Part IV · Molecular biology and therapeutic vulnerabilities · Chapter 20

Epigenetics, transcriptional control, and lineage identity

Identity is maintained rather than given, and anything maintained can be lost.

1 · DNA methylation and chromatin state across subtypes

A methyl mark on DNA has no fixed meaning. What it does depends on where it sits, and the two common situations point in opposite directions. Hypermethylation of a promoter CpG island silences the gene beneath it. Loss of methylation across gene bodies and repeats destabilises the genome. A tumour can show both at once, and reporting a single global methylation figure discards that distinction.

The integrated view of breast cancer placed methylation alongside copy number, exome sequencing, expression and protein data¹. Combining five platforms resolved four main classes, each internally heterogeneous. Only three genes were mutated in more than 10% of breast cancers overall: TP53, PIK3CA and GATA3. That is the arithmetic behind this whole chapter. The mutational landscape is too sparse to explain the phenotypic range, and the regulatory layer has to carry the remainder.

The clinically sharpest example is BRCA1. In 237 triple-negative tumours from a population-based series, promoter hypermethylation of BRCA1 was twice as frequent as a BRCA1 pathogenic variant in early-stage disease². The hypermethylated cases carried a homologous recombination deficiency phenotype resembling the mutated cases, with similar genome-wide methylation, transcriptional and immune features. Prognosis after adjuvant chemotherapy was similarly improved in both groups. An epigenetic lesion produced the functional state that defines Homologous recombination deficiency, and no germline test will find it. The functional assessment of that state is set out in DNA repair, replication stress, and genomic instability.

2 · Enhancer landscapes and transcription factor dependency

The oestrogen receptor is a transcription factor. Its phenotype is therefore not how much of it is present but where in the genome it sits, and that placement is set by other proteins.

FOXA1 is the determining one. Almost all receptor interactions with chromatin, and almost all oestrogen-driven expression changes, depended on FOXA1 in breast cancer cells³. FOXA1 also shaped chromatin accessibility genome-wide, with CTCF acting as an upstream negative regulator of its binding. The dependency did not disappear with resistance. In tamoxifen-resistant cells receptor binding became independent of ligand while remaining absolutely dependent on FOXA1. Expressing FOXA1 in cells that are not breast cancer cells was enough to alter where the receptor bound.

Mapping receptor binding in human tumours rather than cell lines gave the finding that matters most for this book. Drug-resistant cancers still recruited the receptor to chromatin, but to a partly different set of sites⁴. Tumours from patients who went on to relapse had acquired unique binding regions, and the resulting signature predicted outcome in receptor positive disease only. The redistribution was not attributable to selection of a rare pre-existing subpopulation. The redistribution was FOXA1-mediated reprogramming on a rapid timescale.

That distinction is the open question carried through Estrogen receptor signaling and endocrine resistance. Adaptive cistromic rewiring and clonal selection produce the same observation at a biopsy and imply different actions. A rewired cistrome may be reversible and may be attacked at the chromatin level. A selected clone will not revert. Genetic evolution and reversible cell state change as different problems sets out why the two are currently indistinguishable at progression.

Dependency is not confined to one receptor. GATA3 is one of the three genes mutated above 10% across breast cancer¹, and it is a lineage factor rather than a signalling one. The progesterone receptor, when present, redirects oestrogen receptor binding rather than simply adding to it⁵. Androgen receptor signalling is a transcriptional dependency in a subset of oestrogen receptor negative disease⁶. A receptor result is a statement about a protein's abundance. Its cistrome is a statement about what that protein is doing, and the two are measured differently.

3 · Chromatin remodelers and the BRD8 and EP400 axis

Remodellers do not decide which genes to transcribe. They decide which genes are reachable. Changing one therefore changes the menu available to every transcription factor at once, which is why remodeller lesions produce identity changes rather than growth changes.

ARID1A is the clearest example in endocrine-resistant disease. Inactivating ARID1A mutations are present at high frequency in advanced endocrine-resistant receptor positive breast cancer, and an epigenome-wide CRISPR knockout screen returned ARID1A as the top determinant of resistance to fulvestrant⁷. The mechanism was a change of cell identity rather than a change of signalling. Loss of ARID1A removed SWI/SNF targeting from the genomic sites of the luminal lineage-determining factors, including the oestrogen receptor, FOXA1 and GATA3. Cells switched from receptor-dependent luminal to receptor-independent basal-like.

The BRD8 and EP400 work supplies the missing machinery for a different and long-described escape route. Upregulation of oestrogen receptor signalling has been recognised for a decade as an adaptive survival mechanism in HER2-positive tumours treated with anti-HER2 therapy⁸. What was not known was how the chromatin gets opened. Single-cell transcriptomic and epigenomic profiling of receptor positive, HER2-positive models identified BRD8 as the mediator⁹. BRD8 is an acetyl-lysine reader within the EP400 histone acetyltransferase complex. Its expression rose rapidly after anti-HER2 treatment. Chromatin regions that opened after treatment were enriched for oestrogen receptor, FOX and ETS motifs, and the genes there were activated in a BRD8-dependent manner through EP400-mediated deposition of acetylated H2A.Z.

The functional consequences ran in the expected direction. Depleting BRD8 disrupted the interaction between the receptor and HER2 and increased drug sensitivity. It abolished neratinib-induced receptor activation and restored sensitivity in resistant cells. A three-gene BRD8 signature predicted response to anti-HER2 therapy in two human clinical trials.

This is the same architecture seen elsewhere in the pathway literature. PI3K signalling controls receptor-dependent transcription through the chromatin writer KMT2D¹⁰. The HER2 and HER3 heterodimer acts on hormone response by changing FOXA1 chromatin binding through deacetylation¹¹. In each case a blocked signalling pathway escapes through a chromatin-modifying enzyme that repositions a lineage transcription factor. That explains a feature clinicians observe and rarely have a mechanism for. Escape of this kind appears within days, requires no mutation, and can reverse.

4 · Non-coding RNA and regulatory layers

This is the layer with the highest ratio of published mechanisms to established clinical facts. Saying so first is not dismissal. It is the reason the two findings below are worth separating from the several thousand that are not.

Two findings in breast cancer are worth carrying, and they support different claims. The first is that a long non-coding RNA can act as a targeting module for a chromatin complex. HOTAIR is increased in primary breast tumours and in metastases, and its level in the primary tumour was associated with subsequent metastasis and death¹². Enforced expression retargeted polycomb repressive complex 2 across the genome, altered histone H3 lysine 27 methylation, and increased invasiveness in a manner that depended on that complex. The RNA was not itself repressive. It determined where a repressive complex went.

The second is that the non-coding transcriptome carries subtype information independently. Using a compendium of 58,648 long non-coding RNAs to profile 947 breast cancer samples recovered the known molecular subtypes¹³. The top oestrogen-regulated transcript, DSCAM-AS1, mediated progression and tamoxifen resistance through an interaction with hnRNPL.

Those two claims are independent. A transcript can be a mechanism without being a marker, and it can be a marker without being a mechanism. Most of the confusion in this field comes from reporting one and concluding the other, which is the same failure documented for amplification and protein in HER2 heterogeneity.

The measurement problem here is severe enough to explain why so little has reached the clinic. Long non-coding RNAs are low in abundance, their annotation is not stable between reference releases, and quantification varies with library preparation and with tumour cellularity. A transcript that tracks proliferation or tumour content will look prognostic in any cohort. The standards that a candidate would have to meet are set out in Prognostic and predictive assays and somatic profiling.

5 · Epigenetic memory, heritable drug tolerance, and reversibility

Genetic evolution and reversible cell state change as different problems separates genetic change from reversible state change by a test rather than by inspection. A genetic change survives drug-free passage and a state change does not. That test is correct and it hides the interesting question, which is how a state not written in DNA survives mitosis at all.

The original observation set the shape of the answer. Within drug-sensitive tumour cell lines, a small subpopulation survived acute treatment with a reduction in drug sensitivity of more than a hundredfold¹⁴. Survival required IGF-1 receptor signalling and an altered chromatin state dependent on the histone demethylase KDM5A. The state was transiently acquired and relinquished at low frequency by individual cells, so the population maintained a small reservoir at all times. It could be ablated selectively by IGF-1 receptor inhibitors or by chromatin-modifying agents.

Breast cancer models reproduced this and settled a mechanistic question. Cell-state heterogeneity is high in triple-negative and basal-like disease, and persister states emerged under a wide range of pathway-targeted agents¹⁵. Those states arose through cell-state transitions rather than through selection of pre-existing subpopulations, and the transitions involved dynamic remodelling of open chromatin. Co-treatment with a PI3K and mTOR inhibitor and a BET inhibitor prevented the chromatin changes and prevented the persister state from forming.

Persisters are not one population. Cycling and non-cycling persisters arise from different lineages carrying different transcriptional and metabolic programmes¹⁶. Antioxidant programmes and a shift to fatty acid oxidation were associated with the capacity to keep dividing under drug, which is the metabolic argument made in Redox homeostasis, NRF2, and nucleotide synthesis. The programmes marking cycling persisters were induced in minimal residual disease in human tumours under several targeted therapies.

What is inherited here is a probability rather than a phenotype. A daughter cell inherits a chromatin configuration that makes the tolerant state more or less likely, not the state itself. That is why a population regains sensitivity after an interval off the drug, and it is why the reservoir never quite empties. It also has a scheduling consequence. Continuous exposure and intermittent exposure select differently against a stochastically entered state, which is the mechanistic case behind the strategies in Acting on evolution, intermittent dosing, adaptive therapy, pre-emptive switching.

6 · Lineage plasticity and transcriptional reprogramming under therapy

Lineage identity is not conferred once. It is held in place by continuous transcription factor occupancy and continuous chromatin maintenance. Anything held in place can be released, and therapy is one of the things that releases it.

Two routes have to be kept apart, because they have the same appearance on a slide. One is genetic in origin and epigenetic in mechanism. ARID1A inactivation is a mutation, and what it produces is a luminal to basal-like identity switch with loss of receptor dependence⁷. The other route requires no mutation at all. Persister states in breast cancer models arose by transition rather than by selection, with chromatin remodelling as the mechanism¹⁵. Both end with a receptor-independent cell. Only one of them can revert.

Transcription factors hold the reprogrammed state as well as the original one. Inhibiting NR2F2 restored hormone therapy response in endocrine-refractory breast cancer¹⁷. The HER2Δ16 variant directs luminal cell identity and receptor signalling in HER2-positive disease¹⁸, which makes a receptor variant an input to lineage rather than only to growth. Lineage switching to neuroendocrine and squamous phenotypes is taken up in Lineage switching, ER loss, neuroendocrine and squamous transformation and is not repeated here.

The clinical form of all this is the receptor-negative biopsy taken at progression. It has at least three explanations, and they are not distinguishable on one immunohistochemistry slide. The sampled region may not represent the lesion. A pre-existing receptor-negative compartment may have been selected. Cells that were receptor positive may have been reprogrammed. Receptor discordance and conversion deals with the first, Temporal heterogeneity and clonal evolution with the second, and this chapter with the third.

That last point should carry its evidence. Exemestane was combined with the histone deacetylase inhibitor entinostat or with placebo in 608 patients with advanced hormone receptor positive, HER2-negative breast cancer progressing after a non-steroidal aromatase inhibitor¹⁹. Median progression-free survival was 3.3 months against 3.1 months, with a hazard ratio of 0.87 and a 95% confidence interval of 0.67 to 1.13. Median overall survival was 23.4 months against 21.7 months, a hazard ratio of 0.99. The objective response rate was 5.8% against 5.6%.

The detail that matters is easy to miss. Pharmacodynamic analysis confirmed that the drug hit its target in treated patients¹⁹. Target engagement was achieved and benefit was not. That is the cleanest statement available of where this chapter stands. The regulatory layer is mechanistically central to resistance and to identity in breast cancer, and drugging it broadly has so far failed. The likely reason is developed in Targeting the plastic state rather than the resistant clone and in Epigenetic and transcriptional therapeutics. A chromatin enzyme is used by every cell, so an agent against it has no therapeutic index unless something about the tumour makes the dependency selective. The BRD8 and EP400 axis in Chromatin remodelers and the BRD8 and EP400 axis is interesting precisely because it names such a context rather than a general mechanism.

References

1. Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature 2012 490:61-70. PMID 23000897
2. Glodzik D, Bosch A, Hartman J, et al. Comprehensive molecular comparison of BRCA1 hypermethylated and BRCA1 mutated triple negative breast cancers. Nat Commun 2020 11:3747. PMID 32719340
3. Hurtado A, Holmes KA, Ross-Innes CS, Schmidt D, Carroll JS. FOXA1 is a key determinant of estrogen receptor function and endocrine response. Nat Genet 2011 43:27-33. PMID 21151129
4. Ross-Innes CS, Stark R, Teschendorff AE, et al. Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. Nature 2012 481:389-393. PMID 22217937
5. Mohammed H, et al. Progesterone receptor modulates ERα action in breast cancer. Nature 2015. PMID 26153859
6. Ni M, et al. Targeting androgen receptor in estrogen receptor-negative breast cancer. Cancer Cell 2011. PMID 21741601
7. Xu G, Chhangawala S, Cocco E, et al. ARID1A determines luminal identity and therapeutic response in estrogen-receptor-positive breast cancer. Nat Genet 2020 52:198-207. PMID 31932695
8. Giuliano M, et al. Upregulation of ER signaling as an adaptive mechanism of cell survival in HER2-positive breast tumors treated with anti-HER2 therapy. Clin Cancer Res 2015. PMID 26015514
9. Gao A, et al. Integrated multiomic profiling identifies BRD8/EP400 as a pivotal chromatin module mediating anti-HER2 response in HR+/HER2+ breast cancer. Cancer Res 2026. PMID 41886605
10. Toska E, et al. PI3K pathway regulates ER-dependent transcription in breast cancer through the epigenetic regulator KMT2D. Science 2017. PMID 28336670
11. Wang S, et al. The harmonized activities of HER2–HER3 heterodimer and deacetylated FOXA1 evade hormone response by regulating FOXA1 chromatin binding. Nucleic Acids Res 2025. PMID 41224124
12. Gupta RA, Shah N, Wang KC, et al. Long non-coding RNA HOTAIR reprograms chromatin state to promote cancer metastasis. Nature 2010 464:1071-1076. PMID 20393566
13. Niknafs YS, Han S, Ma T, et al. The lncRNA landscape of breast cancer reveals a role for DSCAM-AS1 in breast cancer progression. Nat Commun 2016 7:12791. PMID 27666543
14. Sharma SV, Lee DY, Li B, et al. A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell 2010 141:69-80. PMID 20371346
15. Risom T, Langer EM, Chapman MP, et al. Differentiation-state plasticity is a targetable resistance mechanism in basal-like breast cancer. Nat Commun 2018 9:3815. PMID 30232459
16. Oren Y, Tsabar M, Cuoco MS, et al. Cycling cancer persister cells arise from lineages with distinct programs. Nature 2021 596:576-582. PMID 34381210
17. Cai Y, et al. Inhibition of NR2F2 restores hormone therapy response to endocrine refractory breast cancers. Sci Transl Med 2025. PMID 40465692
18. Proud H, et al. HER2Δ16 directs luminal cell identity and estrogen receptor signaling in HER2+ breast cancer. Nat Commun 2026. PMID 42297801
19. Connolly RM, Zhao F, Miller KD, et al. E2112: randomized phase III trial of endocrine therapy plus entinostat or placebo in hormone receptor-positive advanced breast cancer. J Clin Oncol 2021 39:3171-3181. PMID 34357781