Extended Data Fig. 1: Summary of investigated global chromatin architecture features and technical strategy for confinement analysis.

A. Hi-C data generated from brain tumors and normal brain tissues are analyzed at genome-wide scales. UMAP embedding based on genome-wide comparison across Hi-C contact matrices at three different scales: compartmentalization (first principal component / compartment score), topologically associating domain organization (RobusTAD boundary score), and matrix similarity (HiCRep coefficient). H3K27M pHGGs do not separate from H3 WT pHGGs by any of the three modalities examined. B. From Hi-C datasets in A, silhouette width based on inter-sample similarity in terms of three different modalities, with more positive values indicating that a sample is closer to other samples belonging to the same class whereas more negative samples indicating lack of cohesion (that is, class label is not reflected by high inter-sample similarity for those belonging to the same class). H3K27M pHGGs emerge as the only tumor subtype demonstrating lack of distinct signatures across all three scales considered, generally showing negative silhouette scores (that is, less similar to other H3K27M pHGGs than to tumors of another type). H3K27M thus does not impose a specific signature on large-scale genome organization. C. Euler diagram of CTCF peaks identified in isogenic H3K27M pHGG cell lines and their KO counterparts, demonstrating a substantial overlap. D. Pile-up of pairwise Hi-C interactions among the union CTCF peak set across all H3K27M and KO samples; only pairs of sites with convergent motif orientations were considered. This reveals a lack of global differences in CTCF interaction strength between isogenic H3K27M and H3K27M-KO pHGG cells. E. Correlation of compartment/insulation score differences (H3K27M versus KO/WT) between isogenic comparisons wherein H3K27M is knocked out of BT245, DIPGXIII and HSJ019 glioma cell lines and H3.3K27M is overexpressed in the G477 histone-WT glioma line. The weak correlation coefficients demonstrate lack of consistent changes in compartment/domain structures upon the removal or overexpression of H3K27M. F. Representative tracks of experimental and simulated ChIP–seq datasets, demonstrating the distinction between confined versus diffuse profiles of H3K27me3 or CTCF. G. Genome-wide fragment cluster score computed at either 1 kb shift distance and simulated H3K27me3 at varying shift distances. Our choice for measuring ‘confinement’ can quantitatively distinguish confined versus diffuse experimental ChIP–seq profiles. H. Metaplots showing aggregate depth-normalized H3K27me3 signals from simulated datasets with varying degrees of confinement, with hypothetically no difference in true modification levels at the very center. This reinforces that depth-normalization (for example, CPM) of a more diffuse profile will yield the impression of a lower peak as compared to confined profile, despite no difference in the absolute value at the center (that is, a by-product of ChIP–seq depth-normalization). This phenomenon can be important to consider when assessing normalized metaplots. I. Confinement scores of H3K27me3 (fragment cluster score at 10 kb, see methods) for published ChIP–seq data from the developing mouse brain, ranging from embryonic day 10.5 (E10.5) to birth (P0), in Gorkin et al. (2020). Diminishing scores indicate the spread of H3K27me3 accompanies early brain development.