A generalized pipeline for developing and comparing interaction-based, variance-based, and standard polygenic scores identified significant PGS-by-BMI interactions for 16/20 cardiometabolic risk factors, most consistently using the interaction PGS approach, with many interactions replicating in All of Us.
Key Findings
Results
The interaction PGS (iPGS) approach most consistently detected significant PGS-by-BMI interactions across cardiometabolic risk factors in UK Biobank.
Significant PGS×BMI interactions were identified for 16 out of 20 cardiometabolic risk factors tested.
The iPGS approach outperformed both standard PGSs and variance PGSs (vPGSs) in detecting PGS×exposure interactions.
The pipeline was applied to UK Biobank as the primary discovery cohort.
Three PGS types were systematically compared: standard PGSs, interaction-based iPGSs, and variance-effect-based vPGSs.
Results
Many PGS×BMI interactions identified in UK Biobank replicated in the All of Us (AoU) research program.
Replication was assessed in the All of Us cohort, which has greater ancestral diversity than UK Biobank.
A 72% larger BMI-alanine aminotransferase association was observed in the top iPGS decile in AoU.
The replication in AoU supports the generalizability of the identified gene-by-adiposity interactions across cohorts.
Background
PGSs based on interactions (iPGSs) and variance effects (vPGSs) were hypothesized to be more powerful than standard PGSs for detecting PGS×exposure interactions.
Standard PGSs are typically constructed to predict trait levels, not differential responses to exposures.
iPGSs are built directly from SNP-by-exposure interaction effects.
vPGSs leverage variance-quantitative trait loci (vQTLs) as proxies for interaction effects.
Prior to this study, these three PGS types had not been systematically compared for PGS×E detection.
Methods
The study developed a generalized pipeline for constructing and comparing multiple PGS types in the context of gene-by-environment interaction detection.
The pipeline encompasses development of standard PGSs, iPGSs, and vPGSs.
The pipeline was applied to the relationship between adiposity (measured by BMI) and a broad set of cardiometabolic risk factors.
Twenty cardiometabolic risk factors were analyzed.
The framework is described as applicable to future efforts toward clinically useful response-focused PGSs.
Results
Genetic modification of the adiposity-cardiometabolic risk factor relationship was detectable using observational data via PGS×exposure interaction analyses.
PGS×BMI interactions were detectable as observational signals consistent with genetic modification of the BMI-risk factor relationship.
16 of 20 cardiometabolic risk factors showed significant PGS×BMI interactions in UK Biobank.
The analysis supports the premise that PGS×E interactions in observational datasets can identify individuals with differential responses to exposures such as adiposity.
Results
The iPGS for alanine aminotransferase (ALT) showed a particularly strong interaction with BMI, with the top decile showing a 72% larger BMI-ALT association in All of Us.
The BMI-ALT association was 72% larger in the top iPGS decile compared to the reference group in AoU.
This finding exemplifies the clinical relevance of response-focused PGSs for liver-related metabolic risk.
ALT is a marker of liver function and hepatic steatosis, which is closely related to adiposity.