Genetic factors in nonsmokers with age-related macular degeneration revealed through genome-wide gene-environment interaction analysis

Abstract

Relatively little is known about the interaction between genes and environment in the complex etiology of age-related macular degeneration (AMD). This study aimed to identify novel factors associated with AMD by analyzing gene-smoking interactions in a genome-wide association study of 1207 AMD cases and 686 controls of Caucasian background with genotype data on 668,238 single nucleotide polymorphisms (SNPs) after quality control. Participants' history of smoking at least 100 cigarettes lifetime was determined by a self-administered questionnaire. SNP associations modeled the effect of the minor allele additively on AMD using logistic regression, with adjustment for age, sex, and ever/never smoking. Joint effects of SNPs and smoking were examined comparing a null model containing only age, sex, and smoking against an extended model including genotypic and interaction terms. Genome-wide significant main effects were detected at three known AMD loci: CFH (P = 7.51×10(-30) ), ARMS2 (P = 1.94×10(-23) ), and RDBP/CFB/C2 (P = 4.37×10(-10) ), while joint effects analysis revealed three genomic regions with P < 10(-5) . Analyses stratified by smoking found genetic associations largely restricted to nonsmokers, with one notable exception: the chromosome 18q22.1 intergenic SNP rs17073641 (between SERPINB8 and CDH7), more strongly associated in nonsmokers (OR = 0.57, P = 2.73 × 10(-5) ), with an inverse association among smokers (OR = 1.42, P = 0.00228), suggesting that smoking modifies the effect of some genetic polymorphisms on AMD risk.

Figure 1

Manhattan plots of observed –log P-values for AMD (y-axis) by chromosome (x-axis) from analysis of the discovery dataset of 1,207 AMD cases and 686 controls in the basic model adjusting for sex and age-at-exam. All SNPs with associations below P < 10⁻⁵ are highlighted in red. SNPs with associations P < 10⁻¹³ that reside in the CFH and ARMS2 regions are not depicted here. The dotted line demarcates the threshold for genome-wide statistical significance, P < 5.00×10⁻⁸.

Figure 2

Manhattan plots of observed –log P-values for AMD (y-axis) by chromosome (x-axis) from analysis of the discovery dataset of 720 grade 5 AMD cases and 513 grade 1 controls in the basic model adjusting for sex and age-at-exam. All SNPs with associations below P < 10⁻⁵ are highlighted in red. SNPs with associations P < 10⁻¹³ that reside in the CFH and ARMS2 regions are not depicted here. The dotted line demarcates the threshold for genome-wide statistical significance, P < 5.00×10⁻⁸.

Figure 3

Manhattan plots of observed –log P-values for the 2-df test of association with AMD from gene-smoking interaction analysis (y-axis) by chromosome (x-axis). Analyses were performed on the discovery dataset of 894 AMD cases and 525 controls for whom data on smoking history were available, and modeling adjusted for sex and age-at-exam, as well as ever/never smoking. The two degree-of-freedom test evaluates statistical significance of the inclusion of both genetic main effect and interaction terms with the potential to identify associated variants masked by the effect of smoking. All SNPs with associations below P < 10⁻⁵ are highlighted in red. SNPs with associations P < 10⁻¹³ that reside in the CFH and ARMS2 regions are not depicted here. The dotted line demarcates the threshold for genome-wide statistical significance, P < 5.00×10⁻⁸.

Figure 4

Quantile-quantile plots of observed (y-axis) vs. expected (x-axis) P-values for LOAD from the AMD discovery sample under the basic model adjusting for age-at-exam and sex. Plot (A) examines the results from all cases vs. all controls in the discovery sample, whereas (B) examines the results from association among grade 5 cases vs. grade 1 controls. The genomic inflation factor for analysis of all cases and controls is estimated to be λ = 1.07, whereas for the analysis of grade 5 cases vs. grade 1 controls, λ = 1.03.