A transdisciplinary approach to understand the epigenetic basis of race/ethnicity health disparities

The goal of this manuscript is to provide a transdisciplinary perspective on the contribution of epigenetics to the understanding of health disparities among socially defined racial/ethnic groups, drawing on expertise from scientists in the fields of biological anthropology, bioinformatics and molecular epidemiology. Herein, we summarize theoretical, statistical and practical elements of investigating racial/ethnic health disparities using epigenetics and provide examples of successful applications from cancer research using this approach. This work will aid in shaping how researchers understand and therefore approach problems of health disparities when incorporating epigenetic data.

What is epigenetics?

Epigenetics is the study of heritable phenotypic variations that do not involve changes in the DNA sequence, often involving control for gene activity and expression [1,2]. Examples that produce such changes are histone and chromatin modification, and DNA methylation. DNA methylation is the best-studied type of epigenetic modification in humans, playing a critical role in the regulation of gene expression. This primarily occurs by reducing the transcription of genes at the promoter and enhancer regions, although more complex regulatory mechanisms have been described in other gene contexts [1093,4]. DNA methylation is one of the key players in cellular differentiation, providing cell identity [5]. Genetic variation and environmental factors can affect the cell subpopulation selection through epigenetic adaptation, and under extreme pressures, can alter cell differentiation and lead to abnormal phenotypes, including cancer cells [6–8]. This plasticity makes epigenetic processes lie at the interface of the environment and transcriptional control.

Using epigenetics to understand the development of health disparities

Health disparities are differences in disease risk between populations groups, and oftentimes, are accompanied by a higher than expected mortality burden. Health disparities are common across a variety of social contexts and are documented for a wide range of morbidities that occur across the life course. Such disparities may emerge according to race. Race is defined here as the social construct of human variability based on perceptions of biological differences (e.g., skin color or other aspects of physical appearance). While race categories do not reflect genetically distinct groups, experiences of racism and structural violence can adversely impact the biology and health of racialized minorities. Similarly, health disparities may also emerge according to ethnicity. Ethnicity, according to Mersha et al., refers to a ‘multidimensional construct reflecting biological factors, geographical origins, historical influences, as well as shared customs, beliefs, and traditions among populations that may or may not have a common genetic origin’ [9]. As such, each ‘race’ or ‘ethnicity’ may include multiple subgroups (e.g., Hispanic/Latinx are comprised of Cubans, Panamanians, Ecuadorians, Argentinians, etc.).

Given that racial/ethnic groups are social categories that do not necessarily align with underlying patterns of genetic variation [10], genetic factors alone are insufficient to explain how racial/ethnic health disparities emerge. It is therefore critical to evaluate how differences in experience and environment shape health outcomes. In fact, it is now becoming clear that environmental differences are important in shaping more complex phenotypes that are of interest in public health regarding racial/ethnic inequalities, such as low birth weight, preterm birth, asthma, cancer and cardiovascular disease. Epigenetic studies may help to understand how differences in environmental experience translate into differences in phenotype. Although a limited number of studies have directly connected DNA methylation to health disparities, a few studies have reported intriguing results. Socioeconomic status [11], as well as factors that vary according to socioeconomic statuses such as psychosocial stress exposure [12,13] and diet [14,15], have been associated with variation in DNA methylation.

Early life experiences may be particularly important for shaping health disparities. The Developmental Origins of Health and Disease hypothesis suggests that a mother’s experience during her life and her pregnancy may shape the epigenome and future health trajectory of her infant [16,17]. Moreover, different environmental exposures during pregnancy (maternal lifestyles, diseases and exposures to environmental toxicants) have been associated with alterations in DNA methylation in the placenta or the umbilical cord blood of the newborn [18–21]. However, there has been criticism of the overemphasis on maternal effects relative to paternal effects in predicting health outcomes via epigenetic mechanisms [22]. While it has been less frequently investigated, growing evidence suggests that paternal environmental experience can also affect offspring through epigenetic processes. For example, paternal obesity in the peripartum period is associated with significant differences in offspring methylation at imprinted genes important for regulating growth and development [23]. These findings suggest that socially patterned exposure to stressors in both parents could potentially affect offspring health via changes in the offspring epigenome.

Among historically marginalized communities, the ancestral experience of trauma (i.e., historical trauma) shapes disparities in later health generations [24–26]. In addition to being related to the intergenerational effects described above, the health impacts of historical trauma likely also reflects within-generation epigenetic impacts of environmental conditions shaped by ancestral experience [27,28]. For example, forced displacement of ancestors increases the likelihood that members of the contemporary generation experience poverty and therefore associated health sequelae. Likewise, the parental experience of trauma could shape both the intrauterine environment and patterns of parental care, both of which affect the developing epigenome of offspring [27,29]. Therefore, historical trauma should be considered as an additional conceptual model for explaining observed health disparities.

Statistical methodology for studying the epigenetic basis of health disparities

Studies investigating the DNA methylation basis of health disparities have generally employed global DNA methylation, targeted gene methylation, single variant methylation or network-based analysis [30–32]. With the rapid development of high-throughput technologies in recent years, population-based epigenome-wide association studies at a single-nucleotide resolution became a popular approach utilized in epigenetic studies. The common supervised selection strategy is to select CpG sites affecting phenotypic differences, noted as differentially methylated cytosines [33]. Differentially methylated cytosines can be selected based on the absolute difference in mean beta values or test statistics from a t-test, Wilcoxon test or multivariable regression model. Another selection approach is to consider differential variance in methylation between two traits, noted as differentially variable cytosines, using, for instance, the Bartlett test or the Levene’s test [34,35]. In genomics, variance-based selection of SNPs approaches could be used to prioritize those SNPs for subsequent gene–gene and gene–environment testing [36]. Although some tools have been developed for this goal [37], epigenome-wide approaches for epigenome–environment and genome–epigenome interactions are infrequently studied beyond locus-specific interactions due to computational burden and concerns over model assumptions when using untargeted approaches. On the other hand, the unsupervised selection procedure is to rank and filter CpG sites by variance, aiming for the selection of the most variably methylated cytosines [38]. The most variably methylated cytosines generally represent various levels of DNA methylation and may contain those driven by SNPs or cell heterogeneity (e.g., different immune cell populations in blood/saliva or immune-cell infiltration in solid tissues). Studies have also employed the most variably methylated cytosines approach to filter CpGs before single variant methylation tests to reduce the burden of multiple hypothesis testing.

Differences at any individual site may be small; however, if these differences are persistent across a region or a certain group of genes, statistical power to detect them may be greater. Several methods have been developed to identify sets of neighboring CpGs sites that are correlated with each other, known as differentially methylated regions, and link them with traits of interest, including DMRcate [39], bump hunting [40] and the A-clustering method [41]. Other methods aiming to build gene co-methylation networks have also been proposed. For example, weighted gene co-methylation network analysis aims to describe the correlation patterns among genes across microarray samples, find clusters of highly correlated genes, and relate such clusters to a phenotype of interest via enrichment analysis or network eigengenes (the top principal component of genes in the cluster) [42].

Challenges & opportunities in statistical methodology

Although advances in epigenetic studies are expected to help understand racial/ethnic health disparities, there are notable challenges and limitations to consider. Epigenetic studies are potentially impacted by a range of confounding factors, including but not limited to population genetic patterns, cell-type, environmental confounders related to ethnicity and sample processing batch [43]. Population stratification is another critical source of confounding for studies including heterogeneous populations. DNA methylation signatures of target tissue (e.g., saliva, whole blood, placenta, adipose and tumors) are an average of cell-type-specific methylation levels. Hence, the cell-type proportion is generally related to the measured DNA methylation levels, and in many cases, is also associated with race/ethnicity [44] and traits of interest [45]. Various statistical methods have been proposed to adjust for this potential bias. Statistical models for cell-type deconvolution are classified into three categories called reference-based [46], reference-free [47] and semi-reference-free [48–51], the last of which alleviates some of the problems of both reference-based and reference-free methods. The choice of the appropriate method for cell-type deconvolution mainly depends on the availability of a proper reference database for DNA methylation of the cell types involved [52–54]. Some other methods developed specifically for methylation data, or for general purposes, can be used to control for all unmeasured confounding, including; surrogate variable analysis (SVA) [55], independent SVA [56], smartSVA [57], remove unwanted variation [58,59] and principal component analysis [60]. Several of these methods have been used to adapt reference-free approaches and semi-reference-free approaches and have been reviewed and compared elsewhere [2,61]. However, residual confounding may still be possible after adjustment. In statistical genetics, genomic inflation represents the excess of false positives in genomic analyses. In epigenome-wide association studies, the genomic inflation factor calculation and the quantile-quantile plots have been used to quantify the excess inflation in statistics, however, in most of these analyses, genomic inflation is not corrected. The application of genomic control correction has shown to be ineffective due to the small differences detected in epigenome-wide association studies, with only a few methods that have been adapted specifically for DNA methylation analyses but are still not widely used in the field [62]. Finally, several biomarkers derived from DNA methylation information have been developed which could offer global measures of epigenetic drift related to various phenotypic variations of interest. Among those, we have age acceleration using the DNA methylation age measures [63,64], fetal cell of origin [65], inference of multiple retrotransposons using epigenome-wide information [66] or global methylation changes [67]. The selection of specific methods should be adapted to the specific hypothesis being tested.

Insights & applications from cancer research

Locus specific changes and differentially methylated cytosines related to race/ethnicity have been identified among cancer biologists interested in health disparities and have been extensively reviewed in the past [30,68–70]. Beyond the locus-specific promoter changes, epigenome-wide association studies have continuously reported variation in DNA methylation patterns between different populations such as Europeans, Hispanics, Africans and Asians [71]. Determining how these race/ethnicity variations are associated with disease outcomes will further help to understand health disparities. Another approach is to describe demographic and environmental factor-associated and disease-associated differentially methylated cytosines in different race/ethnicity groups.

Cancer health disparities research in the USA is largely focused on race/ethnicity, where cancer incidence is highest among African–Americans or Blacks, followed by non-Hispanic whites, Hispanics and Asian/Pacific Islanders [72]. Generally, African–Americans also have the highest mortality rates and worse survival outcomes in comparison with all other race/ethnicity groups. The disparity gap between whites and African–Americans for cancer incidence and mortality has narrowed over time, but there is still a notable 14% difference in the mortality rate [72]. While African–Americans are disproportionately affected overall, other race/ethnicity groups (e.g., Hispanics, Asian/Pacific Islanders) have a greater cancer burden or worse survival for certain cancers. For example, Hispanics have a higher incidence of infection-associated cancers (e.g., liver, stomach and cervical cancer) [73]. Cancer is also more prevalent among socially, economically or environmentally disadvantaged populations. Higher cancer incidence and mortality rates, as well as lower survival, are experienced by cancer patients with low educational attainment or residents of impoverished neighborhoods compared with more educated individuals and residents of affluent areas [74]. As the socioeconomic status and race/ethnicity are inextricably linked to one another, it is often difficult to disentangle their independent effects on cancer disparities.

The underlying causes of cancer health disparities are complex and multifactorial. While a portion of the disease burden is due to the marginalization of minority populations, disease susceptibility is a combination of population isolation, genetic burden and selection of specific phenotypes that are advantageous for certain environments [75]. One example is the trends in cancer subtype susceptibility for certain race/ethnicity groups. Skin cancer distributions differ across race/ethnicity as the risk of squamous cell carcinoma is higher in Eurasian descendants and anecdotally in African populations with albinism [76]. Triple-negative breast cancer and aggressive prostate cancer are much more frequent in African–Americans compared with other racial/ethnic groups in the USA. Even after accounting for healthcare access and other social factors, African–Americans with these subtypes have a worse prognosis compared with white–Americans. Interestingly, Hispanic cancer patients have better outcomes than African–Americans despite similar sociodemographic characteristics, also known as the ‘Hispanic paradox’ [77], while at the same time, this group is still adversely affected by other health outcomes, such as infectious diseases, disabilities and diabetes compared with non-Hispanic whites. As neighborhood socioeconomic status has been shown to contribute to survival disparities in Black and Hispanic cancer patients, but not Asian/Pacific Islanders [78], neighborhood socioeconomic status does not represent the only source of variability contributing to health disparities for these groups.

Considerable strides have been made in cancer research to investigate the link between DNA methylation and cancer health disparities, primarily for the most common cancers. In breast cancer, several studies have identified differentially methylated loci when comparing tumors from African–American and European–American women [79–84], with the most differences observed in women with estrogen receptor (ER) negative tumors and younger women. Another study uncovered seven genes hypermethylated in Korean versus European women, which again, was particularly seen among ER and progesterone receptor (PR) negative tumors and women aged ≤50 years [85]. Similarly, work in prostate cancer found several CpG sites that are differentially methylated among tumors from Black versus white men [86–90], with studies consistently implicating CD44 and GSTP1 [86,87,90]. Besides breast and prostate cancer, the literature investigating the epigenetic basis of race/ethnicity disparities in other cancer types is fairly sparse [91–97], especially for rare cancers where challenges arise due to the limited number of cases in existing studies, particularly within minority or underserved populations. Many additional studies in the literature investigate DNA methylation of cancer patients within a specific race/ethnicity but do not compare with other race/ethnicity groups. While these studies uncover the unique epigenetic alterations within different populations, they do not provide a comparison group to elucidate a potential racial/ethnic disparity and are not discussed herein.

Challenges in epigenetic health disparities research

Race/ethnicity is the most common disparity investigated in cancer epigenetics as well as other disease disparities, with the majority of studies comparing African–American/Black and European–American/White populations. Little emphasis has been placed on other race/ethnicity groups (e.g., Hispanics, Asian/Pacific Islanders, American–Indians/Alaskan Natives), although these under-represented groups have notable disparities for many chronic and acute diseases and are a growing proportion of the US population. Moreover, any racial/ethnic categorization encompasses very heterogeneous populations. For instance, Hispanics and Asian/Pacific Islanders represent a variety of ethnic subgroups (e.g., Cuban, Mexican, Filipino and Chinese) that have different experiences and risk profiles. These racial/ethnic groups are often understudied due to inadequate sample sizes or the under-representation of these populations within any one study. Adding to this challenge, there is typically a lack of studies with available biospecimens to conduct epigenetic disparities research.

Promoter methylation of candidate genes or a preselected panel of genes has been the approach used most often to measure DNA methylation. The array-based methodology offers an agnostic approach to move beyond a single gene or gene promoter. However, the majority of studies using arrays to quantify DNA methylation levels have used the Illumina 27 or 450K array. Both are now obsolete after the introduction of the MethylationEPIC (or 850K) array, which provides comprehensive genome-wide coverage and captures additional enhancer and intergenic regions of the genome that were not included on the older versions of the array. As genome-wide association studies note the importance of noncoding regions of the genome in disease susceptibility, the EPIC array will be able to shed light on whether this is also true for epigenetic alterations. DNA methylation sequencing (e.g., reduced representation bisulphite sequencing or whole genome bisulphite sequencing) has been used in a few health disparities studies [98], however, this technology cost is higher than the microarrays and depending on the biospecimen, the genome coverage may not be consistent.

Conclusion

Epigenetic markers have shown several interesting associations that could be driving health disparities from a biological perspective. When investigating the association between racially/ethnically different epigenetic variations and disease outcomes in cross-sectional settings, determining causality is challenging [99]. Prospective follow-up studies among racially/ethnically heterogeneous populations would allow researchers to identify methylation changes involved in different pathways preceding disease onset. Using a Mendelian randomization approach to integrate genotype and epigenetic data may also prove useful in determining causality [100]. However, there are several limitations of this approach including but not limited to low statistical power, population stratification generating spurious genetic variants, re-introduced confounding through pleiotropy and linkage disequilibrium with multiple causal genetic variants of the epigenetic variation [100].

The use of machine learning is of great interest in disease prediction and classification. For example, the elastic net, a penalized regression model, has been applied in predicting human age with DNA methylation data in the USA [63], Chinese and multiracial/multiethnic populations [101,102]. Integrating the epigenome with other types of -omics data such as the genome, transcriptome, proteome, metabolome and the microbiome has the potential to unlock the ‘black box’ in health disparities. Although current technologies are still facing challenges, researchers have found intriguing results [103,104]. Future enhanced bioinformatics and analytical tools [105] will enable a more comprehensive analysis of human observational and interventional studies in a systematic way.

Future perspective

Some studies have devised an integrative approach including a comprehensive analysis of the social and environmental exposures of specific race/ethnicity associated epigenetic changes [32,106]. Newer longitudinal cohorts are trying to recruit more diverse populations representing minorities that were not included in traditional cohort studies [107]. Transdisciplinary approaches to understand the roots of health disparities are required to improve the outcomes of minorities and marginalized populations. From the genetic point of view, researchers are moving beyond self-reported race/ethnicity to the use of ancestry informative genetic markers. In highly admixed populations, such as Latin–Americans or African–American populations, ancestry informative genetic markers reveal a different layer of information about population migration and in some cases, clusters of disease susceptibility that may not be associated when using only self-reported race/ethnicity [9]. Ancestry information will provide broad geographically relevant population information (population migration and inbreeding); however, so far, the utility of genetic markers has been limited for interventions [108]. Epigenetics, on the other hand, provide a unique opportunity to fully integrate the genetic, social and environmental contributors to health disparities, while offering a potential for intervention.