Selecting instruments for Mendelian randomization in the wake of genome-wide association studies.
Swerdlow DI, Kuchenbaecker KB, Shah S, Sofat R, Holmes MV, White J, Mindell JS, Kivimaki M, Brunner EJ, Whittaker JC, Casas JP, Hingorani AD. · 2016
Read Selecting instruments for Mendelian randomization in the wake of genome-wide association studies. by Swerdlow DI, Kuchenbaecker KB, Shah S, Sofat R, Holmes MV, White J, Mindell JS, Kivimaki M, Brunner EJ, Whittaker JC, Casas JP, Hingorani AD. (2016) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5100611.txt.
Introduction
Daniel I. Swerdlow and a wide team of co-authors offered guidance in the International Journal of Epidemiology in 2016. Their subject is Mendelian randomization. That method uses gene variants as stand-ins for exposures such as blood markers, to test cause and effect. Such studies grew more than tenfold between 2004 and 2015, fed by genome-wide association studies.
Those big gene hunts now supply far more candidate variants than early studies had, and picking the wrong ones can mislead. The authors lay out what to weigh. They start with the method's core assumptions and the biology of the marker. They add how its variants are spread and how strong they are. They also ask whether a variant affects only the marker of interest. Their advice aims to help planners design cleaner studies and readers judge them.
Catalog record
| CREATOR | Swerdlow DI, Kuchenbaecker KB, Shah S, Sofat R, Holmes MV, White J, Mindell JS, Kivimaki M, Brunner EJ, Whittaker JC, Casas JP, Hingorani AD. |
|---|---|
| YEAR | 2016 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Selecting instruments for Mendelian randomization in the wake of genome-wide association studies By Daniel I Swerdlow, Karoline B Kuchenbaecker, Sonia Shah, Reecha Sofat, Michael V Holmes, Jon White, Jennifer S Mindell, Mika Kivimaki, Eric J Brunner, John C Whittaker, Juan P Casas, Aroon D Hingorani Published in: International Journal of Epidemiology DOI: 10.1093/ije/dyw088 ABSTRACT Abstract Mendelian randomization (MR) studies typically assess the pathogenic relevance of environmental exposures or disease biomarkers, using genetic variants that instrument these exposures. The approach is gaining popularity—our systematic review reveals a greater than 10-fold increase in MR studies published between 2004 and 2015. When the MR paradigm was first proposed, few biomarker- or exposure-related genetic variants were known, most having been identified by candidate gene studies. However, genome-wide association studies (GWAS) are now providing a rich source of potential instruments for MR analysis. Many early reviews covering the concept, applications and analytical aspects of the MR technique preceded the surge in GWAS, and thus the question of how best to select instruments for MR studies from the now extensive pool of available variants has received insufficient attention. Here we focus on the most common category of MR studies—those concerning disease biomarkers. We consider how the selection of instruments for MR analysis from GWAS requires consideration of: the assumptions underlying the MR approach; the biology of the biomarker; the genome-wide distribution, frequency and effect size of biomarker-associated variants (the genetic architecture); and the specificity of the genetic associations. Based on this, we develop guidance that may help investigators to plan and readers interpret MR studies. Keywords: Mendelian randomization, genome-wide association study, biomarkers, causal inference Key MessagesMR offers novel opportunities for reliable causal inference within the framework of observational research designs.The findings from an MR analysis can provide insight into the pathophysiology of complex disease and have translational relevance, including the prioritization of drug targets.The emerging genetic architecture of disease biomarkers now allows more informed selection of genetic variants for MR studies than was hitherto possible.As the number of biomarker-associated variants grows though genome-wide association studies and, more recently, metabolomics and proteomics, selection of the most appropriate instruments for MR analysis will become an increasingly important issue.We have proposed a set of principles that should inform the selection process to aid the design, analysis and interpretation of MR studies. INTRODUCTION Adverse environmental influences, such as smoking and alcohol consumption, are associated with a higher risk of many chronic, non-communicable diseases. Individuals at higher risk also exhibit alterations in numerous quantitative biological traits (also known as disease biomarkers or intermediate phenotypes), years before disease onset (Supplementary Table 1, available as Supplementary data at IJE online). These associations have been identified mainly through non-genetic observational studies. However, observational epidemiological studies of this type can be subject to a variety of biases. Importantly, it can be difficult to separate causal associations from those that arise from confounding or reverse causation. Effect estimates from such studies may also be prone to regression dilution bias1 and errors in the measurement of the biomarker for technical or biological reasons.1 Mendelian randomization (MR) is an evolving paradigm in which genetic variants (usually single nucleotide polymorphisms, SNPs) are used to help distinguish causal from non-causal associations between environmental exposures or biomarkers and disease outcomes.2 Two unique attributes of genotype make this possible. First, the random allocation of parental alleles to zygotes at meiosis, independent of environmental exposures, reduces the potential for confounding in genetic association studies in the same way as randomized treatment allocation in clinical trials3,4 (Figure 1a). Second, the invariant nature of the DNA sequence and unidirectional flow of biological information, from gene sequence through intermediate phenotypes to disease, avoids reverse causation, though it should not be taken to imply a stability of genetic effect which in theory could be modified in a context-dependent fashion.5 [Figure: A: Mendelian randomization is a natural analogue of the classical randomized controlled trial (RCT). Random allocation of alleles at conception and the unidirectional flow of information from DNA sequence to endogenous biomarker phenotype allow causal inference of the type possible within the RCT framework. Genotype is generally unrelated to environmental exposures, thus reducing confounding. B: the Mendelian randomization model: the causal role of an exposure, P, on a disease state, D, is being evaluated. A genetic variant, G, is associated with biomarker P but not with confounders, U. Variant G is also associated with disease D and acts only through its effects on biomarker P. The model rests on three core assumptions: (i) the genetic instrument (G) is associated with the exposure or biomarker of interest (P); (ii) the genetic instrument (G) is independent of potential confounding factors (U) in the relationship between the exposure/biomarker (P) and the outcome (D); (iii) the outcome (D) is associated with the genetic instrument (G) only through the effect of the exposure/biomarker (P), and is in all other respects independent.] An MR study typically considers three types of association: (i) the association of a biomarker (or environmental exposure) with the disease outcome; (ii) the association of a genetic variant with biomarker or environmental exposure; and (iii) the association of the same variant with disease risk6 (Figure 1b). Provided certain assumptions are met (Figure 1), consistency in direction and magnitude of the three estimates provides evidence on causal relevance of the environmental exposure or biomarker. The causal effect can be quantified within a formal statistical framework, using instrumental variables methods which have been adopted and adapted from the econometric literature.7,8 Some illustrative examples of the early use of MR are outlined in Box 1 and Table 1, and more recent examples that have exploited certain enhancements to the MR approach, are described in more detail later in this. It is notable that several important MR studies of certain disease biomarkers have identified inconsistency between effect estimates obtained in non-genetic observational studies and those through MR analysis that have altered thinking on the causal relevance of those biomarkers, as we describe later. BOX 1. Applications of Mendelian randomizationMR analysis has been applied to assess whether CRP, a circulating marker of inflammation, plays a true causal role in the development of CHD. Despite the robust association of CRP level with CHD in observational studies, CRP variants used to instrument long-term elevations in CRP concentration did not provide evidence of a causal role for this biomarker in the development of CHD, based on meta-analysis of up to 47 studies including 46 557 cases.78,82,125 The observational association between CRP and CHD is more likely explained by confounding or reverse causation. HDL-cholesterol (HDL-C) exhibits an inverse association with CHD risk in observational studies, but whether this association is causal has been in dispute. An MR study used variants in the LIPG gene, encoding hepatic lipase, as an instrument for HDL-C and examined its relationship with myocardial infarction (MI) risk.104 Although higher HDL-C is observationally associated with lower MI risk, MR analysis based on LIPG variants, both alone and within allele scores to instrument HDL-C concentration, did not find evidence for a causal role for HDL-cholesterol in CHD. [Table: Illustrative examples of different types of MR study. Examples are provided of MR studies of exogenous exposures, cis-MR for drug target validation, and disease biomarker MR analysis] A systematic