AKT and EZH2 inhibitors kill TNBCs by hijacking mechanisms of involution

Schade, Amy E.; Perurena, Naiara; Yang, Yoona; Rodriguez, Carrie L.; Krishnan, Anjana; Gardner, Alycia; Loi, Patrick; Xu, Yilin; Nguyen, Van T. M.; Mastellone, G. M.; Pilla, Natalie F.; Watanabe, Marina; Ota, Keiichi; Davis, Rachel A.; Mattioli, Kaia; Xiang, Dongxi; Zoeller, Jason J.; Lin, Jia-Ren; Morganti, Stefania; Garrido-Castro, Ana C.; Tolaney, Sara M.; Li, Zhe; Barbie, David A.; Sorger, Peter K.; Helin, Kristian; Santagata, Sandro; Knott, Simon R. V.; Cichowski, Karen

doi:10.1038/s41586-024-08031-6

Download PDF

Article
Open access
Published: 09 October 2024

AKT and EZH2 inhibitors kill TNBCs by hijacking mechanisms of involution

Nature volume 635, pages 755–763 (2024)Cite this article

36k Accesses
17 Citations
117 Altmetric
Metrics details

Subjects

Abstract

Triple-negative breast cancer (TNBC) is the most aggressive breast cancer subtype and has the highest rate of recurrence¹. The predominant standard of care for advanced TNBC is systemic chemotherapy with or without immunotherapy; however, responses are typically short lived^1,2. Thus, there is an urgent need to develop more effective treatments. Components of the PI3K pathway represent plausible therapeutic targets; more than 70% of TNBCs have alterations in PIK3CA, AKT1 or PTEN^3,4,5,6. However, in contrast to hormone-receptor-positive tumours, it is still unclear whether or how triple-negative disease will respond to PI3K pathway inhibitors⁷. Here we describe a promising AKT-inhibitor-based therapeutic combination for TNBC. Specifically, we show that AKT inhibitors synergize with agents that suppress the histone methyltransferase EZH2 and promote robust tumour regression in multiple TNBC models in vivo. AKT and EZH2 inhibitors exert these effects by first cooperatively driving basal-like TNBC cells into a more differentiated, luminal-like state, which cannot be effectively induced by either agent alone. Once TNBCs are differentiated, these agents kill them by hijacking signals that normally drive mammary gland involution. Using a machine learning approach, we developed a classifier that can be used to predict sensitivity. Together, these findings identify a promising therapeutic strategy for this highly aggressive tumour type and illustrate how deregulated epigenetic enzymes can insulate tumours from oncogenic vulnerabilities. These studies also reveal how developmental tissue-specific cell death pathways may be co-opted for therapeutic benefit.

Practical classification of triple-negative breast cancer: intratumoral heterogeneity, mechanisms of drug resistance, and novel therapies

Article Open access 16 October 2020

Treatment landscape of triple-negative breast cancer — expanded options, evolving needs

Article 09 November 2021

ASAH1 facilitates TNBC by DUSP5 suppression-driven activation of MAP kinase pathway and represents a therapeutic vulnerability

Article Open access 26 June 2024

Main

The PI3K pathway is frequently hyperactivated in breast cancers^3,4,5. In hormone receptor-expressing (HR⁺) luminal tumours, the most common mutations affect PIK3CA, which encodes the catalytic p110α subunit of PI3K^8,9. Accordingly, a PI3Kα-selective inhibitor, combined with an oestrogen receptor (ER) antagonist, is approved for advanced HR⁺ disease^10,11. However, PTEN mutations are more prevalent in basal-like TNBCs, which can function through PI3K p110β^3,10. Thus, agents that target downstream components, such as AKT, may be required to suppress PI3K signalling in these tumours. Nevertheless, an AKT inhibitor plus paclitaxel did not improve outcomes in a phase III TNBC trial⁷, although studies with other agents are pending¹². As such, it is still unclear whether or how PI3K pathway inhibitors will be useful in this aggressive breast cancer subtype.

Given the efficacy of PI3K inhibitors in luminal tumours, we investigated whether TNBCs could be sensitized to AKT inhibitors by modulating cell state. EZH2 is the catalytic component of PRC2, which promotes gene silencing by methylating histone H3 on lysine 27 (H3K27me3)¹³. EZH2 has a broad role in stem cell maintenance, and its overexpression promotes the development of many cancers, including TNBC^13,14,15,16. EZH2 is also required to maintain luminal progenitors and restrict luminal cell differentiation in mice^17,18, whereas its targeted overexpression drives intraductal hyperplasia¹⁹. Although EZH2 inhibitors do not affect the growth of established breast cancers, EZH2 suppression impairs TNBC metastasis²⁰. We therefore initially hypothesized that EZH2 inhibitors might sensitize TNBCs to AKT inhibitors by converting these poorly differentiated tumours to a more luminal-like state. Fortunately, several EZH2 inhibitors have been developed and one, tazemetostat, is FDA approved for other indications^21,22.

AKT and EZH2 inhibitors kill TNBCs

EZH2 is overexpressed in 82% of TNBCs and is readily detected in tumour cell nuclei (Fig. 1a,b and Extended Data Fig. 1a). A panel of TNBC cell lines were pretreated with tazemetostat for 5 days, followed by the addition of the AKT inhibitor ipatasertib (Extended Data Fig. 1b). 60% of cell lines were highly sensitive to this combination, resulting in a substantial depletion of cells within 4 days (Fig. 1c and Extended Data Fig. 1c,d). Live-cell imaging further demonstrated that AKT and EZH2 inhibitors do not effectively kill TNBCs as single agents but do so when combined, in a dose-dependent manner (Fig. 1d and Extended Data Fig. 1e). Notably, AKT and EZH2 inhibitors were highly synergistic in all sensitive cell lines, with highest single agent (HSA) scores of ≥10 (Fig. 1e and Extended Data Fig. 1f). Similar effects were observed when ipatasertib was replaced with a different AKT inhibitor (Extended Data Fig. 2a) or when tazemetostat was substituted with an EED inhibitor, which suppresses another obligate PRC2 component (Extended Data Fig. 2b). These agents did not exert cytotoxic effects in non-transformed epithelial cells or in other primary or unrelated cell lines (Extended Data Fig. 2c,d). Responses to combined agents were durable, in contrast to the effects of ipatasertib alone, which slowed cell proliferation but permitted regrowth (Extended Data Fig. 2e). Sequential dosing of tazemetostat followed by ipatasertib exerted a similar cytotoxic response, but the effects were less robust when a 4 day gap between drugs was included, suggesting that, although concomitant dosing may not be required, maximal responses require a sustained loss of H3K27me3 (Extended Data Fig. 2f).

**Fig. 1: AKT and EZH2 inhibitors potently synergize, kill TNBCs and promote substantial tumour regression in vivo.**

EZH2 was overexpressed in all cell lines, and the relative expression levels did not correlate with sensitivity (Extended Data Fig. 3a); however, PTEN loss was enriched in sensitive cells (Fisher’s exact test, P = 0.045; Extended Data Fig. 3b). Specifically, PTEN was absent in all of the sensitive lines except for one that had a PIK3CA mutation (Extended Data Fig. 3b), and was retained in all but one of the resistant lines, although a second line had a PIK3CA mutation (Extended Data Fig. 3b). Thus, although aberrant PI3K pathway activation, often conferred by PTEN loss in TNBCs, probably contributes to sensitivity to this combination, it is not the only determinant.

Xenograft, GEMM-derived allograft and patient-derived xenograft (PDX) models were used to evaluate efficacy in vivo. Mice with established orthotopic tumours (>100 mm³) were pretreated with EZH2 inhibitor (EZH2i) for 1 week, before addition of the AKT inhibitor (AKTi). No or modest cytostatic responses were observed with single agents (Fig. 1f–l and Extended Data Fig. 3c), but, when combined, EZH2i + AKTi (EZH2/AKTi) triggered potent tumour regression in all five models. MDA-MB-468 and SUM149PT xenografts regressed 42–73% over 28 days (Fig. 1f,g), and survival increased from 46 to 116 days, despite treatment cessation (Extended Data Fig. 3d). Notably, this drug combination was well tolerated in vivo (Extended Data Fig. 3e). EZH2/AKT inhibitors similarly induced regression of p53-mutant GEMM-derived TNBC organoid allografts (Fig. 1h), in which PI3K–AKT is hyperactivated by spontaneous AKT3 and MET copy-number gains^23,24 and EZH2 is amplified and overexpressed (Extended Data Fig. 3g,h). Importantly, EZH2/AKT inhibitors were effective in patient-derived xenograft models. HCI-004 tumours²⁵, which exhibit PTEN copy loss and overexpress EZH2, regressed 60–88%, with one complete response (Fig. 1i). Tumour regression was durable throughout the entire study (7.5 weeks) (Fig. 1j). Multiplexed cyclic immunofluorescence (CycIF) imaging of PDX tumours treated in vivo demonstrated a potent loss of Ki-67 (pink) and concomitant induction of cleaved PARP (white) within 1 day of treatment (Fig. 1k). HCI-025 tumours²⁵, which have a PIK3CA mutation, PTEN copy loss and EZH2 overexpression, also regressed 28–57% (Fig. 1l). Although frank regression was more variable, histological analysis of tumours that appeared to arrest rather than shrink revealed a substantial loss of tumour cells internally with extensive necrosis, fibrosis and calcification (Extended Data Fig. 4). Residual viable tumour cells at the end point comprised only 10–40% of the area and were largely devoid of Ki-67 (Extended Data Fig. 4). Thus, the remaining tissue consisted mostly of extracellular matrix and necrotic deposits, with a small number of arrested cells, consistent with the durable response of HCI-004 tumours (Fig. 1i). Potent tumour regression, confirmed here in multiple PDX, xenograft and GEMM models, is rarely observed in TNBC, underscoring the potential therapeutic use of this combination.

Differentiation is required for the response

RNA sequencing (RNA-seq) analysis was performed on cells treated with EZH2 and/or AKT inhibitors after 24 h, before the onset of substantial cell death. Notably, the combination triggered a potent suppression of genes associated with basal and mammary stem cells (such as KRT5, KRT14 and VIM), and the concomitant upregulation of luminal markers (such as GATA3 and ELF3)²⁶ (Extended Data Fig. 5a). While EZH2 inhibitors have been reported to modestly increase the expression of some luminal markers in a TNBC model²⁰, single-sample gene set enrichment analysis (ssGSEA) revealed that EZH2 and AKT inhibitors were both required to maximally induce luminal signatures and suppress basal and mammary stem cell signatures in TNBCs²⁶ (Fig. 2a and Extended Data Fig. 5b). Transcriptional heat maps of the most highly upregulated and downregulated genes illustrate the cooperative nature of this response (Fig. 2b and Extended Data Fig. 5c).

**Fig. 2: Combined EZH2/AKT inhibitors are required for maximal differentiation and the therapeutic response.**

Transcriptional profiles have been used to classify TNBC tumours into basal-like 1 (BL1), basal-like 2 (BL2), mesenchymal and luminal-AR subtypes (termed TNBCtypes)^27,28. In SUM149PT cells EZH2/AKTi cooperatively suppressed the BL2 signature and induced the TNBCtype luminal-AR tumour signature (Fig. 2c). A similar suppression of basal-like TNBCtype signatures was observed in all five sensitive lines evaluated (Fig. 2c and Extended Data Fig. 5d). Notably, ER was not upregulated by EZH2/AKTi, but AR expression did increase slightly (Extended Data Fig. 5e,f). Regardless, these findings suggest that EZH2 and AKT inhibitors cooperatively drive basal TNBCs to a more luminal-like cell state.

To determine whether this transformation was occurring in vivo, further CyCIF imaging was conducted on HCI-004 PDX tumours (Fig. 2d). Vehicle-treated tumours predominantly expressed the basal cytokeratin (CK14, cyan) and minimally expressed the luminal cytokeratin (CK8, magenta) (Fig. 2d). However, EZH2/AKTi induced a marked shift in cell state, with a concomitant loss of basal and gain of luminal markers (that is, CK14 to CK8) (Fig. 2d (top)). Moreover, although cleaved PARP was minimally detected in vehicle-treated tumours, it was highly expressed in CK8 expressing cells after 1–2 days of treatment, indicating that cells with luminal features were rapidly undergoing apoptosis (Fig. 2d (bottom)). A ring of CK14-positive cells remained on the exterior of the tumour for the first few days; however, after 30 days it appeared as if the hollowed tumour core collapsed, leaving CK14-positive cells and a smaller fraction of CK8-positive cells (Fig. 2d). At this timepoint, tumour cells were no longer dying, illustrated by the absence of PARP cleavage (Fig. 2d (bottom)), but residual tumour remnants exhibited minimal proliferation compared to vehicle-treated tumours (Fig. 2e), similar to what was observed in PDX HCI-025 (Extended Data Fig. 4). These results demonstrate that EZH2 and AKT inhibitors drive TNBCs into a luminal-like state in vivo, which results in rapid cell death and durable tumour regression.

Consistent with the induction of a luminal-like cell state, AKTi and EZH2i were both required to induce maximal expression of GATA3, a master developmental transcription factor that controls luminal differentiation^29,30, which was confirmed by mRNA and protein expression (Fig. 2f,g and Extended Data Fig. 5g). Moreover, GATA3 ablation using siRNAs, short hairpin RNAs (shRNAs) or sgRNAs prevented the cytotoxic response and the synergy observed between AKTi and EZH2i in multiple cell lines in vitro (Fig. 2h and Extended Data Fig. 5h–k) and in vivo (Fig. 2i). Notably, in the absence of GATA3, EZH2/AKTi still exerted cytostatic effects (Fig. 2h,i), consistent with known individual effects of AKTi and EZH2i on the expression of various cell cycle genes^31,32. While the individual contributions of EZH2i and AKTi were not investigated exhaustively, upregulation of p27 and concomitant suppression of the RB–E2F pathway was consistently observed in response to ipatasertib, which was unaffected by GATA3 ablation (Extended Data Fig. 5l–n). These findings suggest that GATA3-mediated differentiation is essential for tumour regression, but AKTi and EZH2i can exert effects on other targets, which may impede the proliferation of the remaining cells.

Predicting response by machine learning

This luminal-like differentiation program was defective in resistant cells. For example, studies using assay for transposase-accessible chromatin with sequencing (ATAC–seq) revealed that GATA3-binding sites became accessible in response to EZH2/AKTi only in sensitive cells (Fig. 3a) and that GATA3 expression was not induced by the combination in resistant cells (Fig. 3b). Principal component analysis of ATAC–seq data from untreated cells further showed that sensitive and resistant cells could be readily differentiated by baseline epigenetic differences (Fig. 3c). Notably, in sensitive cells, the top enriched genes associated with open chromatin were luminal and basal genes, whereas mesenchymal gene loci were identified as the most differentially accessible in resistant cells (Extended Data Fig. 6a). This was accompanied by the enrichment of either basal or mesenchymal genes in sensitive versus resistant cells, respectively (Extended Data Fig. 6b,c). These findings suggest that sensitive cells may be poised to undergo a luminal-like state change, whereas resistant cells appear to be inherently less differentiated or more mesenchymal, with a chromatin state that may restrict this differentiation program.

**Fig. 3: Identifying a sensitive TNBC cell state using a machine learning approach.**

Despite the baseline differences, TNBCtype or broad mesenchymal signatures were not sufficient to predict sensitivity versus resistance to EZH2/AKTi. Thus, a machine learning approach was performed using CCLE RNA-seq data, which showed little variance from transcriptional data generated in this study (Fig. 3d and Extended Data Fig. 6d). Specifically, we used a leave-one-out cross-validation strategy using transcriptional and sensitivity data from the 12 cell lines in Fig. 1c together with 5 additional cell lines (Extended Data Fig. 6e). For each iteration, either the most variable or the most differentially expressed genes between sensitive and insensitive lines were used in the training set, as specified in Fig. 3d. Random forest (RF) and support vector machine (SVM) algorithms were then evaluated for their ability to predict the sensitivity status of the left-out cell lines. We found that models that were trained using the RF algorithm on genes differentially expressed with an absolute log-transformed fold change of >5 could most accurately predict sensitivity (Extended Data Fig. 6f). When this model was applied to transcriptional data from TCGA Firehose TNBC tumour samples, 55% of tumours were predicted to be sensitive to EZH2/AKTi, (Fig. 3d), similar to the empirical observation that 60% of TNBC cell lines were indeed sensitive to these agents. Notably, while PTEN alterations were enriched in sensitive cell lines, closer analysis of TCGA tumours suggests that around 93% of TNBCs have at least one alteration in a PI3K pathway component (PTEN, PI3KCA, AKT3 and INPP4B; Extended Data Fig. 6g). We therefore postulate that the signature generated here will probably be a better predictor of sensitivity, which can be evaluated in future clinical trials.

Cooperative GATA3 induction through FOXO1

As GATA3 was essential for the therapeutic response, we sought to elucidate the precise contribution of EZH2 and AKT inhibition in its induction. GATA3 is known to be a direct target of PRC2³³. Nevertheless, maximal GATA3 expression required both EZH2 and AKT inhibitors (Fig. 2f,g). Hi-C data were therefore used to identify GATA3 enhancer sequences (Extended Data Fig. 7a). Integration of Hi-C and CUT&RUN data showed that GATA3 enhancer sequences were bound by H3K27me3, which was lost in response to EZH2 inhibition (Fig. 4a and Extended Data Fig. 7b). ATAC–seq analysis further demonstrated that EZH2i and AKTi induced accessibility at multiple regions within this enhancer (Fig. 4a and Extended Data Fig. 7b).

**Fig. 4: EZH2 and AKT inhibitors drive differentiation by opening chromatin and cooperatively engaging FOXO1.**

To investigate transcription factor(s) that are involved in GATA3 induction, we first identified proteins with binding sites present in at least one enhancer peak and eliminated those that are not expressed in TNBC lines (the strategy is shown in Extended Data Fig. 7c). We then used single-cell ATAC–seq data from normal mammary glands to identify biologically relevant factors and selected those with motifs that were differentially accessible in luminal versus basal cell populations³⁴. We then focused solely on transcription factors with binding sites that became accessible in response to EZH2/AKTi within the enhancer and also had binding sites within the GATA3 promoter.

A screen using small interfering RNA (siRNA) was performed to identify which of these transcription factors, when suppressed, could prevent cell death in response to EZH2/AKTi (Fig. 4b). Transcription factors that scored in this assay were then assessed for their ability to suppress GATA3 induction (Fig. 4c). This analysis identified FOXO1 as a critical regulator of the therapeutic response (Fig. 4b) and the most potent regulator of GATA3 induction (Fig. 4c). Importantly, we confirmed that FOXO1 ablation could suppress GATA3 induction and the therapeutic response in multiple cell lines (Fig. 4d,e).

AKT is known to inhibit nuclear localization and transcriptional activity of FOXO1 through phosphorylation³⁵. Indeed, AKTi rapidly triggered FOXO1 dephosphorylation at negative-regulatory Ser256 in multiple TNBC cell lines (Extended Data Fig. 8a). However, we noted that EZH2 inhibitors also induced total FOXO1 expression in these cells (Extended Data Fig. 8a,b). FOXO1 has been reported to be a repressed EZH2 target³⁶, and, consistent with this notion, we found that tazemetostat inhibited H3K27me3 at the FOXO1 promoter and induced its transcription (Extended Data Fig. 8a–c). These findings suggest that AKT and EZH2 inhibitors both positively regulate FOXO1 in TNBCs. AKT inhibitors permit its transcriptional activity in the nucleus through its dephosphorylation, but EZH2 inhibitors may further enhance these effects by increasing its expression.

While EZH2 inhibitors were sufficient to promote H3K27me3 and open chromatin at the GATA3 enhancer (Fig. 4a), we predicted that AKT inhibitors would be required for maximal FOXO1 binding. FOXO1 chromatin immunoprecipitation followed by sequencing (ChIP–seq) data were used to precisely define FOXO1-binding sites in the GATA3 enhancer and promoter (Extended Data Fig. 8d,e). EZH2 and AKT inhibitors cooperatively induced FOXO1 binding at all of these sites (Fig. 4f). Taken together, these data demonstrate that FOXO1 is required for both the induction of GATA3 and the cytotoxic response to EZH2 and AKT inhibitors. Mechanistically, this occurs because tazemetostat opens chromatin at the GATA3 enhancer and increases FOXO1 expression, while ipatasertib is required to induce dephosphorylation of FOXO1 at critical negative regulatory sites.

Death is triggered by BMF upregulation

In addition to the upregulation of luminal markers, ssGSEA revealed that EZH2/AKTi cooperatively induced apoptotic signatures (Extended Data Fig. 9a). BMF, a pro-apoptotic BH3-only protein, was the most significantly and cooperatively upregulated proapoptotic gene and was induced in all sensitive cell lines^37,38 (Fig. 5a and Extended Data Fig. 9b). As BMF antibodies are not suitable for immunoblotting, an HA tag was engineered into the endogenous BMF locus using CRISPR technology. We indeed confirmed that the BMF protein was cooperatively upregulated by EZH2/AKTi (Extended Data Fig. 9c). Suppression of BMF by siRNAs prevented the cytotoxic response in multiple cell lines (Fig. 5b,c and Extended Data Fig. 9d,e), demonstrating that its upregulation is also essential for the therapeutic response to EZH2/AKTi.

**Fig. 5: AKT and EZH2 inhibitors trigger cell death by co-opting signals that normally drive involution.**

EZH2/AKTi co-opt involution signals

This finding was intriguing because BMF induction has a critical role in triggering cell death during normal mammary gland involution—the programmed destruction of luminal ductal and alveolar cells after the cessation of lactation^37,38. Conversely, EZH2 is most highly expressed in ductal cells and alveoli during pregnancy and is substantially reduced during involution¹⁷. Notably, ectopic expression of either EZH2 or AKT in mice has been shown to delay involution, whereas EZH2 or AKT1 ablation accelerates involution^{19,39,40,41,42}. Thus, in addition to shifting TNBCs to a more differentiated luminal-like state, we hypothesized that EZH2/AKTi might be killing TNBCs by activating signals that normally drive the involution of luminal cells.

Involution is regulated by JAK1–STAT3, which triggers a STAT3-dependent transcriptional signature that includes BMF^37,43,44. Notably, we found that EZH2/AKTi cooperatively induced involution-associated transcriptional signatures in SUM149PT and MDA-MB-468 cells (Fig. 5d). Transcriptional heat maps demonstrate that around three-quarters of involution-associated genes are upregulated in drug treated TNBC cells and that AKT and EZH2 inhibitors both contribute to this response (Fig. 5e and Extended Data Fig. 9f). Moreover, STAT3 ablation prevented the upregulation of BMF (Fig. 5f) and other involution signature genes, as shown by quantitative PCR (qPCR; Extended Data Fig. 9g). JAK1 ablation similarly suppressed BMF expression, which was confirmed at the protein level (Fig. 5g). Finally, AKT and EZH2 inhibitors cooperatively induced STAT3 phosphorylation in TNBC cells (Fig. 5h). The induction of involution signatures and activation of the JAK1–STAT3–BMF pathway were not a general consequence of cell death signals, because concentrations of docetaxel that effectively killed TNBCs did not induce STAT3 phosphorylation, increase BMF expression or stimulate the expression of involution-related genes, at any timepoint (Extended Data Fig. 9h–l). Thus, EZH2/AKTi triggers a transcriptional program in TNBCs that resembles the transcriptional response induced during mammary gland involution. This JAK1–STAT3-driven pathway culminates in the potent upregulation of BMF, which is required for cell death.

Finally, we confirmed that genetic inhibition of STAT3 (Fig. 5i and Extended Data Fig. 9m,n), as well as genetic or chemical inhibition of JAK1 (Fig. 5j,k and Extended Data Fig. 9o), prevented cell death in response to EZH2/AKTi in multiple TNBC lines. Consistent with the cooperative induction of STAT3 phosphorylation and BMF expression, EZH2 and AKT inhibitors were both required to maximally induce STAT3 binding to sites within the BMF locus that regulate its expression during involution (Fig. 5l). Thus, AKT and EZH2 inhibitors kill TNBCs by first cooperatively driving luminal differentiation and then together activating an involution-like cell death program.

Involution-related cytokines induce BMF

In considering how EZH2 and AKT inhibitors might cooperatively trigger BMF expression and cell death, we noted that both agents were required for maximal JAK1–STAT3 activation (Fig. 5). During involution, JAK1 is activated by the IL-6 family of cytokines, including LIF, OSM and IL-6 itself. While OSM was not appreciably expressed and LIF was not consistently induced in TNBCs, IL-6 was cooperatively upregulated by EZH2 and AKT inhibitors, which was confirmed by enzyme-linked immunosorbent assay (ELISA; Extended Data Fig. 10a). Genetic ablation of the IL-6 receptor (IL-6R) potently blocked the cytotoxic response to EZH2 and AKT inhibitors in multiple cell lines (Fig. 6a and Extended Data Fig. 10b), which could also be recapitulated by anti-IL-6R antibodies (Extended Data Fig. 10c). Importantly, IL-6R was necessary for the induction of BMF expression (Fig. 6b) and activation of STAT3, as measured by its phosphorylation, in response to EZH2 and AKT inhibitors (Fig. 6c). These results suggest that the cooperativity of EZH2 and AKT inhibitors on the JAK1–STAT3–BMF pathway was conferred at the level of cytokine production.

**Fig. 6: AKT and EZH2 inhibitors trigger involution-related cytokine production by cooperatively activating STING–TBK1 signalling.**

STING–TBK1 activation induces cytokines

EZH2 inhibitors are known to activate the STING–TBK1 pathway in a variety of tumour types^45,46,47, which can subsequently induce IRF3 and type I interferons or NF-kB and cytokines^48,49. Notably, both genetic ablation of STING and TBK1 inhibition prevented cell death in response to EZH2 and AKT inhibitors (Fig. 6d) and potently blocked IL-6 production (Fig. 6e and Extended Data Fig. 10d). The STING–TBK1 pathway is activated by CGAS-mediated production of 2′3′-cGAMP, which promotes oligomerization of active STING dimers that subsequently bind to and activate TBK1. Notably, EZH2 inhibition alone induced 2′3′-cGAMP production, which was not enhanced by AKT inhibition (Extended Data Fig. 10e). Consistent with previous observations that EZH2 inhibition can induce 2′3′-cGAMP by derepressing retrotransposable elements and endogenous retroviruses^47,50,51,52, we observed increased expression of endogenous retroviruses in MDA-MB-468 cells treated with tazemetostat alone (Extended Data Fig. 10f). Nevertheless, both EZH2i and AKTi contributed to the formation of STING–TBK1 complexes (Fig. 6f) and subsequent TBK1 activation, as evidenced by autophosphorylation at Ser172 (Fig. 6g). These findings demonstrate that the STING–TBK1 axis is essential for IL-6 production in this context and that both EZH2 and AKT inhibitors promote this response. EZH2 inhibitors do so by activating CGAS/2′3′-cGAMP, whereas AKT inhibitors function further downstream, by enhancing the physical interaction between STING and TBK1. We did not detect any changes in TBK1 or STING phosphorylation using AKT substrate antibodies; however, the precise mechanisms that regulate the NF-κB/cytokine axis of the STING pathway are still unclear^48,49. Regardless, these findings provide a defined launching point for future analysis and reveal a role for AKT in this process.

GATA3 and STING agonists reprogram TNBCs

Together, these data suggest that the therapeutic effects of EZH2 and AKT inhibitors require (1) differentiation driven by GATA3 and (2) the induction of BMF through STING. As shown in Fig. 3b, EZH2 and AKT inhibitors do not induce GATA3 in resistant cell lines. However, we also noted that BMF expression minimally increased, if at all (Extended Data Fig. 10g). We therefore investigated whether resistant cells could be sensitized by ectopically enforcing both signals. Indeed, overexpression of GATA3 was not sufficient to sensitize cells to EZH2 and AKT inhibitors, and STING agonists alone were also not able to do so (Fig. 6h and Extended Data Fig. 10h). However, cells that both overexpressed GATA3 and were exposed to STING agonists became sensitive to EZH2/AKTi (Extended Data Fig. 10h), with cooperative effects on cell death readily observed by live-cell imaging (Fig. 6h). As expected, GATA3 expression and STING agonists alone were not sufficient to kill cells on their own (Fig. 6h) because EZH2 and AKT inhibitors are still required to globally open chromatin at GATA3 sites and engage TBK1; however, these findings demonstrate that GATA3 and STING agonists can reprogram resistant TNBCs, therefore conferring sensitivity to EZH2 and AKT inhibitors. A model illustrating this mechanism is shown in Fig. 6i.

Discussion

Standard chemotherapy is not sufficiently effective in advanced TNBC. Even in individuals with PD-L1-positive tumours, who are candidates for chemotherapy combined with immune checkpoint inhibitors, the median overall survival is around 2 years^53,54. Thus, developing improved treatments for TNBC is of paramount importance. The success of PI3K inhibitors in HR⁺ tumours has fuelled enthusiasm for targeting this pathway in TNBC^3,4,10,11. Nevertheless, clinical trials with PI3K pathway inhibitors in this indication to date have been disappointing⁵⁵. Here we demonstrate how PI3K-pathway inhibitors, specifically AKT inhibitors, can be rendered effective in TNBC, revealing a promising therapeutic strategy for this aggressive disease.

We report that AKT and EZH2 inhibitors synergize and together promote frank tumour regression in multiple TNBC models in vivo. Cell death is preceded by differentiation of TNBCs to a more luminal-like state, which requires both AKT and EZH2 suppression. Importantly, this shift in cell state is required for the therapeutic response; cell death is blocked by inhibiting it and resistant tumour cells do not differentiate in response to EZH2/AKTi.

Mechanistically, EZH2 inhibition opens chromatin at enhancer sequences that regulate GATA3. AKTi then suppresses FOXO1 phosphorylation at critical negative regulatory sites and triggers FOXO1 binding at enhancer and promoter sequences. Once differentiated, EZH2 and AKT inhibitors drive apoptosis by hijacking the same regulatory signals that kill luminal cells during the process of involution. This occurs through the induction of an IL-6–JAK1–STAT3 pathway, which triggers the expression of the pro-apoptotic protein BMF. Moreover, in this therapeutic context, IL-6 induction is regulated by the STING–TBK1 pathway, which is also cooperatively regulated by EZH2 and AKT inhibitors.

Sensitive and resistant cells can be readily distinguished by ATAC–seq profiles, suggesting important baseline differences in their epigenetic state. This observation is consistent with the inability of resistant cells to differentiate into a more luminal-like state. Fortunately, a machine learning approach yielded a transcriptional classifier that can be used to select individuals with tumours that are most likely to respond. Moreover, we found that resistant cells can be reprogramed by ectopically enforcing differentiation and cell death signals through GATA3 expression and STING agonists, suggesting that resistance is reversible.

Together, these studies suggest that EZH2 functions as an epigenetic insulator in TNBCs, shielding tumour cells from luminal differentiation and cell death, by suppressing the expression of GATA3 and BMF. However, AKT has an equally important role in this process, as EZH2 and AKT suppression are both required for maximal upregulation of these genes and cell death through mechanisms described here. Notably, EZH2 and AKT (suppression) have independently been implicated in mammary gland involution in mice, and the expression of both proteins plummets at the onset of involution^{17,19,39,40,42}. However, the precise mechanisms by which EZH2 and/or AKT function in this context are not well understood. In TNBCs, we have shown that these proteins cooperatively regulate the activation of IL-6, JAK1, STAT3 and BMF—key components that regulate cell death during involution. However, we cannot exclude the possibility that involution genes might be generally triggered by some apoptotic inducers and/or that other apoptotic genes not normally involved in involution may contribute to the response. Nevertheless, JAK1, STAT3 and BMF are three components that are essential for cell death in both contexts.

It will be interesting to determine whether EZH2 and AKT inhibitors synergize in other tumour types and, if so, whether identical or other tissue-specific mechanisms are engaged. Certainly, any shift in cell state, if observed or required, would probably be mediated by tissue-specific factors and/or could be instructed by tissue-specific epigenetic states. It will also be equally interesting to investigate whether this paradigm extends beyond these two proteins and their inhibitors in breast cancer and whether epigenetic agents may more broadly sensitize tumours to canonical oncogenic inhibitors. If so, deconstructing how specific epigenetic and oncogenic pathways converge will be critical for the rational design of future combinatorial strategies. Regardless, the data presented here provide compelling support for the development of clinical trials exploring the combined effects of AKT and EZH2 inhibitors in TNBC.

Methods

Cell counting

To measure cellular proliferation and cytotoxicity, manual cell counting experiments were conducted. On day −6, cells were seeded at 40% confluency in 10 cm dishes. On day −5, cells were treated with DMSO or tazemetostat. On day −3, cells were passaged 1:2 and maintained in tazemetostat or DMSO. On day −1, cells were seeded at 250,000 cells per well into six-well tissue culture dishes for counting (enough for day 0 and for day 4 counts) and 625,000 cells per 6 cm dish for protein lysate collection. For cell counting assays including siRNA, cells were transfected on day −1 and incubated with siRNA for 6–8 h before seeding. On day 0, starting cell counts were measured using a haemocytometer and the remaining cells were dosed with DMSO, ipatasertib, tazemetostat or the combination. On day 1, protein lysates were collected. On day 4, remaining cells were counted using a haemocytometer. The log₂-transformed fold change in cell number was calculated by normalizing day 4 cell count values to day 0 cell count values. For days −6 to −1, cells were cultured in appropriate medium containing 10% FBS. For days 0 to 4, cells were cultured in appropriate medium containing 2% FBS. All of the experiments were completed at least three times. Statistical methods were not used to pre-determine sample size. Blinding was not conducted. Cells were seeded from a master solution and allocated across all compared conditions to control for seeding and population differences.

IncuCyte live-cell imaging

Live-cell imaging was completed using IncuCyte live-cell imaging. After 5 days of pretreatment with tazemetostat or DMSO, cells were seeded at 5,000 cells per well in 96-well plates. Then, 24 h after seeding, cells were treated with tazemetostat and/or ipatasertib in medium containing NucLight Rapid Red Reagent (Sartorius, 4717) to label nuclei and Cytotox Green (Sartorius, 4633) to label cytotoxic cells. Plates were placed inside the IncuCyte machine and images were acquired every 2 h for 5 days. The percentage of cytotoxic cells was determined using the integrated IncuCyte software by quantifying the overlap of NucLight- and Cytotox-green-positive cells divided by the total number of NucLight-positive cells. Four images per well were taken and each condition was seeded in duplicate. All of the experiments were completed at least three times.

Synergy score analysis

To measure synergistic interactions between EZH2 inhibitor tazemetostat combined with PI3K–AKT pathway inhibitors, cells were seeded in 96-well white flat bottom plates at 5,000 cells per well. For each condition, two technical replicates were seeded per combination treatment. After 5 days of pretreatment with EZH2 inhibitor, cells were dosed with PI3K–AKT pathways inhibitors. Drugs were added at the following concentrations: EZH2 inhibitor tazemetostat (0, 1.25, 2.5, 5, 10 μM), and AKT inhibitor Ipatasertib (0, 0.625, 1.25, 2.5, 5, 10 μM). After 96 h of treatment, cell viability was quantified using CellTiter-Glo (Promega, G9291) and normalized to DMSO to calculate the inhibitory response. SynergyFinder was then used to calculate the synergy score using the Gaddum’s non-interaction model—highest single agent (HSA) model, where a value of greater than 10 indicates a synergistic interaction⁵⁶.

Orthotopic xenografts

All mouse work was done in compliance with the Institutional Animal Care and Use Committee (IACUC) at Brigham and Women’s Hospital. The maximum tumour size was 20 mm in one dimension of the tumour. SUM149PT and MDA-MB-468 xenografts were generated by injecting 3 × 10⁶ cells using a 1:1 ratio of DMEM/F12 and Matrigel (Corning, 356231) into the fourth inguinal mammary fat pad of nude mice. For the organoid allograft, organoids were generated from Trp53^fl/fl mice and cultured as previously described^23,24 and 50,000 cells were implanted into the fourth inguinal mammary fat pad of nude mice. For SUM149PT and MDA-MB-468 xenograft studies, two tumours per mouse were implanted. For the organoid allograft, MDA-MB-468 xenograft survival study and PDXs, one tumour per mouse was implanted. Tumour fragments of HCI-025 and HCI-004 were implanted into NSG mice. Tumours arose after 2 weeks and were enrolled when they reached 100–250 mm³. The tumour volume was measured using vernier callipers with the formula (width × width × length × 0.52). For the treatment schedule, mice were pretreated with tazemetostat or vehicle for 7 days before being treated with vehicle (ipatasertib vehicle + tazemetostat vehicle), ipatasertib (ipatasertib + tazemetostat vehicle), tazemetostat (ipatasertib vehicle + tazemetostat) or combination (tazemetostat + ipatasertib). Ipatasertib was dosed at 50 mg per kg once per day through oral gavage. Ipatasertib was prepared in 0.5% methylcellulose, viscosity 4,000 cP, with 2% Tween-80 at pH 7.0 and kept at 4 °C for up to 1 week. Tazemetostat was dosed at 250 mg per kg twice a day through oral gavage. Tazemetostat was prepared in 0.5% methylcellulose containing 0.1% Tween-80 and kept at 4 °C for up to 4 weeks. PDXs HCI-025 and HCI-004 were prepared and passaged as previously described^25,57. Sample size was pre-determined: we calculated that if we used 10 mice per treatment group, we would have 87% power to detect a 50% reduction in tumor size, and every effort was made to reach that sample size. Animals were randomized into treatment arms. Blinding was not conducted. The temperature in rooms housing mice was kept stably between 68 and 75 °C. The humidity range of rooms housing mice was kept stably at 35–65%, with 50% humidity being considered optimal. Animals experienced 12 h of light during the day and 12 h of dark during the night with a light intensity between 100–250 lux. Mice were offered diet ab libitum in food hoppers hung from the top of the cage with Pico 5052 Irradiated Rodent Chow or Pico 5058 Irradiated Mouse Chow. Mouse were offered water ad libitum. Mice were housed in Allentown ventilated cages with bedding approximately 1/2 inch in depth with Alpha-blend or crinkled paper (nude mice received paper-based bedding only). Mice were offered gnawing enrichment in the form of nestlets or popsicle sticks (nude mice were offered only popsicle sticks). Raw animal data are provided in Supplementary Data 2.

Cell lines

Cell lines were purchased directly from ATCC or authenticated using STR analysis (Labcorp). SUM149PT and SUM159PT cells were cultured in DMEM/F12 (Gibco, 11330-057), 293T, RPE1, MDA-MB-157, MDA-MB-231, MDA-MB-436 and MDA-MB-468 cells were cultured in DMEM (Corning, 10-013-CV), and BT20, BT549, DU4457, HCC38, HCC70, HCC1143, HCC1187, HCC1395, HCC1806, HCC1937 and HCC2157 cells were cultured in RPMI (Corning, 10-040-CV), IMR90 and BJ fibroblast cell lines were cultured in EMEM (Thermo Fisher Scientific, 10-009-CV). All media were supplemented with 10% FBS, and 1× concentration of penicillin–streptomycin–glutamine (Gibco, 10378016). TP53^−/− organoids were generated and cultured as previously reported^23,24. MCF10A cells were cultured as previously reported⁵⁸. No commonly misidentified cell lines were used in this study. Cell lines were routinely found to have no mycoplasma contamination using Lonza MycoAlert PLUS Mycoplasma Detection Kit.

Crystal violet

Cells were seeded at 50,000 cells per well into 12-well plates and then stained with crystal violet at the indicated times. In brief, after the medium was removed, wells were washed with PBS three times before fixation with 2% formaldehyde (1:2 dilution of Formalde-Fresh Solution; SF94-4) for 15 min a room temperature. After fixation, wells were washed with PBS once before staining with 0.025% crystal violet (Sigma-Aldrich, C6158) for 2 h. After staining, wells were washed with water four times before air drying and imaging using a standard scanner.

Immunohistochemistry

Tumours were collected for immunohistochemistry 4 h after final dose of drug treatment by removing tumour and fixing in Formaldehyde-Fresh (Thermo Fisher Scientific, SF94-4) for 24 h. After 24 h, tumours were stored in 70% ethanol before sectioning and analysis. For haematoxylin and eosin staining, sectioning and staining was performed at the Harvard Medical School Rodent Histopathology Core. For Ki-67 and EZH2 (BD Bio, 612667) immunohistochemistry, sectioning and staining was performed at the Brigham & Women’s Hospital Pathology Core.

siRNA and CRISPR

Cells were incubated for 6–8 h with 0.1 µM siRNA constructs using a 1:400 dilution of Lipofectamine RNAiMAX Transfection Reagent (Invitrogen, 13778075) in antibiotic-free medium. siRNA ON-TARGETplus pools were purchased from Horizon Biosciences: control (D-001810-10-50), GATA3 (L-003781-00-0005), JAK1 (L-004393-00-0010), BMF (L-003145-00-0005), STAT3 (L-003544-00-0005), IL-6R (007994-00-0005), BACH2 (L-009787-00), HSF1 (L-012109-02), MEF2D (L-009884-00), NFIA (L-008661-00), NFIX (L-009250-00), PRDM1 (L-009322-00), RELB (L-004767-00), ZNF143 (L-013965-00), ZNF354C (L-014199-02), BACH1 (L-007750-00), ELF3 (L-016080-00-0005), FOS (L-003265-00), FOXO1 (003006-00-0005), FOXO3 (003007-00-0005), HOXC11 (L-017602-00-0005), HOXD11 (L-013095-00-0005), HOXD12 (L-013096-02-0005), JUN (L-003268-00), MAFK (L-008580-00), SIX2 (L-017024-00), ETV1 (L-003801-00), ETV4 (L-004207-00), FOSL1 (L-004341-00), FOSL2 (L-004110-00), HOXC10 (L-017601-00), JDP2 (L-008321-01), JUNB (L-003269-00), JUND (L-003900-00), MAFF (L-003903-00), MAFG (L-009109-00), MAFK (L-008580-00), MITF (L-008674-00), MXI1 (L-009947-00), PBX3 (L-020121-00), PKNOX1 (L-021413-00), SIX1 (L-020093-00), SREBF2 (L-009549-00), TBX1 (L-012195-00), USF1 (L-003617-00), USF2 (L-003618-00) and VDR (L-003448-00). shRNA constructs in the pLKO.1 backbone against non-targeting and GATA3 (TRCN0000019299) were purchased from Sigma-Aldrich. sgRNA constructs in the pLentiCRISPR-v2 vector expressing Cas9 against control, STING and GATA3 were purchased from GenScript.

Inhibitors

For all assays except synergy assays, tazemetosat was used at 5 μM (S7128, Selleckchem), ipatasertib was used at 5 μM (S2808, Selleckchem), BYL719 was used at 5 μM (S2814, Selleckchem), GDC0077 was used at 5 μM (S8668, Selleckchem), GDC0941 was used at 3 μM (S1065, Selleckchem), MAK683 was used at 5 μM (S8983, Selleckchem) and itacitinib (JAK1i) was used at 500 nM (S7812, Selleckchem). TBK1 inhibitor GSK8612 was used at 16 μM (S8872, Selleckchem). Anti-IL-6R antibody is tocilizumab and was supplemented at 50 μg ml⁻¹ (A2012-5MG, Selleckchem). STING agonist ADU-S100 (MIW815) was supplemented at 50 μM (CT-ADUS100, Chemietek). Docetaxel was supplemented at 2 nM or 10 nM (S1148-10MM/1ML, Selleckchem).

ELISA

Human IL-6 (Thermo Fisher Scientific, KHC0061) and 2′3′-cGAMP (Cayman Chemical, 501700) were detected with ELISAs according to the manufacturer’s instructions. Samples for IL-6 (conditioned media) or 2′3′-cGAMP (protein lysates) were collected 24 h after addition of AKTi and/or TBK1i. Values represent the average of three or four biological replicates normalized to the DMSO sample.

GATA3 expression plasmid

GATA3 ORF was purchased from Addgene (plasmid 81902) in pDONR221 vector⁵⁹ and cloned into pHAGE-puro expression vector (Addgene plasmid 118692) using LR clonase II (11791020; Thermo Fisher Scientific) according to the manufacturer’s specifications. GATA3 overexpression was enforced in SUM159PT cells by lentiviral transduction and puromycin selection (1 μg ml⁻¹) for 72 h.

Western blotting

Protein was collected after 24 h of vehicle, single agent or combination treatment. Cells were washed with PBS (806544-500ML, Sigma-Aldrich) and then boiling 1% SDS lysis buffer (1% SDS (15553-035, Invitrogen), 10 mM Tris-HCl pH 7.5 (77-86-1, Sigma-Aldrich), 100 mM NaCl (S5586, Sigma-Aldrich)) was directly added to plates. Plates were scraped and protein lysate was sheared using 20 G needle 5–6 times. Proteins were boiled at 95 °C for 10 min and centrifuged at maximum speed for 3 min. The protein concentration was determined using the BCA quantification method (23222, Bio-Rad). Proteins were run in SDS–PAGE gels (4561084, Bio-Rad) and transferred to Immobilon PVDF membranes (IPVH00010, Sigma-Aldrich). Membranes were blocked with 5% milk in TBST for 1 h before incubation with primary antibodies overnight. The membranes were incubated with HRP-conjugated secondary antibody corresponding to the species of the primary antibody. HRP signal was measured using autoradiography film or by BioRad Chemidoc. Antibodies for western blotting were as follows: H3K27me3 (9733S, Cell Signaling Technologies, 1:1,000), H3 (4499S, Cell Signaling Technologies, 1:1,000), PRAS40 (2610S, Cell Signaling Technologies, 1:1,000), pPRAS40 (2997S, Cell Signaling Technologies, 1:1,000), GAPDH (2118S, Cell Signaling Technologies, 1:1,000), GATA3 (558686, BD BioSciences, 1:250), pSTAT3 (sc-8059, Santa Cruz, 1:500), STAT3 (9139, Cell Signaling Technologies, 1:1,000), JAK1 (3332S, Cell Signaling Technologies, 1:1,000), P27 (sc-528, Santa Cruz, 1:1,000), FOXO1 (9454, Cell Signaling Technologies, 1:1,000), FOXO1-pS265 (9461, Cell Signaling Technologies, 1:500), TBK1 (3504S, Cell Signaling Technologies, 1:1,000), TBK1-pS172 (5483S, Cell Signaling Technologies, 1:1,000), STING (13647S, Cell Signaling Technologies, 1:1,000), anti-rabbit secondary (111-035-144, Jackson ImmunoResearch, 1:5,000), anti-mouse secondary (115-035-166, Jackson ImmunoResearch, 1:5,000). All uncropped images of immunoblots are provided in Supplementary Data 1.

STING–TBK1 co-IP

Cells were treated with the indicated drugs and then collected after 8 h of treatment by washing with PBS and scraping. Pelleted cells were lysed using EBC lysis buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 0.5% NP-40, 1:10,000 β mercaptoethanol, 0.5 mM EDTA) supplemented with protease inhibitors (Sigma-Aldrich, 11836153001) and phosphatase inhibitors (Sigma-Aldrich, 4906837001). Cell lysates were incubated on ice for 20 min and then cleared by centrifugation at 16,000g for 20 min. Protein extracts were quantified using BCA reagent and then equal quantities were incubated with STING (Cell Signaling, 13647S) or IgG (Cell Signaling, 2729S) antibody and magnetic protein A/G beads (Pierce) overnight with end-over-end agitation at 4 °C. Beads were washed five times with EBC lysis buffer before immunoprecipitated proteins were eluted using SDS sample buffer and analysed by western blotting (described above).

RNA-seq

RNA was isolated after 8 or 24 h of treatment using the RNeasy Plus kit (74134, Qiagen). RNA was sequenced at the Dana-Farber Cancer Institute Molecular Biology Core Facility (SUM149PT, MDA-MB-468) or Novogene (MDA-MB-468 docetaxel experiment, HCC38, HCC1395, HCC1937) using the Illumina NextSeq 500 system. Raw data were mapped to the Hg19 (SUM149PT, MDA-MB-468) or hg38 (MDA-MB-468 docetaxel experiment, HCC38, HCC1395, HCC1937) genome using STAR and count files were made using HTSeq^60,61. DESeq2 was used to normalize counts (mean-ratio method), calculate total reads and determine differentially expressed genes⁶². Volcano plots were generated using EnhancedVolcano and heat maps were generated using pheatmap^63,64. DESeq2 was used to determine differentially expressed genes for public RNA-seq raw counts collected from CCLE database. Volcano plots were generated using the ggplot2 package in R and heat maps were generated using ComplexHeatmap package in R.

ATAC–seq

ATAC–seq data were sequenced using the Illumina NovaSeq 6000 system. Using the ENCODE pipeline (https://www.encodeproject.org/atac-seq/), raw sequencing FASTQ data were read and aligned to the hg38 reference genome. MACS2 was used to identify peaks indicating open chromatin region. We filtered out non-reproducible peaks by selecting peaks that were detected in replicates of each condition for each cell line, then peaks were merged for overlapping region across all conditions and samples using the GenomicRanges package in R. The consensus peaks were annotated using ChIPseeker package in R. Using featureCounts in Rsubread package in R, we counted all reads over the consensus peak regions, then performed differential accessible peak analysis using DESeq2. Motif discovery and analysis was performed using RGT-HINT. Transcription-factor-binding sites were identified with JASPAR motifs. Transcription factor activity scores were estimated based on the transcription factor footprint profiles. We ranked genes based on chromatin accessibility differences in promotor. If multiple peaks were annotated to single gene, we estimate the major changes by taking maximum/minimum log-transformed fold changes if <80% of peaks are more open/closed. The ranked genes were used for GSEA using the fgsea package in R. Gene sets were collected from the entire MSigDB using the msigdbr package in R.

Machine learning model

For the initial training set, TPMs of RNA-seq data for 17 cell lines (10 sensitive, 7 resistant) were collected from the CCLE dataset and standardized using z-score normalization. Feature selection is important for machine learning training as too many irrelevant features can interrupt model training (for example, overfitting). We therefore selected subset of genes by ranking them in terms of its variance or log-transformed fold change. First, we selected genes that were differently expressed between sensitive and resistant cell lines with FDR <0.05. We then ranked the genes based on their variance and selected the top 50, 100 and 200 highly variable genes (represented as top 50, top 100, and top 200, respectively). We also selected genes based on a cut-offs for the log-transformed fold change in sensitive versus resistant of 2 and 5, that is, |LFC| > 2 and |LFC| > 5. The models were trained using two different algorithms, RF and SVM, on the training sets with the subset of genes, then evaluated by its accuracy. We validated the models using leave-one-out cross validation. During the validation, the dataset is divided into N subsets where N is the total number of cell lines. Each subset includes N − 1 cell lines by removing one cell line as the validation set. For each subset, we performed differential expression analysis, ranked differentially expressed genes, defined a subset of genes based on its rank, trained machine learning models on the subset then tested the models using the validation cell line that was not used in the entire process at all. This was repeated N times by removing one cell line each time. In total, we have 2 × 5 × N models as we use two algorithms (RF, SVM) and five gene sets (top 50, top 100, top 200, |LFC| > 2 and |LFC| > 5). Cross validation accuracy is calculated by the number of correctly predicted unused cell line/N. On the basis of the validation and test accuracy, we selected the best model (RF, |LFC | > 5), then retrained the model using 17 cell lines. We used the final model to predict the sensitivity of TNBC tumours in the TCGA database. Similarly, TPMs of RNA-seq data were obtained then scaled using z-score normalization. Feature selection, including differential expressions and rank genes, were implemented in R. All of the machine learning methods, including training, validation, and testing, were implemented using the scikit-learn library in Python.

Copy-number analysis

To measure copy number alterations in TP53^−/− organoid allograft tumours, we completed ultra-low-pass whole-genome sequencing with the Dana-Farber Cancer Institute Molecular Biology Core Facility. Genomic DNA was isolated from a tumour from an untreated tumour-bearing mouse, fragmented and sequenced. Approximately 5 million library reads were analysed for copy variation at 100 kb resolution. We used CNVkit pipeline to process the germline sequencing raw data. The scatter plot was generated using the CNVkit library in Python.

ssGSEA and GSEA

ssGSEA was performed using GenePattern (https://www.genepattern.org/)^65,66. GSEA was performed using software available online (http://www.gsea-msigdb.org/)^65,67. Gene sets are published on the MSigDB for hallmark apoptosis (M5902, HALLMARK_APOPTOSIS)⁶⁸, mammary stem cells (M2573, LIM_MAMMARY_STEM_CELL_UP)²⁶ and mature luminal mammary cells (M2578: LIM_MAMMARY_LUMINAL_MATURE_UP)²⁶. Mammary involution gene set was derived from a previous study that identified genes upregulated in involution that were dependent on JAK1 expression in the mammary gland³⁷.

TNBCtype analysis

Molecular subtyping of samples using RNA-seq values was performed using the TNBCtype tool^27,69. Correlation coefficient with most significantly altered subtypes was reported for each cell line.

BioRender

The model presented in Fig. 6i was created using BioRender (license number IY273VQW9J).

ChIP–qPCR

MDA-MB-468 cells were plated at 1.2 × 10⁶ in 15 cm plates and treated the same as for cell counting and immunoblotting. After 24 h of vehicle, single agent or combination treatment, cells were collected in native medium. Cells were cross-linked first with 2 mM DSG (20593, Pierce) for 45 min at room temperature followed by 1% formaldehyde for 10 min at 37 °C before quenching with 0.125 M glycine for 5 min. Chromatin was isolated by first incubating with buffer 1 (50 mM HEPES-KOH pH 7.5 (15630-080, Thermo Fisher Scientific), 140 mM NaCl (S5586, Sigma-Aldrich), 1 mM EDTA pH 8 (15575-038, Invitrogen), 10% glycerol (G5516, Sigma-Aldrich), 0.0033% NP-40 (493015, Calbiochem) and 0.25% Triton X-100 (T8787, Sigma-Aldrich)), then resuspending the pellet in buffer 2 (10 mM Tris-HCl pH 8 (15567-025, Invitrogen), 200 mM NaCl, 1 mM EDTA pH 8 and 0.5 mM EGTA pH 8), and finally resuspending the cell pellet with buffer 3 (10 mM Tris-HCl pH 8, 100 mM NaCl, 1 mM EDTA pH 8, 0.5 mM EGTA pH 8, 0.1% sodium deoxycolate (D6750, Sigma-Aldrich) and 0.5% N-laurosarcosine (L7414, Sigma-Aldrich)). Isolated chromatin was then sonicated for 30 cycles of 30 s on, 30 s off using a Diagenode BioRuptor Plus sonicator on ‘high’. Soluble chromatin was immunoprecipitated with FOXO1 (2880, Cell Signaling Technologies), IgG (2729S, Cell Signaling Technologies) or STAT3 (9139, Cell Signaling Technologies) antibodies overnight. Immunoprecipitated chromatin was captured with magnetic protein A/G beads (88803, Pierce). Immunoprecipitated chromatin was washed with low-salt buffer three times (0.1% SDS, (15553-035, Invitrogen), 1% Triton X-100, 2 mM EDTA pH 8, 20 mM Tris-HCl pH 8 and 150 mM NaCl), high-salt buffer three times (0.1% SDS, 1% Triton X-100, 2 mM EDTA pH 8, 20 mM Tris-HC-l pH 8, 500 mM NaCl), LiCl buffer three times (0.25 M LiCl (62476, Sigma-Aldrich), 1% NP-40, 1 mM EDTA pH 8, 10 mM Tris-HCl pH 8, 1% sodium deoxycholate) and TE buffer (10 mM Tris-HCl pH 8, 10 mM EDTA pH 8) before elution in elution buffer (50 mM Tris-HCl pH 8, 10 mM EDTA, 1% SDS). Eluted chromatin was purified using the QIAquick PCR purification kit (28106, Qiagen). Relative abundances of BMF and GATA3 promoters in immunoprecipitated chromatin was measured using qPCR using PerfeCTa SYBR Green SuperMix Reaction Mixes (95054-500, QuantaBio) and primers listed below. The percentage input values were calculated for each immunoprecipitation. Fold change enrichment values were calculated by normalizing to the percentage input of the appropriate IgG control for each immunoprecipitation.

HA knock-in BMF

An N-terminal HA-tag was introduced into the endogenous locus of BMF as previously described⁷⁰. The crRNA (5′-TTGCCCCCTCACAGGAGAGA-3′) was hybridized with Alt-R CRISPR-Cas9 tracRNA (IDT, 1073190) at 95 °C for 5 min, at an equimolar ratio (0.375 nmol). The mix was cooled down to room temperature on the benchtop for 10 min before adding 2 µl of Alt-R S.p. Cas9 Nuclease V3 (IDT, 1081059) and 5 µl of 100 µM single-stranded donor oligonucleotide (5′-GCTGAGGGGGCAGTCCAGTAGGCTCTGGGCAAACAGGTCAGCAGAGAGCAAGCTCCCGGGTTGGGTCACCGGCTCCCCATCCTCTGGTTGGAACACATCATCCTCCAGCTCCTCCACACACTGAGATGGCTCAGCGTAATCTGGTACGTCGTATGGGTACATCTCTCCTGTGAGGGGGCAACGCAGGCATCTGGGCTGCT-3′). The mix was incubated for 20 min at room temperature and added to 1.5 million (MDA-MB-468) or 1 million (HCC1395) cells resuspended in 100 µl of SF Cell Line Nucleofector solution (Lonza, V4XC-2012). Cells were then transferred to a cuvette and nucleofected using the EO-117 program in a 4D-Nucleofector X Unit (Lonza).

CUT&RUN

CUT& RUN was performed as previously described⁷¹. In brief, 500,000 nuclei from MDA-MB-468 cells treated with DMSO, 5 μM ipatasertib, 5 μM tazemetostat or a combination of both drugs were isolated using nuclear extraction buffer (20 mM HEPES pH 7.9, 10 mM KCl, 0.1% Triton X-100, 20% glycerol and 1 mM MnCl₂). Nucleus samples were then immobilized to Biomag Plus Concanavalin A (Con A)-coated magnetic beads (Bangs laboratories) that were activated by washing three times with cold bead activation buffer (20 mM HEPES pH 7.9, 10 mM KCl, 1 mM CaCl₂, 1 mM MnCl₂). ConA bead/cell mixtures were resuspended in cold antibody buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM spermidine, 1× Roche cOmplete, Mini, EDTA-free protease inhibitor, 0.01% digitonin, 2 mM EDTA), then incubated with 0.5 μg primary antibodies: H3K27me3 (CST, 9733) and H3K4me3 (CST, 9733) or IgG (Epicypher) overnight in a 4 °C cold room. Unbound antibodies were washed three times each with cold digitonin buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM spermidine, 1× Roche cOmplete, Mini, EDTA-free protease inhibitor, 0.01% digitonin). ConA bead/cell mixtures were then resuspended in 50 μl cold digitonin buffer and incubated with our homemade pAG-MNase in a 4 °C cold room for an hour on a nutator. Unbound pAG-MNase was washed three times with cold digitonin Buffer. MNase was activated by addition of CaCl₂ and incubated in a 4 °C cold room for 30 min on a nutator to cleave and release antibody-bound chromatin. The reaction was stopped by adding cold stop buffer (340 mM NaCl, 20 mM EDTA, 4 mM EGTA, 50 μg ml⁻¹ RNase A, 50 μg ml⁻¹ glycogen, 1 pg μl⁻¹ E. coli spike-in DNA). Cleaved chromatin was then released by incubating at 37 °C for 10 min. CUT&RUN enriched DNA in the supernatant can be collected using magnetic beads and purified using the Monarch PCR and DNA cleanup kit (NEB T1030L). CUT&RUN libraries were prepared with 10 ng CUT&RUN DNA using NEBNext Ultra II DNA library prep kit (NEB), according to manufacturer’s protocol. Libraries were sequenced on the Illumina NextSeq 2000 system, 2 × 50 bp paired-end reads.

Paired-end fastq files were aligned to hg38 reference genome using Bowtie2 with the settings ‘--very-sensitive --no-mixed --no-discordant --phred33 -I 10 -X 700’⁷². Sequencing reads were also aligned to the E. coli genome to map spike-in reads. For spike-in normalization, the total number of mapped reads to E. coli genome was used to calculate normalization factor for CUT&RUN samples. SAM files were converted to BAM files using samtools⁷³. Bigwig files were generated from BAM files using Deeptools⁷⁴. Genome browser tracks of big files were generated using the Integrative Genomics Viewer (IGV)⁷⁵. Peaks were called using MACS2 using callpeak function with ‘-f BAME -keep-dup 1 -q 0.05’ and IgG was used as a control⁷⁶. DeSeq2 analysis from DiffBind R package with option ‘spikein=True’ was used to compare differential binding between conditions (FDR < 0.1)⁷⁷.

RT–qPCR

RNA was isolated using the RNeasy Plus kit (74134, Qiagen). RNA was reverse transcribed into cDNA using the High-Capacity cDNA Reverse Transcription Kit (4368814, Thermo Fisher Scientific). qPCR was completed using PerfeCTa SYBR Green SuperMix Reaction Mixes (95054-500, QuantaBio) and the primers listed below. C_q values were normalized to STAU1 as a reference gene.

CyCIF analysis

Human breast cancer tissue array BRC482 was purchased from Pantomics and consists of 16 cases, with one normal breast tissue core paired with two tumour tissue cores from each patient. FFPE sections of PDX samples and a tissue microarray were prepared and stained with a 23-plex and 33-plex antibody panel, respectively, using tissue-based cyclic immunofluorescence (CyCIF) as previously outlined. Baking and dewaxing: to prepare the samples for antibody staining, tissue sections on glass slides underwent an automated process facilitated by the Leica Bond RX machine. The protocol started with a 30 min baking step at 60 °C. Subsequently, the slides were subjected to dewaxing at 72 °C in BOND Dewax Solution, followed by an antigen-retrieval step conducted at 100 °C for 20 min using BOND Epitope Retrieval Solution 1 (ER1). Photo-bleaching and autofluorescence reduction: after the baking and dewaxing steps, the slides were immersed in a bleaching solution (4.5% H₂O₂, 20 mM NaOH in 1× PBS) and exposed to LED light for 1 h to mitigate autofluorescence. Excess bleaching solution was removed by washing the slides six times with 1× PBS (10–15 s each). Subsequently, the slides were incubated overnight at 4 °C in darkness with secondary antibodies diluted in SuperBlock Blocking Buffer (1:1,000; Thermo Fisher Scientific, 37515). After this incubation, the slides were washed six times with 1× PBS and were photobleached again for 1 h. Antibody staining, coverslip mounting and imaging: for each CyCIF cycle, the samples were incubated overnight at 4 °C in the absence of light with Hoechst 33342 (1:10,000; Thermo Fisher Scientific, 62249) for nuclear staining. Simultaneously, primary antibodies were introduced, which were either conjugated or unconjugated, diluted in SuperBlock Blocking Buffer (Thermo Fisher Scientific, 37515). In cases in which primary antibodies were unconjugated, subsequent incubation with secondary antibodies was performed at room temperature for 1 h in the dark. After staining, the slides underwent a series of six 10 s washes with 1× PBS, and then they were mounted with 24 × 60 mm coverslips, using 80–100 µl of 50% glycerol, and left to dry at room temperature for 30 min. Once coverslipped, automatic imaging was conducted using the RareCyte Cytefinder II HT system with the following channels: UV, cy3, cy5 and cy7. Imaging parameters remained consistent as follows: binning: 1 × 1 for PDX whole-slide images; binning 2 × 2 for the TMA; objective: ×20; numerical aperture: 0.75; resolution: 0.325 mm px⁻¹. Image exposures were fine-tuned for each channel to prevent signal saturation but were maintained uniformly across samples. Between CyCIF cycles, the coverslips were demounted by immersing slides in containers of 1× PBS (5 slides per container) for 10 min at room temperature. Before initiating the subsequent antibody staining cycle, slides were photobleached for 1 h to deactivate the fluorophores and washed again in PBS six times to remove residual bleaching solution. The acquired images were processed using the Docker-based NextFlow pipeline MCMICRO. Initially, raw images from the Cytefinder II HT underwent illumination correction via the BaSic module. Subsequently, they were stitched and registered using the ASHLAR module, resulting in the generation of assembled OME.TIFF files for each slide. Full code is available at GitHub (https://github.com/labsyspharm/mcmicro).

Antibodies were as follows: CK14 (Abcam, ab77684, LL002, conjugated to FITC, 1:500), CK8 (eBioscience, 11-9938-80, LP3K, conjugated to FITC, 1:400), pan-CK (eBioscience, 53-9003-82, AE1/AE3, conjugated to AF488, 1:1,000), EZH2 (CST, 45638, D2C9, conjugated to AF647, 1:100), cPARP (CST, 6987S, D64E10, conjugated to AF647, 1:100), E-cadherin (CST, 9835S, 24E10, conjugated to AF647, 1:600), anti-rabbit IgG (Invitrogen A-31572, conjugated to AF555, 1:1,000) and anti-mouse IgG (Invitrogen, A-31571, conjugated to AF647, 1:1,000).

ERV expression

To count reads for ERVs, ERV locus information was collected from the ERVmap database⁷⁸. Salmon (v.0.14.1)⁷⁹ was used to create a customized reference transcriptome set that includes human genome transcriptome (hg38) and ERVmap database. Using the database, RNA-seq data of MDA-MB-468 cells treated with DMSO or EZH2i tazemetostat were quantified by Salmon. DESeq2 was used for counts normalization and differential expression analysis.

Oligonucleotides

Primers for RT–qPCR were as follows in 5′–3′ orientation: STAU1 (fwd: GGATGAGTTCAGGATGCCTTAT, rev: GGTGTGATGTCCTTGACTAACT), BMF (fwd: ACTTCAGCTCTTCCCTCTCA, rev: GAGTCTGGGTAGCTTTGTCTTC), GATA3 (fwd: CTCATTAAGCCCAAGCGAAG, rev: GTCTGACAGTTCGCACAGGA), FOS (fwd: GTCTTCCTTCGTCTTCACCTAC, rev: GAGTCAGAGGAAGGCTCATTG), CYBRD1 (fwd: AGATCCTGCATACAGTACATTCC, rev: CATTGCGGTCTGGTGACTAT), PDK4 (fwd: GCCTTCCCTTACACCAATAGAG, rev: GTTGGTGCAGTGGAGTATGT), SCNN1A (fwd: GGCTGTGCCTACATCTTCTATC, rev: GAGAAGTCAACCTGGAGCTTATAG), STAT5A (fwd: GCCACCATCACGGACATTAT, rev: CAAACTTGGTCTGGGTCTTCA), IGFBP5 (fwd: GAAGAAGGACCGCAGAAAGAA, rev: CTCAGACTCCTGTCTCATCTCA), TNFRSF21 (fwd: CCAACTCTTCTGCCTCTGTTAG, rev: GAGGGTCTTGTTCACGTCTTC), GDA (fwd: CGCACACTGTCCCAATTCTA, rev: TCTGTACCCAGCCCTATCTT). Primers for ChIP–qPCR were as follows in 5′–3′ orientation: GATA3 promoter (fwd: TTGGTGCTCGCGATTGAA, rev: AAATGCTGACTTCTGAGGCTAA), GATA3 enhancer (fwd: ATCCATCAGCCCTTCTTTCTG, rev: GCGCCATCTACTGGGTTATT).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

Data generated in this study: RNA-seq data of TNBC cell lines treated with EZH2i and AKTi have been deposited at the Gene Expression Omnibus for MDA-MB-468 (GSE205729), SUM149PT (GSE205730), and HCC38, HCC1937 and HCC1395 (GSE251708). RNA-seq data for MDA-MB-468 cells treated with docetaxel or EZH2/AKTi have been deposited under GSE250632. CUT&RUN data are available under PRJNA1054805 and ATAC–seq data have been deposited under GSE252382. CyCIF images of PDX004 tumours are available online (https://s3.amazonaws.com/www.cycif.org/cichowski-santagata-sorger-2024/index.html). Publicly available cancer datasets: to determine EZH2 mRNA levels in TNBC tumours and matched normal breast tissue, data were extracted from the TCGA Firehose Legacy dataset⁸⁰. TNBC tumours were identified from the dataset if they were negative for ER, PR and HER2 in the clinical annotation. RNA-seq read counts were collected from the CCLE DepMap portal (DepMap Public 21Q4). For machine learning, RNA-seq raw counts and corresponding clinical data were also obtained from of BRCA TNBC samples in the TCGA using TCGAbiolinks package in R⁸⁰. Mutation and copy-number variation and the mRNA levels for each TCGA tumour were collected from the Firehose Legacy dataset. Hi-C data from MCF10A (ENCFF977XWK) and MCF7 (ENCFF420JTA) cells were extracted from ENCODE and visualized in the IGV. FOXO1 ChIP–seq data were extracted from ENCODE (ENCSR321OAA) and visualized in the IGV.

References

Waks, A. G. & Winer, E. P. Breast cancer treatment: a review. JAMA 321, 288–300 (2019).
Article CAS PubMed Google Scholar
Garrido-Castro, A. C., Lin, N. U. & Polyak, K. Insights into molecular classifications of triple-negative breast cancer: improving patient selection for treatment. Cancer Discov. 9, 176–198 (2019).
Article CAS PubMed PubMed Central Google Scholar
Nik-Zainal, S. et al. Landscape of somatic mutations in 560 breast cancer whole-genome sequences. Nature 534, 47–54 (2016).
Article CAS PubMed PubMed Central ADS Google Scholar
Stephens, P. J. et al. The landscape of cancer genes and mutational processes in breast cancer. Nature 486, 400–404 (2012).
Article CAS PubMed PubMed Central Google Scholar
Kim, M. et al. A protein interaction landscape of breast cancer. Science 374, eabf3066 (2021).
Article CAS PubMed PubMed Central Google Scholar
Khoury, K. et al. Prevalence of phosphatidylinositol-3-kinase (PI3K) pathway alterations and co-alteration of other molecular markers in breast cancer. Front. Oncol. 10, 1475 (2020).
Article PubMed PubMed Central Google Scholar
Turner, N. et al. Ipatasertib plus paclitaxel for PIK3CA/AKT1/PTEN-altered hormone receptor-positive HER2-negative advanced breast cancer: primary results from cohort B of the IPATunity130 randomized phase 3 trial. Breast Cancer Res. Treat. 191, 565–576 (2022).
Saal, L. H. et al. PIK3CA mutations correlate with hormone receptors, node metastasis, and ERBB2, and are mutually exclusive with PTEN loss in human breast carcinoma. Cancer Res. 65, 2554–2559 (2005).
Article CAS PubMed Google Scholar
Stemke-Hale, K. et al. An integrative genomic and proteomic analysis of PIK3CA, PTEN, and AKT mutations in breast cancer. Cancer Res. 68, 6084–6091 (2008).
Article CAS PubMed PubMed Central Google Scholar
André, F. et al. Alpelisib for PIK3CA-mutated, hormone receptor-positive advanced breast cancer. N. Engl. J. Med. 380, 1929–1940 (2019).
Article PubMed Google Scholar
André, F. et al. Alpelisib plus fulvestrant for PIK3CA-mutated, hormone receptor-positive, human epidermal growth factor receptor-2-negative advanced breast cancer: final overall survival results from SOLAR-1. Ann. Oncol. 32, 208–217 (2021).
Article PubMed Google Scholar
Schmid, P. et al. A phase III trial of capivasertib and paclitaxel in first-line treatment of patients with metastatic triple-negative breast cancer (CAPItello290). J. Clin. Oncol. 38, TPS1109 (2020).
Article Google Scholar
Margueron, R. & Reinberg, D. The Polycomb complex PRC2 and its mark in life. Nature 469, 343–349 (2011).
Article CAS PubMed PubMed Central ADS Google Scholar
Comet, I., Riising, E. M., Leblanc, B. & Helin, K. Maintaining cell identity: PRC2-mediated regulation of transcription and cancer. Nat. Rev. Cancer 16, 803–810 (2016).
Article CAS PubMed Google Scholar
Laugesen, A., Højfeldt, J. W. & Helin, K. Molecular mechanisms directing PRC2 recruitment and H3K27 methylation. Mol. Cell 74, 8–18 (2019).
Article CAS PubMed PubMed Central Google Scholar
Kleer, C. G. et al. EZH2 is a marker of aggressive breast cancer and promotes neoplastic transformation of breast epithelial cells. Proc. Natl Acad. Sci. USA 100, 11606–11611 (2003).
Article CAS PubMed PubMed Central ADS Google Scholar
Pal, B. et al. Global changes in the mammary epigenome are induced by hormonal cues and coordinated by Ezh2. Cell Rep. 3, 411–426 (2013).
Article CAS PubMed Google Scholar
Michalak, E. M. et al. Polycomb group gene Ezh2 regulates mammary gland morphogenesis and maintains the luminal progenitor pool. Stem Cells 31, 1910–1920 (2013).
Article CAS PubMed Google Scholar
Li, X. et al. Targeted overexpression of EZH2 in the mammary gland disrupts ductal morphogenesis and causes epithelial hyperplasia. Am. J. Pathol. 175, 1246–1254 (2009).
Article CAS PubMed PubMed Central Google Scholar
Yomtoubian, S. et al. Inhibition of EZH2 catalytic activity selectively targets a metastatic subpopulation in triple-negative breast cancer. Cell Rep. 30, 755–770 (2020).
Article CAS PubMed Google Scholar
Hoy, S. M. Tazemetostat: first approval. Drugs https://doi.org/10.1007/s40265-020-01288-x (2020).
Stacchiotti, S. et al. Safety and efficacy of tazemetostat, a first-in-class EZH2 inhibitor, in patients (pts) with epithelioid sarcoma (ES) (NCT02601950). J. Clin. Oncol. 37, 11003–11003 (2019).
Article Google Scholar
Wang, H. et al. Inadequate DNA damage repair promotes mammary transdifferentiation, leading to BRCA1 breast cancer. Cell 178, 135–151 (2019).
Article CAS PubMed PubMed Central Google Scholar
Tao, L., Xiang, D., Xie, Y., Bronson, R. T. & Li, Z. Induced p53 loss in mouse luminal cells causes clonal expansion and development of mammary tumours. Nat. Commun. 8, 14431 (2017).
Article CAS PubMed PubMed Central ADS Google Scholar
Guillen, K. P. et al. A human breast cancer-derived xenograft and organoid platform for drug discovery and precision oncology. Nat. Cancer 3, 232–250 (2022).
Article PubMed PubMed Central Google Scholar
Lim, E. et al. Transcriptome analyses of mouse and human mammary cell subpopulations reveal multiple conserved genes and pathways. Breast Cancer Res. 12, R21 (2010).
Article PubMed PubMed Central Google Scholar
Lehmann, B. D. et al. Identification of human triple-negative breast cancer subtypes and preclinical models for selection of targeted therapies. J. Clin. Invest. 121, 2750–2767 (2011).
Article CAS PubMed PubMed Central Google Scholar
Lehmann, B. D. et al. Refinement of triple-negative breast cancer molecular subtypes: implications for neoadjuvant chemotherapy selection. PLoS ONE 11, e0157368 (2016).
Article PubMed PubMed Central Google Scholar
Asselin-Labat, M.-L. et al. Gata-3 is an essential regulator of mammary-gland morphogenesis and luminal-cell differentiation. Nat. Cell Biol. 9, 201–209 (2007).
Article CAS PubMed Google Scholar
Kouros-Mehr, H., Slorach, E. M., Sternlicht, M. D. & Werb, Z. GATA-3 maintains the differentiation of the luminal cell fate in the mammary gland. Cell 127, 1041–1055 (2006).
Article CAS PubMed PubMed Central Google Scholar
Bracken, A. P. et al. The Polycomb group proteins bind throughout the INK4A-ARF locus and are disassociated in senescent cells. Genes Dev. 21, 525–530 (2007).
Article CAS PubMed PubMed Central Google Scholar
Ecker, K. & Hengst, L. Skp2: caught in the Akt. Nat. Cell Biol. 11, 377–379 (2009).
Article CAS PubMed Google Scholar
Ben-Porath, I. et al. An embryonic stem cell-like gene expression signature in poorly differentiated aggressive human tumors. Nat. Genet. 40, 499–507 (2008).
Article CAS PubMed PubMed Central Google Scholar
Raths, F. et al. The molecular consequences of androgen activity in the human breast. Cell Genom. 3, 100272 (2023).
Article CAS PubMed PubMed Central Google Scholar
Greer, E. L. & Brunet, A. FOXO transcription factors at the interface between longevity and tumor suppression. Oncogene 24, 7410–7425 (2005).
Article CAS PubMed Google Scholar
Ma, L. et al. Overcoming EZH2 inhibitor resistance by taxane in PTEN-mutated cancer. Theranostics 9, 5020–5034 (2019).
Article CAS PubMed PubMed Central Google Scholar
Sakamoto, K. et al. Janus kinase 1 is essential for inflammatory cytokine signaling and mammary gland remodeling. Mol. Cell. Biol. 36, 1673–1690 (2016).
Article CAS PubMed PubMed Central Google Scholar
Schmelzle, T. et al. Functional role and oncogene-regulated expression of the BH3-only factor Bmf in mammary epithelial anoikis and morphogenesis. Proc. Natl Acad. Sci. USA 104, 3787–3792 (2007).
Article CAS PubMed PubMed Central ADS Google Scholar
Schwertfeger, K. L., Richert, M. M. & Anderson, S. M. Mammary gland involution is delayed by activated Akt in transgenic mice. Mol. Endocrinol. 15, 867–881 (2001).
Article CAS PubMed Google Scholar
Ackler, S., Ahmad, S., Tobias, C., Johnson, M. D. & Glazer, R. I. Delayed mammary gland involution in MMTV-AKT1 transgenic mice. Oncogene 21, 198–206 (2002).
Article CAS PubMed Google Scholar
Schwertfeger, K. L., McManaman, J. L., Palmer, C. A., Neville, M. C. & Anderson, S. M. Expression of constitutively activated Akt in the mammary gland leads to excess lipid synthesis during pregnancy and lactation. J. Lipid Res. 44, 1100–1112 (2003).
Article CAS PubMed Google Scholar
Yoo, K. H. et al. Loss of EZH2 results in precocious mammary gland development and activation of STAT5-dependent genes. Nucleic Acids Res. 43, 8774–8789 (2015).
Article CAS PubMed PubMed Central Google Scholar
Hughes, K., Wickenden, J. A., Allen, J. E. & Watson, C. J. Conditional deletion of Stat3 in mammary epithelium impairs the acute phase response and modulates immune cell numbers during post‐lactational regression. J. Pathol. 227, 106–117 (2012).
Article CAS PubMed PubMed Central Google Scholar
Chapman, R. S. et al. Suppression of epithelial apoptosis and delayed mammary gland involution in mice with a conditional knockout of Stat3. Genes Dev. 13, 2604–2616 (1999).
Article CAS PubMed PubMed Central Google Scholar
Kitajima, S. et al. Suppression of STING associated with LKB1 loss in KRAS-driven lung cancer. Cancer Discov. 9, 34–45 (2019).
Article CAS PubMed Google Scholar
Mahadevan, N. R. et al. Intrinsic Immunogenicity of Small Cell Lung Carcinoma Revealed by Its Cellular Plasticity. Cancer Discov. 11, 1952–1969 (2021).
Article MathSciNet CAS PubMed PubMed Central Google Scholar
Morel, K. L. et al. EZH2 inhibition activates a dsRNA-STING-interferon stress axis that potentiates response to PD-1 checkpoint blockade in prostate cancer. Nat. Cancer 2, 444–456 (2021).
Article CAS PubMed PubMed Central Google Scholar
Hopfner, K.-P. & Hornung, V. Molecular mechanisms and cellular functions of cGAS-STING signalling. Nat. Rev. Mol. Cell Biol. 21, 501–521 (2020).
Article CAS PubMed Google Scholar
Yum, S., Li, M., Fang, Y. & Chen, Z. J. TBK1 recruitment to STING activates both IRF3 and NF-κB that mediate immune defense against tumors and viral infections. Proc. Natl Acad. Sci. USA 118, e2100225118 (2021).
Article CAS PubMed PubMed Central Google Scholar
Deblois, G. et al. Epigenetic switch-induced viral mimicry evasion in chemotherapy-resistant breast cancer. Cancer Discov. 10, 1312–1329 (2020).
Article CAS PubMed Google Scholar
Cañadas, I. et al. Tumor innate immunity primed by specific interferon-stimulated endogenous retroviruses. Nat. Med. 24, 1143–1150 (2018).
Article PubMed PubMed Central Google Scholar
Kim, S. J., Kiser, P. K., Asfaha, S., DeKoter, R. P. & Dick, F. A. EZH2 inhibition stimulates repetitive element expression and viral mimicry in resting splenic B cells. EMBO J. 42, e114462 (2023).
Article CAS PubMed PubMed Central Google Scholar
Emens, L. A. et al. First-line atezolizumab plus nab-paclitaxel for unresectable, locally advanced, or metastatic triple-negative breast cancer: IMpassion130 final overall survival analysis. Ann. Oncol. 32, 983–993 (2021).
Article CAS PubMed Google Scholar
Rugo, H. S. et al. LBA16 KEYNOTE-355: final results from a randomized, double-blind phase III study of first-line pembrolizumab + chemotherapy vs placebo + chemotherapy for metastatic TNBC. Ann. Oncol. 32, S1289–S1290 (2021).
Dent, R. et al. Final results of the double-blind placebo-controlled randomized phase 2 LOTUS trial of first-line ipatasertib plus paclitaxel for inoperable locally advanced/metastatic triple-negative breast cancer. Breast Cancer Res. Treat. 189, 377–386 (2021).
Article CAS PubMed Google Scholar
Ianevski, A., Giri, A. K. & Aittokallio, T. SynergyFinder 3.0: an interactive analysis and consensus interpretation of multi-drug synergies across multiple samples. Nucleic Acids Res. https://doi.org/10.1093/nar/gkac382 (2022).
DeRose, Y. S. et al. Patient-derived models of human breast cancer: protocols for in vitro and in vivo applications in tumor biology and translational medicine. Curr. Protoc. Pharmacol. 60, 14.23.1–14.23.43 (2013).
Google Scholar
Debnath, J., Muthuswamy, S. K. & Brugge, J. S. Morphogenesis and oncogenesis of MCF-10A mammary epithelial acini grown in three-dimensional basement membrane cultures. Methods 30, 256–268 (2003).
Article CAS PubMed Google Scholar
Kim, E. et al. Systematic functional interrogation of rare cancer variants identifies oncogenic alleles. Cancer Discov. 6, 714–726 (2016).
Article CAS PubMed PubMed Central Google Scholar
Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013).
Article CAS PubMed Google Scholar
Anders, S., Pyl, P. T. & Huber, W. HTSeq-a Python framework to work with high-throughput sequencing data. Bioinformatics 31, 166–169 (2015).
Article CAS PubMed Google Scholar
Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).
Article PubMed PubMed Central Google Scholar
Blighe, K. EnhancedVolcano. https://doi.org/10.18129/B9.bioc.EnhancedVolcano (2018).
Kolde, R. pheatmap (2019).
Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).
Article CAS PubMed PubMed Central ADS Google Scholar
Barbie, D. A. et al. Systematic RNA interference reveals that oncogenic KRAS-driven cancers require TBK1. Nature 462, 108–112 (2009).
Article CAS PubMed PubMed Central ADS Google Scholar
Mootha, V. K. et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 34, 267–273 (2003).
Article CAS PubMed Google Scholar
Liberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 1, 417–425 (2015).
Article CAS PubMed PubMed Central Google Scholar
Chen, X. et al. TNBCtype: a subtyping tool for triple-negative breast cancer. Cancer Inform. 11, CIN.S9983 (2012).
Article Google Scholar
Kurppa, K. J. et al. Treatment-induced tumor dormancy through YAP-mediated transcriptional reprogramming of the apoptotic pathway. Cancer Cell 37, 104–122 (2020).
Article CAS PubMed PubMed Central Google Scholar
Skene, P. J. & Henikoff, S. An efficient targeted nuclease strategy for high-resolution mapping of DNA binding sites. eLife 6, e21856 (2017).
Article PubMed PubMed Central Google Scholar
Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359 (2012).
Article CAS PubMed PubMed Central Google Scholar
Langmead, B. Aligning short sequencing reads with Bowtie. Curr. Protoc. Bioinform. 32, 11.7.1–11.7.14 (2010).
Article Google Scholar
Ramírez, F., Dündar, F., Diehl, S., Grüning, B. A. & Manke, T. deepTools: a flexible platform for exploring deep-sequencing data. Nucleic Acids Res. 42, W187–W191 (2014).
Article PubMed PubMed Central Google Scholar
Thorvaldsdóttir, H., Robinson, J. T. & Mesirov, J. P. Integrative Genomics Viewer (IGV): high-performance genomics data visualization and exploration. Brief. Bioinform. 14, 178–192 (2013).
Article PubMed Google Scholar
Zhang, Y. et al. Model-based Analysis of ChIP-Seq (MACS). Genome Biol. 9, R137 (2008).
Article PubMed PubMed Central Google Scholar
Stark, R. & Brown, G. DiffBind. https://doi.org/10.18129/B9.BIOC.DIFFBIND (Bioconductor, 2017).
Tokuyama, M. et al. ERVmap analysis reveals genome-wide transcription of human endogenous retroviruses. Proc. Natl Acad. Sci. USA 115, 12565–12572 (2018).
Article CAS PubMed PubMed Central ADS Google Scholar
Patro, R., Duggal, G., Love, M. I., Irizarry, R. A. & Kingsford, C. Salmon provides fast and bias-aware quantification of transcript expression. Nat. Methods 14, 417–419 (2017).
Article CAS PubMed PubMed Central Google Scholar
Colaprico, A. et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. Nucleic Acids Res. 44, e71 (2016).
Article PubMed Google Scholar

Download references

Acknowledgements

We thank A. Welm for use of TNBC PDXs HCI-025 and HCI-004; the staff at the Dana-Farber Cancer Institute Molecular Biology Core Facilities for assistance with RNA-seq; and the Dana-Farber Cancer Institute Center for Biomedical Imaging in Oncology for the study involving the TNBC PDX model. This work was supported by a grant from the Cancer Research UK Grand Challenge and the Mark Foundation for Cancer Research to the SPECIFICANCER team (K.C.) and a DOD BC201085P1 Transformative Breast Cancer Consortium Award (K.C.). A.E.S. was supported by an American Cancer Society Postdoctoral Fellowship (PF-22-040-01-ET), a Terri Brodeur Breast Cancer Foundation Grant and a METAvivor Early Career Investigator Award; N.P. by AACR-AstraZeneca Breast Cancer Research Fellowship (20-40-12-PERU) and a Terri Brodeur Breast Cancer Foundation Grant; K.M. by NHGRI F32 1F32HG012318-01; S.S. by National Cancer Institute grant U54-CA225088 and the BWH President’s Scholar Award; Z.L. by NIH grant R01 CA222560;and S.M. by Fondazione Gianni Bonadonna and Associazione Italiana per la Ricerca contro il Cancro 2022-2024.

Author information

Authors and Affiliations

Division of Genetics, Brigham and Women’s Hospital, Boston, MA, USA
Amy E. Schade, Naiara Perurena, Carrie L. Rodriguez, Anjana Krishnan, Alycia Gardner, Patrick Loi, G. M. Mastellone, Natalie F. Pilla, Marina Watanabe, Rachel A. Davis, Kaia Mattioli, Dongxi Xiang, Zhe Li & Karen Cichowski
Department of Medicine, Harvard Medical School, Boston, MA, USA
Amy E. Schade, Naiara Perurena, Carrie L. Rodriguez, Anjana Krishnan, Alycia Gardner, Patrick Loi, G. M. Mastellone, Natalie F. Pilla, Marina Watanabe, Rachel A. Davis, Kaia Mattioli, Dongxi Xiang, Stefania Morganti, Ana C. Garrido-Castro, Sara M. Tolaney, Zhe Li & Karen Cichowski
Cedars-Sinai Samuel Oschin Comprehensive Cancer Institute, Los Angeles, CA, USA
Yoona Yang & Simon R. V. Knott
Department of Biomedical Sciences, Cedars-Sinai Medical Center, Los Angeles, CA, USA
Yoona Yang & Simon R. V. Knott
Ludwig Center at Harvard, Harvard Medical School, Boston, MA, USA
Yilin Xu, Jason J. Zoeller, Jia-Ren Lin, Peter K. Sorger, Sandro Santagata & Karen Cichowski
Laboratory of Systems Pharmacology, Harvard Medical School, Boston, MA, USA
Yilin Xu, Jia-Ren Lin, Peter K. Sorger & Sandro Santagata
Department of Pathology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA
Yilin Xu, Jia-Ren Lin & Sandro Santagata
Division of Cancer Biology, Institute of Cancer Research, London, UK
Van T. M. Nguyen & Kristian Helin
Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA, USA
Keiichi Ota, Stefania Morganti, Ana C. Garrido-Castro, Sara M. Tolaney & David A. Barbie
Department of Cell Biology, Harvard Medical School, Boston, MA, USA
Jason J. Zoeller
Department of Systems Biology, Harvard Medical School, Boston, MA, USA
Peter K. Sorger & Sandro Santagata

Authors

Amy E. Schade
View author publications
Search author on:PubMed Google Scholar
Naiara Perurena
View author publications
Search author on:PubMed Google Scholar
Yoona Yang
View author publications
Search author on:PubMed Google Scholar
Carrie L. Rodriguez
View author publications
Search author on:PubMed Google Scholar
Anjana Krishnan
View author publications
Search author on:PubMed Google Scholar
Alycia Gardner
View author publications
Search author on:PubMed Google Scholar
Patrick Loi
View author publications
Search author on:PubMed Google Scholar
Yilin Xu
View author publications
Search author on:PubMed Google Scholar
Van T. M. Nguyen
View author publications
Search author on:PubMed Google Scholar
G. M. Mastellone
View author publications
Search author on:PubMed Google Scholar
Natalie F. Pilla
View author publications
Search author on:PubMed Google Scholar
Marina Watanabe
View author publications
Search author on:PubMed Google Scholar
Keiichi Ota
View author publications
Search author on:PubMed Google Scholar
Rachel A. Davis
View author publications
Search author on:PubMed Google Scholar
Kaia Mattioli
View author publications
Search author on:PubMed Google Scholar
Dongxi Xiang
View author publications
Search author on:PubMed Google Scholar
Jason J. Zoeller
View author publications
Search author on:PubMed Google Scholar
Jia-Ren Lin
View author publications
Search author on:PubMed Google Scholar
Stefania Morganti
View author publications
Search author on:PubMed Google Scholar
Ana C. Garrido-Castro
View author publications
Search author on:PubMed Google Scholar
Sara M. Tolaney
View author publications
Search author on:PubMed Google Scholar
Zhe Li
View author publications
Search author on:PubMed Google Scholar
David A. Barbie
View author publications
Search author on:PubMed Google Scholar
Peter K. Sorger
View author publications
Search author on:PubMed Google Scholar
Kristian Helin
View author publications
Search author on:PubMed Google Scholar
Sandro Santagata
View author publications
Search author on:PubMed Google Scholar
Simon R. V. Knott
View author publications
Search author on:PubMed Google Scholar
Karen Cichowski
View author publications
Search author on:PubMed Google Scholar

Contributions

A.E.S. and K.C. conceived the study. A.E.S., K.C., N.P., Y.Y., V.T.M.N., J.-R.L., P.K.S., K.H., S.S. and S.R.V.K. designed the study. A.E.S., N.P., Y.Y., C.L.R., A.K., A.G., P.L., Y.X., V.T.M.N., G.M.M., N.F.P., M.W., K.O., R.A.D., D.X., J.J.Z. and J.-R.L. acquired the data and/or generated key resources. A.E.S., K.C., N.P., Y.Y., C.L.R., A.K., Y.X., K.M., K.O., J.-R.L., P.K.S., K.H., S.S. and S.R.V.K. analysed the data. A.E.S., K.C., N.P., Y.Y., Y.X., K.D., S.M., A.C.G.-C., S.M.T., Z.L., D.A.B., P.K.S., K.H., S.S. and S.R.V.K. interpreted the data. A.E.S., K.C., Y.Y. and S.R.V.K. drafted and edited the manuscript.

Corresponding author

Correspondence to Karen Cichowski.

Ethics declarations

Competing interests

K.C. is an advisor at Genentech and serves on the scientific advisory board of Erasca. S.M.T. has consulting or advisory roles at Novartis, Pfizer, Merck, Lilly, Nektar, NanoString Technologies, AstraZeneca, Puma Biotechnology, Genentech/Roche, Eisai, Sanofi, Bristol Myers Squibb, Seattle Genetics, Odonate Therapeutics, OncoPep, Kyowa Hakko Kirin, Samsung Bioepis, CytomX Therapeutics, Daiichi Sankyo, Athenex, Gilead, Mersana, Certara, Chugai Pharma, Ellipses Pharma, Infinity, 4D Pharma, OncoSec Medical, BeyondSpring Pharmaceuticals, OncXerna, Zymeworks, Zentalis, Blueprint Medicines, Reveal Genomics and ARC Therapeutics; and received institutional research funding from Genentech/Roche, Merck, Exelixis, Pfizer, Lilly, Novartis, Bristol Myers Squibb, Eisai, AstraZeneca, NanoString Technologies, Cyclacel, Nektar, Gilead, Odonate Therapeutics, Sanofi and Seattle Genetics. P.K.S. is a co-founder and member of the BOD of Glencoe Software, member of the BOD for Applied Biomath, member of the scientific advisory board for RareCyte, NanoString and Montai Health, holds equity in Glencoe, Applied Biomath and RareCyte, consults for Merck, and has received research funding to the institution from Novartis and Merck in the past 5 years. A.C.G.-C. has grant/research funding to institution from Gilead Sciences, AstraZeneca, Daiichi-Sankyo, Merck, Zenith Epigenetics, Bristol-Myers Squibb, Novartis, Foundation Medicine and Biovica; serves as a consultant and member of the scientific advisory board for AstraZeneca, Daiichi-Sankyo and Novartis; has received honoraria from AstraZeneca and Daiichi-Sankyo; and has other financial or materials support from Roche/Genentech, Gilead Sciences, AstraZeneca, Novartis and Merck. K.H. is a consultant for and a co-founder of Dania Therapeutics Aps and a scientific advisor for Hannibal Health Innovation. S.R.V.K. is a founder and consultant at Faeth Therapeutics and Transomic Technologies. D.A.B. is a consultant for N of One/QIAGEN and Tango Therapeutics; is a founder and shareholder in Xsphera Biosciences; has received honoraria from Merck, H3 Biomedicine/Esai, EMD Serono, Gilead Sciences, Abbvie and Madalon Consulting; and has received research grants from BMS, Takeda, Novartis, Gilead and Lilly. The other authors declare no competing interests.

Peer review

Peer review information

Nature thanks Charles Sawyers and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Extended data in relation to Fig. 1a–e.

a, Related to Fig. 1a. EZH2 mRNA levels within normal tissue (n = 111), or TNBC tumours resected at stage I (n = 19), stage II (n = 74), stage III (n = 19), or stage IV (n = 2). Box shows 10–90 percentiles with line at median. b, Treatment schema for all in vitro studies. c-d, Immunoblot depicting relative levels of AKT target inhibition (pPRAS40) and EZH2 target inhibition (H3K27me3) and relevant loading controls in sensitive (c) and resistant (d) cell lines from panel in Fig. 1c. Experiment was repeated at least 3 times. e, Dose dependence of cell death induced by the combination of EZH2i (tazemetostat) and AKTi (ipatasertib) measured by Incucyte Live Cell Imaging. Data for 5 μM AKTi are presented as the 72 h time point in Fig. 1d. n = 8. p *** <0.001 measured by unpaired one-tailed heteroscedastic Student’s T-test. Data are images from 8 independent wells. f, Synergy plots using Gaddum’s non-interaction model (HSA) for cells treated with AKTi (ipatasertib) and/or EZH2i (tazemetostat).

Extended Data Fig. 2 Extended data in relation to Fig. 1.

a, Relative change in cell number after 4 days of treatment with vehicle, AKTi (MK2206) and/or EZH2i (tazemetostat) (Day 4 versus Day 0). b, Relative change in TNBC cell numbers after 4 days of treatment with vehicle, AKTi (ipatasertib), EZH2i (tazemetostat), and/or EEDi (MAK683) (Day 4 versus Day 0). Immunoblot (repeated at least 3 times) depicting relative levels of AKT target inhibition (pPRAS40) and EZH2 target inhibition (H3K27me3) and relevant loading controls. c, Relative change in cell numbers after 4 days of treatment with vehicle, AKTi (ipatasertib) and/or EZH2i (tazemetostat) (Day 4 versus Day 0) in indicated cell lines. d, Immunoblot depicting relative levels of AKT target inhibition (pPRAS40) and EZH2 target inhibition (H3K27me3) and relevant loading controls in indicated cell lines from panel in (c). e, Crystal violet staining of SUM149PT cells treated with vehicle, AKTi (ipatasertib) and/or EZH2i (tazemetostat) and fixed after 10, 20, or 30 days of treatment. f, Relative change in cell number of MDA-MB-468 cells after treatment. Left – pre-treatment with EZH2i for 5 days followed by 4 days of combined treatment of EZH2i and/or AKTi. Middle – pre-treatment with EZH2i for 5 days followed by switch to vehicle or AKTi for an additional 4 days. Right – pre-treatment with EZH2i for 5 days followed by drug holiday for 4 days, followed by vehicle or AKTi treatment for an additional 4 days. Immunoblot on right depicts relative levels of AKT target inhibition (pPRAS40) and EZH2 target inhibition (H3K27me3) and relevant loading controls. For all subfigures, n = 3, data are mean ± s.d. of biological independent samples and p values were measured by unpaired one-tailed heteroscedastic Student’s T-test.

Extended Data Fig. 3 Extended data in relation to Fig. 1f–j.

a, EZH2 mRNA expression levels in sensitive and resistant TNBC cell lines from CCLE RNAseq dataset. Sensitive cells are ranked in order of sensitivity measured by amount of cell death observed in Fig. 1c. b, Immunoblot depicting PTEN expression and PIK3CA mutational status of TNBC cell lines with Fisher’s exact test panel. Experiment was completed at least 3 times with similar results. c, Graph depicting the relative tumour volume of MDA-MB-468 orthotopic xenografts over time treated with Vehicle, AKTi (ipatasertib) and/or EZH2i (tazemetostat). Note that established tumours (100mm³) were first pretreated with vehicle or tazemetostat for 7 days prior to the addition of these agents. p value determined using mixed effects models (REML) ANOVA between AKTi and Combo. n = 10–13 tumours per arm. Data are mean ± s.e. of biological independent samples. d, Kaplan-Meier survival plot of mice bearing MDA-MB-468 orthotopic xenografts treated with vehicle or EZH2i+AKTi. Treatment ended after day 28 and animals were monitored twice weekly for humane endpoint. Significance measured by Mantel-Cox Log Rank test. e, Graph depicting percent change in body weight of tumour bearing mice treated with the indicated drugs during the treatment duration. Data are mean ± s.e. of biological independent samples. n = 6 or 7 mice per condition. g, Copy number plots of chromosomes bearing Akt3 (e), Met (f), and Ezh2 (g) isolated from untreated GEMM organoid allograft tumours from K8-CreER; Trp53^fl/fl mice implanted orthotopically into nude mice. h, Photographs of H&E-stained (left) or EZH2 immunohistochemistry (right) sections of a normal mouse mammary fat pad or tumour from Fig. 1h. Scale bar = 150 μm.

Extended Data Fig. 4 Extended data in relation to Fig. 1l.

Photographs of H&E-stained (top) or Ki67 immunohistochemistry (bottom) tumour sections at the end of study depicted in 1 l. F = fibrosis; N = necrosis; C = calcification. Scale bar = 150 μm.

Extended Data Fig. 5 Extended data in relation to Fig. 2.

a, differentially expressed genes between EZH2i+AKTi and DMSO treated SUM149PT cells. A subset of luminal and basal breast markers are highlighted. Padj is DESeq2 p value adjusted for multiple hypothesis testing. b, ssGSEA z scores of LIM_MAMMARY_LUMINAL_MATURE gene signatures in MDA-MB-468 cells. c, Transcriptional heatmap of significantly differentially expressed genes in gene signatures from (Fig. 2a) in MDA-MB-468 cells. d, Correlation coefficient of cell lines treated indicated drugs with basal TNBCtypes. e-f, RNA-seq normalized counts of ER (e) and AR (f) in SUM149PT and MDA-MB-468 cells. g, GATA3 protein expression in two sensitive cell lines treated with indicated drugs. h, Immunoblot of SUM149PT cells transfected with siControl or siGATA3 and treated with indicated drugs. i, Synergy scores using Gaddum’s non-interaction model (HSA) in cells stably transduced with shRNA or sgRNA against GATA3 or a control sequence. Cells were treated with various concentrations of AKTi and/or EZH2i. j, Synergy plots using Gaddum’s non-interaction model (HSA) for cells stably transduced with shRNA against GATA3 or a control sequence and then treated with indicated drugs. k, Relative change in cell number of MDA-MB-468 stably transduced with sgRNAs or transfected with siRNAs against a control sequence or GATA3 and then treated with indicated drugs. l, immunoblot of p27 and loading control of MDA-MB-468 cells treated with indicated drugs for 16 h. m, ssGSEA z scores of Hallmarks – E2F target gene signature in SUM149PT and MDA-MB-468 cells. n, immunoblot of p27 and loading control GAPDH of protein lysates from MDA-MB-468 cells treated with indicated drugs, in the presence of siRNAs against a control sequence or GATA3. For all subfigures, data are mean ± s.d. of biological independent samples (n = 3), immunoblots were repeated at least 3 times, box plots show range of data with line at mean, AKTi = ipatasertib, and EZH2i = tazemetostat.

Extended Data Fig. 6 Data related to Fig. 3.

a, gene set enrichment analysis of open peaks enriched in sensitive and resistant TNBC cell lines at baseline as assessed by ATAC-seq. Adjusted p values are FDR calculated using the Benjamini Hochberg method within the GSEA software. b-c, leading edge plot of gene set enrichment analysis of CHARAFE BREAST CANCER BASAL VS MESENCHYMAL DOWN (mesenchymal genes, (b)) and CHARAFE BREAST CANCER BASAL VS MESENCHYMAL UP (basal genes, (c)) in sensitive and resistant cell lines at baseline as measured by RNA-seq. Adjusted p values are FDR calculated using the Benjamini Hochberg method within the GSEA software. d, heatmap of correlation coefficients of differentially expressed genes in TNBC cell lines generated in this study as DMSO treated samples or sourced from the CCLE dataset. e, relative cell counts of five additional TNBC cell lines predicted to be sensitive or resistant by machine learning models. Data are mean ± s.d. of biological independent samples. n = 3. f, performance metrics for machine learning models. g, oncoprint of TNBC tumours from TCGA firehose dataset with alterations for AKT3, PTEN, INPP4B, and PIK3CA denoted.

Extended Data Fig. 7 Data related to Fig. 4.

a, Chromatin looping architecture of GATA3 enhancer linked to GATA3 promoter using MCF10A HiC data from ENCODE. b, Genome browser view of GATA3 enhancer with ATAC-seq and CUT&RUN for H3K27me3 data from MDA-MB-468 cells treated with EZH2i and/or AKTi. Data is overlaid with HiC data for MCF10A and MCF7 cells. Note the ATAC-seq and CUT&RUN data are identical to in Fig. 4a for direct comparison purposes. c, schematic for siRNA screen completed in Fig. 4b.

Extended Data Fig. 8 Data related to Fig. 4.

a, immunoblot of FOXO1 and pFOXO1 expression in TNBC cell lines treated with EZH2i and/or AKTi. Experiment was repeated at least 3 times. b, RT-qPCR of FOXO1 expression after EZH2i and/or AKTi treatment in three sensitive TNBC cell lines. Values indicate z scores of three independent biological samples. c, genome browser view of FOXO1 locus with CUT&RUN data using H3K27me3 and H3K4me3 antibodies in chromatin from MDA-MB-468 cells treated with EZH2i and/or AKTi. d, genome browser view of GATA3 promoter with FOXO1 ChIP-seq occupancy (accessed from ENCODE). e, genome browser view of GATA3 enhancer with FOXO1 ChIP-seq occupancy (accessed from ENCODE).

Extended Data Fig. 9 Data related to Fig. 5.

a, ssGSEA z scores of HALLMARK_APOPTOSIS gene signatures in SUM149PT and MDA-MB-468 cells treated with AKTi and/or EZH2i. b, Relative BMF mRNA expression (RT-qPCR) in SUM149PT cells treated with EZH2i and/or AKTi. c, Protein expression of BMF in HCC1395 cells treated with EZH2i and/or AKTi with stable HA knocked-into the endogenous BMF locus. d-e, Relative change in cell number (f) or BMF expression (e) of SUM149PT cells transfected with siControl or siBMF e and then treated with EZH2i and/or AKTi. f, Transcriptional heatmap depicting expression of significantly differentially expressed genes from Fig. 5d in MDA-MB-468 cells. g, involution gene mRNA expression measured by RT-qPCR in SUM149PT cells transfected with siControl or siSTAT3 in response to indicated drugs. h, Relative change in cell number of MDA-MB-468 cells treated with EZH2i+AKTi or docetaxel (10 nM) after 4 days. i, BMF expression (z-score of DESeq2 normalized counts) in MDA-MB-468 cells treated with EZH2i+AKTi or docetaxel. j, Immunoblot depicting activation of STAT3 (pSTAT3) and target inhibition after treatment with EZH2i+AKTi or docetaxel in MDA-MB-468 cells. k -l, ssGSEA z scores (k) or heatmap (l) of Mammary Gland Involution gene signatures in MDA-MB-468 cells treated with indicated drugs. m,n, Immunoblots depicting confirmation of STAT3 knockdown and inhibition of AKT and EZH2 targets in SUM149PT (m) or MDA-MB-468 (n) cells. Related to Fig. 5i. o, immunoblot of SUM149PT cells transfected with siControl or siJAK1 and then treated with indicated drugs. Related to Fig. 5j. For all subfigures, data are mean ± s.d. of biological independent samples (n = 3), p values measured by unpaired one-tailed heteroscedastic Student’s T-test, box plots show range of data with line at mean, immunoblots were repeated at least 3 times, EZH2i = tazemetostat, AKTi = ipatasertib, and docetaxel dosed at 10 nM.

Extended Data Fig. 10 Data related to Fig. 6.

a, relative IL6 protein (z score) in SUM149PT and MDA-MB-468 cells after 24 h of treatment with EZH2i (tazemetostat) and/or AKTi (ipatasertib). b, Relative IL6R expression in SUM149PT and MDA-MB-468 cells transfected with siRNA against a control sequence or IL6R and then treated with vehicle or EZH2i+AKTi. Related to Fig. 6a. c, Relative cell numbers of SUM149PT and MDA-MB-468 cells treated with vehicle, EZH2i+AKTi, and/or neutralizing antibody against IL6R (tocilizumab). d, relative IL6 protein level (ELISA) in MDA-MB-468 cells transfected with siControl or siSTING and treated with indicated drugs for 24 hr. n = 2. e, 2′3′-cGAMP levels measured by ELISA in MDA-MB-468 cells after treatment with indicated drugs. f, volcano plot depicting -log10(FDR) versus log2(fold change) for expression of endogenous retroviruses in EZH2i vs DMSO treated MDA-MB-468 cell line. g, transcriptional heatmap of BMF levels in a panel of TNBC cell lines treated with EZH2i and/or AKTi and harvested after 24 h. BMF levels were normalized to STAU1 and then relative expression was normalized to each cell line’s DMSO samples. Data for sensitive cell lines is generated from same experiment as Fig. 5a. Heatmap values indicate fold change. h, relative cell numbers of SUM159PT cells transduced with GFP or GATA3 overexpression construct treated with EZH2i+AKTi and/or STING agonist ADUS100. For all subfigures, data are mean ± s.d. of biological independent samples. Unless indicated, all experiments are n = 3. p values measured by unpaired one-tailed heteroscedastic Student’s T-test.

Supplementary information

Reporting Summary

Supplementary Data 1

Immunoblots: source data for all immunoblots in the paper.

Supplementary Data 2

Raw animal data: source data for all studies that used animals in the paper.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Schade, A.E., Perurena, N., Yang, Y. et al. AKT and EZH2 inhibitors kill TNBCs by hijacking mechanisms of involution. Nature 635, 755–763 (2024). https://doi.org/10.1038/s41586-024-08031-6

Download citation

Received: 03 June 2022
Accepted: 06 September 2024
Published: 09 October 2024
Issue Date: 21 November 2024
DOI: https://doi.org/10.1038/s41586-024-08031-6

This article is cited by

Enhancer reprogramming: critical roles in cancer and promising therapeutic strategies
- Jinshou Yang
- Feihan Zhou
- Yupei Zhao
Cell Death Discovery (2025)
Protective effects of berberine on MASLD: regulation of glucose and lipid metabolism through PI3K/Akt and STING pathways
- Jing Ji
- Ying Li
- Fafeng Cheng
Naunyn-Schmiedeberg's Archives of Pharmacology (2025)