In a head and neck squamous cell carcinoma (HNSCC) “window of opportunity” clinical trial, we reported that trametinib reduced MEK-Erk1/2 activation and resulted in tumor responses in a subset of patients. Here, we investigated resistance to trametinib and molecular correlates in HNSCC cell lines and patient samples.
HNSCC cell lines were treated with trametinib to generate resistant lines. Candidate bypass pathways were assessed using immunoblotting, CRISPR knockout, and survival assays. Effectiveness of combined trametinib and verteporfin targeting was evaluated. Patient-derived xenografts (PDXs) from responder patients were treated with trametinib and resistant tumors were analyzed. Window trial clinical samples were subjected to whole-exome and RNA sequencing.
HNSCC cell lines developed resistance (CAL27-TR and HSC3-TR) after prolonged trametinib exposure. Downstream effectors of the Hippo pathway were activated in CAL27-TR and HSC3-TR, and combined trametinib and verteporfin treatment resulted in synergistic treatment response. We defined the Hippo pathway effector Yap1 as an induced survival pathway promoting resistance to trametinib in HSC3-TR. Yap1 was necessary for HSC3-TR trametinib resistance, and constitutively active Yap1 was sufficient to confer resistance in parental HSC3. Analysis of trametinib neoadjuvant trial patient tumors indicated canonical MEK-Erk1/2 pathway activating mutations were infrequent, and Yap1 activity increased following trametinib treatment. Trametinib treatment of a PDX from a responder patient resulted in evolution of resistance with increased Yap1 expression and activity.
These studies identify a Yap1-dependent resistance to trametinib therapy in HNSCCs. Combined Yap1 and MEK targeting may represent a strategy to enhance HNSCC response.
Targeting of the MEK/Erk pathway results in biomarker and clinical responses in only a subset of patients with head and neck squamous cell carcinoma (HNSCC). Defining adaptive resistance mechanisms that evolve in response to trametinib therapy may thus inform therapeutic approaches for these patients. Prolonged trametinib exposure resulted in evolution of resistance in cell line models, and YAP1 was identified as a pathway contributing to this resistance. Combination Yap1 and MEK targeting resulted in synergistic activity in a resistant model. Analysis of patient-derived samples and xenografts revealed upregulation of YAP1 pathways in response to trametinib. Our results highlight implications of a combinatorial MEK and YAP1 therapeutic strategy approach in patients with HNSCC.
Annually, more than 500,000 people worldwide receive the diagnosis of head and neck squamous cell carcinoma (HNSCC; refs. 1 and 2), and 60% succumb to their disease within 5 years (3). Of the reported head and neck cancer cases, approximately half were primary oral cavity squamous cell carcinomas (OCSCC; ref. 2). Despite significant advances in detection, surgical methods, and adjuvant radiation and chemotherapy, the prognosis for HNSCC has remained stable for years (1, 2, 4).
Genomic analysis has identified multiple signaling and regulatory proteins participating in HNSCC development and progression. These include aberrations in TP53, CDKN2A, PIK3CA, PTEN, and NOTCH1 (5, 6). In addition, upregulation of the Ras/Raf/MEK/Erk1/2 (MAPK) signaling pathway has been reported in OCSCC (7–11). The MAPK pathway regulates a breadth of fundamental cellular differentiation, proliferation, and migratory pathways, and as a result its dysregulation is a major driver in cancer progression and development (12). In other malignancies, oncogenic activation of the MAPK pathway often occurs through activating mutations in BRAF or RAS genes (13, 14). However, data from The Cancer Genome Atlas (TCGA) analysis has shown that Ras and Raf proteins are uncommonly mutated in HNSCC. HRAS-activating mutations are observed in 4%–8% of cases and NRAS, KRAS, or BRAF mutations are only rarely identified (15, 16).
Despite infrequent RAS/RAF mutation in HNSCC, the presence of an activated MAPK pathway is common. Several groups have reported IHC analysis of phosphorylated Erk1/2 (p-Erk1/2) in primary HNSCCs (7–11). These include two detailed HNSCC studies showing that 98% and 91% of tumors were positive for p-Erk1/2 staining, respectively (7, 17). In our previous study, we identified increased p-Erk1/2 activation to be associated with more aggressive tumor growth in a carcinogen-induced mouse model of OCSCC (18). Interestingly, p-Erk1/2 staining in patients with HNSCC was associated with advanced nodal stages and was increased in recurrent tumors compared with the tumor at initial presentation (7, 17), providing a rationale for therapeutically targeting the MEK pathway in HNSCCs. In a completed “window-of-opportunity” clinical trial, our group performed a neoadjuvant (preoperative) trial to determine the biomarker and tumor response of OCSCC to the allosteric MEK1/2 inhibitor trametinib (19). We demonstrated that brief trametinib treatment resulted in significant reduction in Erk1/2 pathway activation and both clinical and metabolic tumor responses in patients with OCSCC (19).
In BRAF-mutant melanoma, therapies targeting the MAPK pathway (Raf and MEK inhibitors) are transiently effective (13, 14), with eventual development of resistance. Numerous studies have described multiple resistance mechanisms to MAPK inhibition in Ras/Raf-mutant cancers: reactivation of the MAPK pathway leading to sustained Erk activation in the presence of BRAF inhibition (20), increased receptor tyrosine kinase (RTK)–mediated signaling of alternative survival pathways such as the PI3K-Akt-mTOR pathway (21), increased activation of the Src/FAK/STAT3 signaling axis (22, 23), and upregulation of the Hippo effector Yap1 (14). However, the mechanistic details of trametinib response and resistance have yet to be evaluated in HNSCC.
In this study, we generated HNSCC models of acquired trametinib resistance. Importantly, we identified that in cell line and HNSCC patient-derived xenograft (PDX) models, increased Yap1 expression and activity contributed to trametinib resistance without reactivation of the MAPK pathway. In addition, genetic or pharmacologic inhibition of Yap1 resensitized resistant cells to trametinib. Combination therapy using trametinib and the Yap1 inhibitor verteporfin synergized to enhance treatment response in HNSCC-resistant models. These studies support further clinical evaluation of combination MEK and Yap1 inhibitors in HNSCC therapy.
Materials and Methods
The HSC3 [JCRB, catalog No. JCRB0623, Research Resource Identifier (RRID):CVCL_1288], CAL27 (ATCC, catalog No. CRL-2095, RRID:CVCL_1107), and Detroit562 (ATCC, catalog No. CCL-138, RRID:CVCL_1171) lines were authenticated by short tandem repeat (STR) analysis, and experiments using HSC3, CAL27, and Detroit562 cells were from the same stocks that were validated by STR analysis. HSC3 cells were grown in RPMI1640 (Life Technologies). Detroit562 and CAL27 were grown in DMEM/F12 supplemented with glutaMAX (Life Technologies). HEK293T (NCBI_Iran, catalog No. C498, RRID:CVCL_0063) cells were grown in DMEM (Life Technologies). All cell lines were cultured in media supplemented with 10% heat-inactivated FBS (Hyclone) and penicillin/streptomycin (Lonza). All cell lines tested negative for Mycoplasma.
Patient sample acquisition
The protocols for tumor sample acquisition and correlative studies were approved by the Washington University (St. Louis, MO) Human Research Protection Office and Animal Studies Committee. Samples were obtained, following informed patient consent, from either surgical biopsies (prior to trametinib treatment) or resections (following trametinib treatment) from patents enrolled in a neoadjuvant clinical trial with trametinib (NCT01553851; WU IRB:201205124; ref. 19). All patients provided written, signed, informed consent to participate. This study followed ethical guidelines of the Declaration of Helsinki, Belmont Report and the U.S. Common Rule.
Generation of trametinib-resistant HNSCC models
The HSC3 and CAL27 trametinib-resistant (HSC3-TR, CAL27-TR) lines were generated by culturing HSC3 and CAL27 parental lines with sequentially increasing concentrations of trametinib (Selleckchem) up to 1 μmol/L trametinib. Simultaneously, HSC3 and CAL27 parental cells were cultured in media containing DMSO (0.01%, Sigma-Aldrich) to generate control lines (HSC3-C and CAL27-C).
Tumor biopsies from an HNSCC trametinib neoadjuvant phase II clinical trial (NCT01553851) were obtained under institutionally approved protocol (WU IRB 201205124). Primary xenografts were generated as described previously (24). Briefly, patient tumor biopsies were collected in sterile DMEM containing 10% FCS and 1% amphotericin. Biopsies were sectioned into multiple pieces with a portion used specifically for xenograft generation and another for next-generation sequencing analysis.
PDXs were generated as described previously, and tumor pieces from PDX2233 (P0 passage; refs. 19 and 25) were engrafted to a 5-week-old female NSG mouse (Jackson Laboratory). This PDX was further passaged into ten NSG mice to generate PDX2233 P1. Once tumor volumes reached 500 mm3, trametinib (3 mg/kg) was administered to tumor-bearing NSG mice by daily oral gavage. Trametinib (Selleckchem) was dissolved in DMSO (10 mg/mL) and diluted into an aqueous 100 μL dose containing 0.5% hypromellose (Sigma-Aldrich) and 2% Tween-80 (Sigma-Aldrich). Vehicle-treated mice received this same carrier solution (100 μL) administered daily by oral gavage to generate comparable control PDXs (PDX2233-C). Daily treatment began 7 days after implantation and continued for 150 days when tumor regrowth was typically observed. Tumor dimensions were measured twice weekly. All animal studies were performed under protocols approved by The Institutional Animal Care and Use Committees (IACUC) of Washington University (St. Louis, MO) and the Dana-Farber Cancer Institute (Boston, MA).
Cells (500,000) were plated in 6-well plates 24 hours before drug treatment. Cells were collected in lysis buffer (RIPA buffer, Cell Signaling Technology) supplemented with protease inhibitors (Sigma-Aldrich) and phosphatase inhibitor (EMD Millipore), and lysates were clarified by centrifugation. PDX tissues were pestle homogenized in lysis buffer and then clarified. Equal amounts of protein were separated by SDS-PAGE (4%–12%) and transferred onto nitrocellulose membranes (Life Technologies) for immunoblot analysis. Membranes were blocked with TBS blocking buffer (LI-COR) for an hour, then incubated with primary antibody overnight at 4°C, washed and incubated with secondary antibodies (LI-COR). Membranes were developed using a fluorescence system (LI-COR), and bands were quantified using LI-COR Image Studio Software. Primary antibody information is listed in Supplementary Material and Methods.
HNSCC cells were plated in 96-well cell culture plates (Corning) at 1,500 cells per well and incubated for 24 hours prior to treatment with varying doses of trametinib alone, verteporfin (Sigma-Aldrich) alone, or combination for 96 hours. As a surrogate for cell viability, ATP concentration was assessed by CellTiter-Glo Luminescent Cell Viability Assay (Promega) according to the manufacturer's instructions. Luminescent signal was measured using a TECAN plate reader according to the manufacturer's protocol. Cell metabolic data were fitted to variable slope normalized dose–response curves with media without cells representing maximum effect as has been reported elsewhere (26). IC50 values were calculated using GraphPad Prism (GraphPad Software). All experiments were performed at least three times with three technical replicates.
Combination therapy experiments were performed using a series of constant-ratio drug combinations with several concentration points above and below the IC50 values. The combination index (CI) was determined by the Chou–Talalay method using Calcusyn (Biosoft) software as described previously (24, 27). CI values were (27, 28) for synergy, additivity or antagonism of the drug combinations tested (24). All experiments were performed at least three times with at least two technical replicates.
Cell viability assay
Cells were plated 1,500 cells per well into 96-well culture plates and incubated for 24 hours prior to treatment. Varying doses of trametinib were then added, and treatment continued for 96 hours. Cell toxicity was assayed using CellTox Green (Promega G8472) according to the manufacturer's instructions for an endpoint method. Fluorescence was measured using a TECAN plate reader. Dose–response cell viability was analyzed using a variable slope (GraphPad Software). All experiments were performed at least twice with three technical replicates.
Cell proliferation assay
Prior to plating, a 96-well xCelligence plate with 50 μL of media per well was placed in the xCelligence RTCA SP instrument for 30 minutes to obtain a background reading. HNSCC cells were then added at 5,000 cells per well. The cell index was measured at 15-minute intervals for more than 120 hours to obtain a complete growth curve. Doubling times were calculated using the xCelligence RTCA software beginning 24 hours after plating and ending 10 hours before the plateau of the cell index. All experiments were performed at least three times with a minimum of three technical replicates.
RNA isolation, cDNA synthesis, and quantitative real-time RT-PCR
Total RNA was extracted from cells using the RNeasy Mini Kit (Qiagen). The cDNA was transcribed from total RNA using High Capacity RNA-to-cDNA Kit (Applied Biosystems), per manufacturer's protocol. The cDNA was used as a template for real-time PCR detection using TaqMan Fast Advanced Master Mix on an Applied Biosystems Step One Plus (Applied Biosystems). Expression of human YAP1, ANKRD1, NPPB, CTGF genes and endogenous control gene GAPDH were detected using Taqman probesets (Life Technologies). Relative target gene expression levels normalized to GAPDH were determined by ΔΔC(t) method (29).
Luciferase reporter assay
The 8 × GTIIC-luciferase plasmid, a synthetic TEAD luciferase reporter plasmid (RRID:Addgene_34615), and Renilla luciferase pRL-TK (RRID:Addgene_11313) were cotransfected into cell lines in a 6-well plate format. Plasmids were transfected using FuGENE HD Transfection Reagent (Promega) according to the manufacturer's protocol. After 48–72 hours, transfected cells were harvested, and lysates were analyzed using the Dual-Luciferase Reporter Assay Kit (Promega).
pLenti-CRISPRv2 containing sgNT (single-guide nontargeting RNA), sgYAP1 #1–5 pLKO.1 containing scramble-shRNA (short hairpin RNA) control, shYAP1 #1–5 pLKO.1 containing YAP1 targeting sequences (see Supplementary Table S1), pLX304-empty vector (RRID:Addgene_25890), pLX304-human YAP1-WT, and pLX304-human YAP1–5SA were generously provided by David Barbie (Dana-Farber Cancer Institute, Boston, MA). HEK293T cells were transfected with 1 μg lentivirus-based expression vectors and packaging plasmids using X-tremeGENE 9 transfection reagent (Roche) according to the manufacturer's instructions. After 48–72 hours incubation, the media containing lentivirus particles was collected, passed through a 0.45-μm filter and added to the target cells in the presence of 8 μg/mL polybrene. For selection of virally infected cells, 1–2 μg/mL of puromycin [pLKO.1 (RRID:Addgene_8453) and pLenti-CRISPRv2 guide-puro vectors] or 5 μg/mL of blasticidin (pLX304 vectors) was used 48 hours after transfection. Infected cell lines were selected for 1 week before being used for further experiments and were maintained in selection media.
In vivo treatment efficacy studies
HSC3-C and HSC3-TR cells (2 × 106 in PBS) were injected subcutaneously into the flanks of 6- to 8-week-old NSG mice. For PDX treatment studies, equal-sized pieces from 2233-TR sgNT and 2233-TR sgYAP1 #2 PDXs were implanted subcutaneously into NSG mice. Once tumors reached 100 mm3, trametinib was administered to tumor-bearing NSG mice by daily oral gavage. DMSO in hypromellose (0.5%) and 2% Tween-80 (250 μL) was used as vehicle solution. Digital caliper measurements of the tumors were taken twice weekly, and tumor volume (0.5 × length × width2) was calculated. These in vivo studies were performed using protocols approved by the Dana-Farber Cancer Institute (Boston, MA) IACUC.
Sequencing and data analysis
Nucleic acid extraction was performed by the Siteman Cancer Center Tissue Processing Core and sequencing was performed at the McDonnell Genome Institute at Washington University School of Medicine (St. Louis, MO), as described previously (25).
Genomic DNA was isolated from either tumor tissue, xenograft samples, or blood samples using the DNeasy Blood and Tissue Kit (Qiagen), and fragmented using the Covaris E210 DNA Sonicator (Covaris). Dual-indexed whole-exome sequencing (WES) libraries were constructed according to the manufacturer's instructions using the Kapa Auto Illumina (Kapa Biosystems) library preparation kit and captured using the NimbleGen SeqCap EZ Human Exome Library v3.0 Kit (Roche NimbleGen). WES libraries were pooled and sequenced on the Illumina HiSeq 4000 platform (2 × 150-bp reads). Sequencing data derived from xenograft samples were filtered in silico to remove reads that aligned to the mouse reference genome (mm10) by Xenome (30). The Genome Modeling System was used for all analysis, including reference genome alignment and somatic variant detection, as described previously (25, 31). We designed a custom capture reagent (NimbleGen SeqCap EZ Choice), targeting all single-nucleotide variants (SNV) and Indels detected by WES in tumors and xenografts. Validation capture sequencing was performed on variants in all WES libraries (20 normal blood, 20 baseline tumor biopsies, all xenograft samples); custom capture libraries were pooled and sequenced on the Illumina Novaseq platform (S4 flow cell). Data visualization was performed in R v3.3.2 using the ggplot2 R package (v2.2.1; ref. 32) and GenVisR v1.8.0 (33).
Total RNA was isolated from tumor and xenograft samples using the Qiagen RNeasy kits, and single-indexed RNA sequencing (RNA-seq) libraries were prepared using the Illumina TruSeq Stranded Total RNA kit with 500 ng of starting material according to the manufacturer's recommendations as described previously (25) and sequenced on the Illumina HiSeq 4000 platform (2 × 150-bp reads). RNA-seq reads derived from xenograft samples were aligned competitively against the human reference genome (NCBI build 38, GRCh38) and the mouse reference genome (Genome Reference Consortium Mouse Build 38, mm10) using Xenome (v1.0.0) to filter out murine-specific reads (30). Gene expression levels were quantified using Cufflinks v2.1.1 (RRID:SCR_014597; ref. 34) and HTSeq-count v0.5.4p1(RRID:SCR_011867; ref. 35). Differential expression analysis and statistics were performed on gene expression raw counts (from HTSeq) using the DESeq2 R package (RRID:SCR_000154; ref. 36). Subsequent gene set enrichment analysis and statistical analysis was performed on the log2 fold change (from DESeq2) in gene expression using the fgsea R package (37) across the Hallmark, KEGG (RRID:SCR_012773), and Reactome gene sets (38–40).
Genomic data have been deposited with dbGaP under study accession phs001623.
Statistical analyses were performed as indicated in the figure legends as either an unpaired t test or a one-way ANOVA with Dunnet multiple comparisons test. Asterisks were used to denote significance as follows: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Variable slope dose–response curves were used to fit cell metabolic activity data and to determine IC50 values. All calculations were carried out using GraphPad Prism (RRID:SCR_002798) software. Details of the statistics and number of experimental replicates can be found in figure legends.
Generation of trametinib-resistant HNSCC models
To delineate mechanisms of HNSCC trametinib response and resistance, we assessed the trametinib sensitivity of seven HNSCC lines (Fig. 1A). Of the seven HNSCC lines tested, HSC3, CAL27, and SCC25 were the most sensitive to trametinib, having IC50s less than 30 nmol/L. In addition, we determined that 100 nmol/L trametinib treatment suppressed Erk1/2 phosphorylation in all seven HNSCC cell lines (Fig. 1B), whereas 10 nmol/L trametinib treatment was insufficient to suppress Erk1/2 phosphorylation in HSC3, CAL27, and SCC25 dose response experiments (Fig. 1C). HSC3 and SCC25 do not harbor canonical RAS/BRAF-activating mutations whereas CAL27 harbors non-hotspot NRAS D92N and R68T mutations (41). To generate acquired trametinib-resistant cell line models as tools to define mechanisms of resistance, HSC3, CAL27, and SCC25 were subjected to prolonged trametinib treatment. Progressively increasing doses of trametinib (increased every 2–3 weeks) were applied to these three lines over a period of 6–12 months until they displayed little to no sensitivity to the drug. Resistance to trametinib developed in HSC3 (HSC3-TR) after 6 months and CAL27 (CAL27-TR) after 12 months, and their response to trametinib was compared with their respective control cells (HSC3-C and CAL27-C; Fig. 1D). Note that SCC25 did not develop resistance despite 12 months of exposure (not shown). HSC3 displayed a baseline IC50 of 0.027 nmol/L which increased to 9.06 μmol/L, whereas CAL27 had a baseline IC50 of 0.023 nmol/L which increased to 2.29 μmol/L. Cell growth doubling times did not differ between HSC3-C and HSC3-TR nor between CAL27-C and CAL27-TR (Fig. 1E). Thus, two of three lines displayed acquired resistance to trametinib.
Acquired trametinib-resistant HNSCC model reveals upregulation of Yap1 expression and activity
Multiple mechanisms of acquired resistance to MAPK inhibition have been reported in cancers harboring RAS/RAF mutations. These include reactivation of the MAPK pathway (20), activation of parallel signaling pathways such as the PI3K-AKT-mTOR pathway (21), increased activation of the Src/FAK/STAT3 signaling axis (22, 23), and upregulation of the Hippo effector Yap1 (13, 14). On the basis of these findings, we used a candidate approach to ask whether any of these alterations were present in the models we developed. We evaluated the activity of the MAPK, PI3K-AKT-mTOR, and STAT3 signaling pathways and expression of Yap1 by immunoblot. We noted unabated activation of the MAPK pathway in the presence of trametinib only in CAL27-TR cells, as p-Erk1/2 levels were sustained in this line following trametinib treatment (Fig. 2A). As reported previously (23), we observed increased STAT3 phosphorylation following MEK inhibition in CAL27-C and HSC3-C (Fig. 2A). This increase in p-STAT3 was more modest in trametinib-resistant model cell line HSC3-TR compared with HSC3-C, but this difference was not statistically significant (Fig. 2A). Phospho-AKT (p-AKT) expression was reduced in CAL27-C but sustained in CAL27-TR following trametinib treatment, whereas it increased in HSC3-C and was generally unchanged in HSC3-TR lines following trametinib treatment. (Fig. 2A). Despite sustained p-AKT levels in CAL27-TR compared with CAL27-C, we did not observe elevated activity of the downstream effectors of the AKT pathway such as phosphorylated ribosomal protein S6 kinase (p-S6; ref. 42), and phospho 4E-BP1 (p-4E-BP1; ref. 43) in CAL27-TR as compared with CAL27-C (Supplementary Fig. S1).
HSC3-TR did not show increased p-STAT3, reactivation of the MAPK pathway, nor increased activity of the PI3K-AKT-mTOR pathways compared with HSC3-C (Fig. 2A; Supplementary Fig. S1). Conversely, HSC3-TR showed a distinct pattern where we observed increased Yap1 protein expression in HSC3-TR (Fig. 2A) as compared with parental HSC-3. This increased Yap1 protein expression occurred with a 3-fold increase in YAP1 mRNA expression in HSC3-TR as compared with HSC3-C (Fig. 2B). We next asked whether elevated YAP1 expression was associated with increased functional activity using a TEAD-Yap1 luciferase-based gene expression reporter assay (44). Consistent with increased protein expression, HSC3-TR cells showed enhanced Yap1 activity compared with the control HSC3 (Fig. 2C; P < 0.0001). Furthermore, we found that transcripts of Yap1 target genes, including CTGF, ANKRD1, and NPPB (45) were increased in HSC3-TR (Fig. 2D). CAL27-TR, which did exhibit sustained MAPK signaling in the presence of trametinib, did not have increased Yap1 protein nor mRNA expression compared with CAL27-C (Fig. 2A and B). However, TEAD luciferase activity and expression of Yap1 target genes CTGF, ANKRD1, and NPPB were significantly increased in CAL27-TR compared with CAL27-C (Fig. 2C and D), suggesting the possibility of Yap1 activation in CAL27-TR. We decided to pursue the significance of increased YAP pathway activation in mediating acquired resistance to trametinib in these models.
Pharmacologic inhibition of Yap1 is synergistic with trametinib in acquired and intrinsic trametinib-resistant HNSCC models
Because Yap1 was implicated in trametinib resistance, we sought to define whether treatment with the Yap inhibitor verteporfin in combination with MEK targeting could bypass trametinib resistance in the HSC3-TR and CAL27-TR HNSCC models of acquired resistance. We determined the verteporfin IC50 in HSC3-TR and CAL27-TR to be 4.93 and 6.54 μmol/L, respectively. We then assessed trametinib and verteporfin synergy using the Chou–Talalay method (27). The resulting isobolograms and CIs (Fig. 3A and B) provided evidence of synergy between trametinib and verteporfin in these two HNSCC models of acquired trametinib resistance. We also assessed verteporfin and trametinib synergy in HSC3-C and CAL27-C using the same methodology (Supplementary Fig. S2). Verteporfin IC50s were determined to be 5.25 and 4.41 μmol/L for HSC3-C and CAL27-C, respectively. The ED50 and ED75 CIs for HSC3-C were 1.13 and 0.98, indicating slight antagonism and nearly additive effects, respectively. This was in contrast to the ED50 and ED75 CIs for HSC3-TR, which were 0.87 and 0.68, indicating slight synergism and synergism, respectively. These analyses support the acquisition of greater synergy of targeting Yap and MEK in HSC3-TR compared with HSC3-C. CAL27-TR and CAL27-C had more similar trametinib-verteporfin synergies, as both lines displayed moderate synergism at ED75 and ED90, whereas the ED50 CIs indicated verteporfin and trametinib had slight synergism in CAL27-TR and were nearly additive in CAL27-C. We attempted to extend these studies to the in vivo setting for testing trametinib resistance with Yap1 cotargeting by engrafting the HSC3-TR cells into NSG mice. However, resistance to trametinib was lost upon engraftment, with tumors displaying drug resensitization, negating our ability to assess in vivo combination effects (data not shown).
Confirming Yap1 targeting with verteporfin treatment in vitro, we noted a significant reduction in CTGF, ANKRD1, and NPPB mRNAs upon verteporfin treatment in HSC3-TR and CAL27-TR (Fig. 3C). These data confirmed appropriate “on target” activity of verteporfin in these lines.
Subsequently, we sought to determine whether Yap1 contributed to intrinsic trametinib resistance in distinct HNSCC cell lines. Yap1 expression levels were evaluated using a panel of 12 HNSCC cell lines, including HSC3-C, HSC3-TR, and CAL27. Detroit562 had the highest level of Yap1 protein (Supplementary Fig. S3A) and displayed reduced sensitivity to trametinib (IC50 = 1.89 μmol/L), compared with three selected HNSCC cell lines, SNU899, HSC2, and BICR56 (Supplementary Fig. S3B). Of note, 100 nmol/L trametinib treatment inhibited Erk1/2 phosphorylation in Detroit562 (Supplementary Fig. S3C). Trametinib and verteporfin IC50s in Detroit562 cells were determined to be 1.89 and 5.16 μmol/L, respectively. Performing a series of constant-ratio drug combinations with several concentration points above and below the IC50, we determined the CIs for verteporfin and trametinib. Only the ED75 CI demonstrated synergy in Detroit562 (Fig. 3A, right and 3B). The CIs at ED50 and ED90 indicated antagonism, suggesting that the synergism in Detroit562 occurred within a more narrow dosing range than observed with HSC3-TR and CAL27-TR. We confirmed that verteporfin treatment of Detroit562 resulted in reduced expression of the Yap target genes CTGF and NPPB, whereas expression of ANKRD1 was not significantly reduced following verteporfin treatment at the doses evaluated (Fig. 3C). Together, our results suggest that pharmacologic Yap1 targeting can counteract intrinsic and acquired trametinib resistance in these HNSCC models.
Yap1 mediates trametinib resistance in HNSCC models
As verteporfin has targets in addition to Yap1, we next asked whether Yap1 was specifically necessary for mediating trametinib resistance in these models. We employed multiple approaches to deplete Yap1 expression. First, we used CRISPR/Cas9 to knockout YAP1 in HSC3-TR using five distinct YAP1 sgRNAs or a nontargeting sgRNA (sgNT). All five YAP1 sgRNAs were found to attenuate Yap1 levels. sgYAP1 #2 and sgYAP1 #4 resulted in the most effective Yap1 reduction and were, thus, selected for subsequent functional studies (Fig. 4A). Using these Yap1-deficient lines, we assessed Yap1 functional activity using the TEAD-Yap1 reporter and observed decreased activity in HSC3-TR-sgYAP1 #2 and HSC3-TR-sgYAP1 #4 cell lines as compared with the control line HSC3-TR-sgNT (Fig. 4B; P < 0.0001 and P < 0.0005, respectively). Expression of Yap1 target genes ANKRD1 and NPPB in HSC3-TR Yap1-depleted cell lines was significantly diminished (Fig. 4C; P < 0.0001). These data confirmed effective knockdown of Yap1 expression and activity in the HSC3-TR line.
We next assessed whether YAP1 knockout restored trametinib sensitivity in HSC3-TR. CRISPR/Cas9 YAP1 knockout (Fig. 4D) resulted in increased sensitivity to trametinib, with sgYAP1 #2 showing an IC50 = 0.064 μmol/L and sgYAP1 #4 IC50 = 0.099 μmol/L compared with an sgNT IC50 = 12.46 μmol/L in HSC3-TR cells. Doubling times in the YAP1 knockdown lines did not differ from the NT control (Supplementary Fig. S4A). Knockdown of YAP with sgYAP#1, sgYAP#3, and sgYAP#5 also resulted in increased HSC3-TR sensitivity to trametinib, though not to the same extent as sgYAP#2 and sgYAP#4 (Supplementary Fig. S4B). In parallel, we also utilized shRNAs to knock down Yap1 in HSC3-TR cells and determined that YAP1 knockdown by shRNA increased sensitivity to trametinib, although to a lesser degree (Supplementary Fig. S4C and S4D). Conversely, we tested whether enforced expression of wild-type (YAP1 WT) or constitutively active YAP (YAP1 5SA) were sufficient to increase trametinib resistance in trametinib-sensitive HNSCC lines (Fig. 4E–G). Interestingly, we found that although Yap1 WT and YAP1 5SA were similarly overexpressed and similarly enhanced TEAD activity (Fig. 4E), only stable expression of constitutively active Yap 5SA substantially increased expression of all three Yap target genes and decreased sensitivity to trametinib in HSC3 cells, resulting in an IC50 = 30.42 μmol/L as compared with HSC3 infected with its empty vector control IC50 = 0.082 μmol/L (Fig. 4G). Stable overexpression of YAP1 WT did not decrease trametinib sensitivity in HSC3 (IC50 = 0.18 μmol/L; Fig. 4G). Together, these data confirm the contribution of Yap1 activity to trametinib sensitivity in HSC3 cells.
Similarly, we evaluated whether YAP1 knockout restored trametinib sensitivity in CAL27-TR cells (Supplementary Fig. S5). YAP1 knockout with sgYAP1 #1, sgYAP1 #2, and sgYAP1 #5 all yielded reduced Yap1 protein by immunoblot (Supplementary Fig. S5A). However, trametinib sensitivity in the CAL27-TR YAP1 knockdowns did not differ from the parental line (Supplementary Fig. S5B). CAL27-TR transduced with the NT sgRNA had substantially increased trametinib IC50 compared with the parental line, 19.48 and 2.95 μmol/L, respectively, possibly reflecting off-target effects of the NT sgRNA in these cells. Consistent with YAP1 knockdown, we observed Yap1 downstream target genes had reduced expression in CAL27-TR with sgYAP #1, sgYAP1 #2, and sgYAP1 #5 compared with the parental CAL27-TR and CAL27TR-NT sgRNA controls (Supplementary Fig. S5C). ANKRD1 expression remained 4-fold, 15-fold and 20-fold higher than CAL27-C expression levels in CAL27-TR expressing YAP1sg #2, YAP1sg #5, and YAP1sg #1, respectively. We did not test sgYAP#3 or sgYAP#4 in the CAL27 models.
To investigate whether Yap1 contributed to Detroit562 response to trametinib, we employed CRISPR/Cas9 to deplete YAP1 (Fig. 4H). In Detroit562 sgYAP1 #2 and sgYAP#5, there was significant downregulation of Yap1-target genes as compared with the control Detroit 562-sgNT (Fig. 4I) and a concomitant 6-fold and 60-fold reduction in IC50 for sgRNA #2 (IC50 = 0.25 μmol/L) and sgRNA #5 (IC50 = 0.019 μmol/L), respectively, compared with the nontargeting control (IC50 = 1.211 μmol/L; Fig. 4J). Together, these data indicate that Yap1 modulates trametinib resistance in HSC3 and Detroit562 HNSCC models.
All studies of in vitro response to trametinib and/or verteporfin described above were performed using ATP concentration as a surrogate readout for cell viability. To directly assess cell cytotoxicity in response to drug treatment and/or Yap1 knockdown, we measured cytotoxicity directly in the HSC3-C and HSC3-TR models using Celltox green (Supplementary Fig. S6). We observed trametinib treatment induced a dose-dependent cytotoxicity in HSC3-C cells (IC50 = 2.13 μmol/L), which was higher than the IC50 based on cellular ATP levels. HSC3-TR cells exhibited only modest cytotoxicity in response to high doses of trametinib (IC50 = 26.06 μmol/L; Supplementary Fig. S6A). Neither verteporfin in combination with trametinib nor Yap1 knockdown in combination with trametinib induced cytotoxicity in HSC3-TR by these assays (Supplementary Fig. S6B–S6D). These data indicate that although targeting Yap1 in combination with trametinib reduced cellular ATP concentration, this combined targeting did not induce cell death in this model of acquired trametinib resistance.
An HNSCC PDX with acquired trametinib resistance displays increased Yap1 expression
Having examined cell line model response and resistance, we next queried whether PDX models derived from patient tumors from our clinical trial retained sensitivity to trametinib. As reported previously, of 40 tumor biopsies (20 baseline; 20 posttreatment surgical specimens), 22 (55%) showed successful engraftment and displayed genomic fidelity with primary tumors (25). As our window trial was limited to only a 2-week period, we utilized PDX models from the baseline biopsy of two clinical trial patients for an extended treatment approach. We previously described one of these PDXs, which showed a partial response to trametinib, paralleling the clinical response observed in the patient (25). A second sample, PDX2233, was derived from a patient whose tumor displayed attenuation of Erk1/2 phosphorylation during the window phase (Supplementary Fig. S7) and significant metabolic tumor response (19). Mice bearing PDX2233 treated with vehicle showed progressive tumor growth, whereas daily trametinib treatment resulted in near complete tumor regression. However, after extended daily trametinib treatment, resistant tumor outgrowth was observed (Fig. 5A and B). Tumors were harvested and analyzed by immunoblotting, RNA-seq, and WES. When we interrogated for specific candidate pathways, four of four acquired trametinib-resistant PDX2233 (PDX2233-TR) escape PDXs showed increased Yap1 levels whereas one of three vehicle-treated, trametinib-sensitive tumors showed modest Yap1 expression (Fig. 5C). Note that the PDX2233-TR tumor samples showed sustained p-Erk1/2 levels. To ask whether this sustained Erk1/2 phosphorylation could be due to in vivo trametinib levels and timing of tumor harvest (Fig. 5C), we generated transient cultures from PDX2233-TR for sustained trametinib exposure. Following 4 hours of trametinib treatment in vitro, primary cell lines derived from PDX2233-TR displayed ablation of p-Erk1/2 activity (Supplementary Fig. S8A and S8B). Similar to HSC3-TR, PDX2233-TR did not show increased PI3K-AKT or STAT3 signaling nor reactivation of the MAPK pathway activity (Fig. 5C and D). Furthermore, we transiently cultured PDX2233-TR to knockout YAP1 using CRISPR/Cas9 (Supplementary Fig. S8C and S8D). YAP1 depletion in PDX2233-TR (PDX2233-TR-sgYAP #2) resulted in decreased in vivo growth compared with the control PDX2233-TR-sgNT (Supplementary Fig. S8E). Therefore, we were unable to assess the impact of trametinib in these modified lines.
To explore potential mechanisms by which PDX2233-TR was no longer responding to trametinib, we completed WES, high-depth custom-capture validation sequencing, and RNA-seq to explore both genomic and expression-based mechanisms driving this escape phenotype in vivo. There were 94 somatic SNVs and small insertions or deletions (indels) detected in PDX2233 (passage 0, PDX2233-P0) derived from this patient; 90 of these variants were also detected in the primary tumor (Supplementary Table S2). The four variants specific to PDX2233-P0 were present at 10.5%–30.1% DNA VAF (missense mutations in ZNF326, FRG1, POU6F2, and MAGEC1). PDX2233-TR samples (n = 3; passage 3) shared 92–93 variants with the P0 xenograft; PDX2233-C samples (n = 3; passage 3) shared 91–92 variants with the P0 xenograft. There were 7–9 (median, 7) new variants in PDX2233-C samples and 7–10 (median, 9) new variants in PDX2233-TR samples. Three variants were shared by PDX2233-C and PDX2233-TR samples, and remaining variants were present at lower VAF, suggesting that these were acquired over passaging since PDX2233-P0 (Supplementary Table S2) and not associated with acquired resistance to trametinib. In addition, de novo variants in PDX2233-TR (passage 3) did not have a clear biological association with the YAP pathway. Collectively, these results suggest that trametinib resistance may not be driven by a genetic event.
Differentially expressed gene sets in PDX2233-TR samples, compared with PDX2233-C samples, were associated with general transcriptional regulation (REACTOME GENERIC TRANSCRIPTION PATHWAY, Wald test; P = 0.001, FDR = 0.077) or xenobiotic metabolism (REACTOME PHASE 1 FUNCTIONALIZATION OF COMPOUNDS; Wald test, P = 0.001, FDR = 0.077). PDX2233-C samples showed upregulation of cell-cycle processes, compared with PDX2233-TR samples (Supplementary Tables S3 and S4), consistent with the rapid growth observed in these xenografts. YAP1 was not differentially expressed between the PDX2233-TR and PDX2233-C xenografts, and neither were its downstream effectors CTGF and CYR61. However, its effector ANKRD1 was more highly expressed in the PDX2233-TR (Wald test, log2FC = 3.39, P = 1.44e-5, FDR = 0.003; Supplementary Table S3).
Genomic correlates from HNSCC trametinib window trial
We investigated whether HNSCC clinical trial patients displayed patterns of YAP1 activity with relation to trametinib treatment (19), including whether patients exhibit patterns of YAP1 pathway activity at baseline or following treatment that may suggest intrinsic or acquired resistance evolution to trametinib. WES was performed on baseline tumor biopsies (n = 20) and RNA-seq analysis on paired baseline tumor biopsies (n = 16) and posttreatment surgical resections (n = 15; 10 matched). Tumor DNA was sequenced with an average of 69× coverage, and matched normal DNA was sequenced with an average of 72× coverage across the targeted exome. SNVs and indels were detected by comparing tumor samples with matched normal blood; there were a total of 1,853 nonsilent SNVs and indels detected across the 20 primary tumor samples (average 101 variants per patient; Supplementary Table S5). SNVs and indels were validated by high-depth sequencing in all samples (average 544× coverage across targeted regions). The genomic landscape of our cohort was highly concordant with the genomic landscape of human papillomavirus-negative HNSCC, including recurrent mutations in TP53, CDKN2A, USH2A, NOTCH1, and CASP8 (Fig. 6; ref. 16). Known activating mutations were observed in PIK3CA (E545K, responder patient GSK2899 and nonresponder GSK4795) and HRAS (G13I, patient GSK1978, response not evaluable). There were no clear differences in the genomic profiles between responders and nonresponders.
YAP1 pathway activity in patient samples
Based upon our observations in preclinical models, we evaluated whether patient samples depicted patterns of either intrinsic or acquired resistance evolution to trametinib via the YAP1 pathway by evaluating the expression of YAP1 and its downstream targets. Interestingly, YAP1 was more highly expressed at baseline in trametinib responders (n = 4), compared with nonresponders (n = 12; log2FC = 1.89, P = 1.73e-6, q = 0.002; Supplementary Table S6); however, this significance was strongly driven by 2 of the 4 patients (Fig. 6A–C). Yap targets CTGF, CYR61, and ANKRD1 were not differentially expressed at baseline between responders and nonresponders. When pretreatment and posttreatment samples were compared, YAP1 expression was not significantly modulated in either responders or nonresponders. All three Yap1 targets were significantly higher in nonresponders posttreatment (P < 0.05, q < 0.20 for all genes), and responders showed trends of increased Yap1-target expression posttreatment (P = 0.103–0.456; q = 1). Together, these results suggest that this gene program may be increased because of trametinib treatment.
Patient GSK2233 and its corresponding PDX samples consistently exhibited high expression of YAP1, compared with the rest of the cohort. Only ANKRD1 was significantly upregulated in PDX2233-TR, compared with PDX2233-C (P = 1.44e-05, FDR = 0.0027; Fig. 6E); CTGF and CYR61 were not differentially expressed between PDX2233-C and PDX2233-TR.
Although the RAS/MAPK/Erk pathway is infrequently mutated in HNSCCs, increased pathway activity in tumor cells highlight this pathway as a potential therapeutic target. To address validity of targeting of this pathway in patients, we completed a window of opportunity clinical trial (NCT01553851) with trametinib and a p-Erk1/2 biomarker endpoint (19). Interestingly, we identified several patients who not only showed biomarker responses but also metabolic and clinical activity in response to trametinib. Here, we further explored these findings in laboratory models to begin to define response and resistance mechanisms to MEK inhibition in HNSCCs. We observed increased Yap1 activity as a shared mechanism of trametinib resistance in models of both intrinsic (Detroit562) and acquired resistance (HSC3-TR). In these settings, resistance was established or acquired by a mechanism other than sustained Erk signaling. We found that Yap1 function was necessary and sufficient for this activity as enforced expression of constitutively active Yap1 conferred resistance in a sensitive model and genetic ablation resulted in sensitivity in resistant models. Confirming that Yap1 acts as a bypass pathway for MEK inhibition, pharmacologic blockade of Yap1 resulted in synergy with trametinib on resistant cell lines. Extending these data to a PDX derived from a patient who exhibited a clinical response to trametinib, we observed development of acquired resistance accompanied by increased Yap1 expression and activity. Patient tumor baseline YAP1 expression was not associated with expression of its target genes (ANKRD1, CTGF, CYR61); however, the expression of Yap1 targets were overall increased following 14 days of trametinib treatment. These findings confirm the importance of YAP1 pathway activity in mediating trametinib resistance and suggest inhibition of Yap1 as a potential strategy to enhance the efficacy of trametinib in patients with HNSCC.
Despite the activation of Hippo pathway effectors, we were unable to establish Yap1 dependency for trametinib resistance in CAL27-TR, whereby resistance was acquired concomitantly with maintenance of Erk signaling. Sustained Erk activity despite the presence of trametinib suggests CAL27-TR specific compensatory signaling pathway(s) that differ from the other trametinib-resistant models presented, which bypass Erk signaling with Yap1 dependency. It remains plausible that a Hippo pathway signaling component other than Yap1, for example, TAZ, is a primary mediator of trametinib resistance dependence. At present, the mechanism for acquired trametinib resistance in this model remains undefined.
TCGA analysis identified 5% of HNSCC tumors with canonical HRAS mutations (codon 12, 13, or 61) with very low mutation rates in KRAS, NRAS, or BRAF genes. Genomic analysis of our 20-patient cohort showed that 19 patient tumors did not bear any canonical drivers of MEK dependency and one patient tumor (5%) harbored an HRAS mutation. This latter patient did not complete the trial and thus was not evaluable. Similarly, we were not able to identify driver(s) of MEK dependency in HSC3, CAL27, or Detroit562. However, published proteomic analysis (7, 17) shows increased Erk1/2 phosphorylation in HNSCCs consistent with other drivers impinging on this pathway. Similar to breast cancer, where again RAS pathway mutations are rare, this activation is likely the result of aberrant signaling from RTKs (46). These data formed the basis of our initial studies targeting MEK as a central and important node in key signaling pathways in HNSCCs.
Toward further defining mechanisms of resistance, we used a candidate approach and identified increased YAP1 expression and activity in HSC3-TR and a similar correlation in the escape tumors from a PDX derived from a trametinib responder. Yap1 was identified as a bypass pathway for BRAF inhibitor therapy in BRAF-mutant cells using an unbiased signaling pathway shRNA library approach (9). Our current work in non-RAS/RAF–mutant HNSCC models and a responder patient is consistent with YAP1 expression and pathway/target activation as drivers of acquired resistance to MEK inhibition. Of note, our results show that elevated Yap1 protein expression and activity is supported by its pathway/target activation without necessarily its mRNA upregulation in trametinib-resistant HNSCC models. In addition, WES did not reveal selective growth of tumor subclones in PDXs or specific protein-coding alterations to explain development of resistance. RNA-seq showed shifts in expression, but these general transcription pathways are modulated by a variety of transcription factors. The epigenetic landscape was not queried in this study, but our results suggest that resistance is mediated by nongenetic mechanisms.
Several questions remain regarding the mechanism by which Yap1 induces trametinib resistance in HNSCC. Previous reports have shown that Yap1 signaling reduces apoptosis and promotes resistance in RAS/RAF-mutant malignancies (14, 47). As a result, Yap1 inhibition resensitized resistant models to MAPK inhibitors by inducing greater apoptosis (14, 47). Furthermore, YAP1 has been implicated in the epithelial-to-mesenchymal transition program, a critical regulator of the cancer stem cell (CSC) phenotype (48). CSC resistance to chemotherapeutic agents has been attributed to increased expression of antiapoptotic proteins, elevated levels of ATP-binding cassette transporters, transmembrane protein transporters that are important for drug efflux and, thus, result in multidrug resistant cancer cells (48). Bromodomain containing protein-4 (BRD4) has been shown to mediate cetuximab resistance by upregulating the transcription of various RTKs in HNSCC (49). Because Yap1 is known to physically interact with BRD4 (50), this association may confer a transcriptional advantage of genes that mediate trametinib resistance in HNSCC.
There are several limitations of this study including the number of cell line models tested and patient samples available, definition of the mechanism of Yap1-mediated bypass activity and limited transcriptomic data with respect to matched patient samples. We subjected three HNSCC cell lines to extended trametinib treatment based on the number of lines we were able to identify with baseline drug sensitivity. We tested two PDXs for extended trametinib treatment [PDX2233 shown in this work and PDX4817 (25)] again as we focused on PDXs that were available from responders in our clinical trial. We tested multiple potential Yap1-mediated resistance pathways (negative data, not shown) and our future work aims to further explore potential nongenetic mechanisms. Finally, in vivo validation of combination Yap1 and MEK targeting in HSC3-TR was not feasible as resistance was not maintained when resistant xenografts were reimplanted into untreated mice. The cytostatic impact with reduced cellular ATP without cytotoxicity limits combination therapeutic targeting clinically and is commonly seen with other targeted therapeutics (51). For PDX2233, YAP1 was required for growth thus limiting a CRISPR-based approach (Supplementary Fig. S8). This rapid reversibility of resistance may suggest against permanent genetic changes in these models as an underlying mechanism. This may be addressed by targeting resistance as it evolves for example in the PDX2233 model. We will integrate this approach in future work as we aim to further delineate Yap1 contribution to HNSCC progression and resistance to therapy.
In conclusion, as trametinib therapy resulted in clinical responses, we extended these findings and identified Yap1 as a resistance mechanism in HNSCCs. These data are consistent with findings in BRAF/RAS-mutant tumors and our work extends these data to HNSCC where RAS mutations are less common. Our current work is aimed at defining mechanistic aspects of Yap1-mediated trametinib resistance and approaches integrating MEK targeting with immunotherapy in HNSCCs.
K.M. Campbell is a shareholder in Geneoscopy LLC and received consulting fees from PACT Pharma. Z.L. Skidmore reported grants from NIH during the conduct of the study. E.K. Barnell reported personal fees from Geneoscopy outside the submitted work; in addition, E.K. Barnell had a patent for Geneoscopy pending. D.R. Adkins reported grants from GSK during the conduct of the study; grants and personal fees from Pfizer, Celgene/BMS, Eli Lilly, Merck, and Cue Biopharma; grants from Aduro, Novartis, AstraZeneca, Atara Bio, Blueprint Medicine, Celldex, Kura, Exelixis, Innate, Sensei, Matrix Biomed, Hookipa, CoFactor, Medimmune, and DeNovo outside the submitted work. O.L. Griffith reported grants from the NIH during the conduct of the study. R. Uppaluri reported grants from the NCCN and the NIH during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.
T. Mudianto: Conceptualization, investigation, writing–review and editing. K.M. Campbell: Conceptualization, software, formal analysis, investigation, visualization, writing–review and editing. J. Webb: Investigation, methodology, writing–review and editing. P. Zolkind: Investigation, writing–review and editing. Z.L. Skidmore: Formal analysis, investigation. R. Riley: Investigation. E.K. Barnell: Data curation, investigation, visualization. I. Ozgenc: Investigation. T. Giri: Investigation. G.P. Dunn: Resources, writing–review and editing. D.R. Adkins: Resources, funding acquisition, writing–review and editing. M. Griffith: Formal analysis, supervision, writing–review and editing. A.M. Egloff: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration. O.L. Griffith: Conceptualization, software, formal analysis, supervision, investigation, writing–review and editing. R. Uppaluri: Conceptualization, resources, formal analysis, supervision, funding acquisition, visualization, methodology, writing–original draft, project administration.
R. Uppaluri was supported by NIH/NIDCR DE024403, DE027736 and a V Foundation Translational Research Award. Additional funding was from the National Comprehensive Cancer Network (NCCN) Oncology Research Program from general research support provided by Novartis Pharmaceutical Corporation (Novartis). O.L. Griffith was supported by the NCI (NIH NCI K22CA188163, NIH NCI U01CA209936, and NIH NCI U24CA237719) and a Cancer Research Foundation Young Investigator Award. We thank S. Kitajima and D. Barbie for YAP1 CRISPR and shRNA plasmids.
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.