Introduction

Hypertension is one of the most important risk factors for cardiovascular diseases, which is the leading cause of death worldwide [1]. As with other complex traits, evidence from familial studies suggests that hypertension is caused by a combination of genetic and environmental factors. The heritability of blood pressure (BP) has been estimated at approximately 40–60% [2]. Large-scale meta-analysis studies of genome-wide association studies (GWAS) have identified numerous single-nucleotide polymorphisms (SNPs) for BP [3,4,5]. In addition, BP-associated missense variants have been identified by exome-wide studies of large samples [6,7,8].

Although GWAS have revolutionized the understanding of the genetic architecture of BP, the interpretation of the GWAS results is still a major challenge. Delineation of GWAS variants by distinguishing the functional variants from the rest should be helpful for this. Nonsynonymous variants, as well as genetic variants that have the capacity to alter protein binding [9] and affect RNA splicing [10] and editing [11], are potential BP functional variants.

N6-methyladenosine (m6A) is a pervasive RNA modification in eukaryotes. It has become a hot research topic because of its critical roles not only in the regulation of gene expression [12], messenger RNA (mRNA) stability [13], and homeostasis [14] but also in the pathogenesis of various diseases [15]. Recent studies have shown that genetic variants influence m6A by changing the RNA sequences of the target sites or key flanking nucleotides [16]. This kind of putative functional SNP is called an m6A-associated SNP (m6A-SNP). If m6A modification was interrupted by an m6A-SNP, the biological process by which m6A functions would be disturbed [16]. Thus, m6A-SNPs may have regulatory potential to affect gene expression and mRNA stability and homeostasis, which may consequently affect diseases such as hypertension.

Evaluation of the effect of genetic variants on m6A modification will increase our understanding of the variants’ pathogenic molecular mechanisms and uncover new causal variants. However, although GWAS have identified >400 loci that harbor DNA sequence variants that influence BP [5, 17], identification of m6A-SNPs is very rare, and thus far, the relationship between m6A-SNPs and BP is still unclear. In addition, determining the association between m6A and BP in large samples on a genome-wide scale is difficult to achieve. Using the GWAS identified BP-associated m6A-SNPs as a bridge, we can indirectly assess the relationship between m6A and hypertension. Thus, in this study, we investigated the association between m6A-SNPs and BP in large-scale GWAS and aimed to identify the enrichment of m6A-SNPs among the SNPs that were associated with BP.

Methods

Determination of m6A-SNPs for BP

In this study, we first investigated the effect of a new functional variant, m6A-SNPs, on systolic BP (SBP) and diastolic BP (DBP) in the published summary data of three large-scale GWAS [3,4,5]. The first is the 2011 GWAS, which comprised 69,395 individuals. The raw data used in the present analysis were the downloaded summary results from the initial GWAS, which included association P values of almost 2.5 million SNPs. The second is the 2016 GWAS, which examined the association between 196,725 variants of the custom genotyping microarray Cardio-Metabochip and SBP and DBP in 201,529 individuals of European ancestry. The available summary data contained P values for 128,272 variants. We also evaluated the effect of the m6A-SNPs on BP in the summary data from the 2018 GWAS, the third study [5]. This GWAS dataset comprised the summary results for the association between more than 10 million SNPs and SBP and DBP, which were evaluated in 317,756 individuals from the UK Biobank [18].

To screen out the m6A-SNPs from these millions of SNPs, we annotated them using a list of m6A-SNPs that were downloaded from the m6AVar database (http://m6avar.renlab.org/). The list contains 13,703 high (miCLIP/PA-m6A-seq experiments), 54,222 medium (MeRIP-Seq experiments), and 284,089 low (genome-wide prediction based on Random Forest algorithm) confidence level m6A-SNPs for humans [16]. After annotation of the SNPs in the GWAS summary dataset by the list of m6A-SNPs (merged the entire set of m6A-SNPs with each of the BP GWAS datasets), we identified the m6A-SNPs that were associated with BP. Those m6A-SNPs with P values <0.05 were considered in the following analyses.

Enrichment analysis

Among BP-associated SNPs, we determined if m6A-SNPs were overrepresented compared to what would be expected by chance. We randomly sampled 1000 sets of non-m6A-SNPs (the same number of m6A-SNPs) from the GWAS datasets for SBP and DBP as the matched background and then determined if the proportion of m6A-SNPs with P value <0.05 was significantly higher than the proportion of non-m6A-SNPs with P value <0.05 in the 1000 sets for each trait.

In addition, fgwas was used to assess the functional enrichment of m6A-SNPs as a functional annotation in SBP and DBP. The fgwas, which was a practical tool for genetic and developmental analysis of complex traits or diseases, takes advantage of summary statistics of genome-wide association and incorporates functional annotation information into a GWAS to estimate the enrichment of GWAS findings in the annotation type (e.g., m6A-SNPs) [19]. The software (version 0.3.6) can be downloaded at https://github.com/joepickrell/fgwas.

Association in East Asian populations

Because of substantial ethnic differences in allele frequencies and disease susceptibility, BP association has not been necessarily replicated in East Asians [20]. We looked for evidence on associations between the identified m6A-SNPs and BP traits in three GWAS from East Asian populations. The first study analyzed genome-wide (475,157) SNP data of 400 matched pairs of young-onset hypertensive patients and normotensive controls genotyped with the Illumina HumanHap550-Duo BeadChip [21]. The second study was a GWAS meta-analysis of mean arterial pressure and pulse pressure among 26,600 East Asian participants [22]. Genome-wide data for approximately 2.4 million SNPs were available. Third, the supplementary data from a meta-analysis of GWASs of BP and hypertension in 11,816 Chinese individuals followed by replication studies, including 69,146 additional individuals were also searched [23]. The m6A-SNPs with P values <0.05 were considered.

Association with coronary artery disease and stroke

We evaluated the associations between the identified BP-associated m6A-SNPs and coronary artery disease (CAD) using the summary data from a large-scale 1000 Genomes-based GWAS meta-analysis of CAD carried out by the CARDIoGRAMplusC4D consortium [24]. This study comprised approximately 185,000 individuals who were mainly (77%) of European descent. This dataset is publicly available at http://www.cardiogramplusc4d.org/data-downloads/.

We also evaluated the associations between the identified m6A-SNPs and ischemic stroke using the summary data from previous GWAS, which comprised 60,341 ischemic stroke patients and approximately 400,000 controls [25]. The raw data used in the present analysis were the downloaded summary results from the initial GWAS, which included association P values of almost eight million SNPs and indels for ischemic stroke. The dataset is publicly available at the MEGASTROKE website (http://megastroke.org/). The m6A-SNPs with CAD or stroke association P values <0.05 were considered in this analysis.

Gene expression analysis

The m6A-SNPs may take part in gene expression regulation through exerting influence on RNA modification; thus, they may be associated with gene expression levels. We carried out expression quantitative locus (eQTL) analysis to obtain evidence on associations between the identified BP-associated m6A-SNPs and gene expression using the HaploReg browser (https://pubs.broadinstitute.org/mammals/haploreg/haploreg.php). HaploReg is a database that manually collates and updates data from large sample eQTL studies and contains data from the ENCODE project [26]. Furthermore, we tested genotypes and mRNA expression levels in peripheral blood mononuclear cells (PBMCs) of 40 unrelated Chinese Han individuals (age range from 27 to 67 years) using reverse transcription-polymerase chain reaction to obtain additional evidence to support the top BP-associated m6A-SNPs.

Results

BP-associated m6A-SNPs

The first step of this study was to select m6A-SNPs from the BP GWAS dataset according to the annotation information of the 352,014 m6A-SNPs in the m6AVar database. We found 1712 unique m6A-SNPs in the 2011 GWAS dataset. Among these SNPs, 97 (5.7%) and 129 (7.5%) (166 unique) were nominally (P < 0.05) associated with SBP and DBP, respectively. Two of the 1712 tested m6A-SNPs reached a significance level of 1.5 × 10–5. The most significant m6A-SNP for SBP (P = 2.68 × 10–6) and DBP (P = 9.58 × 10–8) was rs7398833 in the CUX2 gene. The second m6A-SNP was rs13096477 in the SLC4A7 gene (P = 2.85 × 10–6 for DBP).

We found 94 unique m6A-SNPs in the 2016 GWAS dataset, and a total of 32 m6A-SNPs were associated with SBP and/or DBP among the 128,272 variants (P < 0.05). Rs9847953 (ZNF589), rs1801253 (ADRB1), and rs7398833 (CUX2) reached the genome-wide significance level (P < 5 × 10–8) (Table 1). The association between rs197922 (GOSR2) and SBP was also significant (P = 9.58 × 10–8).

Table 1 The significant m6A -SNPs for SBP and DBP in GWAS-2016
Full size table

In the UK Biobank data of GWAS-2018, 761 (11.1%) and 799 (12.1%) (1236 unique) m6A-SNPs were nominally (P < 0.05) associated with SBP and DBP (Supplementary Tables 1 and 2), respectively. Among these SNPs, 13 and 26 (33 unique) were associated with SBP and DBP at the genome-wide significance level (Tables 2 and 3), respectively. These genome-wide significant m6A-SNPs were in linkage disequilibrium (r2 ≥ 0.2 in Europeans) with the sentinel SNPs of the corresponding BP-associated loci (Supplementary Table 3). The four significant m6A-SNPs found in GWAS-2016 were all significant in GWAS-2018. The 1236 m6A-SNPs were associated with 1241 m6A sites, 4.2% and 18.0% of which belong to the high and medium confidence levels, respectively. Among these high and medium confidence level SNPs, 2 and 8 (9 unique) were associated with SBP and DBP at the genome-wide significance level (Tables 2 and 3), respectively.

Table 2 The significant m6A-SNPs for SBP in GWAS-2018
Full size table
Table 3 The significant m6A -SNPs for DBP in GWAS-2018
Full size table

Enrichment of m6A-SNPs

By applying the genome-wide association data of GWAS-2011 and GWAS-2018, we carried out the enrichment analyses. The proportion of non-m6A-SNPs with GWAS P values <0.05 for DBP (95% confidence interval [CI]: [4.4%, 6.5%], P = 6.68 × 10–5) was significantly lower than that of the m6A-SNPs in the GWAS-2011 data. However, it was not significant for SBP (95% CI: [4.3%, 6.5%], P = 0.31). In the GWAS-2018 data, the proportion of non-m6A-SNPs with GWAS P values <0.05 for SBP (95% CI: [9.5%, 10.9%], P = 3.47 × 10–3) and DBP (95% CI: [9.8%, 11.2%], P = 5.81 × 10–6) was significantly lower than the m6A-SNPs. Using fgwas, we found that SNPs associated with DBP (P < 5 × 10–8) were significantly enriched with m6A-SNPs (log 2 enrichment of 2.67, 95% CI: [0.42, 3.68]).

Association in East Asian populations

In the three GWAS from East Asian populations, we found that 57 of the 1236 identified m6A-SNPs were nominally associated with BP (Supplementary Tables 1 and 2). Among the m6A-SNPs that were significantly associated with BP at a genome-wide significance level in Europeans, rs56001051 (C1orf167, P = 0.046) and rs197922 (GOSR2, P = 0.041) were associated with hypertension. rs853678 (ZSCAN31, P = 0.021), rs9657518 (RP1L1, P = 0.024), rs197922 (GOSR2, P = 0.031), and rs740406 (DOT1L, P = 0.018) were associated with mean arterial pressure. rs740406 (DOT1L, P = 5.0 × 10–4) was associated with pulse pressure in East Asian populations.

Association with CAD and stroke

To further evaluate whether the 1236 BP-associated m6A-SNPs were associated with CAD, we tested for association in approximately 185,000 individuals by using the CARDIoGRAMplusC4D GWAS dataset [24]. We found that 78 of the 761 SBP-associated m6A-SNPs and 72 of the 799 DBP-associated m6A-SNPs seemed to be associated with CAD (P < 0.05) (Supplementary Tables 1 and 2). The most significant (P = 4.50 × 10–9) m6A-SNP for CAD was the nonsense SNP rs12286 of the ADAMTS7 gene (Supplementary Table 2). We found that 56 of the 761 SBP-associated m6A-SNPs and 61 of the 799 DBP-associated m6A-SNPs seemed to be associated with ischemic stroke (P < 0.05) (Supplementary Tables 1 and 2). The most significant (P = 1.78 × 10–5) m6A-SNP for ischemic stroke was rs7398833 in the 3′-untranslated region (UTR) of the CUX2 gene. For the 33 BP-associated m6A-SNPs that passed the genome-wide significance level, rs7398833, rs197922, rs740406, rs8024, rs9847953, rs564449, and rs6066802 were associated with CAD, and rs7398833, rs8024, rs1353776, rs853678, and rs12828755 were associated with ischemic stroke (P < 0.05) (Tables 2 and 3). Among them, rs7398833 in CUX2 and rs8024 in IPO9 were associated with both CAD (P = 1.22 × 10–3 and 1.78 × 10–5) and ischemic stroke (P = 3.60 × 10–5 and 7.03 × 10–3) with consistent effects, respectively.

Gene expression analysis

To further clarify the possible functional mechanisms underlying the identified m6A-SNPs in association with BP, we investigated whether they were associated with the expression levels of local genes. Most of these m6A-SNPs were strongly associated with gene expression in cis or trans effects. In total, 359 (217 for SBP and 246 for DBP) of the 1236 BP-associated m6A-SNPs (P < 0.05) may have a cis effect on the 339 corresponding genes in different cells or tissues (Supplementary Tables 1 and 2). Some of these 359 m6A-SNPs were found in multiple tissues or cells, and 20 of them could be considered cis-eQTLs (P < 5 × 10–8) (Tables 2 and 3). These cis-eQTLs are proxies of the sentinel variants in the BP-associated loci (Supplementary Table 3). Fifty-eight of the 359 cis-effect m6A-SNPs showed associations with CAD, and 33 showed associations with ischemic stroke. The most significant cis-eQTLs for SBP, DBP, CAD, and ischemic stroke were rs56001051 in C1orf167 (P = 1.49 × 10–21), rs2275155 in SDCCAG8 (P = 1.06 × 10–21), rs12286 in ADAMTS7 (P = 4.50 × 10–9), and rs5213 in KCNJ11 (P = 2.46 × 10–5), respectively.

We noticed that rs56001051 (C1orf167), rs197922 (GOSR2), rs740406 (DOT1L), rs8024 (IPO9), rs9847953 (ZNF589), rs853678 (ZSCAN31), rs12828755 (PITPNM2), and rs6066802 (PREX1) were significantly associated with BP (P < 5 × 10–8) and cis-eQTLs in Europeans and showed evidence of association with BP traits in the East Asian populations or association with CAD or stroke. We tested the relationship between these eight BP-associated m6A-SNPs and mRNA expression levels of the corresponding genes in PBMCs from Chinese individuals. We found that rs56001051, rs9847953, rs197922, and rs740406 were associated with C1orf167 (P = 0.019), ZNF589 (P = 0.013), GOSR2 (P = 0.001), and DOT1L (P = 0.032) expression levels, respectively (Fig. 1).

Fig. 1
figure 1

Association between top blood pressure (BP)-associated N6-methyladenosine-associated single-nucleotide polymorphisms (m6A-SNPs) and messenger RNA (mRNA) expression levels. The association between m6A-SNPs and mRNA expression levels was evaluated in peripheral blood mononuclear cells (PBMCs) of 40 Chinese individuals. The minor allele G carriers of rs56001051 tended to have higher C1orf167 gene expression levels. The minor allele G carriers of rs9847953 tended to have higher ZNF589 gene expression levels. The minor allele A carriers of rs197922 tended to have higher GOSR2 gene expression levels. The minor allele G carriers of rs740406 tended to have higher DOT1L gene expression levels

Full size image

Discussion

This study represents the first effort to identify m6A-SNPs that influence BP by excavating data from large-scale GWAS. We found many m6A-SNPs that were associated with BP. These m6A-SNPs (e.g., rs9847953 and rs197922) might be functional variants that have the potential to affect gene expression (e.g., ZNF589, GOSR2), by which they affect BP.

Delineation of m6A-SNPs for BP has the potential to pinpoint this kind of multifunctional SNP as a causal variant. Studies have shown that m6A modification plays a pivotal role in the regulation of downstream molecular events, such as nuclear export, stability, translatability, splicing, and miRNA processing [27]. The m6A-SNPs, which were quite close to the methylation sites or were the exact methylation site, would lead to gain or loss of m6A methylation sites [16]. Therefore, one of the functional interpretations for the association between m6A-SNPs and BP is that the SNPs could influence m6A methylation. On the other hand, we noticed that many of the identified m6A-SNPs were missense mutations or variants that could influence transcription. Indeed, most of the identified m6A-SNPs have the potential to alter the binding of regulatory elements, which may in turn regulate gene expression. The nonsynonymous variants could not only change the encoded amino acids but also influence promoter activity, mRNA stability, and structure, and subcellular localization of proteins [28]. SNPs in UTRs are known to have the ability to alter microRNA or transcription factor binding. Thus, the functional interpretations for the missense and regulatory m6A-SNPs should include these well-known inherent attributes. However, although we have identified many BP-associated m6A-SNPs and first demonstrated that m6A-SNPs may play important roles in BP regulation, to date, how m6A affects BP is unclear. The underlying mechanisms still need to be elucidated.

Some of the identified m6A-SNPs and genes showed strong associations with BP. For example, the nonsynonymous m6A-SNP rs9847953 in ZNF589 has the potential to alter the binding of 32 proteins and a regulatory motif LBP-1 and is located in CpG islands and DNase I hypersensitive sites (Supplementary Fig. 1). ZNF589 protein, also known as stem cell zinc-finger 1 (SZF1), belongs to the large family of Krüppel-associated box domain zinc-finger transcription factors. A recent study has established the role of SZF1/ZNF589 as a new functional regulator of the hematopoietic system [29]. Data from the Atherosclerosis Risk in Communities study and the Women’s Genome Health Study have shown that the missense (Lys67Arg) m6A-SNP rs197922 in GOSR2 is associated with hypertension in white individuals [30]. As our analyses showed that rs9847953 and rs197922 were strongly associated with gene expression levels, the identified m6A-SNPs may have the potential to regulate BP by altering gene expression. The nonsynonymous m6A-SNP rs1801253 in ADRB1 has the ability to alter the binding of proteins ELF1 and MAX and a regulatory motif STAT and is located in CpG islands and DNase I hypersensitive sites (Supplementary Fig. 1). The β1-adrenergic receptor is crucial in regulating cardiac output and agents by preventing lower blood pressure [31]. The association between rs1801253 (known as Arg389Gly, which was identified in 1999) and hypertension has been well studied [31,32,33]. This polymorphism and β-blocker treatment duration are independent factors associated with β-blocker treatment outcome [34]. The detected BP-associated m6A-SNP in 12q24.11 was rs7398833 in the CUX2 gene, while the reported SNP was rs3184504 in SH2B3 [3]. The CUX2 gene has been shown to directly regulate the expression of NeuroD and function at multiple levels during spinal cord neurogenesis [35]. The association between rs7398833 and BP has not been reported before, but SNPs in this gene have been shown to be associated with type 1 diabetes [36] and CAD [37]. The present study showed that the rs7398833 C allele may result in the gain of an m6A site and may lead to lower SBP and DBP levels and lower risk of CAD and ischemic stroke. The associations between these m6A-SNPs and BP were not significant in GWAS-2011. However, as the sample size increased in GWAS-2016 and GWAS-2018, the associations were confirmed. These m6A-SNPs and genes should be important candidates for further genetic association and functional studies.

This study has several limitations. First, the sample sizes of BP GWAS from East Asian populations were small, so very few associations between m6A-SNPs and BP have been identified. Second, only a very small proportion of m6A SNPs were examined in this study (1712 of 313,000). To fully recognize the impact of m6A-SNPs on BP regulation, the effects of a large number of m6A-SNPs (especially the rare variants) on BP should be evaluated in larger datasets (e.g., sequencing data). Finally, the functionalities of the detected SNPs, especially the effects on m6A modification, have not been validated by technical and biological experiments. Further experimental studies are needed to determine their functions, for example, by performing gene editing at the single nucleotide of the SNP and observing the effect on gene expression in cell lines or the effect on BP in animals.

In summary, the present study found many BP-associated m6A-SNPs and demonstrated their potential functionalities. This study increased our understanding of the regulation patterns of SNPs and may provide new clues for further detection of the functional mechanism underlying the associations between SNPs and hypertension. Although we found supplementary functional information to support the significant findings, further studies are needed to elucidate the mechanisms.