Dong, X.; Jiao, W.-B.; Campoy, J. A.; Rabanal, F. A.; Ton, J.; Smith, L. M.; Weigel, D.; Schneeberger, K.

Centromeres are specialized chromosome regions essential for sister chromatid cohesion and spindle attachment during mitosis. Many centromeres comprise highly variable, megabase-scale satellite DNA arrays, yet the mutation spectrum driving this variability remains poorly understood. Using replicated genome assemblies of six Arabidopsis mutation accumulation lines, we identified centromeric mutations consisting almost exclusively of point mutations and structure-preserving, in-frame indels spanning a few kilobases. Centromeric point mutations occurred at a ninefold higher rate (6.1x10^-8/bp/gen) than in chromosome arms, frequently introduced by non-allelic gene conversions from closely linked repeat units. Forward-in-time simulations based on the observed mutation spectrum recapitulated the emergence of megabase-scale higher-order repeat (HOR) structures, including long-range sequence similarities, without requiring large-scale rearrangements, closely mirroring centromeric divergence among natural genomes. Our results show that centromere evolution is driven by a unique mutational spectrum, providing a quantitative framework to understand how small-scale mutations shape and maintain the large-scale architecture of centromeric DNA.