ABSTRACT
An increasing number of field studies show that the phenotype of an individual plant depends not only on its genotype but also on those of neighboring plants; however, this fact is not taken into consideration in genome-wide association studies (GWAS). Based on the Ising model of ferromagnetism, we incorporated neighbor genotypic identity into a regression model in this study. The proposed method, named “neighbor GWAS”, was applied to simulated and real phenotypes using Arabidopsis thaliana accessions. Our simulations showed that phenotypic variation explained by neighbor effects approached a plateau when an effective spatial scale became narrow. Thus, the effective scale of neighbor effects could be estimated by patterns of the phenotypic variation explained. The power to detect causal variants of neighbor effects was moderate to strong when a trait was governed by tens of variants. In contrast, there was a reasonable power down when hundreds of variants underlay a single trait. We applied the neighbor GWAS to field herbivory data on 200 accessions of A. thaliana, and found that the neighbor effects more largely contributed to the observed damage variation than self-genotype effects. Interestingly, several defensin family genes were associated with neighbor effects on the herbivory, while self-genotype effects were related to flavin-monooxygenase glucosinolate S-oxygenase 2 (FMO GS-OX2). Overall, the neighbor GWAS highlights the overlooked but significant role of plant neighborhood effects in shaping phenotypic variation, thereby providing a novel and powerful tool to dissect complex traits in spatially structured environments.