EZH2 directs HER2+ breast cancer progression through the modulation of epithelial plasticity

Abstract

Breast cancer remains a leading cause of death among women, with the HER2+ subtype being particularly aggressive due to acquired resistance to HER2-targeted therapies. Enhancer of Zeste Homolog 2 (EZH2), the catalytic subunit of Polycomb Repressive Complex 2, represses the expression of genetic programs crucial for differentiation, proliferation, and apoptosis. To investigate the role of EZH2 in HER2+ tumor progression, we crossed a genetically engineered mouse model of HER2-driven breast cancer with a conditional Ezh2 knockout strain and showed that Ezh2 is essential for accelerating tumor initiation and metastatic dissemination. Combined bulk and single cell RNA sequencing analyses revealed a significant downregulation of basal cell populations in the absence of Ezh2, and an upregulation of luminal progenitor cell populations, driven by crucial transcription factors such as Esr1. Further, inhibition of EZH2 in vitro resulted in increased expression of ER in HER2+ human breast cancer cell lines and conferred sensitivity to Tamoxifen. These findings demonstrate that EZH2 dictates cancer plasticity and provides rationale for combining EZH2 inhibitors with endocrine therapies to improve HER2+ breast cancer outcomes.

Synopsis

This study establishes a role for EZH2 in HER2+ breast cancer progression through driving basal identity. Loss or inhibition of EZH2 reactivates a luminal cellular differentiation program, sensitizing HER2+ breast cancer cells to ER-targeted therapy.

• Genetic ablation of Ezh2 in a HER2-driven mouse model inhibits tumor initiation, progression and lung metastasis.

• Ezh2 loss in HER2+ tumors result in loss of basal/EMT cells and acquisition of a luminal phenotype.

• ChIP analyses reveal Esr1 as a direct target of PRC2-mediated H3K27me3 repression.

• EZH2 inhibitors and Tamoxifen synergistically reduce proliferation of human HER2+ breast cancer cell lines in vitro.

This study establishes a role for EZH2 in HER2+ breast cancer progression through driving basal identity. Loss or inhibition of EZH2 reactivates a luminal cellular differentiation program, sensitizing HER2+ breast cancer cells to ER-targeted therapy.

Introduction

Breast cancer continues to be a leading cause of mortality among women worldwide largely due to its molecular heterogeneity and the development of resistance to therapies. Breast cancer is clinically classified into four major molecular subtypes based on the expression of estrogen receptor alpha (ERα), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2/ERBB2): Luminal A and B (ERα and/or PR-positive), HER2-enriched, and triple-negative breast cancer (TNBC), which lacks expression of all three markers. Among these subtypes, HER2-positive (HER2+) tumors account for 15–20% of cases and are characterized by aggressive clinical behaviour, with worse prognosis and a heightened propensity for metastasis compared to luminal subtypes (Exman and Tolaney, 2021). Despite advancements in early detection methods and treatment strategies, 20–30% of those initially diagnosed with early-stage breast cancer ultimately die from metastases (Early Breast Cancer Trialists’ Collaborative G, 2005; Riggio et al, 2021), highlighting a need for improvement in understanding and predicting metastatic progression, as well as developing novel treatment strategies to prevent or eradicate metastatic disease.

The treatment of HER2+ breast cancer has been greatly improved due to the development of HER2-targeted therapies, however multiple mechanisms of resistance can be acquired, leading to disease progression and metastasis (Swain et al, 2023). This underscores the need for alternative treatment strategies to overcome resistance, which will require a more extensive understanding of the molecular mechanisms of HER2-dependent breast cancer dissemination and of acquired resistance to HER2-targeted agents. Recently, HER2+ breast cancers have been stratified into five distinct molecular subtypes based on transcriptomic profiling of tumor samples from patients with HER2+ breast cancer: ERBB2-dependent, immune-enriched, luminal, mesenchymal/stroma-enriched, and proliferative/metabolic-enriched (Rediti et al, 2024). Interestingly, the luminal HER2+ breast cancer subtype was characterized by high ERα mRNA (ESR1) expression and concurrent positive ERα protein expression, representing a subset of patients which are HER2+ER+ (Rediti et al, 2024). 10% of all breast cancer cases are HER2+ER+, with these cases displaying reduced responsiveness to HER2-targeted therapies (Montemurro et al, 2012; Zhao et al, 2018). Accordingly, despite having a better prognosis than other subtypes, the luminal HER2+ subtype had only a 9% probability of achieving complete response to standard-of-care treatment with the anti-HER2 monoclonal antibody (mAb) Trastuzumab, the lowest of all 5 subtypes (Rediti et al, 2024). Collectively, these findings demonstrate the clinical significance of ESR1 expression within the HER2+ subtype, as it represents one mechanism of resistance to HER2-targeted therapy. By better defining the molecular mechanisms which dictate a luminal program within the HER2+ subtype, we can potentially develop novel therapeutic combinations that target these poorly responsive HER2+ER+ breast cancer tumors.

A key epigenetic regulator which is overexpressed and implicated as a pro-tumorigenic in breast cancer is EZH2, a lysine methyltransferase which acts as the catalytic subunit of the polycomb repressor complex 2 (PRC2) (Holm et al, 2012; Kim and Roberts, 2016; Kleer et al, 2003). EZH2’s canonical role mediates chromatin compaction and transcriptional silencing through tri-methylation of histone H3 at lysine 27 (H3K27me3) and it is widely recognized for its role in embryonic development and control of cell lineage specification, stem cell differentiation and mammary gland development (Margueron and Reinberg, 2011; Michalak et al, 2013; Pal et al, 2013). Specifically in breast cancer, accumulating evidence identifies EZH2 as a key dictator of multiple stages of breast cancer progression, particularly in more aggressive subtypes (Hirukawa et al, 2019; Hirukawa et al, 2018; Kleer et al, 2003; Liu et al, 2023; Yu et al, 2023). We previously showed that, in a luminal B genetically engineered mouse model (GEMM), EZH2 was essential for driving metastatic dissemination through epigenetic repression of an FOXC1-dependent anti-metastatic transcriptional network (Hirukawa et al, 2018). More recently, in the same Luminal B GEMM we showed that EZH2 promotes the initiation phase of breast cancer progression through activation of mTOR and Wnt signalling pathways (Liu et al, 2023). Other evidence argues that EZH2 can modulate cellular plasticity as it is associated with poorly differentiated carcinomas and leads to increased risk of metastasis (Alford et al, 2012; Raaphorst et al, 2003). Despite these insights, its specific role in regulating cell plasticity and metastasis in HER2+ breast cancer remains unclear and requires further investigation.

By developing a GEMM that combines temporally controlled expression of HER2 (Attalla et al, 2023) and deletion of Ezh2 specifically in the mammary epithelium, we showed that mammary epithelial expression of Ezh2 is required for efficient HER2-driven tumor initiation, and that its deletion severely attenuates metastatic progression. Transcriptional profiling of the resulting tumors identified distinct pathways regulated by Ezh2, linking the dramatic loss of metastatic capacity with a Mesenchymal to Epithelial Transition (MET) in Ezh2 deficient tumors. The marked reduction in markers of mesenchymal identity and increased epithelial marker expression correlated with evidence of a shift from a basal to a luminal epithelial cell lineage, with activation of Esr1 expression and coordinated increases in luminal markers including cytokeratin 8 (Ck8), E-cadherin, Gata3 and Foxa1. Moreover, we showed that EZH2 expression is inversely correlated with ESR1 expression in human patients, both across subtypes and within the HER2-enriched subtype. Treatment of human HER2+ cell lines with well-established EZH2 methyltransferase inhibitors increased the expression of ERα and induced Tamoxifen sensitivity in these cells, including in cell lines that had acquired resistance to the HER2 tyrosine kinase inhibitor (TKI) Lapatinib, or have established mechanisms of resistance to Trastuzumab. Together, our results identify EZH2 as a key driver of early tumor progression and a critical regulator of the Epithelial to Mesenchymal Transition (EMT) and metastasis in HER2-driven tumor progression. Further, we provide rationale for combining EZH2 inhibitors with ER-targeted therapy as a viable alternative strategy for patients with HER2+ breast cancer, particularly following HER2-targeted therapy resistance.

Results

Genetic ablation of Ezh2 inhibits tumor progression and lung metastasis in an HER2-driven mouse model

Previous studies have shown that EZH2 expression is elevated in many breast cancer subtypes (Holm et al, 2012; Kleer et al, 2003). Utilizing an online dataset, we confirmed that EZH2 expression levels are significantly elevated in breast cancer (Appendix Fig. S1A), particularly in the more aggressive HER2+ and TNBC subtypes (Appendix Fig. S1B), and that elevated EZH2 expression is associated with poor overall survival in breast cancer (Appendix Fig. S1C). Interestingly, we noted that EZH2 expression is correlated with ERBB2 expression, indicating a strong connection between HER2-driven tumorigenesis and EZH2 overexpression (Appendix Fig. S1D). To further investigate the function of EZH2 in HER2+ breast cancer, we crossed a doxycycline-inducible HER2-driven GEMM (EIC), which faithfully recapitulates the progression of HER2+ breast cancer in human patients, with a conditional Ezh2^fl/fl model (Attalla et al, 2023; Shen et al, 2008). In this inducible GEMM, upon addition of doxycycline, the activated reverse tetracycline-dependent transactivator (rtTA) binds to the tetracycline operator (TetO) to drive the co-expression of HER2 and Cre recombinase, leading to the excision of the loxP-flanked Ezh2 allele exclusively in the mammary epithelium (Fig. 1A). Tumor onset was significantly delayed in Ezh2 deficient mice compared to Ezh2 proficient counterparts upon doxycycline induction (Fig. 1B; Appendix Fig. S2A). After tumor onset, a significant delay in tumor growth was also observed (Fig. 1C; Appendix Fig. S2B), with Ezh2 null tumors taking around 16 weeks to reach endpoint compared to 11 weeks for Ezh2 wild-type tumors (Appendix Fig. S2C). No difference in the number of tumor foci (Appendix Fig. S2D) or tumor burden (Appendix Fig. S2E) was noted at endpoint.

Figure 1. Genetic ablation of Ezh2 inhibits mammary tumor progression and lung metastases.

(A) Schematic of the doxycycline-induced ERBB2-IRES-CRE (EIC) and the conditional Ezh2 (Ezh2^fl/fl) model. (B) Kaplan–Meier survival curve of mammary tumor onset in EIC mice with wild-type (Ezh2^wt/wt), heterozygous (Ezh2^wt/fl) or homozygous (Ezh2^fl/fl) alleles. ****P < 0.0001, log-rank test. (C) Kaplan–Meier survival curve from tumor onset to endpoint for mice with wild-type (Ezh2^wt/wt), heterozygous (Ezh2^wt/fl) or homozygous (Ezh2^fl/fl) alleles. *P < 0.05, log-rank test. (D) Top: Immunofluorescence staining of Ezh2, H3K27me3 and HER2 on 12-week induced mammary glands from Ezh2^wt/wt and Ezh2^fl/fl mice (n = 6 mice per genotype). Scale bar: 100 µm. Bottom: Quantification performed using HALO, plotted as mean ± SD. *P < 0.05, **P < 0.01, unpaired, two-tailed Student’s t test. (E) Left: Representative images of H&E-stained sections obtained at 12-weeks (12 W) and endpoint. Scale bar: 100 µm. Right: Quantification of hyperplasic/tumor area of Ezh2^wt/wt (n = 12) and Ezh2^fl/fl (n = 6) mice at 12 W using HALO, plotted as mean ± SD. *P< 0.05, unpaired, two-tailed Student’s t test. (F) Top: Immunofluorescence staining of Ki67 and HER2 on 12-week induced mammary glands from Ezh2^wt/wt and Ezh2^fl/fl mice (n = 6 mice per genotype). Scale bar: 100 µm. Bottom: Quantification performed using HALO, plotted as mean ± SD. *P < 0.05, unpaired, two-tailed Student’s t test. (G) Left: Representative images of H&E-stained lung sections obtained at endpoint. Top scale bar: 5 mm. Bottom scale bar: 1 mm. Top right: Percentage lung metastatic penetrance of Ezh2^wt/wt (n = 17) and Ezh2^fl/fl (n = 35) endpoint mice, plotted as mean ± SD. **P < 0.01, Mann–Whitney test. Bottom right: Quantified percentage burden of lung metastasis from Ezh2^wt/wt and Ezh2^fl/fl endpoint mice (n = 10 mice per genotype), plotted as mean ± SD. **P < 0.01, unpaired, two-tailed Student’s t test. All statistical analysis: Individual P values are provided in Dataset EV1. Source data are available online for this figure.

To further investigate the role of Ezh2 in HER2-driven tumorigenesis, we analysed early mammary epithelial hyperplastic lesions that were present at 12-weeks (12W) post-doxycycline induction. Ezh2 deficient HER2+ epithelial cells exhibited significantly reduced levels of the H3K27me3 mark (Fig. 1D). The overall burden of hyperplastic mammary lesions at 12-weeks post-induction was also significantly reduced in the Ezh2 null cohort compared to Ezh2 proficient controls (Fig. 1E). Consistent with the observed delay in tumor progression, we showed that Ezh2 deficient mammary glands exhibited a marked decrease in Ki67 expression and EdU incorporation, indicating a block in proliferation (Figs. 1F and EV1A). Importantly, there was no difference in HER2 expression in epithelial cells, indicating that the delay in tumor onset was not due to reduced oncogene expression (Fig. EV1A). A cell cycle progression block specifically into the S phase was evidenced by significant loss of Pcna and Cyclin D1 expression in Ezh2 deficient hyperplastic lesions (Fig. EV1A,B). This was further supported by loss of LaminB1 expression and a reduction in phosphorylated Rb, a phenotype consistent with cellular senescence (Freund et al, 2012; Gonzalez-Gualda et al, 2021) (Fig. EV1C). Collectively, these data argue that a decrease in hyperplastic lesions following Ezh2 loss is associated with coordinated loss of epithelial cell cycle progression markers and subsequent cell cycle arrest. To evaluate the impact of mammary epithelial-specific ablation of Ezh2 on metastasis, lungs from endpoint mice with equivalent tumor burden were collected to assess the presence of metastatic lesions. Tumor bearing mice lacking Ezh2 were found to have a lower metastatic penetrance (28.6%) compared to control mice (70.6%), as well as significantly lower metastatic burden (Fig. 1G). Together, these observations demonstrate that Ezh2 function plays a critical role in both the initiation and metastatic phases of mammary tumor progression.

Figure EV1. Ezh2 ablation decreases cell cycle progression at early mammary tumorigenesis.

(A) Flow cytometric analysis of 12-week doxycycline-induced Ezh2^wt/wt (n = 6) and Ezh2^fl/fl (n = 8) mice, minimum of 250,000 cells per sample, plotted as mean ± SD. (B) Left: Immunofluorescence staining of Pcna, Cyclin D1 and HER2 on 12-weeks post-doxycycline induced mammary glands collected from Ezh2^wt/wt and Ezh2^fl/fl mice (n = 10 mice per genotype). Scale bar: 100 µm. Right: Quantification using HALO, plotted as mean ± SD. (C) Left: Immunofluorescence staining of LaminB1, p-Rb and HER2 on 12-weeks post-doxycycline induced mammary glands collected from Ezh2^wt/wt and Ezh2^fl/fl mice (n = 6 mice per genotype). Scale bar: 100 µm. Right: Quantification using HALO, plotted as mean ± SD. All statistical analysis: ns not significant, **P < 0.01, unpaired, two-tailed Student’s t test. Individual P values are provided in Dataset EV1. Source data are available online for this figure.

Ezh2 is required for migration and invasive phenotypes of HER2+ tumors

To further investigate the observed block in the metastatic cascade in Ezh2 deficient mice, tumors endpoint from both Ezh2 proficient and Ezh2 deficient mice were examined for the expression of HER2 and Ezh2, as well as PRC2 activity (H3K27me3). We confirmed complete loss of Ezh2 protein with a corresponding reduction in the H3K27me3 mark in HER2+ epithelial cells (Fig. 2A). The observed remaining H3K27me3 in Ezh2 null mammary glands likely reflects compensation by the related Ezh1 methyltransferase that is present in these tissues (Shen et al, 2008) (Fig. EV2A), however Ezh1 is unable to fully compensate for Ezh2 loss (Fig. EV2B). HER2 expression and activity were maintained (Fig. 2B), and multiple signaling pathways related to HER2 activity (Akt, Erk1/2, Ampk) were unaltered, apart from an increase in Akt phosphorylation on S473 (Appendix Fig. S3A). Furthermore, no difference in mTOR signaling (phosphorylation of S6 and 4E-bp1) was observed (Appendix Fig. S3B,C). These data demonstrate that HER2 downstream signaling remains active in Ezh2 deficient tumors. In addition, unlike the early preneoplastic lesions, Ezh2 deficient endpoint tumors have no difference in Ki67 or Ccp3 levels compared to the wild-type Ezh2 tumors, indicating that they have overcome the initial proliferative defect (Appendix Fig. S3D).

Figure 2. Ablation of Ezh2 reduces cell migration ability and EMT.

(A) Top: Immunohistofluorescence staining for Ezh2, H3K27me3, and HER2 on endpoint tumors from Ezh2^wt/wt and Ezh2^fl/fl mice (n = 8 mice per genotype). Scale bar: 100 µm. Bottom: Quantification using HALO, plotted as mean ± SD. (B) Top: Representative immunoblot of Ezh2, p-HER2 and HER2 expression in endpoint tumors from Ezh2^wt/wt (n = 6) and Ezh2^fl/fl (n = 6) mice. Bottom: Quantification normalized to α-Tubulin, plotted as mean ± SD. Each lane represents a biological sample. All bands shown are from the same blot. (C) Top: Representative immunoblot of E-cadherin and Snail expression in endpoint tumors from Ezh2^wt/wt (n = 6) and Ezh2^fl/fl (n = 6) mice. Bottom: Quantification, plotted as mean ± SD, of E-cadherin (normalized to β-Actin) and Snail (normalized to α-Tubulin) expression in endpoint tumors from Ezh2^wt/wt (n = 6) and Ezh2^fl/fl (n = 6) mice. Each lane represents a biological sample. E-cadherin and β-Actin bands were from blots processed in parallel, Snail and α-Tubulin bands were from the same blot. (D) Boyden chamber migration assay using Ezh2^wt/wt (n = 4) and Ezh2^fl/fl (n = 3) cell lines (n = 2–3 technical replicates per cell line). Scale bar: 500 µm. Quantification performed using Fiji and plotted as mean ± SD. (E) Tumor growth rate following mammary fat pad injections of Ezh2^wt/wt (n = 3) and Ezh2^fl/fl (n = 4) cell lines (n = 2 technical replicates per cell line), plotted as mean ± SD. *P < 0.05, multiple, unpaired, two-tailed Student’s t-tests. (F) Top: Representative images of H&E-stained lung sections obtained after tail vein injection of Ezh2^wt/wt (n = 4) and Ezh2^fl/fl (n = 4) cell lines into NSG mice. Top scale bar: 5 mm. Bottom scale bar: 1 mm. Bottom: Quantified total burden of lung metastasis area and number of metastatic lesions (mets) >0.1 mm², plotted as mean ± SD. *P < 0.05, Mann–Whitney test. All statistical analysis unless stated: ns not significant, *P < 0.05, ****P < 0.0001, unpaired, two-tailed Student’s t test. Individual P values for all analysis are provided in Dataset EV1. Source data are available online for this figure.

Figure EV2. Ezh1 cannot compensate for Ezh2 loss in Ezh2^fl/fl tumors.

(A) Left: Immunofluorescence staining of Ezh2, Ezh1 and H3K27me3 on Ezh2^wt/wt and Ezh2^fl/fl endpoint tumors (n = 6 mice per genotype). Scale bar: 100 µm. Right: Quantification of Ezh1 using HALO, plotted as mean ± SD. (B) Left: Immunoblot of H3K27me3 and H3 expression H3K27me3 in histone lysates extracted from endpoint tumors of Ezh2^wt/wt (n = 6) and Ezh2^fl/fl (n = 6) mice. Right: Quantification of H3K27me3 normalized to H3. Each lane represents a biological sample. Samples derive from the same experiment, where blots were processed in parallel. Data plotted as mean ± SD. All statistical analysis: ns not significant, ****P < 0.0001, unpaired, two-tailed Student’s t test. Individual P values are provided in Dataset EV1. Source data are available online for this figure.

One of the hallmarks of metastatic progression involves a cellular process known as Epithelial to Mesenchymal Transition (EMT) (Wang and Zhou, 2011). To explore whether the metastatic defect observed in the Ezh2 deficient tumors involve alterations in the EMT program, we investigated the expression levels of the epithelial marker, E-cadherin, and master EMT transcription factor, Snail (Cano et al, 2000). Loss of Ezh2 in HER2+ tumors was associated with an increase in E-cadherin levels with a corresponding decrease in Snail (Fig. 2C). To further investigate the consequences of this Mesenchymal to Epithelial Transition (MET), tumor-derived cell lines were generated from Ezh2 proficient and Ezh2 deficient tumors for further in vitro and in vivo analyses. Consistent with acquisition of an MET phenotype, the Ezh2 deficient cell lines exhibited impaired migration (Fig. 2D) and, upon mammary fat pad injection in immunodeficient mice, were delayed in both tumor initiation and outgrowth (Fig. 2E). We validated that this was not due to a loss in proliferative capacity, with no significant difference in EdU incorporation (Appendix Fig. S4A) or growth rate via IncuCyte proliferation assay (Appendix Fig. S4B). Because Ezh2 deficient cells maintain H3K27me3 levels (Fig. 2A; Appendix Fig. S4A), we were interested to see how Ezh2 proficient cells respond to complete H3K27me3 loss. Interestingly, proliferation was not affected in Ezh2 proficient cells upon treatment with EED inhibitor A395 or SAM-competitive inhibitor EPZ6438, despite more complete H3K27me3 loss (Appendix Fig. S4C,D). This argues that neither Ezh2/Ezh1 canonical nor Ezh2 non-canonical functions are responsible for proliferative capacity of HER2+ tumors and reflects the ability of cells to escape dependence on Ezh2 function.

Finally, to evaluate whether Ezh2 deficient HER2+ cells were capable of colonization and outgrowth in the lungs, we used an in vivo experimental metastasis assay. By contrast to wild-type HER2+ cells, which induced metastatic lesions in 100% of recipient mice, the Ezh2 deficient cells only formed metastatic lesions in 50% of animals which in turn were significantly smaller (Fig. 2F). Taken together, these data argue that Ezh2 plays a crucial role in the metastatic progression in HER2-driven breast cancer by promoting multiple steps of the metastatic cascade, including migration and colonization of the secondary site.

Ezh2 loss in HER2+ tumors results in acquisition of a luminal estrogen receptor (Esr1) positive phenotype

To elucidate the potential Ezh2 transcriptional targets altered by deletion of Ezh2, we performed transcriptomic analysis (RNA-Seq) of wild-type and Ezh2 deficient endpoint tumors. Ezh2 ablation was associated with an altered transcriptomic profile (Fig. 3A) comprising 264 upregulated genes and 420 downregulated genes relative to wild-type controls (Fig. 3B). Downregulated genes are likely to reflect targets which are controlled through non-canonical Ezh2 functions, such as through interaction with transcription factors or other oncogenic partners (Zimmerman et al, 2023). Multiple bioinformatic analyses of this transcriptional signature revealed a significant downregulation of four common EMT signatures (Figs. 3C and EV3A). Accordingly, EMT was identified as the most significantly downregulated hallmark in Ezh2 deficient samples (Fig. EV3B). Additionally, the downregulated genes were most significantly associated with tissue invasion and metastasis cancer hallmark, as defined by Hanahan and Weinberg (Hanahan and Weinberg, 2011), which is consistent with the EMT defect observed in Ezh2 deficient tumors (Fig. EV3C). Transcription factor motif data analyses (Chen et al, 2013; Kuleshov et al, 2016; Xie et al, 2021) identified TP63, a transcription factor and marker of basal cells (Laakso et al, 2005), as a major candidate regulator of the EMT signatures with decreased expression in Ezh2 deficient samples (Fig. 3D). Conversely, genes upregulated in Ezh2 deficient samples were associated with ESR1 and its transcriptional targets (Figs. 3E and EV3D). To validate the bioinformatic data, we performed RNA In Situ Hybridization (ISH) with a Esr1 specific probe on both Ezh2 proficient and deficient tumor samples. Consistent with the acquisition of a luminal state, we found a significant upregulation of HER2+Esr1+ cells in Ezh2 deficient tumors (Fig. 3F). We confirmed significant upregulation of Esr1 transcript as well as three ERα target genes (Myc, Tgfα, Bcl2) in Ezh2 deficient HER2 samples, while Rprm, which is repressed by ERα, was significantly downregulated (Fig. 3G,H). Importantly, the transcriptomic signature associated with Ezh2-null tumors was associated with better overall patient survival (Fig. EV3E). Collectively, these analyses argue that loss of Ezh2 in the mammary epithelium leads to the acquisition of an Esr1-driven luminal identity (Dou et al, 2017; Guttilla et al, 2012) and support previous findings that a luminal HER2+ subtype has better prognostic outcome (Rediti et al, 2024).

Figure 3. Ezh2 deficient tumors lose EMT ability and retain a more luminal identity through upregulation of Esr1 activity.

(A) Heatmap of differentially expressed genes identified by RNA-Seq from endpoint tumors from Ezh2^wt/wt (n = 5) and Ezh2^fl/fl (n = 5) tumors. (B) Volcano plot of differentially expressed genes in Ezh2^fl/fl (n = 5) group vs Ezh2^wt/wt (n = 5) groups. Red points = Padj < 0.05. (C) Differentially expressed EMT signature Z-Scores of Ezh2^wt/wt (n = 5) and Ezh2^fl/fl (n = 5) groups, plotted as minimum to maximum, centre: median, bounds: 25th and 75th percentiles. *P < 0.05, multiple, unpaired, two-tailed Student’s t tests. (D) ENCODE and ChEA analysis of transcriptional factors associated with the significantly downregulated genes in Ezh2^fl/fl (n = 5) group vs Ezh2^wt/wt (n = 5) group. (E) ENCODE and ChEA consensus analysis of transcriptional factors associated with the significantly upregulated genes in Ezh2^fl/fl (n = 5) group vs Ezh2^wt/wt (n = 5) group. (F) Left: RNA-ISH performed using Esr1 probe and Immunohistofluorescence for HER2 on Ezh2^wt/wt (n = 9) and Ezh2^fl/fl (n = 12) samples. Scale bar: 100 µm. Right: Quantification performed using HALO, plotted as mean ± SD. Each datapoint represents an endpoint tumor. (G) qRT-PCR analysis of Esr1 expression normalized to β-Actin in endpoint tumors (n = 6 biological samples per genotype, n = 2 technical replicates), plotted as mean ± SD. (H) qRT-PCR analysis of ERα target genes normalized to β-Actin in endpoint tumors (n = 6 biological samples per genotype, n = 2 technical replicates), plotted as mean ± SD. All statistical analysis: *P < 0.05, **P < 0.01, ****P < 0.0001, unpaired, two-tailed Student’s t test. Individual P values are provided in Dataset EV1. Source data are available online for this figure.

Figure EV3. RNA sequencing reveals an anti-metastatic phenotype.

(A) Gene Set Enrichment Analysis for EMT Signatures in Ezh2^fl/fl (n = 5) vs Ezh2^wt/wt (n = 5) tumors. (B) MsigDB hallmarks 2020 analysis of downregulated pathways in Ezh2^fl/fl (n = 5) vs Ezh2^wt/wt (n = 5) tumors. (C) Cancer hallmark analysis of significantly downregulated genes in Ezh2^fl/fl (n = 5) tumors, created with cancerhallmarks.com. (D) MsigDB hallmarks 2020 analysis of upregulated pathways in Ezh2^fl/fl (n = 5) vs Ezh2^wt/wt (n = 5) tumors. (E) Kaplan–Meier survival curve and median survival of the differentially expressed genes in Ezh2^fl/fl vs Ezh2^wt/wt tumors in breast cancer patients (n = 1210) from TCGA database. *P < 0.05, log-rank test. All statistical analysis: Individual P values are provided in Dataset EV1. Source data are available online for this figure.

To confirm the effects of Ezh2 deficiency on mammary epithelial cell lineage identity in HER2+ tumors, we analysed multiple well-established luminal and basal markers at the transcript and protein levels. As expected, the luminal markers Gata3, Foxa1 and Notch3 (Albergaria et al, 2009; Dou et al, 2017) were upregulated in Ezh2 deficient tumors (Fig. 4A), while basal markers Krt5 and Cd44 were significantly downregulated (Fig. 4B). Ezh2 deficient HER2+ tumors also expressed higher levels of Gata3 and Foxa1 protein, as well as the epithelial marker E-cadherin (Fig. 4C). Consistent with the importance of Ezh2 as a negative regulator of luminal differentiation, Ezh2 deficient tumors expressed significantly elevated levels of the Notch3 internal cleaved domain (NICD3), indicating active Notch3 signalling which is a major driver of the luminal differentiation program (Dou et al, 2017) (Fig. 4D). Conversely, markers of basal cell identity and EMT (P63, Ck5, Ck14 and Vimentin) were lost in Ezh2 null samples (Fig. 4E,F). Taken together, these observations argue that Ezh2 drives a basal/EMT phenotype in HER2+ breast cancer that is associated with metastatic progression (Dang et al, 2015; Laakso et al, 2005).

Figure 4. Validation of a luminal cellular identity and loss of basal cellular markers in Ezh2 deficient tumors.

(A) qRT-PCR analysis of luminal gene (Gata3, Foxa1, Notch3) expression normalized to β-Actin in endpoint tumors (n = 6 biological samples per genotype, n = 2 technical replicates). Data plotted as mean ± SD. (B) qRT-PCR analysis of basal gene (Krt5, Cd44) expression normalized to β-Actin in endpoint tumors (n = 6 biological samples per genotype, n = 2 technical replicates). Data plotted as mean ± SD. (C) Top: Immunohistofluorescence staining for Gata3, Foxa1 and E-cadherin on Ezh2^wt/wt and Ezh2^fl/fl endpoint tumors (n = 6 mice per genotype). Scale bar: 100 µm. Bottom: Quantification performed using HALO, plotted as mean ± SD. (D) Top: Immunoblot of Notch3 and Gapdh expression in endpoint tumors from Ezh2^wt/wt (n = 6) and Ezh2^fl/fl (n = 6) mice. Bottom: Quantification of Notch3 and Nicd expression (normalized to Gapdh), plotted as mean ± SD. Each lane represents a biological sample. All bands shown are from the same blot. (E) Top: Immunohistofluorescence staining for P63, Ck5 and HER2 on Ezh2^wt/wt and Ezh2^fl/fl endpoint tumors (n = 8 mice per genotype). Scale bar: 100 µm. Bottom: Quantification performed using HALO, plotted as mean ± SD. (F) Top: Immunohistofluorescence staining for Ck8, Ck14, and Vimentin on Ezh2^wt/wt and Ezh2^fl/fl endpoint tumors (n = 6 mice per genotype). Scale bar: 100 µm. Bottom: Quantification performed using HALO, plotted as mean ± SD. All statistical analysis: *P < 0.05, **P < 0.01, ****P < 0.0001 unpaired, two-tailed Student’s t test. Individual P values are provided in Dataset EV1. Source data are available online for this figure.

To further examine the cellular composition and lineage identity of Ezh2 null tumors, we performed single cell RNA sequencing (scRNA-Seq) on Ezh2 proficient and deficient tumors. We annotated eight clusters of cells based on the differential expression of key genes (Cao et al, 2020) (Fig. 5A; Appendix Fig. S5A). These clusters identified that Ezh2-deficient HER2+ tumors contained a significantly reduced basal population (Fig. 5B). Specific clustering of epithelial cells (HER2+Cre+) identified 8 distinct groups, based on their gene expression profiles (Fig. 5C; Appendix Fig. S5B). This analysis revealed significant shifts in HER2-expressing epithelial populations including a consistently higher proportion of progenitor-like cells (PLCs) in Ezh2 deficient samples, which are characterized by markers suggesting activation of differentiation pathways including Notch signalling (Rbpj) and Wnt signalling (Lgr6, Tcf7l1, Wnt5a) (Fig. 5D). Furthermore, epithelial-specific expression of genes such as Prom1 and Foxp1 suggest a luminal progenitor cellular identity within this cluster. Conversely, the Ezh2 deficient samples possessed lower proportions of EMT-like cells (Fig. 5D), characterized by high Zeb1/2, Vim and Cd44 expression. The single-cell analysis confirmed that the luminal markers Notch3 and Esr1 are upregulated in Ezh2 null epithelial cells (Fig. 5E). Taken together, these data confirm that loss of Ezh2 results in the generation of a luminal progenitor-like state at the expense of the basal cell populations.

Figure 5. Single cell sequencing reveals variation in cellular identity between Ezh2-proficient and Ezh2-deficient tumors.

(A) UMAP plot of scRNA-seq data displaying the integrated clusters of mammary tumors from Ezh2^wt/wt (n = 2) and Ezh2^fl/fl (n = 2) samples. Clusters: EC epithelial cells (Patj+ , Pard3+, Sox5,9,10+), BC basal cells (Krt14+, Cd9,63,81+), macrophages: (Cd68+, Csf1r + ), proliferating cells: (Top2A+ , Pola1+, Rad51b+ ), TILS tumor infiltrating lymphocytes (Iigp1+, Herc6+, Stat1,2+), myofibroblasts: (Fbn1+, Col12a1+, Col12a1+), endothelial cells: (Pecam1+, Cdh5+, Cd34+), SCLCs stem cell-like cells (Angpt1+, Wnt5b+ ). (B) Quantification of cell proportions of Ezh2^wt/wt (n = 2) and Ezh2^fl/fl (n = 2) cells across each cluster. (C) UMAP plots displaying the integrated clusters of CRE + /HER2+ epithelial cells from Ezh2^wt/wt and Ezh2^fl/fl tumor samples. Clusters: Rb1-enriched: (Rb1+), PLCs progenitor-like cells (Lgr6+, Tcf7l1+, Foxp1+, Prom1+, Rbpj+ ) TA translational active (Rpl23+ and enriched with Rps) with intermediate transition state (Krt8+/Krt14+), PCs proliferating cells (Top2A+ , Pola1+, Brac+ , Mki67+), HACs hypoxia-adapted cells (Egln3+, Hey1+), SACs: Secretary alveolar cells (Slc34a2+,Csn2+, Wap+ ), IAECs: Immune-associated epithelial cells (Gbp4+, Ifit3+), EMT-like cells: (Zeb1/2+/Vim+ /Cd44+). (D) Cell proportions of Ezh2^wt/wt (n = 2) and Ezh2^fl/fl (n = 2) tumors across each cluster within the CRE+ /HER2+ populations. (E) UMAP plot of Ezh2^wt/wt and Ezh2^fl/fl tumor samples showing the relative expression of markers of luminal signature genes, Notch3 and Esr1.

Targeting EZH2 in human HER2 positive breast cancer cells restore their sensitivity to endocrine therapies

To further explore the clinical relevance of the relationship between EZH2 and ESR1 expression, we examined breast cancer patient transcriptomic data. Expression of EZH2 was inversely correlated with the levels of ESR1 across all breast cancer molecular subtypes (Figs. 6A and EV4A) and specifically within the HER2-enriched subtype (Fig. 6B), where we observed that elevated ESR1 expression correlated with better overall survival, in line with other studies (Rediti et al, 2024) (Fig. 6C). ESR1 expression did not predict overall survival in luminal B or TNBC patients (Fig. EV4B). Collectively, these observations argue that ESR1 expression, which in turn is inversely correlated with EZH2 expression, is predictive of good clinical outcome in HER2+ patients.

Figure 6. EZH2 inhibition upregulates ESR1 through loss of H3K27me3-mediated repression.

(A) Analysis of ESR1 mRNA expression correlated with EZH2 mRNA expression in human breast tumor samples from TCGA database (n = 1247), created with Xenabrowser.net. (B) Correlation of EZH2 and ESR1 expression in HER2+ patient samples (n = 224) from TCGA database. (C) Kaplan–Meier overall survival analysis according to ESR1 expression in HER2+ patient samples, created with kmplot.com. (D) Left: Representative immunoblot of ERα, EZH2, HER2 and p-HER2 following 10 d treatment with EZH2 inhibitors GSK-126 (GSK, 2 µM) or EPZ6438 (EPZ, 2 µM) in SK-BR-3 parental (p) and Lapatinib-resistant (LR) cells. Right: Quantification normalized to β-ACTIN and DMSO (n = 6 biological replicates), plotted as mean ± SD of ERα expression. ERα and β-ACTIN were always quantified from the same blot. HER2, p-HER2, EZH2 and β-ACTIN shown below are from the same blot. *P < 0.05, **P < 0.01, Kruskal–Wallis test with Dunn’s multiple comparisons test. (E) qRT-PCR analysis of relative ESR1 expression normalized to β-ACTIN expression following 10 d treatment with GSK-126 (2 µM) or EPZ6438 (2 µM) in SK-BR-3 p and LR cells (n = 3 technical replicates), plotted as mean ± SD. ***P < 0.001, ****P < 0.0001, one-way ANOVA with Dunnett’s multiple comparison test. (F) Immunoblot analysis of ERα (with β-ACTIN control) and H3K27me3 (with total H3 control) expression following 2 d, 4 d, 6 d, 8 d and 10 d treatment with EPZ6438 (2 µM) in SK-BR-3 p cells. ERα and β-ACTIN bands are from the same blot. H3K27me3 and Total H3 bands are from blots that were processed in parallel from the same samples. (G) H3K27me3 ChIP-Seq intensity in the top 2000 differential H3K27me3 genes between SK-BR-3 p cells treated for 10 d with DMSO or EPZ6438 (n = 2 technical replicates per condition). (H) Representative ChIP-Seq enrichment over ESR1 gene in SK-BR-3 p cells treated for 10 d with DMSO or EPZ. (I) EZH2 ChIP-qPCR of SK-BR-3 p and LR cells (n = 2 technical replicates per condition). All statistical analysis: Individual P values are provided in Dataset EV1. Source data are available online for this figure.

Figure EV4. EZH2 inhibition directly upregulates ESR1 across breast cancer.

(A) Correlation of EZH2 and ESR1 expression in all breast cancer patient samples (n = 1865) from TCGA database. (B) Kaplan–Meier survival analysis of overall survival in patients with low and high ESR1 expression across different breast cancer subtypes. Created with kmplot.com. (C) Quantification of ERα expression normalized to β-ACTIN in SK-BR-3 p and LR cells (n = 7 biological replicates, lines connect values calculated from the same experiment). *P < 0.05, paired, two-tailed Student’s t test. (D) Representative immunoblot analysis of ERα expression following 10 d treatment with GSK-126 (GSK, 2 µM) or EPZ6438 (EPZ, 2 µM) in HCC1954 and JIMT-1 cells. ERα and β-ACTIN bands are from the same blot for each cell line. (E) Representative quantification of immunofluorescence of H3K27me3 expression following 10 d treatment with GSK-126 (GSK, 2 µM) or EPZ6438 (EPZ, 2 µM) in SK-BR-3 parental (p) and SK-BR-3 Lapatinib-resistant (LR), HCC1954 and JIMT-1 cells (n = 4 technical replicates per cell line). Data plotted as mean ± SD. ****P < 0.0001, one-way ANOVA with Dunnett’s multiple comparisons test. (F) H3K27me3 ChIP-seq intensity in the top 2000 differential H3K27me3 genes following ChIP-Seq of Ezh2^wt/wt and Ezh2^fl/fl tumors (n = 2 technical replicates per condition). (G) ChIP-Seq H3K27me3 enrichment across the Esr1 gene in representative Ezh2^wt/wt and Ezh2^fl/fl tumors, and respective input samples. All statistical analysis: Individual P values are provided in Dataset EV1. Source data are available online for this figure.

Given that genetic ablation of Ezh2 in HER2+ tumors resulted in the induction of Esr1 expression (Fig. 3F,G), we next evaluated whether inhibiting EZH2 activity in HER2+ breast cancer cells could restore ESR1 expression and confer a luminal cellular identity. For these experiments we used multiple HER2+ breast cancer cell lines, including Lapatinib-resistant (LR) cells that had acquired resistance to the HER2 tyrosine kinase inhibitor (TKI) Lapatinib via continuous culture in increasing concentrations of the drug (Deblois et al, 2016), as well as two established cell lines known to be resistant to Trastuzumab, HCC1954 and JIMT-1 (Kataoka et al, 2010; Tanner et al, 2004). Treatment of both SK-BR-3 parental and Lapatinib-resistant (LR) cells for 10 days with either of two independent small molecule EZH2 methyltransferase inhibitors (GSK-126, EPZ6438) robustly induced ERα expression at the protein and transcript levels (Fig. 6D,E). Interestingly, SK-BR-3 LR cells already possessed elevated basal ERα protein expression which was further increased upon inhibition of EZH2 (Figs. 6D and EV4C). These observations are consistent with upregulation of ER signalling pathways as a mechanism of acquired resistance to HER2 inhibitors (Giuliano et al, 2015). Time course analyses revealed that total loss of H3K27me3 mark following treatment with inhibitors was achieved approximately 6 days post-treatment and resulted in stepwise induction of ERα expression (Fig. 6F). Consistent with these observations, treatment of HCC1954 and JIMT-1 cells by EZH2 targeted inhibitors resulted in robust induction of ERα expression accompanied by total H3K27me3 loss (Fig. EV4D,E).

To establish whether ESR1 is a direct target of H3K27me3-mediated repression, we performed ChIP-Seq analysis to reveal genome-wide distribution of the H3K27me3 mark in SK-BR-3 parental cells treated with either DMSO or EPZ6438. As expected, we observed global loss of H3K27me3 enrichment upon EPZ6438 treatment (Fig. 6G). Enrichment of H3K27me3 was observed at a promoter/enhancer in exon 2 of ESR1 in SK-BR-3 parental cells, which was lost upon EPZ6438 treatment (Fig. 6H). In our HER2 +GEMM, we similarly observed widespread loss of H3K27me3 enrichment in Ezh2 deficient tumors (Fig. EV4F). Consistent with the data from human cells, we identified enrichment of H3K27me3 in the corresponding promoter/enhancer area of Esr1 in wild-type Ezh2 tumors, which was decreased in Ezh2 null tumors (Fig. EV4G). To validate that ESR1 is directly regulated by EZH2, we performed ChIP-qPCR in SK-BR-3 parental and LR cells and confirmed enrichment at the ESR1 locus in comparison to control loci, RPL30 and GAPDH (Fig. 6I). Collectively, these data indicate that ESR1 is a direct target of transcriptional repression by EZH2.

Based on our findings that EZH2 controls ESR1 expression and prior evidence that EZH2 can mediate resistance to Tamoxifen (Wu et al, 2018), a selective estrogen receptor modulator (SERM), we interrogated a patient dataset with information on response to ER-targeted therapy. This revealed that EZH2 expression was significantly lower in patients who did not experience relapse during five years of ER-targeted therapy (responders) compared to patients whose tumors relapsed (non-responders) (Fig. EV5A). High EZH2 expression was found to be predictive of significantly lower overall survival in patients receiving endocrine therapy (Fig. EV5B). Further, EZH2 inhibition led to a partially restored luminal identity, with significant upregulation of FOXA1, GATA3 and NOTCH3 (Fig. EV5C), indicating ERα is active in EZH2 inhibitor-treated cells. Conversely, basal markers SNAI1, TP63 and KRT5 were significantly downregulated, although CD44 and CDH2 were upregulated, indicating mesenchymal identity is not fully lost upon EZH2 inhibition (Fig. EV5D).

Figure EV5. EZH2 inhibition partially restores luminal identity and predicts patient Tamoxifen response.

(A) EZH2 expression in breast cancer patients receiving endocrine therapy based on relapse-free survival (responder = alive at 5 years), based on data from rocplot.org (responder, n = 771, non-responder, n = 136). Box and whisker plotted as 5–95th percentile, centre: median, bounds: 25th and 75th percentiles and individual data points are shown. **P < 0.01, unpaired, two-tailed Students t test. (B) Kaplan–Meier overall survival analysis according to EZH2 expression in breast cancer patients only receiving endocrine therapy, created in kmplot.com. (C) qRT-PCR analysis of luminal gene (GATA3, FOXA1, NOTCH3, EPCAM, NCOA3) expression normalized to β-ACTIN following 10 d treatment with GSK-126 (2 µM) or EPZ6438 (2 µM) in SK-BR-3 p and LR cells (n = 3 technical replicates), plotted as mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way ANOVA with Dunnett’s multiple comparison test. (D) qRT-PCR analysis of basal/mesenchymal gene (SNAI1, KRT5, TP63, CD44, CDH2) expression normalized to β-ACTIN following 10 d treatment with GSK-126 (2 µM) or EPZ6438 (2 µM) in SK-BR-3 p and LR cells (n = 3 technical replicates), plotted as mean ± SD. ns not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way ANOVA with Dunnett’s multiple comparison test. All statistical analysis: Individual P values are provided in Dataset EV1. Source data are available online for this figure.

An important implication of ERα activation is potential sensitization of ER-/HER2+ breast cancers to endocrine therapies by targeting EZH2. To directly test this possibility, we evaluated whether combining EZH2 inhibitors and Tamoxifen would impact on the proliferative capacity of SK-BR-3 cells. The combination treatment profoundly inhibited cell proliferation compared to either inhibitor alone (Fig. 7A; Appendix Fig. S6A). Interestingly, the combination of Tamoxifen and EZH2 inhibitor also reduced the proliferation of SK-BR-3 LR cells, arguing that this could be a strategy to target HER2+ breast cancers with acquired resistance to HER2 targeted therapies. To further test this as a viable therapeutic strategy, we tested Tamoxifen and EZH2 inhibition in HCC1954 and JIMT-1 cells, resistant to HER2-targeted therapies. At the 2 μM dose of Tamoxifen, HCC1954 cells were more resistant to the combination treatment, whereas JIMT-1 cells were seen to be highly susceptible to even Tamoxifen alone and exhibited a complete loss of growth in combination with EZH2 inhibitors (Appendix Fig. S6B). At a higher dose of 5 μM, HCC1954 were seen to have the most significant proliferative defect with a combination of Tamoxifen with EZH2 inhibition (Fig. 7B; Appendix Fig. S6C). Similarly, at a lower Tamoxifen dose of 1 μM, JIMT-1 cells had no effect with Tamoxifen alone, but still a striking complete loss of cell growth in combination with EZH2 inhibitors (Fig. 7B; Appendix Fig. S6C). Collectively, these data indicate that EZH2 plays a critical role in driving a basal cellular differentiation program in emerging HER2+ breast cancers, correlating with the acquisition of an EMT phenotype and metastatic potential (Fig. 7C). Moreover, our observations indicate that dual targeting of EZH2 and ERα may be an effective strategy in the therapeutic management of HER2+ breast cancers, particularly following acquired resistance to HER2-targeted therapies.

Figure 7. EZH2 inhibition sensitizes human HER2+ breast cancer cells to Tamoxifen.

(A) Proliferation assay quantification of relative confluency of SK-BR-3 p and LR cells (n = 4 technical replicates) after 3 d pretreatment with GSK-126 (GSK, 2 µM) or EPZ6438 (EPZ, 2 µM), followed by combination treatment with Tamoxifen (Tam, 2 µM) for 7 d. Plot indicates mean ± SD. (B) Proliferation assay quantification of relative confluency of HCC1954 and JIMT-1 cells (n = 6 technical replicates) after 3 d pretreatment with GSK-126 (2 µM) or EPZ6438 (2 µM), followed by combination treatment with Tamoxifen (HCC1954: 5 µM, JIMT-1: 1 µM) for 7 d. Plot indicates mean ± SD. (C) Left panel: EZH2 activity leads to the repression of luminal genes and activation of a basal cellular program. Right panel: Following EZH2 loss, ESR1 is upregulated, leading to reactivation of a luminal program and allowing sensitivity to ER-targeted therapy. All statistical analysis: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way ANOVA with Tukey’s multiple comparison test. Individual P values are provided in Dataset EV1. Source data are available online for this figure.

Discussion

Previous studies have established the importance of EZH2 in driving tumor initiation, maintaining aggressive breast cancer stem cells, and modulating breast cancer metastasis (Chang et al, 2011; Gonzalez et al, 2014; Hirukawa et al, 2018; Liu et al, 2023; Smith et al, 2019), as well as predicting resistance to targeted therapies (Bao et al, 2020; Hirukawa et al, 2019; Wu et al, 2018). Despite evidence implicating EZH2 as an important target in breast cancer progression, EZH2 inhibitors are not yet approved to treat breast cancer, highlighting the need to identify strategies for patient stratification. In this study, we investigated the role of EZH2 in the progression of HER2-driven breast cancer, a highly aggressive subtype with increased metastatic potential that is also prone to developing resistance to therapeutic interventions (Exman and Tolaney, 2021). We demonstrated that Ezh2 promotes mammary tumor initiation and lung metastasis in a GEMM (EIC) which accurately recapitulates human HER2+ breast cancer progression. The role of Ezh2 in driving proliferation and the progression of early hyperplastic lesions in the inducible EIC model aligns with our previous findings using an earlier MMTV-NIC driven model, where only 20% of Ezh2 null mice developed tumors (Smith et al, 2019). Collectively, these findings demonstrate the necessity of Ezh2 for ErbB2-driven mammary epithelial transformation (Smith et al, 2019). The observed delay in tumor onset was not due to effects on HER2 expression, as robust epithelial-specific staining of HER2 could be detected at 12 weeks post-induction in both control and Ezh2 deficient mice. Rather, we obtained evidence for a profound block in cell cycle progression, with reduced entry of cells into S phase and a loss of markers of cell cycle progression and proliferation (Fig. EV1). These data are consistent with previous studies using genetic targeting of Ezh2 in a GEMM driven by Polyoma Virus middle T antigen (PyVmT), where a similar defect in epithelial proliferation was observed (Hirukawa et al, 2018; Liu et al, 2023).

The incomplete loss of H3K27me3 and small proportion of genes displaying higher H3K27me3 in mammary epithelial cells upon Ezh2 deletion may be due to remarkable compensation by Ezh1 (Shen et al, 2008) (Figs. 2 and EV2). While H3K27me3 was largely Ezh2-dependent in our previous studies using an MMTV-driven ErbB2 model and a doxycycline inducible PyVmT model (Hirukawa et al, 2018; Hirukawa et al, 2019; Liu et al, 2023), which resembles the Luminal B subtype, the relative contributions of Ezh2 and Ezh1 may be context and subtype-dependent and should be further explored in HER2-driven breast cancer. For example, a study of double deletion of Brca1 and Ezh2 in mammary stem cells showed no significant impact on mammary neoplasia due to unaltered H3K27me3 levels (Bae et al, 2015). Also, while inhibition of Ezh2 methyltransferase activity in HER2+ breast cancer cells recapitulated some aspects of the Ezh2 knockout phenotype, such as ESR1 upregulation, we cannot rule out the involvement of non-canonical functions of EZH2 that do not involve its methyltransferase activity (Zimmerman et al, 2023). These functions may depend on post-translational modifications of EZH2 and are also likely to be highly context-dependent. For example, EZH2 phosphorylation at S21 is associated with HER2+ breast cancer invasiveness and phosphorylation at T367 and cytoplasmic localization of EZH2 can promote breast cancer metastasis (Anwar et al, 2018; Yu et al, 2023). Further, the CDK2-mediated phosphorylation of EZH2 at T416 was shown to be crucial for EZH2-mediated ESR1 upregulation in TNBC (Nie et al, 2019).

Among the most striking observations of the present study was a major reduction in the metastatic potential of HER2-driven tumours by genetic ablation of Ezh2. Both the penetrance of lung metastasis and the size of metastatic lesions were decreased in Ezh2 deficient mice or mice bearing Ezh2 deficient cells compared to the wild-type controls. These data indicated they were significantly impaired in their ability to migrate and to colonize the lungs, which was not attributed to a proliferative defect (Figs. 1 and 2; Appendix Fig. S4). These findings highlight a central role for Ezh2 in driving the metastatic progression of HER2+ breast cancer in this model. Multi-modal analysis including bulk and single-cell transcriptomics and quantitative histopathology revealed that this block in metastasis correlated with a profound impact on lineage identity and plasticity, with reduced basal-like and EMT cell populations, a significant loss of key EMT regulators including Snail and P63 (Cano et al, 2000; Dang et al, 2015; Laakso et al, 2005; Smith et al, 2014), and reduced expression of markers of the basal lineage (e.g. Ck5/Ck14) in Ezh2 deficient tumours (Figs. 4 and 5). This was accompanied by upregulation of E-cadherin, an important marker of epithelial cell identity which can be repressed by Snail and EZH2 in cancer (Cano et al, 2000; Cao et al, 2008), and the induction of a luminal cell fate that included upregulation of estrogen receptor alpha (Esr1), a key driver of the luminal differentiation program (Guttilla et al, 2012). In addition to Esr1, Ezh2 null HER2+ tumors expressed and activated Notch3, which is crucial for guiding bipotent progenitors toward the luminal lineage, a function that cannot be replaced by other Notch receptors (Raouf et al, 2008). Interestingly, Notch3 itself can drive activation of ERα, and this was shown to directly suppress metastatic function independently of EZH2 (Dou et al, 2017). Further studies have shown that reactivation of GATA3, which is upregulated in our Ezh2 deficient GEMM and upon EZH2 inhibition, is essential for conversion of basal-like TNBC cells to a more luminal-like state (Schade et al, 2024; Yomtoubian et al, 2020). Taken together, these observations argue that regulation of Ezh2 activity plays a vital role in regulating epithelial lineage plasticity and EMT during HER2+ breast cancer progression, leading to the acquisition of a mesenchymal metastatic phenotype.

The exact role of EZH2 in EMT is widely debated and largely context-dependent, with some studies pointing to EZH2 inhibition suppressing EMT (Cao et al, 2008), whilst others argue that it promotes EMT through loss of H3K27me3-mediated repression of mesenchymal genes (Gallardo et al, 2024; Hirukawa et al, 2018; Zhang et al, 2022). In our study, although upregulation of luminal genes could be recapitulated using EZH2 inhibitors, some mesenchymal genes were also upregulated following EZH2 inhibition, which is attributed to them being directed H3K27me3 targets and consistent with other studies (Fig. EV5) (Gallardo et al, 2024). These findings indicate the possibility of non-canonical roles of EZH2 in suppressing EMT, as well as the importance of Ezh1 in retention of H3K27me3, likely preventing reactivation of mesenchymal gene expression. Both of these points warrant further exploration in HER2-specific contexts. Further, it highlights the potential utility of targeting Ezh2 via degradation rather than canonical inhibition to effectively reverse EMT and overcome metastasis in aggressive HER2-driven breast cancers (Velez et al, 2024; Zimmerman et al, 2023).

Another interesting aspect of our study is the observation that loss of EZH2 catalytic activity in HER2+ breast cancer cell lines was associated with induction of ESR1 expression (Fig. 6). Indeed, we showed that the ESR1 gene is direct target of EZH2-mediated epigenetic repression. We further showed that a combination of EZH2 methyltransferase inhibitors and SERMs (e.g. Tamoxifen) can act synergistically to inhibit HER2+ breast cancer growth (Fig. 7). In addition, we demonstrate that HER2+ cells that have resistance to HER2 targeted therapies can be sensitized to ER-targeted therapies by combining EZH2 inhibitors with Tamoxifen. Consistent with these results, recent studies showed that EZH2 inhibition alone is sufficient to induce a luminal phenotype, block metastasis and sensitize tumors to ER-targeted therapy in TNBC (Nie et al, 2019; Yomtoubian et al, 2020). These studies emphasize the importance of combination therapy strategies that restrict the cellular plasticity exhibited during the evolution and progression of breast cancer cells. Future clinical validation of this strategy to prevent breast cancer progression will have important implications for the treatment of HER2 positive breast cancers.

Methods

Reagents and tools table

Reagent/resource	Reference or source	Identifier or catalog number
Experimental models
TetO-ERBB2-IRES-CRE (M. Musculus)	Attalla et al, 2023
MMTV-rtTA (M. Musculus)	Gunther et al, 2002
Ezh2fl/fl (M. Musculus)	Shen et al, 2008
FVB (M. Musculus)	Charles River Laboratory	# 207
MMTV-rtTA/EIC/Ezh2wt/wt cell line (M. Musculus)	Attalla et al, 2023, this study
MMTV-rtTA/EIC/Ezh2fl/fl cell line (M. Musculus)	This study
NOD-SCID-Gamma (M. Musculus)	Jackson Laboratory	# 005557
SK-BR-3 p	ATCC	Cat # HTB-30
SK-BR-3 LR	Deblois et al, 2016
HCC1954	ATCC	Cat # CRL-2338
JIMT-1	Tanner et al, 2004
Antibodies
Rabbit anti-EZH2	Cell Signaling	5246
Rabbit anti-H3K27me3	Cell Signaling	9733
Mouse anti-HER2	DAKO	A0485
Rabbit anti-Ki67	Cell Signaling	12202
Mouse anti-PCNA	Cell Signaling	2586
Rabbit anti-Cyclin D1	Cell Signaling	2978
Rabbit anti-LaminB1	Cell Signaling	17416
Rabbit anti-p-Rb (S807/S811)	Cell Signaling	8516
Rabbit anti-CCP3	Cell Signaling	9661
Rabbit anti-p-S6 (S240/244)	Cell Signaling	5364
Rabbit anti-p-4EBP1 (T37/T46)	Cell Signaling	2855
Rabbit anti-Ezh1	Invitrogen	PA5-115205
Rabbit anti-p63	Abcam	ab124762
Rabbit anti-CK5	BioLegend	905504
Mouse anti-Gata3	Santa Cruz	sc-268
Rabbit anti-Foxa1	Abcam	ab173287
Mouse anti-E-cadherin	BD Biosciences	610182
Mouse anti-CK8	BioLegend	904804
Rabbit anti-CK14	Roche	760-4803
Rabbit anti-Vimentin	Abcam	ab92547
Mouse anti-β-Actin	BioLegend	A5441
Mouse anti-α-Tubulin	Cell Signaling	3873
Mouse anti-HER2	Cell Signaling	2248
Rabbit anti-p-HER2 (Y1248)	Cell Signaling	2247
Rabbit anti-H3	Cell Signaling	4499
Rabbit anti-HER3	Cell Signaling	12708
Rabbit anti-p-HER3 (Y1289)	Cell Signaling	4791
Rabbit anti-Erk1/2	Cell Signaling	9102
Rabbit anti-p-Erk1/2 (T202/Y204)	Cell Signaling	4370
Rabbit anti-Ampk	Cell Signaling	5831
Rabbit anti-p-Ampk (T172)	Cell Signaling	2535
Mouse anti-pan Akt	Cell Signaling	2920
Rabbit anti-p-Akt (S473)	Cell Signaling	9018
Rabbit anti-p-Akt (T308)	Cell Signaling	13038
Rabbit anti-GAPDH	Cell Signaling	2118
Rabbit anti-Notch3	Abcam	ab23426
Rabbit anti-Snail	Cell Signaling	3879
Rabbit anti-ERα	Abcam	ab108398
Oligonucleotides and other sequence-based reagents
MTB_Forward_ACCGTACTCGTCAATTCCAAGGG	Gunther et al, 2002
MTB_Reverse_TGCCGCCATTATTACGACAAGC	Gunther et al, 2002
HER2_Forward1_GTGACCTGTTTTGGACCGGA	Attalla et al, 2023
HER2_Forward2_TGTGTGGACCTGGATGACAAGG	Attalla et al, 2023
HER2_Reverse_TCTCCGCATCGTGTACTTCC	Attalla et al, 2023
Ezh2_Forward1_CATGTTTAAGGGCATAGTGACATG	Shen et al, 2008
Ezh2_Forward2_TCGAGGGACCTAATAACTCGTATAGCA	Shen et al, 2008
Ezh2_Reverse_ATGTGCAGGTCAGTCAGCAACTTCAG	Shen et al, 2008
mEsr1_Forward_ CCTCCCGCCTTCTACAGGT	PrimerBank, this study
mEsr1_Reverse_ CACACGGCACAGTAGCGAG	PrimerBank, this study
mβ-Actin_Forward_ TCCATCATGAAGTGTGACGT	Taifour et al, 2023
mβ-Actin_Reverse_ GAGCAATGATCTTGATCTTCAT	Taifour et al, 2023
mGapdh_Forward_CTGCACCACCAACTGCTTAG	Bui et al, 2022
mGapdh_Reverse_GTCTTCTGGGTGGCAGTGAT	Bui et al, 2022
mKrt5_Forward_TCCAGTGTGTCCTTCCGAAGT	PrimerBank, this study
mKrt5_Reverse_TGCCTCCGCCAGAACTGTA	PrimerBank, this study
mCd44_Forward_TCGATTTGAATGTAACCTGCCG	PrimerBank, this study
mCd44_Reverse_CAGTCCGGGAGATACTGTAGC	PrimerBank, this study
mNotch3_Forward_CTCTGTGGTGATGCTGGAGA	Su, Wei et al, 2018
mNotch3_Reverse_AATCAAGTCGCTCCACTGCT	Su, Wei et al, 2018
mGata3_Forward_CCTCTGGAGGAGGAACGCTAAT	Origene	NM_008091
mGata3_Reverse_GTTTCGGGTCTGGATGCCTTCT	Origene	NM_008091
mTgfα_Forward_GGAACCTGCCGGTTTTTGG	This study
mTgfα_Reverse_CACAGCGAACACCCACGTA	This study
mBcl2_Forward_ATGCCTTTGTGGAACTATATGGC	Meister et al, 2007
mBcl2_Reverse_GGTATGCACCCAGAGTGATGC	Meister et al, 2007
mFoxa1_Forward_ATGAGAGCAACGACTGGAACA	PrimerBank, this study
mFoxa1_Reverse_TCATGGAGTTCATAGAGCCCA	PrimerBank, this study
mMyc_Forward_ATGCCCCTCAACGTGAACTTC	PrimerBank, this study
mMyc_Reverse_GTCGCAGATGAAATAGGGCTG	PrimerBank, this study
mRprm_Forward_GTGGTGCAGATCGCAGTCAT	PrimerBank, this study
mRprm_Reverse_CGGTCCTTCACTAGGAAGTTGA	PrimerBank, this study
hESR1_Forward_GGGAAGTATGGCTATGGAATCTG	PrimerBank, this study
hESR1_Reverse_TGGCTGGACACATATAGTCGTT	PrimerBank, this study
hβ-ACTIN_Forward_AGAGCTACGAGCTGCCTGAC	Widely used
hβ-ACTIN_Reverse_AGCACTGTGTTGGCGTACAG	Widely used
hGATA3_Forward_GCCCCTCATTAAGCCCAAG	PrimerBank, this study
hGATA3_Reverse_TTGTGGTGGTCTGACAGTTCG	PrimerBank, this study
hFOXA1_Forward_GCAATACTCGCCTTACGGCT	PrimerBank, this study
hFOXA1_Reverse_TACACACCTTGGTAGTACGCC	Harvard Primer Bank, this study
hNOTCH3_Forward_GAGCCAATGCCAACTGAAGAG	Merfeld-Clauss et al, 2023
hNOTCH3_Reverse_GGCAGATCAGGTCGGAGATG	Merfeld-Clauss et al, 2023
hEPCAM_Forward_TGATCCTGACTGCGATGAGAG	PrimerBank, this study
hEPCAM_Reverse_CTTGTCTGTTCTTCTGACCCC	PrimerBank, this study
hNCOA3_Forward_ACATAAACGCCAGTCCTGAAAT	PrimerBank, this study
hNCOA3_Reverse_CCTTCCTCCATCATAGCTCGT	PrimerBank, this study
hSNAI1_Forward_TCGGAAGCCTAACTACAGCGA	PrimerBank, this study
hSNAI1_Reverse_AGATGAGCATTGGCAGCGAG	PrimerBank, this study
hKRT5_Forward_TGACCTCCGCAACACCAAG	PrimerBank, this study
hKRT5_Reverse_CAGATTGGCGCACTGTTTCTT	PrimerBank, this study
hTP63_Forward_GTCATTTGATTCGAGTAGAGGGG	PrimerBank, this study
hTP63_Reverse_CTGGGGTGGCTCATAAGGT	PrimerBank, this study
hCD44_Forward_CTGCCGCTTTGCAGGTGTA	PrimerBank, this study
hCD44_Reverse_CATTGTGGGCAAGGTGCTATT	PrimerBank, this study
hCDH2_Forward_TGCGGTACAGTGTAACTGGG	PrimerBank, this study
hCDH2_Reverse_GAAACCGGGCTATCTGCTCG	PrimerBank, this study
RPL30_ChIP_Forward_TGAAGCCGAGCCCCTTAAAC	This study
RPL30_ChIP_Reverse_CCAGTGGCCCTCTCAATGAC	This study
GAPDH_ChIP_Forward_CTATAAATTGAGCCCGCAGCC	This study
GAPDH_ChIP_Reverse_GCGACGCAAAAGAAGATGCG	This study
ESR1_ChIP_Forward_CAAATAAACACGGGGCGCTT	This study
ESR1_ChIP_Reverse_AAATGCCAAGAGCGAGACCT	This study
Chemicals, enzymes and other reagents
Lapatinib	MedChemExpress	Cat # HY-50898
GSK-126	MedChemExpress	Cat # HY-13470
A395	MedChemExpress	Cat # HY-101512
EPZ6438	MedChemExpress	Cat # HY-13803
Tamoxifen	MedChemExpress	Cat # HY-13757A
EdU	MedChemExpress	Cat # HY-118411
Software
HALO, v3.5.3577	Indica Labs
GraphPad Prism, v9.0	GraphPad Software, Inc.
ImageStudio, v6.1	LICORbio
Fiji	Schindelin et al, 2012
IncuCyte S3 System, v2019A	ESSEN Bioscience
IGV, v2.19.5	Robinson and Oshlack, 2010
FACSDiva Software, v8	BD Biosciences
FlowJo, v10	BD Biosciences
R software, v4.4.2	R Core
Other
Aperio-XT slide scanner	Leica Biosystems
BD LSR Fortessa flow cytometer	BD Biosciences
AxioScan Z1 digital slide scanner	ZEISS
Odyssey CL-X imaging system	Li-COR Biosciences
LightCycler 480	Roche
RNAscope 2.5 HD Assay-RED Kit	ACD	Cat # 322360
Histone Extraction Kit - Rapid/Ultra-pure	Abcam	Cat # 221031
FavorPrep Tissue Total RNA Mini Kit	Favorgen	Cat # FATRK001
TransScript All-in-One cDNA Synthesis SuperMix Kit	Transgen Biotech	Cat # AT341-01
MinElute PCR purification kit	Qiagen	Cat # 28004
NEBNext Ultra II DNA Library Prep kit	New England Biolabs	Cat # E7103L

Animal models and tissue collection

Ethical approval was obtained for the use of all animals in this study, and all animal work and experiments were approved by the Animal Care Committee of McGill University and completed in accordance with the ethical standards mandated by the Canadian Council of Animal Care (CCAC). All strains were housed in the animal facility at the Goodman Cancer Research Institute and maintained on an FVB/N background. The TetO-ERBB2-IRES-CRE (EIC) transgenic mice have been described previously (Attalla et al, 2023), and were mated with the MMTV-rtTA (MTB) strain (Gunther et al, 2002), and Ezh2 (Shen et al, 2008) conditional knockout strain. Mice were genotyped at 4 weeks of age for MTB, HER2, and CRE transgenes, as well as Ezh2 status. Genotyping primers were obtained from Integrated DNA Technologies, used at a concentration of 10 µM and are listed in the Reagents and Tools Table. Experimental female mice were induced with water supplemented with 2 mg/mL doxycycline (Wisent) at 8–12 weeks of age and were palpated and measured weekly with a caliper. Tumor onset was determined when the first palpable lesion occurred. Mice continued on doxycycline water until they reached a time-determined experimental point (12 W post-doxycycline induction or humane tumor endpoint). Humane tumor endpoint was defined as either: one tumor with a volume of 2.5 cm³ or total tumor burden of 6 cm³. Mice were excluded if they reached humane endpoint not related to tumor burden. Upon sacrifice, mice were weighed both before and after tumors were removed to determine tumor burden. Tissues of interest (breast tumor/mammary glands and lungs) were formalin-fixed (Sigma-Aldrich) for 24–36 h and paraffin-embedded (Leica), or flash-frozen in liquid nitrogen and stored long term at −80 °C. Fixed and embedded materials were sectioned at 4 μm thickness and mounted onto glass slides, and samples were stained with hematoxylin and eosin (H&E). To quantify lung metastases, lung H&E slides were scanned at ×20 magnification using an Aperio-XT slide scanner (Leica Biosystems), and quantification was performed by HALO software (Indica Labs, v3.5.3577).

Cell culture

Primary mouse tumor cell lines were established from MMTV-rtTA/EIC/Ezh2^wt/wt or Ezh2^fl/fl endpoint tumors as previously described (Attalla et al, 2023). Briefly, tissue was dissected with a McIIwain Tissue Chopper until homogenous, followed by incubation with DMEM (Wisent) containing 2.4 mg/mL collagenase B (Roche) and dispase II (Roche) for 1.5 h at 37 °C with constant agitation. Cell suspensions were centrifuged for 10 min at 1000 rpm and washed with PBS. Red blood cells were lysed with ammonium-chloride-potassium (ACK) lysis buffer (0.15 mol/L ammonium chloride, 0.1 mol/L potassium bicarbonate, and 0.0001 mol/L EDTA; Bioshop) for 5 min, and obtained primary mouse cells were resuspended and maintained in DMEM (Wisent) containing 5% Fetal Bovine Serum (FBS), 1 μg/mL hydrocortisone (Millipore), 5 ng/mL epidermal growth factor (Wisent), and 35 μg/mL bovine pituitary extract (Hammond CellTech) with 1% penicillin/streptomycin (Wisent), 1% Amphotericin B (Wisent) 0.1% Gentamycin (Wisent) and supplemented with Doxycycline (Wisent, 10 μg/mL).

SK-BR-3 and HCC1954 cells were purchased from ATCC. SK-BR-3 Lapatinib-resistant (LR) cells were developed by exposing SK-BR-3 cells to increasing concentrations of Lapatinib over multiple passages, until no further cell death was observed (Deblois et al, 2016). Cells were grown and maintained in either DMEM (Wisent) for SK-BR-3 and JIMT-1 cells or RPMI 1640 (Wisent) for HCC1954 cells, supplemented with 10% FBS, 1% penicillin/streptomycin (Wisent), 1% Amphotericin B (Wisent) and 0.1% Gentamycin (Wisent) and incubated at 37 °C in 5% CO₂. All cell cultures were routinely passaged every 3–4 days once they reached 80–90% confluency. Cell lines were routinely tested for mycoplasma contamination.

Orthotopic transplantation of cells

For orthotopic transplantation, 1 × 10⁵ cells from established MMTV-rtTA/EIC/Ezh2^wt/wt and MMTV-rtTA/EIC/Ezh2^fl/fl cell lines were resuspended in phosphate-buffered saline (PBS) and injected bilaterally into the inguinal mammary fat pads of 8–12-week-old NOD-SCID-Gamma (NSG) mice. Mice were maintained on water supplemented with doxycycline (2 mg/ml, Wisent). Tumors were monitored twice weekly in a blinded experiment and measurements were performed with a caliper.

Tail vein injections

For assessment of lung colonization, 1 × 105 cells from established MMTV-rtTA/EIC/Ezh2^wt/wt and MMTV-rtTA/EIC/Ezh2^fl/fl cell lines were resuspended in PBS and injected into the tail vein of NSG mice. Mice were maintained on water supplemented with doxycycline (2 mg/ml, Wisent), monitored weekly and sacrificed after 8 weeks, where lungs were taken and analyzed as described.

Boyden chamber migration assay

MMTV-rtTA/EIC/Ezh2^wt/wt and MMTV-rtTA/EIC/Ezh2^fl/fl cells were harvested using trypsin-EDTA and counted, then 1 × 10⁵ cells/mL were resuspended in serum-free DMEM. Transwell inserts were placed into a 24-well plate, where 500 μL of DMEM (supplemented with 10% FBS) was added to the lower chamber and 500 μL of the cell suspension was loaded to the upper chamber. The plate was incubated at 37 °C with 5% CO₂ for 24 h. After incubation, medium was removed and cells were fixed by adding formalin for 20 min, then stained with crystal violet for 20 min. The membrane was washed twice with PBS and dried. The trans-well inserts were wiped using a cotton swab to remove the cells from the upper side of the membrane and left to dry overnight at room temperature (RT). Images were taken using an EVOS microscope and analyzed using Fiji.

Inhibitors

Lapatinib (MedChemExpress, HY-50898), GSK-126 (MedChemExpress, HY-13470), EPZ6438 (MedChemExpress, HY-13803), A395 (MedChemExpress, HY-101512) and Tamoxifen (MedChemExpress, HY-13757A) were dissolved in DMSO and used at the given concentration. DMSO was used as a control.

Incucyte cell proliferation assay

For proliferation assays, cells were pretreated for 72 h with either GSK-126 (2 μM), EPZ6438 (2 μM), A395 (1 μM) or DMSO control, and drugs were replenished every 48 h across the experiment course. In total, 5000 cells (SK-BR-3 cells and MMTV-rtTA/EIC cells) or 2000 cells (HCC1954 and JIMT-1 cells) were seeded per well in quadruplicate or sextuplicate in 96-well optical-bottom plates (Nunc). After 24 h of seeding, Tamoxifen or DMSO control were added, and live cell imaging was performed using the IncuCyte S3 system (ESSEN BioSciences, Ann Arbor, MI, USA) at ×10 magnification every 6 h for the given period. Percentage confluence was determined using the IncuCyte S3 Analysis software (v2019A, ESSEN BioSciences).

EdU and immunofluorescence

For immunofluorescence, human cells were treated with GSK-126, EPZ6438 or DMSO control for 9 days, and MMTV-rtTA/EIC/Ezh2^wt/wt cells were treated with A395, EPZ6438 or DMSO for 5 days before they were seeded on cover slips and incubated for 24 h at 37 °C with 5% CO₂. For 5-ethynyl-2′-deoxyuridine (EdU, 10 μM) incorporation, cells were treated for 2 h prior to fixing. Samples were fixed for 15 min at RT with 4% paraformaldehyde, then permeabilized with PBS (+0.2% Triton-X100) for 10 min at RT. Blocking was performed with PBS + (+0.2%Triton-X100, 0.05% Tween-20, 2% BSA) for 30 min. Detection of EdU-DNA was performed with a homemade kit according to the Invitrogen Click-iT EdU Alexa Fluor 488 Flow Cytometry Kit with copper II sulfate CuSO₄ (ALDON CORP SE), CF®405 M Dye azide (Biotium) and Sodium Ascorbate (Bioshop). Samples were incubated with primary antibody (H3K27me3 1:1000, cs9733) for 1 h at RT, followed by 3 washes with PBS (+0.2%Triton-X100, 0.05% Tween-20). Samples were then incubated with secondary antibody (Alexa Fluor 555, Donkey anti-Rabbit) for 45 min at RT, followed by 3 further washes. Samples were counterstained with DAPI (Sigma D9542, 1 μg/mL) for 5 min, then washed and mounted onto slides using Epredia Immu-Mount (Thermo Fisher). Samples were imaged using a confocal microscope, quantification was performed on the HALO software and figures were made using GraphPad Prism.

Flow cytometry and analysis

In all, 12-week induced MMTV-rtTA/EIC/Ezh2^wt/wt and MMTV-rtTA/EIC/Ezh2^fl/fl mice were injected with 5-ethynyl-2′-deoxyuridine (EdU, 0.1 mg/1 g body weight) (MedChemExpress). After 3 h, mammary glands were collected, finely chopped, and dissociated with 2 mg/mL each of collagenase B/ Dispase II (Roche) in 10 mL DMEM for 1.5 h at 37 °C. Cells were incubated with ACK Lysis Buffer for 5 min, washed with FACs buffer (PBS, 2 mM EDTA, 2% FBS) and then passed through a 70 μm strainer. 1 million cells were stained with TruStain FcX block (Biolegend) followed by incubation with HER2 APC-750 (Biolegend) for 30 min on ice. Cells were washed twice with FACs buffer, fixed for 15 min at RT with BD Cytofix/Cytoperm Fixation/permeabilization buffer kit and washed twice with BD Perm/Wash™ buffer. Detection of EdU-DNA was performed with a homemade kit according to the Invitrogen Click-iT EdU Alexa Fluor 488 Flow Cytometry Kit with copper II sulfate CuSO₄ (ALDON CORP SE), CF®405 M Dye azide (Biotium) and Sodium Ascorbate (Bioshop). Cell pellets were incubated in 250 μl of reaction buffer for 30 min at RT protected from light. Staining Intracellular Antigen for Ki67-Pecy7 (Biolegend) was performed for 1 h at RT, followed by staining of DNA with 7-AAD (7-amino-actinomycin D) (Biolegend). Cells were then analyzed by flow cytometry using a 4-laser BD LSR Fortessa flow cytometer (BD Biosciences, San Jose, CA). A minimum of 250000 events were acquired per experiment in slow rate mode to avoid doublets. Sample measurements were performed with FACSDiva Software (v8) (BD Biosciences). Data analysis was performed with FlowJo Software. Cell debris and aggregates were excluded from the analysis using pulse processing SSC-H vs SSC-W and 7-AAD-A vs 7-AAD-W.

Multiplex immunohistofluorescence

IHF was performed similarly to as previously described (Bui et al, 2022). Briefly, 4μm-thick mounted tissue sections were deparaffinized in three xylene washes, hydrated in decreasing concentrations of ethanol, and then washed thoroughly with deionized water. Antigen retrieval was performed using either citrate buffer (pH 6; Vector Laboratories) or EDTA buffer (pH 9; Vector Laboratories) in a pressure cooker for 10 min, then cooled in a running water bath. The slides were then incubated in 3% hydrogen peroxide for 10 min and blocked in 10% casein-based buffer (Vector Laboratories) for 5 min. Primary antibodies were diluted in 2% BSA-TBS-T and incubated for 30 min at RT in a humidity chamber, followed by washing in TBS-T three times. All primary antibodies used are listed in the Reagents and Tools Table. Incubation with secondary antibodies (Vector Laboratories; VECTMP745250 and VECTMP740150) was performed for 30 min at RT in a humidity chamber, followed by three further washes in TBS-T. The slides were then incubated with Opal fluorophores (Akoya Biosciences; 520, 570, 620, 690) at RT for 10 min and washed with distilled water. For additional rounds, the process was repeated from antigen retrieval to Opal detection. Samples were counterstained with DAPI (Sigma D9542, 1 μg/mL) for 10 min at RT and rinsed with distilled water, then mounted using Epredia Immu-Mount (Thermo Fisher). Slides were scanned using the Zeiss AxioScan Z1 digital slide scanner. Quantification was performed on HALO (Indica Labs) and figures were made using HALO and GraphPad Prism.

RNA in-situ hybridization

RNA in situ hybridization (RNAScope) was carried out as per the manufacturer’s protocol using the RNAscope 2.5 HD Assay-RED Kit (ACD, 322360) and probes against Esr1 (ACD, 478201), and control (Ppib, ACD, 313911). Briefly, 4μm tumor sections were deparaffinized in xylene, followed by 100% ethanol, and dried at 60 °C. Each section was covered with 3% hydrogen peroxide for 10 min at RT. The slides were then placed in antigen retrieval reagent (ACD), heated in a pressure cooker for 10 min, rinsed in 100% ethanol, and dried at 60 °C for 5 min. The tissues were digested using protease III (ACD) for 30 min at 40 °C. The probes were hybridized to the tissues for 2 h at 40 °C. Samples were exposed using Fast-Red alkaline phosphatase substrate and horse radish peroxidase teal substrate (ACD). Counterstaining was performed for HER2 and DAPI and slides were mounted as described for IHF. The stained tissue was scanned using the Zeiss AxioScan Z1 digital slide scanner and analyzed using HALO as performed for IHF slides.

Protein and histone extraction

For total lysate protein extraction, cells or flash-frozen tumor pieces (crushed with mortar and pestle) were lysed in ice-cold RIPA lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5% Sodium Deoxycholate, 1% NP-40, 0.1% SDS, 1 mM EDTA) containing protease inhibitors: 1 mM PMSF, 10 μg/μL aprotinin, 10 μg/μL leupeptin, and phosphatase inhibitors: 25 mM β-glycerophosphate, 1 mM sodium orthovanadate, and 10 mM sodium fluoride. Histone extraction from flash-frozen tumor pieces was performed using Abcam’s histone extraction kit (ab221031) according to the manufacturer’s instructions. Protein concentrations were determined by Bradford assay (Bio-Rad), and samples were prepared with 6× SDS dye and RIPA lysis buffer to a concentration of 4 μg/μL, followed by boiling at 95 °C for 5 min.

Immunoblotting

In all, 40–80 μg of protein samples or 10 μg of histone extracts were run with an SDS-PAGE gel at 80 V–120 V. Proteins were transferred onto a PVDF membrane for 90 min at 30 V at 4 °C. The membrane was blocked in blocking buffer (Li-COR Biosciences) for 1 h at RT, before incubation with primary antibodies for 1 h at RT or overnight at 4 °C. The membrane was washed three times with TBS-T for 5 min, followed by incubation with Li-COR secondary antibodies (IRDye 800CW Donkey anti-Rabbit, IRDye 680RD Donkey anti-Mouse) for 1 h and a further three washes with TBS-T. All antibodies used are described in the Reagents and Tools Table. Gel was imaged using the Odyssey CL-X imaging system (Li-COR Biosciences). Proteins of interest were quantified directly from this gel image using ImageStudio (LICORbio, v6.1) and normalized to loading controls.

RNA extraction

Total RNA from flash-frozen tumor pieces or cell pellets was extracted and purified using the FavorPrep Tissue Total RNA Mini Kit (Favorgen, FATRK001) according to manufacturer’s protocol. Total RNA concentration and purity was assessed using a Nanodrop 2000 (Thermo Fisher Scientific, ND2000CLAPTOP).

cDNA synthesis and quantitative reverse transcriptase PCR (qRT-PCR)

cDNA was synthesized from 1μg RNA using the TransScript All-in-One cDNA Synthesis SuperMix kit (Transgen Biotech, AT341-01), according to the manufacturer’s instructions. qRT-PCR was conducted with SYBR Green MasterMix (Roche) using the LightCycler 480 instrument (Roche). All samples were run in duplicate or triplicate, and data were normalized to β-Actin or Gapdh expression. Human and mouse primer sequences were designed using the Harvard PrimerBank (Wang et al, 2012), obtained from Integrated DNA Technologies and are described in the Reagents and Tools Table.

RNA sequencing and analysis

RNA Sequencing (RNA-Seq) was performed by Novogene on RNA extracted from flash-frozen endpoint tumors. Sequencing data were uploaded to the Galaxy web platform, and preprocessing was performed using the public server at usegalaxy.eu (Galaxy, 2024). Quality control of raw reads was performed using FastQC (Galaxy Version 0.74) (Andrews, 2010). Raw reads were aligned to the mm39 mouse genome obtained from UCSC (Perez et al, 2025) using STAR (Galaxy Version 2.7.11a) (Dobin et al, 2013). Gene expression was quantified with FeatureCounts (Galaxy Version 2.0.6) (Liao et al, 2014). Raw count data were processed in R (v.4.3.3) and ggplot2 was used for visualization. DESeq2 (v.1.42.1)(Love et al, 2014) was employed for normalization and differential expression analysis. Genes with fewer than 10 counts across at least three samples were filtered out. The Wald test was used to identify differentially expressed genes (DEGs), and log-fold changes were adjusted using apeglm (v.1.24.0)(Zhu et al, 2019) to minimize noise. Differentially expressed genes were analyzed by EnrichR (Chen et al, 2013; Kuleshov et al, 2016; Xie et al, 2021). EMT-related signatures were collected from the MSigDB mouse collection (v.2024.1)(Castanza et al, 2023; Subramanian et al, 2005). Gene set enrichment analysis (GSEA) was conducted with ClusterProfiler (v.4.10.1) with significantly changed pathways and enrichplot (v.1.22.0) was used for visualization. Downregulated genes in Ezh2^fl/fl tumors were assessed for association to cancer hallmarks using cancerhallmarks.com (Menyhart et al, 2024). Orthologous genes from the DEG list were mapped to human genes using DIOPT (v9.0) (Hu et al, 2011), applying the high-rank threshold (best score both ways and DIOPT score ≥2). BRCA-TCGA gene expression data and clinical information were retrieved by using TCGAbiolinks (v.2.30.4) (Colaprico et al, 2016). Kaplan–Meier survival analyses were performed using survival (Therneau, 2024; Therneau and Grambsch, 2000) (v.3.5.8) and survminer (v.0.5.0) packages, with high- and low-expression groups determined by median expression values of selected signatures.

Single cell sequencing (scRNA-Seq) and analysis

Flash-frozen tumor samples were sent to Novogene for scRNA-Seq analysis. Droplet libraries were processed using Cell Ranger (10X Genomics) (Zheng et al, 2017). Sequencing reads were aligned to the reference genome, and transcript counts quantified for each annotated gene within every cell. Count matrices (genes × cells) were loaded into the R package Seurat (Hao et al, 2021) for quality control and downstream analyses. Low-quality cells were filtered out using the following criteria: (1) the number of detected genes is >500; (2) percentage of mitochondrial RNA is <5% per cell. Cell doublets were detected and removed using the R package scDblFinder (Germain et al, 2021). Following SCTransform normalization, individual samples were integrated using the Harmony procedure. UMAP dimension reduction was generated based on the first 15 principal components. A nearest-neighbor graph using the first 15 principal components was calculated using FindNeighbors function, followed by clustering using FindClusters function. Cluster-specific marker genes were identified using the function FindMarkers with cutoffs: log2 fold-change >0.5 and adjusted P value < 0.05 (upregulated genes only). Clusters were annotated to cell types using SCSA (Cao et al, 2020). Differential composition analysis was performed using the R package sccomp (Mangiola et al, 2023).

Per-cluster differential expression testing between two groups was conducted using a pseudo-bulk approach. This is accomplished by summing counts together for all cells with the same combination of cluster and sample. We excluded lowly-expressed genes with an average read count lower than 10 across all samples/clusters. Raw counts were normalized using edgeR’s TMM with singleton pairing algorithm (Robinson and Oshlack, 2010) and were then transformed to log2-counts per million (log2CPM) using the voomLmFit function implemented in the R package limma (Ritchie et al, 2015). To assess differences in gene expression levels, we fitted a linear model using the lmfit function taking into account batch effects. Nominal P values were corrected for multiple testing using the Benjamini-Hochberg method. Gene set enrichment analysis based on pre-ranked gene list by t-statistic was performed using the R package fgsea (http://bioconductor.org/packages/fgsea/).

Chromatin immunoprecipitation sequencing (ChIP-Seq) and ChIP-qPCR

ChIP processing was performed on 2–3 mm² frozen tissue pieces and SK-BR-3 cells. Tissue samples were first minced and resuspended in ice-cold PBS with protease inhibitor cocktail before further being disaggregated with a dounce homogenizer, then filtered through a 70 μm filter to collect a single cell suspension. Collected cell pellets were resuspended in PBS containing 1% of formaldehyde for 10 min and the cross-linking reaction was quenched with glycine. Cell lysis was performed and isolated nuclei samples were transferred to a 1 ml tube for chromatin shearing (settings: 30 s on, 30 s off, 15 cycles). Fragmented chromatin were immunoprecipitated using an H3K27me3 antibody (Cell Signaling, Cat#9733, 1:100) for ChIP-Seq or an EZH2 antibody (Cell Signaling, Cat#5246, 1:100) and isolated with Cytiva Protein A/G Magnetic Beads. DNA cleanup was performed to manufacture instructions with the MinElute PCR purification kit (Qiagen, 28004). ChIP-qPCR was performed with SYBR Green MasterMix (Roche) and the LightCycler 480 instrument (Roche) in duplicate, and percentage of input method was used for analysis. Primer sequences were designed using Primer-BLAST (Ye et al, 2012), obtained from Integrated DNA Technologies and are described in the Reagents and Tools Table. For ChIP-Seq, post-sonication of isolated fragments was performed in 200ul tubes (Diagenode Bioruptor Sonicator) for further chromatin shearing to ~150 bp (settings: 30 sec on, 30 sec off, 20 cycles) for library preparation. Libraries were generated with NEBNext Ultra II DNA Library Prep kits (NEB, E7103L) and sequenced at ***SickKids TCAG facilities.

Analysis of publicly available datasets

EZH2 expression graphs in patients and according to breast cancer subtype were created using UALCAN (Chandrashekar et al, 2017; Chandrashekar et al, 2022). Kaplan–Meier survival graphs were generated from data available on KM Plotter (Gyorffy, 2024a, b). Human gene correlations of ESR1 and EZH2 expression were generated using XenaBrowser (Goldman et al, 2020) and cBioPortal (Cerami et al, 2012; de Bruijn et al, 2023; Gao et al, 2013) using publicly available datasets (METABRIC (Curtis et al, 2012; Pereira et al, 2016)). Relapse-free survival based on endocrine therapy was generated using ROC Plotter (Fekete and Gyorffy, 2019).

Statistical analysis

All graphs and statistical analyses were generated using GraphPad Prism 9.0 software unless otherwise indicated. The error bars shown in all figures are indicative of SD, unless otherwise noted. The methods for statistical tests, sample sizes (n), and P values (*P  <  0.05, **P  <  0.01, ***P  <  0.001, ****P  <  0.0001) are indicated in figure legends as appropriate, and a summary of statistical testing and individual P values for all relevant Figures are summarised in Dataset EV1.

Author contributions

Linshan Liu: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Writing—original draft; Writing—review and editing. Ellie J Massey: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Writing—original draft; Writing—review and editing. Dongmei Zuo: Data curation; Investigation; Methodology. Alain Pacis: Formal analysis; Methodology. Mert Demirdizen: Formal analysis; Methodology. Elizabeth Podleszanski: Data curation; Formal analysis; Investigation; Writing—review and editing. Jessica Cinkornpumin: Investigation; Methodology. Yu Gu: Investigation; Writing—review and editing. Hailey Proud: Investigation; Writing—review and editing. Virginie Sanguin-Gendreau: Data curation; Formal analysis; Investigation; Methodology. Vasilios Papavasiliou: Investigation. Ishtiaque Hossain: Investigation. Zhengze Jiang: Investigation. Harvey W Smith: Investigation; Writing—review and editing. William A Pastor: Supervision; Methodology; Writing—review and editing. Paolo Ceppi: Supervision; Methodology; Writing—review and editing. William J Muller: Conceptualization; Resources; Supervision; Funding acquisition; Writing—original draft; Project administration; Writing—review and editing.

Source data underlying figure panels in this paper may have individual authorship assigned. Where available, figure panel/source data authorship is listed in the following database record: biostudies:S-SCDT-10_1038-S44319-026-00691-x.

Data availability

The datasets produced in this study are available in the following databases: RNA-Seq data: Gene Expression Omnibus GSE303503. ChIP-Seq data: Gene Expression Omnibus GSE303502. scRNA-Seq Data: Gene Expression Omnibus GSE303527.

The source data of this paper are collected in the following database record: biostudies:S-SCDT-10_1038-S44319-026-00691-x.

Disclosure and competing interests statement

The authors declare no competing interests.

Supplementary information

Expanded view data, supplementary information, appendices are available for this paper at 10.1038/s44319-026-00691-x.