NF2 Loss Promotes Oncogenic RAS-Induced Thyroid Cancers via YAP-Dependent Transactivation of RAS Proteins and Sensitizes Them to MEK Inhibition

Neurofibromatosis type 2 is an autosomal-dominant syndrome caused by germline heterozygous mutations of NF2, which encodes merlin. It is characterized by tumors of the nervous system, such as schwannomas, meningiomas, and ependymomas. The associated neoplasias often harbor LOH of the chromosome 22q region encompassing the NF2 gene (1). Other malignancies associated with NF2 defects include mesotheliomas, melanomas, and clear-cell renal cancers.

Merlin inhibits cell growth in response to cell contact. It interacts with multiple partners to modulate distinct pathways, including Hippo (2), receptor tyrosine kinases (RTK; refs. 3, 4), RAC/CDC42/p21-activated kinases (PAK; refs. 5–7), and mTOR (8, 9). Merlin also has a nuclear function by inhibiting the CRL4^DCAF1 E3 ubiquitin ligase (10). Hippo is an evolutionarily conserved kinase cascade that suppresses tissue overgrowth through phosphorylation of YAP, leading to its sequestration in the cytoplasm and disrupting its ability to promote transcriptional enhancer activation domain (TEAD)–dependent transcription of genes involved in proliferation and survival (11–14). Despite the critical role of the Hippo pathway in growth control, NF2 is the only commonly mutated cancer gene in this pathway (15). The lineage-specific properties and the genetic repertoire intrinsic to different cancer types may predispose NF2-deficient cells to be preferentially addicted to distinct pathways. For instance, merlin loss activates effectors of mTOR in meningiomas, schwannomas, and mesotheliomas and confers sensitivity to rapamycin (8, 9, 16). In glial cells, merlin loss induces cell growth in an ERBB2-dependent manner (17). In contrast, hepatocellular carcinomas in mice with hepatocyte-targeted deletion of Nf2 have been variously reported to be dependent on Hippo (18) or on EGFR signaling (19).

Papillary thyroid cancers (PTC) are indolent tumors associated with mutually exclusive mutations of BRAF and RAS and of fusion RTK oncogenes, such as RET, NTRK1, and NTRK3 (20). The driver frequency is different in poorly differentiated thyroid cancers (PDTC) and anaplastic thyroid cancers (ATC), in that the latter are enriched for RAS mutations (21–23). Here, we show that NF2 is a novel thyroid tumor suppressor, preferentially associated with RAS mutations. Although loss of Nf2 or Ras activation is insufficient to independently induce thyroid cancers in mice, their combination is highly tumorigenic. NF2 loss cooperates with mutant RAS to increase signaling via MAPK, acting in part through YAP-induced transcriptional activation of oncogenic and wild-type (WT) RAS, providing a novel mechanism of promotion of RAS-induced tumorigenesis. This has therapeutic implications, as these and other inputs resulting from merlin deficiency converge to confer preferential sensitivity to selective MEK inhibitors in vitro and in mouse genetic models of the disease. In addition, pharmacologic disruption of the YAP–TEAD transcriptional complex decreases expression of oncogenic and WT RAS and inhibits tumor cell growth.

The Cancer Genome Atlas recently completed an analysis of approximately 400 PTCs, which showed a high frequency of ch22q loss in RAS-mutant PTC (45% had 22q LOH; ref. 24; Supplementary Table S1). The association was particularly striking for HRAS: 10/14 (71%; P < 5 × E–6; OR >10). Ch22q is the only region of copy-number variation in the genomes of most of these cancers, which are otherwise diploid, suggesting that one or more tumor suppressors on 22q play an important role in tumorigenesis. We found that PDTCs also had a high frequency of 22q LOH (14/63; 22%) as determined by SNP-comparative genomic hybridization (CGH) and/or copy-number analysis of sequence reads derived from an exon capture next-generation sequencing panel of 341 cancer genes, 6 of which mapped to Ch22q (Supplementary Table S1; Supplementary Fig. S1A–S1D). Although most tumors had LOH of all 6 genes, spanning the majority of the chromosome arm, CHEK2, NF2, and EP300 were consistently lost. As was the case in PTC, Ch22q LOH in PDTCs was seen preferentially in association with RAS (8/16; 50%) as compared with BRAF-mutant tumors (0/26). Five of the 20 ATCs were RAS mutant, one of which had 22q LOH (Supplementary Table S1).

Of the cancer genes mapping to Ch22q, we focused in greater detail on NF2 because 3 of 40 thyroid cancer cell lines had homozygous nonsense mutations of this gene (Cal62: c.643G>T, pE215*; 8505c: c.385G>T, p.E129* and TCO-1: c.303T>A, p.Y101*). In addition, the KHM-5M ATC cell line had a homozygous deletion of exon 4 of NF2 that disrupts the central FERM domain of merlin, previously reported in neurofibromatosis patients (ref. 25; Supplementary Fig. S2). Consistent with the low frequency of homozygous NF2 inactivation in cell lines, there are limited data supporting biallelic NF2 inactivation in primary thyroid cancers. Indeed, NF2 mutations in tumor samples were rare other than for one ATC with a somatic G>A substitution at the −1 position of the intron 14/exon 15 boundary (splice donor site), which removes exon 15 and impairs the biologic effects of merlin (26). As ATCs are heavily infiltrated with macrophages, which decrease sensitivity of genomic profiling, we expanded the analysis of NF2 copy number by performing FISH on ATC tissue microarrays, which showed that 10 of 16 had NF2 LOH, one of which had a homozygous deletion (Supplementary Fig. S3A and S3B).

Several thyroid cancer cell lines that were WT or hemizygous for NF2 had very low or absent merlin mRNA and/or protein levels (Supplementary Fig. S4A and S4B). Despite lower NF2 mRNA, we did not detect aberrant methylation patterns of CpG islands in the promoter of NF2 in cell lines or tumors (not shown). Interestingly, the Hth74 ATC cell line had a markedly decreased NF2 mRNA half-life (Supplementary Fig. S4C). Hence, as reported in other lineages, loss of merlin in thyroid cancers occurs through diverse mechanisms (Supplementary Fig. S4D): LOH or intragenic deletions, somatic base substitutions as well as posttranscriptional events (27–30).

In view of the strong association between NF2 and RAS in human thyroid cancers, we next explored the potential biologic significance of this interaction in mouse models. Endogenous expression of Hras^G12V in thyroid cells was achieved by activating the latent FR-Hras^G12V allele by crosses with TPO-Cre mice (ref. 31; Fig. 1A). None of the TPO-Cre/FR-Hras^G12V heterozygous or homozygous animals developed thyroid cancer. Although approximately 65% of mice with thyrocyte-specific homozygous deletion of Nf2 (TPO-Cre/Nf2^flox2) exhibited mild nodular hyperplasia after 18 months, none developed cancer (Supplementary Table S2). In contrast, TPO-Cre/FR-Hras^G12V/Nf2^flox2 mice developed large thyroid cancers with high penetrance (Fig. 1B). Most of these were PDTCs and were associated with a marked increase in phosphorylated ERK (pERK; Fig. 1C).

We next investigated mechanisms accounting for the NF2–RAS interactions seen in vivo in RAS-mutant human thyroid cancer cell lines. Re-expression of WT merlin in RAS-mutant/NF2-null Cal62 and Hth83 cells decreased growth and colony formation in soft agar, whereas the loss-of-function mutant NF2^L64P (32) was without effect (Fig. 2A). NF2, but not NF2^L64P, also decreased pMEK and pERK in these cells (Fig. 2B). Conversely, knockdown of merlin in HRAS-mutant/NF2-WT C643 cells enhanced growth and MAPK signaling (Fig. 2C).

Merlin loss enables sequestration of RICH1, a GTPase-activating protein, by a tight junction protein complex, which de-represses RAC1, leading to activation of RAF–MEK signaling (33). We found that this pathway was also operative in the setting of constitutively activated RAS (Supplementary Fig. S5). Expression of merlin suppressed RAC1-GTP as well as phosphorylation of its effector PAK at S141 and T423. This in turn decreased CRAF and MEK phosphorylation at S298, a PAK substrate. In a reciprocal experiment, merlin silencing was associated with increased pPAK-T423 (Supplementary Fig. S5A and S5B).

Expression of a dominant-negative PAK construct (PAK1 83-149) modestly suppressed growth of KRAS^G12R/NF2-null Cal62 cells. As predicted, this was associated with decreased phosphorylation of the PAK substrate S298-MEK. However, pS217/p221-MEK and pERK were not significantly reduced, suggesting that, in the context of mutant RAS, merlin loss augments MAPK signaling through alternative mechanisms (Supplementary Fig. S5C). Consistent with this, the effects of merlin on PAK signaling were apparent 24 hours after doxycycline (dox) induction (Supplementary Fig. S5B), whereas pS217/p221-MEK and pERK were seen to decrease beginning at 48 hours (Fig. 2B). The ATP-competitive PAK kinase inhibitor FRAX597 also preferentially suppressed PAK phosphorylation in merlin-silenced C643 cells, with no apparent effects on pERK (Supplementary Fig. S5D). It also inhibited cell growth after merlin knockdown, although it was toxic at higher concentrations, possibly through off-target effects.

Upon cell–cell contact, merlin reportedly prevents internalization and signaling of the EGFR by sequestering it into an insoluble membrane compartment (3, 4). The impairment of EGFR signaling by merlin also manifests in RAS-mutant thyroid cancer cell lines. Although merlin loss augments EGFR signaling, this does not appear to contribute significantly to growth regulation (Supplementary Fig. S6).

To explore the possible contribution of other inputs upstream of RAS to MAPK activation, we measured RAS-GTP levels prior to and 72 hours after merlin expression. Unexpectedly, merlin inhibited both WT and mutant RAS-GTP, which was associated with decreased protein and mRNA levels of all RAS isoforms (Fig. 3A and B). Conversely, knockdown of NF2 increased mutant and WT RAS mRNA and protein levels in NF2-WT cells (Fig. 3C). Strikingly, this was also the case in the mouse models of the disease. Thus, protein levels of the three Ras genes were markedly increased in thyroid tissues with conditional deletion of Nf2. This was associated with increased pERK and pMEK in TPO-Cre/Hras^G12V/Nf2^flox2 mice, presumably because of overexpression of the oncogenic Hras allele (Fig. 3D). Copy-number abnormalities of Hras were found in a subset of these tumors (Supplementary Fig. S7). However, this does not account for the consistent increase in HRAS in all tumors we examined, or the higher expression of the WT RAS proteins.

Consistent with the known interaction of merlin with the Hippo pathway, expression of WT merlin in NF2-null RAS-mutant cell lines resulted in YAP phosphorylation and retention in the cytoplasm, with reciprocal changes seen after knockdown of merlin in NF2-WT cells (Fig. 4A and B). YAP is a required component of a transcriptional regulatory complex that includes its close homologs TAZ and TEAD. Silencing of YAP in Cal62 (KRAS^G12R, NF2-null) cells decreased mRNA levels of all RAS isoforms, as well as of the canonical YAP-transcriptional target connective tissue growth factor (CTGF; Fig. 4C). This was associated with a profound decrease of all three RAS proteins (Fig. 4D). YAP knockdown also prevented the induction of RAS protein and of MAPK signaling by merlin silencing in C643 (HRAS^G13R, NF2-WT) cells (Fig. 4E) and inhibited growth of NF2-null cell lines (Fig. 4F). Consistent with this, expression of constitutively active YAP1^S127A in C643 cells (NF2-WT) induced RAS gene expression (Fig. 4G) and promoted cell growth (Fig. 4H). YAP1^S127A also rescued the growth inhibition (Fig. 4I) and the suppression of RAS mRNA levels by merlin in Cal62 cells (Fig. 4J). In silico analysis identified consensus TEAD-binding sites in the promoters of the three RAS genes (Fig. 5A). The functional relevance of this prediction is supported by chromatin immunoprecipitation (ChIP)-PCR with antibodies to either YAP or TEAD1 (TEAD1 is the most abundant TEAD isoform in thyroid cancer cells; see Supplementary Fig. S8A). This showed that merlin expression markedly diminished occupancy by YAP and TEAD1 of TEAD consensus motifs in the promoters of the three RAS genes (Fig. 5B and Supplementary Fig. S8B).

Verteporfin is an FDA-approved drug used as a photosensitizer for photodynamic ablation of abnormal blood vessels in patients with macular degeneration (34). Verteporfin was identified in a drug screen for compounds that could disrupt YAP–TEAD-driven transcription, and shown to inhibit YAP-dependent growth (14). We found that verteporfin inhibited growth of NF2-null Cal62 (Fig. 5C) and Hth83 cells (Supplementary Fig. S8C), an effect that was dampened by merlin expression (Supplementary Fig. S8D). Verteporfin decreased YAP and TEAD, lowered RAS protein abundance, and inhibited MAPK signaling. Reciprocal findings were seen in C643 cells after merlin silencing (Supplementary Fig. S8E). Verteporfin treatment inhibited TEAD occupancy of the oncogenic KRAS gene promoter in Cal62 cells (Fig. 5D). Hence, genetic and pharmacologic disruption of YAP–TEAD leads to decreased oncogenic RAS gene expression, MAPK signaling, and growth.

We next explored whether the increased MAPK signaling seen in RAS-mutant/NF2-deficient thyroid cancer cell lines rendered them more dependent on this pathway for growth. This was explored in isogenic lines derived from C643 cells (HRAS^G13R, NF2-WT) modified to stably express shNF2 (Fig. 6A). Although expression of oncogenic HRAS, as well as baseline pMEK and pERK, was markedly higher in the merlin-depleted shNF2.M4 line, the inhibition of the pathway by the MEK inhibitor was comparable. Despite this, growth suppression by AZD6244 was greater in cells depleted of merlin. We also tested a set of 7 RAS-mutant thyroid cancer cell lines for growth response to MEK inhibition. There was a trend for greater sensitivity to AZD6244 in RAS-mutant thyroid cancer cell lines that were merlin-null or low as compared with merlin WT cell lines (Supplementary Fig. S9A). To explore the contribution of distinct inputs into MEK (i.e., RAC1–PAK vs. canonical RAS–RAF), we explored growth in response to AZD6244, FRAX597, or their combination in C643 cells with or without merlin silencing. Consistent with the data shown in Fig. 6, C643-shNF2.M2 cells are exquisitely sensitive to the MEK inhibitor. Although knockdown of merlin also sensitizes cells to growth inhibition by the PAK inhibitor, the effects were comparatively modest. Combined inhibition showed only modest additive effects (Supplementary Fig. S9B).

We also examined the effects of AZD6244 in three models of HRAS^G12V-driven mouse poorly differentiated thyroid cancers, arising in the context of Nf2, Pten, or Trp53 homozygous loss. All of these gave rise to PDTCs, defined based on the presence of necrosis and a high mitotic rate on histology. These tumors differed, however, in tumor doubling time (Pten = Nf2>>Trp53; Supplementary Fig. S10A) and overall survival (Supplementary Fig. S10B). Hras/Nf2 and Hras/Trp53 tumors stained intensely for pERK, whereas Hras/Pten tumors had low pERK and increased pAKT staining (Fig. 6B). Treatment of mice with Hras^G12V/Nf2-null thyroid cancers with AZD6244 resulted in a greater and more consistent reduction of tumor size than in cancers arising in Hras^G12V/Pten-null or Hras^G12V/Trp53-null mice (Fig. 6C). IHC of treated tumors showed marked reduction of pERK staining in all three tumor types, whereas pAKT IHC remained very high in Hras/Pten tumors, possibly accounting for their lack of response to therapy (Fig. 6D).

The striking tissue overgrowth phenotypes induced by genetically disabling the core components of the Hippo pathway in Drosophila melanogaster prompted exploration of its role in mammalian cell proliferation and cancer development (13). Hippo inactivation converges to regulate gene expression through the transcriptional coactivator YAP and/or its close homolog TAZ. This raises the question of how the YAP/TEAD transcriptional targets may affect the biology of cancers driven by oncogenes, such as RAS, which nominally operate through a distinct canonical signaling network. Our finding that RAS genes are themselves transcriptionally regulated by YAP may have profound implications for RAS-driven tumors.

The importance of Ras-mutant allele gene dosage for transformation is well established. A requirement for intensification of mutant RAS signaling through copy-number imbalances is supported by studies in Hras-mutant fibroblasts, in which transformation is strongly associated with amplification of the mutant allele (35). Increased mutant allele copy number is an obligate early event in Hras^G12V-induced papilloma development (31, 36, 37). Similarly, the aggressiveness of myeloproliferative neoplasms expressing NRAS^G12D is augmented when the mutant allele is homozygous (38). Moreover, the expression of oncogenic RAS in human cancer cell lines is consistently higher than that of the other WT RAS proteins (38). Although HRAS gene amplification has been reported in thyroid cancer (39), the overall frequency, at least in PTC, appears to be low (24). Despite the critical significance of mutant RAS protein abundance on its signaling and transforming properties, there is so far limited information on how other oncogenic inputs may alter its expression. The 3′ untranslated regions of human RAS genes contain multiple complementary sites to the let-7 family of miRNAs, which regulate RAS expression by translational inhibition and/or by decreasing mRNA stability (40, 41). Loss of let-7 miRNAs increases RAS protein levels, and, conversely, let-7 overexpression decreases them and attenuates oncogenic RAS-induced tumorigenesis (41, 42).

Regulation of mutant RAS gene transcription as a functional consequence of disrupting a parallel oncogenic pathway has, to our knowledge, not been previously implicated as a mechanism of tumor promotion. The YAP–TEAD complex modulates expression of a diverse array of transcriptional targets (12). However, comparatively little is known about the specific YAP-regulated genes responsible for mediating the effects of Hippo inactivation on growth, invasiveness, metastases, or senescence (43–46). CTGF is a prototypical target of YAP–TEAD and has been functionally implicated in hepatocellular growth and tumorigenesis (47). YAP1 was recently reported to rescue viability of pancreatic cancer cells conditionally expressing mutant KRAS after oncoprotein withdrawal in mice. The authors implicated genes co-regulated by YAP1 and E2F in this process, many of which were cell-cycle regulators (48). In a screen for genes promoting survival of KRAS-mutant cells following oncogene silencing in vitro, YAP1 was the most prominent hit among those that could also activate either MAPK or PI3K signaling. YAP1 was found to regulate a set of immediate response genes that were also activated by KRAS, ultimately converging on the transcription factor FOS, which also rescued cells after KRAS silencing (49). YAP is also required for KRAS^G12D-induced pancreatic transformation in mice, by inducing expression of genes encoding secreted factors that promote cancer cell growth and the establishment of a tumorigenic stromal microenvironment (50). We do not yet know whether YAP1-mediated transcriptional activation of RAS mRNAs is ubiquitous or context dependent, or if this may have contributed to some of the observations in the lineages and models described above. Our analysis of previously published genome-wide ChIP sequencing data in mouse embryonic stem cells points to clear YAP-binding peaks in all three RAS genes (not shown; ref. 51), suggesting that these are likely to be regulated via Hippo inactivation in other cell types as well. However, tissue overgrowth induced by Hippo silencing likely requires the concerted regulation of genes involved at multiple key steps of the process. As RAS proteins are critical signaling nodes, modulation of their expression would affect the signal amplitude emanating from a wide array of upstream inputs.

The studies described above implicate YAP as an important effector in RAS-mutant cancer. However, the mechanisms causing YAP activation in these contexts are not known. In the case of RAS-mutant thyroid cancer, merlin loss of function is a key event (Fig. 7A and B). Although germline or somatic NF2 mutations are primarily implicated in the pathogenesis of nervous system tumors, mesotheliomas, melanomas, and renal cell carcinomas, mice with heterozygous Nf2 mutation develop a broader range of tumors, which lose the WT allele (52). Loss of function of merlin also occurs in the absence of homozygous deletion or mutation, through aberrant splicing, mRNA loss, calpain-mediated proteolysis, proteasomal degradation, or phosphorylation (27–30, 53). Consistent with this, several thyroid cancer cell lines that were merlin-null or merlin-low did not have homozygous genomic NF2 mutations. This is particularly relevant in advanced thyroid cancer, as intragenic NF2 mutations are rare in tumors with 22q LOH. The extent to which NF2 haploinsufficiency is associated with impaired expression of the WT allele in human tumors is currently uncertain. Hence, sensitive protein-based assays may need to be developed to screen cancers for merlin loss of function, particularly if this proves to be an actionable event.

There are potential therapeutic opportunities resulting from Hippo pathway inactivation in the context of RAS-induced tumorigenesis. We found that these tumors hyperactivate MAPK signaling and consequently generate increased dependency on this pathway for viability. Thus, although merlin loss leads to a global increase in YAP–TEAD transcriptional output, the genes that augment susceptibility to MEK inhibition are paramount in this setting. Moreover, as illustrated by our data with verteporfin, targeting of YAP may prove to be viable strategy for thyroid cancer, and perhaps other cancers, driven by oncogenic RAS.

A total of 83 advanced thyroid tumors, including 20 ATCs and 63 PDTCs, were assessed. Both frozen (n = 37) and formalin-fixed paraffin-embedded (FFPE; n = 46) specimens were included. Forty-three thyroid cancer–derived cell lines, previously authenticated using short tandem repeat and SNP array analysis (54), were also evaluated. Genomic DNA was extracted from all specimens using the DNeasy Tissue Kit (Qiagen). Patient samples were studied under a protocol approved by the Memorial Sloan Kettering Institutional Review Board.

Two hundred fifty nanograms of genomic DNA derived from all 126 thyroid samples and matched normal tissues (when available) were subjected to deep-coverage targeted sequencing with MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets), an exon-capture next-generation sequencing platform covering the entire coding sequence and intron–exon boundaries of 341 cancer genes (55). Sequence data were analyzed to identify three classes of somatic alterations: single-nucleotide variants, small insertions/deletions (indels), and copy-number alterations (CNA). We focused in greater detail on the six genes mapping on chromosome 22q included in the platform: CRKL, MAPK1, SMARCB1, CHEK2, NF2, and EP300.

To confirm the ability of IMPACT to accurately call CNAs, we subjected our subset of 37 frozen tumors to array comparative genomic hybridization (aCGH). Briefly, 3 μg of DNA was digested and labeled by random priming using Cy3 or Cy5-dUTP–labeled primers (Invitrogen). Labeled tumor DNA was cohybridized to Agilent aCGH microarrays with a pool of reference normal DNA for 40 hours at 60°C. After washing, the slides were scanned and images were quantified using Feature Extraction 9.1 (Agilent Technologies). Raw copy-number estimates were normalized and segmented with Circular Binary Segmentation (56). Data were also analyzed using the RAE algorithm (57). IMPACT-sequencing reads and segmented copy-number data from both platforms were visualized in the Integrative Genomics Viewer (58), and all genome coordinates were standardized to NCBI build 37 (hg19) of the reference human genome.

FISH analysis for NF2 status was performed by hybridizing a tissue microarray containing 16 ATC specimens with bacterial artificial chromosome probes for NF2 (RP11-551L12, 22q12.2, red) and BCR (22q11.2, green). Deletion of NF2 was defined as tumors with more than 25% of cells with single or no NF2 hybridization signals (a minimum of 200 cells were scored for each tumor). This threshold was selected because of the high admixture of stromal cells, particularly tumor-associated macrophages, in these cancers.

TPO-Cre mice express Cre recombinase under the control of the thyroid peroxidase gene promoter, which is active only in thyroid follicular cells beginning at E14.5 (59). FR-Hras^G12V mice conditionally express a latent Hras^G12V allele under the regulatory control of its endogenous gene promoter (31). To generate triple transgenic mice, TPO-Cre/FR-Hras mice were bred with Nf2^flox2 (60), Trp53^flox2 (61), or Pten^flox2 (62). Both male and female mice were used. Mice were in the following backgrounds: TPO-Cre/FR-Hras^G12V/Nf2^flox2: ∼7% 129sv, 56% C57BL/6, 12% Swiss black, 25% FVB/n; TPO-Cre/FR-Hras^G12V/Trp53^flox2: ∼56% 129sv, 7% C57BL/6, 12% Swiss black and 25% FVB/n. TPO-Cre/FR-Hras^G12V/Pten^flox2 mice were backcrossed 5 generations into 129sv.

TPO-Cre/FR-Hras^G12V/Nf2^flox2 and TPO-Cre/FR-Hras^G12V/Pten^flox2 mice were treated 5 days a week for 4 weeks with AZD6244 (25 mg/kg/twice a day) dissolved in 0.1% Tween 80 + 0.5% methylcellulose or vehicle. TPO-Cre/FR-Hras^G12V/Trp53^flox2 mice were treated for 7 days a week for 2 weeks. Animal care and all procedures were approved by the MSKCC Institutional Animal Care and Use Committee.

Mice were anesthetized by inhalation of 2% to 3% isoflurane with 1% O₂, neck hair removed with defoliating agent, and placed on the heated stage. An aqueous ultrasonic gel was applied to the skin overlying the thyroid glands. Thyroid tumors were imaged with the VisualSonics Vevo 770 In Vivo High-Resolution Micro-Imaging System (VisualSonics Inc.). Using the Vevo 770 scan module, the entire thyroid bed was imaged with captures every 250 microns. Using the instrument's software, the volume was calculated by manually tracing the margin of the tumor every 250 microns. The instrument is calibrated to allow measurements to be determined accurately. Animals were included in the study if they had a tumor present as demonstrated by ultrasound and at a size that could be accurately measured by the ultrasound probe.

Thyroid tissues were immediately placed in 4% paraformaldehyde and incubated overnight at 4°C. The next day, tissue was washed twice with PBS for 30 minutes followed by a single 30-minute 50% ethanol wash. The fixed tissue was then placed in 70% ethanol, paraffin-embedded, and sectioned into 4-μm paraffin sections. Hematoxylin and eosin (H&E)–stained slides were evaluated by a board-certified pathologist (R. Ghossein). Mouse thyroid sections were deparaffinized and immunostained with an antibody to pERK (#9101; Cell Signaling) or pPAK^S473 (#4051S; Cell Signaling) at the Memorial Sloan Kettering Cancer Center Molecular Cytology Core Facility.

Cancer cell lines were maintained at 37°C and 5% CO₂ in humidified atmosphere and grown in RPMI-1640 (C643, Hth83, and Cal62), DMEM (Hth7 and ACT1), or DMEM:RPMI (ASH3 and KMH2) supplemented with 10% of FBS, 2 mmoL/L glutamine, 50 U/mL penicillin (GIBCO), and 50 μg/mL streptomycin. Cell lines C643, Hth7, and Hth83 were obtained from Dr. Nils-Erik Heldin in December 2006, September 2007, and December 2006, respectively. Cal62 cells were obtained from Dr. Jeanine Gioanni in December 2006. The ACT1 line was obtained from Dr. Onoda Osaka in April 2006. ASH3 and KMH2 were obtained from JCRB in April 2010. All thyroid cancer cell lines used in this study were authenticated using short tandem repeat and SNP array analysis between 2006 and 2010 (54). For cell growth assays, cells were plated in triplicate into 6-well plates at 50,000 cells per well, and incubated for 24 hours. The cells were treated with vehicle or concentrations of the indicated drug in media with 1% of FBS. For transient transfections, 20,000 cells/well were plated into 24-well plates in media containing 10% FBS without antibiotics. Twenty-four hours after the transfection was performed as indicated in the “Gene Expression and Silencing” section, cells were collected by trypsinization and counted in a Vi-Cell series Cell Viability Analyzer (Beckman Coulter) at times indicated. IC₅₀ values were calculated by nonlinear regression using Prism v5.04 (GraphPad Software).

Cells were seeded in triplicate at 50,000 cells per 35 mm dish. Dishes were first coated with a bottom layer of 0.4% agar in RPMI. Cells were resuspended in a top layer of 0.2% agar in RPMI with 10% FBS, and then fed every other day by adding drops of media onto the top layer. After 9 days, the colonies were stained with crystal violet and counted in a GelCount colony counter (Oxford OPTRONIX). Minimum diameter of the colonies was 50 μm.

PCR-amplified full-length cDNAs of human NF2-WT and NF2^L64P were cloned into the pLVX-Tight-puro vector (Clontech) using the following primers containing BamHI and EcoRI restriction sites:5′-GAGAGGATCCTCACCATGGCCGGAGCCATC-3′; 5′-GAGAGAATTCTTCGAACCGCGGGCCCTCTA-3′. Cal62 and Hth83 thyroidcancer cells were coinfected with pLVX-Tight-puro-NF2 or NF2-L64P and with pLVX-tet-on Advanced vector (Clontech). C643 were infected with pLKO.1 and 5 different hairpins for NF2 (M1#TRCN0000039973; M2#TRCN0000039975; M3#TRCN00000 39977; M4#TRCN0000010397; M5#TRCN0000018 338 from Open Biosystems). For infection, cells were incubated with infectious particles in the presence of 10 ng/mL hexadimethrine bromide (Sigma) overnight. After recovery in complete medium for 24 hours, cells were placed under selection in 1 μg/mL puromycin or 300 μg/μL G418 for Cal62 and Hth83 or 1 μg/mL puromycin for C643. The production of viral particles was performed with Mission Lentiviral Packing Mix from SIGMA. Transient transfections were performed using Lipofectamine 2000 (Invitrogen) or Amaxa Nucleofector System (Lonza). Dominant negative of PAK (PAK1 83-149) and YAP1^S127A were purchased from Addgene (#17790 and #12214, respectively; refs. 63, 64). Short interfering RNAs for YAP were from ORIGENE (SR307060). The media (10% FBS without antibiotics) were changed 4 to 6 hours after transfection. Efficiency of knockdown or overexpression was verified by immunoblotting.

Total mRNA from snap-frozen thyroid tissue was extracted with the PrepEase Kit (USB Corporation); total mRNA from cells was isolated by TRIzol (Invitrogen). Equal amounts of isolated RNA (2 μg) were subjected to DNase I Amplification Grade (Invitrogen) and subsequently reverse transcribed into cDNA using the SuperScript III Reverse Transcriptase (Invitrogen) according to the manufacturer's instructions. qPCR was then performed with the Power SYBR Green PCR Master Mix (Applied Biosystems). See Supplementary Table S3 for list of primers used. Expression was analyzed with the ΔΔCt method. The Ct values of the target genes were normalized to that of the housekeeping genes β-actin or GAPDH.

Genomic DNA from mouse thyroid tissues was used as a template for PCR amplification with primers that distinguish mutant (668 bp product due to insertion of loxp) from WT-Hras alleles (622 bp). The primers and PCR conditions were previously described (31).

Cells were treated with 1 μmol/L actinomycin D for 1, 3, 6, 12, and 24 hours prior to harvesting. The mRNA levels were measured by quantitative RT-PCR using Power SYBR Green PCR Master Mix (Applied Biosystems).

Cells were washed twice with cold PBS and lysed in lysis buffer (containing 125 mmol/L HEPES, pH 7.5, 750 mmol/L NaCl, 5% Igepal CA-630, 50 mmol/L MgCl₂, 5 mmol/L EDTA, and 10% glycerol) supplemented with proteinase and phosphatase inhibitors (Roche). Protein concentration was determined using the BCA Kit (Thermo Scientific). Western blots were conducted on 20 μg protein separated by 4% to 12% Bis-Tris SDS-PAGE gels (Invitrogen), transferred to PVDF membranes, and immunoblotted after blocking with 5% skim milk with the corresponding primary antibodies (listed below) in 5% BSA (SIGMA). This was followed by incubation for 1 hour with secondary antibodies conjugated with goat anti-rabbit horseradish peroxidase (HRP)–conjugated antibody (1:5,000; Santa Cruz; sc-2054) or goat anti-mouse HRP-conjugated antibody (1:5,000; Santa Cruz; sc-2005). Bound antibodies were detected by chemiluminescence with the ECL detection system (GE Healthcare Biosciences). In Figs. 2C, 3C, and 4C we show the same representative Merlin and Actin blots as aliquots of the same shNF2 clone M2-transduced C643 cell lysate were used for multiple different blots. Supplementary Fig. 11 shows the individual blots and the original films corresponding to Figs. 2C, 3C, and 4A. Aliquots of the same shNF2 clone M4-transduced C643 cell lysates were used for all blots in Figs. 2C and 3C.

RAS-GTP or RAC1-GTP immunoprecipitation was conducted using the RAS or RAC1 activation assay kits, respectively, from Millipore according to the manufacturer's protocol, and subjected to Western blotting with the indicated antibodies. Cells were treated during 72 hours with dox in media with 1% of FBS.

Subcellular fractions were prepared using the Subcellular Protein Fractionation Kit for cultured cells following the manufacturer's instructions (Thermo Fisher Scientific). Fractions were subjected to Western blotting with the indicated antibodies.

ChIP lysates isolated from Cal62, Hth83, and C643 cells were cross-linked with 1% formaldehyde for 10 minutes at room temperature and stopped by the addition of glycine to a final concentration of 125 mmol/L followed by 2 washes with PBS. The cell pellet was resuspended in Lysis buffer (ChIP-IT Express Enzymatic Kit-Active Motif) supplemented with phenylmethylsulfonylfluoride and protease inhibitors and incubated on ice for 30 minutes. After 20 strokes with a homogenizer and centrifugation for 10 minutes at 5,000 rpm, the pellets were digested with an enzymatic shearing cocktail for 7 minutes at 37°C, to produce chromatin fragments of 200 to 500 bp on average. Equal amounts of cross-linked DNA were immunoprecipitated overnight at 4°C with antibodies against YAP (sc-15407; Santa Cruz) or TEAD1 (#8526; Cell Signaling). Immunoprecipitates were incubated with proteinase K overnight, and the DNA recovered by purification (QIAquick PCR purification kit from Qiagen) prior to real-time PCR with primers bracketing the indicated promoter motifs. The ChIP ratio was calculated as enrichment over noise normalized to input.

Doxycycline (2 μg/mL), hEGF (5 ng/mL), and verteporfin were from Sigma. AZD6244 was from AstraZeneca. FRAX597 was from Selleckchem.

Statistical analyses of the results were performed by unpaired two-tailed Student t test or Mann–Whitney test according to assumptions of the test, using Prism v5.04 (GraphPad Software). Graphs represent mean value, and error bars represent SD. Similar variance between groups was tested by F Test; if different, Welch correction was applied. The P values are presented in figure legends where P < 5 × E-2 was considered statistically significant. No randomization or blinding was used.

All antibodies were used at 1:1,000, except where indicated. pMEK S217/221 (9121L), pERK (#4376), CCND1 (#2978P), pYAP S127 (#4911), YAP/TAZ (#8418), TUBULIN (#2148), pPAK T423 (#2601), PAK (#2604S), pCRAF S338 (#9427), pMEK S298 (#9128), pEGFR Y1068 (#3777S), EGFR (#4267S), GST (#2622), TEAD1 (#8526), and FLAG (#2368S) were from Cell Signaling Technology; NF2/MERLIN [(a-19) sc-331], HRAS (sc-520), KRAS (sc-30), NRAS (sc-31), CTGF (sc-14939), and YAP for ChIP (sc-15407) were from Santa Cruz Biotechnology; ACTIN (A2228; 1:5,000) and pPAK S141 (p7871) were from Sigma; and Na⁺K⁺ ATPase (ab818) and TATA binding (ab7671; 1:2,000) were from Abcam.

No potential conflicts of interest were disclosed.

Conception and design: M.E.R. Garcia-Rendueles, J.C. Ricarte-Filho, B.R. Untch, J.A. Knauf, J.A. Fagin

Development of methodology: M.E.R. Garcia-Rendueles, J.C. Ricarte- Filho, B.R. Untch, G. Oler, S.C. Jhanwar, K.H. Huberman, J.A. Fagin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.E.R. Garcia-Rendueles, J.C. Ricarte-Filho, B.R. Untch, V.E. Smith, I. Ganly, Y. Persaud, Y. Fang, A. Viale, A. Heguy, F. Giancotti, R. Ghossein, J.A. Fagin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.E.R. Garcia-Rendueles, J.C. Ricarte-Filho, B.R. Untch, I. Landa, J.A. Knauf, B.S. Taylor, Y. Fang, S.C. Jhanwar, F. Giancotti, R. Ghossein, J.A. Fagin

Writing, review, and/or revision of the manuscript: M.E.R. Garcia-Rendueles, J.C. Ricarte-Filho, B.R. Untch, I. Landa, I. Ganly, Y. Fang, S.C. Jhanwar, F. Giancotti, J.A. Fagin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.E.R. Garcia-Rendueles, J.C. Ricarte-Filho, J.A. Knauf, F. Voza, K.H. Huberman, J.A. Fagin

Study supervision: J.A. Knauf, J.A. Fagin

The authors thank the Molecular Cytology, Comparative Pathology, Animal Imaging, and Mouse Genetics Core Facilities for support of this project.

This work was supported, in part, by NIH grants CA50706, CA72597, P50-CA72012, and P30-CA008748; the Margot Rosenberg Pulitzer and Lefkovsky Family Foundations; the Society of Memorial Sloan Kettering, and the Byrne fund.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.