BPS2026 – Chromatin folding dynamics from feature-optimized adaptive sampling of MSMs

Chromatin, composed of DNA wrapped around histone and non-histone proteins, exhibits dynamic and heterogeneous folding that regulates both gene expression and nuclear architecture. Quantitative analysis of these processes is often hampered by intrinsic self-association and the interplay of numerous in vivo factors. To address this, we investigate a minimal fiber comprising tetra-nucleosome arrays to delineate the thermodynamics and kinetics of chromatin folding. Using a chemically specific coarse-grained (CG) model implemented in OpenMM, coupled with adaptive sampling and Markov state models (MSMs), we map conformational ensembles and transitions between metastable states while systematically varying DNA linker length and nucleosomal DNA wrapping. Simulations with this new CG model, benchmarked against experimental measurements and established theoretical frameworks, reveal free-energy landscapes that depend strongly on linker length. MSMs resolve metastable states and identify kinetically and thermodynamically favorable routes, highlighting slow collective modes in inter-nucleosome rearrangements. We further employ deep learning-based optimization of low-dimensional reaction coordinates, yielding MSMs with enhanced kinetic fidelity. This integrated computational framework provides a systematic, scale-bridging interrogation of chromatin folding mechanisms that remain challenging to probe experimentally and clarifies how DNA geometry modulates self-association and energy barriers during the early stages of chromatin organization.