Chromatin maps reveal two hidden states driving colorectal cancer

Colorectal cancer kills close to a million people every year, and for decades the story of how it arises has been told almost entirely through genetics: a stepwise accumulation of driver mutations in genes such as APC, KRAS, SMAD4 and TP53. APC is lost in roughly 80 percent of patients, while KRAS and BRAF are frequently mutated in microsatellite-stable tumors. Yet a growing body of evidence shows that this genetic narrative cannot explain everything. Genetically identical cancer cells can adopt radically different behaviors, tumors with similar mutational profiles behave in profoundly different ways clinically, and therapies such as MAPK inhibition can push cancer cells into resistant states without any new mutation appearing. A comprehensive review published in Experimental & Molecular Medicine by Dain Kang, Jun Ho Lee and Inkyung Jung of the Korea Advanced Institute of Science and Technology now argues that the missing layer is epigenetic, and that single-cell profiling of chromatin accessibility is finally making it visible.

The technology at the heart of this shift is the single-cell or single-nucleus assay for transposase-accessible chromatin sequencing, known as scATAC-seq or snATAC-seq. Rather than measuring which genes are switched on, as single-cell RNA sequencing does, these methods measure where the genome is physically open for business. Chromatin accessibility reveals the cis-regulatory landscape: the enhancers and promoters that transcription factors can reach, and therefore the regulatory programs a cell is competent to activate. snATAC-seq has proven especially robust for human colorectal cancer specimens because it works with frozen and archived tissue, integrates cleanly with transcriptomic data, and captures an upstream, mechanistic readout of cellular identity that gene expression alone cannot provide. In effect, it reads the control panel of the cell rather than just the lights that happen to be on.

Applying this lens has produced a striking reframing of how colorectal cancer should be classified. The traditional consensus molecular subtype scheme, based on bulk transcriptomes, sorts tumors into four groups, but single-cell RNA sequencing showed that this classification mixes in signals from stromal and immune cells, obscuring the tumor cells themselves. A refined framework, the intrinsic consensus molecular subtype system, isolates the epithelial compartment and collapses the picture into two states: iCMS2 and iCMS3. Crucially, large-scale snATAC-seq analyses have now shown that this dichotomy is written into the chromatin. iCMS2 tumors retain open regulatory elements bearing the footprints of intestinal lineage factors such as HNF4A, CDX2, PPARA and ASCL2, controlling genes for differentiation, lipid metabolism and Wnt-driven stem cell renewal. In essence, these tumors co-opt a corrupted version of normal intestinal stem cell circuitry. iCMS3 tumors instead display a dedifferentiated landscape, with accessibility at stress-responsive and mesenchymal enhancers enriched for AP-1, TEAD, SOX and MAFK motifs, reflecting an active disengagement from intestinal identity in favor of a primitive, stress-tolerant configuration.

The two states even map onto anatomy. iCMS2 tumors arise predominantly in the left colon, iCMS3 mainly in the right, hinting that the split may trace back to differences in embryonic endodermal origin. Patient-derived organoids faithfully maintain their tumor-specific regulatory landscapes through extended culture, and deep learning models trained on pan-cancer chromatin data can predict tumor lineage, differentiation state and oncogenic dependencies from accessibility patterns alone. Together these findings suggest that chromatin encodes a cellular identity fingerprint, one that explains why genetically similar tumors can behave so differently in the clinic.

The review then traces how this regulatory architecture is rewired as tumors progress, proposing a four-stage temporal model. The earliest detectable changes occur as normal mucosa becomes adenomatous polyp: enhancers bearing motifs for differentiation factors such as GATA4/6, KLF4/5 and HOX family members progressively close, while distal regulatory elements tied to Wnt and beta-catenin signaling, including elements targeting MYC, open up. Notably, this Wnt-associated chromatin remodeling appears in adenoma-resident stem cells but not in isolated Apc-null stem cells in mouse models, implying that interactions within the mutant crypt environment drive early epigenetic change. That finding challenges the linear genetic model and positions epigenetic plasticity as a potential initiating layer of tumorigenesis itself.

The most extensive remodeling happens at the transition from late adenoma to invasive carcinoma, where tumors commit to one of the two epigenetic trajectories. In iCMS2 tumors, accessibility at intestinal factor loci such as HNF4A dips during early adenoma formation and then rebounds in established cancer, a stage-dependent shift that may let cells exploit altered metabolic programs while staying partially differentiated. iCMS3 tumors take the opposite route, with widespread AP-1-enriched accessibility; organoid studies suggest AP-1 actively suppresses intestinal lineage factors like CDX2 and HNF4A to drive dedifferentiation toward a fetal state, and NF-kappa-B signaling, which alone can dedifferentiate intestinal epithelium in mice, may fuel the inflammatory character of this state.

A particularly viral concept emerging from this literature is oncofetal reprogramming, the reactivation of fetal gene programs in cancer. LGR5-positive cells behave as classic cancer stem cells, yet ablating them does not shrink tumors because LGR5-negative cells replenish the pool. Lineage tracing shows that it is the LGR5-negative cells that preferentially migrate to metastatic sites and then switch back to the LGR5-positive state to proliferate. Multiple studies now indicate that these LGR5-negative cells reactivate fetal enhancer programs, and in mouse tumoroids resembling iCMS2, chemotherapy with FOLFIRI upregulated oncofetal genes, with the oncofetal cells showing reduced drug sensitivity. Spatial transcriptomics places oncofetal cells at the invasive front of early human tumors, driven by TGF-beta signals from specialized fibroblasts. At the mechanistic level, enhancer accessibility differs sharply between adult and fetal intestine even when promoter accessibility and three-dimensional genome architecture look similar, and YAP/TAZ-TEAD motifs mark the oncofetal state while ASCL2 and TCF-LEF motifs mark the stem state. Whether chemotherapy selects for pre-existing oncofetal cells or induces them, and whether YAP/TAZ signaling operates as a shared, subtype-agnostic vulnerability across both iCMS states, remain key open questions.

Metastasis adds a further twist: organotropic chromatin remodeling. In mouse models of liver colonization, snATAC-seq shows metastatic cells progressively losing access to colon-specific enhancers bearing IRF1, ELF1 and STAT1 motifs while gaining access to liver-specific elements bearing HNF4A, FOXA2 and CTCF motifs, generating hybrid cells that express both epithelial and hepatocyte-like programs. The acquisition of HNF4A-driven landscapes in liver metastases echoes the iCMS2 signature, and the clinical observation that left-sided, iCMS2-enriched tumors preferentially spread to the liver raises the tantalizing possibility that primary tumor epigenotype predicts metastatic destination. Co-accessibility analyses suggest that long-range enhancer-promoter contacts expand around HNF4A and FOXA2, wiring metastatic cells into liver-specific growth and metabolic signals.

The tumor microenvironment is not a bystander in this process. Cancer-associated fibroblasts split into tumor-promoting subtypes with accessibility enriched for RUNX1, AP-1, NF-kappa-B and HIF1A motifs, and tumor-restraining subtypes that retain homeostatic extracellular matrix programs and are progressively depleted during malignant transformation. Normal fibroblasts acquire CAF-like histone marks when exposed to tumor-conditioned media, confirming that this divergence is epigenetic rather than genetic. In the immune compartment, CD8-positive T cells lose accessibility at the TCF7 locus, a pioneer factor for T cell stemness, marking terminal exhaustion that is already partially established in adenomas, suggesting immunosuppression is an enabling event rather than a late consequence. Regulatory T cells split into IL-10-positive and IL-10-negative subtypes with opposite prognostic implications, potentially resolving the paradox that T regulatory infiltration predicts good outcomes in colorectal cancer but poor outcomes elsewhere. Macrophages divide into C1QC-positive homeostatic and SPP1-positive tumor-enriched subtypes with distinct regulatory architectures. The authors propose viewing all of this as an iCMS-defined epigenomic ecosystem, in which malignant, stromal and immune compartments co-evolve through chromatin-level crosstalk.

Genomic alterations and chromatin states also feed back on each other. Recurrent noncoding mutations rewire transcription factor binding, as in the ETS2 enhancer that promotes epithelial-mesenchymal transition, or TERT promoter mutations that create de novo ETS sites and reactivate telomerase. Structural variations hijack enhancers to supercharge MYC and CCND1, and extrachromosomal DNA circles carry amplified oncogenes with their enhancers as mobile regulatory platforms, segregating nonrandomly during cell division. The same mutation can even have opposite accessibility effects in malignant versus stromal cells, as deep learning models trained on single-cell chromatin data have revealed. Translationally, these insights are already bearing fruit: DNA methyltransferase and EZH2 inhibitors can reopen silenced immune gene enhancers, and early-phase trials combining low-dose DNMT inhibitors with anti-PD-1 antibodies show promising responses in microsatellite-stable colorectal cancer, a population that normally ignores immunotherapy. Cell-free ATAC-seq promises liquid biopsies that read a tumor’s global regulatory state rather than single mutations, spatial epigenomic methods are mapping regulatory domains in intact tissue, and AI foundation models trained on single-cell epigenomes may one day simulate the iCMS2-to-iCMS3 transition and nominate the transcription factors whose manipulation could reverse it. The review’s boldest suggestion is that a therapy capable of reversing the epigenetic clock, restoring oncofetal and dedifferentiated cells to mature, lineage-committed states, could constrain tumor plasticity at its source, turning the cancer’s own regulatory logic against it.

Subject of Research: Single-cell epigenomic profiling of chromatin accessibility in colorectal cancer

Article Title: Single-cell epigenomics of colorectal cancer

Article References: Kang, D., Lee, J. H., & Jung, I. (2026). Single-cell epigenomics of colorectal cancer. Experimental & Molecular Medicine. https://doi.org/10.1038/s12276-026-01855-4

Image Credits: AI Generated

DOI: 10.1038/s12276-026-01855-4

Keywords: colorectal cancer, single-cell epigenomics, snATAC-seq, chromatin accessibility, iCMS subtypes, oncofetal reprogramming, tumor microenvironment, cancer-associated fibroblasts, T cell exhaustion, enhancer hijacking, extrachromosomal DNA, liquid biopsy