Chromatin architecture reprogramming reveals novel epigenetic dependencies in breast cancer

Abstract

Chromatin architecture plays a key role in development and cancer, yet most studies lack mechanistic depth due to widespread epigenomic remodeling. To address this, we tracked chromatin structure dynamics during the progression of endocrine resistance in ER⁺ breast cancer using Hi-C, chromatin accessibility, epigenomic, and transcriptomic profiling. We uncovered a critical role for H3K9 methylation and the demethylase KDM4C association with SWI/SNF in driving proliferation of cells fated to become resistant through a nongenomic estrogen-mediated mechanism. These findings highlight the mechanistic contribution of chromatin regulation in therapy resistance and offer a blueprint for studying similar processes in cancer, development, and cell fate decisions.

Most breast cancers express the estrogen receptor (ER), and agents targeting the ER signaling pathway as well as estrogen production are the main treatment modality (Guan et al. 2019; Burstein 2021; Garcia-Martinez et al. 2021). Despite the success of antiestrogen therapies, therapeutic resistance still represents a major hurdle (Jeselsohn et al. 2015; Razavi et al. 2018; Hanker et al. 2020). Acquisition of genetic alterations that emerge upon treatment pressure can activate downstream effectors and alternative oncogenic pathways to promote resistance to current therapies (Razavi et al. 2018). For instance, mutations in ESR1, the gene encoding for ER, result in constitutive ER hyperactivation and represent a leading driver of acquired resistance (Fanning et al. 2016). Other alterations such as mutations and amplification of genes such as FOXA1 and CCND1 are common in treatment-resistant ER⁺ breast cancer (Razavi et al. 2018; Fu et al. 2019). Overcoming these outcomes represents a major challenge in the therapeutic arena.

Chromatin organization in a three-dimensional (3D) structure is essential for gene regulation, cell fate decisions, and cancer. However, there is still a gap in the knowledge about how chromatin architecture is established, reset, and maintained during therapeutic pressure. The human genome is organized in multiple structural layers including chromosome territories, chromatin compartments, topologically associating domains (TADs), and chromatin loops (Tettey et al. 2023). These different levels of organization collectively contribute to gene regulation and are often disrupted in cancer (Achinger-Kawecka et al. 2016).

A better understanding of the interplay between chromatin architecture and oncogenic regulatory networks can be exploited to identify biomarkers to ultimately improve the outcome of current therapies. For instance, chromatin architecture dynamically changes across a time course of estrogen stimulation in ER⁺ breast cancer cells. Also, recent studies suggested that changes in the 3D chromatin architecture and deregulation of the epigenome orchestrate resistant phenotypes in multiple cancers, including breast cancer (Zhou et al. 2019; Achinger-Kawecka et al. 2020, 2024). In line with these observations, epigenetic therapy stands up as a promising therapy for the treatment of endocrine-resistant breast cancer (Achinger-Kawecka et al. 2024). However, how these structural changes translate into targetable actions in cancer remains largely unknown. Here, we sought to characterize the impact of chromatin remodeling in endocrine resistance acquisition and evolution to identify novel epigenetic dependencies in cancer. To address this, we used a multipronged approach and integrated chromatin architecture, accessibility, transcriptomics, and epigenomic profiles with functional studies in a model of aromatase inhibitor mimics and in cells with constitutively active ESR1.

Results and Discussion

Discrete therapeutic pressures result in specific chromatin architecture changes

Therapeutic response is coordinated by transcriptional programs that integrate signaling pathways, epigenetic mechanisms, and chromatin organization (Garcia-Martinez et al. 2021; Arruabarrena-Aristorena and Toska 2022; Achinger-Kawecka et al. 2024). How the epigenome and the 3D chromatin organization are remodeled during evolution of endocrine resistance is largely unknown. To address this, we used two different endocrine-resistant models. First, we exploited a model that recapitulates ER⁺-to-ER⁻ conversion, a switch in the ER status found in patients who developed resistance to aromatase inhibitors (AIs) (Lopez-Tarruella and Schiff 2007; Ma et al. 2015; Garcia-Martinez et al. 2022). Second, we used CRISPR/Cas9-edited ER⁺ cells carrying a heterozygous Y537S mutation at ER, a common mutation in ESR1 that confers constitutive ligand-independent activity and is found in patients with metastatic disease who receive endocrine therapies such as tamoxifen and aromatase inhibitors (Fig. 1A; Jeselsohn et al. 2015; Ma et al. 2015).

Figure 1.

Transcriptional and chromatin 3D architecture reprogramming during acquisition of endocrine resistance in ER⁺ breast cancer. (A) Schematic representation of two models of endocrine therapy acquisition by estrogen deprivation (parental, primed, and reprogrammed) or mutations in the estrogen receptor (T47D^WT/Y537S). (B) UMAP projection of scRNA-seq reveals distinct clustering of parental and primed cells. (C) Pathway enrichment analysis of differentially expressed genes in the different clusters highlighting the enrichment of pathways related to proliferation, migration, and endocrine therapy resistance in basal and EMT clusters. (D) Gene expression distribution of selected epithelial transcription factors (e.g., ESR1, GATA3, and FOXA1) and cell surface markers (CD24 and CD44) across different clusters, showing downregulation of luminal identity and upregulation of basal and EMT markers during resistance acquisition. (E) Pseudotime trajectory inference depicting a progressive cellular transition from parental to primed cells. (F) Principal component analysis. (G) Distribution of genomic interactions by distance. (H) Comparative differential contact matrices across comparisons, highlighting changes in interchromosomal interactions across the different resistant states. (I) Hi-C contact heat maps of a region in chromosome 17 show progressive reorganization of topologically associating domains (TADs) during the resistance transition; loop changes are highlighted by dashed circles.

To mimic the AI endocrine-resistant modality, ER⁺ parental cells, which are luminal and express CD24, were cultured in the absence of estrogen (E2). Months after E2 depletion, an intermediate or primed resistant state emerged, which was characterized by a mixed population of CD44⁺ (basal marker) (Honeth et al. 2008) and CD24⁺ (luminal marker) and reduced levels of ER and ER cofactors (Supplemental Fig. S1A; Garcia-Martinez et al. 2021). After 9–12 months of estrogen deprivation, cells acquired a final reprogrammed state characterized by being 100% CD44⁺, transition to a basal state, and a complete loss of ESR1 expression (Supplemental Fig. S1A,B; Sleeman et al. 2006; Shipitsin et al. 2007; see Garcia-Martinez et al. 2022 for full characterization of the model; . One of the biggest challenges in breast cancer treatment is the heterogeneity of the disease. To characterize the cellular heterogeneity of our model, we performed single-cell RNA-seq (scRNA-seq) in parental and primed cells. By clustering single cells, we annotated five clusters (Fig. 1B). On the one hand, the transcriptomic profile of the parental cells confirmed their epithelial and luminal identity. On the other hand, primed cells exhibit high transcriptional heterogeneity, and the signature of most of the cells was identified as basal or epithelial-to-mesenchymal transition (EMT). Enrichment analysis of the different clusters identified the MAPK and PI3K–AKT signaling pathways, as well as pathways related to motility and morphological changes, as being enriched in the EMT and basal clusters (Fig. 1C). Interestingly, we also identified a subset of cells with an epithelial profile similar to one of the clusters in the parental cells, suggesting that these cells may represent the origin of the emerging resistant population. As expected, in primed cells, we found higher expression of CD24, ESR1, GATA3, and FOXA1 in the luminal and epithelial clusters, while CD44 and JUN levels were elevated in basal and EMT clusters (Fig. 1D; Supplemental Fig. S1C). Trajectory analysis reflects evolution from parental to primed cells (Fig. 1E), suggesting a gradual reprogramming characterized by the accumulation of transcriptional changes, leading to loss of ER dependence and increased plasticity. These results agree with the cellular heterogeneity that we observed by flow cytometry (Supplemental Fig. S1A). Due to the heterogeneous CD24/CD44⁺ population of the primed cells, we sorted the CD44⁺ population (Supplemental Fig. S1A,B). This approach enabled us to monitor chromatin changes throughout the development of resistance and assess whether primed CD44⁺ cells represent a transient intermediate state capable of further evolution or instead constitute a stable, biologically distinct population. Notably, CD44⁺ cells are more aggressive in vitro and in vivo than CD24⁺ cells (Garcia-Martinez et al. 2022), and tumors composed mostly of CD44⁺ cells may have worse clinical outcomes than tumors predominantly composed of CD24⁺ cells (Shipitsin et al. 2007). To assess how the 3D chromatin organization is remodeled during endocrine resistance, we performed Hi-C experiments in parental CD24⁺ cells, CD44⁺ primed cells, and fully reprogrammed CD44⁺ cells. To compare how the chromatin organization is altered in another endocrine-resistant model, we performed Hi-C assays in T47D^WT/Y537S cells (Bahreini et al. 2017; Garcia-Martinez et al. 2022).

Principal component analysis (PCA) showed clear separation between parental, primed, and reprogrammed cells, suggesting a gradual, dynamic, and global architectural change during loss of ER. In contrast, parental and T47D^WT/Y537S cluster together, suggesting that 3D chromatin architecture is not significantly altered in the mutant cells (Fig. 1F; Supplemental Fig. S1D-F). These results suggest that chromatin architecture has a minor role in endocrine resistance mediated by ESR1 mutations and plays a major role upon loss of ER (Jeselsohn et al. 2015; Garcia-Martinez et al. 2022).

The human genome is organized in multiple layers ranging from chromosome territories to smaller substructures such as TADs and chromatin loops (Tettey et al. 2023). To determine changes in chromatin interactions across the different resistant models, we divided chromatin interactions into four groups: short range (0–20 kb), mid to long range (20 kb to 2 Mb), >2 Mb, and interchromosomal (Fig. 1G). In parental and T47D^WT/Y537S cells, we found a higher proportion (38.38% and 37.89%, respectively) of short-range interactions compared with primed (30.03%) and reprogrammed (31.15%) cells. These differences suggest that parental and T47D^WT/Y537S cells retain stable local chromatin contacts and a more restricted gene expression program. In contrast, the drop in short-range interactions in primed and reprogrammed CD44⁺ cells suggests more chromatin relaxation and a more suitable environment for transcriptional rewiring. Importantly, in primed cells, the percentage of short-range interactions was the lowest among all cellular states, which agrees with the plasticity of these cells. Moreover, ~8% of the short interactions in the parental cells were lost in primed and reprogrammed cells, while 20 kb to 2 Mb interactions were gained. The increased number of mid-size-range interactions in primed (40.8%) and reprogrammed (40%) cells suggest the reorganization of regulatory elements such as TADs. Long-range interactions remained stable with a slight increase in primed (14.5%) and reprogrammed (14.1%) cells, suggesting a shift toward a more decondensed chromatin state and more transcriptional flexibility. On the other hand, there were no differences in interchromosomal interactions across all the cellular states, suggesting that nuclear compartmentalization is not dramatically altered during resistance acquisition in these two different models (Fig. 1G,H).

Next, we analyzed interchromosomal interactions and compared those present in parental cells with the resistant cells. Comparative heat maps suggest that differences in interchromosomal interactions are more noticeable in parental and reprogrammed cells, while primed and T47D^WT/Y537S cells are more similar to parental cells. Notably, we identified differentially regulated regions in reprogrammed and primed cells compared with parental cells. For instance, interactions between chromosomes 8 and 14, 7 and 15, and 9 and 15 are weaker in primed and reprogrammed cells (Fig. 1H; Supplemental Fig. S1G). Importantly, these chromosomes contain genes highly relevant to breast cancer, such as FOXA1, MYC, EGFR, EZH2, NOTCH, and CDKN2A, suggesting a coordinated regulation of different oncogenic and cell cycle-related pathways during the evolution of endocrine resistance. These results agree with the changes in the mid-size range of interactions and with the idea that many genes are regulated in a coordinate fashion inside the TADs. After a closer look at chromosome 7, we found a myriad of changes in the strength of the interactions inside TADs and at more extensive distances through changes in loops (Fig. 1I; Supplemental Fig. S1G-I). In the case of T47D^WT/Y537S cells, we did not observe major changes in interchromosomal interactions compared with parental cells (Fig. 1H,I; Supplemental Fig. S1G-I).

Chromatin accessibility changes during endocrine resistance

The interplay between chromatin architectural changes and chromatin accessibility is crucial to regulate gene expression. To investigate the dynamic changes in chromatin accessibility during cellular reprogramming, we performed ATAC-seq across the three cellular states in the ER loss model (Fig. 2A). Unsupervised clustering of ATAC-seq peaks identified four main clusters of sites undergoing dynamic changes in chromatin accessibility during the evolution of resistance (Fig. 2A). Cluster 1 maintained accessibility across states, while clusters 2 and 3 showed activation and repression patterns, respectively, reflecting regulatory changes during reprogramming. Motif enrichment analysis linked specific transcription factor families—such as KLF, AP-1, and CTCF—to each cluster, suggesting that distinct regulatory programs underlie the chromatin organization patterns observed. These results highlight coordinated shifts in chromatin accessibility and transcription factor binding that underpin cell state transitions (Fig. 2B), which is in agreement with our previous studies showing that in the fully reprogrammed cells, ER and FOXA1 are transcriptionally silenced and c-JUN is the main oncogene that drives their proliferation (Garcia-Martinez et al. 2022). To determine how the ESR1, FOXA1, and JUN loci are remodeled during the evolution of resistance, we used the Hi-C data enhanced by deep learning algorithms (Liu and Wang 2019) that can achieve a higher resolution, making it feasible to study these small gene-size genomic regions and the interactions between promoter and enhancer of a gene (Fig. 2C-E). Additionally, we modeled the 3D chromatin structures of these loci using simulated annealing and Metropolis–Hastings simulations (Zhu et al. 2019). When we examined the ESR1 locus, we found that in the parental population, the enhancer (Fig. 2C, white) and the promoter (Fig. 2C, blue) are interacting and both regions are in a more relaxed conformation, while in the primed and reprogrammed cells, the promoter–enhancer interaction is lost but with a more compact overall conformation (Fig. 2C). In the case of JUN, we observed that in the transition from sensitive to resistant cells, the enhancer and promoter are in a more relaxed composition (Fig. 2E). These 3D models generated from Hi-C highlight the relevance of chromatin architecture changes in gene expression and as drivers of resistance to endocrine therapies.

Figure 2.

Chromatin accessibility and 3D genome architecture reorganization at key regulators of cell identity during resistance evolution. (A) Heat maps of ATAC-seq signal dynamics and clustering in the parental, primed, and reprogrammed states. (B) Motif binding enrichment analysis within each cluster. (C–E) Enhanced Hi-C contact maps (top) and predicted 3D chromatin structures (bottom) at the ESR1, FOXA1, and JUN loci. (White) Enhancers, (blue) promoters. (F,G) Zoomed-in Hi-C maps and epigenomic profiling at the ESR1 and JUN loci showing coordinated changes in chromatin looping, accessibility (ATAC-seq), transcription factor binding (CTCF and cohesin), active enhancer marks (H3K27ac), and RNA Pol II occupancy in parental and reprogrammed cells. (H) Differential contact maps at the ESR1 superenhancer and promoter regions comparing primed versus parental (left) and reprogrammed versus primed (right) states, overlaid with H3K27ac enrichment.

Integration of chromatin architecture changes and the epigenetic landscape is crucial to understand the role of chromatin remodeling in plasticity and drug resistance. To characterize the epigenetic changes associated with chromatin architecture remodeling during the loss of ER, we performed ChIP-seq of two chromatin architectural factors (CTCF and cohesin) and the active enhancer and promoter histone modification H3K27ac and RNA Pol II, as well as chromatin accessibility assays by ATAC-seq in parental and reprogrammed cells (Garcia-Martinez et al. 2022). Interestingly, CTCF binding increased at sites that become less accessible during reprogramming and, more modestly, at regions that gain accessibility (Supplemental Fig. S2A). As expected, H3K27ac—and, weaker, RNA Pol II—correlated with open chromatin in both parental and reprogrammed cells (Supplemental Fig. S2B,C). On the one hand, the stronger accumulation of CTCF at less accessible sites suggests a critical role of this insulator in creating a repressive environment rather than guaranteeing chromatin accessibility in reprogrammed cells. On the other hand, H3K27ac signal and chromatin accessibility are strongly correlated, but the modest correlation with RNA Pol II suggests that chromatin accessibility and H3K27ac are necessary for gene transcription but are not sufficient.

Analysis of the ESR1 locus showed a marked reduction in CTCF and cohesin chromatin occupancy in the reprogrammed cells (Fig. 2F, bottom panel). CTCF and cohesin are critical regulators of TADs and loop formation and maintenance (Gabriele et al. 2022). Thus, these results agree with the strongest interactions observed in the parental cells compared with reprogrammed cells (Fig. 2F, top panel). In addition, H3K27ac levels and RNA Pol II binding, which correlate with transcriptional activity, as well as chromatin accessibility are also reduced in cells with ER loss (Fig. 2F, bottom panel). Importantly, these changes correlate with the alterations in 3D architecture at this locus, suggesting that chromatin 3D interactions could directly affect ESR1 expression in different cellular states. Moreover, we assessed genome-wide how TAD remodeling relates to transcription, RNA Pol II occupancy, and H3K27ac. Genes that gain expression within TADs formed during reprogramming show increased RNA Pol II and H3K27ac, whereas those that lose expression within these new TADs show the opposite. Conversely, genes that decrease expression within TADs lost upon reprogramming display reduced RNA Pol II and H3K27ac levels, while genes that gain expression within lost TADs show increased RNA Pol II occupancy (Supplemental Fig. S2D-G). Next, we calculated the differential Hi-C value for the different cellular states in the genomic region containing the ESR1 promoter, gene body, and enhancer. We found that there is a general loss of chromatin interactions in primed and reprogrammed cells compared with parental cells, suggesting chromatin decompaction or remodeling during ER loss (Fig. 2H). Next, we analyzed the JUN, EGFR, and MAPK1 loci (Fig. 2G; Supplemental Fig. S2H,I) and found that reprogrammed cells showed changes in chromatin structure and altered CTCF and cohesin binding and levels of H3K27ac and RNA Pol II occupancy. These changes are consistent with increased c-JUN expression, which is linked to aggressive and metastatic breast cancer (Smith et al. 1999; Lukey et al. 2016) as well as activation of the MAPK pathway during resistance evolution.

Chromatin architecture changes underlie dynamic transcriptional regulation during endocrine therapy resistance

Chromatin compartments are large-scale 3D structures of the genome that reflect the segregation of active (compartment A) and inactive (compartment B) chromatin states within the nucleus. (Hildebrand and Dekker 2020). Chromatin compartments are determinants of gene regulation, chromatin dynamics, and cell type-specific 3D genome architecture (Dixon et al. 2015), and aberrant compartmentalization is observed in cancer, aging, and other diseases (Gridina and Fishman 2022). To quantify and compare the degree of chromatin compartmentalization across the different cellular states, we generated saddle profiles and strength analyses (Fig. 3A; Supplemental Fig. S3A,B). Saddle profiles define the average frequency of chromatin interactions across different genomic regions sorted by their compartment identity. A strong compartmentalization is characterized by the lack of interactions (Fig. 3A, blue) and stronger interactions (Fig. 3A, red scale). Parental cells showed strong and consistent compartmentalization with stronger intracompartment interactions compared with primed and reprogrammed cells (Fig. 3A). In addition, in the primed and reprogrammed cells, we observed an increase in the frequency of the intercompartment interactions, suggesting a weaker compartmentalization of the chromatin. In the case of T47D^WT/Y537S cells, the saddle profile is very similar to parental cell lines, suggesting that this ER mutation has no effect on chromatin compartmentalization (Fig. 3A). These results suggest that reduced compartmentalization in primed and reprogrammed cells agrees with the plasticity and loss of the epithelial identity of these populations.

Figure 3.

Dynamics of A/B compartments during resistance evolution. (A) Saddle profile from Hi-C interaction contact frequency maps in parental, primed, reprogrammed, and T47D^WT/Y537S cells. The X- and Y-axes represent genomic bins, and the color intensity indicates the interaction frequency. (B) Alluvial plot of A/B compartment switch in the different comparisons. (C) Pie charts quantifying the percentage of stable A (AA), stable B (BB), A-to-B (AB), and B-to-A (BA) compartments. (D) Bar plot representing the number of genes that switch compartments in primed, reprogrammed, and T47D^WT/Y537S cells compared with parental cells. For resistance evolution in the ER loss model, gene changes in reprogrammed cells reflect the accumulation of genes that change compartment in the transition from parental to primed cells (orange and blue) as well as unique changes from primed to reprogrammed cells (red or green). (E) Distribution of A (red) and B (green) compartments at the ESR1 and JUN enhancers and the gene body in parental, primed, and reprogrammed cells. (F) Box plots showing the gene expression (log₂FC) of genes that undergo A-to-B or B-to-A compartment switches in parental versus reprogrammed cells. (****) P < 2.22 × 10⁻¹⁶. (G) Box plots showing the gene expression (log₂FC) of genes that undergo A-to-B or B-to-A compartment switches in parental versus T47D^WT/Y537S cells. (****) P < 9.9 × 10⁻⁵. (H) Enriched gene ontology (GO) analysis of activated and suppressed genes that switch from A to B compartments during reprogramming. The size of the dots represents the number of genes, and the color indicates the adjusted P-value. (I) Box plot of ATAC signal in reprogrammed cells during compartment transition from parental to reprogrammed cells at 100 kb resolution. (****) P < 2.22 × 10⁻¹⁶.

Strength profiling is an additional compartmentalization method. Primed and T47D^WT/Y537S cells showed a slight reduction in compartmentalization strength compared with the parental cells. In contrast, the strength profile in reprogrammed cells suggested that while short-range compartment interactions are strong, long-range compartmentalization is weaker or less stable (Supplemental Fig. S3B). We further examined the compartment dynamics of the evolution of resistance. We found that 15% of compartment A in parental cells switched to B in primed and reprogrammed cells, while B-to-A transitions increased from 8% in primed cells to 11% in reprogrammed cells (Fig. 3B,C; Supplemental Fig. S3C). In our ER loss model, compartmental changes were highly dynamic, with some of the regions remaining stable, while others switched to drive the molecular changes required for resistance evolution. In agreement with our previous work (Garcia-Martinez et al. 2022), we identified the primed cells as an intermediate state in resistance acquisition based on compartment organization (Fig. 3B). Interestingly, 747 compartments switched from A to B in primed cells and reverted to A in reprogrammed cells, showing that this set of genes is repressed temporarily and later reactivated (Fig. 3B). In contrast, T47D^WT/Y537S cells exhibited less changes in A/B compartments compared with parental cells, and transitions from A to B and from B to A were equally found (Fig. 3B). Moreover, the number of genes switching compartments in the different cellular states compared with parental cells was higher in primed and reprogrammed cells than in T47D^WT/Y537S cells (Supplemental Fig. S3D). Collectively, these findings indicate that compartmentalization changes are important during resistance evolution but not in resistant cells expressing a mutant ER.

Importantly, A/B compartment changes were associated with changes in gene transcription (Fig. 3D). For instance, in the transition from parental to reprogrammed cells, 8548 genes, including ESR1 and FOXA1, switched from compartment A to B and 5764 genes, including JUN, switched from B to A (Fig. 3D-F; Supplemental Fig. S3E). The number of genes switching from compartment A to B was higher than from B to A, in agreement with more compartment transitions from A to B in primed and reprogrammed cells.

By integrating RNA-seq data with our A/B compartment analyses, we demonstrated that in reprogrammed and T47D^WT/Y537S cells, genes transitioning from compartment A to B exhibited significantly lower fold change expression levels compared with parental cells, whereas genes switching from B to A significantly increased the expression levels (Fig. 3F,G). Next, we selected the subset of genes undergoing compartment and transcriptional changes and performed gene set enrichment analysis (GSEA) in reprogrammed cells. Interestingly, we found that activated genes (B-to-A switch and higher expression levels compared with parental cells) were related to migration pathways, while the suppressed genes (A-to-B switch and lower expression levels compared with parental cells) were associated with cell adhesion, indicating that compartmental changes enable the acquisition of an invasive gene expression program in the reprogrammed cells (Fig. 3H). We next sought to determine how chromatin accessibility in reprogrammed cells changes at compartments that switch from parental and reprogrammed cells. As expected, chromatin is compact at compartments that switch from A to B and becomes more accessible at compartments that switch from B to A (Fig. 3I).

Furthermore, we interrogated the expression levels for genes belonging to the epithelial and invasive signatures and whether some of these genes switch compartments in the reprogrammed cells compared with the parental population. While the epithelial subset of genes was more expressed in parental cells, in the reprogrammed cells, the invasive signature was highly expressed. Epithelial genes such as FOXA1 and PGR, which encodes for progesterone receptor, switched from A to B, while genes such as VIM and KLF4 switched from B to A (Supplemental Fig. S3F). These results agree with the mesenchymal morphology, aggressiveness, and metastatic features of the reprogrammed cells (Garcia-Martinez et al. 2022). In T47D^WT/Y537S cells, only genes switching from B to A were related to biological processes, including many neuronal, developmental, and cell growth-related pathways (Supplemental Fig. S3H). Compartmental changes in T47D^WT/Y537S cells are less dramatic than in the ER loss model but drive structural remodeling, motility, and cell growth in these cells. Altogether, our results suggest that chromatin remodeling in different endocrine resistance models enables the activation of transcriptional programs essential for growth, motility, and survival under therapeutic pressure. Moreover, chromatin architecture changes also account for both the morphology changes and increased aggressiveness and metastatic capacity observed in the model of ER loss (Garcia-Martinez et al. 2022).

TAD rearrangement and loop formation drive oncogenic expression programs

Chromosomes are spatially segregated into TADs, which are regions of high local contact frequency with sharp boundaries enriched for insulator proteins such as CTCF in a convergent motif manner. The cohesin complex, together with CTCF, has a crucial role in organizing chromatin fibers into loops and TADs (Tettey et al. 2023). TADs are typically visualized as contiguous square domains along the diagonal of Hi-C maps, in which regions within the same TAD interact more frequently than regions located in adjacent domains (Lieberman-Aiden et al. 2009). The dynamics of TADs can have a profound effect on gene expression by maintaining an isolated regulatory environment (Schoenfelder and Fraser 2019). Thus, we next sought to determine how TADs are dynamically formed during the evolution of resistance. First, we plotted the insulation score, which is a quantitative measure used to identify TAD boundaries and reflects the local contact frequency within a genomic region (Crane et al. 2015). As expected, the insulation score is lower at TAD boundaries, indicative of reduced interactions across the boundaries. Notably, on the one hand, parental and T47D^WT/Y537S cells showed a similar TAD boundary strength, highlighting the role of ER signaling in the maintenance of chromatin insulation. On the other hand, the TAD boundary dip is slightly reduced in primed cells and further reduced in reprogrammed cells, suggesting TAD reorganization as a mechanism to drive ER-independent transcription (Fig. 4A). Next, we analyzed the insulation score at regions far away from the boundaries (±500 kb). The higher insulation score compared with the boundaries is consistent with the hierarchical organization of the chromatin in which interactions within a TAD are more frequent than across boundaries. Primed and reprogrammed cells exhibit a higher distant insulation score (Fig. 4A), suggesting that during ER loss, chromatin adopts a more permissive state that promotes new transcriptional programs. In the case of parental and T47D^WT/Y537S cells, the lower distal insulation score, compared with that of primed and reprogrammed cells, suggests a tighter organization of the TADs and regulatory interactions most likely driven by ER activation. In addition, the number of TADs shared among parental and T47D^WT/Y537S cells was higher than that shared between parental and primed or reprogrammed cells (Fig. 4B).

Figure 4.

TAD and loop reorganization during endocrine therapy resistance. (A) Average insulation score profiles at TAD boundaries in parental and endocrine-resistant cell lines. (B) Venn diagrams illustrating the overlap and differences in the number of TADs identified across parental, primed, reprogrammed, and T47D^WT/Y537S cells. (C) Quantification of TAD dynamics showing the number of stable, split, merged, and rearranged TADs. (D) Representative Hi-C interaction matrices and tracks for TAD boundaries, CTCF, and cohesin in parental, primed, reprogrammed, and T47D^WT/Y537S cells at TADs lost in reprogrammed cells. (E) Gene ontology (GO) enrichment analysis of genes located within lost and gained TADs during reprogramming, indicating pathways potentially impacted by chromatin reorganization. (F) Venn diagrams showing the overlap and unique chromatin loops identified in parental, primed, reprogrammed, and resistant cells. (G) Pie chart summarizing the dynamics of chromatin loops across the different cellular states. (H) Volcano plot displaying changes in gene expression associated with gained chromatin loops in reprogrammed cells at 40 kb resolution.

Next, we examined the dynamics of TADs and identified the number of TADs that remain stable or change (split, merged, and rearranged) and found that TAD rearrangement is more frequent in reprogrammed cells compared with parental cells and less frequent in T47D^WT/Y537S cells (Fig. 4C). Moreover, we identified specific TADs gained or lost (Fig. 4D; Supplemental Fig. S4A, B) in reprogrammed cells compared with parental cells, which illustrate how these chromatin changes are enhanced from primed to reprogrammed cells and not present in T47D^WT/Y537S cells. Importantly, we identified TAD changes in regions such as MYC and EGFR during resistance evolution (Supplemental Fig. S4A,B). We also found changes in cohesin and CTCF binding, in agreement with changes in TADs (Fig. 4D; Supplemental Fig. S4A). GSEA revealed that changes in TADs correlate with changes in the expression of genes related to migration, cell–cell signaling, and the WNT pathway (Fig. 4E). Overall, these results highlight the role of ER as a chromatin regulator by distinct mechanisms such as the reinforcement of TAD boundaries. In the ER loss model, TAD boundary strength and intra-TAD interactions drive the expression of genes and lead to a resistant state.

Chromatin loops play a critical role in regulating gene expression by bringing distant regulatory elements, such as enhancers and promoters, into proximity within the 3D structure of the genome. Thus, we also determined the number of loops that change during the evolution of resistance. We found that most of the loops were unique to each cell state, in agreement with the dynamics of this system and profound changes in chromatin architecture and gene transcription upon ER loss. However, primed and reprogrammed cells share more loops than any other two cell states, in agreement with the gradual loss of ER and progressive changes of chromatin architecture in the transition toward a reprogrammed state. Accordingly, parental and reprogrammed cells have a unique signature of loops, with few loops shared between them (Fig. 4F-H; Supplemental Fig. S4C,D). Importantly, loops gained in reprogrammed cells involve the regulation of the expression of different genes such as OVOL2, an evolutionarily conserved regulator of epithelial lineage determination and epithelial-to-mesenchymal transition (EMT) (Fig. 4H; Roca et al. 2013; Wu et al. 2017). In the ER loss model, the reprogrammed cells gained a new regulatory loop that contained OVOL2 and resulted in reduced expression of this gene. Loss or reduced expression of OVOL2 is associated with more aggressive breast cancer, particularly in triple-negative breast cancer subtypes (Wu et al. 2017). In addition, we identified regulatory loops lost in reprogrammed cells that result in increased expression levels of genes such as HDAC9 (Supplemental Fig. S4E), which is overexpressed in aggressive breast tumors, leading to increased proliferation (Lapierre et al. 2016). We also identified loop changes encompassing classic ER target genes such as FMN1 in T47D^WT/Y537S cells (Supplemental Fig. S4F,G). Overall, our results showed the profound effect of chromatin architecture changes to drive specific transcriptional programs in cells that are resistant to endocrine therapies. The extensive reorganization of chromatin architecture in reprogrammed cells compared with the modest changes in ESR1 mutant cells highlights the importance of ER activity as a guardian of chromatin organization. While this study provides novel mechanistic insights into chromatin architectural changes during loss of ER upon treatment pressure, the results should be validated using patient samples that lose ESR1 expression in the metastatic setting.

KDM4C is a novel regulator of the estrogen pathway and interacts with SWI/SNF to regulate both H3K9me3 demethylation and chromatin accessibility

To identify novel transcription factors (TFs) and chromatin regulators involved in the regulation of endocrine sensitivity and resistance, we employed a multipronged approach integrating chromatin architecture and transcriptional changes. We first identified 156 TFs that were either upregulated (63 endocrine resistance candidates) or downregulated (93 endocrine sensitivity candidates) by at least twofold during the transition from parental to reprogrammed cells. Next, we assessed whether these genes underwent changes in TADs and A/B compartment shifts (either from A to B or from B to A) during cellular reprogramming. This analysis revealed 21 candidate genes potentially involved in regulating the cell identity of parental cells (Fig. 5A; Supplemental Fig. S5A). Notably, all but two of the genes that met our criteria as potential novel regulators of endocrine sensitivity are already recognized as key regulators of the estrogen pathway, highlighting the effectiveness of our approach. Among the newly identified candidates, we identified the histone H3 lysine 9 and lysine 36 demethylase KDM4C (Fig. 5A; Supplemental Fig. S5A; Cloos et al. 2006; Berry and Janknecht 2013). On the other hand, NR2F1 emerged as a potential transcription factor to promote proliferation or cell identity of endocrine-resistant breast cancer cells that lose ESR1 expression (Supplemental Fig. S5A).

Figure 5.

KDM4C is a novel epigenetic dependency in ER⁺ breast cancer. (A) Workflow describing the strategy used. Transcriptional and chromatin architecture data were combined to identify novel dependencies in ER⁺ breast cancer. (B) Western blot of total proteins extracted from T47D shCTRL and shKDM4C cells. Vinculin was used as a loading control. (C) Bar plot showing proliferation differences between T47D cells expressing shCTRL and shKDM4C on day 7 (right) and colony assay with T47D cells expressing shCTRL and shKDM4C (left). (D) Proliferation of shCTRL and shKDM4C cells under different treatment conditions (DMSO, 1 μM tamoxifen, or 1 μM fulvestrant). Statistical significance was determined at the end point. (****) P < 0.0001, (**) P < 0.01, (*) P < 0.05, (ns) not significant. Data are presented as mean ± SEM. (E) Proliferation of parental cells treated with DMSO, 1 μM KDM4 pan inhibitor GC3652, or 1 μM KDM4C-INH-1. Statistical significance was determined at the end point. (****) P < 0.0001. Data are presented as mean ± SEM. (F) Proliferation of T47D^WT/Y537S shCTRL and shKDM4C cells. Statistical significance was determined at the end point. (****) P < 0.0001. Data are presented as mean ± SEM. (G) Proliferation of T47D^WT/Y537S cells under different treatment conditions (DMSO, 1 μM tamoxifen, 1 μM fulvestrant, and 0.5 μM KDM4C-INH-1 alone and in combination with 1 μM tamoxifen or 1 μM fulvestrant). Statistical significance was determined at the end point. (****) P < 0.0001. Data are presented as mean ± SEM. (H) Violin plot of H3K9me3 CUT&RUN signal in T47D siCTRL cells in hormone-deprived (HD) media or after 8 h of estrogen (E2) stimulation. H3K9me3 signal significantly decreased after estrogen stimulation, (***) P < 2.22 × 10⁻¹⁶, Wilcoxon test. (I) Average profile of H3K9me3 CUT&RUN signal in T47D siCTRL cells in HD and E2. (J) Violin plot of H3K9me3 signal in siCTRL and siKDM4C cells in HD and E2. H3K9me3 signal significantly increases after estrogen stimulation in siKDM4C cells compared with shCTRL cells . (***) P-value < 2.22 × 10⁻¹⁶, Wilcoxon test. (K) Heat map of H3K9me3-specific signal in siCTRL and siKDM4C cells. (L) Western blot analysis of acid-extracted histones in siCTRL and siKDM4C parental cells in HD and E2. Histone H3 was used as a loading control. (M) Chromatin interaction heat map showing interactions around TFF1 and ChIP-seq tracks of ER in siCTRL cells after 45 min of estrogen stimulation (top panel) and H3K9me3 signal in siCTRL and siKDM4C cells in HD and E2 conditions in the same region (bottom panel). (N) Significant KDM4C interactions assayed by endogenous immunoprecipitation followed by LC-MS/MS. (Red) SWI/SNF subunits. n = 3 biological independent replicates. (O) Box plot of ATAC signal in control cells at sites that gained H3K9me3 upon KDM4C loss in HD and E2 (P = 0.00018). (P) Violin plot of H3K9me3 signal at sites that gain accessibility upon 8 h of estrogen stimulation (P < 2.22 × 10⁻¹⁶). (Q) Model of chromatin architecture remodeling in endocrine therapy resistance.

We confirmed by both RNA and protein that KDM4C was downregulated during reprogramming (Supplemental Fig. S5B,C) concomitantly with an increase of H3K9me3 but not H3K36me3 (Supplemental Fig. S5D). KDM4C depletion markedly reduced the growth and fitness of ER⁺ breast cancer parental T47D and MCF7 cells (Fig. 5B,C; Supplemental Fig. S5G). We used a second shRNA (#2) that resulted in a modest effect on proliferation in agreement with its lower knockdown efficiency (Supplemental Fig. S5E,F). Importantly, treatment with tamoxifen or fulvestrant was more effective in KDM4C knockdown cells (Fig. 5D; Supplemental Fig. S5H). Additionally, treatment of parental T47D cells with either a pan-KDM4 inhibitor (GC6352) or a selective KDM4C inhibitor (KDM4C-INH-1; KDM4i) significantly impaired their proliferation (Fig. 5E; Supplemental Fig. S5J,K). Because KDM4C expression was not altered in T47D^WT/Y537S cells (Supplemental Fig. S5C), we next depleted KDM4C or treated these cells with KDM4i. Both genetic depletion and pharmacologic inhibition of KDM4C markedly reduced their proliferation and cellular fitness (Fig. 5F,G; Supplemental Fig. S5I). Together, these findings indicate that KDM4C inhibition may represent a promising therapeutic strategy for patients with ER⁺ breast cancer, including both endocrine therapy-sensitive tumors and advanced tumors harboring ESR1 mutations.

Although ER levels were downregulated in cells depleted by KDM4C (Fig. 5B), expression of classical E2-responsive genes remained unaffected (Supplemental Fig. S5L). Additionally, ER recruitment to enhancers and promoters was also largely unchanged (Supplemental Fig. S5M). ER can regulate cell proliferation by nongenomic mechanisms that include activation of the PI3K/AKT and MAPK signaling pathways (Björnström and Sjöberg 2005; Garcia-Martinez et al. 2021). Western blots of key effectors of the PI3K/AKT pathway remained unaffected, and effectors of the MAPK pathway were inconsistent between experiments (data not shown). Therefore, we next sought to determine whether the enzymatic activity of KDM4C regulates cell proliferation. To test this, we performed H3K9me3 CUT&RUN in hormone-deprived (HD) conditions and after E2 stimulation. Notably, H3K9me3 levels were markedly decreased genome-wide upon E2 (Fig. 5H,I), suggesting a potential role of H3K9me3 in repressing pathways activated by E2. We then performed KDM4C ChIP-seq and H3K9me3 CUT&RUN in control and knockdown cells both in HD conditions and after E2 stimulation. Multiple attempts at KDM4C ChIP-seq were unsuccessful, but we were able to demonstrate that the global H3K9 demethylation after E2 stimulation is KDM4C-dependent (Fig. 5J, cf. third and fourth violin plots). Specifically, we found >40,000 new H3K9me3 peaks upon KDM4C depletion and E2 administration (Fig. 5K; Supplemental Fig. S5N). No changes in H3K36me3 were observed by Western blot, suggesting that in this context, KDM4C predominantly demethylates H3K9me3 (Fig. 5L). HOMER motif analysis revealed that these sites were enriched for binding sites of several members of the SOX family (Supplemental Fig. S5O). The SOX family of transcription factors consists of well-established regulators of cell fate decisions during development (Sarkar and Hochedlinger 2013), and in breast cancer, SOX4 is associated with poor survival, increased tumor size, and metastasis formation and is considered a master regulator of EMT (Tiwari et al. 2013; Vervoort et al. 2018). Unfortunately, SOX4 ChIP-grade antibodies are not commercially available to corroborate whether SOX4 binding is regulated by H3K39me3. We next asked whether H3K9me3 was decorating repressed ER target genes and enhancers prior to E2 stimulation. Surprisingly, only a small proportion of ER-bound sites were not decorated with H3K9me3 either before or after E2, indicating that H3K9me3 is not the main mechanism of repression of ER targets (Fig. 5m; Supplemental Fig. S5P). Surprisingly, upon E2, we also found little overlap of ER-bound sites and the PRC2-associated H3K27me3 in the HD condition (Supplemental Fig. S5Q). Indeed, H3K9me3 and H3K27me3 occupy discrete genomic regions (Supplemental Fig. S6A). Although ER binding does not directly overlap with H3K9me3 at classic ER target genes such as PGR and TFF1, H3K9me3 was consistently enriched near their promoters and enhancers (Fig. 5m; Supplemental Fig. S5L). Notably, the H3K9me3 signal intensifies with increasing distance from ER binding sites, suggesting the establishment of a repressive chromatin environment surrounding, but not directly obstructing, ER occupancy (Supplemental Fig. S6B).

Finally, we interrogated the potential contribution of H3K9me3 and H3K27me3 in regions undergoing A/B compartment transitions during evolution of resistance compared with regions that remain unaffected. Although both H3K9me3 and H3K27me3 signals are elevated in regions undergoing compartment transitions, the increase in H3K9me3 is more pronounced than that of H3K27me3, indicating that H3K9me3 plays a dominant role in marking regions with compartment dynamics. These findings suggest that chromatin regions switching compartments during reprogramming are preferentially targeted by H3K9me3, highlighting its importance in driving compartmental reorganization (Supplemental Fig. S6C).

Finally, we reasoned that removal of H3K9me3 would promote chromatin accessibility and hypothesized that KDM4C might act in concert with the SWI/SNF remodeling machinery. To explore this possibility, we performed endogenous immunoprecipitation in parental T47D cells using a KDM4C antibody, with IgG as a negative control, followed by LC-MS/MS analysis. This revealed that KDM4C associates with several subunits of the BAF complex (Fig. 5n; Supplemental Fig. S6D). We then investigated how H3K9me3 dynamics influence chromatin accessibility. ATAC-seq analysis of sites showing at least a twofold increase in accessibility after 8 h of E2 stimulation (Supplemental Fig. S6E) demonstrated that chromatin opening occurred in control cells at loci that became enriched for H3K9me3 upon KDM4C depletion (Fig. 5O). Moreover, we observed elevated H3K9me3 levels precisely at the regions that gained accessibility in control cells after E2 treatment (Fig. 5P).

Our integrated analyses reveal that the KDM4C–BAF axis and H3K9me3 are central regulators of chromatin organization and cellular proliferation in ER⁺ breast cancer. Loss of KDM4C impairs breast cancer cell fitness via modulating chromatin organization and triggers widespread accumulation of H3K9me3, particularly at genomic regions undergoing compartment transitions. Unlike H3K27me3, H3K9me3 predominantly marks these dynamically reorganizing regions, implicating it as a key driver of chromatin compartmentalization during endocrine resistance. Notably, H3K9me3-mediated repression occurs independently of ER binding, suggesting that KDM4C-dependent H3K9 demethylation and chromatin accessibility are essential for maintaining an epigenetic landscape permissive for proliferation and endocrine response. While validation in patient data sets and patient-derived models is still needed, these findings position the KDM4C–BAF–H3K9me3 axis as a promising therapeutic target to counteract resistance and progression in ER⁺ breast cancer (Fig. 5Q).

Materials and methods

Cell lines

T47D primed and reprogrammed cell lines were derived from parental T47D cells (ATCC HTB-113) cultured in long-term estrogen deprivation (LTED) conditions. T47D^WT/Y537S cells were kindly provided by Dr. Steffi Oesterreich. All cells were maintained at 37°C with 5% CO₂ and split every 2–3 days according to ATCC recommendations. Complete culture media for the different cell lines were as follows: for T47D and T47D^WT/Y537S, RPMI 1640 (Lonza 12-167Q) with 10% FBS; for T47D primed and reprogrammed, improved MEM and Richter’s modification without phenol red and glutamine (Corning 10-026-CV) with 10% charcoal:dextran-stripped FBS (BenchMark 100-119); and for MCF7 (ATCC HTB22), Eagle’s minimum essential medium with 10% FBS and 0.01 mg/mL human recombinant insulin. All cell culture media were supplemented with 10,000 U/mL 1× penicillin/streptomycin (Thermo Fisher Scientific 15140-122) and glutaMAX (Thermo Fisher Scientific 35050-061). For the experiments performed after estrogen stimulation, cells were maintained in phenol-red-free media and 5% charcoal-depleted FBS for 72 h and then treated with ethanol (vehicle) or 10 nM E2 (Sigma-Aldrich E2758-250MG) for 8 h.

shRNA and siRNA

293T cells (2 × 10⁶ cells; ATCC CRL-3216) were plated into a 10 cm² plate and transfected 16 h later with 8 μg of pLKO-shRNAs (Addgene 10879 for shCTRL, TRCN0000022058 for shKDM4C, VSC11709 SMARTvector for nontargeting hEF1a-TurboGFP, and V3SH11240-230436410 SMARTvector for lentiviral human KDM4C hEF1a-TurboGFP shRNA), 2.25 μg of pMD2.G, and 4.5 μg of psPAX2 with calcium phosphate. Seventy-two hours after transfection, viral supernatant was collected, passed through a 0.45 μm polyethersulfone filter, and used to transduce cells with 8 μg/mL polybrene (MilliporeSigma TR-1003-G). Cells were allowed to recover in complete cell culture media and were selected with 2 μg/mL puromycin after 24 h (BioGems 5855822). For siRNA experiments, 5 × 10⁵ T47D cells were seeded into 6 well plates in hormone-deprived media 1 day before siRNA transfection and maintained in antibiotic-free culture medium. siRNAs (25 nM; Sigma-Aldrich SIC007 for control and EHU001431 for KDM4C) were transfected using Lipofectamine RNAiMAX (Thermo Fisher Scientific 13778150) following the manufacturer’s instructions.

Analysis of A/B compartments

The function “cooltools eigs-cis” in the cooltools software (Open2C et al. 2024) was used to call the A/B compartments based on the Hi-C data at 100 kb resolution and the GC contents, which were used to determine the orientation of eigenvectors. The A/B compartment changes (A to A, A to B, B to A, and B to B) were reported in alluvial plots, Venn diagrams, and proportional stacked bar plots using R. To plot the saddle plots and saddle strength plots, we followed the steps and computer programs defined in the compartments and saddleplot notebook (https://cooltools.readthedocs.io/en/latest/notebooks/compartments_and_saddles.html) by using the cooltools package. Alluvial plots in Figure 3B were plotted using R after A/B compartments were called from cooltools. To make the scatter plots in Supplemental Figure S2A in R, we extracted the eigenvector values from the A/B compartment calling step using cooltools and then used the values of the same compartment regions of different cells as the X- and Y-axes.

Analysis of TADs

The topologically associating domains (TADs) were called from 40 kb resolution Hi-C contact matrices using the “insulation” function in the software cooltools, which was also used to generate the insulation scores. The ChIP-seq enrichment values of specific genes were computed using pyBigWig (https://github.com/deeptools/pyBigWig) at 40 kb resolution. The Arima-SV pipeline (https://github.com/ArimaGenomics/Arima-SV-Pipeline) was used to determine the significant changes in the A/B compartments, TADs, and loops to find the structural variants of the chromosomal structures. We used the tool Singularity to run the hic_breakfinder program of the Arima-SV pipeline to find all of the Hi-C structural breaks for parental, primed, and reprogrammed cells. We manually classify different types of Hi-C structure breaks.

Finding differentially expressed human TFs associated with genome structural changes

We selected the genes in the compartments that did not change, changed from A to B, and changed from B to A between parental and reprogrammed cells and then plotted the log₂FC values (P < 0.05) of their gene expression with P-values. To generate Figure 5A and Supplemental Figure S4A, we ranked all the human TFs based on the ratio of the average RNA-seq TPM value of reprogrammed cells divided by the average RNA-seq TPM value for parental cells. We only kept the TFs that had a ratio of ≤0.5 and found the TFs that were associated with gain or loss of TAD and the transition of A/B compartments between the primed and reprogrammed conditions.

Analysis of chromatin loops

We detected differential and common Hi-C peaks using Mustache (Roayaei Ardakany et al. 2020) with the parameters “-pt 0.05 -pt2 0.1 -st 0.8” at multiple resolutions. Based on all the loops detected in each condition, we extracted a 41 × 41 Hi-C contact matrix for each loop, with the loop located in the center of the matrix. We then aggregated all 41 × 41 Hi-C contact matrices and generated the aggregation heat maps for all the loops in each condition. To analyze the gene expression profiles of gained chromatin loops, we filtered out all the genes that overlapped with either of the two anchors of a gained loop and then highlighted the differentially expressed genes.

Building the 3D genome structures

Hi-C data resolution was enhanced by using the deep learning algorithm HiCNN (Liu and Wang 2019). Specifically, a 54 layer deep convolutional neural network was trained with lower-resolution Hi-C matrices as input and higher-resolution Hi-C matrices as output or target values. When making predictions, we input the lower-resolution Hi-C matrices obtained from this research into the pretrained convolutional neural network to obtain the resolution-enhanced Hi-C matrices. More details, such as the training and blind test data and evaluation results of the convolutional neural network, are available in the study by Liu and Wang (2019). The 3D genome structures were generated using the classical conversion formula $d_{i,j}=c_{i,j}^{-1∕3}$, where $d_{i,j}$ is the target distance between DNA beads $i$ and $j$, and $c_{i,j}$ is the Hi-C contact values between DNA beads $i$ and $j$. A segment of the chromosome was represented using the “beads on a string” representation, with each bead representing a certain number of DNA base pairs, which was referred to as the resolution. We created a random structure of the segment of the chromosome and put it into a 3D cube with a volume of $V=S^3$, where $S=5n$, with $n$ as the number of beads representing that segment of the chromosome. Simulated annealing and Metropolis–Hastings were used to build the 3D genome structures following the protocols defined by Zhu et al. (2019).

Supplementary Material

Supplemental material is available for this article.

Data availability

All raw and processed NSG data were deposited in the NCBI Gene Expression Omnibus under accession numbers GSE296421-GSE296426. Mass spectrophotometry raw files were deposited in the public repository Chorus (https://www.chorusproject.org) with the project number 1763.