Abstract [unknown]

CHG methylation, but had a significant effect on CHH methylation, levels of which were 14% higher at 16˚C than at 10˚C, on average ( Figure 1A) . To investigate the genetic basis of DNA methylation, we performed genome-wide association studies (GWAS) using different facets of average methylation as the phenotype. For global CG and CHG methylation, no associations reached genome-wide significance, while for CHH methylation a clear peak of association was observed on chromosome 4 (Figure 1-figure
more » ... 4 (Figure 1-figure supplement 1). The association was even more significant when restricting attention to average CHH methylation of large transposons ( Figure 1B ), in agreement with the notion that this type of methylation mostly occurs in transposons in Arabidopsis (Stroud et al., 2013). The association centered around a SNP at 10,459,127 on chromosome 4, 38 kb downstream from the locus AT4G19020, which encodes a homolog of the CHG methyltransferase chromo-methylase-3 (Lindroth et al., 2001) that has recently been shown to catalyze both CHH and CHG methylation on transposons, and is thus an excellent candidate (Zemach et al., 2013; Stroud et al., 2014) . A multi-locus mixed model (Segura et al., 2012) that included the identified SNP (CMT2a) as a fixed effect revealed another SNP downstream of CMT2, at position 10,454,628 (CMT2b), 4.5 kb closer to CMT2 than CMT2a, and in complete linkage disequilibrium with it (i.e., the non-reference alleles at CMT2a and CMT2b are never seen together). Repeating the GWAS with both CMT2a and CMT2b as cofactors identified no further loci (Figure 1-figure supplement 2) . Both non-reference alleles are common in southern Sweden, but are also found in the north (22.6% vs 9.5% and 30.6% vs 7.9% for CMT2a and CMT2b, respectively). Accessions with the non-reference CMT2a allele have on average more CHH methylation on transposons than those with the reference haplotype (p = 1.1e-03), while those with the non-reference CMT2b allele have lower levels of CHH methylation than the reference haplotype (p = 8.1e-03; Figure 1C ). The associations were readily confirmed using an F2 population generated by
doi:10.7554/elife.05255.001 fatcat:4pblcwusx5hj5d3h72omdny7hq