Candidate gene-environment interaction research: reflections and recommendations

Abstract

Studying how genetic predispositions come together with environmental factors to contribute to complex behavioral outcomes has great potential for advancing the understanding of the development of psychopathology. It represents a clear theoretical advance over studying these factors in isolation. However, research at the intersection of multiple fields creates many challenges. We review several reasons why the rapidly expanding candidate gene-environment interaction (cG×E) literature should be considered with a degree of caution. We discuss lessons learned about candidate gene main effects from the evolving genetics literature and how these inform the study of cG×E. We review the importance of the measurement of the gene and environment of interest in cG×E studies. We discuss statistical concerns with modeling cG×E that are frequently overlooked. Furthermore, we review other challenges that have likely contributed to the cG×E literature being difficult to interpret, including low power and publication bias. Many of these issues are similar to other concerns about research integrity (e.g., high false-positive rates) that have received increasing attention in the social sciences. We provide recommendations for rigorous research practices for cG×E studies that we believe will advance its potential to contribute more robustly to the understanding of complex behavioral phenotypes.

Keywords: G×E; candidate genes; genetics; gene–environment interaction.

Figure 1

Overview of common gene finding strategies. Results of a linkage analysis are often depicted using a Logarithm of Odds (LOD) score plot that depicts the genomic region(s) (measured in genetic distance, or centiMorgans, where 1cM roughly equals 1,000,000 base pairs) with the linkage peak(s), or the highest LOD scores. The LOD is the ratio of the likelihood that there is excess allele sharing to the null hypothesis of no excess allele sharing. The adjacent diagram illustrates these results for one chromosome. Elevated LOD scores indicate that the genomic region is shared by affected relative pairs more often than expected by chance alone, suggesting there is a gene in the region contributing to the outcome under study. Results from a classical candidate gene study are illustrated. The prevalence of psychopathology increases in an additive fashion with increasing copies of the risk allele “G”. The adjacent figure illustrates the results of a genome-wide association study (GWAS). Each dot represents the negative logarithm (base 10) of the p-value for an individual association test (usually hundreds of thousands or millions of SNPs tested across the genome). Therefore, a p-value of 5 × 10⁻⁸, or the threshold for genome-wide significance, as denoted by the horizontal solid line, is noted at the midway point between 7 and 8. The dotted line reflects a p-value of 1 × 10⁻⁵, indicating SNPs of interest. The x-axis denotes physical positions (in base pairs) across each of the 22 autosomal chromosomes. In the hypothetical example, there are three “hits” (SNPs) that surpass the genome-wide significance threshold, and many more SNPs that are suggestive.

Figure 2

Alternative portrayals of a candidate genotype (in this example a single nucleotide polymorphism, SNP) × environment (in this example high versus low adversity) interaction. Panel A emphasizes that the strongest impact of high adversity on psychopathology is found for individuals with the GG genotype. Alternatively, Panel B emphasizes that the strongest association of the genotype with psychopathology is among those experiencing high adversity. These are alternate presentations of the same interaction.

Figure 3

The figure presents simulated phenotypic data for three genotypic groups (G=0, 1, 2, indicating groups of individuals who carry 0, 1, or 2 copies of a particular allele) each shown in a different color. The four parameter model corresponds to the case where the interaction term is modeled by a cross-product term only. Although a significant interaction is detected, the corresponding linear regression lines do not match the data points and the slopes are incorrectly ordered from 0 to 1 to 2, based on the constraints imposed by the use of the cross-product term to model the interaction. Thus, although the model would produce a “significant interaction”, the regression lines implied by the model inaccurately represent the data and would be misleading as to the nature of the interaction. The data can be accurately reproduced by an extended parameterization of the regression model (six parameter model) as detailed in (Aliev, Latendresse, Bacanu, Neale, & Dick, 2014).

Figure 4

Recommendations for rigorous GxE research practices.