Commentary

Effect modification and the limits of biological inference from epidemiologic data

The relationship of exposure to benzo[a]pyrene (BaP) with lung cancer risk has been firmly established, but whether this association could be modified by other environmental or genetic factors remains to be explored. To investigate whether and how zinc (Zn) and genetic predisposition modify the association between BaP and lung cancer, we performed a case-cohort study with a 5.4-year median follow-up duration, comprising a representative subcohort of 1399 participants and 359 incident lung cancer cases. The baseline concentrations of benzo[a]pyrene diol epoxide-albumin adduct (BPDE-Alb) and Zn were quantified. We also genotyped the participants and computed the polygenic risk score (PRS) for lung cancer. Our findings indicated that elevated BPDE-Alb and PRS were linked to increased lung cancer risk, with the HR (95%CI) of 1.54 (1.36, 1.74) per SD increment in ln-transformed BPDE-Alb and 1.27 (1.14, 1.41) per SD increment in PRS, but high plasma Zn level was linked to a lower lung cancer risk [HR (95%CI)=0.77 (0.66, 0.91) per SD increment in ln-transformed Zn]. There was evidence of effect modification by Zn on BaP-lung cancer association (P for multiplicative interaction = 0.008). As Zn concentrations increased from the lowest to the highest tertile, the lung cancer risk per SD increment in ln-transformed BPDE-Alb decreased from 2.07 (1.48, 2.89) to 1.33 (0.90, 1.95). Additionally, we observed a significant synergistic interaction of BPDE-Alb and PRS [RERI (95%CI) = 0.85 (0.03, 1.67)], with 42% of the incident lung cancer cases among individuals with high BPDE-Alb and high PRS attributable to their additive effect [AP (95%CI) = 0.42 (0.14, 0.69)]. This study provided the first prospective epidemiological evidence that Zn has protective effect against BaP-induced lung tumorigenesis, whereas high genetic risk can enhance the harmful effect of BaP. These findings may provide novel insight into the environment-environment and environment-gene interaction underlying lung cancer development, which may help to develop prevention and intervention strategies to manage BaP-induced lung cancer.

Effect modification and interaction are important concepts for answering causal questions about interdependent effects of two (or more) exposures on some outcome of interest. Although conceptually alike and often mistakenly regarded as synonymous, effect modification and interaction actually refer to slightly different concepts when considered from a causal perspective. Their subtle yet relevant distinction lies in how the interplay between exposures is defined and the causal roles attributed to the exposures involved in the effect modification and interaction. To gain more insight into similarities and differences between the concepts of effect modification and interaction, the counterfactual theory of causation, albeit complicated, can be very helpful. Therefore, this article presents a nontechnical explanation of the counterfactual definition of effect modification and interaction. Essentially, effect modification and interaction are reflections of the reality and complexity of multicausality. The causal effect of an exposure of interest often depends on the levels of other exposures (effect modification) or causal effects of other exposures (interaction). Consequently, exposure effects should not be regarded in isolation but in combination. Understanding the underlying principles of effect modification and interaction on a conceptual level enables researchers to better anticipate, detect, and interpret these causal phenomena when setting up, analyzing, and reporting findings of (clinical) epidemiological studies. Effect modification and interaction are not biases to be avoided but properties of causal effects that ought to be unveiled. Hence, evidence for effect modification and interaction needs to be shown in order to delineate in whom and which instances causal effects occur.

“Quantile-dependent expressivity” occurs when the effect size of a genetic variant depends upon whether the phenotype (e.g. cholesterol) is high or low relative to its distribution. We have previously shown that the effect of a 52-SNP genetic-risk score was 3-fold larger at the 90th percentile of the total cholesterol distribution than at its 10th percentile. The objective of this study is to assess quantile-dependent expressivity for total cholesterol in 7006 offspring with parents and 2112 sibships from Framingham Heart Study.

Quantile-specific heritability (h²) was estimated as twice the offspring-parent regression slope as robustly estimated by quantile regression with nonparametric significance assigned from 1000 bootstrap samples.

Quantile-specific h² increased linearly with increasing percentiles of the offspring's cholesterol distribution (P = 3.0 × 10⁻⁹), i.e. h² = 0.38 at the 10th percentile, h² = 0.45 at the 25th percentile, h² = 0.52 at the 50th, h² = 0.61 at the 75th percentile, and h² = 0.65 at the 90th percentile of the total cholesterol distribution. Average h² decreased from 0.55 to 0.34 in 3564 offspring who started cholesterol-lowering medications, but this was attributable to quantile-dependent expressivity and the offspring's 0.94 mmol/L average drop in total cholesterol. Quantile-dependent expressivity likely explains the reported effect of the CELSR2/PSRC1/SORT1 rs646776 and APOE rs7412 gene loci on statin efficacy. Specifically, a smaller genetic effect size at the lower (post-treatment) than higher (pre-treatment) cholesterol concentrations mandates that the trajectories of the genotypes cannot move in parallel when cholesterol is decreased pharmacologically.

Cholesterol concentrations exhibit quantile-dependent expressivity, which may provide an alternative interpretation to pharmacogenetic and nutrigenetic interactions.

The contribution of gene-by-environment (GxE) interactions for many human traits and diseases is poorly characterized. We propose a Bayesian whole-genome regression model for joint modeling of main genetic effects and GxE interactions in large-scale datasets, such as the UK Biobank, where many environmental variables have been measured. The method is called LEMMA (Linear Environment Mixed Model Analysis) and estimates a linear combination of environmental variables, called an environmental score (ES), that interacts with genetic markers throughout the genome. The ES provides a readily interpretable way to examine the combined effect of many environmental variables. The ES can be used both to estimate the proportion of phenotypic variance attributable to GxE effects and to test for GxE effects at genetic variants across the genome. GxE effects can induce heteroskedasticity in quantitative traits, and LEMMA accounts for this by using robust standard error estimates when testing for GxE effects. When applied to body mass index, systolic blood pressure, diastolic blood pressure, and pulse pressure in the UK Biobank, we estimate that $9.3\mathit{\%}$, $3.9\mathit{\%}$, $1.6\mathit{\%}$, and $12.5\mathit{\%}$, respectively, of phenotypic variance is explained by GxE interactions and that low-frequency variants explain most of this variance. We also identify three loci that interact with the estimated environmental scores ($-\mathit{lo}{\mathit{g}}_{10}\mathrm{p}>7.3$).

The etiology of most complex diseases involves genetic variants, environmental factors, and gene-environment interaction (G × E) effects. Compared with marginal genetic association studies, G × E analysis requires more samples and detailed measure of environmental exposures, and this limits the possible discoveries. Large-scale population-based biobanks with detailed phenotypic and environmental information, such as UK-Biobank, can be ideal resources for identifying G × E effects. However, due to the large computation cost and the presence of case-control imbalance, existing methods often fail. Here we propose a scalable and accurate method, SPAGE (SaddlePoint Approximation implementation of G × E analysis), that is applicable for genome-wide scale phenome-wide G × E studies. SPAGE fits a genotype-independent logistic model only once across the genome-wide analysis in order to reduce computation cost, and SPAGE uses a saddlepoint approximation (SPA) to calibrate the test statistics for analysis of phenotypes with unbalanced case-control ratios. Simulation studies show that SPAGE is 33–79 times faster than the Wald test and 72–439 times faster than the Firth’s test, and SPAGE can control type I error rates at the genome-wide significance level even when case-control ratios are extremely unbalanced. Through the analysis of UK-Biobank data of 344,341 white British European-ancestry samples, we show that SPAGE can efficiently analyze large samples while controlling for unbalanced case-control ratios.

Body mass index (BMI) is criticized for not distinguishing fat from lean mass and ignoring fat distribution, leaving its ability to detect health effects unclear.

The aim of this study was to compare BMI with total and regional fat indexes from dual-energy x-ray absorptiometry in their associations with cardiometabolic traits. Duration of exposure to and change in each index across adolescence were examined in relation to detailed traits in young adulthood.

BMI was examined alongside total, trunk, arm, and leg fat indexes (each in kilograms per square meter) from dual-energy x-ray absorptiometry at ages 10 and 18 years in relation to 230 traits from targeted metabolomics at age 18 years in 2,840 offspring from the Avon Longitudinal Study of Parents and Children.

Higher total fat mass index and BMI at age 10 years were similarly associated with cardiometabolic traits at age 18 years, including higher systolic and diastolic blood pressure, higher very low-density lipoprotein and low-density lipoprotein cholesterol, lower high-density lipoprotein cholesterol, higher triglycerides, and higher insulin and glycoprotein acetyls. Associations were stronger for both indexes measured at age 18 years and for gains in each index from age 10 to 18 years (e.g., 0.45 SDs [95% confidence interval: 0.38 to 0.53] in glycoprotein acetyls per SD unit gain in fat mass index vs. 0.38 SDs [95% confidence interval: 0.27 to 0.48] per SD unit gain in BMI). Associations resembled those for trunk fat index. Higher lean mass index was weakly associated with traits and was not protective against higher fat mass index.

The results of this study support abdominal fatness as a primary driver of cardiometabolic dysfunction and BMI as a useful tool for detecting its effects.