Multivariable Mendelian randomization (MVMR) is a form of instrumental variable analysis which estimates the direct effect of multiple exposures on an outcome using genetic variants as instruments. ...Mendelian randomization and MVMR are frequently conducted using two‐sample summary data where the association of the genetic variants with the exposures and outcome are obtained from separate samples. If the genetic variants are only weakly associated with the exposures either individually or conditionally, given the other exposures in the model, then standard inverse variance weighting will yield biased estimates for the effect of each exposure. Here, we develop a two‐sample conditional F‐statistic to test whether the genetic variants strongly predict each exposure conditional on the other exposures included in a MVMR model. We show formally that this test is equivalent to the individual level data conditional F‐statistic, indicating that conventional rule‐of‐thumb critical values of F> 10, can be used to test for weak instruments. We then demonstrate how reliable estimates of the causal effect of each exposure on the outcome can be obtained in the presence of weak instruments and pleiotropy, by repurposing a commonly used heterogeneity Q‐statistic as an estimating equation. Furthermore, the minimized value of this Q‐statistic yields an exact test for heterogeneity due to pleiotropy. We illustrate our methods with an application to estimate the causal effect of blood lipid fractions on age‐related macular degeneration.
Abstract
Background
Mendelian randomization (MR) is a powerful tool in epidemiology that can be used to estimate the causal effect of an exposure on an outcome in the presence of unobserved ...confounding, by utilizing genetic variants that are instrumental variables (IVs) for the exposure. This has been extended to multivariable MR (MVMR) to estimate the effect of two or more exposures on an outcome.
Methods and results
We use simulations and theory to clarify the interpretation of estimated effects in a MVMR analysis under a range of underlying scenarios, where a secondary exposure acts variously as a confounder, a mediator, a pleiotropic pathway and a collider. We then describe how instrument strength and validity can be assessed for an MVMR analysis in the single-sample setting, and develop tests to assess these assumptions in the popular two-sample summary data setting. We illustrate our methods using data from UK Biobank to estimate the effect of education and cognitive ability on body mass index.
Conclusion
MVMR analysis consistently estimates the direct causal effect of an exposure, or exposures, of interest and provides a powerful tool for determining causal effects in a wide range of scenarios with either individual- or summary-level data.
Mendelian randomization (MR) is a powerful tool through which the causal effects of modifiable exposures on outcomes can be estimated from observational data. Most exposures vary throughout the life ...course, but MR is commonly applied to one measurement of an exposure (e.g. weight measured once between ages 40 and 60 years). It has been argued that MR provides biased causal effect estimates when applied to one measure of an exposure that varies over time.
We propose an approach that emphasizes the liability that causes the entire exposure trajectory. We demonstrate this approach using simulations and an applied example.
We show that rather than estimating the direct or total causal effect of changing the exposure value at a given time, MR estimates the causal effect of changing the underlying liability for the exposure, scaled to the effect of the liability on the exposure at that time. As such, results from MR conducted at different time points are expected to differ (unless the effect of the liability on exposure is constant over time), as we illustrate by estimating the effect of body mass index measured at different ages on systolic blood pressure.
Univariable MR results should not be interpreted as time-point-specific direct or total causal effects, but as the effect of changing the liability for the exposure. Estimates of how the effects of a genetic variant on an exposure vary over time, together with biological knowledge that provides evidence regarding likely effective exposure periods, are required to interpret time-point-specific causal effects.
Aims/hypothesis
Several studies have identified associations between type 2 diabetes and DNA methylation (DNAm). However, the causal role of these associations remains unclear. This study aimed to ...provide evidence for a causal relationship between DNAm and type 2 diabetes.
Methods
We used bidirectional two-sample Mendelian randomisation (2SMR) to evaluate causality at 58 CpG sites previously detected in a meta-analysis of epigenome-wide association studies (meta-EWAS) of prevalent type 2 diabetes in European populations. We retrieved genetic proxies for type 2 diabetes and DNAm from the largest genome-wide association study (GWAS) available. We also used data from the Avon Longitudinal Study of Parents and Children (ALSPAC, UK) when associations of interest were not available in the larger datasets. We identified 62 independent SNPs as proxies for type 2 diabetes, and 39 methylation quantitative trait loci as proxies for 30 of the 58 type 2 diabetes-related CpGs. We applied the Bonferroni correction for multiple testing and inferred causality based on
p
<0.001 for the type 2 diabetes to DNAm direction and
p
<0.002 for the opposing DNAm to type 2 diabetes direction in the 2SMR analysis.
Results
We found strong evidence of a causal effect of DNAm at cg25536676 (
DHCR24
) on type 2 diabetes. An increase in transformed residuals of DNAm at this site was associated with a 43% (OR 1.43, 95% CI 1.15, 1.78,
p
=0.001) higher risk of type 2 diabetes. We inferred a likely causal direction for the remaining CpG sites assessed. In silico analyses showed that the CpGs analysed were enriched for expression quantitative trait methylation sites (eQTMs) and for specific traits, dependent on the direction of causality predicted by the 2SMR analysis.
Conclusions/interpretation
We identified one CpG mapping to a gene related to the metabolism of lipids (
DHCR24
) as a novel causal biomarker for risk of type 2 diabetes. CpGs within the same gene region have previously been associated with type 2 diabetes-related traits in observational studies (BMI, waist circumference, HDL-cholesterol, insulin) and in Mendelian randomisation analyses (LDL-cholesterol). Thus, we hypothesise that our candidate CpG in
DHCR24
may be a causal mediator of the association between known modifiable risk factors and type 2 diabetes. Formal causal mediation analysis should be implemented to further validate this assumption.
Graphical Abstract
Mendelian randomization (MR) is the use of genetic variants associated with an exposure to estimate the causal effect of that exposure on an outcome. Mediation analysis is the method of decomposing ...the effects of an exposure on an outcome, which act directly, and those that act via mediating variables. These effects are decomposed through the use of multivariable analysis to estimate the causal effects between three types of variables: exposures, mediators, and an outcome. Multivariable MR (MVMR) is a recent extension to MR that uses genetic variants associated with multiple, potentially related exposures to estimate the effect of each exposure on a single outcome. MVMR allows for equivalent analysis to mediation within the MR framework and therefore can also be used to estimate mediation effects. This approach retains the benefits of using genetic instruments for causal inference, such as avoiding bias due to confounding, while allowing for estimation of the different effects required for mediation analysis. This review explains MVMR, what is estimated when one exposure is a mediator of another in an MVMR estimation, and how MR and MVMR can therefore be used to estimate mediated effects. This review then goes on to consider the advantages and limitations of using MR and MVMR to conduct mediation analysis.
Circulating lipoprotein lipids cause coronary heart disease (CHD). However, the precise way in which one or more lipoprotein lipid-related entities account for this relationship remains unclear. ...Using genetic instruments for lipoprotein lipid traits implemented through multivariable Mendelian randomisation (MR), we sought to compare their causal roles in the aetiology of CHD.
We conducted a genome-wide association study (GWAS) of circulating non-fasted lipoprotein lipid traits in the UK Biobank (UKBB) for low-density lipoprotein (LDL) cholesterol, triglycerides, and apolipoprotein B to identify lipid-associated single nucleotide polymorphisms (SNPs). Using data from CARDIoGRAMplusC4D for CHD (consisting of 60,801 cases and 123,504 controls), we performed univariable and multivariable MR analyses. Similar GWAS and MR analyses were conducted for high-density lipoprotein (HDL) cholesterol and apolipoprotein A-I. The GWAS of lipids and apolipoproteins in the UKBB included between 393,193 and 441,016 individuals in whom the mean age was 56.9 y (range 39-73 y) and of whom 54.2% were women. The mean (standard deviation) lipid concentrations were LDL cholesterol 3.57 (0.87) mmol/L and HDL cholesterol 1.45 (0.38) mmol/L, and the median triglycerides was 1.50 (IQR = 1.11) mmol/L. The mean (standard deviation) values for apolipoproteins B and A-I were 1.03 (0.24) g/L and 1.54 (0.27) g/L, respectively. The GWAS identified multiple independent SNPs associated at P < 5 × 10-8 for LDL cholesterol (220), apolipoprotein B (n = 255), triglycerides (440), HDL cholesterol (534), and apolipoprotein A-I (440). Between 56%-93% of SNPs identified for each lipid trait had not been previously reported in large-scale GWASs. Almost half (46%) of these SNPs were associated at P < 5 × 10-8 with more than one lipid-related trait. Assessed individually using MR, LDL cholesterol (odds ratio OR 1.66 per 1-standard-deviation-higher trait; 95% CI: 1.49-1.86; P < 0.001), triglycerides (OR 1.34; 95% CI: 1.25-1.44; P < 0.001) and apolipoprotein B (OR 1.73; 95% CI: 1.56-1.91; P < 0.001) had effect estimates consistent with a higher risk of CHD. In multivariable MR, only apolipoprotein B (OR 1.92; 95% CI: 1.31-2.81; P < 0.001) retained a robust effect, with the estimate for LDL cholesterol (OR 0.85; 95% CI: 0.57-1.27; P = 0.44) reversing and that of triglycerides (OR 1.12; 95% CI: 1.02-1.23; P = 0.01) becoming weaker. Individual MR analyses showed a 1-standard-deviation-higher HDL cholesterol (OR 0.80; 95% CI: 0.75-0.86; P < 0.001) and apolipoprotein A-I (OR 0.83; 95% CI: 0.77-0.89; P < 0.001) to lower the risk of CHD, but these effect estimates attenuated substantially to the null on accounting for apolipoprotein B. A limitation is that, owing to the nature of lipoprotein metabolism, measures related to the composition of lipoprotein particles are highly correlated, creating a challenge in making exclusive interpretations on causation of individual components.
These findings suggest that apolipoprotein B is the predominant trait that accounts for the aetiological relationship of lipoprotein lipids with risk of CHD.
Mendelian Randomisation (MR) is a powerful tool in epidemiology that can be used to estimate the causal effect of an exposure on an outcome in the presence of unobserved confounding, by utilising ...genetic variants as instrumental variables (IVs) for the exposure. The effect estimates obtained from MR studies are often interpreted as the lifetime effect of the exposure in question. However, the causal effects of some exposures are thought to vary throughout an individual's lifetime with periods during which an exposure has a greater effect on a particular outcome. Multivariable MR (MVMR) is an extension of MR that allows for multiple, potentially highly related, exposures to be included in an MR estimation. MVMR estimates the direct effect of each exposure on the outcome conditional on all the other exposures included in the estimation. We explore the use of MVMR to estimate the direct effect of a single exposure at different time points in an individual's lifetime on an outcome. We use simulations to illustrate the interpretation of the results from such analyses and the key assumptions required. We show that causal effects at different time periods can be estimated through MVMR when the association between the genetic variants used as instruments and the exposure measured at those time periods varies. However, this estimation will not necessarily identify exact time periods over which an exposure has the most effect on the outcome. Prior knowledge regarding the biological basis of exposure trajectories can help interpretation. We illustrate the method through estimation of the causal effects of childhood and adult BMI on C-Reactive protein and smoking behaviour.
Abstract
Objectives
To examine whether educational attainment and intelligence have causal effects on risk of Alzheimer’s disease (AD), independently of each other.
Design
Two-sample univariable and ...multivariable Mendelian randomization (MR) to estimate the causal effects of education on intelligence and vice versa, and the total and independent causal effects of both education and intelligence on AD risk.
Participants
17 008 AD cases and 37 154 controls from the International Genomics of Alzheimer’s Project (IGAP) consortium.
Main outcome measure
Odds ratio (OR) of AD per standardized deviation increase in years of schooling (SD = 3.6 years) and intelligence (SD = 15 points on intelligence test).
Results
There was strong evidence of a causal, bidirectional relationship between intelligence and educational attainment, with the magnitude of effect being similar in both directions OR for intelligence on education = 0.51 SD units, 95% confidence interval (CI): 0.49, 0.54; OR for education on intelligence = 0.57 SD units, 95% CI: 0.48, 0.66. Similar overall effects were observed for both educational attainment and intelligence on AD risk in the univariable MR analysis; with each SD increase in years of schooling and intelligence, odds of AD were, on average, 37% (95% CI: 23–49%) and 35% (95% CI: 25–43%) lower, respectively. There was little evidence from the multivariable MR analysis that educational attainment affected AD risk once intelligence was taken into account (OR = 1.15, 95% CI: 0.68–1.93), but intelligence affected AD risk independently of educational attainment to a similar magnitude observed in the univariate analysis (OR = 0.69, 95% CI: 0.44–0.88).
Conclusions
There is robust evidence for an independent, causal effect of intelligence in lowering AD risk. The causal effect of educational attainment on AD risk is likely to be mediated by intelligence.
In cross-platform analyses of 174 metabolites, we identify 499 associations (P < 4.9 × 10
) characterized by pleiotropy, allelic heterogeneity, large and nonlinear effects and enrichment for ...nonsynonymous variation. We identify a signal at GLP2R (p.Asp470Asn) shared among higher citrulline levels, body mass index, fasting glucose-dependent insulinotropic peptide and type 2 diabetes, with β-arrestin signaling as the underlying mechanism. Genetically higher serine levels are shown to reduce the likelihood (by 95%) and predict development of macular telangiectasia type 2, a rare degenerative retinal disease. Integration of genomic and small molecule data across platforms enables the discovery of regulators of human metabolism and translation into clinical insights.
In Mendelian randomization (MR) analysis, variants that exert horizontal pleiotropy are typically treated as a nuisance. However, they could be valuable in identifying alternative pathways to the ...traits under investigation. Here, we develop MR-TRYX, a framework that exploits horizontal pleiotropy to discover putative risk factors for disease. We begin by detecting outliers in a single exposure-outcome MR analysis, hypothesising they are due to horizontal pleiotropy. We search across hundreds of complete GWAS summary datasets to systematically identify other (candidate) traits that associate with the outliers. We develop a multi-trait pleiotropy model of the heterogeneity in the exposure-outcome analysis due to pathways through candidate traits. Through detailed investigation of several causal relationships, many pleiotropic pathways are uncovered with already established causal effects, validating the approach, but also alternative putative causal pathways. Adjustment for pleiotropic pathways reduces the heterogeneity across the analyses.