To characterize the nature of the relationship between epigenetic status and diet and lifestyle for TG, we first identified DMSs associated with TG by conducting an EWAS, then examined the relationship between those TG-associated DMSs and diet and lifestyle habits over a period of ~ 13 years. While there was a trend for more factors associated with TG-epigenetic marks in the last exam than in the earlier exams, several dietary factors showed a consistent correlation with epigenetic marks over all four exams. The most impactful dietary and lifestyle factors include alcohol and carbohydrate intake, total sugar, smoking, vitamins B1 and B2, dairy desserts, calcium, saturated fat, total fat, vitamin D, protein, and sweet baked foods (Table S3).
TG is a causal risk factor for CVD (13), in addition to LDL-C. EWAS identified 19 independent DMSs, which accounted for a substantial amount of total TG variation (15%). Over four exams, we observed many associations between the 19 TG-associated DMSs and diet and lifestyle factors, representing 102 of these factors. The strongest and most consistent associations are alcohol and carbohydrate intake, representing 11 of 19 DMSs. Alcohol intake accounts for 13.3% of cg06690548 methylation variation at SLC7A11.
Although high alcohol intake (1–2 drinks/day) was associated with increased TG (24, 25), other studies have indicated that alcohol intake is associated with increased HDL and decreased TG, and increased risk of hypertension, coronary heart disease, and myocardial infarction (26). Lifetime average consumption of alcohol is positively associated with accelerated biological aging, as estimated by GrimAge (27). Our study found that 13 of 19 DMSs were associated with alcohol intake. From mediation analysis, the results further support that the effects of alcohol intake increased TG via differential DNA methylation of seven DMSs at PHGDH, TXNIP, SLC7A11, GARS, SLC43A1, SREBF1, and SLC1A5. Different types of alcoholic drinks, notably (beer, red wine, white wine, and liquor,) all showed consistent mediated effects on TG through CpG methylation at SLC7A11. Although the amount of total alcohol intake decreased from exam 5 to exam 8 (28), all 11 DMSs exhibited a solid and consistent association with alcohol intake across the four exams (Table 3). Alcohol could have a cumulative effect on DMSs from exam 5 to exam 8 (Table 3), but this remains to be unequivocally illustrated. The high consumption of alcohol affecting risk of CVD, myocardial infarction, and aging could be confounded by unhealthy lifestyle choices such as smoking. Nevertheless, our results suggest that alcohol and carbohydrate intakes, and smoking are the most critical lifestyle factors acting epigenetically to modulate TG.
Alcohol is a more energy dense nutrient than carbohydrate. In this study, our results indicated that alcohol intake and carbohydrate intake exhibited opposite effects on TG through epigenetic mechanisms. Alcohol intake was strongly associated with nine DMSs across four exams. Mediation analysis implied that alcohol intake was associated with increased TG through seven DMSs in seven gene regions (SARS, PHGDH, TXNIP, SLC7A11, GARS, SLC43A1, CPT1A, SREBF1, SLC1A5) as mediators. On the contrary, carbohydrate intake was strongly correlated with six of the same DMSs (excluding PHGDH), but in the opposite direction (Table 3). Mediation results support that carbohydrate shows negative effects on (decreased) TG through two DMSs (cg00574958 and cg06690548 as mediators. In a prior study using data from two cohorts, we demonstrated with mediation analysis that carbohydrate intake induces CPT1A methylation at cg00574958, and observed negative indirect effects on (decreased) BMI, glucose, hypertension, TG, type 2 diabetes, and metabolic syndrome, and this then reduces the risk of metabolic diseases (11). In addition, that research observed that CPT1A mRNA expression was negatively associated with carbohydrate intake.
From a mechanistic perspective, male C57BL/6J mice fed an ethanol-containing diet exhibited higher levels of liver TG, indicating hepatic steatosis and, interestingly, altered diurnal oscillations of core clock genes in the liver but not in the suprachiasmatic nucleus, compared to control mice (29). These chrono-disruptions in the liver propagated to specific clock-controlled genes and several metabolic genes, including Cpt1a (29). The prominent findings in the current analysis are the consistent associations between TG-associated DMSs and alcohol, with alcohol acting as the mediator to affect TG. Many of those same DMSs have been observed as associated with alcohol and diseases consequential to heavy drinking. For example, a recently published EWAS identified the same CpG sites noted here in SLC7A11, SLC43A1, and PHGDH, with a different CpG observed in SLC1A5, all associated with alcohol consumption (30). The top EWAS probe cg06690548, mapped to cystine/glutamate transporter SLC7A11, was replicated in the second cohort of alcohol use disorders (AUD) and control participants showing strong hypomethylation in AUD (P < 10–17). Importantly, it was observed that decreased methylation at cg06690548 in SLC7A11 was consistently associated with clinical measures, including increased heavy drinking days. Additionally, hypomethylation at cg06690548 was associated with elevated total cholesterol and TG levels (30). Regarding PHGDH, encoding phosphoglycerate dehydrogenase, increased lipid accumulation and reduced NAD + activity were seen in mouse Phgdh-knockout primary hepatocytes incubated with free fatty acids, effects that were reversed upon Phgdh overexpression, including reduced hepatic TG accumulation (31). SLC1A5 is known as a transporter of alanine, serine, and cysteine but transports glutamine in a Na+-dependent manner in the liver (32). A comparison of rats fed a high-alcohol diet either supplemented with glutamine (at 0.84%) or not indicated that hepatic fat deposition, inflammation, altered liver function, and hyperammonemia in the glutamine group were all attenuated (33).
Parallel to the stress that alcohol intake places on the hepatic biological clocks are oxidative stress in the liver and its induction of TXNIP (34). In cultured hepatocytes and mouse livers, alcohol exposure inhibited the expression of FoxO1, identified as a transcriptional regulator for microRNA MIR148A, which is a direct inhibitor of TXNIP expression (35). Furthermore, hepatocytes treated with ethanol exhibited TXNIP overexpression and activation of the NLRP3 inflammasome and caspase-1-mediated pyroptosis (35). Similarly, it was reported that exposure of the liver to high levels of alcohol results in reduced capacity to methylate proteins and DNA, as observed with protein phosphatase PP2A. Reduced action of this phosphatase permits phosphorylation and nuclear exclusion of FoxO1, leading to increased expression of TXNIP, which caused hepatic lipid accumulation (36). Lastly, numerous reports connect lipogenesis and glucose metabolism regulator SREBF1 and its encoded proteins to the effects of alcohol, for example (37). In sum, our results examined in the context of these previous reports clearly show that the observed associations between methylation levels at specific CpGs and outcomes related to metabolic diseases can be strongly mediated by various exposures. Thus, EWAS must consider the impact that dietary and other lifestyle exposures impart on those CpGs that are sensitive to such in ways that manifest as altered risk of disease.
Importantly, the dietary assessment of the FOS cohort from exams 5 to 8 over 13 years uses data from four standardized exams (18), making their use in such analyses as presented here a distinct advantage. This study examined all dietary intakes measured at four different time points. Several key foods, like alcohol, carbohydrate, ice cream, and sugar, plus smoking, all show consistent correlation with these DMSs across four exams. However, there was a trend for alcohol consumption, sugar intake, and smoking exposure to decrease from exam 5 to exam 8 (18).
This study is not without its limitations. One of those is the measurements of epigenetic status were performed in PBMCs, which may not be the optimal tissue for epigenetic signals of diet and lifestyle habits as related to TG. Yet DNA methylation measured from blood DNA can accurately predict biological age (3), which is associated with environmental exposure (7). Second, the loci described here are from the study population alone, and are not to be considered as general-use biomarkers of exposure to alcoholic drinks or other dietary factors, as equating the methylation status at specific loci with exposure to alcohol would be unethical (38). In addition, while there is no replication of these results in another cohort, the associations between TG-associated DMSs at exam 8 and diet and lifestyle habits were observed in four exams over 13 years. Although that consistency strengthens the findings, it must be recognized that such epigenetic marks of diet and lifestyle could be specific to given environments and populations. Hence, the conclusions based on findings from the current study must be interpreted with caution.