What is gene-environment equivalence?

The intuition behind the introduction of MR was that germline genetic variants reliably mimicking the effects of an exposure can contribute to assessing the evidence for causality, strengthening the translation of observational findings into successful interventions [1]. Many of the early examples of MR involved genetic variants that modified the effect of an exposure on a disease, without actually relating to the exposure under investigation [1,6]. In such situations the variant is clearly not an IV—the first IV assumption (Fig 1A) is violated—yet provides powerful evidence regarding the causal effect of the exposure under investigation. Currently, however, the large majority of MR studies are implemented within an IV framework, in which the genetic variant reliably associates with differences in levels of an exposure. In this setting, G-EE would be seen if the effect on the disease of interest generated by the genetic variants matches the effect that would be seen if the same exposure difference were produced through certain environmental changes (including in lifestyle, treatment and modifiable endogenous factors). This is G-EE, and whilst it has been implicitly [1] or explicitly [2,6–8] discussed in the literature since the introduction of MR, it has been neglected in comparison to the IV assumptions.

G-EE has its origins in clinical, ecological, and developmental genetics [6,9,10]. For example, G-EE is assumed by the breeder’s equation, which extrapolates an estimate of the selection coefficient to make a prediction regarding trait change in the next generation [11]. G-EE can also be seen in the phenotypic similarity of the recessive genetic condition Hartnup’s syndrome and the disease pellagra: both involve characteristic skin rashes, neurological symptoms, gastrointestinal problems and an episodic course. Clinically, both would be in a differential diagnosis for a patient presenting with such symptoms. Their mechanistic overlap is that in Hartnup’s syndrome genetic variation leading to low niacin activity drives the condition; in pellagra, dietary niacin deficiency is the cause [6].

Such mechanistic overlap is the core of G-EE. Cutting across this are the dimensions of timing and magnitude of exposure. The same exposure perturbation can have different effects depending on the stage during the life course in which it occurs. During development, many irreversible changes occur, and compensation for insults—be they environmental or genetic—can be entrained. This highlights the much-neglected issue of canalization—whereby biological compensation during development attenuates the effects of genetic variation or environmental changes on a trait, potentially compromising MR interpretations [1]. Furthermore, genetic perturbations are often much smaller than those produced by, e.g., pharmaceutical interventions. Thus, predicted effects from MR are generally estimated from linear extrapolation from a narrow actual range of demonstrable associations. These biological complexities are often difficult to express and evaluate solely within the IV framework, and designing MR studies with G-EE in mind is important for appropriately integrating their findings within evidence triangulation [8].

Examples of gene-environment equivalence in MR

An early MR example used variants in the MTHFR gene, which encodes an enzyme in folate-dependent one-carbon metabolism; maternal variants that reduce its activity alter folate-cycle intermediates (including pre-conceptually and during pregnancy) and are associated with increased neural tube defect risk in offspring [1,12]. The effect of lower antenatal folate levels on NTD risk is supported by randomized controlled trials (RCTs) of supplementation starting before conception or during early pregnancy. For the conceptus, it is irrelevant whether the difference in maternal folate activity were generated by folate supplementation or by different maternal MTHFR alleles; i.e., there is G-EE. As the exposure period for offspring is only during their gestation, genetic variation and the intervention would affect them over similar time periods (Fig 1B).

A second early example relates to cholesterol and coronary heart disease (CHD) [1]. One study reported that heterozygous carriers of rare large effect variants mostly in the LDLR gene region (which encodes a receptor that helps clear cholesterol from the blood) had higher risk of CHD [13]; this was twice the magnitude of the effects of cholesterol on CHD predicted by RCTs of statins [1]. However, while the genetic variants reported in this study are known to produce differences in cholesterol levels that can be seen from birth onwards, trials often only last ≤5 years [5]. G-EE is partial in relation to duration of exposure, generating the observed difference in effect size (Fig 1C).

Heritable confounding and gene-environment equivalence

A common issue in MR studies and G-EE is heritable confounding, whereby genetic factors influence both an exposure and an outcome. Consider the link between C-reactive protein (CRP) and cardiometabolic disease [7]. CRP is an acute phase reactant, elevated by many causes of physiological dysregulation, e.g., cigarette smoking, obesity, atherosclerosis and infection. CRP strongly associates with CHD risk, and pharmaceutical manipulation of CRP was being considered as an intervention; however, MR studies using cis variants in the CRP gene suggested that circulating CRP levels had no causal effect on CHD, discouraging the development of such drugs [2]. Genome wide association studies (GWAS), however, now identify hundreds of genetic variants influencing CRP levels, many working through obvious upstream factors, such as obesity or cigarette smoking [7]. Such genetic variants would not have effects equivalent to the pharmaceutical manipulation of circulating CRP, since obesity and smoking will influence health outcomes entirely independently of their effects on CRP; G-EE is clearly invalid for these variants. As GWAS studies become ever larger and identify genetic variants of very small effect size working upstream of the exposure of interest, this becomes a general issue for MR [7]. In some situations, where the important heritable confounders are known, multivariable MR may be able to recover sensible estimates [7], but in other contexts such heritable confounding contributes to the tsunami of published nonsensical MR studies [8]. Consider the recent stream of papers performing MR of breakfast skipping as the exposure and a vast array of outcomes. It is not plausible that a genetic variant can directly influence breakfast skipping as its principal effect; rather, breakfast skipping will be downstream of a series of processes that will likely have effects through other pathways (Fig 1D). In such circumstances, G-EE is very unlikely to hold.

The plausibility of gene-environment equivalence

The STROBE-MR reporting guidelines [14] state that G-EE should be justified in any MR study. This is clearly not a binary situation, in which there is either complete equivalence or none at all; there will be degrees of reasonableness of the principle. Evaluation should consider the mechanisms through which germline genetic variants influence the exposure, timing over the life course of the exposure, potential dose-response relationships, and cell type and tissue location of genetic effects. Reviewers and editors should consider this evaluation when deciding whether a paper deserves consideration for publication.

The emphasis on “certain” environmental modifications in our introduction of G-EE above indicates that the range of modifications being considered for each question needs to be carefully considered. For example, reduced levels of adiposity can be caused by many factors, including infections, cancers, endocrine disorders, cigarette smoking, alcoholism, severe anorexia nervosa, depression or environmental poisons. For many of these examples, genetic proxies exist, but these would not be equivalent to any actual or hypothetical interventions that would be implemented to reduce obesity levels and their consequent health burdens.

In the context of many contemporary MR studies, the notion of there being specific genetic influences on many of the exposures examined is simply implausible [8]. If attempts to construct G-EE stories lead to researcher reflexivity it could go a long way to stemming the tide of noncontributory (and often paper-mill produced) MR publications [3,4,8], and free up the time of scientists, reviewers, and editors to focus on the issues to which MR could be contributory. The reconnection with biology, and moving on from mindless application of IV analysis, could return MR to being a useful contributor to medical knowledge [8].