Statin therapy is associated with epigenetic modifications in individuals with Type 2 diabetes

Statins are one of the most commonly prescribed medicines worldwide. They inhibit HMG-CoA reductase and reduce hepatic cholesterol levels which leads to reduced plasma low-density lipoproteins (LDL) levels. Due to their high tolerability, statins have been widely accepted to be effective for cardiovascular disease (CVD) prevention [1]. Since CVD is one of the main causes of mortality in individuals with Type 2 diabetes, statins are a common treatment among these patients. Conversely, there is evidence that statins increase the risk of Type 2 diabetes [2]. However, it is not yet fully understood what role statins play in individuals with Type 2 diabetes themselves.

Epigenetics, including DNA methylation, is often referred to as the link between environment and genetics and may give insights into how statins affect patients on this therapy. While statin therapy was recently linked to epigenetics in nondiabetic individuals [3], there are, to our knowledge, no studies investigating the effects of statins on DNA methylation in Type 2 diabetes patients. The increased use of statins in this at-risk population due to comorbidities, such as CVD, presents a need to better understand the molecular mechanisms associated with this treatment. In the future, this could potentially ameliorate personalized medicine in Type 2 diabetes.

This study, therefore, aimed to elucidate whether statin treatment in newly diagnosed patients with Type 2 diabetes is associated with altered blood DNA methylation and whether these epigenetic changes may indicate how statin treatment affects these patients.

Materials & methods

Study population

This study included a discovery cohort of 198 newly diagnosed individuals with Type 2 diabetes from the All New Diabetics in Scania (ANDIS, http://andis.ludc.med.lu.se) study, consisting of 88 cases and 110 controls regarding statin therapy at diagnosis of Type 2 diabetes, as well as an independent replication cohort including 58 cases on statin therapy and 108 naive controls from the All New Diabetics In Uppsala County (ANDiU, http://www.andiu.se/) study (Table 1). Individuals in ANDIS were considered cases if patients picked up statins (anatomical therapeutic chemical [ATC]-code: C10) from the pharmacy and in ANDiU if statin treatment was registered in the clinical health journal within 6 months before blood was sampled. Written consent was obtained from all patients and ANDIS and ANDiU research protocols were approved by the Ethical Review Board in Lund (no.: 584/2006, 2011/354, 2014/198) and Uppsala (2011/155), respectively.

Individuals with missing information regarding phenotypes included in the regression analysis, who were on diabetes medication at baseline, whose DNA samples did not pass quality control or who were on statins ≤7 days, were excluded. Statin positive cases were thus defined as newly diagnosed Type 2 diabetes individuals having started statin therapy >7 days before blood samples were taken. Cases and controls in ANDIS were balanced for antihypertensive medication (ATC codes: C02-C03, C08-C09) (Table 1 & Supplementary Figure 1) to minimize its effect on our results/data. Patients from the ANDiU replication cohort were matched for age, BMI, HbA1c, HOMA2-B, HOMA2-IR, antihypertensives using the R package MatchIt using nearest neighbour matching (https://github.com/kosukeimai/MatchIt), in order to be comparable with ANDIS (Table 1 & Supplementary Figure 1). Whether patients were on antihypertensives was assessed in the same measure as for statins described above.

DNA methylation analysis

DNA was extracted from blood, collected at registration, using the Gentra Puregene Blood kit (Qiagen, Germany). Nucleic acid concentration was measured with the NanoDrop 1000 spectrophotometer (NanoDrop Technologies, DE, USA). The EZ DNA methylation kit (Zymo Research, CA, USA) was used for bisulphite conversion of genomic DNA (500–1000 ng). Samples were randomized across chips before being amplified and fragmented following the Infinium HD assay methylation protocol (Illumina Inc., CA, USA). DNA methylation was analyzed using the Illumina MethylationEPIC BeadChip microarray. Illumina iScan was used for imaging of BeadChips and methylation scores were retrieved using GenomeStudio Methylation module. All samples had high bisulphite conversion efficiency (intensity signal >4000) and passed GenomeStudio quality control. Bioconductor (https://www.bioconductor.org/) was used for further analysis. Probes with mean detection p-value of > 0.01 and cross-reactive, reference SNP, annotation based polymorphic [4] and Y-chromosome probes were excluded.

Statistical analyses

Phenotypical differences were assessed using Mann–Whitney U test for continuous variables and chi-squared test for categorical variables. p < 0.05 was considered significant.

To identify differently methylated sites, we conducted a linear regression analysis in ANDIS adjusted for age, BMI, sex and HbA1c, with DNA methylation being the dependent variable and statin therapy being the independent variable. We further adjusted the analysis for antihypertensives in ANDiU due to the significant difference between cases and controls. Reference-free Houseman method was included in the models to adjust for cell composition [5]. A false discovery rate (FDR) correction based on the Benjamini–Hochberg procedure was applied. FDR <5% (q <0.05) was considered significant for the discovery cohort and FDR <10% in the replication due to a reduction of statistical power related to the sample size of this cohort. Kyoto Encyclopedia of Genes and Genomes (KEGG) gene set analyses were conducted using WebGestaltR [6]. Causal mediation analyses were performed with the mediation R package using the default settings [7]. Linear regressions to assess correlation between significant sites and HbA1c, HOMA2-B and HOMA2-IR were adjusted for age, sex and statin medication. p < 0.05 was considered statistically significant. All statistical analyses were performed using R [8].

Results

We investigated whether statin therapy is associated with altered DNA methylation in blood from 198 Type 2 diabetes patients in ANDIS and 166 patients in ANDiU. LDL levels were higher in controls versus cases on statin therapy, whereas no significant differences were found for the other variables (Table 1).

After correction for multiple testing, we found 79 differentially methylated sites between cases and controls for statin therapy in ANDIS (q <0.05), with nine sites having >3% absolute differences in DNA methylation (Supplementary Table 1). Interestingly, DNA methylation of three out of five previously identified sites (cg17901584 DHCR24, cg27243685 ABCG1, cg05119988 SC4MOL) associated with statin therapy in nondiabetic individuals [3] were among our significant sites in the discovery cohort (Figure 1A). A large proportion (∼2/3) of these 79 sites were hypermethylated (Figure 1B & C). We next performed KEGG pathway analyses of the 72 genes annotated to our 79 sites. We identified 2 KEGG pathways, arginine and proline metabolism and thiamine metabolism, with p < 0.05 (Supplementary Table 2), however, no pathway remains significant after correction for multiple testing. Furthermore, two of our 72 unique genes with differential methylation, NRXN3 and CAMK1D, have previously been associated with Type 2 diabetes in genome-wide association studies (GWAS) (GWAS catalogue, accessed 18 October 2019, maximum p-value of 9e^–6, Supplementary Table 1).

We then performed replication testing of sites significantly associated with statin therapy in ANDIS using an independent replication cohort, ANDiU (Table 1). DNA methylation of two sites, cg17901584 (DHCR24) and cg23011663 (ARIH2), was also associated with statin therapy in the replication cohort (q <0.1) with directional consistency (Supplementary Table 1 & Figure 1D). In addition, DNA methylation of three additional sites was also nominally associated with statin therapy with directional consistency in the replication cohort (p < 0.05, Supplementary Table 1).

We next used causal mediation analysis to investigate whether DNA methylation of any of the 79 sites found to be associated with statin therapy are part of a pathway through which statins exerts its effects on LDL, HOMA2-B or HOMA2-IR [7]. While causal mediation analyses showed that statin therapy had a significant direct effect on LDL levels, but not on HOMA2-B or HOMA2-IR, there was no mediative effect of DNA methylation of the two replicated sites (cg17901584 and cg23011663). Of our 79 significant sites in the discovery cohort, only DNA methylation of cg23080761 annotated to SH2B3 had an average causal mediation effect regarding statins’ effect on LDL in the ANDIS discovery cohort (q <0.05), but not in the ANDiU replication cohort (Supplementary Table 3).

We finally tested if DNA methylation of the sites associated with statin therapy is also associated with measures of glucose metabolism. We hence correlated DNA methylation of these sites with HbA1C, HOMA2-IR and HOMA2-B in a combined cohort including both ANDIS and ANDiU. We found a significant association of HbA1c and HOMA2-IR with methylation of cg17901584, but there was no association regarding methylation of cg23011663 (Supplementary Table 4).

Discussion

In this study, we found epigenetic differences between Type 2 diabetes individuals taking statins and those who were statin-naive. Notably, methylation of three out of five sites recently associated with statin therapy in nondiabetic individuals [3] were also found in our discovery cohort, indicating that similar epigenetic changes take place in Type 2 diabetes patients and nondiabetics when exposed to statin therapy.

While statins have a direct effect on LDL, only methylation of cg23080761 annotated to SH2B3 had an average causal mediation effect regarding statin’s effect on LDL in ANDIS. Interestingly, a SNP at SH2B3 was associated with SH2B3 expression, HDL, LDL, total cholesterol and rheumatoid arthritis [9]. Moreover, Mendelian randomization analysis showed that SH2B3 expression was associated with total cholesterol [9]. None of the other significant methylation sites were mediating the effect of statins on LDL. The effect of statins on methylation may therefore affect other mechanisms involved in cholesterol synthesis and immune response, since we replicated two genes (DHCR24 and ARIH2) implicated in these pathways and annotated to sites which were differentially methylated between cases and controls. Differential lower methylation of a site annotated to DHCR24 was associated with statin therapy in our discovery (ANDIS) and replication (ANDiU) cohorts as well as in the study by Ochoa-Rosales et al. [3]. DHCR24 encodes DHCR24 protein, which catalyzes reduction of delta-24 double bond of sterol intermediates during cholesterol biosynthesis. Reduced methylation of this gene in people on statin therapy may thereby affect cholesterol biosynthesis (Supplementary Figure 2). Interestingly, we found correlations between DNA methylation of the site annotated to DHCR24 and HbA1c and HOMA2-IR, suggesting that this gene is also linked to glucose metabolism. Furthermore, statin therapy was associated with increased methylation of cg23011663, annotated to ARIH2 which is, just as cholesterol, involved in the activation and de-activation of the NLRP3 inflammasome [10] and may thus affect immune response (Supplementary Figure 2). Other studies further suggest that statins modulate the immune system directly and that some of their beneficial properties are therefore not mediated through lowering cholesterol [11].

A strength of this study is the balanced case–control design targeting a homogenous at-risk population of newly diagnosed Type 2 diabetes patients, thereby reducing possible bias regarding age, sex, HbA1c, BMI and hypertension. This allows to ascertain effects of statin therapy in a population which has an increased need for CVD prevention therapy. The inclusion of a replication cohort is an additional strength. A potential limitation of this study is the fact that statin cases in the discovery cohort are based on whether they picked up the medication from the pharmacy, as it relies on the compliance of the individual patients. Missing clinical data regarding LDL, HDL and HOMA2 may affect the power of the causal mediation analysis and the drawn inferences. Similarly, we could not control the analyses for smoking or medication dosage since this information was not available in both ANDIS and ANDiU. Furthermore, this study was carried out in Northern Europeans and validation in other ethnicities should be conducted in future studies. Additionally, a limited sample size may carry a risk of statistical inflation of the results, however, widely used measures such as the genomic inflation factor, cannot be applied to epigenome-wide association studies since here, in contrast to GWAS, the outcome of interest is expressed in larger numbers of small genetic changes [12]. Nevertheless, our replication in an independent cohort and a previous study including nondiabetic individuals supports statistical robustness of our results.

Conclusion

Our study demonstrates that statin therapy associates with altered DNA methylation in individuals with Type 2 diabetes. Target genes are involved in pathways essential to cholesterol biosynthesis and the NLRP3 inflammasome (Supplementary Figure 2), supporting the key biological role these sites play. Further follow-up studies are necessary to determine the risk of Type 2 diabetes progression in regard to statin therapy.

Future perspective

The findings of our study support that statin therapy affects DNA methylation on sites annotated to genes that play a role in cholesterol synthesis and immune response. This study could help future research targeting larger samples to validate the applicability of these findings to the general population with Type 2 diabetes. Furthermore, it would be of interest to determine whether DNA methylation on the sites associated with statin therapy influences the progression of Type 2 diabetes or CVD using prospective cohorts. This could be of importance for clinical practice, particularly in an at-risk population such as individuals with Type 2 diabetes.