Genome-wide association analysis
GWA analyses were performed using the easyGWAS tool (Grimm et al., 2017)(https://easygwas.ethz.ch/). For each of the complex traits, the averages under limiting, intermediate, or optimal N, the fold change between limiting to optimal N, and the CV across the three N conditions were used as input for GWAS. The analyses were performed with both the 250K SNPs and the 1001 whole sequence datasets, using TAIR10 gene annotation. The data was transformed using the BoxCox method, and each analysis was corrected for population structure using the FaSTLMM algorithm. The minimum allele frequency (MAF) cut-off was set to 10 %, and only associations with correction significance level α < 0.1 using Bonferroni-method were reported (Table 1 and S4).