Epigenetic and Oncogenic Inhibitors Cooperatively Drive Differentiation and Kill KRAS-Mutant Colorectal Cancers

Colorectal cancers represent the second highest cause of cancer mortality worldwide, with a 5-year survival rate of <13% for advanced disease (1–3). Many colorectal cancers exhibit hyperactivated RAS/RAF/MEK/ERK signaling, with activating KRAS mutations occurring in 40% to 50% of all colorectal cancers (4–7). Conventional chemotherapies have limited efficacy in advanced tumors, and single-agent kinase inhibitors, including MEK- and EGFR-targeted therapies, are ineffective against colorectal cancers that harbor KRAS mutations (8, 9). Fortunately, the KRAS^G12C inhibitor, adagrasib, combined with the EGFR-directed antibody, cetuximab, has demonstrated promising clinical activity, resulting in accelerated FDA approval for progressive KRAS^G12C-mutant tumors (10). Nevertheless, KRAS^G12C mutations are relatively infrequent in colorectal cancers, representing only 11% of KRAS mutations in these malignancies, and are only present in 3% of colorectal cancers overall (11). In addition, clinical responses to these agents are generally incomplete and not durable. Thus, there is a critical need for developing additional improved therapies for KRAS-mutant colorectal cancers.

Disease heterogeneity and cellular plasticity both hamper the efficacy of targeted therapies and contribute to de novo and acquired resistance in colorectal cancer (12). In intestinal cells, cellular plasticity is often regulated by epigenetic mechanisms, which allow cells to dynamically alter their differentiation state in response to injury, inflammation, and stress (13). During tumor development, nonmutational epigenetic reprogramming through the dysregulation of epigenetic regulators is also thought to control tumor cell plasticity (14). Therefore, we hypothesized that targeting a traditional oncogenic signal along with a critical epigenetic regulator might offer a powerful therapeutic approach if we were able to drive or lock cells into a more vulnerable cell state.

EZH2, the catalytic component of polycomb repressive complex 2 (PRC2), is a major epigenetic regulator and is commonly overexpressed in solid tumors (15). Polycomb group proteins are highly conserved developmental regulators that maintain cellular identity by dynamically silencing key genes involved in differentiation (16, 17). EZH2 silences transcription at specific genomic sites by methylating histone 3 at lysine 27 (H3K27me3), thereby reducing chromatin accessibility to transcriptional machinery (15, 18). Notably, alterations in EZH2 have been proposed to play an important role in many tumor types (19–21). Although activating mutations in EZH2 have been detected in a subset of cancers, such as lymphomas and melanomas, EZH2 is more commonly overexpressed in solid tumors (22–25). PRC2 is known to play a broad role in stem cell maintenance and has been proposed to regulate the differentiation of intestinal cells (26); however, little is known about how EZH2 overexpression contributes to the development or maintenance of colorectal cancers.

In this study, we show that combined EZH2 and RAS pathway inhibitors potently kill colorectal cancer cells and promote dramatic tumor regression in multiple models in vivo. We further show that these agents converge by upregulating genes that suppress the WNT pathway, drive intestinal differentiation, and trigger apoptosis, all of which contribute to this synergy. Together, these findings reveal a promising therapeutic strategy for advanced colorectal cancer, elucidate the mechanism by which these agents function, and illustrate a paradigm for developing epigenetic-based combination therapies.

Analysis of the Cancer Genome Atlas (TCGA) Firehose Legacy dataset reveals that EZH2 is overexpressed in >66% of all samples from patients with colorectal cancer when compared with normal tissue (Fig. 1A). Notably, EZH2 is overexpressed in a similar percentage of KRAS-mutant tumors (61%), and equally high expression levels are observed in tumors that harbor different KRAS-mutant alleles (Supplementary Fig. S1A). To determine whether EZH2 inhibition might sensitize colorectal cancers to RAS pathway inhibitors, we first evaluated these agents in a panel of human cell lines harboring a variety of KRAS mutations. Tumor cells were pretreated for 5 days with the EZH2 inhibitor tazemetostat (EPZ-6438) to allow for the turnover and loss of H3K27me3 (27), followed by the addition of trametinib (Supplementary Fig. S1B). Because colorectal cancer cell lines exhibit different sensitivities to trametinib, a dose that exerted the maximal cytostatic response was empirically determined for each cell line (1–50 nmol/L; Supplementary Fig. S1C) and was used to facilitate comparison (Fig. 1B). EZH2 inhibition alone exerted minimal or occasional cytostatic effects; however, when exposed to both agents, cells rapidly died, as demonstrated by a loss of 50% to 75% of cells after only 5 days of co-treatment (Fig. 1B). Moreover, whereas single concentrations of trametinib are shown in Fig. 1B, tazemetostat broadly enhanced the effects of trametinib at multiple doses (Supplementary Fig. S1C), and the effects were synergistic, as determined by the highest single agent (HSA) synergy model (Fig. 1C; Supplementary Fig. S1D).

In all cell lines, tazemetostat and trametinib effectively inhibited H3K27me3 and pERK, respectively, and a pattern of cooperative suppression of either target was not observed (Fig. 1D). EZH2 was overexpressed in all of these tumor cell lines, and small differences in EZH2 expression did not correlate with subtle differences in sensitivity to these agents (Supplementary Fig. S1E). Notably, the response to combined EZH2 and MEK inhibition was durable, and cells continued to die for at least 2 weeks (Fig. 1E). Importantly, this drug combination did not kill normal human intestinal epithelial cells (Supplementary Fig. S1F). Finally, to determine whether cells were dying via apoptosis, caspase-3/7 activity was measured using live-cell imaging. Whereas trametinib alone induced low levels of apoptosis, together these agents triggered cell death in 60% to 75% of cells in less than 40 hours (Fig. 1F; Supplementary Fig. S1G).

MEK inhibitors were used for the studies described above because the majority of KRAS mutations found in colorectal cancers are still not directly targetable by clinically approved agents. However, KRAS^G12D accounts for 33% of KRAS mutations in this tumor type (28). To determine if EZH2 inhibition could also enhance the effects of recently developed KRAS^G12D inhibitors, cells were pretreated with tazemetostat and then exposed to increasing concentrations of MRTX1133 (29). Tazemetostat substantially sensitized colorectal cancers to this inhibitor (Fig. 1G). Notably, colorectal cancer cells harboring the KRAS^G12C allele also exhibited enhanced sensitivity to the covalent KRAS^G12C selective inhibitor, MRTX849, when combined with tazemetostat (Fig. 1H). These effects could be recapitulated using a different MEK inhibitor, binimetinib (Supplementary Fig. S1H), or by replacing the EZH2 inhibitor with an embryonic ectoderm development (EED) inhibitor, which suppresses another obligate PRC2 component (Supplementary Fig. S1I).

To determine if these effects might extend to BRAF-mutant colorectal cancers and/or whether they were dependent on RAS pathway activation at all, additional colorectal cancer models were investigated. Four BRAF^V600E colorectal cancer cell lines were pretreated with tazemetostat and then exposed to either trametinib, the BRAF^V600E mutant−selective inhibitor encorafenib, or the recently approved combination of encorafenib plus cetuximab, which inhibits feedback activation of the RAS pathway (30). In all settings, tazemetostat potentiated the effects of these RAS pathway inhibitors (Fig. 1I), which was observed at multiple doses (Supplementary Fig. S1J). By contrast, EZH2 inhibition did not trigger a cooperative cytotoxic response when combined with trametinib at any dose in three additional lines that lacked KRAS or BRAF mutations (Fig. 1J; Supplementary Fig. S1K). Taken together, these findings demonstrate that EZH2 inhibition broadly enhances the cytostatic effects of RAS pathway inhibitors in colorectal cancers but only in tumors driven by mutations in RAS pathway components.

To evaluate the efficacy of this therapeutic strategy in vivo, a cell-derived xenograft (CDX) model (LOVO) and four patient-derived xenograft (PDX) models were generated (COCA74P, derived from a primary rectal tumor with a KRAS^G12V mutation; COCA9, derived from a previously treated liver metastasis with a KRAS^G12V mutation; COCA4, derived from a primary tumor with a KRAS^G12D mutation from an individual with Lynch syndrome; and COCA30, derived from a previously treated primary tumor with a KRAS^G12D mutation; refs. 31, 32). Similar to the design of in vitro studies, mice with established tumors were pretreated with the EZH2 inhibitor or vehicle for 7 days before the addition of trametinib, and treatment continued for a total of 28 days. Notably, EZH2 or MEK inhibition alone had minimal effects on tumor growth throughout the entire duration of treatment in all three models (Fig. 2A–J). However, combined EZH2/MEK inhibitors triggered potent tumor regression (Fig. 2A–J) with no observed toxicity or weight loss (Supplementary Fig. S2A–S2E). Specifically, the LOVO xenograft model regressed up to 75% (Fig. 2A), whereas the PDX models (COCA74P, COCA9, COCA4, and COCA30) regressed up to 74%, 54%, 48%, and 72%, respectively, over 4 weeks (Fig. 2C, E, G, and I).

While longer treatment was not possible because of the need for repeated oral gavaging (3×/day), mice harboring COCA74P tumors were followed up after treatment ended at 28 days. Animals exposed to vehicle, trametinib, or tazemetostat alone died relatively quickly (median survival 42, 48, and 46 days, respectively); however, survival was substantially increased in mice treated with both agents together, despite the cessation of treatment (median survival 77 days; Fig. 2K).

We also evaluated the therapeutic efficacy of these agents in organoids derived from previously characterized genetically engineered mouse models with mutations in Apc, Kras, and Trp53 (termed AKP) or Apc, Kras, Trp53, and Tgfbr1 (termed AKPT; refs. 33, 34). Similar to cell line studies, tumor organoids were pretreated with EZH2 inhibitor for 5 to 7 days, prior to the addition of MEK inhibitor. Interestingly, a subset of organoids treated with the EZH2 inhibitor adopted a branched, gut-like morphology, reminiscent of differentiated organoids grown in WNT-depleted media (Fig. 3A; ref. 35). Additionally, whereas the MEK inhibitor reduced proliferation in both models, only the combination triggered cell death as illustrated by a rapid decrease in cell number (Fig. 3B; Supplementary Fig. S3A). Human tumor organoids derived from a KRAS^G12D-mutant tumor similarly died in response to these agents (Supplementary Fig. S3B).

Based on these interesting morphological changes, we considered the possibility that EZH2 inhibition might be functioning by promoting a shift in the differentiation state. To investigate this possibility, RNA sequencing (RNA-seq) was performed using multiple human colorectal cancer cell lines treated with EZH2 and/or MEK inhibitors or EZH2 and KRAS^G12D inhibitors. mRNA was collected 16 hours after treatment with the combination, prior to the onset of substantial cell death. Single sample gene set enrichment analysis (ssGSEA) revealed that many genes associated with markers of differentiated intestinal cell types (enterocytes, enteroendocrine cells, goblet cells, deep crypt secretory cells, and tuft cells) were upregulated by these agents (Fig. 3C; Supplementary Fig. S3C; refs. 36, 37). Surprisingly, however, we found that both agents contributed to this response. Notably, the signatures that were most cooperatively upregulated were associated with goblet cells, tuft cells, and enteroendocrine cells, suggesting that EZH2 and MEK inhibitors preferentially drive differentiation toward a secretory lineage. Concomitantly, signatures associated with intestinal stem cells decreased (Fig. 3C; Supplementary Fig. S3C; ref. 38). Although the relative contribution of each agent on specific signatures varied between cell lines, in all cases both agents were required for a maximal shift in the differentiation state. This dramatic shift in gene expression was further confirmed by immunoblotting (Fig. 3D; fully quantified in Supplementary Fig. S3D). For example, well-established genes associated with differentiated intestinal cells, including ATOH1, CDX2, and KRT20, were upregulated 4.6- to 57.2-fold, whereas stem-like associated genes including LGR5, SOX9, and PROM1 were downregulated, by 1.6- to 124.2-fold (Fig. 3D; ref. 39). Consistent with RNA-seq data, the relative contribution of each agent to the induction or suppression of different markers varied, but overall, both agents were required for a maximal response.

The WNT pathway plays a critical role in the self-renewal and maintenance of intestinal epithelial cells (40, 41). Moreover, suppression of WNT signaling has been shown to induce differentiation and promote tumor regression in various animal models of colorectal cancer (42–45). Therefore, we investigated whether these agents were suppressing oncogenic intestinal WNT transcriptional signatures (46, 47). ssGSEA of multiple oncogenic intestinal WNT pathway signatures further revealed that, like intestinal differentiation genes, both agents were required for maximal suppression of the pathway (Fig. 3E). Similar cooperative suppressive effects were observed in the KRAS^G12D cells treated with EZH2/KRAS^G12D inhibitors (Supplementary Fig. S3E). These data suggest that EZH2 and RAS pathway inhibitors also cooperatively suppress WNT-driven transcription.

However, to determine whether suppression of oncogenic intestinal WNT signaling was required for cell death in response to EZH2 and MEK inhibitors, colorectal cancer cells were transduced with a constitutively active form of β-catenin, lacking the first N-terminal 90 amino acids that encode critical regulatory sequences (β-catenin Δ90; ref. 48). Importantly, β-catenin Δ90 restored the expression of oncogenic intestinal WNT signatures in cells treated with the combination (Fig. 3F, expression shown in Supplementary Fig. S3F). Moreover, its expression blocked the enhanced apoptosis triggered by EZH2/MEK inhibitors (Fig. 3G) and the subsequent depletion of cells in multiple models (Fig. 3H; Supplementary Fig. S3G), demonstrating that suppression of the WNT/β-catenin pathway is functionally required for cell death. Of note, MEK and EZH2 inhibitors are also known to independently affect the expression of various cell cycle regulators, including p21, p27, and cyclin D1, which we observed (Supplementary Fig. S3H; ref. 49). Accordingly, these agents also cooperatively activated the RB pathway (Supplementary Fig. S3I and S3J), which was not restored by β-catenin Δ90 reconstitution (Supplementary Fig. S3I and S3J). Therefore, we conclude that WNT pathway suppression is required for cell death triggered by combined EZH2 and MEK inhibitors; however, these agents can also restrict the proliferation of any remaining cells through additional mechanisms likely affecting multiple cell-cycle regulators.

Finally, we investigated whether blocking intestinal differentiation might also prevent cell death. As noted, we observed that CDX2, an intestinal-specific homeobox gene known to be critical for intestinal differentiation, was potently upregulated by these agents (50). Importantly, siRNA-mediated suppression of CDX2 also prevented the cytotoxic effects of the combination (Fig. 3I, knockdown shown in Supplementary Fig. S3K). Conversely, expression of SOX9 has been shown to drive a stem cell–like program that blocks intestinal differentiation in colorectal cancer (51). Accordingly, ectopic expression of SOX9 was also able to prevent cell death in response to these agents (Fig. 3J, expression shown in Supplementary Fig. S3L). Taken together, these data demonstrate that suppression of oncogenic WNT signaling and the induction of intestinal differentiation programs are required for the therapeutic response to EZH2 and MEK inhibitors.

Given the dramatic effects of these agents on differentiation markers in vitro, we investigated whether these changes were also occurring in vivo. Combined EZH2/MEKi triggered a rapid and progressive increase in mucinous cells in CDX tumors, which could be readily observed by gross histologic analysis of hematoxylin and eosin–stained sections within 1 to 4 days (Fig. 4A). Cyclic immunofluorescence (CycIF) revealed that this was accompanied by a dramatic decrease in SOX9 expression and a concomitant increase in CDX2 expression (red→green; Fig. 4A). A similar shift in the differentiation state was observed in PDX tumors, which also adopted a mucinous appearance, along with a reduction in SOX9 and increase in KRT20 (red→green; Fig. 4B). Whereas frank tumor regression could be readily measured within 3 days (Fig. 2A–I), drug-induced PARP cleavage could not be accurately quantified in tumor cells in vivo because of the high baseline levels of necrosis in these models and the progressive nature of cell death and clearing.

Consistent with the absence of weight loss, no changes in the expression patterns of CDX2, KRT20, and SOX9 were observed in the mouse colons even after long-term treatment with EZH2 and MEK inhibitors (Fig. 4C). A slight suppression of pERK in normal colon was observed (Supplementary Fig. S4A and S4B), but Ki67 levels did not change (Fig. 4D). These findings suggest that a therapeutic window can be achieved between KRAS-mutant/EZH2-overexpressing colorectal cancers and normal tissue. Although toxicity in humans can only be determined empirically, the observation that these agents do not affect body weight, body score, or colonic differentiation is promising. Indeed, a clinical trial evaluating tazemetostat with dabrafenib/trametinib in metastatic melanoma is underway (NCT04557956).

EZH2 functions by inhibiting the expression of specific genes, through the deposition of repressive H3K27me3 marks. Therefore, to identify de-repressed EZH2 targets required for this response, CUT&RUN analysis was performed on LOVO cells treated with EZH2 and MEK inhibitors using H3K27me3 antibodies. We first identified bona fide EZH2 targets by integrating RNA-seq and CUT&RUN data and found that 2,356 genes were differentially upregulated by the combination, 2,911 genes had differentially lost H3K27me3 binding at promoter regions, and 573 genes were upregulated and lost H3K27me3 binding, representing potential EZH2 targets of interest (Fig. 5A). These targets were further reduced to 299 genes by identifying those that were also transcriptionally upregulated in a second colorectal cancer cell line, SW620.

Among these 299 targets, we prioritized those implicated in colonic or intestinal differentiation or with a proposed tumor suppressive function and enriched this list for transcription factors, which could have broad reprogramming effects. We then initiated a screen to identify genes that were required for cell death in response to EZH2/MEK inhibitors by suppressing each gene individually with pooled siRNA sequences. Strikingly, a hit that blocked the cytotoxic effects of the combination by ≥75% (when suppressed) was identified within the first panel of 15 genes: the TLE4 gene (Fig. 5B, knockdown shown in Supplementary Fig. S5A). Two independent Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) guide sequences confirmed the requirement for TLE4 in the cytotoxic response to EZH2 and MEK inhibitors (Fig. 5C and D; Supplementary Fig. S5B). Immunoblots further demonstrated that the TLE4 protein is potently upregulated by these agents by 5.48-fold (Fig. 5D, quantified in Supplementary Fig. S5C). As expected, EZH2 inhibition exerted a dominant effect on TLE4 expression levels; however, we noted that MEK inhibitors slightly enhanced its expression (Fig. 5D, quantified in Supplementary Fig. S5C).

TLE4 suppression prevented the dramatic loss of cells in multiple cell lines (Fig. 5B and E) and suppressed the enhanced apoptosis triggered by the combination (Supplementary Fig. S5D). Of note, TLE4 ablation had no effect on RAS/ERK signaling. In colorectal cancer models, trametinib rapidly suppresses pERK; however, it becomes partially reactivated within 24 to 48 hours (Supplementary Fig. S5E), which is thought to contribute to its lack of efficacy. As mentioned previously but shown here in more detail, tazemostat did not further suppress RAS signaling or prevent the rebound of pERK in response to trametinib (Supplementary Fig. S5F). TLE4 ablation also had no effect on pERK levels (Supplementary Fig. S5F). Together these findings demonstrate that EZH2 inhibition does not potentiate the therapeutic effects of RAS pathway inhibitors by further suppressing RAS signaling and that TLE4 ablation does not rescue these effects by restoring or enhancing RAS signaling.

Most importantly, TLE4 ablation prevented tumor regression in vivo induced by these agents (Fig. 5F). Consistent with additional known effects on cell-cycle components (Supplementary Fig. S3H–S3J), EZH2 inhibitor and MEK inhibitor (EZH2i/MEKi) was still able to confer cytostasis in the absence of TLE4, similar to observations in in vitro β-catenin Δ90 and SOX9 reconstitution studies (Fig. 3H–J). Thus, TLE4 upregulation is required for apoptosis and the full therapeutic response to EZH2 and MEK inhibitors, but surviving tumor cells can be further restrained by additional mechanisms likely affecting cell-cycle components.

TLE4 is a member of the TLE family of transcriptional repressors involved in the negative regulation of canonical WNT signaling (52). When WNT signaling is active, β-catenin translocates to the nucleus and associates with TCF/LEF factors to drive the transcriptional activity of WNT-responsive targets. However, transcription can be suppressed by Groucho/TLE repressive complexes that competitively associate with TCF/LEF factors (53), providing a potential link between TLE4 induction and WNT pathway suppression in this setting. CUT&RUN analysis further demonstrated that H3K27me3 was bound throughout the entire TLE4 gene in colorectal cancer cells and showed that binding was lost in response to EZH2 inhibitor alone or the combination (Fig. 5G), suggesting that EZH2 inhibitors control the expression of TLE4 by removing repressive H3K27me3 marks at the TLE4 locus. Moreover, analysis of transcriptional profiles from colorectal cancer cells with and without TLE4 demonstrated that TLE4 ablation partially restored the expression of β-catenin–regulated genes in these cells (Fig. 5H). Together, these findings suggest that TLE4 is a critical PRC2 target in colorectal cancers, its de-repression is required for the therapeutic response to EZH2 and MEK inhibitors, and its induction contributes to the WNT pathway suppression. Nevertheless, we expected that additional genes were likely contributing to this response.

In the course of this analysis, we noted that tazemetostat alone increased the expression of genes in which H3K27me3 binding was lost at proximal regulatory sites, as expected [Fig. 6A (third vs. first column)]. Surprisingly, however, trametinib substantially enhanced the effects of tazemetostat, resulting in the de novo or cooperative upregulation of the majority of PRC2 targets [275 and 259 genes, respectively; Fig. 6A (last column)]. Gene Ontology (GO) term analysis revealed that signatures associated with WNT suppression, differentiation, and development were significantly enriched in this list of 534 de novo/cooperatively upregulated PRC2 targets (Fig. 6B), suggesting that combined EZH2 and MEK inhibitors might be exerting broader suppressive effects on these pathways. Notably, the same cooperative upregulation of PRC2 targets was observed in cells treated with KRAS^G12D and EZH2 inhibitors (Supplementary Fig. S6A and S6B).

An unbiased assessment of GO terms associated with these 534 genes indicated that many were components of WNT pathway suppression, cell differentiation, or developmental process signatures (Fig. 6B). Notably, multiple known negative regulators of the Wnt/β-catenin pathway were among the uniquely induced PRC2 targets, although many function upstream of APC (e.g., DKK3, SFRP4, WIF1, and FRZB; Fig. 6C). Therefore, although potentially important in tumors with wild-type APC, the induction of these genes might not be expected to contribute to WNT pathway suppression in our models, which harbor APC mutations. However, we identified some genes (e.g., TRPS1 and SOX6) which were also present in this list and have been proposed to exert more direct suppressive effects on β-catenin (54, 55). In addition, we noted that ATOH1, a transcription factor known to promote differentiation into the secretory lineage, was also cooperatively induced by EZH2 and MEK inhibitors (Fig. 6C; ref. 56). To determine whether upregulation of these genes might also contribute to the therapeutic response, they were individually ablated using siRNA pools, and cells were treated with EZH2 and MEK inhibitors as described in Fig. 5B. Notably, the ablation of TRPS1, SOX6, and to a lesser extent ATOH1 all reduced the cytotoxic response to EZH2 and MEK inhibitors (Fig. 6D; knockdown shown in Supplementary Fig. S6C–S6E). Similar observations were made in two additional colorectal cancer cell lines, with some variation in the degree of rescue of each component, as might be expected (Fig. 6E and F; Supplementary Fig. S6E–S6I). Although we were unable to effectively ablate all three components simultaneously, each of them has been implicated in WNT pathway suppression and/or differentiation, and future studies will be required to understand their precise function. Together these observations demonstrate that TLE4 is a critical PRC2 target that suppresses WNT/β-catenin–driven transcription and mediates the therapeutic response; however, these agents cooperatively induce a larger network of genes that suppress the WNT pathway and/or promote differentiation, which could be further investigated in future studies.

These data indicated that EZH2 and MEK inhibitors cooperatively suppress WNT/β-catenin signaling and drive differentiation, which are required for cell death. However, ectopic expression of TLE4 or genetic ablation of β-catenin was not sufficient to trigger the acute loss of cells alone or when combined with MEK inhibitors (Supplementary Fig. S7A and S7B). Therefore, we also sought to identify specific apoptotic regulators involved in this process. Indeed, EZH2 and MEK inhibitors also cooperatively upregulated genes associated with apoptosis (Fig. 7A). Strikingly, analysis of the differentially expressed genes after EZH2/MEK inhibition revealed that the proapoptotic gene BMF was induced ≥30-fold and was the most significantly upregulated gene on this list (Fig. 7B). CUT&RUN analysis further revealed that, whereas H3K27me3 binding was relatively low at the BMF promoter and the gene body, broad H3K27me3 peaks were present at sequences immediately upstream, which were lost in response to EZH2 inhibition, consistent with a suppressive role for PRC2 at these sequences (Fig. 7C). Similar to the PRC2 targets shown in Fig. 6A, inhibition of both EZH2 and MEK was required for its maximal induction, which was observed in multiple colorectal cancer cell lines (Fig. 7D).

BMF is a BH3-only protein that acts as an apoptotic sensitizer by binding and sequestering antiapoptotic proteins including BCL-2, BCL-xL, and MCL-1 (57, 58). Importantly, we found that RNAi-mediated suppression of the BMF gene inhibited the robust loss of cells triggered by EZH2 and MEK inhibitors in multiple cell lines (Fig. 7E). BMF inhibition also suppressed apoptosis triggered by these agents (Supplementary Fig. S7C). Because BMF antibodies are of poor quality, we introduced a hemagglutin tag (HA) epitope into the endogenous BMF locus via a CRISPR/Cas9 approach into LOVO and SW620 cells (57). Immunoblots illustrate the potent upregulation of the endogenous (tagged) protein and its effective knockdown by pooled siRNA sequences in both cell lines (Fig. 7F; Supplementary Fig. S7D). Knockdown was confirmed in other cell lines by qPCR (Supplementary Fig. S7E and S7F). Notably, these responses could be recapitulated by CRISPR/Cas9 sgRNA–mediated ablation of BMF (Fig. 7G and H; Supplementary Fig. S7G), which also prevented tumor regression in vivo (Fig. 7I). Interestingly, we found that SOX9 expression and CDX2 ablation, both which block differentiation and inhibit the cytotoxic response as shown in Fig. 3I and J, did not prevent the upregulation of BMF (Supplementary Fig. S7H and S7I). Taken together, these findings demonstrate that EZH2 and RAS pathway inhibitors independently drive differentiation and upregulate BMF and that both are required for cell death. This also suggests that a more differentiated state is required for BMF to exert its apoptotic effects. A model summarizing all of these data is shown in Fig. 7J.

Here, we report a promising therapeutic strategy for KRAS-mutant colorectal cancers, one of the most treatment-refractory subtypes of this disease. Specifically, we have shown that EZH2 and RAS pathway inhibitors potently synergize, trigger apoptosis, and induce dramatic tumor regression in multiple PDX tumor models, including a model derived from a previously treated colorectal cancer metastasis. We further demonstrate that these agents function by cooperatively driving differentiation and inducing the expression of BMF, a proapoptotic Bcl-2 family member. Genetic studies indicate that both processes are required for cell death, suggesting that cells in a more differentiated state are more sensitive to this apoptotic signal.

These effects are mediated, in part, by the induction of TLE4, a member of the TLE/Groucho family of transcriptional repressors, which play a well-established role in repressing WNT/β-catenin–regulated transcription. We show that TLE4 is a bona fide PRC2 target gene and is decorated with repressive H3K27me3 marks in colorectal cancer cells. Accordingly, the induction of TLE4 requires the loss of H3K27me3, conferred by EZH2 inhibition. Unexpectedly however, we discovered that RAS pathway inhibitors cooperatively regulate the majority of PRC2 targets in colorectal cancers, triggering their induction but only when repressive H3K27me3 marks are lost in response to EZH2 inhibition. Interestingly, many of these genes also suppress the WNT pathway and/or promote intestinal differentiation, and we confirmed that a subset contributes to the therapeutic response. Finally, we show that EZH2 and RAS pathway inhibitors also converge on critical apoptotic signals via a similar mechanism, specifically the proapoptotic protein BMF. Taken altogether, these studies mechanistically reveal how EZH2 inhibitors epigenetically reprogram colorectal cancers and consequently transform the transcriptional response to RAS pathway inhibitors. This epigenetic reprogramming converts the relatively modest effects of RAS pathway inhibitors to a robust transcriptional response that drives WNT suppression, differentiation, and apoptosis in colorectal cancers.

Interestingly, suppression of WNT signaling has been shown to induce differentiation and promote tumor regression in various mouse models of colorectal cancer (42–45). Nevertheless, WNT pathway inhibition is not universally effective in all human cell line models (59). However, Dow and colleagues found that inhibition of CDK4, a distal downstream effector of the RAS pathway, could synergize with tankyrase inhibition in resistant models and induce cell-cycle arrest and senescence (59). Perhaps EZH2 and RAS pathway inhibitors are functioning in an analogous capacity, as we have shown here that EZH2 and RAS pathway inhibitors cooperatively suppress WNT signaling and activate RB. However, additional induction of the apoptotic regulator BMF, which specifically requires EZH2i-mediated de-repression, may convert cytostatic effects to a cytotoxic response.

Many of the studies presented here describe the combined effects of EZH2 and MEK inhibitors, which could be broadly effective in colorectal cancers with all types of KRAS mutations. However, 8% and 33% of KRAS-mutant colorectal cancers harbor KRAS^G12C and KRAS^G12D mutations, respectively. As such, there is a great deal of enthusiasm for the newly developed KRAS^G12C and KRAS^G12D inhibitors in these settings. Indeed, EZH2 inhibitors similarly suppress WNT signaling, drive differentiation, and kill colorectal cancer cells when combined with these allele-selective inhibitors, providing important support for these combinations as well.

Altogether these studies have defined a promising therapeutic strategy for KRAS-mutant colorectal cancers, elucidated the mechanism by which these agents function, and uncovered important insight about how EZH2 and the RAS pathway cooperatively regulate the WNT pathway, differentiation, and survival/cell death in colorectal cancers. Indeed, the orthogonal activation/upregulation of RAS and EZH2 may provide a failsafe mechanism for suppressing differentiation and apoptosis, which could underlie the treatment-refractory nature of KRAS-mutant colorectal cancers. Regardless, these findings have important clinical implications and provide strong mechanistic support for developing new clinical trials in colorectal cancers.

All cell lines were purchased from ATCC. Cell lines were authenticated using short-tandem repeat profiling (Labcorp). All cell lines were regularly tested for Mycoplasma using the MycoAlert Mycoplasma Detection Kit (Lonza, LT07-318). All cells were used for experiments within 10 to 15 passages from thawing. Cells were cultured in Ham’s F-12 (Gibco, #11765), Eagle’s minimum essential medium (Corning, #10-009-CV), DMEM (Corning, #10-013-CV), RPMI (Corning, #10-040-CV), or OptiMEM (Gibco, #31985), supplemented with FBS and 1× concentration of pencillin/streptomycin/glutamine (Gibco, #10378016). All cell lines were cultured with 10% FBS, except HIEC6 as specified by ATCC. Ten percent FBS was maintained during proliferation studies in all cell lines, except SW620 and SW837, in which 2% FBS was used starting at day 0. The precise concentration of trametinib used for proliferation studies in each cell line was determined by performing dose–response experiments to identify a cytostatic dose but not exceeding 50 nmol/L. Each cell line’s culture media, mutation, and concentrations of trametinib and encorafenib are as follows: LS513 (RPMI, KRAS^G12D, and trametinib 5 nmol/L), SW403 (DMEM, KRAS^G12V, and trametinib 10 nmol/L), SW1116 (DMEM, KRAS^G12A, and trametinib 10 nmol/L), SK-CO1 (Eagle’s minimum essential medium, KRAS^G12V, and trametinib 1 nmol/L), LOVO (Ham’s F-12, KRAS^G13D, and trametinib 50 nmol/L), SW837 (DMEM, KRAS^G12C, and trametinib 50 nmol/L), SW620 (DMEM, KRAS^G12V, and trametinib 50 nmol/L), H747 (RPMI, KRAS^G13D, and trametinib 10 nmol/L), HIEC6 (OptiMEM + 4% FBS and trametinib 50 nmol/L), HT29 (McCoy’s, BRAF^V600E, trametinib 5 nmol/L, and encorafenib 0.5 μmol/L), LS411N (RPMI, BRAF^V600E, trametinib 25 nmol/L, and encorafenib 2.5 μmol/L), COLO205 (RPMI, BRAF^V600E, trametinib 5 nmol/L, and encorafenib 1 μmol/L), COLO201 (RPMI, BRAF^V600E, trametinib 10 nmol/L, and encorafenib 1 μmol/L), KM12 (DMEM and trametinib 50 nmol/L), CACO2 (DMEM and trametinib 50 nmol/L), and SW48 (DMEM and trametinib 50 nmol/L).

Drugs were purchased from vendors: tazemetostat (SelleckChem, #S7128; used at 5 μmol/L), trametinib (LC Laboratories, #T8123), MRTX1113 (MedChemExpress; #HY-134813), MRTX849 (MedChemExpress; #HY-130149), binimetinib (SelleckChem, #S7007; used at 5 μmol/L), MAK683 (SelleckChem, #S8983; used 5 μmol/L), encorafenib (SelleckChem, #S7108), and cetuximab (SelleckChem, #A2000; used at 50 μg/mL).

To determine EZH2 mRNA levels in colorectal cancer tumors and matched normal tissue, data were obtained from TCGA Firehose Legacy dataset for COADREAD (illuminahiseq_rnaseqv2-RSEM_genes_normalized) at https://gdac.broadinstitute.org/. Colorectal cancer tumors were identified from the dataset for stage of disease progression and KRAS mutation status from https://www.cbioportal.org/.

To measure cellular proliferation and cell death, manual cell counting assays were performed. On day −5, cells were split and treated with either DMSO or tazemetostat. On day −3, cells were passaged 1:2 and maintained in DMSO or tazemetostat. On day −1, cells were seeded in triplicate in six-well plates (100,000–150,000 cells/well) for counting for day 0 timepoint and day 5 timepoint (unless otherwise stated) and 10 cm plates for protein lysates. For cell counting assays including siRNAs, cells were transfected with siRNA on day −1 and incubated for at least 8 hours before seeding. Twenty-four hours after seeding, a day 0 count was taken. Cells were trypsinized, spun down, resuspended in PBS, and manually counted in a hemocytometer. Experimental plates were dosed with the indicated compounds by replacing the media. On day 1, protein lysates were collected 16 to 24 hours after dosing, unless otherwise stated. For counting plates, the media and drug were refreshed after 3 days. During a final day 5 timepoint, the cells were counted using a hemocytometer. Final counts were normalized to respective day 0 counts, and log₂ fold change and percent change were calculated for graphical representation. Data points represent technical replicates from a representative experiment. All counting experiments and associated Western blots were repeated at least three times independently.

Similar to cell counting assays, after cells were pretreated for 5 days with tazemetostat or DMSO, the cells were seeded at 3,000 to 6,000 cells per well in 96-well clear-bottom black plates (Thermo Fisher Scientific, #165305) at three to five technical replicates per treatment condition with four images taken per technical replicate. Nuclei were either labeled by stably expressing Incucyte Nuclight Red (Sartorius, #4625) or transiently labeled with Incucyte Nuclight Rapid Red Reagent (Sartorius, #4717). Twenty-four hours after seeding, the cells were treated with media containing the indicated drug and CellEvent Caspase-3/7 Green Detection Reagent (Invitrogen, #C10423) at a final 1:1,000 dilution to detect the presence of caspase activity and apoptosis. Four images were acquired at 10× magnification in phase, green (300 ms acquisition time), and red (400 ms acquisition time) channels every 2 hours. Image acquisition and analysis were conducted using Incucyte S3 software. To determine the number of caspase 3/7–positive cells, the overlap (green and red) counts per well were normalized to red counts/well at each timepoint.

CellTiter-Glo assays were conducted to determine the synergistic interactions between EZH2 inhibitors (0, 1, 2.5, and 5 μmol/L) and MEK inhibitors (0, 1, 5, 10, and 50 nmol/L). Cells were seeded at 3,000 to 6,000 cells per well in a 96-well white flat-bottom plates with three technical replicates per treatment condition. Twenty-four hours after seeding, the cells were dosed with varying indicated concentrations of EZH2 and MEK inhibitors. Five days after combined treatment, cell viability was measured using CellTiter-Glo (Promega, #G9291), normalized to a day 0 measurement and then to DMSO to calculate the inhibitory response. Synergy scores were calculated using the Gaddum’s non-interaction model—HSA with SynergyFinder (60). An HSA value greater than 10 indicates a synergistic interaction.

Cell lysates were harvested after 16 to 24 hours after vehicle or drug treatment. Cells were washed with PBS, lysed in boiling 1% SDS lysis buffer [1% SDS (Invitrogen, #15553-035), 10 mmol/L Tris-HCl pH 7.5 (Sigma, #77-86-1), and 100 mmol/L NaCl (Sigma, #S5586)], scraped, collected, boiled at 95°C for 10 to 15 minutes, and spun down at maximum speed for 3 minutes. Protein concentration was determined using bicinchoninic acid quantification (Bio-Rad, #23222). Proteins were run on SDS-PAGE gels (Bio-Rad, #4561084) and transferred to Immobilon nitrocellulose membranes (Bio-Rad, #1620115). The membranes were blocked in 5% milk in Tris-buffered saline with Tween 20 (TBST) for 30 to 60 minutes and then incubated with primary antibodies overnight at 4°C. Membranes were incubated with infrared dye-conjugated (IRDye 800CW, #926-32213; IRDye 680CW, #926-68072) or HRP-conjugated secondary antibodies for 1 hour, and signal was measured using Odyssey LICOR Fc or autoradiographic film. The following antibodies were used at 1:500 to 1:1,000 dilution unless otherwise stated: pERK (CST, #4370, RRID: AB_2315112), ERK (CST, #9102, RRID:AB_330744), H3K27me3 (CST, #9733, RRID:AB_2616029), Histone H3 (CST, #4499, RRID:AB_10544537), TLE4 (Abcam, #ab64833, RRID:AB_2203850), HA-tag (CST, #3724, RRID:AB_1549585), ATOH1 (Proteintech, #21215-1-AP, RRID:AB_10733126), LGR5 (Abcam, #75850, RRID:AB_1523716), SOX9 (EMD Millipore, #AB5535 1:5,000, RRID:AB_2239761), PROM1 (CST, #64326, RRID:AB_2721172), CDX2 (CST, #12306, RRID:AB_2797879), KRT20 (CST, #13063, RRID:AB_2798106), KLF4 (CST, #4038, RRID:AB_2265207), CDCA7 (Proteintech, #15249-1-AP, RRID:AB_2878119), GAPDH (CST, #2118, RRID:AB_561053), vinculin (Santa Cruz Biotechnology, #sc-25336, RRID:AB_628438), β-catenin (CST, #8814, RRID:AB_11127203), p-RB (CST, #9308, RRID:AB_331472), RB (CST, #9313S, RRID:AB_1904119), p21 (CST, #2947, RRID:AB_823586), and p27 (SantaCruz, #sc-528, RRID:AB_632129).

Total RNA was isolated after 16 to 24 hours of vehicle, single-agent, or combination treatment using the RNeasy Plus Mini kit (Qiagen, #74136) and reverse transcribed to cDNA using the qScript cDNA synthesis kit (Quantabio, #95047-500) following the manufacturer’s instructions. qPCR amplification was done using PerfeCTa SYBR Green SuperMix Reagent (Quantabio, #95054-500), and reactions were run on the Bio-Rad CFX96 cycler in technical triplicates. Relative expression was calculated by first normalizing to housekeeping genes STAU1 or UBC and then to the DMSO sample using the ΔΔCq method.

For siRNA experiments, cells were incubated for 6 to 8 hours with 0.1 μmol/L siRNA constructs with a 1:400 dilution of Lipofectamine RNAiMAX transfection reagent (Invitrogen, #13778075) in antibiotic-free media. Non-targeting BMF, TLE4, CDX2, TRPS1, SOX6, ATOH1, MAF, DACH1, PTPRD, TSHZ2, CSMD1, EYA4, SMOC2, ATOH8, WNT5A, PPARGC1A, DEFA5, CADM1, CELF2, GLI3, CTNNB1 ON-TARGETplus pools and non-target controls were purchased from Horizon Biosciences (catalog #s: D-001810-10, L-004393-00, L-019336-00, L-015636-00, L-009644-00, L-015101-01, L-008915-01, L-003746-00, L-013222-00, L-008527-00, L-018447-02, L-007288-00, L-011853-00, L-013886-02, L-008817-00, L-003939-00, L-005111-00, L-013143-01, L-016565-00, L-012741-00, L-011043-00, and L-003482-00, respectively).

Oligos used for primers for RT-qPCR are listed as follows in 5′-3′ orientation: BMF (fwd: ACTTCAGCTCTTCCCTCTCA, rev: GAGTCTGGGTAGCTTTGTCTTC), UBC (fwd: ATTTGGGTCGCGGTTCTT, rev: TGCCTTGACATTCTCGATGGT), STAU1 (fwd: GGATGAGTTCAGGATGCCTTAT, rev: GGTGTGATGTCCTTGACTAACT), MAF (fwd: TGGAGTCGGAGAAGAACCA, rev: CTGCTCACCAACTTCTCGTATT), DACH1 (fwd: GGAAGGGTGGCTATGTGTTATT, rev: GCACTGTTTGCCGCTTTAC), PTPRD (fwd: CATGGAGAGGAGCAACGAATTA, rev: GTGCAGCCAGACGGAAATA), TSHZ2 (fwd: CCAGACATCAGAGGGCAAATAC, rev: GGTCATTGAGAGTCCCGTAAAC), CSMD1 (fwd: AACCACTACCTTCGTGCATAG, rev: GTTGTCATACGGAGCTGGATAG), EYA4 (fwd: ATCTCTCCCAGGACTGACTAAC, rev: ACTCCTACAGGTTCTCTCATCAA), SMOC2 (fwd: GACCTTCCTTTCCCGTTGT, rev: GCTCCTGGGTATACTTCCTTTC), ATOH8 (fwd: GGCAGAAGCTGTCCAAACT, rev: ACTGTAGTCAAGGTCAGCCA), WNT5A (fwd: CCTAGTGGCTTTGGCCATATT, rev: TCTGACATCTGAACAGGGTTATTC), PPARGC1A (fwd: TGAACTGAGGGACAGTGATTTC, rev: CCCAAGGGTAGCTCAGTTTATC), DEFA5 (fwd: CCATCCTTGCTGCCATTCT, rev: GTTGTAGCCTCATCAGCTCTTT), CADM1 (fwd: GCTTCTGCTGTTGCTCTTCT, rev: CTCGATCACTGTCACGTCTTTC), CELF2 (fwd: GAACCCTCCGCAGAGTAAAG, rev: GGATGATGCATCCCAGGTAAA), TLE4 (fwd: CAGCCTGCTCAACCCTTTA, rev: CCAGCTTCAGACTGTGGTATT), and GLI3 (fwd: TCTGACCGATGGAGGTAGTATAG, rev: GTTGCAGTGGAAATGGTTGAG). Oligos used for sgRNAs for CRISPR are listed as follows in 5′-3′ orientation: BMF (fwd: GGGAGCCGGTGACCCAACCC, rev: GGGTTGGGTCACCGGCTCCC), TLE4 #1 (fwd: CCGGCACTGCTACCGATGGG, rev: CCCATCGGTAGCAGTGCCGG), and TLE4 #2 (fwd: CATGACAATGATCACCAAAG, rev: CTTTGGTGATCATTGTCATG).

For CRISPR knockouts, oligos were purchased from Invitrogen with the described sequences and cloned into CRISPR Lenti-v2 backbone. For ectopic constitutive β-catenin expression, the pLV β-catenin Δ90 construct and its corresponding empty vector control were purchased from Addgene (#36985 and #85139). For ectopic SOX9 expression, the pLX304 SOX9 construct was a gift from Dr. Nilay Sethi from Dana-Farber. For ectopic TLE4 expression, the pLX304 TLE4 construct was purchased from Horizon Biosciences (Cat #OHS6085-213583570) from the CCSB–Broad Lentiviral Expression Collection. Lentivirus was produced in 293T cells using X-tremeGENE 9 DNA Transfection Reagent (Roche, #6365787001) and collected 72 hours after transfection. Cells were infected twice for 24 hours each at 1:2 dilution in the presence of 8 μg/mL polybrene. Cells were recovered from the infection for 24 hours and selected in 2 μg/mL puromycin for 3 to 5 days.

An N-terminal HA-tag was introduced into the endogenous locus of BMF as previously described (57). The crRNA (5′-TTGCCCCCTCACAGGAGAGA-3′) was hybridized with Alt-R CRISPR-Cas9 tracRNA (IDT, #1073190) at 95°C for 5 minutes at an equimolar ratio (0.375 nmol). The mix was cooled down to room temperature on the benchtop for 10 minutes before adding 2 μL of Alt-R S.p.Cas9 Nuclease V3 (IDT, #1081059) and 5 μL of 100 μmol/L single-stranded donor oligonucleotide (5′GCTGAGGGGGCAGTCCAGTAGGCTCTGGGCAAACAGGTCAGCAGAGAGCAAGCTCCCGGGTTGGGTCACCGGCTCCCCATCCTCTGGTTGGAACACATCATCCTCCAGCTCCTCCACACACTGAGATGGCTCAGCGTAATCTGGTACGTCGTATGGGTACATCTCTCCTGTGAGGGGGCAACGCAGGCATCTGGGCTGCT-3′). The mix was incubated for 20 minutes at room temperature and added to 1 million LOVO and SW620 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 E0-117 program in a 4D-Nucleofector X Unit (Lonza).

Total RNA was isolated after 16 to 24 hours of vehicle, single-agent, or combination treatment using the RNeasy Plus Mini kit (Qiagen, #74136). RNA was sequenced at the Dana-Farber Cancer Institute Molecular Biology Core Facility using the Illumina NextSeq 500 system, or with Novogene UC Davis Sequencing Center. Raw data were mapped to the hg38 genome using STAR (RRID:SCR_004463), and count files were generated using HTSeq (RRID:SCR_005514). DESeq2 (RRID:SCR_015687) was used to normalize gene counts (the mean ratio method) and determine differentially expressed genes. Differentially expressed genes between different treatment conditions (combo vs. DMSO, combo vs. EZH2i) were defined as log₂ fold change >1 and Benjamini–Hochberg corrected P value < 0.05.

ssGSEA was performed using GenePattern (http://genepattern.org/) using the normalized gene counts matrix. GSEA was performed using GSEA_4.2.3 software (RRID:SCR_003199) from http://gsea-msigdb.org/. Gene signatures from GO:BP were obtained from Molecular Signatures Database (RRID:SCR_016863; https://www.gsea-msigdb.org/gsea/msigdb). GO analysis was performed with differentially upregulated PRC2 targets using DAVID (RRID:SCR_001881; https://david.ncifcrf.gov/).

CUT&RUN was performed as previously described (61). In brief, 500,000 nuclei from LOVO cells treated with DMSO, 50 nmol/L trametinib, 5 μmol/L tazemetostat, or the combination of both drugs were isolated using nuclear extraction buffer (20 mmol/L HEPES pH 7.9, 10 mmol/L KCl, 0.1% Triton X-100, 20% glycerol, and 1 mmol/L MnCl₂). Nuclei samples were then immobilized to BioMag Plus Concanavalin A (ConA)–coated magnetic beads (Bangs Laboratories, #BP531) that were activated by washing three times with cold bead activation buffer (20 mmol/L HEPES pH 7.9, 10 mmol/L KCl, 1 mmol/L CaCl₂, and 1 mmol/L MnCl₂). ConA bead/cell mixtures were resuspended in cold antibody buffer (20 mmol/L HEPES pH 7.5, 150 mmol/L NaCl, 0.5 mmol/L spermidine, 1× Roche cOmplete, Mini, EDTA-free protease inhibitor, 0.01% digitonin, and 2 mmol/L EDTA), then incubated with 0.5 μg primary antibodies [H3K27me3 (CST, #9733, RRID:AB_2616029) and H3K4me3 (CST, #9751, RRID:AB_2616028) or IgG (Epicypher, #13-0042, RRID:AB_2923178)] overnight at 4°C in the cold room. Unbound antibodies were washed three times each with cold digitonin buffer (20 mmol/L HEPES pH 7.5, 150 mmol/L NaCl, 0.5 mmol/L spermidine, 1× Roche cOmplete, Mini, EDTA-free protease inhibitor, and 0.01% digitonin). ConA bead/cell mixtures were then resuspended in 50 μL cold digitonin buffer and incubated with our homemade pAG-MNase at 4°C in the cold room for an hour on a nutator. Unbound pAG-MNase was washed three times with cold digitonin buffer. MNase was activated by the addition of CaCl₂ and incubated at 4°C in the cold room for 30 minutes on a nutator to cleave and release antibody-bound chromatin. The reaction was stopped by adding cold stop buffer (340 mmol/L NaCl, 20 mmol/L ethylene glycol-bis(2-aminoethylether)-N,N,N′,N′-tetraacetic acid, egtazic acid (EDTA), 4 mmol/L EGTA, 50 μg/mL RNase A, 50 μg/mL glycogen, and 1 pg/μL E. coli spike-in DNA). Cleaved chromatin was then released by incubating at 37°C for 10 minutes. CUT&RUN enriched DNA in the supernatant can be collected using magnetic beads and purified using the Monarch PCR and DNA clean-up kit (NEB, #T1030L). CUT&RUN libraries were prepared with 10 ng CUT&RUN DNA using the NEBNext Ultra II DNA Library Prep Kit (NEB, #E7645), according to the manufacturer’s protocol. Libraries were sequenced on Illumina NextSeq2000, 2 × 50-bp paired-end reads.

Paired-end fastq files were aligned to hg38 reference genome using Bowtie2 (RRID:SCR_016368) with the setting “—very-sensitive –no-mixed –no-discordant –phred33 –I 10 –X 700”. 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 the E. coli genome was used to calculate the normalization factor for CUT&RUN samples. SAM files were converted to bam files using Samtools (RRID:SCR_002105). Bigwig files were generated from bam files using Deeptools with “scale-factor” option of bamCoverage for spike-in normalization (RRID:SCR_016366). Genome browser tracks of big files were generated using Integrative Genomics Viewer (IGV) (RRID:SCR_011793). Peaks were called using MACS2 (RRID:SCR_013291) using the callpeak function with “–f BAME –keep-dup 1 –q 0.05” and IgG was used as controls. DESeq2 analysis from DiffBind R package with option “spikein = True” was used to compare differential binding between conditions (FDR < 0.1).

For the PDX tumors, the tissue specimens to establish the colorectal cancer PDX models were collected in accordance with the Dana-Farber/Harvard Cancer Center Institutional Review Board protocol 03-189, where model generation and sharing of deidentified samples are covered under the consent. Six- to eight-week-old athymic Nu/Nu mice were purchased from Charles River Laboratory. LOVO xenografts were generated by injecting 3 million cells in PBS subcutaneously in the rear flanks of nude mice. For the PDX models, tumors were passaged from seed mice, cut into 2 × 2 × 2-mm chunks, and implanted subcutaneously in the rear flank of nude mice. For the LOVO cell line xenograft, two tumors per mouse were used. For the human PDX models, COCA9, COCA74P, COCA4, and COCA30, one tumor per mouse was used. When tumors reached 50 to 150 mm³, mice were randomized and enrolled into either vehicle or tazemetostat pretreatment arms. At least 5 to 10 tumors were used per treatment arm to detect a 50% change in tumor volume between the combo treatment arms and the single-agent arms with a power of at least 80% and FDR of 5%.

Mice were pretreated for 7 days and then randomized and enrolled into treatment arms: vehicle (tazemetostat vehicle with trametinib vehicle), trametinib (trametinib with tazemetostat vehicle), tazemetostat (tazemetostat with trametinib vehicle), or combination (tazemetostat with trametinib). Tazemetostat was prepared in 0.5% methylcellulose (FUJIFILM Wako Chemicals, #13317815) containing 0.1% Tween-80 and dosed at 250 mg/kg twice a day via oral gavage. Trametinib was prepared in 0.5% hydroxypropyl methylcellulose (Sigma, #H7509) containing 0.2% Tween-80, pH 8.0, and dosed at 0.6 mg/kg once a day via oral gavage. Tumor volume was measured using Vernier calipers two to three times a week and calculated with the formula (width × length² × 0.52).

Organoid lines were derived and cultured as previously described (33, 62). In brief, isolated crypts were resuspended in Matrigel (Corning, #356231), plated in six-well plates, and overlaid with growth medium comprising advanced DMEM/F12 (#12634010) supplemented with penicillin–streptomycin (#15070063); 10 mmol/L HEPES (#15630080); 2 mmol/L GlutaMAX (#35050061), N2 (#17502048), and B27 (#17504044; all from Gibco, Life Technologies); 100 ng/mL Noggin (PeproTech #250-38); and EGF (PeproTech #AF-100-15). As tumor organoid lines harbor Apc mutations, R-spondin and WNT3A were not supplemented. Organoids were dissociated into single cells with TrypLE Express (Gibco, #12604021) and seeded in 50 to 100 cells per μL of Matrigel. For counting experiments, organoids were seeded in 7 × 20 μL domes of Matrigel per well of a six-well plate. Cells were cultured in DMSO or 5 μmol/L of tazemetostat for 2 days, seeded into experimental plates with continued pretreatment of DMSO and tazemetostat, and allowed to form spheres for an additional 3 to 5 days before the addition of 50 nmol/L of trametinib. On days 0 and 3 after combination treatment, organoids were dissociated into single cells with TrypLE Express, spun down, resuspended in PBS, and manually counted in a hemocytometer. Final counts were normalized to respective day 0 counts, and log₂ fold change was calculated for graphical representation.

Human patient–derived organoids were cultured as described (63). In brief, human organoids were cultured in organoid growth media as described above, with the addition of N-acetylcysteine (NAC) (#A9165), nicotinamide (#N3376), SB202190 (#S7067), A83 (#SML0788), prostaglandin E (#P5640; all from Sigma Aldrich); WNT3A (R&D Systems, #5036-WN); R-spondin (PeproTech, #120-38); and FBS. Patient-derived organoids were dissociated into single cells, and 5,000 cells were seeded in 30 μL of Matrigel in a 96-well plate. Cells were cultured in DMSO or 5 μmol/L of tazemetostat for 5 days and then treated on day 0 with DMSO or trametinib for an additional 5 days. Cell number was indirectly assessed using the CellTiter-Glo 3D Viability Assay (Promega, #G9683) according to the manufacturer’s instructions on day 0 prior to the addition of trametinib and on day 5 after combination treatment.

Formalin-fixed, paraffin-embedded sections of mouse colons, CDX, and PDX samples were stained with a 15-plex antibody panel. High-plex images were acquired using the CyCIF technique (64). In brief, slides first underwent dewaxing and antigen retrieval using a preset protocol on the Leica BOND RX. Sections were then bleached with 4.5% H₂O₂ and 25 mmol/L NaOH in 1× PBS under LED light for 1 hour at room temperature. Subsequently, the slides were blocked overnight at 4°C in darkness with secondary antibodies (1:1,000 dilution) in SuperBlock blocking buffer (Thermo Fisher Scientific, #37515) to reduce autofluorescence and nonspecific secondary background. In each staining cycle, samples were incubated overnight at 4°C in darkness with Hoechst 33342 (1:10,000; Thermo Fisher Scientific, cat. # 62249) for nuclear staining and appropriate antibodies diluted in SuperBlock blocking buffer. Post-staining, the slides were washed and mounted/coverslipped with 50% glycerol/PBS for imaging on CyteFinder II HT (RareCyte), using a 20× 0.75NA objective. Between cycles, fluorophores were photobleached with the H₂O₂ solution as described earlier, and the slides were washed thoroughly before incubating antibodies for the next cycle. The multiplexed whole-slide images were processed via the Docker-based NextFlow pipeline, MCMICRO (65). Essentially, the raw images underwent stitching, registration, segmentation, and single-cell quantification of fluorescence intensities. The full codebase is available on GitHub (https://github.com/labsyspharm/mcmicro). Antibodies used were SOX9 (Abcam, #ab202516, 1:200, RRID:AB_2943105), KRT20 (CST, #63126, 1:200, RRID: AB_3101888), pERK (CST, #4344S, 1:200, RRID:AB_2139960), CDX2 (Abcam, #ab195008, 1:400, RRID:AB_2889213), panCK (eBioscience, #53-9003-82, 1:800, RRID:AB_1834350), and SMA (Invitrogen, #41-9760-82, 1:800, RRID:AB_2573631).

To identify relevant PRC2 targets, we analyzed genes that had differentially lost H3K27me3 binding between combo and DMSO-treated LOVO cells from the CUT&RUN data (adjusted P value < 0.05, FDR < 0.05, fold change < −1, and nearest distance to transcription start site annotation < 5 kb). We overlapped these genes with those that were differentially upregulated between combo and DMSO-treated LOVO cells from the RNA-seq data (adjusted P value < 0.05, fold change > 1, and base mean expression > 10).

β-Catenin–regulated signature is a generated set of 175 genes identified from RNA-seq experiments that were simultaneously downregulated by EZH2 and MEK inhibitors versus DMSO (log₂ fold change < −1) and upregulated or rescued in β-catenin Δ90 combo versus empty vector combo (log₂ fold change > 0.5). This signature represents β-catenin–regulated targets that are critical for the cytotoxic effects of EZH2 and MEK inhibitors in colorectal cancer cell lines.

For quantitative measurements in vitro, graphs depict the mean of the indicated number of technical replicates ± SD, unless otherwise noted. The in vivo tumor volume over time graphs represent the mean ± SEM. Statistical tests, including two-tailed unpaired t tests, ANOVA followed by Tukey multiple comparisons test, Mann–Whitney tests, or log-rank tests were used to compare experiment groups, and P values are indicated. A P value less than 0.05 was considered significant. All data were graphed and analyzed using GraphPad Prism v9.2.0 (RRID:SCR_002798).

All mouse work was done in compliance with the Institutional Animal Care and Use Committee at Brigham and Women’s Hospital and Animal Welfare Act (Approved Protocol #2016N000467). Written informed consent was obtained from patients for the use of their tumor specimens and clinical and genomic data. The study was conducted in accordance with recognized ethical guidelines (e.g., Declaration of Helsinki, CIOMS, Belmont Report, U.S. Common Rule) and was approved by the Dana-Farber/Harvard Cancer Center Institutional Review Board.

RNA-seq data for LOVO, SW620, SK-CO1, and LS513 cells treated with EZH2, MEK, and KRAS inhibitors are available under GEO database GSE265926. RNA-seq data for LOVO cells expressing empty vector or β-catenin Δ90 treated with EZH2 and MEK inhibitors are available under GEO database GSE265906. LOVO cells transduced with sgControl or sgTLE4 and treated with EZH2 and MEK inhibitors are available under GEO database GSE265907. H3K27me3 CUT&RUN data are available under BioProject PRJNA1104573.

P. Loi reports grants from the NIH/NCI during the conduct of the study. A.E. Schade reports grants from the American Cancer Society during the conduct of the study. M. Watanabe reports grants from Brigham and Women’s Hospital during the conduct of the study. O. Popow reports grants from Cancer Research UK and the Mark Foundation for Cancer Research during the conduct of the study. N. Gunduz reports grants from Cancer Grand Challenges during the conduct of the study. M. Giannakis reports grants from Janssen and personal fees from Nerviano Medical Sciences outside the submitted work. K. Ng reports nonfinancial support from Pharmavite, grants from Janssen, personal fees from Bayer, GlaxoSmithKline, Pfizer, CytomX, Jazz Pharmaceuticals, Revolution Medicines, AbbVie, Etiome, Seagen, CRICO, and JAMA outside the submitted work. S. Santagata reports being speaker honoraria for Novartis and Roche. K. Helin reports grants from the Cancer Research UK Grand Challenge and the Mark Foundation for Cancer Research to the SPECIFICANCER team during the conduct of the study. O.J. Sansom reports grants from AstraZeneca, Novartis, Boehringer Ingelheim, and Cancer Research Horizons outside the submitted work. K. Cichowski reports grants from Cancer Research UK and the Mark Foundation for Cancer Research during the conduct of the study, as well as other support from Erasca outside the submitted work. No disclosures were reported by the other authors.

P. Loi: Conceptualization, resources, data curation, formal analysis, methodology, writing–original draft, writing–review and editing, design. A.E. Schade: Conceptualization, resources, data curation, formal analysis, methodology, writing–original draft, writing–review and editing, design. C.L. Rodriguez: Resources, data curation, project administration. A. Krishnan: Resources, data curation, project administration. N. Perurena: Resources, data curation. V.T.M. Nguyen: Resources, data curation, formal analysis. Y. Xu: Resources, data curation, formal analysis. M. Watanabe: Resources, data curation. R.A. Davis: Resources, data curation. A. Gardner: Resources, data curation. N.F. Pilla: Resources, data curation. K. Mattioli: Formal analysis. O. Popow: Resources, data curation. N. Gunduz: Resources, data curation. T.R.M. Lannagan: Resources, data curation, project administration. S. Fitzgerald: Resources, data curation, project administration. E.T. Sicinska: Resources, data curation, project administration. J.-R. Lin: Resources, data curation, formal analysis. W. Tan: Project administration. L.K. Brais: Project administration. K.M. Haigis: Resources, data curation. M. Giannakis: Resources, data curation, writing–original draft, writing–review and editing. K. Ng: Resources, data curation. S. Santagata: Resources, data curation, supervision. K. Helin: Resources, data curation. O.J. Sansom: Resources, data curation, supervision. K. Cichowski: Conceptualization, resources, data curation, supervision, methodology, writing–original draft, writing–review and editing, design.

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. Cichowski, K.M. Haigis, O.J. Sansom, and K. Helin). P. Loi was supported by the NIH/NCI Ruth Kirschstein National Research Service Award (F31CA260804). A.E. Schade was supported by an American Cancer Society—Cancer Research Racquet and Women’s Tennis Association—Postdoctoral Fellowship, PF-22-040-01-ET. A. Gardner was supported by a NIH/NCI Predoctoral to Postdoctoral Fellow Transition Award (K00CA245807).