Unidirectional and bidirectional causation between smoking and blood DNA methylation: evidence from twin-based Mendelian randomisation

Abstract

Cigarette smoking is associated with numerous differentially-methylated genomic loci in multiple human tissues. These associations are often assumed to reflect the causal effects of smoking on DNA methylation (DNAm), which may underpin some of the adverse health sequelae of smoking. However, prior causal analyses with Mendelian Randomisation (MR) have found limited support for such effects. Here, we apply an integrated approach combining MR with twin causal models to examine causality between smoking and blood DNAm in the Netherlands Twin Register (N = 2577). Analyses revealed potential causal effects of current smoking on DNAm at > 500 sites in/near genes enriched for functional pathways relevant to known biological effects of smoking (e.g., hemopoiesis, cell- and neuro-development, and immune regulation). Notably, we also found evidence of reverse and bidirectional causation at several DNAm sites, suggesting that variation in DNAm at these sites may influence smoking liability. Seventeen of the loci with putative effects of DNAm on smoking showed highly specific enrichment for gene-regulatory functional elements in the brain, while the top three sites annotated to genes involved in G protein-coupled receptor signalling and innate immune response. These novel findings are partly attributable to the analyses of current smoking in twin models, rather than lifetime smoking typically examined in MR studies, as well as the increased statistical power achieved using multiallelic/polygenic scores as instrumental variables while controlling for potential horizontal pleiotropy. This study highlights the value of twin studies with genotypic and DNAm data for investigating causal relationships of DNAm with health and disease.

Keywords: Causal inference; DNA Methylation; Epigenetics; Mendelian Randomisation; Smoking; Twin modelling.

Fig. 1

Study Design. Overview of the data and MR-DoC models used to examine the causality between cigarette smoking and blood DNA methylation (DNAm) in the Netherlands Twin Register. The models were fitted separately for current (versus never) and former (versus never) smoking. Applying the five MR-DoC models shown in the path diagrams, we obtained a set of three causal estimates in each direction of causation: Smoking (Smk) → DNAm (the blue paths labelled g₁) and DNAm → Smoking (the red paths labelled g₂). In each MR-DOC model, the residual variance of each phenotype (smoking status liability and DNAm levels) is decomposed into latent additive genetic (A) and unique environmental (E) factors. The correlation between the latent A factors of smoking and DNAm (rA) represents confounding due to additive genetic factors, while that between the latent E factors (rE) represents confounding due to unique environmental factors. Note that these models did not include shared environmental (C) variance components, as the AE model was found to be the most parsimonious in univariate twin models (see Supplementary Methods). Note. For better readability, the path diagrams show only the within-individual part of the models fitted to data from twin pairs. The squares/rectangles indicate observed variables, the circles indicate latent (unobserved) variables, the single-headed arrows indicate regression paths, and the double-headed curved arrows indicate (co-)variances.

Fig. 2

Selection of CpGs tested in each MR-DoC model. Previous independent EWAS meta-analysis of cigarette smoking [2] examined DNA methylation (DNAm) at CpGs from the Illumina HumanMethylation450 BeadChip array [22], which was also used to measure DNAm in the NTR biobank. In the unidirectional MR-DoC1 models for Smoking → DNAm, we included autosomal CpGs associated with smoking in the EWAS meta-analysis that also passed the QC metrics in NTR. The MR-DoC1 models for DNAm → Smoking and the bidirectional MR-DoC2 models were restricted to a subset of these sites having cis-mQTL summary statistics from the GoDMC [21] and a resulting mQTL allelic score with F-statistic > 10.

Fig. 3

Putative Causal Effects of Current Smoking on Blood DNA Methylation in MR-DoC Models. The top panel shows an UpSet plot of the intersection of CpGs with statistically significant (FDR < 0.05) estimates of Current Smoking → DNAm in the three MR-DoC models. The matrix consists of the models along the three rows and their intersections along the columns. The horizontal bars on the left represent the number of CpGs with significant (FDR < 0.05) causal estimates in each model. The vertical bars represent the number of CpGs belonging to the respective intersection in the matrix. A similar UpSet plot with Bonferroni correction is shown in Supplementary Fig. S7. The bottom panel shows a Miami plot of the Current Smoking → DNAm causal estimates across 16,940 smoking-associated CpGs. The X-axis shows the genomic positions of the CpGs aligned to Genome Reference Consortium Human Build 37 (GRCh37). The Y-axis shows the Z-statistic of the estimated effect of the liability for current (versus never) smoking on (residualised and standardised) DNA methylation β-values in the MR-DoC1 model with unique environmental confounding (rE). The solid points indicate the 64 sites with significant causal estimates (FDR < 0.05) in all three models (i.e., the blue vertical bar in the UpSet plot). The 14 CpGs with causal estimates significant after Bonferroni correction in more than one model are labelled by their respective nearest gene. Note. The data underlying these plots are in Supplementary Table S1

Fig. 4

Potential reverse and bidirectional effects of blood DNA methylation on current smoking. A Estimates and Wald-type 95% confidence intervals of DNAm → Current Smoking causal effects in each of the three MR-DoC models: bidirectional MR-DoC2, MR-DoC1 with horizontal pleiotropic path, and MR-DoC1 with unique environmental confounding (rE). B An UpSet plot of the intersection of CpGs with statistically significant (FDR < 0.05) estimates of DNAm → Current Smoking in each of the three MR-DoC models. The matrix consists of the models along the three rows and their intersections along the columns. The horizontal bars on the left represent the number of CpGs with significant (FDR < 0.05) causal estimates in each model. The vertical bars represent the number of CpGs belonging to the respective intersection in the matrix. A similar UpSet plot with Bonferroni correction is shown in Supplementary Fig. S8 for comparison. C Estimates and Wald-type 95% confidence intervals of bidirectional causal effects between current smoking and DNA methylation in the three MR-DoC models. In panels A and C, the Y-axis labels indicate the CpG probe IDs and the respective genes in which the CpGs are located. Note. The numerical data underlying these plots are in Supplementary Tables S1-S4.

Fig. 5

Among the CpGs with potential effects of blood DNA methylation on current smoking liability, iterative eFORGE analyses elucidated sites enriched for overlap with brain-related chromatin states and histone marks. The first iteration of eFORGE examined the 64 CpGs with potential effects of blood DNA methylation on current smoking liability (Supplementary Fig. S15), revealing 21 CpGs enriched for overlap with enhancers in the brain (Supplementary Fig. S18/Table S12). In follow-up analyses restricted to these 21 CpGs (eFORGE iteration 2), all 21 probes were also enriched for the brain H3K4me1 marks, while 17 of these probes overlapped with H3K4me3 marks in the brain (Supplementary Fig. S22/Table S16). This iteration also showed significant enrichment (FDR q < 0.01) for histone marks in other tissues, including small and large intestines, adrenal gland, and thymus. So, to identify a subset of these CpGs with potentially more specific enrichment for brain-related functional elements, we restricted further analyses to the 17 sites overlapping with the brain H3K4me3 marks (eFORGE iteration 3). This figure shows that these 17 sites showed highly specific enrichment for enhancers and histone marks in the brain (Supplementary Tables S18, S19). Ten of these sites also overlapped with DNase-I hotspots in the brain (Supplementary Table S20).

Fig. 6

Putative causal effects of former smoking on blood DNA methylation. Estimates and Wald-type 95% confidence intervals of the causal effects of the liability for former (versus never) smoking and (residualised and standardised) DNA methylation beta-values in each of the three MR-DoC models: bidirectional MR-DoC2, MR-DoC1 with horizontal pleiotropic path, and MR-DoC1 with unique environmental confounding (rE). The corresponding estimates for current (versus never) smoking are also shown with dashed lines. The text labels on the left indicate the CpG probe IDs and the genes mapped by the CpGs. Note. The data underlying these plots are in Supplementary Tables S1 and S27 indicated by the column g1_robust.