Introduction

Hypertension is a prevalent disorder among the elderly population. It has been served as a significant risk factor for various age-related diseases, such as cardiovascular disease, stroke, and cognitive decline1. Given the rapid increase in global aging, hypertension will also become a major global public health concern. Therefore, it is imperative to investigate the specific shared genetic architecture between hypertension and human aging.

Aging-related phenotypes have been utilized to assess the degree of aging and are increasingly employed in genetic research2. Recent evidence have showed that biological age measures capture the age-related biological consequences of hypertension3. Further, epigenetic clocks were evaluated as the most prospective biological age estimators4. Herein, our study focused on epigenetic clocks, measures of biological age based on DNA methylation patterns that have emerged as a powerful way to assess aging and its association with various health outcomes5,6. Epigenetic age may be older than chronological age, which is known as epigenetic age acceleration7. Different epigenetic clocks capture different epigenetic features of the complicated aging process.

Each epigenetic clock is based on the unique characteristics of DNA methylation levels reflecting biological aging, as measured at a specific set of cytosine-phosphate-guanine (CpG) loci. HannumAge8 and intrinsic epigenetic age acceleration (IEAA)9 are “first-generation” epigenetic clocks. HannumAge utilizes 71 CpG sites based on Illumina 450 k array and performs best with whole blood samples8. IEAA is a multi-tissue age predictor trained on 353 CpG sites based on Illumina 27k array, allowing one to estimate the age of DNA methylation for most tissues and cell types9. The “second-generation” epigenetic clocks include PhenoAge10 and GrimAge11. PhenoAge was trained on 513 CpG sites and developed using a two-stage procedure12. Besides, PhenoAge surpasses “first-generation” clocks in predicting many age-related diseases and longevity12. GrimAge combined data from 1030 CpGs associated with smoking pack-years and seven plasma proteins (such as, cystatin C, leptin and tissue inhibitor, etc.)11. Genome-wide association studies (GWAS) have identified 137 genetic loci associated with epigenetic aging measured by four epigenetic clocks (GrimAge, HannumAg, IEAA, and PhenoAge)13. All of these four epigenetic clocks are capable of accurately assessing epigenetic aging.

Increasing evidence supported that epigenetic signatures in pre-hypertensive and hypertensive patients could be exploited for early intervention and prevention14,15,16. Based on DNA methylation profiles, Jacob et al. revealed that, in a cohort of women, GrimAge acceleration (OR 1.28; 95% CI 1.14–1.45 per SD-increase) and PhenoAge acceleration (OR 1.16; 95% CI 1.05–1.28 per SD-increase) increased the risk of hypertension3. They concluded that epigenetic age increased preceding the diagnosis of hypertension and remained high for years after diagnosis and medical intervention.

Nevertheless, we perceived some limitations of the available studies on the genetic associations between hypertension and epigenetic aging. First, though many studies have elucidated aging was a risk factor of hypertension, to best of our knowledge, few studies have investigated epigenetic aging with hypertension. Further, there is also a demand for research based on other omics data, as most research in this field focused on DNA methylation profiles. Finally, the limitation of the Jacob et al. study to female participants may have implications for the generalizability of the findings. Incorporating diverse participant groups, including both males and females, would be essential for obtaining a comprehensive understanding and enhancing the overall applicability and relevance of the research findings.

In this study, we employed sophisticated statistical approaches to explore the shared polygenic architecture of hypertension and epigenetic aging, and exploited this overlap to improve the ability to identify more hypertension loci17. We firstly conducted univariate and bivariate causal mixture model (MiXeR) to quantify the number of variants influencing a phenotype regardless of effect direction, and clarified the polygenic architecture of hypertension and epigenetic aging. Furthermore, we performed conditional and conjunctional false discovery rate (cond/conjFDR) analysis to identify shared loci between phenotypes and boost GWAS discovery to identify novel loci associated with hypertension.

Methods

GWAS datasets

We obtained the GWAS summary statistics for hypertension from a published study by Handan et al.18. They acquired the GWAS data, self-reports, and age-at-diagnosis (range from 0 to 70) from UK Biobank (UKBB) covering 484,598 participants. Cases were assessed by self-report survey. The four GWAS summary-level datasets for epigenetic clocks (GrimAge, HannumAge, IEAA, PhenoAge) were obtained from a meta-analysis study13. Each dataset comprises 34,710 European ancestry subjects of 28 cohorts. Briefly, the estimates of the four epigenetic age acceleration were calculated by McCartney et al. using Horvath epigenetic age calculator software (https://dnamage.genetics.ucla.edu/ or standalone scripts provided by Steve Horvath and Ake Lu). Then, quality control and genotype imputation were conducted. For more specific details on experimental methods, please refer to McCartney et al.13.

Hypertension is characterized by increased systolic blood pressure (SBP) or/and diastolic blood pressure (DBP)19. We acquired GWAS summary statistics for SBP and DBP as validation datasets to verify the robustness of the hypertension-associated distinct loci we identified. These two datasets incorporate 1,028,980 participates, respectively.

Linkage disequilibrium score regression analysis

Investigating genetic correlations between complex phenotypes can shed helpful light on aetiology20. Linkage disequilibrium score (LDSC) regression is a widely adopted approach for estimating SNP-based heritability and genetic correlations (\({r}_{g}\)) between complex traits, while effectively accounting for potential biases arising from polygenicity, sample overlap, and population stratification21.

We utilized LDSC software (v1.0.1, available at https://github.com/bulik/ldsc) to calculate linkage disequilibrium (LD) scores with the aim of investigating the genetic associations between hypertension and epigenetic age acceleration22. Pre-calculated LD scores were retrieved from the 1000 Genomes Project Phase III. To ensure robust test statistics, we focused our analysis exclusively only on well-imputed HapMap3 SNPs, recognizing that poor imputation quality may compromise the accuracy of results. Furthermore, we excluded SNPs within the major histocompatibility complex (MHC) region (Chr 6: 28,477,797–33,448,354). Traits with mean \({\chi }^{2}<1.02\) were not suitable for pairwise LDSC due to the absence of polygenic signals.

Causal mixture model

Using MiXeR, we estimated the number of trait-influencing variants (also known as ‘causal’ variants) that explained 90% of the SNP heritability of each phenotype, i.e., polygenicity23. Trait-influencing variants or ‘causal’ variants were genetically considered as a variant with non-zero additive genetic effects on a phenotype24,25. In the Univariate MiXeR model, common genetic variants are treated as capable of exerting causal or non-causal effects on the specific phenotype. Leveraging the univariate model, bivariate MiXeR investigates a pair of traits and facilitates the estimation of both phenotype-specific and shared variants. Dice coefficient indicates the proportion of variations shared between the pair of traits. Considering the substantial differences in the polygenicity of secondary phenotypes, we calculated the ratio of shared variants in terms of the maximum potential overlap (P_overlap) according to the following Eq. (1):

$${\text{P}}\_{\text{overlap}} = {\text{N}}\_{\text{overlap}}/{\text{N}}\_{\text{less}}$$
(1)

N_overlap represents the number of overlapped variants between two phenotypes. N_less represents the number of variants for phenotype with fewer variants.

Conditional and conjunctional false discovery rate analysis

For GWAS data, genetic variants significantly associated with the phenotype were usually identified with a corrected \(p<5\times {10}^{-8}\). To get rid of false-negative results, we employed condFDR and conjFDR approaches to enhance the identification of specific genomic loci that are jointly associated with phenotypes displaying powerful causal relationships17,26. Subsequently, novel loci were identified if the genetic variant did not have any physical overlap with the loci reported in the original GWAS within a range of ± 500 kb, or if they were not reported in the NHGRI-EBI catalog27. The condFDR method extends the standard false discovery rate (FDR) framework by incorporating genetic summary statistics from both the primary trait of interest (e.g., hypertension) and a conditional trait (e.g., each epigenetic clock). Thus, condFDR can reassess the test statistics for the primary trait. Genetic variants exhibiting a condFDR value below 0.01 were deemed to be associated with the primary phenotype. The conjFDR method is an advancement of condFDR, aiming to enhance genetic discovery by examining the cross-trait enrichment between two traits. Genetic variants with a conjFDR value below 0.05 were identified as shared loci28. Besides, we performed quantile–quantile (Q-Q) plots by conditioning hypertension on each four epigenetic clocks (GrimAge, HannumAge, IEAA, and PhenoAge). The Q-Q plots illustrated the polygenic enrichment. Considering the complex LD structure may bias cond/conjFDR estimates, we excluded the SNPs within the extended MHC region, chromosome 8p23.1 (chr8:7,200,000–12,500,000) and MAPT region (chr17:40,000,000–47,000,000) based on human genome build 19 positions before fitting cond/conjFDR model29. cond/conjFDR analyses were conducted by pleiofdr (https://github.com/precimed/pleiofdr).

Genomic loci definition

In this study, independent genomic loci were defined according to the functional mapping and gene annotation (FUMA) protocol. SNPs with conjFDR < 0.05 and \({\text{r}}^{2}\)< 0.1 were considered as independent significant SNPs. The lead SNP was defined as those with \({\text{r}}^{2}\)< 0.05 among independent significant SNPs. The SNPs with conjFDR < 0.05 and \({\text{r}}^{2}\ge\) 0.1 were regarded as candidate SNPs. The loci were merged if their distance was less than 250 kb (i.e., LD blocks < 250 kb apart). Further, the SNP exhibiting the most significant conjFDR was selected as the lead SNP for the merged locus. The candidate SNP located within the boundaries of a genomic locus was defined as belonging to a single independent genomic locus. LD information was acquired and computed using the reference panel provided by the 1000 Genomes Project20. To determine the effect directions for the shared loci, we compared the z-scores in the GWAS summary statistics corresponding to the phenotype.

Functional annotations

We employed combined annotation dependent depletion scores (CADD)30, RegulomeDB scores31, and chromatin states to functionally annotate SNPs. CADD is a widely used computational tool for evaluating the degree of pathogenicity. RegulomeDB scores estimate the potential regulatory functionality of SNPs based on RegulomeDB database. We performed gene-set enrichment, gene expression analysis, and expression quantitative trait locus (eQTL) functional analysis using FUMA and Genotype Tissue Expression (GTEx) resources.

Ethics approval and consent to participate

All GWAS data used in this study are publicly download from the original studies. This study only used publicly available data and hence no ethics approval was required.

Results

Assessment of cross-phenotype polygenic enrichment

We assessed the Q-Q plots to investigate the presence of cross-trait polygenic enrichment. The observed enrichment is characterized by an upward and leftward deviation of the plots, indicating a stronger association between the subset of SNPs and the secondary phenotype26. We found that SNPs associations with hypertension were substantially enriched with increasing SNP associations with all epigenetic clocks, suggesting polygenic overlap (Fig. 1). Reverse conditional Q-Q plots were illustrated in Supplementary Fig. 1.

Figure 1
figure 1

Cross-trait enrichment between hypertension and epigenetic aging. Quantile–quantile (Q-Q) plots illustrate cross-trait enrichment between hypertension and GrimAge (A), HannumAge (B), IEAA (C), PhenoAge (D). Conditional Q-Q plots of nominal versus empirical \(-{log}_{10}p\), in which p represents the p values corrected for inflation, in primary phenotypes below the GWAS significance threshold of \(p<5\times {10}^{-8}\) as a function of significance of association with the second phenotypes, at \(p<0.10\), \(p<0.01\), \(p<0.001\). The dashed lines indicate the null hypothesis. The blue lines indicate all SNPs.

Full size image

Estimating polygenicity and genetic overlap

Univariate MiXeR analysis revealed that hypertension was a polygenic phenotype, with about \(2.8\text{ K}\pm 78\) variants influencing this phenotype (Fig. 2, Supplementary Table 1). The polygenicity estimates for four epigenetic clocks were approximately 0.7 K variants for GrimAge (SD = 0.19 K), 0.2 K variants for HannumAge (SD = 38), 0.3 K variants for HannumAge (SD = 0.77 K), and 0.1 K variants for PhenoAge (SD = 30) (Fig. 2, Supplementary Table 1).

Figure 2
figure 2

Polygenicity of hypertension and epigenetic aging measures. The number of ‘causal’ variants explaining 90% of the heritability.

Full size image

Bivariate MiXeR analysis demonstrated genetic overlap between hypertension and epigenetic clocks (Fig. 3). The number of trait-influenced variants shared between hypertension and GrimAge was 0.4 K \(\pm\) 0.1 K, of which 84% had concordant effect directions. For hypertension and HannumAge, the shared variants accounted for approximately 0.2 K ± 22, with 95% exhibiting concordant effect directions. Similarly, between hypertension and IEAA, the shared variants accounted for approximately 0.2 K ± 34, with 88% exhibiting concordant effect directions. Lastly, between hypertension and PhenoAge, the shared variants accounted for approximately 0.1 K ± 24, of which almost all (98%) had concordant effect directions (Fig. 3, Supplementary Table 2).

Figure 3
figure 3

The number of genetic variants for genetic overlap, shared distinct loci and novel loci between hypertension and epigenetic clocks. The number of genetic variants for genetic overlap, shared distinct loci and novel loci between hypertension and GrimAge (A), HannumAge (B), IEAA (C), PhenoAge (D). The numbers displayed in the colored Venn diagrams represent the counts of shared and phenotype-specific trait-influencing variants, collectively explaining 90% of the heritability in thousands. \({r}_{g}\) represents genome-wide genetic correlation.

Full size image

The genetic overlap between hypertension and four epigenetic clocks, measured by the dice coefficient, was GrimAge (22%), HannumAge (11%), IEAA (10%), and PhenoAge (9%) (Supplementary Table 2). Given the differences in the polygenicity of secondary phenotypes, as calculated by Eq. (1), \({P}_{overlap}\) for hypertension and epigenetic clocks was GrimAge (51%), HannumAge (81%), IEAA (57%), and PhenoAge (95%). The minimum Akaike information criterion (AIC) scores between hypertension and HannumAge, and hypertension and PhenoAge are negative, suggesting that the number of overlapped variations between these two pairs of phenotypes may be lower than the estimated.

Genetic correlations

We estimated genetic correlations between hypertension and four epigenetic clocks (GrimAge, HannumAge, IEAA, PhenoAge), respectively. The single-trait LDSC analysis revealed heritability estimates of 0.11, 0.10, 0.12, 0.17, and 0.10 for hypertension, GrimAge, HannumAge, IEAA, and PhenoAge, respectively. No traits were excluded in the further pairwise LDSC analyses (All mean \({\chi }^{2}>1.02\)). Pairwise LDSC showed a strong genetic correlation between hypertension and PhenoAge (\({r}_{g}=0.157\), p = 0.001, Table 1). We also found a significant genetic correlation between hypertension and IEAA (\({r}_{g}=0.091\), p = 0.038, Table 1).

Table 1 Genome-wide genetic correlation between hypertension and epigenetic age acceleration.
Full size table

Shared loci between hypertension and epigenetic aging

After identifying polygenic overlap, we conducted bi-directional cross-trait enrichment using conjFDR analysis to detect shared genomic loci, enhancing statistical power. We identified a total of 32 distinct genomic loci shared between hypertension and epigenetic clocks—two with GrimAge, seven with HannumAge, 21 with IEAA, and four with PhenoAge (Figs. 3 and 4, Table 2, Supplementary Tables 3–6). Specially, among these loci, we found that the novel locus rs1849209 was also a shared loci jointly associated with three epigenetic clocks (HannumAge, IEAA, PhenoAge) (Table 2, Supplementary Tables 4–6). Two distinct shared loci between hypertension and GrimAge exhibit concordant effect directions. Besides, distinct shared loci between hypertension and HannumAge (2/7, 28.6%), hypertension and IEAA (12/21, 57.1%), as well as hypertension and PhenoAge (2/4, 50%) had concordant effect directions (Table 2, Supplementary Tables 3–6). Further, we identified 25 novel loci for hypertension (Table 2).

Figure 4
figure 4

Shared loci between hypertension and epigenetic clocks at conjFDR < 0.05. Common genetic variants jointly associated with hypertension and GrimAge (A), HannumAge (B), IEAA (C), PhenoAge (D) at conjFDR < 0.05. Independent lead SNPs are circled in black.

Full size image
Table 2 All distinct loci associated with hypertension at conjFDR < 0.05.
Full size table

In validation datasets (i.e., SBP and DBP), we checked the p-values and beta of the distinct loci we identified as associated with hypertension. If these distinct loci had beta > 0 and p < 5e−08 in the validation datasets, we considered the conclusion that these loci were associated with hypertension to be relatively more robust. Totally, nine of 32 distinct loci were validated as hypertension-associated in SBP and DBP validation datasets (rs1982200, rs2859868, rs11944870, rs7223364, rs117778193, rs12940887, rs1849209, rs6456686, and rs7119934, Supplementary Table 7). Particularly, five of the nine were identified as hypertension-associated in both validation datasets (i.e., rs1982200, rs2859868, rs7223364, rs12940887, and rs1849209). In addition, we conducted perturbation experiments by randomly extracting the same number of loci (n = 32) 10,000 times to verify the reproduction in validation datasets. The result performed that the nine validated loci were significantly higher than random occurrence (p < 0.0001).

Functional annotations

Functional annotation revealed that most of the candidate SNPs located in intergenic or intronic region (Supplementary Tables 8–11). Among these loci, we detected a total of 92 candidate SNPs had CADD > 12.37, indicating their detrimental effects (Supplementary Tables 8–11). Distinct loci were mapped to 545 protein coding genes using the three-way gene mapping strategy (Supplementary Tables 12–15). Totally, we detected 42 gene-sets significantly enriched with the genes mapped to the loci shared between hypertension and GrimAge (Supplementary Table 16), HannumAge (Supplementary Table 17), and IEAA (Supplementary Table 18). GO analysis of genes located in loci shared between hypertension and IEAA revealed significant enrichment in biological processes and molecular function associated with the sensory perception of smell, nervous system processes, and odorant binding (Supplementary 18). The gene-set analysis of shared loci between hypertension and PhenoAge is underpowered. Finally, we analyzed the differential gene expression patterns of the mapped genes across 54 GTEx tissue types (Supplementary Figs. 2–5). Tissue enrichment for hypertension and HannumAge indicated that the differential gene expression patterns of the mapped genes were significantly up-regulated in bladder and colon sigmoid (Supplementary Fig. 3). Otherwise, we did not find any other significant tissue enrichment between hypertension and the other three epigenetic clocks (Supplementary Figs. 2, 4, and 5).

Discussion

In this comprehensive analysis utilizing GWAS summary statistic, we uncovered polygenetic architecture between hypertension and epigenetic aging, and identified novel loci of hypertension. First, our analysis of the genetic architecture of hypertension revealed a strong degree of polygenicity, with approximately 2.8 K common variants. Since hypertension is more polygenic than epigenetic aging (0.1 K–0.7 K common variants), overlapping genomic loci account for a greater proportion of the genetic architecture of epigenetic aging compared with hypertension. Second, the statistical power was enhanced by employing conjFDR to uncover a substantial cross-trait enrichment between hypertension. We identified a total of 32 distinct loci shared between hypertension and epigenetic aging, 25 of which were identified as novel loci for hypertension. Especially, nine of 32 distinct loci were validated as hypertension-associated in two validation datasets. Further, our study provides novel perspective on their shared polygenetic architecture.

Recent studies showed that DNA methylation metrics of biological aging (GrimAge and PhenoAge) are associated with hypertension1,3. Based on DNA methylation profiles, another two studies showed that hypertension was associated with epigenetic age acceleration measured by HannumAge32,33. A study of 1100 hypertensive African Americans reported that HannumAge acceleration was associated with hypertension34. Overall, the findings of our study were in line with these above studies. However, a prospective study of 2543 black participates indicated that epigenetic acceleration measured by HanummAge was not associated with hypertension35. The inconsistency of this study’s conclusion with ours may be due to different participant populations between the datasets used for analysis.

The allelic effect directions of distinct shared loci between hypertension and epigenetic aging were in mixed patterns, indicating the complex genetic relationships, which is consistent with the results of genome-wide genetic overlapping. Among the 25 novel loci for hypertension, one was shared with GrimAge, five were shared with HunnumAge, 17 were shared with IEAA, and two were shared with PhenoAge (Table 2). Additionally, rs1849209 (near gene CTC-782O7.1, also named as LINC01478) was jointly associated with three measures of epigenetic aging (HannumAge, IEAA, and PhenoAge). CTC-782O7.1 was reported to be associated with at least one phenotype of hypertension by Li et al.36. Therefore, whether rs1849209 is associated with hypertension and epigenetic age acceleration in different populations deserves further investigation. rs1825819 (near gene RP11-83C7.1, also named as LINC02513) was identified as a novel locus for hypertension associated with HannumAge. Paul et al. concluded that rs17499404 (near LINC02513) was a life-extending locus and was associated with reduced cardiovascular disease phenotypes2. These previous studies support the validity of our findings.

We mapped all candidate SNPs to genes using the three mapping strategies (positional, eQTL and chromatin interaction mapping) in FUMA protocol37. Among these genes, CYP1B1, CYP1A1 and CYP1A2 were associated with cardiovascular diseases, especially hypertension38. CYP1B1 enzyme belongs to CYP1 gene family, which also includes CYP1A1 and CYP1A2 enzymes. LRP1B belongs to the low-density lipoprotein receptor (LDLR) gene family. It remains unclear whether it is associated with hypertension, although it might be. CACNB2 was shown as the pharmacological target for hypertension39. Besides, some gene-sets are involved in pathways associated with sensory perception of smell. Previous studies have suggested that smell perception might be a risk factor for hypertension40,41,42. A hospital-based study suggested that olfactory dysfunction was a marker for essential hypertension in a drug-naïve adult population40. They found that altered sense of smell in hypertension may be associated with certain pathogenic mechanisms, usually related to olfactory and blood pressure. A longitudinal study concluded that smell perception and taste were associated with hypertension41. Datta et al. indicated that patients with hypertension have a reduced ability to smell40. These genes associated with smell perception may represent novel treatment target, but further experimental validation is needed.

Some may doubt that inconsistent polygenicity, genetic overlap and shared loci between four epigenetic measures and hypertension suggest that these results are not plausible. However, it is important to note that these differences arise from the fact that computational models of different epigenetic clocks are based on different tissues, DNA methylation sites, and training methods to capture different features4. Evidence showed that all these four measures of epigenetic aging were accurate4.

Our study is limited in several ways. Firstly, the conjFDR approach can only indicate which SNPs are associated with a disease or trait, but it cannot determine exactly which SNP is driving the association. This is because some SNPs may be in linkage disequilibrium (LD) with the lead SNP43. In this case, the SNP associated with the phenotype may be only one of the SNPs associated with the true causative or functional association, and not necessarily the one that directly causes the association. Nonetheless, we identified the most likely causal variants within each locus and functionally characterized them for further examination and selection in experimental follow-up studies. Besides, Kresovich et al. suggested that epigenetic age indicators might be valuable for hypertension risk stratification3. Therefore, if individual-level data was available, we could explore risk loci for hypertension in different age cohorts.

Conclusions

In conclusion, we provided evidence for extensive polygenic overlap between hypertension and epigenetic aging. Then, we identified 32 shared distinct loci with mixed effect directions jointly associated with hypertension and epigenetic aging, 25 of which were novel for hypertension. Functional gene-sets pathway analyses indicated that genetic factors associated with hypertension and epigenetic aging might were likely to be implicated in olfaction. However, further biological evidence is necessary for the elucidation of the underlying mechanisms.