Stochastic changes in cytosine methylation are a source of heritable epigenetic and phenotypic diversity in plants. Using the model plant Arabidopsis thaliana, we derive robust estimates of the rate at which methylation is spontaneously gained (forward epimutation) or lost (backward epimutation) at individual cytosines and construct a comprehensive picture of the epimutation landscape in this species. We demonstrate that the dynamic interplay between forward and backward epimutations is modulated by genomic context and show that subtle contextual differences have profoundly shaped patterns of methylation diversity in A. thaliana natural populations over evolutionary timescales. Theoretical arguments indicate that the epimutation rates reported here are high enough to rapidly uncouple genetic from epigenetic variation, but low enough for new epialleles to sustain long-term selection responses. Our results provide new insights into methylome evolution and its population-level consequences.epigenetics | epimutation | DNA methylation | evolution | Arabidopsis P lant genomes make extensive use of cytosine methylation to control the expression of transposable elements (TEs) and genes (1). Despite its tight regulation, methylation losses or gains at individual cytosines or clusters of cytosines can emerge spontaneously, in an event termed "epimutation" (2, 3). Many examples of segregating epimutations have been documented in experimental and wild populations of plants and in some cases contribute to heritable variation in phenotypes independently of DNA sequence variation (4, 5). These observations have led to much speculation about the role of DNA methylation in plant evolution (6-8), and its potential in breeding programs (9). In the model plant Arabidopsis thaliana, spontaneous methylation changes at CG dinucleotides accumulate in a rapid but nonlinear fashion over generations (2,3,10), thus pointing to high forward-backward epimutation rates (11). Precise estimates of these rates are necessary to be able to quantify the long-term dynamics of epigenetic variation under laboratory or natural conditions, and to understand the molecular mechanisms that drive methylome evolution (12-14). Here we combine theoretical modeling with high-resolution methylome analysis of multiple independent A. thaliana mutation accumulation (MA) lines (15), including measurements of methylation changes in continuous generations, to obtain robust estimates of forward and backward epimutation rates. ResultsWe joined whole-genome MethylC-seq (16) data from two earlier MA studies (2, 3) with extensive multigenerational MethylC-seq measurements from three additional MA lines (Fig. 1A and SI Appendix, Tables S1-S6). The first of these new MA lines (MA1 3) was propagated for 30 generations and includes measurements for 13 (nearly) consecutive generations (Fig. 1A). The other two MA lines (MA2 3) were propagated for 17 generations and were measured every four generations on average (Fig. 1A). These new data therefore allowed us to track epimutation...