Choice of phenotype scale is critical in biobank-based GxE tests

Technical Summary: Choice of Phenotype Scale is Critical in Biobank-Based GxE Tests

Problem Statement
The significance of gene-environment interactions (GxE) for complex human traits remains a subject of intense debate. While recent genome-wide association studies (GWAS) utilizing biobank data have identified numerous statistically significant GxE signals, the majority of these findings lack clear evidence of biological significance. This discrepancy suggests that many reported interactions may be statistical artifacts rather than true biological phenomena. A primary, yet currently underappreciated, source of this issue is the choice of phenotype scale; specifically, many apparent GxE signals may simply represent additive effects that manifest differently depending on the mathematical transformation applied to the phenotype.

Methodology
The authors investigate the impact of phenotype scaling on GxE detection and interpretation using data from the UK Biobank. The study employs power transformations, including simple log transformations, to re-evaluate GxE signals across a diverse set of 32 phenotypes. The analysis focuses on distinguishing between true interaction effects and additive effects that are merely scale-dependent. The researchers specifically examine GxSex interactions, comparing results on raw scales versus transformed scales (e.g., log scale) to determine if the interaction persists or vanishes. Furthermore, the study assesses how these scaling choices influence GWAS discovery rates and the construction and evaluation of polygenic scores (PGS).

Key Contributions and Results
The paper provides empirical evidence that the choice of phenotype scale fundamentally alters the interpretation of GxE signals:

• Scale-Dependent Interactions: The study demonstrates that many GxE signals are not robust across scales. For instance, GxSex effects observed on height disappear entirely when the phenotype is analyzed on a log scale, suggesting the original signal was an artifact of the scale rather than a true interaction. Conversely, for testosterone, a log transformation reveals biologically meaningful female-specific effects that were not apparent on the raw scale.
• Quantitative Impact: Across the 32 phenotypes analyzed, the authors find that scaling by a power transformation can explain 46% of PGSxSex interactions, while a simple log transformation explains 23%. Similar patterns are observed for interactions with other environmental factors.
• Downstream Effects: The choice of scale is shown to substantially impact not only the detection of interactions but also the discovery of main effects in GWAS and the subsequent performance of polygenic scores.

Significance and Claims
The paper argues that the current underappreciation of phenotype scaling is a critical factor contributing to the discrepancy between statistical significance and biological relevance in GxE research. By showing that many interactions simplify to additive effects on alternative scales, the authors provide a mechanistic explanation for the proliferation of potentially spurious GxE signals in biobank studies.

The study concludes by offering a set of practical guidelines for researchers to consider when selecting phenotype scales in modern genetic studies. The authors emphasize that rigorous scale selection is not merely a technical detail but a fundamental requirement for distinguishing true biological interactions from statistical artifacts, thereby improving the reliability and interpretability of GxE findings.