Binary pan-cancer classes with distinct vulnerabilities defined by pro- or anti-cancer YAP/TEAD activity

SUMMARY

Cancer heterogeneity impacts therapeutic response, driving efforts to discover over-arching rules that supersede variability. Here, we define pan-cancer binary classes based on distinct expression of YAP and YAP-responsive adhesion regulators. Combining informatics with in vivo and in vitro gain- and loss-of-function studies across multiple murine and human tumor types, we show that opposite pro- or anti-cancer YAP activity functionally defines binary YAP^on or YAP^off cancer classes that express or silence YAP, respectively. YAP^off solid cancers are neural/neuroendocrine and frequently RB1^–/–, such as retinoblastoma, small cell lung cancer, and neuroendocrine prostate cancer. YAP silencing is intrinsic to the cell of origin, or acquired with lineage switching and drug resistance. The binary cancer groups exhibit distinct YAP-dependent adhesive behavior and pharmaceutical vulnerabilities, underscoring clinical relevance. Mechanistically, distinct YAP/TEAD enhancers in YAP^off or YAP^on cancers deploy anti-cancer integrin or pro-cancer proliferative programs, respectively. YAP is thus pivotal across cancer, but in opposite ways, with therapeutic implications.

Graphical Abstract

In brief

Pearson et al. demonstrate that YAP/TAZ, well-known oncogenes, are tumor suppressors in a large group of cancers. Pan-cancer analyses reveal that opposite YAP/TAZ expression, adhesive behavior, and oncogenic versus tumor suppressor YAP/TAZ activity functionally stratify binary cancer classes, which interchange to drive drug resistance. Contrasting YAP^off/YAP^on classes exhibit unique vulnerabilities, facilitating therapeutic selection.

INTRODUCTION

Tumor heterogeneity and plasticity complicate cancer diagnosis and therapeutics, thus there is much interest in deducing over-arching rules that span distinct tumor types (Balanis et al., 2019; Chen et al., 2018, 2019; Hoadley et al., 2018; Priestley et al., 2019). These efforts can expose vulnerabilities with broad therapeutic relevance, such as strategies that exploit different classes of p53 mutation (Hientz et al., 2017). Here, starting with the goal of understanding unique properties of small cell cancers, we delineate a pan-cancer rule based on opposite expression and activity of a single transcriptional complex.

Neuroendocrine (NE) cancers arise from hormone/neurotransmitter-secreting cells, or evolve from adenocarcinomas through lineage conversion (Rindi and Inzani, 2020). Small cell neuroendocrine (SCN) cancers are a highly lethal subset, and often exhibit metastasis at presentation; e.g., small cell lung cancer (SCLC) has a dismal 5-year survival of <6% and originates from endogenous lung NE cells (Huang et al., 2018; Lázaro et al., 2019; Sutherland et al., 2011; Yang et al., 2018), or conversion of non-SCLC (NSCLC) to SCLC linked to drug resistance (reviewed in Rudin et al., 2019). SCN prostate cancer is rare at presentation, but up to 17% of castration-resistant cases develop pure or mixed NE histology, with reduced overall survival versus non-NE castration-resistant disease (Aggarwal et al., 2018). Such lineage switching will rise with increased use of targeted adenocarcinoma therapies, thus there is much interest in new SCN cancer vulnerabilities (e.g., Balanis et al., 2019; Oser et al., 2019; Zhang et al., 2020).

The RB1 tumor suppressor gene is inactivated in most SCLCs, and is lost during NE lineage transitions in both NSCLC and prostate cancer (reviewed in Beltran et al., 2019; Rudin et al., 2019). The RB1 gene was identified in the pediatric cancer, retinoblastoma, which is also a small cell cancer (Rajwanshi et al., 2009), but arises from neurons, as opposed to NE cells (Xu et al., 2014). RB1 loss approaches 100% in these cancers, yet is far rarer in most other solid tumors, but this context specificity is un-explained. SCN cancers grow as semi- or non-adherent cells in culture, akin to hematopoietic tumors, but most solid cancers adhere via interactions between adhesion molecules (e.g., integrins) and extracellular matrix (ECM) components. The driving force behind opposite pan-cancer adhesive behavior is unclear.

The paralogs YAP and TAZ (WWTR1) are oncogenic drivers in human solid cancers (Zanconato et al., 2016; Zheng and Pan, 2019). These transcriptional coactivators interact with DNA-binding transcription factors (TFs). TEAD1–4 recruit YAP to remote enhancers populated with AP1 complexes to induce cell-cycle genes and promote cancer (Liu et al., 2016; Stein et al., 2015; Verfaillie et al., 2015; Zanconato et al., 2015; Zhao et al., 2008). In contrast, ectopically expressing YAP in the liquid cancer, multiple myeloma, promotes p73-mediated apoptosis (Cottini et al., 2014). In line with an anti-cancer function, YAP is absent in hematopoietic cancers. Intriguingly, most cases of SCLC may be YAP deficient (Rudin et al., 2019), but whether that reflects a tumor suppressor role, whether TAZ behaves in this way, and/or the extent to which YAP/TAZ silencing is critical in other solid cancers is unknown. Finally, it is unclear whether contrasting YAP/TAZ levels influence therapeutic vulnerability.

Here, we connect contrasting levels and opposite pro- or anti-cancer activity of YAP/TAZ to context-specific transformation upon RB1 loss, opposite adhesive behavior, and distinct pan-cancer vulnerabilities, which tumors exploit to evade therapeutics. We tie these opposing activities to unique YAP/TEAD enhancer deployment in YAP-deficient or YAP-positive cancers. Divergent behavior of a single transcriptional complex thus defines binary pan-cancer classes, exposing broadly relevant therapeutic opportunities.

RESULTS

Obligate YAP and TAZ absence in the retinoblastoma and SCLC cells of origin

Removing Rb1 and its paralog Rbl1 (p107) causes murine retino-blastoma, but penetrance is partial (Chen et al., 2004). Tumors were Yap/Taz positive (Figure 1A), but despite their oncogenic role in most solid cancers (Zanconato et al., 2016; Zheng and Pan, 2019), penetrance rose when one, and more so both, genes were deleted (Figure 1B and S1A). These data prove Yap/Taz tumor suppressor activity in vivo, and explain low penetrance murine retinoblastoma. Moreover, while ectopic Yap initially promoted growth of Rb1 or Rb1/Rbl1-null clones, division petered out, generating benign Yap⁺ lesions, but never malignancy (Figures S1B–S1E). Thus, both loss-of-function (LoF) and gain-of-function (GoF) assays show that Yap suppresses cancer in Rb1^–/– murine retina.

Figure 1. Obligate YAP/TAZ silencing in retinoblastoma.

(A) Western blots of murine retinal or tumor lysates.

(B) Kaplan-Meyer from αCre mice of the indicated genotypes. *p < 0.01, ***p < 0.0001, log rank (Mantel-Cox) test. DKO/TKO/QKO, double/triple/quadruple knockout; het, heterozygous.

(C) H&E and immunofluorescence (IF) in human retinoblastoma (Rtb) and retinoma (Rtm). Scale bar, 100 mm.

(D) Western blots of Rtb cell lines and tumors.

(E) Ectopic YAP or hyperactive YAP^5SA, but not YAP^S94A, impedes growth of Rtb lines (n = 3, p < 0.01).

(F) Soft agar assays of Rtb lines expressing the indicated YAP constructs. Scale bars, 3 mm.

(G) Ectopic TAZ impedes growth of Rtb lines (n = 3, p < 0.001).

(H) IF in human Fw17 RB1^–/– retinal explants. Orange arrowheads, YAP^– dividing (MCM3⁺) cones (RXRγ⁺); white brackets, YAP⁺ progenitors/Müller glia. Scale bar, 20 mm.

(I) Western blots of human RB1-depleted cone line (Co^shRB). NCI-H661, control for RB1, YAP/TAZ. See also Figure S1.

Strikingly, YAP/TAZ were absent in human retinoblastoma, but expressed in retinoma (Figure 1C), an RB1^–/–, but benign, postmitotic ‘‘dead-end’’ or barrier for progression (Dimaras et al., 2008). Matching the immunofluorescence data, YAP/TAZ were undetectable/low in human retinoblastoma tumors and cell lines (Figure 1D). Expressing YAP or TAZ inhibited retinoblastoma 2D or 3D growth, which overactive YAP (YAP^5SA) enhanced, and deleting ectopic YAP ablated (Figures 1E–1G and S1F–S1H). Thus, unlike other solid cancers, YAP/TAZ suppress retinoblastoma.

Human retinoblastoma arises from cones, which are normally post-mitotic, but divide without RB1 (Xu et al., 2014). In human fetal explants, dividing progenitors expressed YAP/TAZ, but ectopically dividing RB1^–/– cones did not (Figure 1H and S1I). Moreover, RB1-deficient primary cones we derived from fetal retina also lacked YAP/TAZ (Figure 1I). Thus, YAP/TAZ absence is intrinsic to the retinoblastoma cell of origin.

Next, we assessed SCLC, a highly lethal RB1^–/– cancer. Among seven cell lines, representing three subtypes (A: ASCL1; N: NEUROD1; P: POU2F3, Figure S2A) (Mollaoglu et al., 2017; Rudin et al., 2019), none were YAP⁺, and only one expressed TAZ, but weakly relative to NSCLC (Figure 2A). Immunohistochemistry (IHC) showed that primary SCLC lacks YAP/TAZ, unlike NSCLC (Figure 2B). Transcriptome analysis revealed low YAP/TAZ mRNA levels in NE versus other lung cancers, and in SCLC patient-derived xenografts (PDXs), YAP/TAZ proteins were well below NSCLC PDX levels (Figures 2C and 2D). Yap/ Taz mRNAs were low in murine SCLC (Figure 2E), and in both Cgrp^CreER;Rb1^f/f;p53^f/f (Song et al., 2012) and adenoviral (Ad)-Cre-Rb1^f/f;p53^f/f SCLC, Yap/Taz proteins were absent, contrasting normal lung and murine EGFR-driven NSCLC (Figures 2F and 2G). Thus, retinoblastoma and SCLC are predominantly YAP/TAZ^– (‘‘YAP^off’’) cancers, contrasting YAP/TAZ⁺ (‘‘YAP^on’’) cancers.

Figure 2. Obligate YAP/TAZ silencing in SCLC.

(A) Western blots of SCLC, NSCLC cell lines.

(B) YAP/TAZ and p16 (marker for RB1 loss) IHC in primary human SCLC and NSCLC PDXs (left) and scoring (right). Scale bars, 300 µm.

(C) YAP and TAZ mRNA in primary human lung tumors. *p < 0.05, **p < 0.01, ***p < 0.001, *****p < 1 × 10^–5.

(D) Western blots of YAP/TAZ in SCLC and NSCLC PDXs. PDXs are from different donors, except the paired treatment naive (N) and cisplatin/etoposide-resistant (R) LX108 samples.

(E) Violin plot of low Yap and Taz mRNA in murine SCLC.

(F) Yap/Taz IF in Cgrp^CreER;Rb1^f/f;p53^f/f SCLC. Neighboring airway/alveolar cells express Yap/Taz but not dividing Syp⁺ NE cells. Note rare Syp^–;Ki67^–;Yap/Taz⁺ cells within tumor. Scale bar, 20 µm.

(G) IHC for Ccsp-rtTA;tetO-EGFR^L858R NSCLC and Ad-CMV-Cre-Rb1^f/f;p53^f/f;Pten^f/+ SCLC. Scale bar, 100 µm.

(H) IF on embryonic day 18.5 (E18.5) Spc-rtTA;tetO-Cre;Rb1^f/f;p53^f/f lungs. Orange arrows, Yap^– dividing (Ki67⁺) NE cells (Syp⁺); white arrows, Yap⁺ dividing non-NE cells (Syp^–). Scale bar, 20 µm.

(I) Ectopic YAP or hyperactive YAP^5SA (n = 3, p < 0.05) but not YAP^S94A impede growth of SCLC lines.

(J) Ectopic TAZ suppresses growth of SCLC lines (n = 3, p < 0.0001).

(K) Outline for murine SCLC +/– Yap (left), images (middle), and area (right) of control and Yap⁺ SCLC 3 months after tam/dox. NEB, NE bundle. t test with Welch’s correction. Scale bars, 50 µm.

(L) Graphs show number (left) and size (middle) of dividing NEB in Rb1^f/f;p53^f/f or Rb1^f/f;p53^f/f;Yap^f/f;Taz^f/f mice 2 or 3 months after Ad-CMV-Cre administration. Mean ± SD, n = 3 (3 months) or 4 (2 months) mice/genotype; *p < 0.05, **p < 0.01. Representative images (right) are from 3 month mice. Scale bars, 200 µm. See also Figure S2.

Rb1/p53-deficient pulmonary NE cells (PNECs) generate SCLC (Semenova et al., 2015; Sutherland et al., 2011), so we asked whether, like human RB1^–/– cones, they are naturally YAP/TAZ^–. In SPC-rtTA;tetO-Cre;Rb1^f/f;p53^f/f mice, doxycycline (dox) treatment during gestation induces ectopic division in many Rb1/p53 null lung lineages (Akeno et al., 2017; Simpson et al., 2009). After birth, ectopic division ceases, except in PNECs. At embryonic day 18.5, abnormally dividing but cancer-resistant ATII and Club cells expressed Yap/Taz, but cancer-prone PNECs did not (Figures 2H and S2B). Lineage tracing in tamoxifen-treated Cgrp^CreER;Rb1^f/f;p53^f/f mice, which specifically targets NE cells (Song et al., 2012), confirmed that Rb1/p53-null PNECs lack Yap/Taz, as did adult wild-type PNECs (Figures S2C–S2D). Thus, Rb1/p53 loss triggers division in many lung cell types, but abundant Yap⁺ lineages are tumor-resistant while rare tumor-prone NE cells (<0.5% of the lung), are Yap^–. Our lung and retina data tie YAP/TAZ status to context-dependent susceptibility to cancer after RB1 loss.

Next, we performed GoF studies in models of SCLC. Ectopic YAP/TAZ inhibited growth of human YAP^off SCLC lines, but not YAP^on NSCLC lines (Figures 2I, 2J, and S2E–S2I). To assess whether YAP blocks SCLC initiation in vivo, we combined Cgrp^CreER;Rb1^f/f;p53^f/f mice with a Cre-inducible, tetracycline (tet)-activated transactivator (lox-STOP-lox-rtTA-GFP) and tetresponsive Yap transgene (TRE-Yap) (Gregorieff et al., 2015) (Figure 2K). Yap dramatically reduced expansion of Rb1/p53-null PNECs (Figure 2K). Rare, large, dividing NE clusters emerged, but expressed minimal Yap (Figure S2J). To test whether Yap also inhibits established SCLC, we treated Rb1^f/f;p53^f/f;lox-STOP-lox-rtTA-GFP;TRE-Yap mice with Ad-Cre, then dox-induced Yap 3 months after tumor initiation. Despite only mosaic induction, Yap reduced tumor load and prolonged survival (Figure S2K), and Yap⁺ regions retained an NE phenotype and NE marker expression, but exhibited large cell instead of small cell morphology and significantly reduced division (Figures S2L–S2M). Our extensive GoF data suggest that reactivating YAP expression is a therapeutic option for SCLC.

The observation that YAP/TAZ are undetectable in the SCLC cell of origin, coupled with our GoF data, suggest tumor suppressor activity. Conceivably, however, trace levels could be essential for SCLC. To test this idea, we compared tumorigenesis in Rb1^f/f;p53^f/f and Rb1^f/f;p53^f/f; Yap^f/f;Taz^ff mice treated with Ad-Cre. Yap/Taz loss did not inhibit tumorigenesis, and instead tumors were larger (Figure 2L). The increase may reflect tumor suppression by trace Yap/Taz in NE cells, more efficient conversion of non-NE cells (Ferone et al., 2020), and/or loss of indirect tumor suppression from neighboring cells (Moya et al., 2019), although these were Yap/Taz⁺ (Figure S2N). Irrespective, these in vivo data show, unequivocally, that Yap/Taz are not essential oncogenes in SCLC, and instead suppress this cancer.

YAP/TAZ and integrin/adhesion genes stratify binary pan-cancer classes

To define other YAP/TAZ-deficient cancers, we surveyed 1,036 lines in the Cancer Cell Line Encyclopedia (CCLE). YAP/TAZ exhibited a striking off/on pattern (Figure 3A), raising the hypothesis that binary expression and opposite functions of YAP/TAZ may provide a useful stratification for human cancers. As an unbiased test, we used principal-component analysis (PCA) to assess CCLE transcriptomes. YAP/TAZ were top scoring genes on the positive side of principal component 1 (PC1⁺), which defined binary YAP^off and YAP^on classes (Figures 3B and S3A; Table S1). PC2 and, more effectively, PC3 distinguished two groups of YAP^off cancers: hematopoietic (cluster A), and solid neural/NE cancers (cluster B) from many tissues, including lung, prostate, thyroid, breast, gastrointestinal, and ganglia (Figures 3B, 3C, S3B, and S3C). PC3 also parsed YAP^on cancers into two groups (clusters C/D). PCA of Sanger lines gave similar results (Figure S3D). Western blots confirmed no/very low YAP/TAZ in YAP-^off cancers of the breast, pancreas, ganglia, and prostate, similar to SCLC (Figure 3D). RB1 mutation was enriched in cluster B, extending our retinoblastoma/SCLC data (Figure 3E). Gene set enrichment analysis (GSEA) showed that PC1⁺ genes are Hippo signaling (e.g., YAP, TAZ, CYR61, AMOTL2), adhesion, and ECM components (e.g., the integrin ITGB5; Figures 3C and 3F; Table S1). PC3⁺ and PC1^–, highest in blood cancers (cluster A, YAP^off), were enriched for hematopoietic genes. PC3^– was enriched for REST-repressed neural genes, consistent with high levels in solid YAP^off cancers, which have neural/NE features. Thus, YAP/TAZ and correlated integrin/ECM genes in PC1⁺ parse cancers into binary classes that are further defined by PC3 genes.

Figure 3. YAP, TAZ, and their integrin/ECM/adhesion targets stratify binary cancer classes.

(A) YAP/TAZ mRNA across the CCLE.

(B) PCA on CCLE transcriptome data.

(C) k-Means clustering of the top (>1.8%) PC1/PC3 genes.

(D) Western blots of YAP^off and YAP^on lines.

(E) RB1 mutation frequency (clusters from C); Fisher’s exact test.

(F) GSEA of KEGG pathways or ENCODE/ChEA TF binding to ranked PC1/PC3 gene lists.

(G) Expression of the top 330 PC1/3 genes in TCGA/TARGET cancers.

(H) GSEA of the top PC1/3 gene sets from (C) versus an SCN cancer signature (Balanis et al., 2019).

(I) YAP/TAZ mRNA expression in the indicated TCGA cancers.

(J) YAP IHC in lung and gastrointestinal well-differentiated NETs and metastases (met). Similar results with a YAP/TAZ antibody. LN, lymph node. *YAP⁺ stroma/ endothelia. Scale bars, 100 µm.

See also Figure S3 and Tables S1 and S2.

Next, we assessed primary cancers. Strikingly, the top 330 PC1/3 genes stratified 10,485 TCGA/TARGET cancers into YAP^off/YAP^on classes (Figures 3G, S3B, and S3C). Like cell lines, YAP^off cancers were cluster A/blood and cluster B/solid cancers from neural/NE tissues, while YAP^on cancers comprised clusters C/D similar to the CCLE (Figure S3C). PC1⁺/PC3⁺ genes were down-regulated and PC3^- genes up-regulated in a signature of SCN cancers (Figure 3H) (Balanis et al., 2019). TCGA ‘‘adenocarcinomas’’ include SCN variants of pancreas, bladder, breast, lung, and stomach cancer (Balanis et al., 2019), and YAP/TAZ were down-regulated in these tumors (Figure 3I). Well-differentiated NE tumors (NETs) are histologically distinct from SCN cancers, and among 143 NETs from stomach, lung, duodenum, small and large intestine, pancreas, and appendix, as well as liver and lymph node metastases, we easily detected YAP/TAZ in mesenchyme and vasculature, but they were absent in 97.9% (140) of these tumors (Figure 3J; Table S2). While there may be exceptions, our data suggest that most cancers stratify into binary YAP^off and YAP^on classes.

Low levels of YAP/TAZ-correlated adhesion genes could reflect transcriptional or post-transcriptional regulation. Hippo pathway kinases can destabilize YAP/TAZ, but depleting MST1 or overexpressing dominant-negative LATS1 did not induce YAP or TAZ (Figure S3E). Mining chromatin immunoprecipitation sequencing (ChIP-seq) data from YAP^off lines revealed YAP and YAP-correlated PC1⁺ gene promoters lack RNA polymerase II, unlike anti-correlated PC1^– genes, and the opposite pattern applied in YAP^on cells (Figure S3F). ChIP-qPCR of activating and repressive epigenetic marks at YAP, TAZ, AJUBA, and CYR61 promoters indicated opposite patterns in YAP^off versus YAP^on cell lines (Figure S3G). Thus, epigenetic regulation defines the binary YAP^off or YAP^on pan-cancer states, respectively.

Binary classes exhibit opposite YAP/TAZ function, contrasting vulnerabilities, and therapy-driven drug resistance

YAP/TAZ suppress solid YAP^off cancers (Figures 1 and 2), and YAP also inhibits multiple myeloma (Cottini et al., 2014), a liquid YAP^off cancer (Figures 3C and S3C). In contrast, many YAP^on solid cancers require YAP/TAZ (Zanconato et al., 2016; Zheng and Pan, 2019). Thus, an intriguing hypothesis is that opposite YAP/TAZ function defines most cancers. Mining the Cancer Dependency Map (DepMap) (Meyers et al., 2017), a functional genomics effort to define essential genes in hundreds of cancer cell lines, revealed that deleting YAP or TAZ did not affect 92% of YAP^off cases, contrasting dependency on either gene in 65% of YAP^on lines, and in 5/5 YAP^on lines where either YAP or TAZ are dispensable, deleting both revealed redundancy (Figures 4A–4C, S4A, and S4B). One study claimed that RNAi-depletion of TAZ weakly impedes retinoblastoma cell line growth (Zhang et al., 2015), but dual YAP/TAZ-deletion had no effect in these or other YAP^off cell lines (Figures 4C and S4B). Moreover, as in 11 retinoblastoma/SCLC cell lines (Figures 1E and 2I), expressing YAP inhibited growth of breast, pancreas, and prostate YAP^off NE cancer lines (Figures 4D and S4C). Combined with our in vivo and in vitro GoF/LoF studies in YAP^off cancers, these pan-cancer data suggest that it is not only opposite expression, but also opposite YAP/TAZ function that defines pan-cancer binary classes.

Figure 4. Binary cancer classes exhibit contrasting YAP/TAZ function and vulnerabilities.

(A) YAP/TAZ dependency and mRNA levels in DepMap cell lines. The indicated YAP^on lines were tested in dual KO assays (see panel C).

(B) YAP or TAZ dependency in YAP^off or YAP^on lines; Fisher’s test.

(C) YAP/TAZ KO suppresses growth of YAP^on (n = 3, p < 0.05) but not YAP^off cells. D, I, n/a, YAP/TAZ-dependent, -independent, or not available in DepMap.

(D) Ectopic YAP or YAP^5SA, but not YAP^S94A, impede growth of indicated YAP^off lines (n = 3, p < 0.05).

(E) Drug map of YAP^off solid cancer-selective (blue) or YAP^on cancer-selective (red) drugs in CTRPv2 and GDSC databases.

(F) Area under the curve (AUC) of select top scoring drugs in CTRPv2 database.

(G) Validation of indicated YAP^off-selective drugs.

(H) Effect of the indicated expression vectors on the sensitivity of YAP^off lines to YAP^off-selective drugs. Mean ± SD, n ≥ 3; *p < 0.05, **p < 0.01.

(I) Top dependency genes selective for cluster A/B (YAP^off) or cluster C/D (YAP^on) cell lines. MYC/MAX-bound targets and metabolic (Metab.) genes (YAP^off), or genes related to integrin signaling, ECM, or cytoskeletal regulation (Int./ECM/Cyto., YAP^on) are indicated; +, from Enrichr; (+), from the literature.

(J) DepMap-defined dependency of different cancer classes on select gene combinations. Fisher’s test: *p < 0.05, **p < 0.01, ***p < 0.0001.

(K) MYCN or MYCL amplification in cancer classes.

(L) MYC and MYCN signature scores in cancer classes. Pairwise Wilcoxon test, Bonferroni correction.

See also Figure S4 and Tables S3 and S4.

Next, we searched for class-selective vulnerabilities, which could have broad therapeutic implications. Employing CTRP and GDSC drug databases, we identified several high-confidence cluster B-selective drugs. Inhibitors of the NAD synthesis enzyme NAMPT, aurora kinases, the anti-apoptotic BCL family, and histone deacetylases (HDACs), were hits in both databases (Figure 4E), and inhibitors of the nuclear transporter XPO1 and the elongation factor EIF4 had multiple hits in the CTRP database (Table S3). Moreover, these drugs also potently inhibited cluster A cancers, revealing general YAP^off selectivity (Figures 4F and S4D). Notably, aurora kinase inhibitors are synthetic lethal with RB1 loss (Gong et al., 2019; Oser et al., 2019), which is enriched in YAP^off solid cancers (Figure 3E), and XPO1 and BCL or HDAC inhibitors are potent against leukemia and SCN tumors, respectively (Balanis et al., 2019; Verbeke et al., 2020). We validated several YAP^off-selective drugs, and expressing YAP and/or constitutively active YAP^S5A reduced sensitivity in several YAP^off cancer types (Figures 4G and 4H). Thus, the unique drug profile of YAP^off cancers depends on YAP silencing.

To further validate and expand therapeutic targets, we assessed genetic vulnerabilities in YAP^off cancers. Using DepMap we calculated differential gene dependency scores between YAP^off and YAP^on cancers (Figure 4I; Table S4). Genes for drug targets that are indispensable in all cells (e.g., XPO1, EIF4) cannot be validated genetically, and single-gene deletion for redundant families (e.g., BCL/HDAC) may have little/no effect. Nevertheless, the fraction of lines that required BCL2 and/or BCL2L2 were significantly higher in YAP^off cancers (Figure 4J). In striking agreement with YAP^off sensitivity to NAMPT inhibitors (Figure 4E), top single-gene dependencies included both NAMPT and its partner, NAD synthesis enzyme, NMNAT (Figure 4I). YAP^off-dependency genes were enriched for metabolic enzymes, including three of the six-member purinosome that drives de novo purine synthesis (Figures 4I and S4E; Table S4). This gene set was also enriched for direct MYC/MAX binding (Figure 4I; Table S4), and MYC drives these metabolic pathways (Stine et al., 2015). Furthermore, MYCN or MYCL amplification was enriched in solid (cluster B) YAP^off cancers (Figure 4K), and both cluster A and cluster B YAP^off cancers had higher levels of MYC- or MYCN-induced gene signatures than YAP^on cancers (Figure 4L). Contrasting the YAP^off drug map, YAP^on cancers exhibited elevated sensitivity to tyrosine (e.g., SRC, ERBB, EGFR) or serine/threonine (e.g., BRAF) kinase inhibitors and, in agreement, the fraction of YAP^on cancers dependent on the corresponding genes was higher in YAP^on versus YAP^off cancers (Figures 4E and 4J). In addition, YAP^on gene dependencies were enriched in integrin/ECM/cytoskeletal regulators (Figure 4I), mirroring high expression in YAP^on cancers (Figures 3C and 3F; Table S1). Thus, YAP^off and YAP^on cancers exhibit distinct drug vulnerabilities, genetic dependencies complement these contrasts, and class specificity mirrors divergent mutational spectra and transcriptional profiles that typify YAP^off and YAP^on cancers.

The discovery of contrasting vulnerabilities in YAP^off versus YAP^on classes suggested that cancer may switch binary classes to escape therapeutics. Indeed, conversion of adenocarcinoma to SCN cancer is a lethal, therapy-driven event associated with RB1 loss, an NE lineage switch, and altered drug sensitivity (Beltran et al., 2019), all features consistent with YAP^on to YAP^off transformation. To explore this hypothesis, we assessed relevant murine cancer models. In lung cancer with mixed adenocarcinoma and NE histology, only the former was Yap⁺ (Figure 5A). Similarly, Yap protein was expressed in Pten^–/– (so-called single KO [SKO]) prostate adeno-carcinoma (PrAd), but absent in two models of late-stage NE prostate cancer (NEPC) involving Rb1/Pten (DKO) or Rb1/ Pten/p53 loss (TKO) (Figure 5B). Mapping our class-defining PC1/PC3 gene set onto transcriptome data revealed that SKO PrAd are YAP^on cancers, whereas advanced DKO or TKO NEPC are YAP^off cancers (cf. Figures 5C and 3C). DKO tumors are initially PrAd, and convert to NEPC starting at around weeks 20–25, mimicking anti-androgen-induced lineage conversion in human prostate cancer (Ku et al., 2017). Strikingly, early-stage 12-week DKO PrAd exhibited a YAP^on transcriptome (Figure 5C). To quantitatively assess these differences, PCA was used to define a ranked gene list differentiating SKO and 12-week DKO PrAd from late-stage DKO and TKO NEPC (Table S5). GSEA validated the YAP^on versus YAP^off classification of PrAd versus NEPC, respectively (Figure 5D). In the TKO model, PrAd to PNEC conversion starts earlier, at 11 weeks and, while we lacked transcriptome data from this stage, nascent NEPC regions were Yap^–, starkly contrasting neighboring Yap⁺ adenocarcinoma (Figure 5E).

Figure 5. Epithelial-to-SCN cancer is a YAP^on to YAP^off switch.

(A) IHC on a murine lung model with mixed NSCLC (Syp^–/Yap⁺) and NE (Syp⁺/Yap^–) histology. Scale bar, 100 µm.

(B) IHC in murine PrAd (SKO: Pten^–/–) or NEPC (DKO: Pten^–/–;Rb1^–/– or TKO:Pten^–/–;Rb1^–/–;p53^–/–). Scale bar, 100 µm.

(C) PC1/3 gene levels in murine prostate cancer (cf. Figure 3C).

(D) PC1/3 gene enrichment in a ranked list (Table S5) distinguishing murine PrAd and NEPC confirms these prostate tumors are YAP^on and YAP^off cancers, respectively.

(E) IHC on emerging NE tumors in 11-week murine TKO prostate. Scale bars, 100 µm (top), 20 µm (bottom).

(F) Multispectral IHC of mixed human NEPC (SYP⁺) and PrAd (SYP^–). Only PrAd (dotted area) and blood vessels (arrows) stain for YAP or YAP/TAZ.

(G) PARCB gene code to reprogram human primary epithelia to SCN cancer (Park et al., 2018).

(H) PC1/3 gene levels in primary human epithelia or SCN cancer derived as in (G) (cf. Figure 3C).

(I) PC1/3 gene enrichment in a ranked list (Table S5) distinguishing the two states in (G) confirms the PARCB genes induce YAP^on to YAP^off reprogramming.

See also Table S5.

Next, we asked whether NE cancer conversion in humans is also a YAP^on to YAP^off class switch. We used H&E staining plus multispectral IHC to analyze 17 pure PrAd, 11 mixed NEPC/PrAd, and 9 NEPC samples. Pure PrAd and NEPC tumors were YAP/TAZ positive and negative, respectively, and, in mixed tumors, YAP/TAZ staining was confined to PrAd regions or blood vessels, and absent from NEPC (Figure 5F). In a second approach, we took advantage of a model in which primary human epithelia are reprogrammed to aggressive SCN cancer through the ‘‘PARCB’’ code involving RB1/p53 inactivation plus cMYC/AKT/BCL2 activation (Figure 5G) (Park et al., 2018). Mining the associated transcriptome changes revealed a YAP^on to YAP^off transformation, both in lung and prostate (Figures 5H and 5I). These murine and human data reveal that highly lethal NE lineage switches in YAP^on lung or PrAd drive YAP^off cancer, highlighting the potential value of YAP^off-selective therapies (Figures 4E–4G, S4D, and S4E).

YAP^on- and YAP^off-specific gene targets drive opposite adhesive behavior of binary cancer classes

To deduce why YAP has opposite pro- or anti-cancer effects, we performed RNA sequencing (RNA-seq) with five empty vector- or YAP-transduced YAP^off lines, and compared published YAP-dependent gene sets from five YAP^on cell lines. YAP mainly induced genes in YAP^off lines (Figure S5A). Either rank-rank hypergeometric overlap or GSEA clustered YAP^off and YAP^on datasets separately, implying distinct YAP targets underpin binary classes (Figures 6A and S5B). We confirmed that YAP induces proliferative genes (G1-S transition, DNA replication, etc.) in YAP^on cells (Hiemer et al., 2015; Kim et al., 2016; Li et al., 2014; Park et al., 2016; Zanconato et al., 2015), but YAP did not affect these gene sets in YAP^off cells and, instead, activated the PC1⁺ integrin/ECM/adhesion genes silenced in this class (Figures 6B, S5C, and S5D; Table S6). Unsupervised clustering of the top YAP targets confirmed induction of integrins, collagens, laminins, and downstream cytoskeletal components (e.g., myosins, FLNA) in YAP^off cells, contrasting induction of cyclins, minichromosome maintenance (MCM), and other cell-cycle genes in YAP^on cells, while well-known YAP targets, such as CTGF and CYR61, were induced in both classes (mixed) (Figure 6C; Table S6).

Figure 6. Distinct YAP targets drive opposite adhesive behavior of binary cancer classes.

(A) Hierarchical clustering of RHO coefficients from rank-rank hypergeometric overlap (RRHO) to compare YAP targets in YAP^on and YAP^off lines.

(B) Hierarchical clustering of KEGG pathways enriched in YAP targets. Numbers: rank (stats in Table S6).

(C) Unsupervised clustering of the top YAP-induced genes (left). Graphs of the top enriched KEGG pathways and ENCODE TFs (right).

(D) YAP/TAZ western blots and adhesion type for indicated cell lines.

(E) PC1⁺ signature scores (left) and levels of select genes (right) in CCLE lines with indicated adhesive behavior.

(F) Proportion of adherent and non/semi-adherent cell lines in cancer classes (left; Fisher’s test); PC1⁺ signature score in the adherent and non/semi-adherent cell lines from clusters B and C (right).

(G) Phalloidin (F-actin) staining of WERI-RB1 Rtb cells transduced with the indicated vectors.

(H) YAP increases adhesion in YAP^off cell lines. n R 3; Fisher’s test: *p < 0.05, **p < 0.01.

(I) Phalloidin staining of retina from aCre mice of the indicated genotypes (abbreviations, Figure 1B). Arrowhead, apical surface; arrows, apical breaks.

See also Figure S5 and Table S6.

YAP^on- or YAP^off -dependent adhesion/ECM genes suggested that binary classes may have distinct adhesive behavior. Indeed, YAP^off cancers from many different tissues were non- or semi-adherent in culture, whereas YAP^on lines were adherent (Figure 6D). The Hodgkin lymphoma line, HDMY-Z, unlike other lymphomas is adherent and it expressed YAP/TAZ at levels seen in YAP^on cancers (Figure 6D). Comparison of hundreds of cell lines showed that PC1⁺ genes exhibit opposite expression across non-adherent versus adherent cell lines (Figure 6E). Among YAP^off lines, all cluster A and most cluster B lines were non-adherent, while most YAP^on lines were adherent, and within clusters B and C, adherence correlated with PC1⁺ gene levels (Figure 6F). Ectopic YAP induced adhesion of human YAP^off cancer lines (Figures 6G and 6H), prompting us to ask if YAP/TAZ influence adhesion in vivo. Basal integrin/ECM signaling engages cytoskeletal components to drive apical constriction required for tissue morphogenesis (Fernandes et al., 2014) and, intriguingly, Yap/Taz loss disrupted the apical surface in the Rb1/Rbl1 null retina at P8, commensurate with the subsequent increase in tumor penetrance in adults (cf. Figures 6I and 1B). Thus, adding to distinct transcriptional, pharmaceutical and genetic profiles, YAP^off and YAP^on cancers also display opposite adhesive properties.

Differential enhancer deployment dictates pro- or anti-cancer YAP/TEAD functions

Next, we investigated the mechanism underlying contrasting YAP targets in binary cancers. In oncogenic contexts, YAP/ TAZ associate with TEADs and AP-1 to promote cancer (Liu et al., 2016; Stein et al., 2015; Verfaillie et al., 2015; Zanconato et al., 2015; Zhao et al., 2008), whereas it binds p73 to drive apoptosis in multiple myeloma (Cottini et al., 2014), a YAP^off liquid cancer. To test YAP-TEAD involvement in suppression of solid YAP^off cancers, we used a mutant (YAP^S94A) that disrupts TEAD binding (Zhao et al., 2008). YAP^S94A failed to induce cytostasis, drug resistance, adhesion, or gene induction in YAP^off cancer cells (Figures 1E, 1F, 2I, 4D, 4H, 6G, 6H, and S5A). In contrast, mutating YAP WW domains, required for binding to p73 and other factors (Piccolo et al., 2014), did not alter YAP-induced cytostasis (Figure S6A). YAP-mediated growth inhibition was linked to reduced S- and increased G2/M-phase cells with minimal apoptosis (Figures S6B and S6C), consistent with p73 independence. We confirmed that TEAD binds ectopic YAP but not YAP^S94A in YAP^off cells (Figures 7A and S6D). Deleting TEAD1, the predominant TEAD in SHP77 SCLC cells blocked, whereas overexpressing TEAD3 or TEAD4 enhanced YAP-mediated cytostasis (Figures 7B and 7C). Moreover, a hybrid protein with TEAD4-DNA-binding domain fused to the VP64-activation domain recapitulated YAP-induced cytostasis (Figure 7D). Thus, YAP suppresses YAP^off cancers through TEAD binding and gene activation, identical to the oncogenic mechanism in YAP^on cancers (He et al., 2019; Kapoor et al., 2014; Ota and Sasaki, 2008).

Figure 7. Enhancer usage dictates opposite YAP/TEAD effects.

(A) TEAD1/YAP coimmunoprecipitation in YAP^on and YAP-transduced YAP^off cells.

(B) TEAD1 deletion prevents YAP-induced cytostasis. Mean ± SD, n = 4.

(C) Ectopic TEAD3 or TEAD4 enhances YAP-meditated cytostasis. Mean ± SD, n = 3.

(D) TEAD4-VP64 impedes YAP^off cell line growth (n = 3, p < 0.001).

(E) Hierarchical clustering of the top 1,000 TEAD4 ChIP-seq peaks from each YAP^on or YAP-transduced YAP^off line.

(F) k-Means clustering of TEAD4 peaks defines 16 enhancer groups.

(G) Unsupervised hierarchical clustering of enriched TF motifs in the 16 TEAD4-enhancer groups. Numbers: motif rank.

(H) Hierarchical clustering of co-bound TFs in the 16 TEAD enhancer groups.

See also Figures S6 and S7 and Table S7.

To understand contrasting YAP/TEAD function, we performed TEAD4 ChIP-seq in four YAP^off lines plus/minus YAP, comparing our results with published TEAD4 ChIP-seq data from four YAP^on lines. YAP enhanced existing binding sites and greatly expanded the target repertoire of TEAD4 in YAP^off cancers, suggesting cooperativity, likely aided by increased TEAD levels (Figures S7A–S7C). We noted tissue-specific TEAD sub-clusters in YAP^off data but, more importantly, YAP^off and YAP^on cancers formed separate superclusters (Figure 7E), indicating distinct targeting in each class.

TEADs cooperate with AP1 to induce cell-cycle genes in YAP^on cancers (Liu et al., 2016; Stein et al., 2015; Verfaillie et al., 2015; Zanconato et al., 2015), thus we asked if they employ distinct factors in YAP^off cancers. For this, we first classified 16 distinct enhancer clusters using k-means clustering (Figures 7F and S7D; Table S7). Most YAP^off enhancers were remote (Figure S7E), similar to YAP^on TEAD enhancers (Liu et al., 2016; Stein et al., 2015; Verfaillie et al., 2015; Zanconato et al., 2015). TEAD-bound clusters 1–4 were common to multiple YAP^off and YAP^on lines, but others were specific to cell lines, cancer types, or cancer classes (Figure 7F). We then searched for features that distinguish YAP/TEAD enhancers in YAP^off and YAP^on cancer classes. Motif analysis revealed TEAD sites at all enhancers, validating the dataset (Figure 7G). However, whereas YAP^on enhancers contained AP1/MafA, FOXM1, or REST motifs, YAP^off enhancers lacked AP1 sites, and were enriched in motifs for lineage-determining basic-helix-loop-helix (bHLH) and homeobox TFs (Figure 7G). To validate these results, we mined 2,829 ChIP-seq datasets covering 485 TFs. Again, unsupervised clustering separated YAP^on and YAP^off enhancers, sub-clustered retino-blastoma and SCLC-specific enhancers, and confirmed enrichment of homeobox (OTX2/PBX3/LHX2/NKX2–1) or bHLH (ASCL1/TCF12/MAX) factors at YAP^off enhancers, but AP1 (JUN/FOS/ATF family) or REST at YAP^on enhancers (Figure 7H). Consistent with tissue of origin, SCLC and retinoblastoma elements bound TFs that influence lung (e.g., NKX) or retinal (e.g., OTX2) development, respectively (Figures 7G and 7H). Thus, while YAP/TEAD cooperate with AP1 in YAP^on cancers, they collaborate with neurogenic bHLH/homeobox TFs in YAP^off cancers, exposing a high-level regulatory rule distinguishing binary cancer classes.

A YAP-integrin axis drives cytostasis in YAP^off cancer cells

To identify genes required to mediate cytostasis in YAP^off cancers we ran CRISPR screens against ~1,000 YAP targets, PC1⁺ genes, and controls. A top hit was integrin-β5 (ITGB5), and validation assays confirmed its loss inhibits YAP-induced cytostasis, not YAP expression or activity (Figures 8A, 8B, S8A, and S8B). ITGB5 is a remarkable hit, because it is a PC1⁺ gene (off in YAP^off, on in YAP^on cancers; Figure 3C), dimerizes with integrin-αV (ITGAV) (Raab-Westphal et al., 2017), a high-rank essential gene in YAP^on cancers (Figure 4I), and was induced by YAP in 4/5 YAP^off cancer cell lines, but unaffected by YAP-loss in 5/5 published YAP^on lines (Figure 6C). We confirmed differential regulation of ITGB5 protein by YAP in multiple YAP^off and YAP^on lines (Figures 8C and S8C). ITGB5 regulation in YAP^on cells may be complex, as it is not YAP dependent, but can be induced by supra-physiological YAP levels (Cordenonsi et al., 2011). Irrespective, our data highlight a class-specific cytostatic YAP-ITGB5 axis, explaining why these genes must be down-regulated in YAP^off cancers.

Figure 8. ITGAV/ITGB5 are effectors of YAP-induced cytostasis.

(A) Scatterplot of fold difference between YAP- and empty vector-transduced or input cells for each single-guide RNA in the CRISPR screen.

(B) ITGB5 knockout rescues YAP-induced cytostasis.

(C) Western blots in YAP-transduced YAP^off or YAP/TAZ-knockout YAP^on lines.

(D) Knockout of ITGAV, but not other integrins, alleviates YAP-induced cytostasis.

(E) ITGAV/B5, but not ITGB1, blocking antibody alleviates YAP-induced cytostasis. All panels: mean ± SD, n ≥ 3. *p < 0.05, **p < 0.01, ***p < 0.001 versus singleguide control or untreated (untr) cells.

(F) Cyclic RGD peptide (cRGD) blocks YAP-induced adhesion (images), but not cytostasis (graphs). Scale bars, 200 µm.

(G) Summary of contrasting YAP/TEAD mechanisms of action in solid YAP^on and YAP^off cancers.

See also Figure S8.

Next, we asked whether the ITGB5 partner, ITGAV, influences growth suppression. Indeed, deleting ITGAV, but not other integrins, rescued cell growth (Figures 8D and S8D). Moreover, an ITGAV/B5, but not ITGB1, blocking antibody alleviated YAP-induced growth suppression in YAP^off lines (Figure 8E). The ITGAV/ITGB5 dimer is thus a key effector of YAP-induced cytostasis in YAP^off cancer cells.

Integrins mediate cell-matrix adhesion, but also have adhesion-independent functions (Ferraris et al., 2014; Seguin et al., 2014). A cyclic RGD peptide that disrupts integrin-ECM interactions prevented adhesion of YAP-transduced YAP^off cells, but did not rescue cytostasis (Figure 8F). These data contrast the genetic and antibody blocking assays above tying ITGAV/B5 to YAP-induced cytostasis. ITGAV/B5-blocking antibodies inhibited YAP-induced adhesion of some (e.g., Y79), but not all (e.g., WERI-RB1), YAP^off lines (Figure S8E). Thus, while YAP drives both adhesion and cytostasis in YAP^off cancer cells, they are uncoupled, focusing attention on adhesion-independent functions of ITGAV/ITGB5. Altogether, our transcriptome, chromatin binding, and GoF and LoF studies reveal that YAP silencing is vital in solid YAP^off cancers to avoid deployment of a cytostatic integrin program centered on ITGAV/B5, contrasting the opposite pro-cancer role in YAP^on cancers (Figure 8G).

DISCUSSION

Over-arching cancer principles can expose therapies with broad relevance. Here, starting with the goal of understanding distinct genetic and adhesive features of small cell versus other solid cancers, we uncovered a binary pan-cancer classification scheme that pivots on contrasting YAP activity. Unlike YAP-dependent YAP^on solid cancers, a large group of neural/NE solid cancers co-silence YAP/TAZ with integrins and other adhesion regulators. Opposite anti- or pro-cancer YAP activity in YAP^off versus YAP^on cancers underlies these transcriptome differences and, mechanistically, a unique YAP/TEAD enhancer repertoire in YAP^off solid cancers deploys integrin-αV/β5 to drive cytostasis. This YAP-centric cancer classification scheme has interesting therapeutic implications, because the binary groups exhibit distinct vulnerabilities, and therapy-driven lineage switching exploits these differences to drive drug resistance.

Small molecules that disrupt YAP/TEAD activity show promise in YAP^on cancers (Crawford et al., 2018; Gill et al., 2018; Kurppa et al., 2020). However, targeting YAP would, by definition, be futile in YAP^off cancers, and in YAP^on prostate and lung adenocarcinomas may favor evolution of drug-resistant YAP^off SCN cancer (Beltran et al., 2016, 2019; Lee et al., 2017; Niederst et al., 2015; Tan et al., 2014). Our LoF and GoF data, in multiple cancer/tissue types, in vivo, in vitro, and in murine and human contexts, expose YAP derepression as a potential avenue to tackle solid YAP^off cancers, which our functional genomics data extends to include the YAP-regulated integrin pair, ITGAV/ITGB5. Methods to derepress these genes await discovery, but our studies also highlight approved drugs that selectively inhibit YAP^off cancers, several of which target metabolic dependencies (e.g., NAD/purine/ ATP synthesis) that correlate with high MYC family activity. High MYC activity provides a logical explanation as to how YAP^off cancers sustain division, contrasting reliance on YAP in YAP^on cancers. We find that expressing YAP reduces sensitivity to YAP^off-selective therapies, predicting YAP^off-to-YAP^on switching as a potential resistance mechanism. YAP expression alone seems insufficient to cause such a switch, because we find it hinders growth of murine retinoblastoma or SCLC in vivo, consistent with YAP presence in benign, cytostatic human retinoma. Conversion to a malignant YAP^on state may require cooperation between YAP and other pathways, such as NOTCH signaling (Ireland et al., 2020; Tlemsani et al., 2020). Irrespective, our binary scheme opens a different perspective on cancer treatment, based on opposite activity of a single transcription complex.

Cancer resistance is the common response to ‘‘oncogenic’’ mutations. For example, RB1 loss predisposes lung neuroendocrine or retinal cone cells to cancer, but these lineages represent only ~0.4% or ~3% of these tissues. We find that the cell of origin in each case is intrinsically YAP deficient, and in cancers where RB1 loss is a late event, we link this genetic event to silencing of YAP and the adhesion gene targets that define binary cancers. Removing YAP/TAZ may sensitize cancer-resistant lineages to transformation upon RB1 loss, as we observed in murine retina, although there are likely additional requirements, such as the elevated MYC/AKT activity plus p53 loss that converts human RB1^–/– epithelia to NE cancer (Park et al., 2018). Notably, cone photoreceptors provide naturally high levels of MYCN and the p53 inhibitor MDM2 (Xu et al., 2009), MYC family amplification and p53 loss is common in SCLC (Rudin et al., 2019), and deleting the AKT inhibitor PTEN promotes murine retinoblastoma and SCLC (Cui et al., 2014; Song et al., 2012; Xie et al., 2015). Why certain RB1^–/– lineages are far more susceptible to transformation than others puzzled the field for decades. Our data expose YAP/TAZ silencing as a key component of the underlying code.

Why TEAD complexes target different enhancers in YAP^off versus YAP^on cancers is an intriguing issue for future studies. The co-bound bHLH/homeobox proteins we identified (e.g., ASCL1, NEUROD1, OTX2) specify YAP/TAZ-deficient lineages that are sensitive to RB1 loss in lung and retina (Borges et al., 1997; Neptune et al., 2008; Nishida et al., 2003). Lineage, therefore, seems to determine opposite YAP/TAZ activity. Potentially, lineage-determining bHLH/homeobox proteins may compete with AP1, consistent with YAP/TEAD/homeobox protein interactions in embryonic stem cells (Beyer et al., 2013). bHLH neurogenic proteins are indispensable in small cell cancers (Rudin et al., 2019), providing a logical explanation as to why YAP/ TAZ must be silenced in these cancers. Extensive proteomics in YAP^off cancers will be required to test this model.

In summary, our results show that YAP and TAZ profoundly influence context-specific responses to RB1 loss, expose a large YAP^off cancer class with pharmaceutical, genetic, metabolic, and adhesive profiles that contrast YAP^on cancers, and pinpoint alternate enhancer engagement and a YAP-integrin axis as a key component of opposite molecular behavior in YAP^off cancers. The data highlight multiple therapeutic avenues for YAP^off cancers, including untreatable SCN cancers of the prostate and lung.

STAR☆METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Rod Bremner (bremner@lunenfeld.ca).

Materials availability

Vectors generated in this study can be obtained from the lead contact.

Data and code availability

• RNA-seq and ChIP-seq data have been deposited at GEO under accession codes GSE144972 and GSE144973. This paper analyzes existing, publicly available data. These accession numbers for the datasets are listed in the key resources table.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell culture, PDXs and virus production

The retinoblastoma cell lines RB1021, RB3823 and RB247 (kindly provided by Brenda Gallie) were grown in Iscove’s media supplemented with 10% FBS, 0.0004% (v/v) β-mercaptoethanol, and 0.010 mg/mL insulin (Thermo Fisher) (To et al., 2012). Y79 and WERI-RB1 retinoblastoma cells, SH-SY5Y and SK-N-SH neuroblastoma lines (kindly provided by Meredith Irwin), along with the DU4475 breast cancer (kindly provided by Eldad Zacksenhaus), BPH-1 (benign prostate), HCC1419 (breast cancer), and NCI-H526 (SCLC) cell lines were cultured in RPMI 1640 media supplemented with 10% FBS. SHP77, NCI-H510A and NCI-H82 SCLC lines, as well as NCI-H661 and NCI-H1650 NSCLC cells were maintained in RPMI 1640 media containing 10% FBS and 10mM HEPES. NCI-H209, NCI-H69 and NCI-N417 SCLC cells were grown in RPMI 1640 supplemented with 7.5% FBS. NCI-H2171 SCLC cells were cultured in DMEM:F12 media with 5% FBS, 0.005 mg/ml insulin, 0.01 mg/ml bovine transferrin (Thermo Fisher), 30 nM sodium selenite (Sigma), 10 nM hydrocortisone (Sigma), 10 nM β-estradiol (Sigma) and extra 2mM L-glutamine (for final conc. of 4.5 mM). NCI-H660 NEPC cells were maintained in RPMI 1640 supplemented the same as NCI-H2171 culture media. The NE pancreatic cell lines, QGP-1 and BON-1 (kindly provided by Dr. Todd McMullen), were grown in DMEM media supplemented with 10% FBS, 0.01 mg/mL insulin, 10 nM hydrocortisone, 0.005 mg/mL transferrin, 10 ng/ml Somatostatin (Sigma) and 10 ng/ml Glycyl-L-histidyl-L-lysine (Sigma). A549 (NSCLC), SW-1573 (NSCLC), MCF7 (breast cancer), MDA-MB-231 (breast cancer), DU145 (prostate cancer), PANC-1 (pancreatic adenocarcinoma), Hs 766T (pancreatic adenocarcinoma) and Lenti-X 293T cells were cultured in DMEM media with 10% FBS. T24 bladder cancer cells were cultured in McCoy’s 5a media with 10% FBS. ALK+ Anaplastic Large Cell Lymphoma (Karpas 299, SUP-M2, SU-DHL-1, SR-786 and UCONN) and the HDMY-Z Hodgkin lymphoma cell lines were cultured in RPMI-1640 containing 10% FBS and 50 µM 2-mercaptoethanol, while Hodgkin lines KM-H2, L428, L540 and HDLM-2 were cultured in similar media with 20% FBS. The leukemia lines OCI-AML2 and OCI-AML3 were cultured in AMEM supplemented with 10% FBS and THP-1, HL-60 and K562 in IMDM with 10% FBS. All cells were maintained at 37^◦C and 5% CO₂, and were routinely screened for mycoplasma contamination by PCR.

SCLC PDXs (kindly provided by Charles Rudin) have been described previously (Gardner et al., 2017; Lok et al., 2017). Lentiviruses were generated using Lenti-X 293T cells and the psPax2 and pMD2.G packaging vectors. Following collection, virus was concentrated by ultracentrifugation. Retroviruses were produced using the Phoenix-eco packaging line and concentrated by ultracentrifugation.

Human retinal explants and generation of a human cone cell line

Human fetal retinas were obtained from the Morgentaler Clinic in Toronto with approval from the Research Ethics Board (REB #13–0132-E) of Mount Sinai Hospital in Toronto, Canada. All donors read the consent form approved by the REB before surgical procedures, and voluntarily donated fetal eye samples. The gestational age was estimated by a combination of clinic intakes, ultrasound, crown-rump, and fetal foot length measurements where possible (FitzSimmons et al., 1994; Shepard, 1975). Eye samples collected were held on ice for up to 6 hr in retina culture medium (IMDM with 10% FBS and 1X Antibiotic-Antimycotic, Life Technologies).

FW17 retina were dissected in sterile IMDM and cut radially. Tissue fragments were transferred onto cell-culture inserts (Millipore) with photoreceptor side down (Jin and Xiang, 2012). Inserts with retinal fragments were quickly put in 6-well plates with 1300 µl of prewarmed retina culture medium, and incubated at 37^◦C with 5% CO₂. Explants were transduced the next day with CRISPR/Cas9 lentivirus expressing a control, non-targeting or RB1-targeting sgRNA. Three weeks later, explants were harvested and processed for immunostaining.

To generate a human cone cell line, a Fw17 eye was sterilized in 70% ethanol for 3 seconds, rinsed twice in cold phosphate-buffered saline (PBS), and transferred to IMDM for retinal dissection. Retina was dissociated with Papain Dissociation System (Worthington Biochemical, Cat# LK003150). Briefly, retina was incubated in papain solution for about 15 min at 37^◦C and 5% CO2, with gentle pipetting every 5 min. After dissociation to ~20-cell clusters, cells were suspended in 10 volumes of sterile phosphate buffered saline (PBS), pelleted by centrifugation at 300g for 10 min, and washed with PBS, followed by digestion in 0.05% trypsin/EDTA with gentle pipetting to produce single cell suspension. Cells were washed in 10 volumes of retina culture medium, resuspended in culture medium, then transduced with shRB1 lentivirus and replicating cells selected. Lineage was confirmed using the expression of the cone markers RXRγ, and L/M Opsin.

Mouse lines and experiments

All animal experiments were conducted with ethical approval from the respective animal care committees of each research institute, and all experiments conformed to the relevant regulatory standards. All mice were maintained on mixed backgrounds, and siblings utilized in comparisons.

Retina

α-Cre;Rb1^f/f;Rbl1^−/− (Chen et al., 2004) and Yap^f/f;Taz^f/f (Reginensi et al., 2013) mice were interbred to generate α-Cre;Rb1^{f/ f};Rbl1^−/−;Yap^f/f;Taz^f/f experimental mice and controls used for survival analysis (Figures 1A, 1B, and S1A) and apical polarity studies (Figure 6I). Mice were a mixed 129Sv x C57BL/6J genetic background. Littermates from multiple breeding pairs were assessed, e.g. To compare Rb1^f/f;p107^−/− and Rb1^f/f;p107^−/−;Yap^f/f mice we used litters from several aCre;Rb1^f/f;p107^−/−;Yap^f/+x Rb1^f/f;p107^−/−;Yap^f/+breeding pairs. Subretinal injections (Livne-Bar et al., 2006) of Cre-GFP or YAP-Cre-GFP retroviruses were performed on newborn P0 Rb1^f/f or Rb1^f/f;Rbl1^−/− pups, which were then harvested for analysis at the indicated time points (Figures S1B–S1E).

Lung

We used six models to study cell-of-origin and/or cancer in lung. The Cgrp^CreER;Rb1^f/f;p53^f/f SCLC model (Figures 2F and S2C) is described in (Song et al., 2012), and was crossed with Rb1^f/f;p53^f/f;ROSA-rtTA^+/+;TRE-Yap^+/− mice (Gregorieff et al., 2015) to generate Rb1^f/f;p53^f/f;Cgrp^CreER+/−;ROSA-rtTA^+/−;TRE-Yap^+/− or TRE-Yap^−/− experimental mice (Figures 2K and S2J). 5–7 wk old mice were injected IP with 0.225 mg/g tamoxifen (Sigma) daily for four consecutive days, then placed on a continuous doxycycline-containing diet. Mice were harvested 3 months after tamoxifen treatment and lungs processed for frozen sections. For Ad-Cre induction of Rb1/p53 null SCLC, combined with Yap/Taz loss (Figures 2L and S2N) or Yap transgene induction (Figures S2K–S2M), Ad5CMVCre (University of Iowa Viral Vector Core Facility, Iowa City, IA) was delivered to the lungs of P2–3 mice of the indicated genotypes by intranasal inhalation (10 µL at 1×10⁸ pfu). For EGFR-driven NSCLC (Figure 2G) we used the previously described CcsprtTA;tetO-EGFR^L858R NSCLC model (Politi et al., 2006; Tichelaar et al., 2000). At weaning, mice were fed chow containing doxycycline, activating rtTA and inducing expression of EGFR^L858R in lung respiratory epithelial cells. Moribund mice were euthanized and tissues harvested. To compare Yap expression in the latter NSCLC model with that in a SCLC model (Figure 2G), we delivered Ad5CMVCre to the lungs of 6–8 week old Pten^f/+;Rb1^f/f;p53^f/f mice by intratracheal injection (50 µL at 1×10⁸ pfu), as described previously (Cui et al., 2014; DuPage et al., 2009). To obtain a mixed model of NSCLC and SCLC (Figure 5A), we used the same intratracheal injection technique, but in Pten^f/f;Rb1^f/f;p53^f/f mice. Finally, we studied Yap/Taz expression in Rb1/p53 null E18.5 embryonic lung using SPC-rtTA;tetO-Cre;Rb1^f/f;p53^f/f mice (Figures 2H and S2B), treated as described previously (Akeno et al., 2015; Perl et al., 2002a; 2002b; Simpson et al., 2009). Briefly, timed pregnant females were fed doxycycline-containing diet throughout gestation to induced Cre and Rb1;p53 gene deletion. E18.5 embryos were then harvested and lungs processed for immunostaining.

Prostate

We used three models of prostate cancer (Figures 5B–5E). Male PB-Cre4; Pten^f/f (SKO) mice were used to model prostate adenocarcinoma, and male PB-Cre4;Pten^f/f;Rb1^f/f (DKO) or PB-Cre4;Pten^f/f;Rb1^f/f;p53^f/f (TKO) mice were used to model NEPC. These models are described in (Ku et al., 2017; Lesche et al., 2002; Sun et al., 2006; Wang et al., 2003; Wu et al., 2001; Zhou et al., 2006). Briefly, recombination of floxed alleles was induced by the Cre4 transgene under control of the prostate epithelial cell-specific probasin promoter (PB-Cre4).

Confirmation of recombined floxed alleles

Cre-mediated recombination of floxed Rb1, p53, Yap and Taz alleles was confirmed in retina and lung models by PCR (data not shown, primers and diagnostic fragment sizes are in Table S8). Tissue was scraped from 14 µm-thick OCT-embedded frozen tissue sections on glass slides using a 30 G x 1/2” hypodermic needle and extracted by alkaline lysis (25 mM NaOH, 0.2 mM EDTA, pH 12) with heating at 95^oC for ~1 hr followed by neutralization (40 mM Tris-HCl, pH 5). Rb1 and p53 primers have been described (Meuwissen et al., 2003), and all primer sequences are listed in Table S8. PCR was performed with 500 nM each primer using REDExtract-N-Amp PCR ReadyMix (Sigma) and the following cycling conditions: 94^oC for 3 min; 35 cycles of 94^oC for 0:40 min, 60^oC (Rb1, p53, Yap and Taz F1-R1) or 63^oC (Taz F2-R2) for 0:40 min, 72^oC for 1 min; 72^oC for 10 min. PCR products were resolved on 2% agarose gels.

METHOD DETAILS

Plasmid cloning

Lentiviral vectors to express YAP, TAZ and TEADs were generated using a modified pLKO.1 backbone (Stewart et al., 2003) from which the U6 promoter and shRNA cassette were removed. For YAP expression in retinoblastoma cells and TAZ in all lines the original hPGK promoter was used. For all other constructs the hPGK was replaced with an EFS promoter. The puromycin (puro) resistance gene in pLKO was replaced with IRES-GFP-T2A-puro or IRES-mCherry sequence. YAP, TAZ or YAP mutants were cloned upstream of the IRES-GFP while TEAD and TEAD4-VP64 constructs were cloned into the mCherry vector. Retroviruses for subretinal injections utilized the pMXIE backbone expressing Yap-IRES-CreGFP or CreGFP alone. CRISPR sgRNA sequences are listed in Table S8 and were cloned into the LentiCRISPR v2 vector (Addgene #52961) (Sanjana et al., 2014). The RB1 shRNA (sequence: CCACATTATTTCT AGTCCAAA, TRCN0000040163) was cloned into the pLKO.1 lentiviral vector modified to carry a GFP-puromycin selection cassette.

Growth curves and soft agar assays

For YAP/TAZ knockouts, cells were transduced with control or a YAP/TAZ sgRNA virus and selected in puromycin. Knockout was confirmed in bulk populations by Western blotting and/or genomic DNA sequencing followed by TIDE analysis (Brinkman et al., 2014) 8–10 days after selection, and cells plated for growth curves. Final cell counts were conducted 9 days after plating. New knockout populations were generated for each biological replicate.

For ectopic expression of YAP, TAZ or TEAD4-VP64, cells were transduced so that >90% expressed the indicated gene as determined by YAP, GFP and/or mCherry staining and flow cytometry and/or fluorescent microscopy. For growth curves, cells were seeded five days (seven days for NCI-H660) after viral transduction, and then counted 10 days later for all lines except slower-growing H660 NEPC cells which were counted 5 wks later. For sgYAP (rescue), sgTEAD1, or sgITG experiments in SHP77 and/or Y79 cells, stable bulk knockout populations were generated, then transduced with control or YAP virus as above. Final cell counts were conducted 15 days after YAP virus transduction. At least three independent knockout populations were generated for each sgRNA. For experiments with co-expression of YAP and TEAD3 or TEAD4, SHP77 cells were transduced with control, YAP-IRES-GFP or TEAD-IRES-mCherry viruses and the percentage of GFP/mCherry double-positive cells was assessed by flow cytometry after 5 days (plateau of YAP/TEAD expression) and 19 days on a Beckman Coulter Gallios flow cytometer and Kaluza analysis software.

For Integrin blocking experiments, cells were transduced with control (Empty) or YAP lentiviruses, then the following day left un-treated or treated with ITGAV/B5 (sc-81632) or ITGB1 (sc-13590) blocking antibodies at 2.5 µg/ml. Fresh antibody was added to the cells every 3–4 days and final cell counts conducted 15 days after YAP transduction.

For soft agar assays, 1×10³ Y79, SHP77 or WERI-RB1 cells were seeded in 0.4% 2-hydroxyethyl agarose (Sigma), which was overlaid on 0.6% bottom agar in a 24-well plate. After 3–4 weeks, cells were fixed in MeOH and stained with crystal violet.

Cell cycle profiling by flow cytometry

SCLC and retinoblastoma cell lines transduced with YAP-GFP or control GFP viruses were pulsed with EdU (10 µm) for 20 min (Y79/ SHP77) or 30 min (WERI/H2171), harvested and fixed in 4% PFA for 10 min. After permeabilization, cells were stained for goat anti-GFP (1 hr), washed and stained using a donkey anti-goat Alexa488 secondary antibody (30 min). EdU was then labelled for 30 min using standard Click chemistry with an Alexa647-conjugated azide (Thermo Fisher). After washing cells were counterstained using the FxCycle violet DNA dye (Thermo Fisher) and analyzed on a Beckman Coulter Gallios flow cytometer and Kaluza analysis software.

Drug studies

For drug studies, six YAP^off (WERI-RB1, NCI-H82, NCI-H69, NCI-H2171, DU4475 and SH-SY5Y) and six YAP^on (NCI-H661, NCI-H1650, SW-1573, PANC-1, DU145 and MDA-MB-231) lines were used. Cells were seeded in 96-well plates ~24 hrs prior to treatment, then left untreated or treated in triplicate with dose curves of the indicated inhibitors (all from Selleck Chemicals). After 72 hrs, viability was measured using Alamar blue. To test the role of YAP in drug sensitivity, cells were transduced with the indicated YAP vectors as above, then seeded five days after transduction and treated as above.

CRISPR/Cas9 screen

A custom pooled sgRNA library of four sgRNAs per gene and 50 non-targeting controls (Table S8) was cloned into the LentiCRISPR v2 lentiviral backbone. sgRNA sequences were from the human Brunello library (Doench et al., 2016) and non-targeting controls from GeCKOv2 (Sanjana et al., 2014). Y79 cells were transduced with the pooled library (MOI = 0.3), then four days later selected in puromycin (2 mg/ml). Six days later, an input gDNA sample was collected, then cells transduced with control (empty vector) or YAP lentiviruses. After 15 days gDNA was collected and was extracted using the Qiagen DNEasy Blood and Tissue Kit and sgRNA sequences amplified with primers containing Illumina i5 and i7 adaptor barcodes. Gel-purified PCR products were quantified by qPCR using SsoAdvanced Universal SYBR Green Supermix (BioRad) on a BioRad CFX96 Touch Real-Time PCR Detection System, and loaded onto an Illumina NextSeq 500 running at 22 dark cycles and 26 light cycles with Illumina NextSeq 500/550 Hi Output Kit v2.5 (75 Cycles). Real-time base call (.bcl) files were converted to FASTQ files using Illumina bcl2fastq2 conversion software v2.17 (on CentOS 6.0 data storage and computation linux servers).

Western blotting and immunoprecipitations

Cells and tissues were lysed in RIPA lysis buffer containing protease/phosphatase inhibitor cocktail (Cell Signaling Technologies) and PMSF, followed by brief sonication then centrifugation to remove insoluble material. Equal protein amounts were resolved by SDS-PAGE, transferred to nitrocellulose membrane, and then probed with the indicated antibodies (see key resources table and Table S8). Blots were imaged using a Li-Cor Odyssey Infrared imaging system. Immunoprecipitations were performed (Lee et al., 2013) with 2 mg of control or anti-TEAD1 antibody and Protein G agarose beads.

Immunostaining

For immunofluorescence on human retinal explants, murine retina and murine embryonic lung samples were harvested and fixed in 4% paraformaldehyde for 1hr (murine retina) or overnight (human retina and mouse lung) at 4^◦C, dehydrated in 30% sucrose over-night and then processed for frozen sections. For adult mice, lungs were inflated with 4% PFA and fixed in 4% PFA overnight, then washed and dehydrated in 30% sucrose overnight prior to processing for frozen sections. After heat-assisted antigen retrieval in pH 6.0 citrate buffer (except human retinal explant samples, which were stained without antigen retrieval), sections were stained with the indicated primary antibodies overnight (see key resources table and Table S8), followed by detection with fluorescently-conjugated secondary antibodies. For Yap or Yap/Taz staining in the mouse lung, the Tyramide SuperBoost amplification kit (Thermo Fisher) was used for detection as per the manufacturer’s protocol. FFPE human retinoblastoma/retinoma samples were deparaffinized and rehydrated, then stained as above.

Immunohistochemistry (IHC) was performed on FFPE tissue microarrays or PFA-fixed mouse tissues following standard procedures. Analysis of murine SCLC +/− Yap induction and scoring of human SCLC, NSCLC PDXs and NET TMAs was performed by expert pathologists (TW, AKW). De-identified prostate tumor tissues were provided under an IRB approved protocol at Roswell Park Cancer Center. FFPE tissues were sectioned at 4 mm for immunohistochemical and multispectral immunostaining with indicated antibodies (see key resources table and Table S8). All stains were validated using recommended tissue controls and evaluated for correct compartment specific staining by pathologists. Staining was performed with a BondRX (Leica Biosystems) automated stainer. Multispectral staining was performed after antibodies were optimized for standard immunohistochemistry and uniplex immunofluorescence staining. Slides were imaged using the Vectra Polaris spectral imaging system (PerkinElmer). Each fluorophore from Opal Polaris reagent kit was visualized using MOTiF™ technology, which generates unmixed whole slide scans at 40x magnification using a separate filter cube corresponding to its emission wavelength. The images were unmixed using a spectral library and individual fluorophores were separated by the InFormTM software. The cell populations were visualized using cell segmentation and phenotype cell tool from Inform 1.1 (PerkinElmer). Threshold for positive staining and accuracy of phenotypes were confirmed by pathologist supervision (AKW).

RNA extraction, qPCR and RNA-seq

Five days after transduction with Empty vector, wild type or S94A YAP (SHP77 and Y79), or Empty vector or 5SA YAP (H209, H2171 and WERI), cells were harvested and RNA extracted using the RNeasy mini kit (Qiagen). For qPCR, cDNA was generated using Superscript II Reverse Transcriptase and qPCR performed using Power SYBR Green PCR master mix (Thermo Fisher) on a Bio-Rad C1000 Touch Real Time PCR system. The primers used are listed in Table S8.

For RNA-seq, RNA was assessed using a Fragment Analyzer, and high quality total RNA was subjected to library preparation using the Illumina TruSeq Stranded mRNA Library Preparation Kit according to the manufacturer’s instructions. Library fragment size was checked using an Agilent Fragment Analyzer, and then quantified with qPCR using SsoAdvanced Universal SYBR Green Supermix (BioRad) on a BioRad CFX96 Touch Real-Time PCR Detection System (WERI, H209 and H2171 lines) or using KAPA SYBR FAST Universal 2X qPCR Master Mix (Kapa Biosystems) running in 7900HT Fast Real Time PCR System (Applied Biosystems) (Y79 and SHP77 cells). Quality checked libraries from WERI, H209 and H2171 cells were loaded onto an Illumina NextSeq 500 run with Illumina NextSeq 500/550 Hi Output Kit v2.5 (75 Cycles). Real-time base call (.bcl) files were converted to FASTQ files using Illumina bcl2fastq2 conversion software v2.17 (on CentOS 6.0 data storage and computation linux servers). Libraries from Y79 and SHP77 cells were loaded on a flowcell for cluster generation using c-Bot and TruSeq PE Cluster Kit v3 (Illumina), and sequencing performed using a HiSeq2000 system and the TruSeq SBS Kit v3 (pair-ended 200 cycles, Illumina). The real-time base call (.bcl) files were converted to fastq files using CASAVA 1.8.2 (Illumina, on CentOS 6.0 data storage and computation linux servers).

For analysis of RNA-seq data, FASTQ files were assessed by fastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc) for sequence quality before further processing and analysis. RSEM (Li and Dewey, 2011) was used for transcript quantification, with reference genome GRCh37.p13, and the resulting gene expression, as a matrix of raw counts, was applied to DESeq2 (Love et al., 2014) for differential gene expression analysis in each cell line. Gene expression data of YAP^on cancer cell lines was obtained from GEO datasets (GSE54617, GSE66949, GSE66082, GSE68599, GSE56157) and differential expression analysis was obtained using the GEO online tool, GEO2R, for each microarray dataset.

To compare YAP regulated genes in different cells across studies on distinct platforms, we used ranked gene lists. We first defined a set of common genes for this study by overlapping official gene symbols annotated from different platforms. Then we ranked genes by signed log2 p-values of differential expression. Rank Rank Hypergeometric Overlap (RRHO) was performed to compare similarity of the ranked gene lists using the R package RRHO and RRHO2 (Cahill et al., 2018; Plaisier et al., 2010), and the Spearman’s rank correlation coefficient (Spearman’s rho) between each pair of ranked gene lists were clustered and displayed as a heatmap using the R package ‘‘NMF’’ (Gaujoux and Seoighe, 2010). Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) was conducted on the pre-ranked gene lists to identify biological processes and pathways regulated by YAP. The gene sets used in this analysis were obtained from Enrichr (Chen et al., 2013; Kuleshov et al., 2016). Terms with frequencies <10 or >2000 were excluded. The top 10 enriched terms from each of the 10 cell lines were merged and clustered by their Normalized Enrichment Score (NES) values. We also compared our RNAseq data in YAP^off lines with an additional four published transcriptome sets from YAP^on lines, generated using RNAseq data from A549 & H1299 NSCLC lung cancer (siRNA depletion of YAP, GSE151200), CAOV2 ovarian cancer (siRNA depletion of YAP, GSE146353), and mel537 melanoma (siRNA depletion of both YAP & TAZ, GSE146918), and obtained the same conclusions using either RRHO or GSEA.

For analysis of top YAP-induced genes we used a ranked list consisting of the top and bottom 2000 ranked genes from the above ranked list. We then selected and merged the top 100 genes from each cell line. To identify targets induced in multiple contexts, we removed genes that were not ranked in the top 500 from at least three different cell lines, which resulted in a final list of 339 genes; the median number of genes induced in YAP^off or YAP^on datasets was > 550, thus 500 was chosen as a conservative cutoff for target selection. The rank of those 339 genes from each cell line was then clustered using the aheatmap function in the R package ‘‘NMF’’. Gene lists from each of the 3 major clusters (class-specific or mixed) were then subjected to Enrichr analysis to obtain top KEGG and ENCODE TF binding terms.

ChIP, MeDIP and ChIP-seq

For TEAD4 ChIPs, control or YAP-expressing SCLC and retinoblastoma lines were harvested 5 days after viral transduction and processed for ChIP (Ni and Bremner, 2007; Ni et al., 2008) with 2 µg of antibody and 20 µl of Dynabeads (sheep anti-mouse or anti-rabbit, Invitrogen). ChIP for histone modifications or RNA POLII were similarly performed using parental cells.

For methylated DNA IPs (MeDIP), RNase-free genomic DNA was sonicated to generate 100–400bp fragments, PCR purified/ concentrated, then heat denatured (95^◦C for 1 min) and cooled on ice. Approximately 5 µg of sheared DNA was added to 5mC-antibody (10 µg) pre-bound to Dynabeads in IP buffer (10 mM Na₃PO₄ pH 7.0, 140 mM NaCl, 0.05% Triton X-100). IPs were incubated 4 hrs at 4^◦C, washed 3 × 10 min in IP buffer and then eluted. Samples were then digested with Proteinase K (2 hrs at 50^◦C) and DNA purified.

For ChIP or MeDIP-qPCR, samples were analyzed by absolute quantification using Power SYBR Green PCR master mix (Thermo Fisher) on a Bio-Rad C1000 Touch Real Time PCR system. The primers used are listed in Table S8. ChIPs were corrected for control IgG samples and calculated as a percent of the input chromatin.

For TEAD4 ChIP-seq, sequencing libraries were prepared using the Illumina TruSeq ChIP Sample Prep Kit (cat#: IP-202–101) according to the manufacturers protocol with 18 cycles for final PCR enrichment. Samples were size selected using PippinHT (Sage Science) with 2% gel, targeting 300 bp fragments (start at 239 bp, end at 361 bp, and tight on Range Flag). Sizes were confirmed using an Agilent Fragment Analyzer and quantified by qPCR using SsoAdvanced Universal SYBR Green Supermix (BioRad) on a BioRad CFX96 Touch Real-Time PCR Detection System. Quality checked libraries were loaded onto an Illumina NextSeq 500, run with Illumina NextSeq 500/550 Hi Output Kit v2.5 (75 Cycles). Real-time base call (.bcl) files were converted to fastq files using Illumina bcl2fastq2 conversion software v2.17 (on CentOS 6.0 data storage and computation linux servers).

The sequence quality of ChIP-seq data were evaluated using fastQC and then aligned to the human reference genome build hg19 by using Bowtie2 (Langmead and Salzberg, 2012) with default parameters. Redundant reads were removed and only uniquely mapping reads were kept for further analysis using SAMtools (Li et al., 2009). To assess the consistency of replicate experiments and identify possible external and/or contamination across different cell lines, the association between duplicate samples within each line and across all four cell lines was examined using the Spearman Correlation of read counts in 10 kb bins over the entire genome after excluding the genomic regions on the ‘‘Black List’’ (Amemiya et al., 2019) using DeepTools (Ramírez et al., 2016).

The Homer package (Heinz et al., 2010) version 4.10 was used for peak calling and de novo motif discovery of the ChIP-seq data. We first called peaks in individual replicates using input as background control in order to assess the consistency between replicates, the resolution of the called peak regions, the IP efficiency and specificity before further analysis. Next, we generated a high-confidence unified peak set with two replicates of each condition in H209, WERI and Y79 cells by calling the getDifferentialPeaksReplicates function. For H2171 cells, only the peaks called in replicate1 were used. The peak regions on the ‘‘Black List’’ were excluded from further analysis. Normalized signal coverage (fragment pileup per million reads, SPMR) was calculated using the MACS2 (Zhang et al., 2008) callpeak function and background signals were subtracted from IP with the bdgcmp function. The result bedGraph files were converted to bigwig files with the UCSC tool bedGraphToBigWig (Kent et al., 2010). The bigwig files were used to display TEAD4 signal coverage with the Integrative Genomics Viewer (IGV) (Robinson et al., 2011) and for further clustering analysis and heatmap.

To compare TEAD4 genomic binding events between two conditions (empty vector control and YAP-expressing cells), we overlapped unified peaks from the two conditions in each cell line using the Homer’s mergePeaks function to obtain a list of merged unique peak regions. We then mapped the normalized signals in bigwig files from each replicate into 10 bp bins of a 300 bp window centered at each of the merged regions for each replicate in each condition correspondingly to get a set of signal coverage matrices. For each line, the matrices were then plotted as heatmaps of signals in: 1) regions sorted by average signal values per region in YAP-expressing cells, and 2) regions clustered by K-means algorithm. Examples of the clustered regions were displayed in the IGV.

To compare TEAD4 binding in YAP-expressing YAP^off cells to YAP^on cells, we compared our data with published TEAD4 ChIP-seq data in four YAP^on cells: A549, HCT116, HepG2 and MDA-MB-231. The aligned reads for A549, HCT116 and HepG2 cells were downloaded from the ENCODE project (http://genome.ucsc.edu/ENCODE/downloads.html) and data for MDA-MB-231 was obtained from GEO dataset GSE66081. A set of unified peaks was called with replicates in each cell using the same parameters as described above. We selected the top 1000 peaks ranked by peak score for each cell line and overlapped them into a list of merged unique peak regions as described above. The normalized genomic signal coverage in bigwig file was obtained by using pooled replicates for each cell, except in H2171 where only replicate1 was used. We evaluated the similarity of genomic TEAD4 binding in each pair of cells by their co-localized binding intensity. To achieve this, we first subset the merged peak set by filtering out singletons, which did not overlap with peaks from other cells, and then generated a matrix of binding intensity scores with overlapped peak regions as rows and cells as columns. Each score was calculated as the average signal coverage in a 200 bp window centered at peak center and scaled into range 0 to 1 in each column by using the rescale function of the R package ‘‘scale’’ (https://scales.r-lib.org). The score matrix was ordered in both columns and rows by hierarchical clustering using the “euclidean” distance and displayed as heatmap with a dendrogram to show the relationship of the cells by their TEAD4 bindings. The aheatmap function of R package ‘‘NMF’’ (Gaujoux and Seoighe, 2010) was used for the clustering and heatmap plotting.

To further explore putative co-bound transcription factors at TEAD4 sites, we first classified the entire merged peak regions into 16 clusters by K-means clustering of TEAD4 binding signals mapped into 10 bp bins of a 300 bp window centered at peak center. We use the ReMap online database and annotation tool (Chè neby et al., 2018) to define overlap between TEAD4 sites and 485 TFs from ENCODE and published studies in each of the 16 clusters. The statistical significance of the enrichment was computed by comparing the number of overlapped regions with the number of overlaps obtained using random regions sampled with the same size and number as the regions in each TEAD4 binding group. We ranked the co-bound TFs by their enrichment p-values and selected the top 10 enriched TFs in each group to form a matrix of ranks with the TFs as rows and clusters as columns. We then used the previously described hierarchical clustering and heatmap to show association of the co-bound TFs with TEAD4-bound clusters. We also conducted de novo motif discovery in each of the 16 TEAD4-bound clusters.

Pan-cancer transcriptomics analysis

Microarray gene expressions data of cancer cell lines, ‘‘CCLE_Expression_Entrez_2012–09-29.gct’’ (18988 probes and 1037 samples) and ‘‘CCLE_sample_info_file_2012–10-18.txt’’ were downloaded from the Cancer Cell Line Encyclopedia (CCLE; https://portals.broadinstitute.org/ccle) (Barretina et al., 2012). Outliers and samples with missing values were removed. Gene expression values were obtained from the mean expression values of probes collapsing by corresponding genes. Genes with zero expression in all samples were removed, and the remaining 1036 samples with expression data for 18900 genes were used for further analysis. Principle component analysis (PCA) was applied on standard scaled data of genes using the ‘‘prcomp’’ function of R. YAP and TAZ expression in each sample, as z-scores, was projected onto the PC1 and PC2/PC3 dimensions to display the pattern of YAP expression across the cell lines distributed in the 2D space. We found that PC3 performed better than PC2 to separate two distinct YAP^off groups. Genes with rotation values >0.018 and <−0.018 in PC1 and PC3 were selected as features for further clustering analysis. Using the scaled expression values of the feature genes, samples were classified into four clusters with the ‘‘k-means’’ function of R and displayed as a heatmap using the ‘‘aheatmap’’ function of the R package NMF (Gaujoux and Seoighe, 2010). The frequency of the ‘‘Site Primary’’ and the ‘‘cancer type’’ of the samples in each super-cluster were displayed as bar charts. The ‘‘cancer type’’ was defined based on the ‘‘Hist Subtype1’’ and combined with the ‘‘Site Primary’’ and ‘‘Histology’’ information. The distribution of YAP and TAZ expression in each cluster (A-D) was displayed as a box plot. The GSEA analysis was applied to pre-ranked gene lists by using PC1 and PC3 rotation values. Mutational data for the four clusters (A-D) were obtained from cBioportal (Cerami1 et al., 2012; Gao et al., 2013). The MYC signature was based on ‘‘validated targets of c-Myc transcriptional activation’’ from the NCI-Nature Pathway Interaction Database, and the MYCN signature was calculated from shMYCN knockdown in IMR32 neuroblastoma cells (Valentijn et al., 2012). Scores were applied to CCLE cell line data as per (McCurdy et al., 2017). Groups were compared using a pair-wise Wilcoxon Test with Bonferroni correction for multiple testing (Figure 2L).

RNA-seq gene expression data from primary tumor samples from TCGA (Cancer Genome Atlas Research Network et al., 2013) and TARGET (Ma et al., 2018) data sets, ‘‘TcgaTargetGtex_RSEM_Hugo_norm_count.txt’’ and ‘‘TcgaTargetGTEX_phenotype.txt’’ were downloaded from UCSC Xena (https://xena.ucsc.edu/). Only cancer samples were extracted for this study. Outliers and samples with missing values were removed. The remaining 10,485 samples were applied for further analysis. The top PC1/PC3 genes mapped above were extracted and scaled for k-means clustering into four clusters. The clustered gene expression matrix with the features was displayed as a heatmap as before. Bar charts for frequencies of the ‘‘Site Primary’’ and the ‘‘cancer type’’ of the cancer samples in each super-clusters were similarly plotted as before. The ‘‘cancer type’’ was defined by the “detailed category” in the phenotype. All plotting except the heatmaps was conducted by using the R package ‘‘ggplot2’’ (Wickham, 2016).

Tissue-specific transcriptomics analysis

Analysis of YAP/TAZ mRNA in human lung cancer subtypes (Figure 2C) was extracted from (Clinical Lung Cancer Genome Project (CLCGP) and Network Genomic Medicine (NGM), 2013) and subtypes with <3 samples were excluded. Statistical analysis was performed using a Kruskal-Wallis H-test (YAP H-value: 99.8, p-value: 5.97 × 10⁻¹⁷; TAZ H-value: 82.7, p-value: 1.48 × 10⁻¹³) followed by Conover’s post-hoc test with a Holm p-value adjustment. Yap/Taz mRNA levels in murine lung cancers (Figure 2E) were extracted from (Semenova et al., 2016). For analysis of the SCN subset in the TCGA (Figure 3I), previously defined SCN samples (Balanis et al., 2019) were used. Cancer types with >2 SCN samples with a SCN score of >50 were included and YAP/TAZ expression was compared to the remaining non-SCN samples for each cancer type.

For the analysis of transcriptome changes during lineage switching (Figure 5) we used murine and human datasets. DESeq2 normalized transcriptome data from murine prostate cancer (Ku et al., 2017) was Log2 transformed and the top 10,000 highly variable genes (hvgs) were used for PCA, which separated adenocarcinoma (SKO and 12 wk DKO) from NEPC (DKO and TKO end stage tumors) in a single dimension (Table S5). This prostate cancer rotation vector was used as a ranked list for GSEA analysis together with the PC1/3 gene-sets from Figure 3C that define YAP^off and YAP^on cancers. For this we identified 74 PC1⁺ (which include Yap & Taz), 33 PC1^-, 61 PC3⁺, and 108 PC3^- murine orthologs of the human PC1/3 gene-sets. We also generated 96 random control gene sets of comparable sizes from background (total genes in the data set after removing low-counts genes with maximum counts < 5). A similar approach was used to analyze transcriptome changes during PARCB induced reprogramming of human primary epithelia to SCN cancer (Figures 5G–5I), except that the transcriptome data (Park et al., 2018) was normalized with UQ-FPKM and Log2 transformed prior to PCA/GSEA analyses.

Cancer class genetic and pharmaceutical dependencies

Data for gene dependencies from genome-wide CRISPR/Cas9 screens were obtained from the Cancer DepMap (DepMap Public 20Q2, https://depmap.org/portal/download/) (Meyers et al., 2017). First, we determined the percentage of cell lines in each cluster (A-D) that are dependent on a particular gene; as per DepMap, genes were considered essential if they had a dependency score > 0.5. Next, we averaged the percentages for cluster-A and -B (YAP^off lines), and subtracted the average of the percentages for clusters-C and -D (YAP^on lines) to obtain a YAP^off vs. YAP^on Differential Dependency Score (DDS), and focused on genes with a DDS of ≥ 25 or ≤ −25. Genes which were essential in over 50% of lines in all four clusters were excluded. In addition, to avoid biases caused by a single cluster, we excluded genes for which the percentage of dependent lines in either A or B was less than 10 percentage points different from the percentage of dependent lines in either C or D. These criteria yielded 16 YAP^off and 31 YAP^on gene dependencies, and we performed Enrichr analysis with all YAP^off and the top 25 YAP^on hits to obtain enriched TF binding features (ENCODE/ChEA), pathways (KEGG, BioPlanet), and GO Biological Processes.

Drug sensitivity data (AUC values) were obtained from the Cancer Therapeutics Response Portal (CTRPv2) (Seashore-Ludlow et al., 2015) or Genomics of Drug Sensitivity in Cancer (GDSC1 & GDSC2) (Iorio et al., 2016). For each compound, p-values comparing the four clusters (A-D) and median AUC values for each cluster were calculated. The mean p-value comparing cluster-B to clusters-C or -D was -log10-transformed and multiplied by the mean fold difference in AUC between the groups to calculate a weighted sensitivity score. Compounds scoring >1.5 or <−1.5 in CTRPv2 were compared to values obtained for the GDSC1/2 databases to obtain high-confidence hits.

QUANTIFICATION AND STATISTICAL ANALYSIS

Unless stated otherwise in the above Method Details or figure legends, unpaired 2-tailed t-tests determined using Prism software (GraphPad) were used for sample comparisons. Statistical details of individual experiments can be found in the related figure legends, where n indicates the number of mice (in vivo experiments) or independent biological replicates (in vitro cell line experiments).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-YAP/TAZ (clone 63.7)	Santa Cruz Biotechnology	Cat# sc-101199; RRID: AB_1131430
Rabbit monoclonal anti-YAP/TAZ (clone D24E4)	Cell Signaling Technology	Cat# 8418; RRID: AB_10950494
Rabbit monoclonal anti-YAP (clone D8H1X)	Cell Signaling Technology	Cat# 14074; RRID: AB_2650491
Mouse monoclonal anti-YAP (clone 1A12)	Cell Signaling Technology	Cat# 12395; RRID: AB_2797897
Mouse monoclonal anti-GFP (clone B-2)	Santa Cruz Biotechnology	Cat# sc-9996; RRID: AB_627695
Goat polyclonal anti-GFP	Abcam	Cat# ab6673; RRID: AB_305643
Chicken polyclonal anti-GFP	Abcam	Cat# ab13970; RRID: AB_300798
Mouse monoclonal anti-Ki67 (clone B56)	BD Biosciences	Cat# 550609; RRID: AB_393778
Rat monoclonal anti-Ki67 (clone SolA15)	ThermoFisher	Cat# 14–5698-82; RRID: AB_10854564
Rabbit monoclonal anti-Human Ki67 (clone SP6)	ThermoFisher	Cat# RM-9106-S1; RRID: AB_149792
Guinea pig polyclonal anti-Synaptophysin	Synaptic Systems	Cat# 101 004; RRID: AB_1210382
Rabbit monoclonal anti-Synaptophysin (clone MRQ-40)	Origene	Cat# TA327740
Rabbit monoclonal anti-Synaptophysin (clone D8F6H)	Cell Signaling Technology	Cat# 36406; RRID: AB_2799098
Rabbit polyclonal anti-CGRP	Sigma-Aldrich	Cat# C8198; RRID: AB_259091
Rabbit polyclonal anti-Androgen Receptor (AR)	Santa Cruz Biotechnology	Cat# sc-816; RRID: AB_1563391
Rabbit polyclonal anti-CCSP (CC10)	Seven Hills Bioreagents	WRAB-3950
Rabbit polyclonal anti-pro SPC	Abcam	Cat# ab40879; RRID: AB_777473
Mouse monoclonal anti-PARP-1 (clone C2–10)	Santa Cruz Biotechnology	Cat# sc-53643; RRID: AB_785086
Mouse monoclonal anti-a TUBULIN (clone DM1A)	Santa Cruz Biotechnology	Cat# sc-32293; RRID: AB_628412
Mouse monoclonal anti-TEAD1 (clone 31/TEF-1)	BD Biosciences	Cat# 610923; RRID: AB_398238
Rabbit monoclonal anti-pan TEAD (clone D3F7L)	Cell Signaling Technology	Cat# 13295; RRID: AB_2687902
Mouse monoclonal anti-TEAD4 (clone N-G2)	Santa Cruz Biotechnology	Cat# sc-101184; RRID: AB_2203086
Mouse monoclonal anti-ASCL1 (clone D-7)	Santa Cruz Biotechnology	Cat# sc-374104; RRID: AB_10918561
Mouse monoclonal anti-NEUROD (clone A-10)	Santa Cruz Biotechnology	Cat# sc-46684; RRID: AB_671759
Mouse monoclonal anti-MYCN (clone B8.4.B)	Santa Cruz Biotechnology	Cat# sc-53993, RRID:AB_831602
Goat polyclonal anti-MYCL	Novus Biologicals	Cat# AF4050
Mouse monoclonal anti-RB1 (clone G99–549)	BD Biosciences	Cat# 554164; RRID: AB_395279
Mouse monoclonal anti-b-ACTIN (clone AC-15)	Sigma-Aldrich	Cat# A5441; RRID: AB_476744
Mouse monoclonal anti-RXRg (clone A-2)	Santa Cruz Biotechnology	Cat# sc-365252; RRID: AB_10850062
Goat polyclonal anti-MCM3 (N-19)	Santa Cruz Biotechnology	Cat# sc-9850; RRID: AB_2142269
Mouse monoclonal anti-ITGB5 (clone F-5)	Santa Cruz Biotechnology	Cat# sc-398214
Rabbit monoclonal anti-ITGB1 (clone D2E5)	Cell Signaling Technology	Cat# 9699; RRID: AB_11178800
Rabbit monoclonal anti-ITGB4 (clone D8P6C)	Cell Signaling Technology	Cat# 14803; RRID: AB_2798620
Rabbit monoclonal anti-ITGA4 (clone D2E1)	Cell Signaling Technology	Cat# 8440; RRID: AB_2797643
Rabbit polyclonal anti-ITGAV (clone D2E5)	Cell Signaling Technology	Cat# 4711; RRID: AB_2128178
Rabbit polyclonal anti-Histone H3 (trimethyl Lys4)	Abcam	Cat# ab8580; RRID: AB_306649
Rabbit polyclonal anti-Histone H3, (trimethyl Lys27)	Millipore	Cat# 07–449; RRID: AB_310624
Rabbit polyclonal anti-Histone H3 (trimethyl Lys9)	Abcam	Cat# ab8898; RRID: AB_306848
Rabbit polyclonal anti-Histone H3 (acetyl Lys27)	Abcam	Cat# ab4729; RRID: AB_2118291
Mouse monoclonal anti-RNA polymerase II CTD repeat YSPTSPS antibody [clone 4H8]	Abcam	Cat# ab5408; RRID: AB_304868
Mouse monoclonal anti-Methylcytidine	Eurogentec	Cat# BI-MECY-0100; RRID: AB_2616058
Mouse monoclonal anti-ITGAV/B5 (clone P1F6)	Santa Cruz Biotechnology	Cat# sc-81632; RRID: AB_1123634
Mouse monoclonal anti-ITGB1 (clone P5D2)	Santa Cruz Biotechnology	Cat# sc-13590, RRID:AB_627008
Bacterial and virus strains
Ad5CMVCre	University of Iowa Viral Vector Core Facility	VVC-U of Iowa-5
Biological samples
Human fetal retinas	Morgentaler Clinic	N/A
SCLC patient derived xenografts	Dr. Charles Rudin	Gardner et al., 2017; Lok et al., 2017
Paraffin-embedded SCLC TMA	Princess Margaret Cancer Centre	N/A
Paraffin-embedded NSCLC PDX TMA	Princess Margaret Cancer Centre	N/A
Paraffin-embedded NET TMA	Roswell Park Cancer Institute	N/A
Paraffin-embedded primary prostate cancer tissue samples	Roswell Park Cancer Institute	N/A
Chemicals, peptides, and recombinant proteins
Insulin, human recombinant, zinc solution	ThermoFisher	Cat# 12585014
Transferrin, Bovine (Holo form)	ThermoFisher	Cat# 11107047
Sodium selenite	Sigma-Aldrich	Cat# S5261
Hydrocortisone	Sigma-Aldrich	Cat# H0888
b-Estradiol	Sigma-Aldrich	Cat# E2758
Somatostatin	Sigma-Aldrich	Cat# S1763
Glycyl-L-histidyl-L-lysine	Sigma-Aldrich	Cat# G7387
REDExtract-N-Amp PCR ReadyMix	Sigma-Aldrich	Cat# R4775
2-hydroxyethyl agarose	Sigma-Aldrich	Cat# A4018
Crystal Violet	Sigma-Aldrich	Cat# C6158
EdU (5-ethynyl-2’-deoxyuridine)	ThermoFisher	Cat# E10187
Paraformaldehyde	Electron Microscopy Sciences	Cat# 15710
Alexa Fluor 647 Azide, Triethylammonium Salt	ThermoFisher	Cat# A10277
FxCycle Violet Stain	ThermoFisher	Cat# F10347
AlamarBlue Cell Viability Reagent	ThermoFisher	Cat# DAL1100
Alisertib	Selleck Chemicals	Cat# S1133
KPT-330 (Selinexor)	Selleck Chemicals	Cat# S7252
Navitoclax	Selleck Chemicals	Cat# S1001
Daporinad	Selleck Chemicals	Cat# S2799
SsoAdvanced Universal SYBR Green Supermix	Bio-Rad	Cat# 1725270
Protease/Phosphatase Inhibitor Cocktail	Cell Signaling Technology	Cat# 5872S
Phenylmethanesulfonyl fluoride (PMSF) solution	Sigma-Aldrich	Cat# 93482
Protein G Agarose Beads	Cell Signaling Technology	Cat# 37478S
Cyclic RGD (cRGD) peptide	Sigma-Aldrich	Cat# SCP0111
Superscript II Reverse Transcriptase	ThermoFisher	Cat# 18064014
Power SYBR Green PCR Master Mix	ThermoFisher	Cat# 4368708
Dynabeads M-280 Sheep Anti-Mouse IgG	ThermoFisher	Cat# 11202D
Dynabeads M-280 Sheep Anti-Rabbit IgG	ThermoFisher	Cat# 11204D
Proteinase K, recombinant, PCR grade	ThermoFisher	Cat# EO0491
Critical commercial assays
Papain Dissociation System	Worthington Biochemical	Cat# LK003150
Alexa Fluor 488 Tyramide SuperBoost Kit, goat anti-rabbit IgG	ThermoFisher	Cat# B40943
Alexa Fluor 594 Tyramide SuperBoost Kit, goat anti-mouse IgG	ThermoFisher	Cat# B40915
TruSeq Stranded mRNA Library Prep Kit	Illumina	Cat# 20020594; Cat# RS-122–2101
TruSeq ChIP Sample Prep Kit	Illumina	Cat# IP-202–101
Deposited data
RNA-seq from YAP-transduced SCLC and retinoblastoma cell lines	This paper	GSE144972
TEAD4 ChIP-seq in SCLC and retinoblastoma cell lines	This paper	GSE144973
Microarray from YAP/TAZ knockdown SK-Hep1 cells	Park et al., 2016	GSE54617
Microarray from YAP/TAZ knockdown UM-SCC-2 cells	Hiemer et al., 2015	GSE66949
Microarray from YAP/TAZ knockdown MDA-MB-231 cells	Zanconato et al., 2015	GSE66082
Microarray from YAP/TAZ knockdown SKMEL28 cells	Kim et al., 2016	GSE68599
Microarray from YAP/TAZ knockdown Meso-33 cells	Li et al., 2014	GSE56157
TEAD4 ChIP-seq from A549, HCT116 and HepG2 cells	ENCODE project	http://genome.ucsc.edu/ENCODE/downloads.html
TEAD4 ChIP-seq from MDA-MB-231 cells	Zanconato et al., 2015	GSE66081
Cancer Cell Line Encyclopedia microarray gene expression data	Barretina et al., 2012	https://portals.broadinstitute.org/ccle
TCGA and TARGET RNA-seq gene expression data from primary tumor samples	Cancer Genome Atlas Research Network et al., 2013; Ma et al., 2018	https://xena.ucsc.edu/
Gene expression data from primary human lung tumors	Clinical Lung Cancer Genome Project (CLCGP) and Network Genomic Medicine (NGM), 2013	http://www.uni-koeln.de/med-fak/clcgp/
RNA-seq from murine prostate cancer models	Ku et al., 2017	GSE90891
RNA-seq from normal epithelia and PARCB-transformed SCN tumors	Park et al., 2018	GSE118206
Cancer DepMap Public 20Q2	Meyers et al., 2017	https://depmap.org/portal/download/
Cancer Therapeutics Response Portal (CTRPv2) drug sensitivity data	Seashore-Ludlow et al., 2015	https://portals.broadinstitute.org/ctrp/
Genomics of Drug Sensitivity in Cancer (GDSC1 & GDSC2) drug sensitivity data	Iorio et al., 2016	https://www.cancerrxgene.org/
Experimental models: Cell lines
RB1021	Dr. Brenda Gallie (To et al., 2012)	RRID:CVCL_S624
RB3823	Dr. Brenda Gallie (To et al., 2012)	RRID:CVCL_ZF07
RB247	Dr. Brenda Gallie (To et al., 2012)	RRID:CVCL_2704
Y79	Rod Bremner lab	RRID: CVCL_1893
WERI-RB1	Rod Bremner lab	RRID: CVCL_1792
NCI-H526	ATCC	Cat# CRL-5811; RRID: CVCL_1569
SHP77	ATCC	Cat# CRL-2195; RRID: CVCL_1693
NCI-H510A	ATCC	Cat# HTB-184; RRID: CVCL_1565
NCI-H82	ATCC	Cat# HTB-175; RRID: CVCL_1591
NCI-H2171	ATCC	Cat# CRL-5929; RRID: CVCL_1536
NCI-H209	Susan P.C. Cole lab	RRID: CVCL_1525
NCI-H69	Susan P.C. Cole lab	RRID: CVCL_1579
NCI-N417	Susan P.C. Cole lab	RRID: CVCL_1602
NCI-H660	ATCC	Cat# CRL-5813; RRID: CVCL_1576
DU4475	Dr. Eldad Zacksenhaus	RRID: CVCL_1183
QGP-1	Dr. Todd McMullen	RRID: CVCL_3143
BON-1	Dr. Todd McMullen	RRID:CVCL_3985
SH-SY5Y	Dr. Meredith Irwin	RRID: CVCL_0019
SK-N-SH	Dr. Meredith Irwin	RRID: CVCL_0531
NCI-H661	ATCC	Cat# HTB-183; RRID: CVCL_1577
NCI-H1650	ATCC	Cat# CRL-5883; RRID: CVCL_1483
A549	Rod Bremner lab	RRID: CVCL_0023
SW-1573	Susan P.C. Cole lab	RRID: CVCL_1720
MCF7	Rod Bremner lab	RRID: CVCL_0031
MDA-MB-231	Rod Bremner lab	RRID: CVCL_0062
HCC1419	Dr. Benjamin Neel	RRID: CVCL_1251
DU145	Dr. Jan Jongstra	RRID: CVCL_0105
BPH-1	Rod Bremner lab	RRID: CVCL_1091
PANC-1	Rod Bremner lab	RRID: CVCL_0480
Hs 766T	Rod Bremner lab	RRID: CVCL_0334
T24	Jeff L. Wrana lab	RRID: CVCL_0554
Karpas 299	Dr. Rob Ingham	RRID: CVCL_1324
SUP-M2	Dr. Rob Ingham	RRID: CVCL_2209
SU-DHL-1	Dr. Rob Ingham	RRID: CVCL_0538
SR-786	Dr. Rob Ingham	RRID: CVCL_1711
UCONN	Dr. Rob Ingham	RRID: CVCL_A693
HDMY-Z	Dr. Rob Ingham	RRID: CVCL_1273
KM-H2	Dr. Rob Ingham	RRID: CVCL_1330
L428	Dr. Rob Ingham	RRID: CVCL_1361
L540	Dr. Rob Ingham	RRID: CVCL_1362
HDLM-2	Dr. Rob Ingham	RRID: CVCL_0009
OCI-AML2	Dr. Mark Minden	RRID: CVCL_1619
OCI-AML3	Dr. Mark Minden	RRID: CVCL_1844
THP-1	Dr. Aaron Schimmer	RRID: CVCL_0006
HL-60	Dr. Aaron Schimmer	RRID: CVCL_0002
K562	Dr. Aaron Schimmer	RRID: CVCL_0004
Lenti-X 293T	Takara/Clontech	Cat# 632180
Experimental models: Organisms/strains
Mouse: αCre: Tg(Pax6-cre,GFP)2Pgr	Dr. Peter Gruss	MGI:3052661
Mouse: Rb1f/f: Rb1tm2Brn	Dr. Anton Berns	MGI:1931018
Mouse: Rbl1−/−: Rbl1tm1Mru	Dr. Michael Rudnicki	MGI:1857798
Mouse: Yapf/f: Yap1tm1Hmc	Reginensi et al., 2013	MGI:5486298
Mouse: Tazf/f: Wwtr1tm1Hmc	Reginensi et al., 2013	MGI:5486299
Mouse: p53f/f: Trp53tm1Brn	NCI Mouse Repository	Strain# 01XC2; MGI:1931011
Mouse: CgrpCreER: Calcatm1.1(cre/ERT2)Ptch	Dr. Pao-Tien Chuang (Song et al., 2012)	MGI:5460801
Mouse: TRE-Yap: TRE-HA-YapTg	Gregorieff et al., 2015	N/A
Mouse: ROSA-rtTA: Gt(ROSA)26Sortm1(rtTA,EGFP)Nagy	Dr. Andras Nagy	MGI:3583817
Mouse: Ccsp-rtTA: Tg(Scgb1a1-rtTA)1Jaw	Tichelaar et al., 2000	MGI:2445707
Mouse: tetO-EGFRL858R: Tg(tetO-EGFR*L858R)56Hev	Politi et al., 2006	MGI:3690078
Mouse: Ptenf/f: Ptentm1Hwu	Lesche et al., 2002	MGI:2156086
Mouse: SPC-rtTA: SFTPC-rtTA	Perl et al., 2002a	N/A
Mouse: tetO-Cre: Tg(tetO-cre)1Jaw	Perl et al., 2002b	MGI:2679524
Mouse: PB-Cre4: Tg(Pbsn-cre)4Prb	Wu et al., 2001	MGI:2385927
Oligonucleotides
See Table S8 for oligonucleotide sequences used
for CRISPR sgRNA vectors, the CRIPSR library,
RT-qPCR, ChIP-qPCR and mouse genotyping.
Recombinant DNA
psPax2	Didier Trono	Addgene plasmid #12260; RRID: Addgene_12260
pMD2.G	Didier Trono	Addgene plasmid #12259; RRID: Addgene_12259
LentiCRISPR v2	Sanjana et al., 2014	Addgene plasmid #52961; RRID: Addgene_52961
pLKO.1-puro	Stewart et al., 2003	Addgene plasmid #8453; RRID: Addgene_8453
Software and algorithms
Kaluza, Flow Cytometry analysis software	Beckman Coulter	https://www.mybeckman.ca/flow-cytometry/software/kaluza
Prism	GraphPad	https://www.graphpad.com/scientific-software/prism/
RSEM	Li and Dewey, 2011	https://deweylab.github.io/RSEM/
DESeq2	Love et al., 2014	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
RRHO	Plaisier et al., 2010	https://bioconductor.org/packages/release/bioc/html/RRHO.html
RRHO2	Cahill et al., 2018	https://github.com/RRHO2/RRHO2
DeepTools	Ramírez et al., 2016	https://github.com/deeptools/deeptools
Homer package	Heinz et al., 2010	http://homer.ucsd.edu/homer/
NMF	Gaujoux and Seoighe, 2010	https://cran.r-project.org/web/packages/NMF/index.html
fastQC	N/A	https://www.bioinformatics.babraham.ac.uk/projects/fastqc
Bowtie2	Langmead and Salzberg, 2012	http://bowtie-bio.sourceforge.net/bowtie2
SAMtools	Li et al., 2009	http://samtools.sourceforge.net
MACS2	Zhang et al., 2008	https://github.com/macs3-project/MACS
Integrative Genomics Viewer (IGV)	Robinson et al., 2011	https://software.broadinstitute.org/software/igv/
UCSC tool bedGraphToBigWig	Kent WJ., 2010	https://anaconda.org/bioconda/ucsc-bedgraphtobigwig
ReMap	Chèneby et al., 2018	http://remap.univ-amu.fr/
GSEA	Subramanian, et al., 2005	https://www.gsea-msigdb.org/gsea/index.jsp
R package ‘‘ggplot2’’	Wickham, 2016	https://cran.r-project.org/web/packages/ggplot2
R package ‘‘scales’’	N/A	https://scales.r-lib.org

Highlights.

• YAP/TAZ silencing in multiple cancers reflects their tumor suppressor activity

• Cancers can be parsed into binary YAP^on and YAP^off classes

• Binary transcriptomes drive distinct adhesive, metabolic, genetic, and drug profiles

• Alternate YAP enhancers explain anti-cancer integrin-aVb5 program in YAP^off cancers