Genome-Wide Interaction Analysis with DASH Diet Score Identified Novel Loci for Systolic Blood Pressure

Abstract

Background:

The Dietary Approaches to Stop Hypertension (DASH) diet score lowers blood pressure (BP). We examined interactions between genotype and the DASH diet score in relation to systolic SBP.

Methods:

We analyzed up to 9,420,585 single nucleotide polymorphisms (SNPs) in up to 127,282 individuals of six population groups (91% of European population) from the Cohorts for Heart and Aging Research in Genomic Epidemiology consortium (CHARGE; n=35,660) and UK Biobank (n=91,622) and performed European population-specific and cross-population meta-analyses.

Results:

We identified three loci in European-specific analyses and an additional four loci in cross-population analyses at P_interaction < 5e-8. We observed a consistent interaction between rs117878928 at 15q25.1 (minor allele frequency = 0.03) and the DASH diet score (P_interaction = 4e-8; P for heterogeneity = 0.35) in European population, where the interaction effect size was 0.42±0.09 mm Hg (P_interaction = 9.4e-7) and 0.20±0.06 mm Hg (P_interaction = 0.001) in CHARGE and the UK Biobank, respectively. The 1 Mb region surrounding rs117878928 was enriched with cis-expression quantitative trait loci (eQTL) variants (P = 4e-273) and cis-DNA methylation quantitative trait loci (mQTL) variants (P = 1e-300). While the closest gene for rs117878928 is MTHFS, the highest narrow sense heritability accounted by SNPs potentially interacting with the DASH diet score in this locus was for gene ST20 at 15q25.1.

Conclusion:

We demonstrated gene-DASH diet score interaction effects on SBP in several loci. Studies with larger diverse populations are needed to validate our findings.

Graphical Abstract

Introduction

The Dietary Approach to Stop Hypertension (DASH) diet promotes consumption of fruits, vegetables, whole grains, nuts and legumes, and low-fat dairy products, and limits the consumption of red and processed meats, sugar-sweetened beverages, and sodium [1]. The DASH diet is based on diets provided in a multicenter, randomized controlled trial, where the primary results successfully reduced blood pressure (BP) among individuals with prehypertension and hypertension and was subsequently shown in several additional trials [1, 2]. In population-based studies, DASH diet scores have been developed to reflect an individual’s adherence to the DASH diet. Several independent studies showed that higher DASH diet scores were associated with a decreased risk of hypertension and cardiovascular disease [3–6].

The genetic architecture of hypertension has been studied extensively [7, 8]. For example, a recent genome-wide association study (GWAS) reported that individuals in the high polygenic score quintile, calculated based on over 2,000 BP-associated loci, had a five-fold greater risk of hypertension compared to those in the lowest quintile with a low polygenic score [8]. An unhealthy lifestyle such as poor diet, smoking, and sedentary lifestyle is also a significant risk factor of hypertension and may interact with genetic factors. For example, an earlier study in UK Biobank showed an interaction between a polygenic score, built with 314 BP-associated loci, and a healthy lifestyle score for systolic BP. Compared to individuals with a low lifestyle score, systolic BP was 4.9 mm Hg lower in those with a high lifestyle score and a low genetic risk and 4.1 mm Hg lower in those with a high lifestyle score and a high polygenic score [9]. Gene-diet interaction has also been examined in other general populations [10–13], however, these studies are limited to investigating specific dietary components such as sodium intake [10], different types of dietary fat [11, 12], or caloric intake [13]. Little is known about the interaction between genes and overall diet quality in relation to BP in the general population. To address this knowledge gap, the Gene-Lifestyle Interactions Working Group, part of the CHARGE consortium [14], investigated genome-wide gene-DASH diet score interaction in relation to systolic BP.

Methods

Study Populations.

All data and materials for the CHS, ARIC, FHS, WHI cohorts have been made publicly available at the dbGaP and can be accessed at https://www.ncbi.nlm.nih.gov/gap/. Requests to access the Raine, NEO, SP & LVB and UK Biobank study datasets from qualified researchers trained in human subject confidentiality protocols may be sent to https://rainestudy.org.au/, http://www.einthovenlaboratory.com/onderzoeken/neo-study/, https://www.kp4cd.org/node/289, and https://www.ukbiobank.ac.uk/ respectively. Cohort specific acknowledgements and information can be found in the supplemental material (Appendices S2–S3). We analyzed six participating cohorts in the CHARGE consortium and the UK Biobank (Appendix S3, Table S1). Descriptions of the CHARGE consortium and the UK Biobank have been published previously [15, 16]. We included participants if they were aged 18 to 80 years and had no missing data on genotype, diet, systolic BP measurements, or covariates. The different populations in these cohorts included African population (AFR), Admixed American population (AMR), Central/South Asian population (CSA), East Asian population (EAS), European population (EUR), and Middle Eastern population (MID). In the UK Biobank, we excluded those who were related to other participants up to the third degree. All participants provided written informed consent. The present study protocol was approved by the Tufts University Institutional Review Board.

Study Design.

The study design is depicted in Figure 1. We developed the analytical protocol based on the method created by the Gene-Lifestyle Interactions Working Group [14]. Each cohort conducted population-specific analysis according to this protocol. Population-specific and cross-population meta-analyses were then conducted centrally. Detailed information regarding systolic BP assessment [7, 14], DASH diet score calculations, and genotyping [17, 18] are presented in Appendix S4.

Figure 1.

Study design. E is the DASH diet score, SNP is the dosage of the genetic variant (coded additively), C is the vector of cohort-specific covariates.

Statistical analysis.

In the Framingham Heart Study (FHS) and the UK Biobank, we examined cross-sectional associations of the DASH diet score with systolic BP in a multivariate model with adjustment for age, age squared, sex, energy intake, and population (only in the UK Biobank) in the first model, and additionally adjusted for body mass index (BMI) in the second model.

Within each cohort, we first performed population-specific interaction analyses with the quantitative DASH diet score by examining the multiplicative interaction with SNP dosage. To explore whether the potential gene-diet interaction was driven by threshold effect, we also analyzed interactions with the DASH diet score dichotomized by its median or lower quartile. Linear regression models or linear mixed effect models (for familial data in the CHARGE cohorts) were run adjusting for age, age squared, sex, energy intake, BMI, field center (for multi-center studies), cohort-specific SNP-based principal components, and additional cohort-specific covariates, if any. Narrow sense heritability was approximated by the R² derived from the regression models. The EasyQC R package with 1000 Genomes data as reference [19] was used to conduct QC of summary statistics across all CHARGE cohorts. We followed the UK Biobank’s QC protocol [17] and only analyzed UK Biobank SNPs if these SNPs were also analyzed in the CHARGE cohorts. An inverse variance-weighted, fixed-effect meta-analysis was performed to combine cohort-specific results using METAL. Genomic control was applied in the meta-analyses.

In secondary analyses, we examined associations for SNP-systolic BP (i.e., a main-effect model) and performed a 2-degree-of-freedom (2df) test [18] to jointly examine both SNP main effect and SNP-DASH diet interaction effect, with adjustment for the same covariates included in the interaction analysis (Figure 1). Similarly, population-specific analyses were conducted in each cohort, and meta-analyses were performed to combine cohort-specific findings. For the interaction analyses and the 2df tests, robust estimates of the standard errors (SE) and covariances were used in meta-analyses to protect against potential misspecification of the mean models [14]. In all analyses, heterogeneity across cohorts was determined based on Cochran’s Q-test. We considered SNPs with joint P <5e-8 that were present in at least two cohorts as significant. Novel loci were identified as SNPs with P-value < 5e-8 that are not in linkage disequilibrium (LD) R² ≥0.1; based on the 1000 Genomes data) or are ±500 kb from any previously validated BP–associated SNPs in the GWAS Catalog [20].

We conducted a colocalization analysis to further evaluate whether SNPs with an interaction effect (with the DASH diet score) were independent of those significant in the main effect GWAS (Appendix S4). Additionally, we carried out analyses to determine whether these interaction-based loci are enriched with eQTL or mQTL variants [21, 22] (Appendix S4).

Results

Participant Characteristics.

We analyzed data from up to 35,660 individuals: 28,478 EUR, 2,751 AFR, and 4,431 EAS participants from cohorts participating in the CHARGE consortium and up to 91,622 unrelated UK Biobank participants comprising six population groups: EUR (N=86,932), AFR (N=1,557), EAS (N=658), CSA (N=1,898), MID (N=312), and AMR (N=265). Mean age ranged from 20 to 75 years in the CHARGE cohorts and 51 to 57 years in the UK Biobank. Both the CHARGE cohorts and the UK Biobank included more women than men, 59.1% and 54.2%, respectively. Demographic characteristics of participants are shown in Table S1.

Association of the DASH diet score with systolic BP.

A higher DASH diet score was inversely associated with systolic BP (Figure S1). Systolic BP was 2.4±0.4 mm Hg lower in the FHS (P = 6.1e-8) and 1.1±0.2 mm Hg lower in the UK Biobank (P = 4.2e-12) per 10 units increase in the DASH diet score. The inverse association became nonsignificant (0.2±0.2 mm Hg; P = 0.24) in UK Biobank, while the association remained significant in FHS (1.4±0.4 mm Hg; P = 7.6e-4) after additionally adjusting for BMI.

Gene-DASH diet score interaction in relation to systolic BP in EUR population.

We examined 8,454,957 common biallelic SNPs available in at least two EUR cohorts. In the meta-analysis of all EUR individuals, we found potential interaction for the quantitative DASH score at three independent loci at 15q25.1 (lead SNP rs117878928, Methenytetrahydrofolate synthase [MTHFS], P for interaction [P_int] =4e-8), 16q23.1 (lead SNP rs28562150, WWOX, P_int =3.9e-8), and 18q21.2 (lead SNP rs138826501, P_int =2.6e-8) (Table 1; Figures S2 and S3). The direction of the interaction between rs117878928 (MTHFS at 15q25.1) and the quantitative DASH score was consistent in all study populations (P_het = 0.35; Figure S4). The interaction effect size (mean and SE) for the lead SNP, rs117878928, was 0.42±0.09 (P_int = 9.4e-7) and 0.20±0.06 (P_int = 0.001) in the CHARGE cohorts and the UK Biobank, respectively. The other two loci were statistically significant in the CHARGE cohorts but not in the UK Biobank (Table 1). These results suggest that the relationship between the DASH diet score and systolic BP may depend on an individual’s genotype, particularly the rs117878928 genotype.

Table 1.

Statistically Significantly interacting loci with the quantitative DASH diet score on systolic BP in meta-analysis of EUR participants.

													UKB				CHARGE
SNP	Gene	Chr	Pos.38	Region	EA	OA	EAF	Effect	SE	Pint	Phet	N	Effect	SE	P	N	Effect	SE	P
rs117878928	MTHFS	15	79841858	15q25.1	A	C	0.03	0.28	0.05	4e-8	0.35	86932	0.20	0.06	0.001	28507	0.42	0.09	9.4e-7
rs28562150	WWOX	16	78530453	16q23.1	A	G	0.01	0.48	0.09	3.9e-8	1.85e-8	86932	0.21	0.11	0.05	13628	0.99	0.15	2.7e-11
rs138826501		18	51789596	18q21.2	T	C	0.01	0.37	0.07	2.6e-8	0.003	86932	0.12	0.11	0.25	6945	0.54	0.09	4.2e-10

Interaction analysis with dichotomized DASH diet scores.

To explore if the interaction analysis was influenced by any irregularities in the distribution of the diet score, we also analyzed the DASH diet score dichotomized by either the cohort-specific median or lower quartile (Figure S2). In EUR participants, correlations of effect size and log10 P_int were slightly stronger in the analysis using the continuous DASH diet score compared to the median or lower quartile dichotomized DASH score (Figure S5). For SNPs with P_int < 1e-3 in the quantitative DASH diet score analysis, their effect sizes and log10 P_int were correlated with that from analyses using the median dichotomized DASH score (Pearson r = 0.92 and 0.31, respectively; Figure S6) and the lower quartile dichotomized DASH score (Pearson r = 0.74 and 0.16, respectively; Figure S6). This analysis supports our primary interaction analysis using the continuous DASH score, although the findings were corroborated by the dichotomized analyses.

Colocalization analysis.

To examine if the relationship of the lead interaction SNPs at the three loci colocalized with known systolic BP associated SNPs at the same loci (±1Mb region surrounding the lead SNPs), we conducted colocalization analysis. At the MTHFS locus (15q25.1), the colocalization analysis demonstrated that rs117878928 (MTHFS at 15q25.1) was a potential causal variant of systolic BP independent of the GWAS SNPs (i.e., SNPs with significant main effect in previous GWAS; posterior probably of H₃ = 0.97; Table 2; Figure S7). Because of low posterior probability of H₁, H₃, or H₄, colocalization analysis did not support that the interaction SNPs identified in the other two loci (16q23.1 and 18q21.2) were causal variants to systolic BP (Table 2). This analysis validated that one of the three interaction loci was causal independent of the GWAS loci.

Table 2.

Colocalization analysis for the three statistically significant interaction loci in the EUR analysis.

Lead SNP	rs117878928	rs28562150	rs138826501
Number of SNPs tested	5672	9885	5304
PP.H0	0.00	0.57	0.00
PP.H1	0.00	0.18	0.00
PP.H2	0.03	0.19	0.91
PP.H3	0.97	0.06	0.09
PP.H4	0.00	0.01	0.00

Stratified analysis of lead SNP at 15q25.1.

Low heterogeneity and the colocalization analysis suggest that, out of the three significant loci, rs117878928 at the MTHFS locus may be causally interacting with the quantitative DASH score and associated with systolic BP. We therefore conducted stratified analyses by rs117878928 genotype, CC (dosage ≤ 0.35, n = 8,1792), CA (dosage ≥ 0.75 and ≤ 1.25, n= 4,967), AA (dosage ≥ 1.65, n= 75) in the UK Biobank. CA and AA were combined due to small sample size (n = 5,042), and participants with ambiguous rs117878928 genotype were excluded. After adjusting for sex, age, age squared, energy intake, and BMI, we observed that one SD higher DASH diet score was associated with 0.15±0.07 (P = 0.02) mm Hg lower systolic BP in individuals with CC genotype, whereas one SD higher DASH diet score was associated with 0.78±0.27 (P = 0.004) mm Hg higher systolic BP in those with CA or AA genotype.

Expression and DNA methylation quantitative loci variants at the MTHFS locus.

To further understand the potential biological function of the MTHFS locus, we analyzed gene expression and DNA methylation associated SNPs (i.e., eQTL and mQTL variants) within a 1Mb region of rs117878928. We found that this locus was enriched with both cis-eQTL variants (Fisher exact test; P = 4e-273) and cis-mQTL variants (P = 1e-300). Figure 2 showed the link between SNPs with P_int < 1e-3 and cis-eQTL and cis-meQTL variants. In this region, we found 419 cis-eQTL variant-gene transcript pairs from 144 unique cis-eQTL variants and seven genes (including four protein coding genes; Table S4) and 1,629 cis-mQTL variant-CpG (i.e., DNA methylation site) pairs from 151 unique cis-mQTL variants and 32 CpGs (mapped to five protein coding genes; Table S5) [21, 22]. For example, we found that a cis-eQTL variant, rs12915498 (P_int = 8.8e-4), accounted for 9.4% of heritability of expression levels of MTHFS and a cis-meQTL variant, rs11856431 (P_int = 6.8e-4), accounted for 22.7% of heritability of methylation levels of cg23855392 (a DNA methylation site ~6 kb away from the transcription start site of MTHFS). The highest heritability accounted by SNPs potentially interacting with the DASH diet score was for ST20 (a neighbor gene of MTHFS at 15q25.1). A cis-eQTL variant, rs35666771 (P_int = 5.2e-5), accounted for 11.1% of heritability of expression levels of ST20 and a cis-mQTL, rs3178646 (P_int = 9.9e-4), accounted for 46.5% of heritability of methylation levels of cg21315874 (a DNA methylation site ~6 kb away from the transcription start site of ST20).

Figure 2.

Expression quantitative trait loci (eQTL) variants and DNA methylation quantitative trait loci (meQTL) variants in the 1mb region surrounding the rs117878928 (MTHFS). The first track represents protein coding genes: MTHFS (red), ST20 (green), FAH (purple), and ARNT2 (blue). The second track represents CpGs or non-protein coding genes. The third track represents SNPs interaction P-values or narrow sense heritability (h²). Green and yellow bands link significantly associated meQTLs and eQTLs with DASH interacting SNPs, respectively.

We found that 139 cis-mQTL variants with P_int < 0.001 at the MTHFS locus were linked to four CpGs (cg13805518, cg02196730, cg26673396, and cg00225070) that were associated with a Mediterranean-style diet score at random-effect meta-analysis P < 0.05 [23]. The top cis-mQTL variants of the four diet-associated CpGs are presented in Table S6. Furthermore, 81.2% of cis-mQTL variants associated with cg13805518 (annotated to ARNT2 at 15q25.1) were cis-mQTLs of cg13148921 (another CpG annotated to ARNT2; Table S7), which was nominally associated with systolic BP in previous meta-analysis (P = 0.01) [24]. For example, rs11072902 (P_int = 2.6e-5) was associated with cg13805518 at P = 1e-46 and cg13148921 at P = 1e-8 [22]. However, none of the cis-mQTLs of cg13805518 and cg13148921 was in strong LD with the lead SNP (rs117878928) for gene-DASH diet score interaction, LD R² < 0.01. This analysis suggests that some genetic variants potentially interacting with the DASH diet at the MTHFS locus may affect systolic BP via a DNA methylation related mechanism.

2df test in EUR participants.

In secondary analysis, we ran a 2df test to jointly evaluate the SNP and diet-SNP interaction effect. Most SNPs (93%; n=1230) with P_joint < 5e-8 in the 2df joint analyses were driven by their main genetic effect on systolic BP (P_main < 5e-8; Figure S8A). We found that 11 loci reached P_joint < 5e-8 with main effect P_main ≥ 0.001 (Table S8; Figure S8A). In these eleven loci, five loci (45%) had P_int < 0.001 (Table S8). We further compared our 2df test statistics with that from the GWAS (i.e. main genetic effect) conducted by the International Consortium for Blood Pressure (ICBP) and the UK Biobank in N~750k EUR individuals [25]. As shown in Figure S8B, 98.7% (1230) SNPs with P_joint < 5e-8 in the 2df test had P < 5e-8 in the previous GWAS for BP [25]. Among these, none of the SNPs with P_joint < 5e-8 in the 2df test had P_int < 0.001, and 43 SNPs had P_int < 0.005. Five loci with P_int < 0.05 were shown in Table S9. Overall, this analysis using the 2 degree-of-freedom test identified a number of additional SNPs with moderate interaction effects with the DASH diet score.

Cross-population analysis.

In this analysis, we included 9,420,585 bi-allelic SNPs available in at least two studies with more than one population group. Five loci reached P_int < 5e-8, including one locus at 16q23.1 with statistical significance in EUR analysis. It should be noted that three of these five loci are driven by low frequency SNPs. (Table 3; Figure S9 for Manhattan plot; Figure S10 for regional plots). The lead SNPs in the five loci are intronic variants to THSD7B, SPATA5, UBE3D, GATA4, and WWOX, respectively. None of the SNPs with P_int < 0.05 at 1 mb region surrounding the five lead SNPs overlapped with SNPs associated with systolic BP (P < 1e-5) reported in the GWAS catalog [20].

Table 3.

Statistically significant SNPs in cross-population meta-analysis of gene-quantitative DASH diet score interaction.

SNP	Chr	Pos.hg38	Gene	Region	N	EA	OA	EAF	Beta	SE	Interaction P	Heterogeneity P
rs145158769	2	137235853	THSD7B	2q22.1	4918	A	T	0.02	−1.73	0.3	6e-9	1.4e-5
rs180939244	4	122940020	SPATA5	4q28.1	102530	C	G	0.71	−0.44	0.08	9.5e-9	4.4e-16
rs117137155	6	82895547	UBE3D	6q14.1	6038	T	C	0.99	2.06	0.28	1e-13	8.3e-11
rs116170345	8	11689753	GATA4	8p23.1	3477	A	C	0.99	−1.26	0.21	1.6e-9	0.04
rs28633096	16	78529552	WWOX	16q23.1	109405	T	C	0.31	0.41	0.08	5e-8	6.7e-6

Discussion

In this genome-wide interaction analysis of the CHARGE cohorts and the UK biobank, we showed that the association of DASH diet score with systolic BP was modified by multiple SNPs: at three loci in EUR analyses and an additional four loci in cross-population analysis. Our interaction SNP hits are independent of known BP loci from BP-GWAS. Furthermore, at the MTHFS locus (15q25.1), we demonstrated that the SNP-DASH diet score interaction may affect systolic BP through regulating levels of DNA methylation at this locus. While limitations exist in this study, our findings provide novel insights into gene-diet interactions in BP with respect to a better understanding of potential mechanisms of BP regulation.

A recent review highlighted several genes that may interact with lifestyle factors to modify the risk of hypertension [26]. As indicated in this review, most studies had modest sample sizes, and the observed interactions were often different across studies. Here, we also observed that most of the loci with significant interactions had high heterogeneity, specifically the poor replication between observations in the CHARGE cohorts and the UK Biobank (Figure S10). This high heterogeneity may be partly due to the diverse food habits that limit the ability of the DASH diet score to consistently reflect overall diet quality across cohorts or different dietary tools used in our participating cohorts. The present study adds novel evidence to the literature regarding interaction between genetic variants and overall diet quality, nonetheless, future analyses with larger sample sizes and better harmonized dietary information are needed to validate our findings.

At 15q25.1, the lead SNP, rs117878928 resides ~2 kb downstream of long noncoding RNA (LOC124903536) and ~2 kb upstream of a protein coding gene (MTHFS). MTHFS encodes methenyltetrahydrofolate synthase that catalyzes the conversion of 5-formyltetrahydrofolate to 5,10-methenyltetrahydrofolate. Activation of MTHFS may accelerate folate catabolism by modifying folate one-carbon forms, leading to impaired methylation reactions such as DNA methylation [27]. Folate metabolism has been implicated in the risk of hypertension [28], although its role is not fully established. Our observation regarding the enrichment of cis-mQTLs in this locus appears to be consistent with the functionality of MTHFS. In FHS, SNP rs117878928 is a cis-mQTL variant for cg21315874 (h² = 0.015, P = 4.4e-15) [22], which is residing at 5’ untranslated region (UTR) of ST20. In the Genetics of DNA Methylation Consortium (GoDMC) [29], rs117878928 is also a cis-mQTL variant for other CpGs (e.g., cg22389121, P=2.3e-42) at 5’UTR or close the transcription start site of ST20. ST20 is adjacent to MTHFS, and it is a tumor suppressing gene involved in several processes such as apoptotic signaling pathway, cellular response to ultraviolet C, and negative regulation of cell growth [30]. ST20-MTHS readthrough transcript can be formed through splicing to produce a fusion protein that shares sequence identity from both genes, which are highly expressed in the liver and kidney (https://www.proteinatlas.org) [31]. Overall, our observations indicate that a DNA methylation related mechanism may be relevant to the gene-DASH diet score interaction observed in this region.

The joint analysis of SNP main effects and interaction effects has been shown to be more powerful than the analysis of SNP main effects or interaction effects alone, when the genetic effects are relatively weak and the interaction effects are moderate [18, 32]. Thus, the joint analysis is a promising approach to identify additional loci relevant to traits of interest. In the present study, we compared the joint analysis results with main effect statistics obtained from our study sample and a larger GWAS for systolic BP [25], and we identified several candidate loci for future validation (Table S8). For example, the significant 2df test for an intronic SNP (rs140635454) of NUP93 was driven by its interaction with the DASH diet score. A recent study reported a significant 2df test for rs76976871 at this gene in a sleep-by-gene interaction analysis for high density lipoprotein cholesterol [33]. Although the interaction term was modest in the sleep study [33], our results and theirs combined highlight the importance of conducting a comprehensive joint analysis to facilitate identification of loci that are potentially modified by diet and other lifestyle factors.

There are several limitations in this study that should be discussed. First, the cross-sectional nature of the study does not allow us to account for reverse causation, which might have occurred if individuals with hypertension were advised to change their dietary patterns to help control their blood pressure. The DASH diet score was calculated using different versions of FFQs in the CHARGE cohorts and multiple 24-hour recalls in the UK Biobank. All these dietary assessments tools are based on self-reported dietary intake, which is subjective to both random and systematic bias [34]. In addition, different versions of FFQs have different food lists and different levels of detail, e.g., a 126-item FFQ was used in FHS, while the ARIC Study used a 66-item FFQ; therefore, some DASH diet score components may include a different numbers and types of food items. These combined may partly explain the high heterogeneity regarding gene-DASH diet score interaction across cohorts, and future analyses should be geared towards cohorts using the same dietary assessment tool to capture intake. Most of our study participants (91%) were EUR; our cross-population analysis was therefore mainly driven by the EUR participants. Furthermore, the numerous associations detected in the lower quartile dichotomized DASH score are probably due to misclassification, therefore larger sample size to replicate the threshold analysis is necessary. Further analyses should also explore the magnitude of gene-diet interactions in individuals with high DASH diet adherence and favorable genetics and low DASH diet adherence and adverse genetics. Diet scores generally have stronger effect sizes in association analysis; however, future studies examining specific dietary components are needed to provide more insights into gene-diet interactions in relation to hypertension.

In conclusion, we demonstrated gene-DASH diet interactions in several loci, particularly at the MTHFS locus (15q25.1). In addition, we showed that DNA methylation may be a plausible underlying mechanism linking gene-diet interaction for systolic BP. Compared to large GWAS, the sample size of the present study is modest; therefore, studies with larger sample size and more diverse populations are needed to validate our findings.

Perspectives

Both diet and genetic variants affect BP. In the present study, we examined interactions between the DASH diet score and over nine million SNPs in relation to systolic BP in up to 127,282 individuals from six population groups. We observed significant gene-diet interactions in three loci in European-specific analyses and an additional four loci in cross-population analyses. At 15q25.1 (lead SNP rs117878928; annotated to MTHFS), our analysis suggests that SNPs potentially interacting with the DASH diet may affect systolic BP via a DNA a methylation-related mechanism.

Pathophysiological Novelty and Relevance.

What is new?

By examining gene-DASH diet score interaction, our study identified novel genetic loci affecting BP and provided new insights into the mechanisms linking the DASH diet and BP. At the MTHFS locus (15q25.1), we showed that the interaction between the DASH diet score and genetic variants may affect DNA methylation levels and subsequently change systolic BP.

What is relevant?

Diet is an important lifestyle factor affecting hypertension risk. The DASH diet has been shown to reduce systolic BP by multiple randomized controlled clinical trials. In population studies, by examining the interaction between the DASH diet score and genetic variants, as well as jointly evaluating the diet-gene interaction with genetic main effect (i.e., the 2df test), we identified novel genetic loci affecting systolic BP that have not been reported in previous genome-wide association studies. Diet-gene interaction analysis also provides insights into the underlying mechanisms suggesting how diet affects BP. Here, we demonstrated that the DASH diet score interacts with several genes, including MTHFS, THSD7B, SPATA5, UBE3D, GATA4, and WWOX, to change systolic BP in general populations.

Clinical/Pathophysiological Implications

Our observation regarding the consistent interaction between the DASH diet score and rs117878928 (at the MTHFS locus) across all participating cohorts indicates folate metabolism may have impact on BP regulation. The enrichment of cis-mQTLs in this locus appears to be consistent with the functionality of MTHFS, suggesting a DNA methylation related mechanism may be relevant to the gene-DASH diet score interaction for systolic BP.

Non-Standard Abbreviations

2df

2-degree-of-freedom

AFR

African population

AMR

Admixed American population

BP

blood pressure

CpG

cytosine-guanine dinucleotides

CSA

Central/South Asian population

DASH

Dietary Approach to Stop Hypertension

eQTL

expression quantitative trait loci

EUR

European population

LD

linkage disequilibrium

MAF

minor allele frequency

MID

Middle Eastern population

mQTL

DNA methylation quantitative trait

MTHFS

methenyltetrahydrofolate synthetase

SNP

Single nucleotide polymorphism