DNA Methylation-based Biomarkers of Protein Levels and Cardiovascular Disease Risk: Opportunities and Challenges for Precision Cardiology

Over the last three decades, high-throughput technologies have emerged as powerful tools for research and clinical applications. A broad range of assays now allow for relatively low-cost generation of omics data including genomics, epigenomics, transcriptomics, proteomics, and metabolomics, opening the door to new approaches for the development of novel therapeutics and precision medicine. Of particular interest is DNA methylation, the addition of a methyl group (-CH₃) to a cytosine residue. DNA methylation is mitotically heritable, occurs at specific loci throughout the genome, is mechanistically linked to the regulation of gene expression, and is relatively more stable over time than other epigenetic markers.¹ Genome-wide DNA methylation levels can now be measured at a single-site resolution in over 900,000 loci with commercially available microarrays.² DNA methylation data from cohort studies have enabled the training of biomarkers related to smoking,^3,4 diseases including cancer,⁵ and chronological and biological aging.⁶

In this issue, Chybowska et al evaluated the potential for predicting cardiovascular disease (CVD) risk using blood DNA methylation-based estimates of protein levels, referred to as protein “EpiScores.”⁷ This study leveraged data from 12,657 adults with 16 years of follow-up in the GS study (Generation Scotland), a family-based cohort initiated in 2006. Using DNA methylation levels measured using the Illumina Infinium MethylationEPIC array, the authors calculated 109 previously estimated EpiScores⁸ for plasma proteins, including those with established associations with CVD (eg, CRP (C-reactive protein), MMP12 (matrix metalloproteinase-12), and OMD (osteomodulin)), as surrogates for the direct measurement of protein concentrations. An EpiScore for cTnI (cardiac troponin I), a known marker of myocardial damage and used in the diagnosis of myocardial infarction, was also developed in GS. The authors found that 36 protein EpiScores were associated with CVD risk after correction for multiple comparisons. The authors also analyzed the predictive ability EpiScores beyond that of cTnI concentrations and the ASSIGN score, a cardiovascular risk score based on clinical factors, family history, and place of residence.⁹ In models including cTnI concentration and ASSIGN, 33 protein EpiScores remained significantly associated with CVD risk, with modest improvements in risk prediction. The composite CVD EpiScore similarly increased prediction accuracy compared with a model with cTnI concentration and ASSIGN. The authors highlight the associations of protein EpiScores with CVD risk independent of a protein marker of myocyte damage and an established risk score, but recognize the limited increase in predictive accuracy and marginal utility of the EpiScores in a clinical setting at this moment.

Before evaluating the utility of epigenetic biomarkers, it is important to consider the definition of a biomarker as “objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention”¹⁰ and the ability to be “accurately measurable across individuals and populations.”¹¹ DNA methylation has demonstrated great potential as a biomarker considering its regulation of gene expression, mitotic stability, dynamic nature yet conserved stability relative to other candidate molecules, accessibility in noninvasive samples, and availability of validated high-throughput detection methods. However, several challenges must be overcome to identify epigenetic biomarkers, particularly those suitable for patient care.

The progress toward developing epigenetic clocks, DNA methylation-based biomarkers of health span and life span, may provide lessons for other diseases including CVD. The observation that DNA methylation varies predictably with age first led to the development of epigenetic clocks as estimates of chronological age. Although these clocks are associated with mortality,⁶ they display weak associations with clinical measures that may be amenable to intervention, for example, blood pressure and glucose.¹² Training on a range of clinical markers or DNA methylation-based surrogate measures has improved prediction of age-related clinical measures, phenotypes, and morbidities, including coronary heart disease.^13,14 Epigenetic clocks have also advanced through novel training approaches. Due to the impact of technical artifacts, epigenetic clocks have suffered from poor reliability. Therefore, despite the utility of epigenetic clocks on the population level in research settings, longitudinal measurements within an individual are difficult to interpret. Novel machine learning training approaches have proven effective in discriminating between biological and technical variation and improving reliability.¹⁵

This advancement from the identification of DNA methylation as a suitable molecular indicator of aging to defining appropriate outcomes with clinical relevance and improving reliability may provide a framework for the development of new epigenetic biomarkers for CVD. For example, training of a CVD biomarker may incorporate readily available clinical risk factors (eg, hypertension and blood lipids), comorbidities (eg, diabetes and obesity), age, and environmental risk factors through direct measurement or epigenetic surrogates.¹⁶ Exploring additional machine learning approaches in training of protein EpiScores or a single risk biomarker may help to mitigate technical noise and improve accuracy across studies and DNA methylation platforms. In addition, leveraging data from cohorts including populations with diverse ancestries will improve biomarker performance and, most importantly, help to ensure health equity in precision medicine.

Despite the rapid increase of publications on epigenetic biomarkers, translation of candidate biomarkers from research to clinical applications has been marginal.⁵ As outlined previously,⁵ an assay suitable for clinical settings must address clinical utility, cost effectiveness, suitability for use with other modes of testing, accessibility and storage of tissues analyzed, and product-market fit. Perhaps the largest barrier to clinical translation is currently the ability to advance from establishing associations between DNA methylation and health outcomes to identifying diagnostic or prognostic uses.¹⁷ Building on advancements in cancer biology research, clinical epigenetic biomarkers have been largely restricted to the field of oncology.^5,11,17,18 Liquid biopsies can provide a minimally or non-invasive source of circulating tumor cells and cell-free circulating tumor DNA, which may better represent heterogeneity within tumors compared with traditional biopsies. DNA methylation-based assays have reached the Centers for Medicare and Medicaid Services, European Union, or Food and Drug Administration premarket approval for use in the detection, monitoring, or predicted treatment response for cancers including bladder, breast, cervical, colon, liver, lung, and glioblastoma.^5,17 These tests can complement other tools in precision oncology to inform decision-making and improve patient outcomes. As highlighted by Chybowska et al⁷ CVD may also be a suitable candidate for DNA methylation-based biomarker development and clinical application.

Early risk detection and intervention are paramount to prevent morbidity and mortality from CVD as a highly preventable chronic disease. Although DNA methylation is commonly measured in peripheral blood leukocytes and distinct from other circulating protein biomarkers of CVD, DNA methylation likely carries signals of CVD risk due to the relationship between leukocyte function, inflammation, and CVD health.^19,20 Moreover, because DNA methylation is mitotically heritable it may be a more stable marker of risk than fluctuations in plasma protein concentrations. Findings presented in the study by Chybowska et al provide some evidence that DNA methylation-based biomarkers of protein concentrations may predict CVD risk beyond existing screening tools.⁷ Furthermore, leveraging DNA methylation measured with nonspecific genome-wide microarrays provides cost-effective opportunities for calculating biomarkers of multiple diseases with a single biospecimen and assay. Future research should prioritize exploration of additional CVD-related clinical markers as training outcomes, ensuring reliability across time and in diverse populations, and evaluation of sensitivity, specificity, and positive predictive value.

Omics-based biomarkers, including those developed using DNA methylation data, are promising tools in precision cardiology for early disease risk assessment, diagnosis, and clinical decision-making. As demonstrated in cancer treatment, DNA methylation-based biomarkers may be particularly informative for guiding treatment approaches by providing insights to underlying molecular pathways. However, the current limited use in clinical settings highlights the challenges of advancing research beyond proof-of-concept studies. Nevertheless, in light of recent increases in approved cancer-related biomarkers, DNA methylation- and other omics-based biomarkers have the potential to revolutionize precision medicine for multiple diseases such as CVD, and ultimately benefit patient health.