TEAD-targeting small molecules induce a cofactor switch to regulate the Hippo pathway

Significance

TEAD proteins are the transcriptional effectors of the Hippo pathway and are a promising target for cancer therapy, with several TEAD inhibitors advancing in the clinic. Drug-like TEAD inhibitors are developed with the aim to disrupt TEAD protein–protein interactions with the transcriptional activators YAP and TAZ. Here, we identify an alternative, unexpected mechanism of action for a subset of TEAD inhibitors. We find that some TEAD-targeting molecules act as molecular glues by bringing together TEAD and its repressive cofactor, VGLL4. Chemically induced VGLL4–TEAD complexes repress transcriptional networks to reduce cancer growth. Our research reveals that TEAD inhibitors can act through multiple mechanisms, and spotlights a class of glue-like small molecules that can repress transcription.

Abstract

TEAD proteins are the main transcriptional effectors of the Hippo signaling pathway and a clinical-stage pharmacological target in oncology. Most TEAD-targeting small molecules are designed to disrupt interaction between TEAD and the oncogenic transcriptional activators YAP and TAZ. Here, we uncover an alternative mechanism for a subset of TEAD lipid pocket-binding molecules. We report that select sulfonamide-containing TEAD-targeting compounds enhance the interaction between TEAD and the transcriptional repressor VGLL4. Chemically induced VGLL4–TEAD complexes confer an antiproliferative effect by outcompeting YAP–TEAD complexes at chromatin. This cofactor switch from YAP to VGLL4 impacts transcriptional networks, including influencing the expression of genes involved in cellular proliferation and mechanosignaling. We demonstrate that VGLL4 is required for an antiproliferative response to these sulfonamide-containing compounds by counteracting YAP. We show that VGLL4 overexpression can confer sensitivity to these compounds in Hippo-driven cell lines, and we further show that genetic deletion of VGLL4 oblates cellular responsiveness to these molecules in cells and in vivo. Our data reveal a category of TEAD inhibitors that act as “molecular glues” toward the repressive VGLL4–TEAD interaction. These findings open up understandings for curbing the oncogenic activity of Hippo pathway deregulation in cancer, and identify glue-like molecules that promote transcriptional repression.

Initially discovered in Drosophila, the Hippo pathway is now realized as a key regulatory mechanism in human development and disease, including cancer (1). The Hippo signaling pathway is a crucial cellular regulator, exerting tumor suppressive activity to tightly control cellular growth, proliferation, and differentiation. The core activity of the Hippo pathway involves upstream factors, including MST1/2, NF2, and LATS1/2, which promote phosphorylation of YAP and TAZ to prevent their nuclear translocation and transcriptional activity with the TEAD family of transcription factors (TEAD1-4). In cancer, deregulation of the Hippo pathway is prominent both as a driver of tumor development [for example NF2 mutations (2) and inactivation of LATS kinases (3)] and as a mechanism of acquired resistance to targeted therapies (4–7). Therefore, modalities that can effectively reactivate the tumor-suppressive function of the Hippo pathway hold significant therapeutic potential.

Understanding how the Hippo pathway functions is essential to targeting it effectively. The key output of the Hippo pathway is transcriptional regulation mediated by TEAD transcription factors. Because TEAD proteins have minimal inherent transcriptional activity, TEAD relies on interactions with other proteins to influence chromatin. Interaction partners include other transcription factors (8, 9), chromatin remodelers (10), and DNA damage repair proteins (11, 12), as well as coactivators and corepressors. TEAD coactivators and corepressors do not have DNA binding domains and thus depend on direct interaction with TEAD to engage with chromatin. The three TEAD cofactors YAP, TAZ, and VGLL4 are notable because they have mutually exclusive, overlapping binding sites on TEAD (13) and engage TEAD to influence a transcriptional balancing act of activation and repression (14–16). YAP and TAZ are TEAD coactivators (1, 17, 18) and can be oncogenic, as they are frequently observed to have high nuclear levels and activity in cancers (4, 18, 19). In contrast, VGLL4 is a TEAD corepressor and is tumor suppressive (16, 20–22). Low expression of VGLL4 is associated with poor prognosis in a variety of malignancies (14, 15, 23–26). Proof-of-concept experiments using VGLL4-mimicking peptides have demonstrated that leveraging the VGLL4–TEAD interaction can counteract YAP and suppress tumor growth (15, 27, 28). However, the extent to which VGLL4–TEAD can be modulated by drug-like small molecules remains largely unknown.

TEAD small molecule inhibitors represent the most prominent strategy for targeting deregulated Hippo signaling in cancer (13, 29). Unlike YAP and TAZ which are intrinsically disordered proteins, TEAD proteins contain a druggable hydrophobic pocket (30, 31). This site is termed the lipid pocket because it is posttranslationally occupied by a lipid modification, S-palmitoylation of a conserved cysteine (32). Small molecules have been designed to specifically engage this pocket with the objective of allosterically disrupting the YAP–TEAD interaction to repress oncogenic transcriptional programs (13, 29, 30). Development of TEAD lipid pocket-binding (LPB) therapeutic candidates is advancing rapidly, with only 6 y between the discovery of this druggable pocket (30, 31, 33) and the entrance of such molecules in the clinic (NCT05228015, NCT04665206). With this rapid advancement, deeper biological insight is needed to understand precisely how chemical engagement at the lipid pocket modulates TEAD, particularly regarding the impact of inhibitory molecules on TEAD protein interactions. Although many TEAD-targeting molecules potently inhibit the YAP–TEAD interaction (4, 29, 34, 35), paradoxically, some molecules inhibit TEAD palmitoylation and repress Hippo target gene expression without disrupting the YAP–TEAD interaction (29, 36–38). If disruption of YAP is not necessary for efficacy of TEAD inhibitors, other mechanistic forces are likely at play.

Here, we challenge the idea that TEAD LPB inhibitors exclusively disrupt the YAP/TAZ–TEAD interaction. Using a global proteomic approach, we examine changes in TEAD interactions upon LPB inhibitor treatment. Our findings reveal a class of TEAD-targeting molecules that enhances the VGLL4–TEAD interaction and shed light on the repressive transcriptional effects of VGLL4. We show that an enhanced VGLL4–TEAD axis can inhibit YAP oncogenic activity, providing an alternative antiproliferative mechanism for TEAD LPB inhibitors. Overall, we detail how small molecules can induce a cofactor switch to modulate TEAD transcriptional activity.

Results

Lipid Pocket Binding Compound, Compound 2, Increases the Interaction of TEAD and VGLL4.

To better understand the cellular effects of TEAD LPB inhibitors, we used the Hippo-driven mesothelioma cell line NCI-H226 (4, 34, 35, 39, 40). We focused on two characterized pan-TEAD LPB small molecules: GNE-7883 (lipid pocket affinity of ~330 nM) (4) and Compound 2 (lipid pocket affinity of ~20 nM) (36) (Fig. 1A). We chose these molecules because they differ in their impact on the YAP/TAZ–TEAD interaction: GNE-7883 allosterically blocks the interaction between TEAD and YAP/TAZ (4), while Compound 2 does not (36). Despite the difference, both molecules decrease the cellular viability of NCI-H226 cancer cells with nanomolar EC₅₀ values (Fig. 1B), similar to what has previously been reported (4, 36).

Fig. 1.

Identification of TEAD interaction partners that are sensitive to LPB TEAD inhibitors. (A) Structures of GNE-7883 and Compound 2. (B) Viability dose–response curves (mean ± SD) for GNE-7883 and Compound 2 in NCI-H226 cells treated for 6 d. n = 3 biological replicates. (C) Schematic of experimental workflow for TEAD-FLAG affinity purification mass spectrometry (AP-MS) using tandem mass tags (TMT). n = 3 biological replicates. (D) Volcano plot of TMT proteomic results for TEAD4 AP-MS comparing GNE-7883 treatment to DMSO vehicle control. The TMT-quantified fold-change is plotted against the adjusted P-value from model-based testing using MSstatsTMT. Proteins meeting a cutoff of adj P-value <0.05 and twofold change are highlighted. (E) Volcano plot for TEAD4 AP-MS as in (D) but for Compound 2 treatment compared to DMSO. (F) Endogenous pan-TEAD coimmunoprecipitation (coIP) assays in NCI-H226 cells. Cellular extracts were subjected to IP with pan-TEAD antibody or an immunoglobulin G (IgG) control. IP samples were probed for coprecipitating endogenous proteins by immunoblot (IB). Inputs are 10% for pan-TEAD and 1% for others. n = 5 biological replicates. Experiment is quantified in SI Appendix, Fig. S1G. (G) Endogenous pan-TEAD was recovered from lysates of the indicated cell lines treated for 24 h and probed for coprecipitating VGLL4, YAP, and TAZ by IB. Inputs for pan-TEAD are 10 and 1% for others. n = 2 or more biological replicates per cell line. Experiments are quantified in SI Appendix, Fig. S1 H–M. See also SI Appendix, Fig. S1 and Datasets S1 and S2.

To learn how these small molecules influence TEAD biology we took a quantitative proteomic approach using AP-MS. We used TEAD1 or TEAD4 as baits because these are the TEAD paralogs that are most highly expressed in NCI-H226 cells (SI Appendix, Fig. S1A). We generated stable cell lines that express inducible, FLAG-tagged forms of TEAD1 or TEAD4, treated the cells with small molecules, recovered proteins by FLAG immunoprecipitation (IP), and analyzed the samples by mass spectrometry (Fig. 1C and SI Appendix, Fig. S1B). GNE-7883 treatment reduced the association of 16 proteins with TEAD4, including YAP, AMOTL2, MPDZ, IGFBP3, and CCN1 (Fig. 1D). Compound 2 treatment decreased the association of TEAD4 with IGFBP3, AMOTL2, and MPDZ, and increased the association of TEAD4 with VGLL4 (Fig. 1E). YAP also decreased with Compound 2 treatment but to an extent that did not reach the significance threshold. Similar AP-MS results were obtained both with TEAD4 (Fig. 1 D and E and Dataset S1) and with TEAD1 (SI Appendix, Fig. S1 C and D and Dataset S2).

The most significantly changed protein in our AP-MS experiments upon treatment with Compound 2 was VGLL4, displaying a five-fold increased association with TEAD4 (SI Appendix, Fig. S1E) and a sevenfold increased association with TEAD1 (SI Appendix, Fig. S1F). Increased VGLL4–TEAD interaction was not observed with GNE-7883 treatment. Endogenous coIP in NCI-H226 cells using a pan-TEAD antibody recovered more VGLL4 when the cells were treated with Compound 2 (Fig. 1F and SI Appendix, Fig. S1G). And coIP in six additional cell lines—three with functional YAP and TAZ, and three with nonfunctional YAP and TAZ—also confirmed a boosted interaction between TEAD and VGLL4 upon treatment with Compound 2 (Fig. 1G and SI Appendix, Fig. S1 H–M). Thus, interaction between TEAD and VGLL4 can be induced with Compound 2 in a variety of cell types, irrespective of YAP/TAZ presence.

Sulfonamide-Containing Compounds Promote TEAD Binding to the TDU2 Domain of VGLL4.

VGLL4 is a direct interacting partner for TEAD that antagonizes YAP/TAZ binding (15, 20, 41) but is rarely assayed during the development of TEAD inhibitors. To investigate the molecular determinants that enable Compound 2 to induce the VGLL4–TEAD interaction, we first examined four published TEAD LPB compounds, MGH-CP1 (42), VT-107 (34), TED-347 (43), and K-975 (35) (SI Appendix, Fig. S2A), and found that none of these compounds induced a change in VGLL4–TEAD association (SI Appendix, Fig. S2B). We next focused on the chemical backbone of Compound 2 and synthesized four additional analogs that engage the lipid pocket of TEADs: N1, N2, S1, and S2 (Fig. 2A and SI Appendix, Fig. S2C and Methods S1). Remarkably, compounds S1 and S2, but not N1 or N2, improve recovery of VGLL4 to levels comparable to that induced by Compound 2 (Fig. 2B and SI Appendix, Fig. S1D). A distinct feature of Compound 2, S1, and S2 is a sulfonamide group appended to the pyridine ring. We next assayed two additional LPB compounds, VT-103 which contains a sulfonamide, and VT-104 which does not contain a sulfonamide (Fig. 2C) (34). Consistently, sulfonamide-containing VT-103 increased VGLL4 recovery to levels comparable to or higher than Compound 2, while sulfonamide-free VT-104 did not impact VGLL4 recovery (Fig. 2D and SI Appendix, Fig. S2E). We also synthesized three additional compounds based on Compound 2 by replacing the sulfonamide group: C2 amino, C2 methyl, and C2 acetyl (SI Appendix, Fig. S2F and Methods S1). However, replacing the sulfonamide caused a loss in affinity for TEAD proteins (SI Appendix, Fig. S2G), and we observed no change in VGLL4 recovery by coIP in two cell lines (SI Appendix, Fig. S2 H and I). Altogether, these results uncover the sulfonamide group as a common feature for the TEAD LPB compounds that promote the VGLL4–TEAD interaction.

Fig. 2.

Sulfonamide-containing LPB compounds facilitate the VGLL4–TEAD interaction. (A) Structures of tool compounds without (N1 and N2) and with (S1 and S2) a sulfonamide functional group. Sulfonamide groups are highlighted in red. (B) CoIP experiment to assay TEAD compounds in NCI-H226 cells. Inputs are 10% for pan-TEAD and 1% for others. n = 3 biological replicates. Experiment is quantified in SI Appendix, Fig. S2D. (C) Structures of Compound 2, VT-103, and VT-104. Sulfonamide groups are highlighted in red. (D) CoIP experiments performed as in (B) but treated with the compounds in (C). n = 3 biological replicates. Experiment is quantified in SI Appendix, Fig. S2E. (E) Amino acid sequence alignment of human VGLL protein family members, VGLL1 (UniProt Q99990), VGLL2 (UniProt Q8N8G2), VGLL3(UniProt A8MV65), and VGLL4 (UniProt Q14135). TONDU (TDU) domain regions are designated in green above the sequences. (F) Schematic illustrating the biotinylated (bio.) VGLL4 peptides (residues 202-259) used in the pulldown assays. The mutations made in TDU1 are H212A and F213A, and the mutations made in TDU2 are H240A and F241A. (G) In vitro peptide pulldown assay. Recombinant TEAD1 (S210-D426) was preincubated with compounds or DMSO and then incubated with the indicated biotinylated VGLL4 peptides. Recovered TEAD1 was visualized by Coomassie staining. Peptide input was visualized by dot blot. n = 3 biological replicates. Experiment is quantified in SI Appendix, Fig. S2J. (H) Table of SPR affinity measurements for recombinant TEAD1 (S210-D426) and the VGLL4 peptides depicted in 2F treated with the indicated compounds or DMSO. Mean ± SD from 4 to 6 independent experiments. (I) Normalized NanoBRET assay with overexpressed wildtype TEAD and VGLL4 in HEK293T cells with a titration of the indicated compounds. Mean ± SEM; n = 3 biological replicates. See also SI Appendix, Figs. S2 and S3 and Methods S1.

We next turned to assess what protein features contribute to the gained VGLL4–TEAD interaction. VGLL4 is unique among the four VGLL protein family members in that VGLL4 contains not one but two TEAD-interacting TONDU (TDU) domains (21) (Fig. 2E). To investigate the TDU domains of VGLL4, we individually mutated each TDU domain and performed a peptide pulldown assay using a minimal in vitro system. Using peptides of VGLL4 (Fig. 2F) and recombinant TEAD1 (C-terminal domain), we observed increased recovery of TEAD1 with VT-103, but little change with Compound 2 or GNE-7883 (Fig. 2G and SI Appendix, Fig. S2J). When we mutate the second TDU domain (TDU2), or both TDU domains, of VGLL4, the interaction is lost almost entirely. We next measured the VGLL4–TEAD interactions using surface plasmon resonance (SPR). First, we observed that C-terminal TEAD1 binds to WT VGLL4 peptide with an affinity of ~500 nM. When we introduced GNE-7883, the VGLL4–TEAD1 K_D decreased ~2-fold, and Compound 2 had no effect on the affinity (Fig. 2H and SI Appendix, Fig. S3 A–M). However, when we introduced VT-103 we observed an approximately 12-fold increase in affinity for the VGLL4–TEAD1 interaction (Fig. 2H and SI Appendix, Fig. S3N) and a slowed off-rate (SI Appendix, Fig. S3 I and M). Results of mutating the first TDU domain of VGLL4 were similar to WT VGLL4, but when we mutated the second TDU domain or both TDU domains simultaneously we did not detect VGLL4–TEAD1 binding (Fig. 2H and SI Appendix, Fig. S3 C and D). These SPR results are in line with the results of the peptide pulldown experiment.

Both in the peptide pulldown experiments and in the SPR experiments we used a truncated recombinant TEAD1 protein (only the C-terminal binding domain) and a 58-mer peptide of VGLL4. It is interesting that in this truncated, simplified system we did not detect an enhanced TEAD–VGLL4 interaction with Compound 2. To expand our biophysical interrogation to full-length proteins in a cellular environment, we performed a NanoBRET assay using tagged, full-length TEAD1 and WT VGLL4. By NanoBRET assay we observed an enhanced interaction of TEAD1 and VGLL4 with both Compound 2 and with VT-103, but not with the sulfonamide-free compounds GNE-7883 or VT-104 (Fig. 2I). We did not observe a change in NanoBRET signal with the TDU1+TDU2 mutant (SI Appendix, Fig. S3O). Based on these data, we conclude that the full-length proteins have a crucial role in mediating the molecular glue effect of Compound 2 that is missing from the recombinant experiments. Notably, the cocrystal structure of truncated TEAD in complex with Compound 2 shows no prominent changes in the overall fold of TEAD (SI Appendix, Fig. S3P) (36). Perhaps full-length TEAD is required to elucidate these structural details. Altogether, these results indicate that sulfonamide-containing LPB TEAD compounds directly modulate the VGLL4–TEAD interaction via the TDU2 motif of VGLL4.

Compound 2 Treatment Phenocopies VGLL4 Overexpression and Counteracts YAP.

VGLL4 counteracts YAP–TEAD transcriptional complexes to exert growth-inhibitory transcriptional repression (15, 16, 21). To determine how cell growth and transcription are affected by enhancing the VGLL4–TEAD interaction, we compared chemical and genetic modulation of VGLL4 using Compound 2 and VGLL4 overexpression, respectively. We used Compound 2 because, unlike VT-103 which is TEAD1-specific (34, 44), Compound 2 is a pan-TEAD molecule. We engineered NCI-H226 cells to inducibly overexpress HA-tagged VGLL4, either wildtype or the TDU1+TDU2 mutant VGLL4 that is unable to interact with TEAD (Fig. 3 A and B and SI Appendix, Fig. S4A) (15, 21, 45). Doxycycline treatment efficiently induced VGLL4 overexpression (Fig. 3B). Simultaneous WT VGLL4 overexpression together with Compound 2 treatment had an additive effect on promoting VGLL4–TEAD association and was accompanied by a concomitant decrease in YAP–TEAD association (SI Appendix, Fig. S4B), consistent with VGLL4 antagonizing YAP–TEAD complexes. We monitored cellular growth over 10 d, and gene expression at 24 h by probing five representative Hippo target genes: CTGF (CCN2), CYR61 (CCN1), ANKRD1, F3, and IGFBP3. Both with Compound 2 treatment (Fig. 3C) and with WT VGLL4 overexpression (Fig. 3D), cellular growth rates slowed and target gene expression decreased. In contrast, overexpression of the TDU1+TDU2 mutant VGLL4 resulted in little to no change in proliferation or gene expression (Fig. 3E). We conclude that Compound 2 treatment is analogous to WT VGLL4 overexpression and that the VGLL4–TEAD interaction can promote an antiproliferative phenotype.

Fig. 3.

Compound 2 treatment phenocopies VGLL4 overexpression to counteract YAP. (A) Schematic of VGLL4. Amino acid sequences of the two TDU motifs are detailed for WT and the TDU1+TDU2 mutant. The four amino acid substitutions that comprise TDU1+TDU2 mutant VGLL4 are highlighted in pink (H212A, F213A, H240A, F241A). (B) Immunoblots of doxycycline (dox) induced overexpression of WT and TDU1+TDU2 mutant VGLL4 in NCI-H226 cells. (C) Characterization of the impact of Compound 2 treatment on transcription and proliferation in NCI-H226 cells. Cells were cultured in 3 µM Compound 2. For gene expression analysis, cells were assayed after 24 h; n = 3 independent biological replicates, mean ± SEM, ***P < 0.001, **P < 0.01, *P < 0.05 by the parametric unpaired two-tailed t test. For proliferation, cells were imaged every 6 h over 10 d; mean ± SEM, n = 3 independent biological replicates. (D) Characterization of the impact of WT VGLL4 overexpression on transcription and proliferation in NCI-H226 cells. Cells were treated with dox to induce overexpression. Data were collected and are represented as in (C). (E) Characterization of the impact of TDU1+TDU2 mutant VGLL4 overexpression on transcription and proliferation in NCI-H226 cells. Cells were treated with dox to induce overexpression. Data were collected and are represented as in (C). (F) Principal component analysis (PCA) performed on RNA-seq experiments in WT VGLL4 and TDU1+TDU2 mutant VGLL4 cell contexts. Treatment conditions are DMSO vehicle control, Compound 2 (3 µM for 24 h), and VGLL4 overexpression. (G) Diagram showing the overlap between gene expression changes that are decreased at 24 h both with Compound 2 treatment and with WT VGLL4 overexpression. (H) Diagram showing the overlap between gene expression changes that are increased at 24 h both with Compound 2 treatment and with WT VGLL4 overexpression. (I) Enriched Hallmark gene sets determined by GSEA of RNA-seq from 24-h Compound 2 treatment or WT VGLL4 overexpression. Twelve top enriched gene sets are shown. (J) Number of transcripts significantly (false discovery rate [FDR] < 0.05) altered by 24 h treatment of cells with GNE-7883 compared with DMSO control. n = 3 biological replicates. (K) Diagram showing the overlap between gene expression changes that are decreased at 24 h both with GNE-7883 treatment and with VGLL4 activity (intersection from G). (L) Diagram showing the overlap between gene expression changes that are increased at 24 h both with GNE-7883 treatment and with VGLL4 activity (intersection from H). (M) Transcript level changes for 52 Hippo target genes upon the indicated treatments as measured by RNA-seq. FDR is represented by size and fold change is represented by color. See also SI Appendix, Figs. S4 and S5 and Table S1 and Dataset S3.

To profile the global transcriptional changes that occur with Compound 2 treatment and with VGLL4 overexpression, we performed RNA sequencing (RNA-seq) (SI Appendix, Fig. S4C). To distinguish changes that depend on interaction with TEAD, we also analyzed the TDU1+TDU2 mutant VGLL4 context. Compound 2 treatment or WT VGLL4 overexpression yielded thousands of significantly changed transcripts, a majority of which were less than two-fold in magnitude (SI Appendix, Fig. S4 D and E). In contrast, overexpression of the TDU1+TDU2 mutant VGLL4 induced a mere 9 transcripts, one of which is VGLL4 (SI Appendix, Fig. S4F). Principal component analysis (PCA) revealed similar clustering of the WT VGLL4 samples for Compound 2 treatment and for VGLL4 overexpression (Fig. 3F), indicating similar transcriptional changes. Indeed, we observed 84% (2,062 of 2,454) overlap of decreased transcripts (Fig. 3G) and 81% (2,154 of 2,649) overlap of increased transcripts (Fig. 3H). Gene set enrichment analysis (GSEA) and Reactome pathway analyses further emphasized similarity between WT VGLL4 overexpression and Compound 2 treatment (Fig. 3I and SI Appendix, Fig. S4 G and H). For both conditions we observed negative scoring of gene categories related to cellular growth and proliferation (such as E2F targets, MYC targets, and G2M checkpoint, Cell Cycle) and positive scoring for gene categories related to inflammatory signaling (such as Interferon alpha response, Inflammatory response, Immune system).

Finally, we asked how the transcriptional effects we observe upon chemical or genetic induction of VGLL4 compared with inhibition of the YAP–TEAD interaction by GNE-7883 treatment (4). Treatment of NCI-H226 cells with GNE-7883 resulted in ~4,000 significantly increased transcripts and ~4,000 significantly decreased transcripts (false discovery rate [FDR] < 0.05, n = 3) (Fig. 3J). We found ~80% overlap between the VGLL4 consensus transcriptional changes (Fig. 3 G and H) and the GNE-7883 transcriptional changes, (Fig. 3 K and L), indicating that activating VGLL4 chemically or genetically exerts a similar transcriptional effect to disrupting YAP. To examine this trend further, we zoomed in on a set of 52 YAP/TAZ and TEAD target genes defined previously (46) (SI Appendix, Table S1) and observed decreased expression for many of the key Hippo pathway-regulated genes with GNE-7883, Compound 2, and WT VGLL4 overexpression, but not with overexpression of the TDU1+TDU2 mutant VGLL4 (Fig. 3M). We also performed pathway analysis to assess how GNE-7883 and VGLL4 may influence transcriptional patterns in unique ways (SI Appendix, Fig. S5 and Dataset S3). We observe some redundancy, such as mitosis-related categories, as well as some distinctly modulated signaling pathways that suggest that promoting VGLL4–TEAD complexes results in a different set of cellular compensatory signaling pathways than disrupting YAP–TEAD complexes. Overall, these transcriptional trends indicate that YAP–TEAD transcriptional activity can be attenuated by VGLL4 and that a TEAD cofactor switch to VGLL4 can be activated by small molecules like Compound 2.

TEAD Sulfonamide LPB Compounds Promote VGLL4 at Chromatin.

Given that VGLL4 binds to TEAD and influences transcriptional outputs, we next asked how boosting the VGLL4–TEAD interaction alters VGLL4 subcellular localization. We performed biochemical subcellular fractionations in cells treated with TEAD LPB compounds. Distributions of TEAD did not change (Fig. 4A and SI Appendix, Fig. S6A) but there were changes in the distributions of VGLL4, YAP, and TAZ. Most notably, in the chromatin-associated fractions, VGLL4 protein levels increased (Fig. 4A and SI Appendix, Fig. S6B), and YAP protein levels decreased (SI Appendix, Fig. S6C) upon treatment with Compound 2 or VT-103. At the same time, TAZ levels increased slightly both in the chromatin-associated fraction and at the whole cell level (SI Appendix, Fig. S6D), indicating an overall redistribution of TEAD cofactors. We also assayed VGLL4 chromatin localization by immunofluorescence. Pretreating cells with Triton X-100 detergent prior to fixation permeabilized the cells, removed soluble proteins, and enabled us to image residual chromatin-bound proteins (47). Because antibodies to endogenous VGLL4 do not give a specific immunofluorescence signal, we used VGLL4-2xHA expressing cells (SI Appendix, Fig. S6E). Altogether, treatment with Compound 2 or VT-103 resulted in a significant increase in chromatin-associated VGLL4 (Fig. 4 B and C).

Fig. 4.

TEAD sulfonamide LPB compounds promote chromatin-bound VGLL4. (A) NCI-H226 cells were treated for 24 h and then fractionated into cytoplasmic, soluble nuclear, and chromatin-associated fractions. Equal amounts of each fraction were analyzed by immunoblotting with the antibodies against the indicated proteins. n = 3 biological replicates. Experiment is quantified in SI Appendix, Fig. S6 A–D. (B) Immunofluorescence staining following triton extraction of soluble proteins. (Scale bar, 100 µm.) (C) Quantification of triton extraction immunofluorescence experiment in (B). Data are plotted as mean with 95% CI, analyzed by ordinary one-way ANOVA, ** adj P-value = 0.0012, *** adj P-value = 0.0008. Note that the y-axis begins at 150. n = 2,000 randomly selected cells for each condition. Data are representative of three replicates. (D) Heatmaps for ChIP-seq peaks in cells expressing empty vector control or VGLL4-2xHA. (E) De novo motif analysis by HOMER for regions bound by HA-tagged VGLL4. (F) Plot of the peak distribution of the 54,979 VGLL4-HA peaks according to their distance from the nearest annotated transcription start site (TSS). (G) Boxplot showing the ChIP-seq read count signal for the three treatment conditions and assessed by the Wilcoxon signed-rank test. n = 3 experiments. (H) Genome browser tracks of VGLL4-HA ChIP-seq at the F3 locus. (I) Genome browser tracks of VGLL4-HA ChIP-seq upstream of the LAMA1 locus. See also SI Appendix, Fig. S6.

To track the genomic binding sites of VGLL4 in NCI-H226 cells, we performed chromatin immunoprecipitation sequencing (ChIP-seq) analysis upon Compound 2 treatment. We monitored VGLL4 using the inducible HA-tagged system, where the levels of epitope-tagged VGLL4 were comparable to endogenous levels of VGLL4 (SI Appendix, Fig. S6F). We identified 54,979 peaks of VGLL4 binding that were specific to the HA epitope-expressing cells (FDR < 0.01, n = 3) (Fig. 4D and SI Appendix, Fig. S6G). Binding sites were enriched primarily for TEAD motifs, as well as AP-1 and GATA sequences (Fig. 4E), and the majority of VGLL4 peaks (>70%) were more than 10 kb from the transcription start site of genes, localized in intergenic or intronic regions (Fig. 4F and SI Appendix, Fig. S6H). Upon treatment with Compound 2, we detected enhanced VGLL4 chromatin binding (Wilcoxon Signed-Rank test, P < 0.0001; Fig. 4 G–I and SI Appendix, Fig. S6I). Overall, these data indicate that the transcriptional influence of Compound 2 stems from facilitating the VGLL4–TEAD interaction at chromatin.

VGLL4 Overexpression Can Sensitize Cells to Sulfonamide LPB Inhibitors.

We next investigated how VGLL4 protein levels impact the cellular response to TEAD LPB compounds in other cancer cell models. We mined publicly available expression data from DepMap and found that NCI-H226 cells have higher RNA expression of VGLL4 than most other cell lines (SI Appendix, Fig. S7A). We selected five cell types with genetic alterations in the Hippo pathway to cross-compare with NF2-null NCI-H226 cells. NCI-H290 (pleural mesothelioma) and MDA-MB-231 (breast adenocarcinoma) are NF2-null (4, 46), MSTO-211H (pleural mesothelioma) is LATS-deficient (48), OVCAR-8 (ovarian serous adenocarcinoma) is YAP-amplified (46), and QGP-1 (pancreatic somatostatinoma) is a YAP/TAZ-nonexpressing cell line (Fig. 1G) (49) which we selected as a Hippo-independent control. Immunoblotting of lysates showed that VGLL4 protein was notably higher in NCI-H226 cells compared to these other cell types (Fig. 5A). We next assayed the sensitivity of these cell lines to TEAD LPB compounds. Although all the Hippo-dependent cell lines were sensitive to GNE-7883 (SI Appendix, Fig. S7B), only NCI-H226 cells responded appreciably to Compound 2 and VT-103 (Fig. 5B). As expected, the YAP/TAZ-deficient QGP-1 cells were insensitive to all three TEAD compounds (Fig. 5B and SI Appendix, Fig. S7B). The observation that cell lines with more VGLL4 are more sensitive to Compound 2 and VT-103 further implicates VGLL4 as the determining factor required for mediating an anticancer response.

Fig. 5.

VGLL4 overexpression sensitizes Hippo-dependent cell lines to sulfonamide LPB compounds. (A) Immunoblots of lysates from the six indicated cell lines. (B) Viability dose–response curves (mean ± SD) for Compound 2 (Left) and VT-103 (Right) in the six indicated cell lines treated for 6 d. n = 3 biological replicates. (C) Immunoblots of lysates from cell lines engineered with inducible VGLL4 overexpression for the five indicated cell types with and without doxycycline (Dox) induction. The TDU1+TDU2 VGLL4 mutant used in these experiments is detailed in Fig. 3A. (D) Viability dose–response curves (mean ± SD) for Compound 2. Viability assays were performed with uniform doxycycline treatment. n = 3 biological replicates. (E) Viability dose–response curves (mean ± SD) for VT-103. Viability assays were performed with uniform doxycycline treatment. n = 3 biological replicates. See also SI Appendix, Fig. S7.

To further test the idea that sensitivity to Compound 2 and VT-103 is linked to VGLL4 protein levels, we next overexpressed VGLL4. We created stable cell lines by introducing inducible VGLL4 overexpression constructs, WT, or the TEAD interaction-deficient TDU1+TDU2 mutant (Figs. 3A and 5C and SI Appendix, Fig. S7C). When we overexpressed WT VGLL4 the previously insensitive OVCAR-8, NCI-H290, MDA-MB-231, and MSTO-211H cells showed dose–responses to Compound 2 (Fig. 5D) and to VT-103 (Fig. 5E). These cell types remained insensitive with the vector control and with overexpression of the TDU1+TDU2 mutant. In contrast, the YAP/TAZ-deficient cell type, QGP-1, remained consistently insensitive in all conditions. We also examined NCI-H226 cells, which were already sensitive to Compound 2 and VT-103, and found that overexpression of WT VGLL4 did not further sensitize cells to Compound 2 or VT-103 (SI Appendix, Fig. S7D), indicating that endogenous VGLL4 levels exert maximal repressive activity in this cell type. Altogether, these results reveal VGLL4 as the critical factor through which these sulfonamide-containing TEAD-targeting molecules exert anticancer efficacy and indicate that YAP/TAZ activity is necessary to elicit an antiproliferative response.

VGLL4 Knockout Confers Resistance to Compound 2 and VT-103 in NCI-H226 Cells.

Given the antiproliferative effects of inducing the VGLL4–TEAD interaction, we next asked how NCI-H226 cells respond to the opposite perturbation, deletion of VGLL4. CRISPR targeting of the VGLL4 locus (Fig. 6A) resulted in efficient cellular depletion of VGLL4 protein (Fig. 6B), with no influence upon cell growth rate (SI Appendix, Fig. S8A). A pooled transfected population of cells showed similar VGLL4 depletion compared to a clonal population (SI Appendix, Fig. S8 B and C), and, to avoid clone-specific effects, we performed all subsequent experiments with the VGLL4 knockout (KO) pool. Parental NCI-H226 cells are acutely sensitive to VT-103 and Compound 2 (Fig. 6C and SI Appendix, Fig. S8D). But remarkably, when we knocked out VGLL4, the cells became resistant to these sulfonamide-containing compounds (Fig. 6D and SI Appendix, Fig. S8E) while remaining sensitive to the nonsulfonamide compounds VT-104 and GNE-7883 (Fig. 6 C and D and SI Appendix, Fig. S8F).

Fig. 6.

VGLL4 is necessary for the anticancer activity of sulfonamide LPB compounds. (A) Schematic of the CRISPR knockout strategy targeting VGLL4. (B) Immunoblots of lysates from parental NCI-H226 cells and from a knockout pool of NCI-H226 cells. (C) Proliferation (mean ± SEM) of NCI-H226 parental cells grown in the presence of 3 µM compound or DMSO. n = 3 experiments. (D) Proliferation (mean ± SEM) of NCI-H226 VGLL4 knockout (KO) cells grown in the presence of 3 µM compound or DMSO. n = 3 experiments. (E) Immunoblots of coIP assays from parental and VGLL4 KO cells. Cells were treated for 24 h with 3 µM of the indicated compounds or DMSO vehicle control and cellular extracts were subjected to IP with pan-TEAD antibody or an IgG control. Samples were probed with antibodies against the indicated endogenous proteins. Inputs are 10% for pan-TEAD and 1% for others. n = 4 biological replicates. Experiment is quantified in SI Appendix, Fig. S8M. (F) Number of transcripts significantly (FDR < 0.05) altered by RNA-seq in the VGLL4 KO pool of cells compared to parental NCI-H226 cells. n = 3 biological replicates. (G) Pathway enrichment analysis on the 797 significantly increased genes from RNA-seq (FDR < 0.05 and |log₂FoldChange| > 1). Reactome Knowledgebase pathways were ranked by FDR and the top 15 pathways are shown. GTPase categories are presented in red text, ECM categories are in purple. Circle color indicates the FDR, size indicates the number of genes. (H) Diagram depicting the overlap of significantly changed transcripts between those increased with VGLL4 knockout and those decreased upon Compound 2 treatment and VGLL4 overexpression (from Fig. 3G). (I) Pathway enrichment analysis on the 356 genes from (H). Reactome Knowledgebase pathways were ranked by FDR and the top 15 significantly enriched pathways are shown. GTPase categories are presented in red text. Circle color indicates the FDR, size indicates the number of genes. (J) Tumor volumes (mean ± SEM) of mice bearing parental NCI-H226 xenograft tumors treated daily with VT-103 at 5 mg/kg, Compound 2 at 200 mg/kg, or MCT vehicle control. n = 6 mice per group. ***P < 0.00001 and by unpaired t tests with Welch correction. (K) As in (J) but for mice bearing KO VGLL4 NCI-H226 xenograft tumors. Endpoint tumor volumes were not significant by the unpaired t test with Welch correction. (L) Tumor masses from mice in (J and K). Tumors were excised 16 d after injection and measured. Box plots show spread and averages weights of tumors from each treatment group, n = 6 per group. *P = 0.012, ***P = 0.0001, and ns = not significant, by Welch’s two-tailed t test. See also SI Appendix, Figs. S8 and S9 and Table S1 and Dataset S4.

We next asked whether the other three VGLL family members contribute to the antiproliferative effects of Compound 2 and VT-103. VGLL1 and VGLL2 have very low expression in NCI-H226 cells, but VGLL3 is expressed (SI Appendix, Fig. S8G). We targeted VGLL1, VGLL2, and VGLL3, and achieved >80% decreased expression for all (SI Appendix, Fig. S8 H and I), except for VGLL1 which had no detectable expression in these cells. Because there are no reliable, commercially available antibodies for VGLL1, VGLL2, or VGLL3, we used RNA expression by qPCR to evaluate efficient knockdown by CRISPR, and, to avoid clonal effects, we used pools of CRISPR-transfected cells. We performed a dose–response viability assay in these CRISPR-modified cells and found that targeting VGLL1, VGLL2, and VGLL3 together or VGLL3 alone had little to no effect on the dose–response curves, while targeting VGLL4 conferred insensitivity to Compound 2 and VT-103 (SI Appendix, Fig. S8 J and K). In all conditions the sensitivity to VT-104 was retained (SI Appendix, Fig. S8L). These results indicate that VGLL4 is the only VGLL family member that has a role in conferring sensitivity to the sulfonamide-containing TEAD-targeting compounds.

Because VGLL4 has been shown to antagonize YAP–TEAD interaction and transcriptional activity in a variety of contexts (14–16, 50, 51), we next interrogated YAP in our VGLL4 KO cell system. Not only did we observe that YAP protein levels are higher in the VGLL4 KO cells compared to unperturbed parental NCI-H226 cells (Fig. 6E and SI Appendix, Fig. S8C), but we also observed a higher level of the YAP–TEAD interaction in the VGLL4 KO cells compared to the parental cells (Fig. 6E and SI Appendix, Fig. S6M). Interestingly, we observed a different trend for TAZ. Upon VGLL4 KO, TAZ levels decreased (Fig. 6E and SI Appendix, Fig. S8C), and there was little to no change in the TAZ–TEAD interaction (Fig. 6E and SI Appendix, Fig. S8M). Thus, VGLL4 depletion influences YAP and YAP–TEAD complexes, but not TAZ complexes, in NCI-H226 cells, suggesting that the VGLL4 cofactor switch favors YAP.

To explore the effects of YAP and TAZ individually, and how their functions differ in the presence of TEAD LPB compounds, we performed a CRISPR experiment. We used CRISPR to disrupt the expression of YAP alone, TAZ alone, or YAP and TAZ together (YAP+TAZ) in NCI-H226 cells (SI Appendix, Fig. S9A) and then monitored cellular growth with or without the presence TEAD LPB compounds over the course of 10 d. Targeting YAP or TAZ impacted cellular proliferation, and the effect of targeting both paralogs together was additive, resulting in very little cell growth (SI Appendix, Fig. S9 B–E). Treatment with compounds further decreased cellular proliferation in all conditions tested. These data demonstrate that YAP and TAZ are not completely redundant and indicate that both YAP and TAZ contribute to the effects of TEAD inhibitors.

To ask how VGLL4 depletion influences gene expression, we performed RNA-seq in parental NCI-H226 cells and in the VGLL4 KO cells. Overall, we identified ~6,000 genes that significantly changed in expression upon VGLL4 depletion (Fig. 6F). Of these changes, ~1,600 were more than two-fold in magnitude (SI Appendix, Fig. S9F). Because VGLL4 is a transcriptional repressor, we anticipated that many YAP/TAZ–TEAD target genes might be derepressed by the deletion of VGLL4. However, most of the 52 YAP/TAZ/TEAD target genes (46) (SI Appendix, Table S1) were only modestly changed in the VGLL4 KO (SI Appendix, Fig. S9G). Instead, the genes that change significantly are enriched in factors connected to extracellular matrix (ECM) formation and Rho GTPase activity (Fig. 6G, colored text and SI Appendix, Fig. S9H). Interestingly, these upregulated pathways are connected to mechanosignaling, which is a major regulatory mechanism for YAP activation (52–55). We next surmised that genes repressed by VGLL4 would increase in expression when we knock out VGLL4 and decrease in expression when we activate VGLL4 chemically or genetically (data from Fig. 3). Intersecting our RNA-seq datasets, we found 356 genes that match these criteria (Fig. 6H and Dataset S4). These 356 genes again show enrichment of Rho GTPase categories (Fig. 6I). Taken together, these results suggest that VGLL4 functions to block YAP not only by competing for binding to TEAD, but also by repressing Rho GTPase and ECM genes that activate YAP upstream.

Finally, to test the impact of VGLL4 deletion in a more tumor-relevant system we employed a NCI-H226 xenograft tumor model, which is frequently used to evaluate efficacy of TEAD-targeting molecules in vivo (4, 34, 35). Using parental NCI-H226 cells or VGLL4 KO NCI-H226 cells, we grafted subcutaneous tumors. No difference in the rate of tumor formation was observed between the parental tumors and the KO tumors (SI Appendix, Fig. S9I). Once tumors were established, we treated mice with vehicle control, Compound 2, or VT-103 for 16 d. Compared to vehicle control, growth of parental NCI-H226 tumors was inhibited in the Compound 2- and VT-103-treated mice (Fig. 6J), but growth of the VGLL4 KO tumors was insensitive to both compound treatments (Fig. 6K). In contrast to the responsive parental NCI-H226 tumors, the VGLL4 KO tumors were not smaller between treatment conditions (Fig. 6L). All treatments were well tolerated and did not cause body weight loss (SI Appendix, Fig. S9J). Altogether, these VGLL4 knockout experiments reveal VGLL4 as the required factor for the efficacy of sulfonamide-containing LPB TEAD inhibitors in slowing cancer growth in vitro and in vivo.

Discussion

Disrupting the oncogenic YAP–TEAD interaction is a predominant strategy for targeting Hippo pathway deregulation in cancer. Our data uncover a surprising, alternate mechanism whereby certain TEAD-targeting compounds act as “molecular glues” that promote the repressive VGLL4–TEAD interaction. We show that chemically boosting the VGLL4–TEAD interaction induces a cofactor switch from YAP to VGLL4. VGLL4–TEAD complexes exert an anticancer effect by dampening YAP activity and modulating key transcriptional networks, including mechanosignaling genes. Our findings suggest that TEAD-targeting molecules can be effective by inducing VGLL4–TEAD complexes, and we propose that VGLL4 should be considered when developing TEAD-based therapies.

Cofactor switching in transcriptional regulation is an established molecular mechanism for integrating different signaling pathways and fine-tuning transcriptional activities (56, 57). In the data presented here, four lines of evidence support the concept of a chemically induced TEAD cofactor switch from YAP to VGLL4. First, the interaction of VGLL4 and YAP with TEAD is mutually exclusive (15, 21). Second, treatment with Compound 2 leads to a concomitant decrease in the YAP–TEAD interaction (Figs. 1 and 2) which is restored upon depletion of VGLL4 (Fig. 6). Third, the transcriptional response to Compound 2 is highly correlated with the transcriptional response to GNE-7883, and we observe a lower magnitude of transcriptional changes with Compound 2 (Fig. 3). This is consistent with partial inhibition of YAP–TEAD by Compound 2 (via VGLL4), and a more complete blockade of YAP–TEAD with GNE-7883. Finally, in cell lines with high YAP levels and low expression of VGLL4 (such as MDA-MB-231 and OVCAR-8), inducing exogenous VGLL4 confers sensitivity to Compound 2 and VT-103 (Fig. 5), indicating that a certain threshold of VGLL4 protein is required to functionally outcompete YAP from TEAD. These data are consistent with studies demonstrating that VGLL4 and YAP compete for binding to TEAD (14–16, 20, 51), and indicate that small molecules can induce VGLL4-mediated transrepression of YAP by cofactor switching. It is notable that Compound 2 and VT-103 do not significantly decrease the TAZ–TEAD interaction. This could be due to functional differences between YAP and TAZ, or differences in how the proteins interact with TEAD (58). Future further structural interrogation of these binding modes, as well as TAZ-centered studies in TAZ-driven cells, will provide further insights.

A key observation from our study is a mechanosignaling gene signature linked to VGLL4. The VGLL4-regulated genes encompass integrins, collagens, fibronectin, and many genes involved in Rho-GTPase signaling. These genes are significant because a rigid extracellular matrix and Rho-GTPase signaling are key mechanical inputs that induce nuclear YAP activity (18, 55). VGLL4-mediated repression of these genes would weaken mechanical signals and release this activating input for nuclear YAP. Thus, based on this gene signature, we suggest a dual mechanism for VGLL4 in antagonizing YAP activity: One node directly competes for YAP–TEAD binding at DNA, and the second node transcriptionally restricts the expression of proteins involved in mechanical signals to activate nuclear YAP. In a context where YAP promotes cell growth, these two mechanisms together allow VGLL4 to exert tighter control over YAP activity. It is well established that multiple upstream inputs influence YAP activity. Accordingly, downstream regulators like VGLL4 are also advantageous, if not imperative. We propose that VGLL4 is one such downstream safeguard against YAP hyperactivity, acting both directly, by competing for YAP binding to TEAD, and indirectly, by repressing the expression of mechanosignaling genes that activate YAP.

So far, the glue-like mechanism by which Compound 2 and VT-103 compounds influence TEAD to recruit VGLL4 is not fully clear. One possibility is that the compounds induce an allosteric conformational change that favors a VGLL4-bound state or simply reduces the in-solution flexibility of TEAD. Alternatively, the sulfonamide could act orthosterically to alter the protein interface of TEAD to enhance the VGLL4–TEAD interaction. Either way, the compound-induced changes are not clear from existing crystal structure data (4, 37, 42, 44, 59–61), and crystal structures often fail to fully explain how LPB molecules disrupt YAP–TEAD (4, 61, 62). The minimal insights from currently available crystallographic data is a considerable limitation for the field and more interrogation is required to dissect how TEAD cofactor interactions are differentially modulated by different LPB molecules.

With small molecules targeting the Hippo pathway in preclinical and clinical development, our research raises an important question on the therapeutic potential of TEAD-targeting molecules. Our findings identify VGLL4 as the key determinant for the response to sulfonamide-containing TEAD-targeting compounds, but recent research suggests a broader relevance for VGLL4. Results of a CRISPR knockout screen identified VGLL4 as a mechanism of resistance to a nonsulfonamide TEAD inhibitor (63). These results—combined with the trend that low VGLL4 expression in tumors correlates with poor prognosis (14, 15, 23–26)—suggest that VGLL4 may play a larger than expected role in mediating the effects of TEAD inhibitors. It is also important to consider that if a cancer cell evolves to inactivate VGLL4 (as frequently occurs), then a therapy that acts through VGLL4 will not work. Loss of VGLL4, acquired or inherent, is a vulnerability. Because of this, perhaps an ideal TEAD-targeting molecule would both increase VGLL4–TEAD and directly disrupt YAP–TEAD, even in the absence of VGLL4. Evaluating the role of VGLL4 within emerging experimental and clinical data from various TEAD inhibitors will better our understanding of VGLL4 in cancer and could be a critical step toward developing effective therapeutics targeting the Hippo pathway.

Materials and Methods

Compound Synthesis and Characterization.

Detailed methods for the synthesis and characterization of compounds S1, N1, S2, N2, C2-Amino, C2-Acetyl, and C2-Methyl are described in the SI Appendix, Methods S1 Chemistry Detail.

Cell Lines and CRISPR Genome Engineering.

Detailed descriptions of the cell lines used and generated in this study can be found in the SI Appendix, Materials and Methods.

Molecular Biology.

Detailed methods and reagents for biochemical assays, RNA-seq, ChIP-seq, and cell-based assays are described in the SI Appendix, Materials and Methods.

Mouse Xenograft Assays.

Female C.B-17 SCID (inbred) mice were obtained from Charles River Laboratories at Hollister. Animals were maintained in accordance with the Guide for the Care and Use of Laboratory Animals (National Research Council 2011). Genentech is an AAALAC-accredited facility and all animal activities in this research study were conducted under protocols approved by the Genentech Institutional Animal Care and Use Committee (IACUC). Experimental methods for xenograft assays are described in the SI Appendix, Materials and Methods.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Data, Materials, and Software Availability

Mass spectrometry proteomics data are deposited at the MassIVE repository (https://massive.ucsd.edu/) under the Accession Nos. MSV000094637 (64) and MSV000094638 (65). RNA-Seq and ChIP-seq data are deposited at the NCBI BioProject database (https://www.ncbi.nlm.nih.gov/bioproject/) under the BioProject Accession PRJNA1156808 (66). All other data are included in the manuscript and/or SI Appendix. Some study data are available; unique reagents generated in this study may be requested through Genentech’s MTA program.