Abstract
Chromosomal loops are CTCF–cohesin-mediated 3D genomic structures in the mammalian nucleus. In addition to these CTCF–cohesin loops, other epigenomic features can also form long-range interactions. Polycomb-targeted loci, regulated by the Polycomb repressive complex and trithorax group during development, form long-range chromatin interactions independent of CTCF–cohesin and are demarcated by regions of low DNA methylation—referred to as DNA methylation canyons.
We previously identified extremely long Polycomb loops occurring between DNA methylation Canyon demarked Polycomb-targeted loci, spanning distances of up to 60 Mb (Zhang et al, 2020). These loops are found exclusively in self-renewing cells, such as human hematopoietic stem cells and mouse embryonic stem cells.
In hematopoietic malignancies, both DNA methylation and Polycomb binding are significantly altered. To explore this further, we conducted a pan-cancer survey of long Polycomb loops across 223 tumor samples, with a focus on hematopoietic malignancies—given that their likely cell-of-origin, the hematopoietic stem cell, exhibits strong long Polycomb loops. Our cohort included 32 acute myeloid leukemias (AMLs), 24 T-cell lymphoblastic leukemias, 61 pediatric brain tumors, 80 prostate cancers, and 26 colon cancers. We found that most cancers—including all prostate and colon cancers—lack long Polycomb loops. However, long Polycomb loops are retained in pediatric brain tumors and a subset of AMLs. Interestingly, strong long Polycomb loop interactions are observed in normal developing brain tissue and hematopoietic stem cell. Our data suggest that the presence of long Polycomb loops is likely inherited from the epigenomic state of the cell of origin, as they are also observed in normal pediatric brain tissue and hematopoietic stem cells.
Loss of long Polycomb loops in primary cancer samples is accompanied by DNA hypermethylation and reduced Polycomb binding at loop anchor loci, and these disruptions are further exacerbated in cultured cell lines. Interestingly, in AML, many of these previously silenced loci—often mesodermal transcription factors—become activated and form de novo 3D interaction anchors. For example, we observed the formation of a new domain around the leukemogenic ZEB2 gene, driven by leukemia-specific HOXA9 binding at an upstream enhancer within the TEX41 locus.
Notably, AML samples displayed a wide range of long Polycomb loop strength. While most AMLs lose these loops, approximately 12% (4 out of 33) retain strong long Polycomb loops comparable to hematopoietic stem cells. These cases recurrently harbor somatic mutations in CEBPA (2 out of 4) and STAG2 (2 out of 4). These genes are not directly linked to Polycomb or DNA methylation machinery, suggesting alternative mechanisms of loop maintenance. The CEBPA mutant AMLs also exhibited H3K27me3 spreading to non-Polycomb target loci across the genome.
We therefore tested whether EZH2 inhibition could disrupt long Polycomb loops in these AMLs and impact disease maintenance. Indeed, EZH2 inhibition attenuated long Polycomb loops, reduced colony-forming capacity, and promoted differentiation in long Polycomb loop–retaining AMLs. Treated AML cells activated a macrophage differentiation program and showed downregulation of cell cycle and DNA replication genes. These findings suggest that AMLs retaining long Polycomb loops are dependent on this 3D chromatin architecture. This rare but strong sensitivity to EZH2 inhibition in AML indicates that long Polycomb loops could serve as an epigenomic biomarker for EZH2 or other Polycomb-targeted therapies in cancer rather than the current genomic marker such as EZH2 gain-of-function mutations and SWI/SNF loss-of-function mutations.
Overall, long Polycomb loops are commonly lost during leukemogenesis due to epigenomic disruption. However, a subset of AMLs maintains these loops from the cell of origin and appears to rely on the Polycomb network to sustain self-renewal.
