Epigenetic regulation of PAR4-related platelet activation: mechanistic links between environmental exposure and cardiovascular disease

Protease-activated receptor 4 (PAR4) is a potent thrombin receptor. Epigenetic control of the F2RL3 locus (which encodes for PAR4) via DNA methylation is associated with both smoking and cardiovascular disease. We examined the association between DNA hypomethylation at F2RL3 and risk of cardiovascular disease, focusing on acute myocardial infarction (AMI) (n=853 cases / 2,352 controls). We used in vitro cell models to dissect the role of DNA methylation in regulating expression of F2RL3. We investigated the interplay between F2RL3 DNA methylation and platelet function in human (n=41). Lastly, we used Mendelian randomization to unify observational and functional work by assessing evidence for causal relationships using data from UK Biobank (n=407,141) and CARDIoGRAMplusC4D (n=184,305). Observationally, one standard deviation (SD) decrease in DNA methylation at F2RL3 was associated with a 25% increase in the odds of AMI. In vitro, short-term exposure of cells to cigarette smoke reduced F2RL3 DNA methylation and increased gene expression. Transcriptional assays flagged a role for a CEBP recognition sequence in modulating the enhancer activity of F2RL3 exon 2. Lower DNA methylation at F2RL3 was associated with increased platelet reactivity in human. The estimated casual odds ratio of ischaemic heart disease was 1.03 (95% CI: 1.00, 1.07) per 1 SD decrease in F2RL3 DNA. In conclusion, we show that DNA methylation-dependent platelet activation is part of a complex system of features contributing to cardiovascular health. Tailoring therapeutic intervention to new knowledge of F2RL3/PAR4 function should be explored to ameliorate the detrimental effects of this risk factor on cardiovascular health. One sentence summary DNA methylation-dependent platelet activation is a likely causal contributor to cardiovascular health.


Introduction
The mechanisms behind the adverse effects of smoking on cardiovascular health remain incompletely understood, but over the last decade increasing evidence suggests that epigenetic modifications might link environmental exposure and pathology. The use of arraybased DNA methylation detection technologies has revealed smoking-related differential DNA methylation patterns in DNA extracted from peripheral blood 1-5 . Specifically, DNA methylation at F2RL3 appears to show a dose-response relationship with smoking and lower DNA methylation at F2RL3 has been associated with mortality from all causes, cardiovascular disease (CVD) and cancer 6,7 . Little evidence currently exists as to the causal impact of these observational effects nor the potential mechanisms or therapeutic implications of this route to disease.
F2RL3 codes for protease-activated receptor 4 (PAR4), a G-protein coupled receptor (GPCR) expressed on the surface of a number of cell types including platelets 8 . Together with protease-activated receptor 1 (PAR-1), PAR4 activates platelets in response to thrombin generated at the site of tissue injury. A small number of missense coding variants in F2RL3 that alter platelet aggregation and function have been described, providing a link between PAR4 and the heritable inter-individual variation in platelet reactivity [9][10][11] . Little is known about the functional consequences of epigenetic modifications at the F2RL3 locus and how regulatory shifts in the complex events controlling platelet function may be manifest in realised health outcomes. We aimed to triangulate evidence from multiple sources in order to not only test hypothetical causal relationships between smoking and methylation related regulation, but also to flag possible targets for therapeutic intervention.

Methods
We used human data, in vitro studies and human testing to investigate the functional consequences of differential DNA methylation at F2RL3. Our methods are described in detail in the Materials and Methods section in the Supplementary Appendix (under the same subheadings as those used below) and an overview of the different components of the study can be found in Fig. 1A.

(i) F2RL3 Epidemiology
Using individual participant data in an observational framework, we explored the relationship between smoking, DNA methylation and acute myocardial infarction (AMI). DNA methylation was measured at four CpG sites (CpG_1-4) in exon 2 of F2RL3 ( Fig. 1B) in 853 AMI cases and 2,352 controls from the Copenhagen City Heart Study (Table S1) using pyrosequencing.
The CpG_3 at position 16,889,774-5 corresponds to the CpG labelled cg03636183 on the Illumina Infinium BeadChip (27k and 450k). Associations between DNA methylation and selfreported smoking behaviour, incidence of AMI and AMI mortality (within AMI cases) were assessed using linear, logistic and Cox regression, respectively, and adjusted for age, sex and smoking status as appropriate. We also improve the adjustment for smoking exposure in these analyses by using a second smoking related DNA methylation site, the aryl hydrocarbon receptor repressor (AHRR), as a more refined measure of long-term exposure to cigarette smoke.

(ii) F2RL3 DNA methylation in a cell model
Two cell types pertinent to CVD were used to evaluate the effect of cigarette smoke on F2RL3 DNA methylation and mRNA expression. Firstly, in line with previously published work using this model 12,13 , human coronary artery endothelial cells (HCAEC) were exposed to three doses of cigarette smoke extract (CSE) then F2RL3 DNA methylation and mRNA expression measured 13 . In HCAEC not exposed to CSE, we investigated the impact of global DNA demethylation on F2RL3 mRNA expression by culture with 5-Azacytidine 14 . In a second set of experiments designed to assess the effect of CSE on a human hematopoietic cell lineage (precursors to platelets) an acute megakaryocytic leukemia cell line (CMK) was used 15 . CMK cells were exposed to four doses of CSE over the course of four days, with F2RL3 DNA methylation and mRNA expression measured on day five. Expression was also measured in the endogenous control, ribosomal protein lateral stalk subunit P0 (RPLP0).

(iii) Functional regulation of F2RL3
We used a pGL3 reporter vector to test for the presence of an enhancer within a fragment of F2RL3 exon 2 containing CpG_1 to CpG_4. The potential mechanisms of effects on F2RL3 expression were explored by transfecting HEK-293 cells with reporter constructs containing different fragments of F2RL3 to drive expression of luciferase. We began with an expression model in HEK-293 cells with a promoter-less pCpGL reporter vector and in order to set a baseline comparator, the F2RL3 putative promotor sequence was inserted immediately upstream of the transcription start site adjacent to the luciferase cDNA in pCpGL (pCpGL_F2RL3pro). Subsequently, the F2RL3 exon 2 fragment only was inserted into the pCPGL vector (pCpGL_exon2) and then both the F2RL3 promoter region and the exon 2 fragment (pCpGL_F2RL3pro_exon2). To investigate whether the CCAAT/enhancer binding protein (CEBP) recognition sequence in F2RL3 exon 2 (see Fig. 1B) was functional and involved in DNA methylation-dependent regulation of F2RL3, we assessed luciferase activity again having deleted the CEBP recognition sequence (pCpGL_F2RL3pro_exon2 CCAAT deletion). Finally, the HCAEC model was revisited in order to examine the impact of differential DNA methylation on DNA-protein interactions at the locus using chromatin immunoprecipitation (ChIP) in cells cultured with 5-Azacytidine.

(iv) Differential platelet function in a human experiment
Forty-one never-smoking volunteers (aged 22-24 years) were recruited from the Accessible Resource for Integrated Epigenomic Studies (ARIES) substudy 16 of the Avon Longitudinal Study of Parents and Children (ALSPAC) 17 based on having high or low methylation at the F2RL3 CpG site cg03636183. Recruited individuals provided fresh blood samples that were immediately analysed for platelet reactivity, as assessed by stimulating platelet-rich plasma (PRP) with different concentrations of AYPGKF peptide, a specific agonist of PAR4. Platelet responses were measured using flow cytometry to detect the open conformation of the platelet aIIbb3 integrin and platelet surface exposure of P-selectin, both markers of platelet activation during haemostasis. To assess the specificity of any differences observed, the same measurements were made after PRP was treated with SFLLRN, a PAR1 specific agonist. DNA was extracted and targeted pyrosequencing of F2RL3 carried out to capture the same four positions described previously (Fig. 1B).
Previous studies have shown that the single nucleotide polymorphism (SNP), rs773902, located at 16,889,821 bp on chromosome 19 (Fig. 1B) is associated with platelet function 9,11 .
Using existing genetic data in the ALSPAC cohort 17 , we explored the potential impact of this variant both on methylation and platelet reactivity. Methods and results relating to this genetic analysis are described in the Supplementary Appendix (under the subheading 'Differential platelet function in a human experiment' in both the Materials and Methods and Additional results sections).

(v) Mendelian randomization
A two-step epigenetic Mendelian randomization (MR) strategy 18,19 was used to estimate the causal relationship between smoking, DNA methylation and CVD outcomes (Fig. S1). In the first step, a genetic variant in the CHRNA5-A3-B4 gene cluster, rs1051730, was used as an instrument for smoking intensity in a one-sample MR design using individual level data from the Copenhagen City Heart Study. Causality was assessed through association of the variant with methylation at CpG_3 in groups stratified by smoking status 20 . In the second step, we carried out a two-sample MR analysis 21,22 using genetic variants reliably associated with F2RL3 DNA methylation at CpG_3 in results from the GoDMC Consortium (N=27,750, unpublished) and their effect estimates from association analyses for CVD outcomes in UK Biobank 23,24 (n=407,141, using individual level data to calculate associations) and CARDIoGRAMplusC4D consortium 25 (max. n=184,305, using publicly available genomewide association study summary statistics accessed via MRBase 26 ). In UK Biobank, where individual level data was available, three disease outcomes were defined from the most general definition of CVD through ischaemic heart disease (IHD) to the most specific definition of AMI.
Approximately equivalent disease definitions were available in CARDIoGRAMplusC4D (myocardial infarction for AMI; coronary heart disease (CHD) for IHD). Sensitivity analyses were performed in UK Biobank to investigate the potential impact of survivorship bias, confounding due to population stratification and smoking on our estimates.

(i) F2RL3 Epidemiology
Observationally, smoking was associated with lower DNA methylation of F2RL3 across all four CpG sites with the strongest association being seen at CpG_1 (Table S2). The percentage DNA methylation at all four sites was highly correlated in current smokers (pairwise correlations all r>0.77) and former smokers (r between 0.58 and 0.76) and was moderately correlated in those who had never smoked (r between 0.28 and 0.56) (Fig. S2).
Amongst smokers, there appeared to be a dose-response relationship such that DNA methylation was lowest in heavier smokers (Table S2).
In a sex-and age-adjusted model, the estimated odds of subsequent AMI observationally was

(iv) Differential platelet function in a human experiment
ALSPAC participants (see Tables S5 and S6 for descriptive statistics) recalled based on high (retrospective) DNA methylation (n=22) had on average higher contemporary DNA methylation values than the group recalled based on low DNA methylation (n=19) (Table S7; 4A). The largest difference in DNA methylation (4.8%) was observed at CpG_1 and the correlation across positions ranged from 0.41 to 0.80 (Table S8). The difference in methylation between the groups is small relative to the difference observed between smokers and non-smokers (Fig. S6). The allele frequency at rs773902 in those invited and recruited to this study is shown in Supplementary Tables S5, S6 and S11.
Comparison of dose response curves revealed lower mean half maximum effective concentration (EC50) values for both integrin activation (EC50, p=0.001) and P-selectin exposure (EC50, p=0.046) in the low DNA methylation group (Table S9; Fig. 4B & C). These results correspond to an increase in responsiveness with lower DNA methylation. For example, at 75µM AYPGKF, the response in high methylation status individuals was 47.6% compared to 68.6% in the low methylation status individuals, for integrin activation. For Pselectin exposure, the equivalent responses were 21.1% versus 35.6%. These differences could not be explained either by differences in hematological measures (Table S10) or by other measured confounders (Table S11). No between-group differences were observed after the stimulation of platelets via a PAR1-specific agonist (Table S9; Fig. 4D & E).
Furthermore, we found no evidence of a between-group difference in the expression of the individual components of integrin (CD41 and CD61) in basal (non-stimulated) samples (Table   S9). Accounting for DNA methylation group (high/low), linear regression of integrin activation EC50 on DNA methylation at CpG_1 gave evidence of a 1.10 µM (95% CI: -0.26,2.47) decrease in EC50 per unit (%) decrease in DNA methylation (Fig. 4F).

(v) Mendelian randomization
In the Copenhagen City Heart Study, each copy of the minor allele of rs1051730, which is associated with an 0.83 increase in the number of cigarettes smoked per day of (95% CI: 0.24, 1.43) amongst current smokers (Table S12), was associated with decreases in F2RL3 DNA methylation at CpG_3 of 1.10% (95% CI: -1.93, -0.28) in current smokers and 0.52% (95% CI -1.23, 0.19) in former smokers (Table S13) (we present results from CpG_3, which is the only F2RL3 site available on the Illumina 450k array (corresponding to cg03636183)).
In never smokers, there was no clear evidence that rs1051730 was associated with F2RL3 DNA methylation (beta per minor allele: -0.08%, 95% CI: -0.78, 0.63). Whilst this apparent lack of association of rs1051730 with methylation in never smokers supports the use of this SNP as an instrument for smoking intensity, power was limited to test for differences between smoking groups (p-value for heterogeneity=0.18). A causal effect estimate was not calculated because the instrument used is not considered to be a good proxy for lifetime exposure to tobacco and therefore, such estimates would likely be biased 20 .
In UK Biobank, causal estimates for the effect of F2RL3 DNA methylation on CHD/IHD disease risk were generated, with the odds of disease being 1.04 (95% CI: 1.00, 1.08) given a one SD decrease in methylation (Fig. 5). The same estimate generated using data from CARDIoGRAMplusC4D for the equivalent phenotype of coronary heart disease yielded an OR of 1.02 (95% CI: 0.95, 1.09) and the meta-analysed estimate from the two datasets was 1.03 (95% CI: 1.00, 1.07). An analysis in UK Biobank of incident fatal events generated imprecise estimates (Fig. S9). When repeated in a subset of UK Biobank participants designated as 'white British', effect estimates are consistent (IHD: OR 1.05 (95% CI: 1.00,1.10)) ( Fig. S10). Additional analysis performed in UK Biobank provided no clear evidence of a difference in estimates for ever smokers versus never smokers (Fig. S11).
There appeared to be no strong evidence for a causal effect of F2RL3 methylation on the broadest categorisation of CVD in UK Biobank which includes non-thrombotic events (OR 0.99 (95% CI: 0.97, 1.02) (Fig. 5). Results for analyses using all SNPs separately (including rs773902) and for sensitivity analyses can be found in Supplementary Appendix (Fig. S8 -S13) for both UK Biobank and CARDIoGRAMplusC4D.

Discussion
Until now, little direct and causal evidence has been available linking environmental exposure, epigenetic regulation and health outcomes. We have been able to unify evidence from multiple independent sources to address this and to specifically target platelet function as a modifiable aetiological component of smoking-related cardiovascular risk.
Observationally, we were able to show that one SD decrease in DNA methylation at F2RL3 was associated with a 25% increase in the odds of AMI. In vitro, short-term exposure of cells to cigarette smoke yielded reductions in F2RL3 DNA methylation and increased gene expression. With this, transcriptional assays flagged a role for a CEBP recognition sequence in modulating the enhancer activity of F2RL3 exon 2 and in human, lower DNA methylation at F2RL3 was associated with increased platelet reactivity.
Cell-based modelling in two CVD-relevant cell types gave evidence of both differential DNA methylation and expression at F2RL3 because of exposure to aqueous CSE. Functional analysis of the expression of PAR4 identified a CEBP binding site in exon 2 of F2RL3 as important and occupancy of this site by CEBP-β was shown by ChIP to increase in response to global demethylation. CEBP binding to an identical CCAAT recognition site at a different locus (MLH1), is known to be reduced by DNA methylation of a CpG residue in an identical relative position to the CCCAT recognition sequence to that observed with CpG_1 28 .
Together, these findings suggest that F2RL3 expression could be in part constrained by constitutive DNA methylation of CpG sites within an exon 2 enhancer, with methylation reducing CEBP-β occupancy and enhancer activity -smoking disturbs this regulation, reducing methylation and increasing CEBP-β occupancy and F2RL3 expression. Evidence in support of this hypothesis comes from a recent transcriptome-wide association study that revealed an association between F2RL3 expression assessed in lymphoblastoid cell lines (LCLs) and smoking (n=92 current versus n=364 never smokers) 29 . However, no relationship between F2RL3 expression and mortality was observed 29 and the same association was not seen in a similar study based on whole blood gene expression (n=1,421 current versus n=4,860 never smokers) 30 .
Recruitment and fresh blood sample collection in participants selected based on DNA methylation at F2RL3 was able not only to recapitulate a DNA methylation gradient independent of smoking, but to also show that this was associated with platelet activation.
Although the increase in platelet responsiveness may be directly related to increased PAR4 expression, leading to a leftward shift in a concentration response curve, mechanistically it is possible that this is a result of a change in heterodimer arrangement for PAR4 with other associated GPCRs. It is known, for example, that PAR4 heterodimerises with PAR1 and with P2Y12, and an alteration in the stoichiometry of the association may lead to an altered responsiveness to PAR4 agonism 31,32 . It will therefore be important to determine whether this underlies part of the increased responsiveness when changes in methylation of F2RL3 gene occur, either naturally or induced by smoking.
Observational evidence was complemented by an MR analysis designed to interrogate the extent to which this relationship is likely to be causal. We were able to estimate the likely causal effect of differential DNA methylation at x 10 -05 ) 37 whilst other nearby variants acted as eQTLs in other tissues (Table S17A) and in blood 38 (Table S17B).
We have begun to characterise the pathway from reduced F2RL3 DNA methylation to increased risk of cardiovascular events but further work is needed to fully understand the mechanisms behind DNA methylation related risk at this locus. It has been proposed that activated platelets have key thromboinflammatory activities with platelets being able to both respond and contribute to inflammatory signals 39,40 . Therefore, the increased activation of platelets that we see in response to decreased DNA methylation may be influenced by or have downstream effects on inflammation 41 . Given that an important part of the development of risk may be associated with increased responsiveness of PAR4 to thrombin, development of PAR4 antagonists as novel antithrombotics is likely to be an attractive intervention 32 .
Indeed this is now being realised through the development of several new drug compounds that enable highly specific inhibition of the platelet PAR4 receptor and which potentially have The translational implications of this research are heightened by our growing understanding of the temporal nature of DNA methylation marks. It has been suggested that it can take many years from smoking cessation for DNA methylation levels to return to the level of never smokers, and that the rate of change is CpG site-specific 2,53,54 . In our own collections, it is evident that smoking-induced DNA hypomethylation of the F2RL3 CpG sites persists for decades after tobacco smoking cessation (Fig. 6). This observation is consistent with the elevated mortality risk that may be seen for ex-cigarette smokers, even after they have given up smoking for many years 55,56 and has obvious implications for the design of therapy.
In summary, the observation that variation in DNA methylation at      Results shown for UK Biobank and CARDIoGRAMplusC4D using a multi-SNP instrument. Where equivalent outcomes were available in both datasets, results were meta-analysed (acute myocardial infarction in UK Biobank matched with myocardial infarction (MI) in CARDIoGRAMplusC4D; ischaemic heart disease matched with coronary heart disease (CHD) in CARDIoGRAMplusC4D). Effect estimates represent the OR (95% CI) for each outcome per one standard deviation unit decrease in DNA methylation at F2RL3 (CpG_3 / cg03636183). Heterogeneity (Cochran's Q Statistic) across SNPs: p>0.35 for all outcomes in UK Biobank; p=0.045 for MI in CARDIoGRAMplusC4D; p=0.094 for CHD in CARDIoGRAMplusC4D.