The formation of condensed heterochromatin is critical for establishing cell-specific transcriptional programs. To reveal structural transitions underlying heterochromatin formation in maturing mouse rod photoreceptors, we apply cryo-EM tomography, AI-assisted denoising, and molecular modeling. We find that chromatin isolated from immature retina cells contains many closely apposed nucleosomes with extremely short or absent nucleosome linkers, which are inconsistent with the typical two-start zigzag chromatin folding. In mature retina cells, the fraction of short-linker nucleosomes is much lower, supporting stronger chromatin compaction. By Cryo-EM-assisted nucleosome interaction capture, we observe that chromatin in immature retina is enriched with i±1 interactions while chromatin in mature retina contains predominantly i±2 interactions typical of the two-start zigzag. By mesoscale modeling and computational simulation, we clarify that the unusually short linkers typical of immature retina are sufficient to inhibit the two-start zigzag and chromatin compaction by the interference of very short linkers with linker DNA stems. We propose that this short linker composition renders nucleosome arrays more open in immature retina and that, as the linker DNA length increases in mature retina, chromatin becomes globally condensed via tight zigzag folding. This mechanism may be broadly utilized to introduce higher chromatin folding entropy for epigenomic plasticity.

Retina chromatin samples with a concentration of about 0.2 mg/ml DNA were mixed with a suspension of 10 nm fiduciary gold particles (Sigma Aldrich cat # 741957) treated with bovine serum albumin to prevent clustering. 3 μl chromatin samples were applied to Quantifoil R2/2 200 mesh copper grids (EMS Q250-CR2). Vitrification by plunging into liquid ethane was conducted using the FEI Vitrobot Mk IV Grid Plunging System at 100% humidity, 4 °C, and setting the blotting strength at 5, and blotting time at 3.5 sec.  Cryo-EM imaging was conducted on a Titan Krios G3i 300 kV electron microscope, equipped with a K3 direct electron detector (Gatan, CA) at the Penn State Hershey cryo-EM core as described  (61). Tilt series were aligned using fiducials, CTF corrected, and SIRT reconstructed using the IMOD software suite (https://bio3d.colorado.edu/imod/).

Regression denoising was accomplished in Dragonfly (ORS) using data synthesized by cryo-TomoSim (CTS, https://github.com/carsonpurnell/cryotomosim_CTS). Regression denoised images were exported as .tiff files and converted to MRC using the mrc2tif program within IMOD. The regression-denoised images were segmented into smaller subtomograms by IMOD/3dmod and inverted using “newstack” to generate subtomograms with positive intensity corresponding to high density. Data was visualized either with IMOD or Chimera. 

For stereological measurements, the SIRT-reconstructed tomograms were segmented into smaller subtomograms by IMOD/3dmod,  inverted using “newstack”, and filtered by nonlinear anisotropic diffusion using IMOD command: “nad_eed_3d -n 30 -f -k 50”  to reduce noise and enhance chromatin edges. The filtered subtomograms (both SIRT-reconstructed and regression-denoised) were exported into UCSF Chimera (RBVI, Univ. San Francisco, CA) for interactive visualization and analysis of nucleosome structures. In Chimera, the filtered volumes were fitted with nucleosome core X-ray crystal structure (pdb 2CV5 (63)) semi-automatically to correspondent electron densities in the volume using the ‘fitmap’ command and the nucleosomes were overlaid with centroids, center-to-center axes, and nucleosome planes using the structure analysis ‘Axes/Planes/Centroids’ tool. Each nucleosome in an array was numbered and the following measurements were recorded: center-to-center distance D to the next nucleosome (n+1) in the array; center-to-center distance N to the nearest nucleosome (n_x) in the 3D space; angle α between the two axes connecting each nucleosome with the previous one (n-1) and the next one (n+1) in a chain: angle β between the planes of consecutive nucleosomes n and n+1; and angle para between the plane of each nucleosome (n) and the plane of the nearest nucleosome (n_x) in the 3D space.

The chromatin mesoscale model combines coarse-grained representations of nucleosome cores, histone tails, linker DNA, and LHs within chromatin fiber arrays. To introduce nucleosome cores with 0 bp linker DNA, we connect two nucleosomes by a spring of 3 nm equilibrium length. Overlapping nucleosomes are allowed due to soft excluded volume terms for nucleosome-nucleosome interactions. We simulate 20 copies each for PN1 and PN56 by Monte Carlo sampling, recording for each an ensemble of 2000 configurations. For each ensemble, we calculate the packing ratio as the number of nucleosomes in 11 nm of fiber, the internucleosome interactions, and the tail interactions. For a single trajectory of PN1 and PN56, we create fan plots to represent, along each nucleosomal plane, the linker DNA cumulative and average positional distribution across a single trajectory. For single trajectories, we also assess quantitatively the degree of stem formation, as follows. We measure the distances between the average positions of each pair of beads on the two linker DNAs associated with the same core and assume that stem formation requires the distance between two DNA beads to be less than 2.5 nm.