DNA Methylation and Age-Independent Cardiovascular Risk, an Epigenome-Wide ApproachHighlights
The REGICOR Study (REgistre GIroní del COR)
Objective—The objectives of this study were to decipher whether age-independent cardiovascular risk is associated with DNA methylation at 5′-cytosine-phosphate-guanine-3′ (CpG) level and to determine whether these differential methylation signatures are associated with the incidence of cardiovascular events.
Approach and Results—We designed a 2-stage, cross-sectional, epigenome-wide association study. Age-independent cardiovascular risk calculation was based on vascular age and on the residuals of the relationship between age and cardiovascular risk. Blood DNA methylomes from 2 independent populations were profiled using the Infinium HumanMethylation450 BeadChip. The discovery stage of these studies was performed in the REGICOR cohort (REgistre GIroní del COR; n=645). Next, we validated the initial findings in the Framingham Offspring Study (n=2542). Eight CpGs located in 4 genes (AHRR, CPT1A, PPIF, and SBNO2) and 3 intergenic regions showed differential methylation in association with age-independent cardiovascular risk (P≤1.17×10−7). These CpGs explained 12.01% to 15.16% of the variability of age-independent cardiovascular risk in REGICOR and 7.51% to 8.53% in Framingham Offspring Study. Four of them were only related to smoking, 3 were related to smoking and body mass index, and 1 to diabetes mellitus, triglycerides levels, and body mass index (P≤7.81×10−4). In addition, we developed methylation risk scores based on these CpGs and observed an association between these scores and cardiovascular disease incidence (hazard ratio=1.32; 95% confidence interval: 1.16–1.51).
Conclusions—Age-independent cardiovascular risk was related to different DNA methylation profiles, with 8 CpGs showing differential methylation patterns. Most of these CpGs were associated with smoking, and 3 of them were also related to body mass index. Risk scores based on these differential methylation patterns were associated with cardiovascular events and could be useful predictive indices.
Coronary heart disease (CHD) is the leading cause of worldwide morbidity and mortality,1,2 and atherosclerosis is its principal mechanism.3 Atherosclerosis development and progression are driven by age4,5 and the presence of vascular risk factors (VRFs), such as smoking, diabetes mellitus, hypertension, and total, high-density lipoprotein, and low-density lipoprotein cholesterol levels.6 The cumulative effect of aging and the presence of these VRFs are considered and summarized in the functions used to determine an individual’s cardiovascular risk (CVR).6
Atherosclerotic vascular burden is also influenced by heritable factors. Epigenetic signatures are heritable traits, but they are also modifiable by age and environmental factors. The best-known epigenetic mechanism regulating gene expression, DNA methylation, is related to atherosclerotic traits.7 Thus, it may be a mediator between aging, environmental factors, VRFs, and atherosclerosis. Because both CVR and DNA methylation are highly determined by age, linking DNA methylation with age-independent CVR might contribute to understand the individual and cumulative effect of the confluence of VRFs on DNA methylation. Two approaches, the difference between vascular8,9 and chronological age (Δage) and the residuals of the correlation between age and CVR, have been used to calculate age-independent CVR.
We hypothesized that specific DNA methylation signatures are influenced by the confluence of VRFs and that, in turn, these signatures are associated with the progression of atherosclerosis and the incidence of cardiovascular events. Our aims were to identify 5′-cytosine-phosphate-guanine-3′ (CpGs) showing differential methylation related to age-independent CVR and to determine the association of these differential methylation signatures with the presence of subclinical atherosclerosis and with the incidence of coronary or cardiovascular events.
Materials and Methods
Materials and Methods are available in the online-only Data Supplement. Briefly, we performed an initial cross-sectional study through a 2-stage epigenome-wide association study (EWAS) strategy, using 2 independent populations. Based on whole-blood differential methylation profiles associated with age-independent CVR, we developed 2 methylation risk scores (MRSs) and assessed their association with subclinical atherosclerosis and with the incidence of clinical coronary and cardiovascular events.
The flowchart showing the study procedure is presented in Figure I in the online-only Data Supplement.
Estimate of the Age-Independent CVR in the Study Populations
We defined 2 approaches to determine age-independent CVR, the residuals and the Δage approach. Two of the 648 individuals with DNA methylation data in the REGICOR study (REgistre GIroní del COR) were excluded after DNA methylation quality control analysis. We also excluded individuals younger and older than the suitable age range (35–79 years) to estimate CVR (n=67), and those individuals with missing values on the VRFs considered in the risk function (n=5). Finally, we estimated the CVR of 574 individuals (residuals approach) and the difference between vascular and chronological age of 465 individuals (Δage approach).
A descriptive analysis of the main sociodemographic and clinical characteristics of the individuals included in the discovery stage of the Δage approach is shown in Table 1; equivalent information for the individuals included in the residual approach is shown in Table I in the online-only Data Supplement.
In the Framingham Offspring Cohort, after the DNA methylation quality control and exclusion of duplicates, we excluded individuals not within the valid age range (n=209) and those with missing values for any VRF used in the risk function (n=161). Therefore, from 2542 individuals remaining after the quality control, we calculated the CVR of 2172 participants and the difference between vascular and chronological age of 1823 participants. The main sociodemographic and clinical characteristics of these individuals are shown in Tables II and III in the online-only Data Supplement.
Discovery Stage of the EWAS
Because we hypothesized that methylation is influenced by the convergence of VRFs other than age, we studied their association through an epigenome-wide strategy. We identified 29 differentially methylated CpGs in association with age-independent CVR using the residuals approach and 24 using the Δage approach, defining an arbitrary P value threshold <1×10−5. Of these, 48 were unique differentially methylated CpGs in association with age-independent CVR (Table IVA and IVA in the online-only Data Supplement). These CpGs were located in 32 genes and 16 intergenic regions. The Manhattan and QQ plots are shown in the Figure.
Validation Stage and Meta-Analysis
We assessed the association between the age-independent CVR and the initially identified 48 CpGs in the Framingham population. Seven CpGs (cg00574958, cg05575921, cg05951221, cg12057156, cg12547807, cg18608055, and cg27537125) were significant (P<1.04×10−3; 0.05/48) in both approaches, and 1 additional (cg19939077) was significant only in the residuals approach (Table VA and VB in the online-only Data Supplement).
By meta-analyzing the results observed in the REGICOR and the Framingham studies, we validated 8 of the 48 initially discovered CpGs in 7 differential loci (Table 2), using a P value threshold <1.17×10−7 that was defined by the Bonferroni correction for multiple comparisons (0.05/427 948). Two of these CpGs were located in the same locus (cg05951221 and cg21566642), therefore 7 independent loci were identified. The model adjusted for age, sex, cell type, and surrogate variables was considered as the main analysis.
Four of those CpGs were validated in both approaches used to estimate the age-independent CVR (cg05575921, located within AHRR; cg18608055, within SBNO2; cg27537125, in an intergenic region close to a noncoding RNA [MIR4425]; and cg12547807, in another intergenic region). Apart from this CpG site, the residuals approach provided 3 CpGs (cg05951221 and cg21566642, located in the same intergenic region close to the gene ALPP2 and the long noncoding RNA ACO68134.6, and cg19939077, within PPIF). The Δage approach contributed one additional CpG (cg00574958, located within CPT1A).
Seven of the 8 identified CpGs were found to be hypomethylated in association with smoking status (cg05575921, cg05951221, cg12547807, cg18608055, cg19939077, cg21566642, and cg27537125; Table 3; Table VI in the online-only Data Supplement). Three of them (cg12547807, cg18608055, cg19939077) were also hypomethylated in association with body mass index (BMI). The eighth CpG (cg00574958) was hypomethylated in association with diabetes mellitus, BMI, and triglycerides levels.
The identified CpGs explained 15.16% of the variability of the age-independent CVR estimated through the residuals approach and 12.01% of the variability of the CVR assessed through the Δage approach in REGICOR (excluding cg21566642 to avoid redundant information already provided by cg00574958). In the Framingham population, these 8 CpGs explained 7.51% and 8.53% of the variability of the CVR from the residuals and the Δage approach, respectively.
We analyzed whether genetic variants located close to the validated CpGs were associated with CHD. Data on CHD have been contributed by CARDIoGRAMplusC4D investigators and were downloaded from www.cardiogramplusc4d.org. We did not find any cardiovascular disease (CVD)–associated SNP (single nucleotide polymorphism) located 10 000 bp downstream or upstream of the identified CpGs.
Development of MRSs and Evaluation of Their Association With Subclinical Atherosclerosis and Clinical Events
We developed MRSs based on the identified CpGs, one for each population used in the residuals and Δage studies (REGICOR, n=574 and n=465, and Framingham, n=2172 and n=1823, respectively). We used 7 CpGs (excluding cg21566642 to avoid redundant information).
Association of MRSs With Subclinical Atherosclerosis
The association between the MRSs and arterial stiffness measurements in the REGICOR population is shown in Table 4A. Measurements related to arterial stiffness were available for 534 individuals included in the EWASs. Both MRSs were directly associated with arterial distensibility coefficient and inversely with pulse wave velocity. In model 2, the association with those parameters was only found in the left common carotid artery.
Association of MRSs With CHD and CVD Incidence
The associations between the MRSs and the incidence of clinical coronary (n=94) and cardiovascular (n=222) events in the Framingham population are shown in Table 4B. The median and interquartile range of the follow-up periods for CVD and CHD incidence were 7.66 (5.44, 9.89) and 7.87 (6.47, 9.26) years, respectively. Both MRSs were associated with higher CVR independently of the classical VRFs considered in the CVR function. The hazard ratios for CVD events per 1-SD increase were similar in all the models of adjustment.
Here, we identify 7 loci (8 CpGs) showing differential DNA methylation associated with age-independent CVR, located in 4 genes (AHRR, CPT1A, PPIF, and SBNO2) and 3 intergenic regions. All but one of these CpGs were associated with smoking, and 3 of them were also associated with BMI (2 not previously reported). The eighth CpG was associated with BMI, diabetes mellitus, and triglycerides. In addition, epigenetic risk scores based on the differentially methylated CpGs identified were associated with the incidence of CVD events.
A growing number of studies published in the past decade support the role of DNA methylation in atherosclerosis and CVDs.7 The development of epigenome-wide techniques has allowed the identification of differentially methylated patterns across the genome that might not have been found through a candidate-gene methylation strategy. The published studies have analyzed the association between DNA methylation and atherosclerosis, mainly focused on either atherosclerotic and cardiovascular events or individual VRFs, but none of them have assessed the cumulative effect of several VRFs. To determine this cumulative effect, we estimated CVR using a validated risk function. Moreover, because age is a strong determinant of CVR,4,5 we estimated age-independent CVR by calculating the difference between vascular and chronological age and the residuals of the association between CVR and age.
Seven of the 8 validated CpGs were associated with smoking, and 3 of them were additionally associated with BMI (2 of them newly discovered: cg12547807 and cg19939077). Methylation at cg12547807, located in an intergenic region of chromosome 1, has been previously identified as related to smoking10 as have both cg19939077 and cg18608055 (false discovery rate <0.05).10 cg19939077 was additionally described as differentially methylated in association with alcoholism11,12; it is located within PPIF, which encodes a protein of the peptidyl-prolyl cis-trans-isomerases, which have an important role in several CVDs.13
cg18608055 is located in SBNO2, which encodes a component of the pathways leading to the anti-inflammatory effect of IL-10 (interleukin-10).14 Previously identified as differentially methylated in association with BMI15,16 and serum C-reactive protein,17 it has also been reported to be associated with the cardiovascular biomarker GDF-15 (growth differentiation factor 15).18 The same study found it to be hypomethylated in individuals who had experienced a myocardial infarction (ncases=47, ncontrols=271) although this association was not statistically significant, probably because of the small population size. Similarly, our results suggest that hypomethylation of the cg18608055 is associated with an increased age-independent CVR.
Methylation status at cg05575921 has been associated with smoking,10,19 obesity,20 and triglycerides,21 but we found no relationship to BMI or lipid levels. This CpG is located within AHRR, the smoking-associated gene that is most epigenetically regulated.22 Its transcription factor is involved in the metabolism of toxins from cigarette smoke.23 cg27537125, cg05951221, and cg21566642 are also identified in association with smoking.10,19 The last 2 are located within an intergenic CpG island of chromosome 2 (chr2:233,283,397-233,285,959). This CpG island should be further analyzed in future studies assessing the association between age-independent CVR and epigenomic regions instead of sites.
Differential methylation at cg00574958 has been associated with metabolic syndrome,24 diabetes mellitus,25,26 BMI,16,27 and lipid levels,19,21 in agreement with our findings. It is located within intron 1 of CPT1A, which encodes a key enzyme in the regulation of mitochondrial fatty acid oxidation.28
We hypothesized that VRFs may lead to vascular dysfunction, and thus epigenetic risk scores will be inversely associated with arterial distensibility and positively associated with pulse wave velocity. Remarkably, the associations between arterial wall properties and MRSs were contrary to that hypothesis. Engelen et al29 previously reported a positive association between carotid distensibility and smoking in a healthy population (n = 3601), including data from 24 centers worldwide. In our study, smoking was related to 6 of the 7 CpGs used in the development of the MRSs, which could explain the unexpected result.
We did observe an association between the age-independent CVR epigenetic risk scores and the incidence of clinical CVD, supporting the consistency of our approach and our results. These associations remained significant when adjusting for classical VRFs considered in the FRESCO (Función de Riesgo ESpañola de acontecimientos Coronarios y Otros, or Spanish Risk Function of Coronary and Other Cardiovascular Events) risk function (ie, age, sex, total and high-density lipoprotein cholesterol, diabetes mellitus, smoking status, systolic blood pressure, and hypertensive treatment). They could be the consequence of a synergistic effect of classical VRFs on DNA methylation. Also, because 6 of the CpGs used to build the MRS were related to smoking, it could be that the observed associations reflect the smoking status more precisely than the applied questionnaires, which consider it as a dichotomous instead of a continuous variable.
Finally, our rationale was that differential DNA methylation was the result of exposure to VRFs accumulated as age-independent CVR. It could act as a mediator, but not a causal determinant, of the association between CVR and CVD. Although we cannot discard the causal association between DNA methylation and CVD, we report that genetic variability in the identified loci was not associated with CHD and calls into question any causal role.
The main strength of this study is the dual-approach design to identify the age-independent CVR. Both methods presented consistent and similar results in all the associations. Regarding the EWASs, we used powerful statistics by fitting models based on robust linear regression, adjusted for confounding variables and further nonmeasured confounding (surrogate variables). Not only did we use methylation data processed according to standardized protocols but also we validated our first findings in a large independent population. Last, we showed the potential clinical importance of our results by applying them to create age-independent CVR epigenetic risk scores and then evaluated them as predictors of clinical cardiovascular events.
Some limitations of our study should be noted. First, our results are based on European origin populations and cannot be extrapolated to other ethnic groups. In addition, we could not adjust our analysis for genetic variability between the 2 populations of the study because we do not have genome-wide genetic data in the REGICOR population. Second, because of the cross-sectional study design and the potential bidirectional relationship between CVR and DNA methylation, we cannot infer causality of the association between these 2 variables. Finally, EWASs have several limitations, such as the type of sample used (peripheral blood circulating cells), the population size, or the data analysis and interpretation.
In summary, we identified 7 loci showing differential methylation in relation to age-independent CVR. Six of these loci were associated with smoking. Moreover, 3 of them were associated with BMI, 2 of them for the first time (to our knowledge). We also showed that age-independent CVR epigenetic risk scores based on differential methylation patterns are related to the incidence of clinical cardiovascular events. They may also have potential value as research tools to unravel atherosclerosis mechanisms.
We thank Elaine M. Lilly, PhD, for her critical reading and revision of the English text.
Sources of Funding
This project was funded by the Carlos III Health Institute–European Regional Development Fund (FIS PI15/00051, CIBER Cardiovascular Diseases, CIBER Epidemiology and Public Health) and the Government of Catalonia through the Agency for Management of University and Research Grants (2014SGR240). S. Sayols-Baixeras was funded by the Instituto de Salud Carlos III-Fondos FEDER (IFI14/00007). A. Fernández-Sanlés was funded by the Spanish Ministry of Economy and Competitivity (BES-2014–069718). The Framingham Heart Study is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with Boston University (Contract No. N01-HC-25195 and HHSN268201500001I). This manuscript was not prepared in collaboration with investigators of the Framingham Heart Study and does not necessarily reflect the opinions or views of the Framingham Heart Study, Boston University, or NHLBI. Additional funding for SABRe was provided by Division of Intramural Research, NHLBI, and Center for Population Studies, NHLBI.
The online-only Data Supplement is available with this article at http://atvb.ahajournals.org/lookup/suppl/doi:10.1161/ATVBAHA.117.310340/-/DC1.
- Nonstandard Abbreviations and Acronyms
- coronary heart disease
- cardiovascular disease
- cardiovascular risk
- epigenome-wide association study
- methylation risk score
- REgistre GIroní del COR
- vascular risk factor
- Received October 5, 2017.
- Accepted December 14, 2017.
- © 2018 American Heart Association, Inc.
- 1.↵GBD 2016 DALYs and HALE Collaborators. Global, regional, and national disability-adjusted life-years (DALYs) for 333 diseases and injuries and healthy life expectancy (HALE) for 195 countries and territories, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet. 2017;390:1260–1344.
- 2.↵WHO. Global Status Report on Noncommunicable Diseases 2014. http://www.who.int/nmh/publications/ncd-status-report-2014/en/. Accessed October 30, 2017.
- Goff DC Jr.,
- Lloyd-Jones DM,
- Bennett G,
- et al
- Fernández-Sanlés A,
- Sayols-Baixeras S,
- Subirana I,
- Degano IR,
- Elosua R
- D’Agostino RB Sr.,
- Vasan RS,
- Pencina MJ,
- Wolf PA,
- Cobain M,
- Massaro JM,
- Kannel WB
- Joehanes R,
- Just AC,
- Marioni RE,
- et al
- Liu C,
- Marioni RE,
- Hedman ÅK,
- et al
- Philibert RA,
- Penaluna B,
- White T,
- Shires S,
- Gunter T,
- Liesveld J,
- Erwin C,
- Hollenbeck N,
- Osborn T
- El Kasmi KC,
- Smith AM,
- Williams L,
- Neale G,
- Panopoulos AD,
- Panopolous A,
- Watowich SS,
- Häcker H,
- Foxwell BM,
- Murray PJ
- Ronn T,
- Volkov P,
- Gillberg L,
- et al
- Ligthart S,
- Marzi C,
- Aslibekyan S,
- et al
- Dekkers KF,
- van Iterson M,
- Slieker RC,
- et al
- Das M,
- Sha J,
- Hidalgo B,
- et al
- Aslibekyan S,
- Demerath EW,
- Mendelson M,
- et al
- Casals N,
- Zammit V,
- Herrero L,
- Fadó R,
- Rodríguez-Rodríguez R,
- Serra D
- Engelen L,
- Bossuyt J,
- Ferreira I,
- van Bortel LM,
- Reesink KD,
- Segers P,
- Stehouwer CD,
- Laurent S,
- Boutouyrie P
We identified 8 differentially methylated 5′-cytosine-phosphate-guanine-3′ in association with age-independent cardiovascular risk.
Seven of those 5′-cytosine-phosphate-guanine-3′ were related to smoking and 3 of them also with body mass index (2 of them newly associated with body mass index). The eighth 5′-cytosine-phosphate-guanine-3′ was also associated with body mass index and diabetes mellitus and triglycerides.
Risk scores based on these differential methylation patterns were related to clinical cardiovascular events. This association was independent of age, sex, and classical vascular risk factors.