Two-step epigenetic Mendelian randomization: a strategy for establishing the causal role of epigenetic processes in pathways to disease

Abstract

The burgeoning interest in the field of epigenetics has precipitated the need to develop approaches to strengthen causal inference when considering the role of epigenetic mediators of environmental exposures on disease risk. Epigenetic markers, like any other molecular biomarker, are vulnerable to confounding and reverse causation. Here, we present a strategy, based on the well-established framework of Mendelian randomization, to interrogate the causal relationships between exposure, DNA methylation and outcome. The two-step approach first uses a genetic proxy for the exposure of interest to assess the causal relationship between exposure and methylation. A second step then utilizes a genetic proxy for DNA methylation to interrogate the causal relationship between DNA methylation and outcome. The rationale, origins, methodology, advantages and limitations of this novel strategy are presented.

Figure 1

Mendelian randomization: using genetic variants as instrumental variables to establish whether an exposure is causally related to a disease or trait. (A) Instrumental variable (genetic variation) [G] acts as a proxy/instrumental variable for environmental exposure [E], postulated to influence disease [Y]. G is independent of unmeasured confounders or [U]. G only influences Y if E →Y is causal (red dashed line). (B) The influence of BMI on blood pressure using the FTO variant as an instrumental variable is shown as an example. A robust association between FTO and blood pressure is indicative of a causal association between BMI and blood pressure.

Figure 2

Integrative genomics: using genetic variants as a causal anchor to demonstrate that modulation of gene expression levels at a specific locus can influence the incidence of a disease or trait and to discount the possibility that reverse causation is at play, i.e. that the disease state is not acting to alter gene expression levels. (A) Instrumental variable (genetic variation) [G] acts as a proxy for gene expression level [R], postulated to influence disease [Y]. G is independent of unmeasured confounder or [U]. G only influences Y if R →Y is causal (red dashed line). (B) The relationship between SNPs at the 17q21 locus (ORMDL3 gene), transcript levels of genes in Epstein–Barr virus-transformed lymphoblastoid cell lines and childhood asthma is shown as an example. The SNPs associated with childhood asthma were consistently and strongly associated (P < 10⁻²²) in cis with transcript levels of ORMDL3, making these SNPs useful instrumental variables in this setting

Figure 3

Mediation: a modifiable causal risk factor [E] for disease [Y] exerts its causal effect (at least in part) via the effect of E on X (the mediator) and through the causal effect of X on Y. U¹ and U² represent all confounders for the association of E with Y and X with Y, respectively. U¹ and U² can include different characteristics. In simple multivariable analyses to test this hypothesis it is tempting to adjust the association of E with Y for U¹ and declare that this is the total causal effect of E on Y and then to adjust further for X; any resulting attenuation of the U¹ adjusted association of E with Y following further adjustment for X is considered to represent the amount of the causal effect of E on Y that is mediated by X. However, by conditioning on X a pathway between U² and E is produced and hence this association (E with Y) is now confounded by U². In this situation X is said to be a collider between and E and U². Furthermore, measurement error in X will bias the assessment of its mediation. Thus, both U¹ and U² require separate consideration and this can be achieved in the two-step epigenetic Mendelian randomization framework

Figure 4

Two-step epigenetic Mendelian randomization: applying the principle of Mendelian randomization to DNA methylation as an intermediate phenotype. Genetic variants can be used as instrumental variables in a two-step framework to establish whether DNA methylation is on the causal pathway between exposure and disease. An overview of the two-step framework of this approach is shown. (A) First, an SNP is used to proxy for the environmentally modifiable exposure of interest and (B) secondly, a different SNP is used to proxy for DNA methylation levels

Figure 5

Two-step epigenetic Mendelian randomization applied to smoking and cardiovascular disease: (A) instrumental variable [G] acts as a proxy for environmental exposure [E], postulated to influence disease [Y] via altered DNA methylation [X]. G is independent of unmeasured confounder [U]. G only influences X if E →X is causal (red dashed line). This is shown in the left hand side of Panel A. This is illustrated by considering the influence of smoking on cardiovascular disease risk using the CHRN3/5 variant as an instrumental variable for smoking intensity, as shown on the right hand side of Panel A. This variant has been used previously as an instrument to assess the causal relationship between smoking and BMI. The CpG site of interest could be the site in F2RL3 recently reported by Breitling et al. (B) In a second step, an alternative proxy, here an SNP in the same gene in which methylation is measured (F2RL3), is used, as shown in the two diagrams in Panel B to assess the causal relationship between X and Y or F2RL3 methylation and CVD

Figure 6

Applying a two-step epigenetic Mendelian randomization approach to in utero influences on offspring outcomes: (A) maternal instrumental variable [G_m] acts as a proxy for environmental influences [E_m] on fetal DNA methylation during pregnancy [X_o] and subsequent offspring outcome [Y_o]. G_m only influences X_o if E_m→X_o is causal (red dashed line). The postulated influence of maternal alcohol intake during pregnancy on offspring cognition using the maternal ADH1B variant as an instrumental variable is shown as an example. (B) An additional proxy, which correlates with methylation at the alcohol responsive CpG site(s), is then applied in a second step. In this instance, the offspring's genotype [G_o] is used as a proxy for methylation

Figure 7

Multiple instruments: a potential advantage of a two-step epigenetic Mendelian randomization approach is the exploitation of genetic heterogeneity where multiple instrumental variables can be combined to strengthen causal inference for a role of DNA methylation in the causal pathway between exposure and disease. (A) Genetic heterogeneity might occur when several genes [G₁, G₂, G₃] are robustly associated with the exposure of interest and can be used as separate instruments to interrogate the association of the exposure and DNA methylation. The influence of multiple lipid-altering SNPs might be one example. Collectively, the combination of lipid-altering variants can be used to repeatedly assess the causal relationship between lipid levels and DNA methylation. (B) A multiple instrument approach is also possible in step 2 where several uncorrelated cis-SNPs [G_a, G_b, G_c] might be identified that impact upon DNA methylation at a particular site or across a particular genomic region. These could then be used as separate instruments to interrogate the relationship between methylation and phenotype