Testing Association with Interactions by Partitioning Chi‐Squares

Genome-wide association studies typically search for marginal associations between a single-nucleotide polymorphism (SNP) and a disease trait while gene-environment (G × E) interactions remain generally unexplored. More powerful methods beyond the simple case-control (CC) approach leverage either marginal effects or CC ascertainment to increase power. However, these potential gains depend on assumptions whose aptness is often unclear a priori. Here, we review G × E methods and use simulations to highlight performance as a function of main and interaction effects and the association of the two factors in the source population. Substantial variation in performance between methods leads to uncertainty as to which approach is most appropriate for any given analysis. We present a framework that (a) balances the robustness of a CC approach with the power of the case-only (CO) approach; (b) incorporates main SNP effects; (c) allows for incorporation of prior information; and (d) allows the data to determine the most appropriate model. Our framework is based on Bayes model averaging, which provides a principled statistical method for incorporating model uncertainty. We average over inclusion of parameters corresponding to the main and G × E interaction effects and the G-E association in controls. The resulting method exploits the joint evidence for main and interaction effects while gaining power from a CO equivalent analysis. Through simulations, we demonstrate that our approach detects SNPs within a wide range of scenarios with increased power over current methods. We illustrate the approach on a gene-environment scan in the USC Children's Health Study.

Early causal heterogeneity but developmental homogeneity before schizophrenia. A prospective study of early life in 15000 people from two general population birth cohorts

Genome-wide association studies (GWAS) and rare variant association studies (RVAS) are applied across many areas of complex disease to analyze variation in whole genomes of thousands of unrelated patients. These approaches are able to identify variants and/or biological pathways which are associated with disease status and, in contrast to traditional linkage studies or candidate gene approaches, do so without requiring multigenerational affected families, prior hypotheses, or known genes of interest. However, the novel associations identified by these methods typically have lower effect sizes than those found in classical family studies. In the motor neuron disease amyotrophic lateral sclerosis (ALS), GWAS, and RVAS have been used to identify multiple disease-associated genes but have not yet resulted in novel therapeutic interventions. There is significant urgency within the ALS community to identify additional genetic markers of disease to uncover novel biological mechanisms, stratify genetic subgroups of disease, and drive drug development. Given the widespread and increasing application of genetic association studies of complex disease, it is important to recognize the strengths and limitations of these approaches. Here, we review ALS gene discovery via GWAS and RVAS.

Genome wide association studies (GWAS) of type 2 diabetes (T2D) have identified hundreds of genetic variants. However, the majority of these studies fail to consider the multifactorial nature and the interplay of genetic and environmental factors on disease risk, e.g., gene x environment interaction analyses. The role of smoking as a modulator in the association between genetic variants and T2D is of particular importance, since it is a modifiable risk factor for T2D. We performed a large-scale gene x smoking analysis on 345,955 individuals from 14 studies retrieved from dbGaP and UK Biobank. Phenotypic data was reviewed and harmonized prior to performing a gene by smoking GWAS for each study. A joint (main and interaction effect) framework was performed for each study. Individual study GWAS statistics were combined through meta-analysis, stratified by racial/ethnic group. The study sample includes 40,994 diabetics and 34,764 current smokers, and 324,834 of European-, 9,040 of African-, 6,125 of Hispanic-, and 5,956 of Asian-ancestry subjects. Results from the meta-analysis in the European ancestry group revealed a genome wide significant signal (p<5x10-8) in the TCF7L2 gene (main effect, p-value 9.59x10-33) and several genome-suggestive (p<1x10-5) loci. We are currently expanding the analyses to include data from a total of 20 studies and 700,000 subjects. Gene by smoking interaction analysis on T2D underscores the importance of both genetic and environmental risk factors and has the potential to unveil novel variants, genes, and biological pathways that influence disease vulnerability typically missed by traditional analytic approaches. Disclosure S. N. Ferdos: None. R. Salem: None. Funding Tobacco Related Disease Research Program (T29IR0770)

Genetic risk variants that have been identified in genome-wide association studies of complex diseases are primarily non-coding1. Translating these risk variants into mechanistic insights requires detailed maps of gene regulation in disease-relevant cell types2. Here we combined two approaches: a genome-wide association study of type 1 diabetes (T1D) using 520,580 samples, and the identification of candidate cis-regulatory elements (cCREs) in pancreas and peripheral blood mononuclear cells using single-nucleus assay for transposase-accessible chromatin with sequencing (snATAC-seq) of 131,554 nuclei. Risk variants for T1D were enriched in cCREs that were active in T cells and other cell types, including acinar and ductal cells of the exocrine pancreas. Risk variants at multiple T1D signals overlapped with exocrine-specific cCREs that were linked to genes with exocrine-specific expression. At the CFTR locus, theT1D risk variant rs7795896 mapped to a ductal-specific cCRE that regulated CFTR; the risk allele reduced transcription factor binding, enhancer activity and CFTR expression in ductal cells. These findings support a role for the exocrine pancreas in the pathogenesis of T1D and highlight the power of large-scale genome-wide association studies and single-cell epigenomics for understanding the cellular origins of complex disease.

Genetic linkage analysis has been a traditional means for identifying regions of the genome with large genetic effects that contribute to a disease. Following linkage analysis, association studies are widely pursued to fine-tune regions with significant linkage signals. For complex diseases which often involve function of multi-genetic variants each with small or moderate effect, linkage analysis has little power compared to association studies. In this chapter, we give a brief review of design issues related to linkage analysis and association studies with human genetic data. We introduce methods commonly used for linkage and association studies and compared the relative merits of the family-based and population-based association studies. Compared to candidate gene studies, a genomewide blind searching of disease variant is proving to be a more powerful approach. We briefly review the commonly used two-stage designs in genome-wide association studies. As more and more biological evidences indicate the role of genomic imprinting in disease, identifying imprinted genes becomes critically important. Design and analysis in genetic mapping imprinted genes are introduced in this chapter. Recent efforts in integrating gene expression analysis and genetic mapping, termed expression quantitative trait loci (eQTLs) mapping or genetical genomics analysis, offer new prospect in elucidating the genetic architecture of gene expression. Designs in genetical genomics analysis are also covered in this chapter.

There are several non‐functional‐form‐based interaction tests for testing additivity in unreplicated two‐way layouts. The distinctness of the various tests in this setting is somewhat unfortunate, as no single test can detect all patterns of possible interaction, and the tests are sensitive to a particular pattern of interaction. Therefore, it is reasonable to combine multiple interaction tests for testing additivity. The main problem is how we can combine multiple dependent tests into a single test such that it has an acceptable performance in terms of controlling the type I error rate and maximizing the power. In this paper, we review six methods for combining interaction tests which utilize six different non‐functional‐form‐based interaction tests. We also propose two novel combination methods based on the Jacobi polynomial expansion and the Gaussian copula. We first compare these eight combination methods in terms of controlling the type I error rate via a simulation study. We then compare the six interaction tests and four combination methods in terms of power. The simulation results show that our proposed combination method that is based on the Jacobi polynomial expansion has the best performance in terms of controlling the type I error rate and maximizing the power. Therefore, we recommend this combination method as a test procedure for testing additivity. The major contribution of this work is to provide researchers with a testing approach which leverages many existing methods to detect disparate forms of non‐additivity. This alleviates the need for researchers to naively choose among highly specialized methods by instead allowing for a general approach that captures a number of interaction effects with reasonable power.

Despite the recent success of genome-wide association studies (GWASs) in identifying loci consistently associated with coronary artery disease (CAD), a large proportion of the genetic components of CAD and its metabolic risk factors, including plasma lipids, type 2 diabetes and body mass index, remain unattributed. Gene-gene and gene-environment interactions might produce a meaningful improvement in quantification of the genetic determinants of CAD. Testing for gene-gene and gene-environment interactions is thus a new frontier for large-scale GWASs of CAD. There are several anecdotal examples of monogenic susceptibility to CAD in which the phenotype was worsened by an adverse environment. In addition, small-scale candidate gene association studies with functional hypotheses have identified gene-environment interactions. For future evaluation of gene-gene and gene-environment interactions to achieve the same success as the single gene associations reported in recent GWASs, it will be important to pre-specify agreed standards of study design and statistical power, environmental exposure measurement, phenomic characterization and analytical strategies. Here we discuss these issues, particularly in relation to the investigation and potential clinical utility of gene-gene and gene-environment interactions in CAD.

This Month in Genetics

Gene-gene interaction plays an important role in the etiology of complex diseases, which may exist without a genetic main effect. Most current statistical approaches, however, focus on assessing an interaction effect in the presence of the gene's main effects. It would be very helpful to develop methods that can detect not only the gene's main effects but also gene-gene interaction effects regardless of the existence of the gene's main effects while adjusting for confounding factors. In addition, when a disease variant is rare or when the sample size is quite limited, the statistical asymptotic properties are not applicable; therefore, approaches based on a reasonable and applicable computational framework would be practical and frequently applied. In this study, we have developed an extended support vector machine (SVM) method and an SVM-based pedigree-based generalized multifactor dimensionality reduction (PGMDR) method to study interactions in the presence or absence of main effects of genes with an adjustment for covariates using limited samples of families. A new test statistic is proposed for classifying the affected and the unaffected in the SVM-based PGMDR approach to improve performance in detecting gene-gene interactions. Simulation studies under various scenarios have been performed to compare the performances of the proposed and the original methods. The proposed and original approaches have been applied to a real data example for illustration and comparison. Both the simulation and real data studies show that the proposed SVM and SVM-based PGMDR methods have great prediction accuracies, consistencies, and power in detecting gene-gene interactions.

This chapter presents set-based approaches that focus on identifying G X E interactions rather than set-based approaches that are based primarily on detecting G main effects (e.g., via marginal effects). The author reviews both his own research and the development of his Set Based Gene EnviRonment InterAction test (SBERIA), as well as another set-based G X E approach referred to as GESAT. GESAT extends the variance component test of the SNP-set Kernel Association Test (SKAT) to evaluate G x E effects while incorporating the main SNP effects as covariates. While both of these approaches (SBERIA and GESAT) have outperformed other benchmark methods (e.g., likelihood ratio test) and have been demonstrated to retain the appropriate Type 1 error rate, in this chapter the author conducts simulation studies to compare findings for SBERIA and GESAT approaches, and identifies associated strengths and limitations of the respective methods.

Accumulating evidence indicates that there is a rich and varied interplay between persons and their environments, which strongly suggests that this involves gene-environment correlations and interactions. We investigated whether familial risk (FR) to externalizing behaviors and prenatal and perinatal risk factors, separately or in interaction with each other, predicted externalizing behaviors. The subjects were 10- to 12-year-old preadolescents who were taking part in TRAILS, a large prospective population-based cohort study (N = 2,230). Regression analyses were used to determine the relative contribution of FR and prenatal and perinatal risks to parent and teacher ratings of inattention, hyperactivity/impulsivity aggression, and delinquency. Regression models explained between 6 and 11% of the variance of externalizing behaviors. We found main effects of FR (vs. no FR), macrosomia (birth weight > 4,500 g), maternal prenatal smoking (MPS), pregnancy and delivery complications (PDCs), and gender that were rather consistent across rater and outcome measures. For some outcome measures, the effect of MPS and PDCs depended on the presence of FR. These included both positive and negative interaction effects. Correlations between FR and prenatal and perinatal risks were significant but rather low. Both main effects and interaction effects of FR and prenatal and perinatal risks contributed to externalizing behaviors in preadolescents, but all effects were of small size. Further research including use of candidate gene polymorphisms is necessary to identify the underlying neurobiological mechanisms of these main and interaction effects.

Toxicity assessment of organic contaminants: Evaluation of mixture effects in model industrial mixtures using 2 n full factorial design

Risk of chronic obstructive pulmonary disease (COPD) is determined by both cigarette smoking and genetic susceptibility, but little is known about gene-by-smoking interactions. We performed a genome-wide association analysis of 179,689 controls and 21,077 COPD cases from UK Biobank subjects of European ancestry recruited from 2006 to 2010, considering genetic main effects and gene-by-smoking interaction effects simultaneously (2-degrees-of-freedom (df) test) as well as interaction effects alone (1-df interaction test). We sought to replicate significant results in COPDGene (United States, 2008-2010) and SpiroMeta Consortium (multiple countries, 1947-2015) data. We considered 2 smoking variables: 1) ever/never and 2) current/noncurrent. In the 1-df test, we identified 1 genome-wide significant locus on 15q25.1 (cholinergic receptor nicotinic β4 subunit, or CHRNB4) for ever- and current smoking and identified PI*Z allele (rs28929474) of serpin family A member 1 (SERPINA1) for ever-smoking and 3q26.2 (MDS1 and EVI1 complex locus, or MECOM) for current smoking in an analysis of previously reported COPD loci. In the 2-df test, most of the significant signals were also significant for genetic marginal effects, aside from 16q22.1 (sphingomyelin phosphodiesterase 3, or SMPD3) and 19q13.2 (Egl-9 family hypoxia inducible factor 2, or EGLN2). The significant effects at 15q25.1 and 19q13.2 loci, both previously described in prior genome-wide association studies of COPD or smoking, were replicated in COPDGene and SpiroMeta. We identified interaction effects at previously reported COPD loci; however, we failed to identify novel susceptibility loci.

Pleiotropic conflict, where a genetic locus has antagonistic effects on different traits, is a common phenomenon observed in animals and plants. While pleiotropy has been widely reported in humans, there is no systematic study of pleiotropic conflict in humans. Here, we leverage summary statistics from genome-wide association studies of complex diseases and traits derived from large-scale population cohorts to identify pleiotropic regions with conflicting effects. Through a multi-trait colocalization approach, we identified 219 independent regions containing variants showing pleiotropic conflict, which cover ~11.4% of linkage disequilibrium blocks in the human genome. Antagonistic variants are observed to be enriched for genetic variants with intermediate minor allele frequencies and antagonistic regions show signatures of positive/balancing selection. Our results suggest that antagonistic variants are pervasive in humans and indicate their role in maintaining phenotypic and genetic diversity in humans.