E − 17) and cg04875128 (r = − 0.2215, p = 8.08E − 17) showed negative correlations. In addition, cg08124030, cg16781885, and cg23379566 showed hypomethylation in abdominal obesity (Δβ = −0.021 to − 0.022) and strong negative correlations with WHR (r = − 0.480, − 0.468, and − 0.486, respectively), whereas cg16867657 and cg04875128 showed hypermethylation (Δβ = 0.020 and 0.021) and positive correlations with WHR (r = 0.459 and 0.409) (Table [78]3 and Supplementary Table S2). Fermented foods showed distinct methylation patterns. Water kimchi consumption showed significant correlations with multiple CpG sites, including positive correlations with cg04875128 (r = 0.2191, p = 1.73E − 16) and cg16867657 (r = 0.2133, p = 1.10E − 15), and negative correlations with cg22943590 (r = − 0.2191, p = 1.76E − 16) and cg02228185 (r = − 0.2183, p = 2.25E − 16). Similarly, fermented soybean soup consumption was significantly correlated with cg22943590 (r = − 0.1659, p = 5.46E − 10), cg02228185 (r = − 0.1636, p = 9.47E − 10), and cg16714760 (r = − 0.1589, p = 2.85E − 09). Figure [79]4A shows that the hierarchical clustering of the correlations between CpG sites and food items revealed distinct patterns in relation to WHR. The heatmap shows two main clusters of CpG sites with opposing correlation patterns. The top cluster (including cg04875128, cg16867657, cg05825244, and cg06032337) showed positive correlations with fermented soybean soup, water kimchi, cabbage, and fermented fish but negative correlations with pork belly, ramen, mushrooms, green tea, cookies, processed meat, fried chicken, and eel. Notably, the top cluster sites showed hypermethylation in individuals with abdominal obesity (positive Δβ values) and positive correlations with WHR (Table [80]3 and Supplementary Table S2). The bottom cluster (including cg18826637, cg23379566, cg08124030, cg16781885, cg10323433, cg10501210, cg14193434, cg27286120, cg13001142, cg15830864, cg13039251, cg02228185, cg22943590, cg00740914, cg11807280, and cg16714760) was negatively correlated with traditional fermented foods but positively correlated with animal-based and processed foods. These CpG sites predominantly showed hypomethylation in abdominal obesity (negative Δβ values) and negative correlations with WHR (correlation coefficient, 0.280–0.521, Supplementary Table S2). Notably, the top DMP for WHR, cg10323433 (HTR2A), showing significant hypomethylation in abdominal obesity (Δβ = −0.023) and a strong negative correlation with WHR (r = − 0.512), was significantly correlated with other mushrooms (mushroom varieties excluding oyster mushrooms, including shiitake, enoki, and king oyster mushrooms; r = 0.2075, p = 6.55E − 15), water kimchi (r = − 0.2073, p = 6.97E − 15), vegetable wraps (r = 0.1820, p = 9.38E − 12), and pork belly (r = 0.1670, p = 4.18E − 10). Similarly, cg10501210, which was the third-ranked DMP for WHR (Δβ = −0.026) with the strongest negative correlation with WHR (r = − 0.521), was significantly correlated with water kimchi (r = − 0.2054, p = 1.26E − 14), other mushrooms (r = 0.2047, p = 1.52E − 14), and vegetable wraps (r = 0.1921, p = 5.87E − 13). Fig. 4. [81]Fig. 4 [82]Open in a new tab Hierarchical clustering analysis of correlations between WHR-associated CpG methylation sites and dietary factors. (A) Heatmap showing clustered correlations between WHR-associated CpG sites (rows) and food items (columns) commonly consumed in the Korean diet. The color scale represents the correlation strength and direction, with red and blue indicating positive (r = 0–0.5) and negative correlations (r = 0 to − 0.5), respectively. Asterisks (*) denote statistically significant correlations (p < 0.05). Two main clusters of CpG sites emerged: the top cluster (cg04875128, cg16867657, cg05825244, and cg06032337) was positively correlated with traditional fermented foods (fermented soybean soup, water kimchi) but negatively correlated with animal-based and processed foods (pork belly, ramen, mushrooms). However, the bottom cluster was positively correlated with animal and processed foods but negatively correlated with traditional fermented foods. These patterns align with the methylation status observed in abdominal obesity, as sites in the top cluster showed hypermethylation in individuals with abdominal obesity, whereas sites in the bottom cluster predominantly exhibited hypomethylation. (B) Heatmap depicting the clustered correlations between WHR-associated CpG sites (rows) and nutrient intakes (columns). Similar to the food correlation patterns, two distinct clusters of CpG sites were evident. The top cluster was positively correlated with carbohydrates and fiber but negatively correlated with fat, cholesterol, and retinol. However, the bottom cluster was strongly positively correlated with fat and cholesterol but negatively correlated with carbohydrates. This inverse relationship between the fat and carbohydrate correlation patterns was consistent across most examined CpG sites and aligned with their methylation status in relation to abdominal obesity. The strongest correlations were observed between fat consumption and methylation at several CpG sites, including cg13001142 (r = 0.3308), cg16714760 (r = 0.3258), and cg11807280 (r = 0.3206), which showed hypomethylation in abdominal obesity, indicating that dietary macronutrient composition may influence DNA methylation patterns associated with abdominal obesity. Analysis of nutrient intake data from Supplementary Table S5 revealed even stronger correlations between methylation at the obesity-associated CpG sites and macronutrient consumption. Fat consumption showed strong positive correlations with methylation at cg13001142 (r = 0.3308, p = 1.22E − 36), cg16714760 (r = 0.3258, p = 1.54E − 35), cg11807280 (r = 0.3206, p = 2.05E − 34), and cg15830864 (r = 0.3190, p = 4.67E − 34), showing hypomethylation in abdominal obesity (negative Δβ values in Table [83]3) and negative correlations with WHR (r = − 0.291 to − 0.338). Interestingly, cg05825244, showing hypermethylation in abdominal obesity (Δβ = 0.021) and a positive correlation with WHR (r = 0.299), showed a significant negative correlation with fat intake (r = − 0.3264, p = 1.15E − 35). Conversely, carbohydrate intake showed predominantly negative correlations with methylation at sites positively correlated with fat intake. For example, cg13001142 (r = − 0.3087, p = 6.75E − 32), cg16714760 (r = − 0.3034, p = 8.22E − 31), cg11807280 (r = − 0.2978, p = 1.07E − 29), and cg15830864 (r = − 0.2967, p = 1.78E − 29) showed strong negative correlations with carbohydrate consumption. Notably, cg05825244, negatively correlated with fat intake, was positively correlated with carbohydrate intake (r = 0.3050, p = 3.88E − 31). This inverse relationship between the fat and carbohydrate correlation patterns was consistent across most of the examined CpG sites and aligned with their methylation status in relation to abdominal obesity. Figure [84]4B shows that the clustering of CpG sites based on nutrient correlations revealed patterns similar to those observed with food items. The top cluster of CpG sites (those with hypermethylation in abdominal obesity and positive WHR correlations) was positively correlated with carbohydrates and fiber but negatively correlated with fat, cholesterol, and retinol. The bottom cluster (those with hypomethylation in abdominal obesity and negative WHR correlations) showed strong positive correlations with fat and cholesterol but negative correlations with carbohydrates. These findings indicate that dietary patterns influence DNA methylation in ways that may affect abdominal obesity risk. Traditional diets rich in fermented foods and carbohydrates seem to be associated with different methylation patterns compared with diets high in animal products and fat, potentially reflecting different epigenetic regulatory mechanisms. The consistent alignment between food correlations, nutrient correlations, and methylation status in abdominal obesity provides strong evidence for diet-induced epigenetic modifications that could contribute to the development or protection of obesity. Association between methylation levels of CpG sites and food intake in relation to BMI The association between DNA methylation and dietary intake showed distinct patterns in relation to BMI-defined obesity, which differed from those observed with abdominal obesity. Supplementary Table S6 shows the strongest correlations between CpG methylation and fruit consumption, particularly apples and grapes. Apple consumption showed the strongest positive correlations with methylation at cg10725542 (r = 0.2748, p = 2.32E − 25), cg23379566 (r = 0.2722, p = 6.70E − 25), cg01791648 (r = 0.2630, p = 2.67E − 23), and cg08124030 (r = 0.2617, p = 4.54E − 23). Table [85]4 showed that these CpG sites exhibited progressive hypomethylation with increasing BMI category (cg10725542: Δβ = −0.005401 to − 0.011192; cg23379566: Δβ = −0.006141 to − 0.012961; cg01791648: Δβ = −0.005547 to − 0.013725; cg08124030: Δβ = −0.006060 to − 0.014513). Supplementary Table S3 showed that these sites also had moderate to strong negative correlations with BMI (cg08124030: r = − 0.434, ranked 14th; cg01791648: r = − 0.413, ranked 13th). Similarly, grape consumption was significantly associated with methylation at the same CpG sites (r = 0.2374–0.2560). Notably, cg15229836 showed a significant negative correlation with apple consumption (r = − 0.2225, p = 5.75E − 17) and exhibited progressive hypermethylation with increasing BMI category (Δβ = 0.005740–0.017446) and a positive correlation with BMI (r = 0.428). Other fruits were also significantly correlated with the methylation patterns. Strawberry consumption was associated with methylation at cg08124030 (r = 0.1914, p = 7.21E − 13), cg01791648 (r = 0.1885, p = 1.62E − 12), and cg10725542 (r = 0.1843, p = 5.00E − 12). Interestingly, dairy products, such as cream, showed correlations with cg11839163 (r = 0.1998, p = 6.54E − 14), cg10323433 (r = 0.1805, p = 1.40E − 11), and a negative correlation with cg15570860 (r = − 0.1905, p = 9.24E − 13). Table [86]4 and Supplementary Table S3 indicate that cg11839163 showed progressive hypomethylation with increasing BMI (Δβ = −0.005797 to − 0.020135) and a strong negative correlation with BMI (r = − 0.428), whereas cg15570860 showed hypermethylation (Δβ = 0.008981–0.027949) and a positive correlation with BMI (r = 0.438). Figure [87]5A showed that the hierarchical clustering of the correlations between CpG sites and food items revealed distinct patterns in relation to BMI. The heatmap shows two main clusters of CpG sites with opposing correlation patterns. The top cluster (including cg13139335, cg07965995, cg15570860, cg15229836, and cg26237810) showed positive correlations with mixed grain rice, water kimchi, dog meat, radish kimchi, and cabbage, but negative correlations with cookies, apple, other mushrooms, cream, coffee, and sugar. These CpG sites predominantly showed hypermethylation in BMI-related obesity (positive Δβ values in Table [88]4) and positive correlations with BMI (r = 0.414–0.438) (Supplementary Table S3). The bottom cluster (including cg09408571, cg07412545, cg10323433, cg06223162, and others) was positively correlated with fruits, coffee, cream, sugar, and mushrooms but negatively correlated with traditional fermented foods and grain-based foods. Table [89]4 showed that these CpG sites exhibited progressive hypomethylation with increasing BMI category (negative Δβ values) and showed negative correlations with BMI (r = − 0.431 to − 0.452) (Supplementary Table S3). CpG sites previously identified as strongly associated with BMI showed significant correlations with dietary patterns. For instance, cg10501210, the top DMP for BMI (Δβ = −0.006895 to − 0.021296) with the strongest negative correlation with BMI (r = − 0.452), was significantly correlated with apple (r = 0.2129, p = 1.25E − 15), grapes (r = 0.1691, p = 2.49E − 10), and strawberry (r = 0.1684, p = 2.95E − 10). Similarly, cg10323433, the second-ranked DMP for BMI (Δβ = −0.005930 to − 0.019079) and had a strong negative correlation with BMI (r = − 0.450), was significantly correlated with cream (r = 0.1805, p = 1.40E − 11), apple (r = 0.1731, p = 9.23E − 11), sugar (r = 0.1593, p = 2.63E − 09), and coffee (r = 0.1581, p = 3.42E − 09). Fig. 5. [90]Fig. 5 [91]Open in a new tab Hierarchical clustering analysis of the correlations between BMI-associated CpG methylation sites and dietary factors. (A) Heatmap showing the clustered correlations between BMI-associated CpG sites (rows) and food items (columns). The color scale represents the correlation strength and direction, with red and blue indicating positive (r = 0–0.5) and negative correlations (r = 0 to − 0.5), respectively. Asterisks (*) denote statistically significant correlations (p < 0.05). Two distinct clusters of CpG sites emerged: the top cluster (cg13139335, cg07965995, cg15570860, cg15229836, cg26237810) was positively correlated with traditional Korean foods (mixed grain rice, water kimchi, dog meat, radish kimchi) but negatively correlated with fruits, coffee, and sugar. However, the bottom cluster was positively correlated with fruits, coffee, cream, and sugar but negatively correlated with traditional fermented foods. These patterns align with the methylation status in BMI-related obesity, as sites in the top cluster showed hypermethylation in individuals with higher BMI and positive correlations with BMI, whereas sites in the bottom cluster predominantly exhibited hypomethylation with increasing BMI. (B) Heatmap depicting the clustered correlations between BMI-associated CpG sites (rows) and nutrient intakes (columns). Similar to the food correlation patterns, two distinct clusters of CpG sites were evident. The top cluster (cg15229836, cg07965995, cg13139335, cg15570860, cg26237810) showed positive correlations with carbohydrates, fiber, and ash content but negative correlations with fat, protein, cholesterol, and retinol. The bottom cluster demonstrates the opposite pattern, with strong positive correlations with fat, protein, and micronutrients from animal sources. This pattern contrasts with those observed for WHR-associated sites, highlighting the distinct dietary patterns associated with BMI-related DNA methylation compared with those associated with WHR. Notably, fruit consumption and related micronutrients (such as vitamin C) showed particularly strong correlations with CpG sites in the bottom cluster (including cg10725542, cg23379566, cg01791648, and cg08124030), showing hypomethylation in BMI-related obesity and negative correlations with BMI, indicating different dietary components may influence DNA methylation in ways specifically affecting general obesity through distinct epigenetic regulatory mechanisms. Analysis of nutrient intake data from Supplementary Table S7 revealed strong correlations between methylation at the obesity-associated CpG sites and macronutrient consumption. In relation to BMI, fat consumption showed the strongest positive correlations with methylation at cg11839163 (r = 0.2610, p = 5.98E − 23), cg10323433 (r = 0.2340, p = 1.22E − 18), and cg06223162 (r = 0.2045, p = 1.65E − 14), which showed hypomethylation in BMI-related obesity (negative Δβ values in Table [92]4) and negative correlations with BMI (r = − 0.428 to − 0.450). Conversely, fat intake was negatively correlated with cg15229836 (r = − 0.2326, p = 1.96E − 18), cg15570860 (r = − 0.2282, p = 8.65E − 18), and cg26237810 (r = − 0.2127, p = 1.32E − 15), which exhibited hypermethylation in BMI-related obesity (positive Δβ values) and positive correlations with BMI (r = 0.428, 0.438, and 0.412, respectively). Carbohydrate intake showed opposing correlation patterns with these same CpG sites, with negative correlations at cg11839163 (r = − 0.2219, p = 7.02E − 17) and cg10323433 (r = − 0.1916, p = 6.84E − 13) and positive correlations with cg15229836 (r = 0.1884, p = 1.67E − 12) and cg15570860 (r = 0.1862, p = 2.99E − 12). This inverse relationship between the fat and carbohydrate correlation patterns was consistent across most of the examined CpG sites and aligned with their methylation status in relation to BMI-defined obesity. Additionally, micronutrient intake was significantly associated with methylation patterns. Vitamin C intake, which is abundant in fruits, showed particularly strong correlations with cg10725542 (r = 0.1800, p = 1.57E − 11), cg23379566 (r = 0.1790, p = 2.07E − 11), and cg01791648 (r = 0.1577, p = 3.77E − 09), showing hypomethylation in BMI-related obesity and negative correlations with BMI, which wereconsistent with the observed positive correlations between fruit consumption and methylation at these same CpG sites. Figure [93]5B shows that the clustering of CpG sites based on nutrient correlations revealed patterns similar to those observed with food items in relation to BMI. The top cluster (including cg15229836, cg07965995, cg13139335, cg15570860, and cg26237810) was positively correlated with carbohydrates, fiber, and ash content but negatively correlated with fat, protein, cholesterol, and retinol. These CpG sites exhibited hypermethylation in BMI-related obesity and positive correlations with BMI. The bottom cluster showed strong positive correlations with fat, protein, and micronutrients from animal sources, and included CpG sites showing hypomethylation in BMI-related obesity and negative correlations with BMI. These findings highlight the distinct dietary patterns associated with BMI-related DNA methylation compared with those associated with WHR. Although WHR-related methylation was strongly associated with fermented foods, pork, and seafood, BMI-related methylation demonstrated stronger correlations with fruit consumption, dairy, and sugar intake, indicating that different dietary components may influence DNA methylation in ways specifically affecting either overall obesity or fat distribution patterns, potentially through distinct epigenetic regulatory mechanisms. The consistent alignment between food correlations, nutrient correlations, methylation status in BMI-related obesity, and correlations with BMI provides compelling evidence for diet-induced epigenetic modifications that may contribute to the overall obesity risk. Identifying fruit consumption as a key correlate of BMI-associated methylation patterns may have implications for dietary recommendations aimed at modifying obesity-related epigenetic signatures. Discussion We present a comprehensive epigenome-wide association analysis of obesity phenotypes in the Korean population, identifying distinct DNA methylation signatures associated with abdominal and BMI-defined obesity. Our findings showed novel epigenetic markers potentially involved in obesity development and progression, as well as significant associations between methylation patterns and dietary intake. These results provide insights into the complex interplay between epigenetics, diet, and obesity, highlighting the potential mechanisms underlying different obesity phenotypes in Korean adults. Distinct epigenetic signatures of abdominal obesity reveal potential regulatory mechanisms Our EWAS identified 23 DMPs significantly associated with abdominal obesity, with 85% exhibiting hypomethylation in individuals with elevated WHR. The most significant association was observed at cg10323433, located in the promoter region of HTR2A (serotonin receptor 2 A), indicating a potential role for serotonin signaling in abdominal fat distribution. Serotonin is a key neurotransmitter involved in appetite regulation and energy homeostasis, and altered serotonin signaling has been implicated in obesity development^[94]21. The hypomethylation at the HTR2A locus may lead to altered serotonin receptor 2 A expression, implicated in central adiposity and metabolic syndrome components, such as hypertension^[95]22. This hypomethylation may lead to dysregulated HTR2A expression, potentially affecting appetite control and metabolic processes. Identifying multiple DMPs in metabolically relevant genes, such as PDZD2, RNF220, and PCDH9, further supports the involvement of epigenetic mechanisms in abdominal obesity. PCDH9, represented by two distinct DMPs with consistent hypomethylation patterns, encodes a protocadherin involved in cell–cell adhesion and has been implicated in various neurological processes^[96]23. It is expressed predominantly in the brain, particularly in the hippocampus and amygdala, where it influences synaptic morphology and function^[97]24. The robust association of these PCDH9-related DMPs with abdominal obesity indicates their potential involvement in adipose tissue remodeling or neuronal regulation of metabolism. Notably, the genomic distribution of these DMPs was predominantly localized in gene bodies (45%) and opensea regions (55%), which are typically more variable and responsive to environmental influences compared with the promoter regions and CpG islands. This distribution pattern aligns with findings from studies on DNA methylation polymorphisms, which show that variably methylated regions often occur in enhancers and 3′UTRs, known for their environmental responsiveness and lower heritability compared with promoter regions and CpG islands^[98]25. This finding aligns with the concept that obesity-related epigenetic modifications may be shaped by environmental factors, such as diet and lifestyle, providing a potential mechanistic link between the environment and phenotype. BMI-related methylation patterns reveal a dose–response relationship with obesity severity Our analysis revealed a compelling dose–response relationship between BMI categories and methylation changes at specific CpG sites, as evidenced by the progressive hypomethylation of cg10501210 and cg10323433 with increasing obesity severity. This gradient effect indicates that these epigenetic modifications may not only be markers of obesity but could potentially contribute to or result from obesity progression, reflecting cumulative metabolic dysregulation. Identifying multiple DMPs within the GPR88 gene, showing consistent hypomethylation patterns with increasing BMI, highlights the potential importance of G protein-coupled receptor signaling in general obesity. Moreover, GPR88 has been implicated in energy homeostasis, and the coordinated methylation changes observed across multiple sites within this gene provide robust evidence for its epigenetic regulation in obesity. Studies showed that GPR88 plays a crucial role in controlling food intake and body composition, as evidenced by studies on Gpr88/mice, which had reduced adiposity and altered feeding behaviors, particularly under high-fat diet conditions^[99]26. Similarly, identifying three DMPs in TM4SF1, a gene encoding a cell surface antigen involved in cellular migration, indicates potential roles in adipocyte function or tissue remodeling associated with obesity^[100]27. This is supported by the understanding that obesity is characterized by significant changes in adipose tissue, including inflammation and fibrosis, influenced by various genetic and epigenetic factors. For instance, the role of Tim-4 in regulating macrophage polarization and maintaining adipose tissue homeostasis through the NF-κB pathway highlights the complex interplay between immune signaling and adipocyte function in obesity^[101]28. The three consecutive DMPs found in an intergenic region on chromosome 3 further point to a potential regulatory hotspot that may influence obesity-related gene expression through long-range interactions^[102]29. Furthermore, the predominance of hypomethylation (75% of DMPs) associated with increasing BMI aligns with our findings in abdominal obesity and supports the hypothesis that obesity is generally characterized by widespread DNA hypomethylation at specific regulatory sites. This epigenetic signature may contribute to altered gene expression profiles in obesity, potentially affecting the pathways involved in adipogenesis, inflammation, and metabolic regulation. Shared and distinct epigenetic mechanisms underlying different obesity phenotypes Our comparative analysis of DMPs associated with WHR and BMI showed shared and distinct epigenetic signatures, providing insights into the molecular mechanisms underlying different obesity phenotypes. Identifying four CpG sites (cg10323433, cg10501210, cg08124030, and cg23379566) associated with WHR and BMI indicates common epigenetic mechanisms influencing overall adiposity and fat distribution. Particularly noteworthy is that cg10323433 in HTR2A may potentially play a central role for serotonin signaling in various obesity manifestations. The shared association of cg10501210, an intergenic site on chromosome 1, further supports the existence of common regulatory mechanisms that potentially affect distant genes through enhancer or insulator functions^[103]30. Conversely, identifying 16 CpG sites specifically associated with abdominal obesity and 16 sites unique to BMI-related obesity highlights the distinct epigenetic architecture underlying these obesity phenotypes. The WHR-specific DMPs were predominantly located in the gene bodies and intergenic regions, whereas BMI-specific markers were enriched in the promoter regions, indicating different regulatory mechanisms. The enrichment of GPR88-related DMPs among BMI-specific markers but not WHR-specific markers indicates that G protein-coupled receptor signaling may be more relevant to overall adiposity than fat distribution, which is supported by the role of GPCRs in regulating energy homeostasis and glucose metabolism, which are critical in obesity and T2DM^[104]31. GPCRs, particularly those expressed in adipose tissues, are involved in adipogenesis and adipocyte function, influencing lipid accumulation and energy expenditure, which are central to obesity development^[105]32,[106]33. Conversely, the presence of PCDH9-related DMPs exclusively among WHR-specific markers indicates that cell adhesion processes may play a more critical role in abdominal obesity. Furthermore, genetic studies have shown that body fat distribution, as measured by WHR, is a heritable trait influenced by distinct genetic loci, separate from those affecting overall adiposity^[107]34. These findings support the clinical observation that abdominal obesity and general obesity represent distinct metabolic phenotypes with different health implications and suggest that targeted interventions might consider the specific epigenetic signatures associated with each phenotype. Comparison with existing obesity-related EWAS studies Our findings show substantial consistency with previous international EWAS studies while contributing novel population-specific insights. Key genes such as ABCG1, CPT1A, and PHGDH, which emerged consistently in our analyses, have been identified across multiple international cohorts, supporting the robustness of these associations across populations. Compared to the previous Korean study by Koh et al. (2020) using KoGES data (n = 902)^[108]10our larger cohort (n = 1,526) and inclusion of abdominal obesity phenotypes revealed both overlapping and distinct findings. While both studies identified obesity-related genes such as ABCG1, our study’s novel identification of CPA3 represents a significant advancement. The identification of 23 WHR-specific versus 1,931 BMI-specific DMPs with only 4 overlapping sites demonstrates distinct epigenetic architectures underlying different obesity phenotypes, which was not captured in previous Korean studies focusing solely on BMI. Cross-population validation from recent studies, including the Chinese twin study examining WHR-related methylation patterns^[109]35 and Scottish population studies^[110]36supports several of our key findings while highlighting population-specific differences. This emphasizes the importance of conducting EWAS across diverse populations to distinguish between universal and population-specific epigenetic markers of obesity. Dietary patterns exhibit differential associations with obesity-related methylation signatures Notably, associations between dietary intake and methylation had distinct patterns at obesity-related CpG sites. For WHR-associated DMPs, the strongest correlations were observed with traditional Korean foods, particularly pork belly, mushrooms, and fermented foods, such as water kimchi and fermented soybean soup. Conversely, BMI-associated DMPs showed the strongest correlations with fruit consumption, particularly apples and grapes, as well as dairy products, such as cream. Studies have shown that hypermethylation at specific genomic sites is positively correlated with higher BMI and associated with traditional Korean food intake, such as mixed grain rice and kimchi. Conversely, these hypermethylated sites are negatively associated with fruit, coffee, and sugar intake^[111]37,[112]38indicating that traditional Korean diet, which is rich in multigrain rice and fermented vegetables, may influence DNA methylation patterns linked to obesity. The hierarchical clustering analysis revealed two distinct correlation patterns between CpG methylation and food intake for both obesity phenotypes. For WHR, sites with hypermethylation (positive Δβ values) and positive correlations with WHR were positively associated with traditional fermented foods but negatively associated with animal products and processed foods. Conversely, sites with hypomethylation (negative Δβ values) and negative correlations with WHR showed the opposite pattern. Similarly, for BMI, sites with hypermethylation and positive correlations with BMI were positively associated with traditional Korean foods, such as mixed grain rice and kimchi, but negatively associated with fruits, coffee, and sugar. Sites with hypomethylation and negative correlations with BMI showed the opposite pattern.These findings indicate that different dietary components may influence DNA methylation in ways specifically affecting either overall obesity or fat distribution patterns. The consistent alignment between food correlations, nutrient correlations, and methylation status provides compelling evidence for diet-induced epigenetic modifications that may contribute to obesity development or protection. Notably, fat and carbohydrate intake correlations had a consistent inverse relationship across most CpG sites. Sites showing positive correlations with fat intake generally were negatively correlated with carbohydrate intake and vice versa. This pattern aligns with the traditional dietary transition hypothesis, where shifts from traditional carbohydrate-based diets to Western high-fat diets are associated with increased prevalence of obesity, which is supported by several studies highlighting the impact of dietary macronutrient composition on body weight and health outcomes. For instance, Gaesser reported an inverse relationship between carbohydrate intake and BMI, indicating that high-carbohydrate diets, particularly those rich in fiber, may be beneficial for weight control and health^[113]39. Identifying fruit consumption as a key correlate of BMI-associated methylation patterns may have implications for dietary recommendations aimed at modifying obesity-related epigenetic signatures. Similarly, the associations between fermented foods and WHR-related methylation patterns highlight the potential metabolic benefits of traditional Korean fermented foods rich in probiotics and bioactive compounds that may influence host metabolism through epigenetic mechanisms, potentially offering metabolic benefits and aiding in obesity management^[114]40,[115]41. Limitations and future directions This study has several limitations. First, the cross-sectional design precludes causal inference regarding the relationships between methylation, diet, and obesity. Longitudinal studies should determine whether the observed methylation changes precede or result from obesity development and whether dietary interventions can modify these epigenetic signatures. Second, the temporal discrepancy between DNA methylation data (collected during the fourth follow-up survey, 2009–2010) and dietary intake data (collected during the second follow-up survey, 2005–2006) represents a significant limitation that requires careful consideration. This 4-year interval introduces several potential confounding factors that may affect the validity of diet-methylation associations. During this period, participants may have experienced changes in dietary patterns, body weight fluctuations, development of comorbidities, medication use, or other lifestyle modifications that could independently influence DNA methylation patterns. Given that DNA methylation is highly dynamic and responsive to environmental factors, the observed correlations between dietary intake and methylation patterns should be interpreted with caution. We acknowledge that ideally, participants with significant weight changes (> 5% body weight change) or major lifestyle interventions during this interval should have been excluded from diet-methylation analyses, or these changes should have been statistically controlled for. Unfortunately, detailed longitudinal data on these parameters were not available for the intervening period. While dietary habits in middle-aged and older Korean adults show relative stability over time, individual variations cannot be ruled out. The correlations observed between dietary factors and DMPs may reflect historical dietary influences on current methylation status, but they may also be confounded by unmeasured changes occurring during the 4-year gap. To address this limitation in future work, we are currently collecting DNA methylation data and semi-quantitative food frequency questionnaire (semi-FFQ) data concurrently from the same participants at the same time points. This ongoing data collection will enable us to examine diet-methylation associations without the temporal discrepancy issue and provide more robust evidence for the relationship between dietary factors and epigenetic modifications. Additionally, predictive modeling approaches that account for temporal relationships and potential mediating factors should be employed in future analyses. Longitudinal studies with multiple time points for both dietary assessment and methylation profiling would provide the most comprehensive evidence for diet-methylation relationships and help distinguish between acute and chronic dietary effects on the epigenome. Third, although we adjusted for key confounders, including age, sex, smoking, and relevant anthropometric measures, residual confounding by unmeasured factors cannot be ruled out. Environmental exposure, physical activity, and genetic variations may influence methylation patterns and obesity phenotypes and should be considered in future analyses. Additionally, our confounder selection strategy lacked a systematic theoretical framework, and we did not include potentially important confounders such as socioeconomic status and alcohol consumption, which could influence both dietary patterns and DNA methylation. The inconsistency in confounder adjustment across different analytical sections, particularly the unadjusted dietary correlation analyses, represents a methodological weakness that limits the comparability of findings across different approaches. Moreover, despite acknowledging sex differences in obesity-related methylation patterns, we did not stratify analyses by menopausal status. Given our participant age range (40–69 years) and the known effects of menopause on body composition, fat distribution, and hormonal regulation, this represents a missed opportunity for more nuanced analysis that could have revealed important subgroup effects specific to different hormonal states in women. Fourth, the sample size, while substantial, may have limited the power to detect more subtle methylation changes, particularly in the severely obese group, comprising 4.3% of the cohort. Larger studies, particularly with greater representation of severely obese individuals, would enhance the detection of obesity-related epigenetic signatures. Finally, we focused on DNA methylation in blood, which may not fully reflect tissue-specific methylation patterns in adipose tissue or other metabolically relevant organs. However, significant correlations have been found between blood and adipose tissue methylation at obesity-related loci, supporting the validity of our findings. Future studies incorporating tissue-specific methylation analysis would provide more comprehensive insights into the epigenetic basis of obesity. Future research should also address methodological improvements including: (1) systematic confounder selection using directed acyclic graphs to better control for potential confounding pathways, (2) inclusion of comprehensive demographic and lifestyle factors such as socioeconomic status and alcohol consumption, (3) sex-stratified analyses with detailed reproductive health information to distinguish between pre- and post-menopausal women, and (4) integration with national health registry data to assess the predictive utility of identified methylation markers for Korean-specific disease burdens such as diabetes and cardiovascular disease. Mendelian randomization approaches using genetic variants as instrumental variables could also help establish causal relationships between DNA methylation and obesity phenotypes. Despite these limitations, this study provides valuable insights into the epigenetic landscape of obesity in the Korean population and highlights the potential role of diet in modulating these epigenetic signatures. Identifying shared and distinct methylation markers for different obesity phenotypes may inform more targeted approaches to obesity prevention and management. Conclusions This EWAS identified distinct DNA methylation signatures associated with abdominal and BMI-defined obesity in Korean adults, revealing both shared and unique epigenetic markers that indicate complex molecular mechanisms underlying different obesity phenotypes. The most significant associations involved genes related to serotonin signaling (HTR2A), G protein-coupled receptor function (GPR88), and cell adhesion (PCDH9), with a predominant pattern of hypomethylation in obesity suggesting widespread epigenetic dysregulation. Our novel integration of dietary pattern analysis revealed distinct associations between traditional Korean foods and WHR-related methylation sites versus fruits and dairy products with BMI-related sites. The consistent inverse relationship between fat and carbohydrate intake correlations across multiple CpG sites provides evidence for the epigenetic impact of macronutrient composition. These findings contribute to understanding the molecular mechanisms underlying different obesity phenotypes in Koreans and may inform targeted approaches to obesity prevention and management through personalized dietary interventions. Future longitudinal studies should establish causality and evaluate the potential for dietary modifications to alter obesity-related epigenetic patterns. Supplementary Information Below is the link to the electronic supplementary material. [116]Supplementary Material 1^ (762.7KB, docx) Acknowledgements