A Novel Test for Gene-Ancestry Interactions in Genome-Wide Association Data

Joanna L. Davies, Jean‐Baptiste Cazier, Malcolm G. Dunlop, Richard S. Houlston, Ian Tomlinson, Chris HolmesPLoS 2012 · DOI

Abstract

Genome-wide association study (GWAS) data on a disease are increasingly available from multiple related populations. In this scenario, meta-analyses can improve power to detect homogeneous genetic associations, but if there exist ancestry-specific effects, via interactions on genetic background or with a causal effect that co-varies with genetic background, then these will typically be obscured. To address this issue, we have developed a robust statistical method for detecting susceptibility gene-ancestry interactions in multi-cohort GWAS based on closely-related populations. We use the leading principal components of the empirical genotype matrix to cluster individuals into "ancestry groups" and then look for evidence of heterogeneous genetic associations with disease or other trait across these clusters. Robustness is improved when there are multiple cohorts, as the signal from true gene-ancestry interactions can then be distinguished from gene-collection artefacts by comparing the observed interaction effect sizes in collection groups relative to ancestry groups. When applied to colorectal cancer, we identified a missense polymorphism in iron-absorption gene CYBRD1 that associated with disease in individuals of English, but not Scottish, ancestry. The association replicated in two additional, independently-collected data sets. Our method can be used to detect associations between genetic variants and disease that have been obscured by population genetic heterogeneity. It can be readily extended to the identification of genetic interactions on other covariates such as measured environmental exposures. We envisage our methodology being of particular interest to researchers with existing GWAS data, as ancestry groups can be easily defined and thus tested for interactions.

Introduction

Genome-wide association studies (GWAS) frequently aggregate data from diverse populations to maximize statistical power. This approach typically assumes that genetic variants exert a uniform effect across different genetic backgrounds. However, if a variant's impact is specific to a particular ancestry—a gene-ancestry interaction—standard meta-analyses may fail to detect the signal.

A significant hurdle in identifying these interactions is the presence of "collection artifacts." These are technical variations introduced by differences in sample recruitment, handling, or genotyping across different geographic sites. To find genuine biological signals, researchers must distinguish true ancestry-driven heterogeneity from these site-specific biases.

Methods

The proposed methodology utilizes a multi-step pipeline designed to isolate heterogeneous genetic associations while filtering out technical noise.

Overview of the statistical pipeline for detecting gene-ancestry interactions.
  1. Defining Ancestry Space: Principal Component Analysis (PCA) is performed on genotype data to map the genetic relatedness of all individuals.
  2. Clustering: Individuals are assigned to "ancestry groups" using k-means clustering on the top principal components. This step often reveals that geographic collection labels (e.g., "Scottish collection") do not perfectly align with genetic ancestry.
  3. The D-Statistic: The core innovation is a comparison between two test statistics:
  4. $T_{anc}$: Evidence for interaction based on genetic ancestry clusters.
  5. $T_{coll}$: Evidence for interaction based on original geographic collection labels.
  6. $D = T_{anc} - T_{coll}$: A positive and significant $D$ suggests that the observed heterogeneity is driven by biological ancestry rather than technical artifacts associated with the collection site.
  7. Validation: Null simulations establish significance thresholds, followed by testing in independent replication cohorts.

Results

The method was applied to colorectal cancer (CRC) susceptibility using a discovery set of 13,690 individuals and two subsequent replication phases.

Discovery of the CYBRD1 Interaction

The analysis identified a missense polymorphism (rs10455) in the CYBRD1 gene, which regulates intestinal iron absorption. This variant was the only marker out of nearly 500,000 to consistently pass the $D$-statistic filter and replicate across independent phases.

Forest plot of odds ratios for the CYBRD1 variant across English and Scottish groups.

As shown in the forest plot, the variant demonstrates a significant protective effect in individuals of English ancestry (Odds Ratio ≈ 0.85–0.93). In contrast, no detectable association was found in individuals of Scottish ancestry (OR ≈ 1.09). In a meta-analysis restricted to English samples, the variant reached a significance level of $p \approx 3.9 \times 10^{-6}$, a signal that would have been diluted if the Scottish samples were included without accounting for the interaction.

Genetics vs. Geography

The study noted that approximately 18-21% of individuals were "misclassified" by their geographic collection labels relative to their genetic ancestry. By re-clustering individuals based on genetic data, the researchers significantly improved the signal-to-noise ratio, enabling the detection of the CYBRD1 interaction.

Discussion

While powerful, this method involves trade-offs. If a genetic effect is truly homogeneous across populations, clustering may result in lower statistical power compared to a standard meta-analysis. Additionally, the test cannot definitively distinguish between a true gene-ancestry interaction and a gene-environment interaction where the environmental factor (such as a regional diet) is perfectly correlated with ancestry.

Future work may involve extending this framework to probabilistic clustering to better accommodate individuals with admixed backgrounds. Furthermore, the $D$-statistic approach could be applied to other covariates, such as environmental exposures or drug response data, to uncover further hidden genetic interactions.

Summary updated

Abstract

Genome-wide association study (GWAS) data on a disease are increasingly available from multiple related populations. In this scenario, meta-analyses can improve power to detect homogeneous genetic associations, but if there exist ancestry-specific effects, via interactions on genetic background or with a causal effect that co-varies with genetic background, then these will typically be obscured. To address this issue, we have developed a robust statistical method for detecting susceptibility gene-ancestry interactions in multi-cohort GWAS based on closely-related populations. We use the leading principal components of the empirical genotype matrix to cluster individuals into "ancestry groups" and then look for evidence of heterogeneous genetic associations with disease or other trait across these clusters. Robustness is improved when there are multiple cohorts, as the signal from true gene-ancestry interactions can then be distinguished from gene-collection artefacts by comparing the observed interaction effect sizes in collection groups relative to ancestry groups. When applied to colorectal cancer, we identified a missense polymorphism in iron-absorption gene CYBRD1 that associated with disease in individuals of English, but not Scottish, ancestry. The association replicated in two additional, independently-collected data sets. Our method can be used to detect associations between genetic variants and disease that have been obscured by population genetic heterogeneity. It can be readily extended to the identification of genetic interactions on other covariates such as measured environmental exposures. We envisage our methodology being of particular interest to researchers with existing GWAS data, as ancestry groups can be easily defined and thus tested for interactions.
BibTeX
@inproceedings{davies2012novel,
  author = {Joanna L. Davies and Jean‐Baptiste Cazier and Malcolm G. Dunlop and Richard S. Houlston and Ian Tomlinson and Chris Holmes},
  title = {{A Novel Test for Gene-Ancestry Interactions in Genome-Wide Association Data}},
  booktitle = {PLoS},
  year = {2012},
  doi = {10.1371/journal.pone.0048687},
}
Figure