Abstract
Lung cancer is the leading cause of cancer-related deaths worldwide. The paralogous transcriptional cofactors Yes-associated protein (YAP) and transcriptional coactivator with PDZ-binding motif (TAZ, also called WWTR1), the main downstream effectors of the Hippo signal transduction pathway, are emerging as pivotal determinants of malignancy in lung cancer. Traditionally, studies have tended to consider YAP and TAZ as functionally redundant transcriptional cofactors with similar biological impact. However, there is growing evidence that each of them also possesses distinct attributes. Here we sought to systematically characterize the division of labor between YAP and TAZ in non–small cell lung cancer (NSCLC), the most common histological subtype of lung cancer. Representative NSCLC cell lines as well as patient-derived data showed that the two paralogs orchestrated nonoverlapping transcriptional programs in this cancer type. YAP preferentially regulated gene sets associated with cell division and cell-cycle progression, whereas TAZ preferentially regulated genes associated with extracellular matrix organization. Depletion of YAP resulted in growth arrest, whereas its overexpression promoted cell proliferation. Likewise, depletion of TAZ compromised cell migration, whereas its overexpression enhanced migration. The differential effects of YAP and TAZ on key cellular processes were also associated with differential response to anticancer therapies. Uncovering the different activities and downstream effects of YAP and TAZ may thus facilitate better stratification of patients with lung cancer for anticancer therapies.
Thease findings show that oncogenic paralogs YAP and TAZ have distinct roles in NSCLC and are associated with differential response to anticancer drugs, knowledge that may assist lung cancer therapy decisions.
Introduction
Lung cancer is the leading cause of cancer-related deaths worldwide. Despite significant progress in understanding the causes of lung cancer, the 5-year survival is still below 15% (1). Therefore, there is an urgent need for better understanding of the molecular mechanisms that drive lung cancer to devise more effective treatments.
The paralogous transcriptional cofactors Yes-associated protein (YAP) and transcriptional coactivator with PDZ-binding motif (TAZ, also called WWTR1), the main downstream effectors of the Hippo signaling pathway, are emerging as pivotal determinants of malignancy in lung cancer (2, 3). However, Hippo pathway genes, including YAP/TAZ, are rarely mutated in tumors, with only a few exceptions (4, 5), and thus YAP/TAZ dysregulation in cancer is likely to be driven by other mechanisms.
One open question in the YAP/TAZ field is to which extent these paralogs are functionally redundant. Although YAP and TAZ possess only nearly 57% amino acid sequence similarity (Supplementary Fig. S1), they are often regarded as functionally redundant, with similar biological impact (3, 6–9). Indeed, both participate in diverse biological processes, including tissue homeostasis, development, organ growth, regeneration, stem cell regulation, and mechanotransduction (10). Yet, a variety of features suggest that they may differ in their regulation and downstream activities (11–13). For example, YAP and TAZ display distinct knockout phenotypes during nephron development (14). Notably, evidence for their different functions in lung cancer is also emerging (15–17). For example, YAP, but not TAZ, inhibits squamous trans-differentiation of lung adenocarcinoma cells (15). Furthermore, TAZ protein levels are higher in lung cancer cell lines relative to normal lung epithelial cells, whereas YAP levels are more or less comparable (17). Moreover, lung cancer-derived cells depleted of TAZ, but not YAP, are more sensitive to bromodomain and extraterminal (BET) domain protein inhibitors (16). Nevertheless, the extent of non-redundancy between YAP and TAZ in lung cancer cells, as well as the functional impact of such non-redundancy, remains to be comprehensively determined.
Here we sought to explore the division of labor between YAP and TAZ in non–small cell lung cancer (NSCLC), the most common histologic subtype of lung cancer (1). Employing representative NSCLC cell lines, we show that the two paralogs orchestrate nonidentical transcriptional programs, giving rise to distinct biological phenotypes. Specifically, YAP preferentially regulates cell division and cell-cycle progression, whereas TAZ preferentially regulates extracellular matrix (ECM), cell adhesion, and cell migration. These findings imply that YAP and TAZ have distinct, yet complementary, roles in lung cancer cells. Identification of differential activities and downstream effects of YAP and TAZ may enable better stratification of lung cancer tumors, and uncover cancer vulnerabilities that might be exploited therapeutically.
Materials and Methods
Cell lines and treatments
Cells were maintained at 37°C with 5% CO2. H1299 and H460 cells were cultured in RPMI1640 (Gibco) supplemented with 10% FBS (Biological Industries, BI) and 1% penicillin–streptomycin (BI). A549 cells were grown in DMEM (BI) supplemented with 10% heat-inactivated fetal bovine serum (Hyclone) and 1% penicillin–streptomycin (BI). mCherry-labeled H1299 cells (18) were a kind gift of Uri Alon (Weizmann Institute of Science). H460 and A549 cells were obtained from ATCC; H460 cells were labeled with mCherry (19). Cell lines were authenticated by STR profiling and tested negative for Mycoplasma, and did not exceed 12 in vitro passages. Taxol [paclitaxel; Sigma (T7402)] was used at a final concentration of 1 μmol/L for 42 hours.
Western blot analysis
Western blot analysis was performed as described (20), using the following antibodies: GAPDH (Millipore, Catalog No. MAB374, RRID:AB_2107445), Flag (Sigma-Aldrich, Catalog No. F3165, RRID:AB_259529), Vinculin (Sigma-Aldrich, Catalog No. V9131, RRID:AB_477629), YAP/TAZ (Cell Signaling Technology, Catalog No. 8418, RRID:AB_10950494), and YAP (Santa Cruz Biotechnology, Catalog No. sc-376830, RRID:AB_2750899). Conjugated anti-mouse or anti-rabbit secondary antibodies were from Jackson ImmunoResearch. Imaging and quantification were performed using ChemiDoc MP Imager with Image Lab 4.1 software (Bio-Rad).
siRNA and plasmid transfections
For siRNA-mediated knockdown, SMARTpools or single oligonucleotides (Dharmacon; Supplementary Table S1) were used with the Dharmafect#1 transfection reagent (Dharmacon), at a final concentration of 30 nmol/L. For siPLK1 knockdown, we used 10 nmol/L PLK1 siRNA and 20 nmol/L siControl. The medium containing oligonucleotides and reagents was replaced after 6 hours. Plasmid transfection was done using jetPEI DNA transfection reagent (Polyplus Transfection). The final DNA amount was 10 μg per 10 cm dish, and the transfection medium was replaced after five hours. pcDNA3-Flag-YAP, with residues 90, 91, 94, 95, 96 replaced with Alanine (YAPmTEAD) and pcDNA3-Flag-TAZ were a generous gift of Yosef Shaul (Weizmann Institute of Science). TAZ-GFP and TAZ S51A (TAZmTEAD; ref. 21) were a generous gift of Kunxin Luo (University of California, Berkeley, CA).
RNA extraction, reverse transcription, and RT-qPCR
Total RNA was isolated using a NucleoSpin RNA Kit (Macherey Nagel). Reverse transcription and qPCR were performed as described (22). Values were normalized to either HPRT or GAPDH. Primers are listed in Supplementary Table S1.
RNA sequencing
Cells were transfected with siControl, siYAP, or siTAZ SMARTpool oligonucleotides for 48 hours and serum-starved for the last 18 hours. RNA was extracted and the polyA fraction was purified from 500 ng of total RNA, followed by fragmentation and generation of double-stranded cDNA, Agencourt Ampure XP beads cleanup (Beckman Coulter), end repair, A base addition, adapter ligation and PCR amplification. Libraries were quantified by Qubit (Thermo Fisher Scientific) and TapeStation (Agilent) and sequenced on HiSeq 2500 (Illumina; single read sequencing).
Transcriptomic analysis
RNA-sequencing (RNA-seq) analysis was done using the UTAP transcriptome analysis pipeline (23). Reads were trimmed to remove adapters and low-quality bases using cutadapt [-a “A(10)” -a “T(10)” –times 2 -q 20 -m 25; ref. 24] and mapped to genome GRCh38 (Gencode, UCSC) using STAR v2.4.2a (–alignEndsType EndToEnd, –outFilterMismatchNoverLmax 0.05, –twopassMode Basic; ref. 25). Reads were counted using STAR, and genes having minimum five reads in at least one sample were considered. Counts normalization and differential expression detection were analyzed with DESeq2 (betaPrior, cooksCutoff and independent filtering parameters set to False; ref. 26). Differentially expressed genes were selected with |log2Fold change|≥1 and adjusted P-value ≤0.05 (Benjamini and Hochberg). For A549 RNA-seq, batch correction was done using sva (3.26.0) R package. Gene expression heatmaps were generated with Partek Genomics Suite 7.0 (Partek Inc.), using log normalized values (rld), with row standardization. Volcano plots were generated with Matlab. Gene Ontology (GO) enrichment analyses were performed with Metascape (27) and Gene Set Enrichment Analysis (GSEA; ref. 28) preranked tool. The data are accessible through GEO Series accession number GSE151201.
Cell-cycle profiling
Cells were grown in 6 cm plates and transfected with the indicated siRNAs or plasmids for 48 hours. Following serum starvation (0% serum) for 18 hours, cells were analyzed with Phase-Flow BrdU Cell Proliferation Kit (BioLegend). Briefly, cells were incubated with BrdU for 75 minutes and labeled with Alexa Fluor-647 conjugated anti-BrdU antibody. Total DNA was stained with DAPI. Then, 50,000 cells were collected and analyzed by multispectral imaging flow cytometry. The percentage of cells in each cell-cycle phase was manually determined on the basis of BrdU intensity and total DNA content, using FlowJo (Becton, Dickinson and Company).
Scratch assays
Cells were transfected with the indicated siRNAs or plasmids for a total of 48 hours, reaching 90% to 100% confluence. A scratch was introduced with a 200 μL pipette tip; detached cells were washed off, and medium was replaced with serum-free medium. Gap closure was imaged at 0 and 24 hours with a Nikon eclipse Ti-E microscope at ×4 magnification, capturing at least four fields for each condition. The migration distance was assessed manually using ImageJ (NIH). Gap closure was calculated as follows: (0 hour gap width – 24 hours gap width)/0 hour gap width.
Transwell migration assays
Cells were seeded on a transwell insert (8 μm pore size; Costar) with medium containing 1% serum. Medium containing 10% serum was added to the bottom well. After 16 hours, inserts were fixed with ice cold methanol (100%) for 5 minutes, and migrated cells were stained with crystal violet for 30 minutes; nonmigrated cells were removed with cotton plugs. Fixed and stained cells were imaged with a Nikon eclipse Ti-E microscope at ×10 magnification, capturing at least four fields for each condition, and crystal violet stained areas were quantified with an ImageJ macro. Coverage by migrating cells was calculated as percentage of stained area relative to total area.
Cell counting
Cells were seeded in a 96-well plate, transfected with the indicated siRNAs or plasmids in medium containing 10% serum for either 24, 48, or 72 hours, and counted with a LUNA automated cell counter (Logos Biosystems).
Cell death assays
Cell death was assessed using the CellTox Green Cytotoxicity Assay (Promega). Cells were seeded in a 96-well plate and transfected with indicated siRNAs or plasmids for a total 48 hours. Following incubation with either 1 μmol/L Taxol or DMSO for 42 hours, including serum starvation for the last 18 hours, CellTox reagent was added and the GFP fluorescence of each well was measured using an Infinite M200 microplate reader (Tecan). Three technical replicates were done for each condition. Cell death was calculated by subtracting the background signal and normalizing to DMSO-treated cells (control).
Statistical analysis
Independent biological replicates were performed and group comparisons were done as detailed in the figure legends. P-values below 0.05 were considered significant. Statistical significance between two experimental groups is indicated by asterisks.
Synthetic lethality
We employed a recently published synthetic lethality (SL) pipeline, ISLE (29) in a four-step process, to compute SL scores for all paralogous gene pairs (n = 6353, derived from ref. 30), from expression profiles of 433 lung adenocarcinoma (LUAD) human tumors from The Cancer Genome Atlas (TCGA; ref. 31) and 87 LUAD cancer cell lines (32). To compute the SL score for a given paralogous pair, we first investigated the gene essentiality upon inactivation of its partner in LUAD cell lines. By definition, it is expected that gene A will be essential only when its SL partner gene B is inactive in a given cancer cell line (29). Using a set of genome-wide shRNA/sgRNA screens (as in ref. 29), we tested whether knockdown/knockout of one paralog is significantly more lethal when the other paralog is lowly expressed (bottom third quantile) versus highly expressed (top third quantile) across LUAD cell lines, via Wilcoxon rank-sum test (FDR corrected P < 0.1). Underrepresented SL pairs were next identified by quantifying the significance of simultaneous inactivation of both paralogs (under-expressed, via a hypergeometric test, FDR corrected P < 0.1) in LUAD TCGA cohort. Then, we investigated the clinical relevance score, by performing a Cox multivariate regression testing whether the patients with an active SL interaction (both paralogs are inactive) have better survival than the rest of the patients, while controlling for cancer types, age, gender, race, tumor purity, genomic instability, and the effect of individual gene activation (FDR corrected P < 0.1). In the last step, we calculated phylogenetic similarity measure for each paralogous pair using a nonnegative matrix factorization, which measures the Euclidean distance while taking into account their phylogenetic profile across 86 species. For a final SL score, we averaged the scores from all steps, and scaled the scores from 0 to 1 (1 indicates high SL interaction). Paralogous pairs were ranked according to their final SL scores.
YAP-differential and TAZ-differential drug sensitivity analysis
To identify YAP or TAZ distinct associations with drug response, we utilized drug sensitivity profiles (PRISM) of 4,518 drugs tested across 47 LUAD cancer cell lines (33). We identified drugs that showed differential response between cell lines with high expression of YAP (top 33%) compared with those with low YAP expression (bottom 33%), and likewise for TAZ. P-value of the comparison was determined by Wilcoxon rank-sum test. Then, by using Drugbank (34) annotation, the targets for each of the YAP-differential or TAZ-differential drugs were defined. YAP-differential and TAZ-differential drugs, their targets and P-values are listed in Supplementary Table S2.
Taxol sensitivity analysis
PRISM dataset
By utilizing single-dose Taxol cell viability data across cancer cell lines (n = 578; ref. 33), we tested whether lines with high expression of YAP (top 33%, n = 192) are differentially more sensitive to Taxol compared with those with low YAP (bottom 33%, n = 192), and likewise for TAZ. P-value of the comparison was determined by Wilcoxon rank-sum test.
GDSC dataset
Taxol sensitivity data was obtained from ref. 35; sensitivity was measured by area under the dose–response curve in 136 lung cancer cell lines. The Spearman correlation coefficient of Taxol sensitivity with either YAP or TAZ log2-transformed protein levels (determined by RPPA in Cancer Cell Line Encyclopedia; ref. 36) was calculated across these cell lines.
TCGA YAP and TAZ correlation analysis
We queried the LUAD expression dataset of TCGA, comprising 20,167 expressed genes and 515 tumor samples (after cleaning and filtering the data). LUAD TCGA mRNA-seq RSEM normalized data were downloaded from http://gdac.broadinstitute.org/. Expression data (X) was transformed to X_transformed = log2(X + 1). The Pearson correlation coefficient with either YAP or TAZ mRNA levels was calculated for each gene. Genes were sorted according to absolute R, and the top 8% (1,600 genes) were selected for further analysis. Nonoverlapping genes between the YAP-correlated and TAZ-correlated lists (1,276 genes) were subjected for GO annotation analysis.
Chromatin immunoprecipitation sequencing
Cells were serum starved for 18 hours. Cross-linking with 1% formaldehyde and chromatin immunoprecipitation (ChIP) were performed, as in ref. 37, using either YAP1 (Thermo Fisher Scientific, Catalog No. PA1-46189, RRID:AB_2219137) or TAZ antibody (Proteintech, Catalog No. 23306-1-AP, RRID:AB_2721185), and Pierce ChIP-grade Protein G magnetic beads (Thermo Fisher Scientific). Reverse crosslinking and DNA purification were with ChIP DNA Clean & Concentrator Kit (Zymo Research). 4 to 20 ng ChIP DNA was processed as described (38). Libraries were prepared and multiplexed at the Crown Institute for Genomics (G-INCPM, Weizmann Institute of Science), using NextSeq 75 cycles high output kit (Illumina; single read sequencing). Reads were preprocessed with cutadapt to remove adapters and low-quality bases (parameters: –times 2 -q 20 -m 25). Reads were mapped to human genome (hg19) using bowtie2 (local mode). Broad peaks were called using MACS2, relative to background input samples that had been cross-linked and sonicated, but not immunoprecipitated (parameters: –bw 300 -B -f –SPMR -keep-dup auto -q 0.01). Peaks were annotated using Homer software. Peaks with at least 30% overlap between replicates were considered. For differential peaks analysis, DiffBind package was used (http://bioconductor.org/packages/release/bioc/html/DiffBind.html). Significant peaks were defined as FDR ≤ 0.05 and |log2Fold change|≥1. Peaks were visualized on genomic features with IGV (39). The data are accessible through GEO Series accession number GSE151201.
Results
YAP and TAZ are associated with distinct transcriptional programs
YAP and TAZ act primarily as transcriptional cofactors (6, 9, 10). To compare YAP and TAZ impact on the transcriptome of lung cancer-derived cells, we performed RNA-seq analysis following siRNA-mediated transient knockdown of either YAP (siYAP) or TAZ (siTAZ) in NSCLC-derived H1299 cell line (knockdown validation in Fig. 1A and Supplementary Fig. S2). As expected, a subset of 82 common genes were impacted similarly by partial depletion of either YAP or TAZ (Fig. 1B; Supplementary Fig. S3). However, expression of a larger number of genes was distinctly affected in a paralog-specific manner. Thus, 204 genes were significantly downregulated or upregulated at least two-fold upon YAP depletion, but were only mildly affected by TAZ depletion (Fig. 1B and C). Conversely, expression of 324 other genes was significantly affected by TAZ depletion, but less so by YAP depletion (Fig. 1B and D). For simplicity, the sets of genes preferentially regulated by YAP or TAZ will hereafter be referred to as YAP- or TAZ-regulated genes, respectively.
YAP preferentially regulates cell-cycle progression, whereas TAZ preferentially regulates cell migration
To identify biological processes regulated by either YAP or TAZ in H1299 cells, we performed GO analysis. Interestingly, genes preferentially regulated by YAP were strongly enriched for cell division- and mitosis-related-terms (Fig. 2A; Supplementary Fig. S4A; Supplementary Table S3), whereas the two top enriched terms for TAZ-regulated genes were related to ECM organization (Fig. 2B; Supplementary Fig. S4B; Supplementary Table S3). Likewise, GSEA on all informative genes, ranked by their expression fold change (relative to control) upon either siYAP or siTAZ, showed that the top downregulated genes upon YAP silencing were strongly enriched for cell-proliferation-related genes (Fig. 2C; ref. 40), whereas genes upregulated upon TAZ silencing were enriched for adhesion- and ECM-related genes (Fig. 2D; ref. 41). To assess the generality of our observations, we performed RNA-seq analysis also with NSCLC-derived A549 cells. Importantly, GSEA revealed again that YAP-regulated genes were enriched for cell-proliferation-related genes whereas TAZ-regulated genes were enriched for adhesion- and ECM-related terms (Fig. 2E and F).
Selected genes from each subset were also validated by RT-qPCR analysis in several NSCLC-derived cell lines. Although the extent of preferential regulation varied between cell lines, canonical cell-cycle regulators (PLK1, CIT, KIF23, CDC20) were indeed preferentially downregulated by siYAP (Fig. 2G), whereas genes affecting cell adhesion and migration [integrin beta8 (ITGB8), laminin subunit alpha-3 (LAMA3), Synaptopodin (SYNPO)] were preferentially upregulated by siTAZ (Fig. 2H).
As expected, expression of common YAP/TAZ-regulated genes was affected more strongly by simultaneous knockdown of YAP and TAZ, as compared with knockdown of each paralog alone (Supplementary Figs. S5A and S5B). In contrast, YAP-regulated genes were affected to a similar extent by YAP single knockdown and YAP/TAZ double knockdown (Supplementary Fig. S5C), and likewise for TAZ-regulated genes (Supplementary Fig. S5D), further confirming their paralog-specific regulation.
Next, to determine whether the differential transcriptomic effects of YAP and TAZ drive different functional outcomes, we examined the effects of depletion of each paralog on the proliferation and migration of NSCLC-derived cells. Notably, in all three cell lines tested, YAP silencing (knockdown validation in Supplementary Figs. S6A–S6C) attenuated markedly the increase in cell number over time (Fig. 3A). In contrast, TAZ knockdown had only a mild (H1299, A549) or marginal (H460) impact. Concordantly, transient overexpression of YAP, but not TAZ (overexpression validation in Supplementary Figs. S7A–S7C), augmented the proliferation of H1299 and A549 cells (Fig. 3B).
Depletion of either PLK1 or CIT, both preferentially regulated by YAP (Fig. 2G), attenuates cell-cycle progression in multiple cancer cell types (42–44), leading to depletion of the S-phase subpopulation and accumulation of cells in G1-phase. Indeed, as seen in Fig. 3C and D, silencing of YAP, but not TAZ, resulted in a prominent S-phase depletion in both H1299 and A549 cells (siYAP/Control ratio = 0.4 and 0.3, respectively) and elicited a mild increase in G1-phase, relative to control. Similarly, PLK1 silencing mimicked the cell-cycle effects of siYAP (Supplementary Figs. S8A and S8B). Conversely, overexpression of YAP, but not TAZ, increased the S-phase fraction (Fig. 3E). Thus, YAP preferentially acts as a positive regulator of cell-cycle progression in these lung cancer cells.
Excessive ECM production can impede cell migration, owing to aberrant cell adhesion (45). Hence, we assessed the impact of YAP and TAZ on cell migration. Indeed, in a gap closure (“scratch”) assay, TAZ knockdown, which augmented the expression of adhesion- and ECM-related genes (Fig. 2D, F, and H), strongly attenuated cell migration in H1299, A549, and H460 cells (Fig. 4A and B; Supplementary Fig. 9A), whereas silencing of YAP had almost no effect. Concordantly, transient overexpression of TAZ, but not YAP, led to a mild increase in cell migration in both H1299 and A549 cells (Fig. 4C; Supplementary Fig. S9B). These observations were further validated in a transwell migration assay (Supplementary Figs. S9C and S9D). Silencing of YAP or TAZ by single siRNA oligonucleotides phenocopied the effects of the corresponding siRNA SMARTpools (Supplementary Figs. S10A–S10D). Altogether, our observations imply that YAP and TAZ have nonoverlapping roles in these lung cancer cells, whereas YAP preferentially promotes cell-cycle progression, TAZ preferentially promotes cell migration.
Partial nonredundancy of YAP and TAZ and association with distinct phenotypes in lung tumors
SL is a broadly accepted indication of functional backup and redundancy between genes (46). SL implies a genetic interaction between two genes, whereby individual inactivation of either gene retains cell fitness, while their combined inactivation reduces fitness (and in extreme cases, is lethal; ref. 29). Redundant paralogous pairs often display strong SL, such that cells expressing intrinsically low levels of one paralog are highly dependent on retention of the other paralog (30). Hence, the extent of SL informs on the extent of functional redundancy of a paralogous pair. The partially nonredundant transcriptional and functional effects of YAP and TAZ in lung cancer cells suggested that these paralogs might display only partial SL. We therefore utilized a SL identification computational pipeline (29) to compute SL scores of human paralogs, as previously defined (n = 6,353; ref. 30), in lung adenocarcinoma (LUAD) cells and tumors. Indeed, we found that YAP-TAZ have a fairly low degree of SL when compared with the ranking of all paralogous pairs (Fig. 5A; SL score = 0.37, Rank = 5643). In contrast, ARID1A and ARID1B, known to be functionally redundant (47), were ranked much higher (SL score = 0.95, Rank = 8, Fig. 5A, “+”). As negative controls, we paired ARID1A-ARID1B with either YAP or TAZ (YAP-ARID1A, YAP-ARID1B, TAZ-ARID1A, TAZ-ARID1B); as expected, these pairings yielded very low SL scores (mean SL score = 0.12, Fig. 5A, “−”). Thus, the SL analysis further confirms that YAP and TAZ are only partially redundant in lung cancer cells and tumors.
Next, we asked whether expression levels of YAP and TAZ in lung tumors are associated with cell-cycle- or ECM-related terms. Toward this aim, we queried the LUAD expression dataset of TCGA. Utilizing an unbiased correlation-based method, the correlation coefficient with either YAP or TAZ mRNA was calculated for each gene in the dataset. Interestingly, YAP and TAZ mRNA levels are only partially correlated (Supplementary Fig. S11). The top 1,600 genes displaying the highest correlation coefficients, regardless of directionality, were further analyzed. Remarkably, only 324 of the 1,600 genes overlapped between YAP-correlated and TAZ-correlated gene sets. Importantly, GO analysis of the nonoverlapping genes (n = 1,276) showed that YAP-correlated genes were enriched in cell division- and cell-cycle-related terms (Fig. 5B; Supplementary Table S4), whereas TAZ-correlated genes were enriched in morphogenesis-, development-, and adhesion-related terms (Fig. 5C; Supplementary Table S4); of note, morphogenesis is closely linked with adhesion (48). Hence, our in vitro observations, implying distinct roles for YAP and TAZ in lung cancer cell biology, also appear to hold for actual human tumors.
Differential association of YAP versus TAZ with response to anticancer drugs
Direct pharmacologic inhibition of YAP/TAZ activity still remains a clinical challenge (49). The distinct functions of YAP and TAZ suggest that the two paralogs may affect differently the response to particular anticancer therapies. Indeed, using drug sensitivity data across 47 LUAD cell lines (PRISM dataset; ref. 33), we identified 142 drugs whose response profile differs preferentially between high versus low YAP mRNA levels, and 38 drugs whose response profile differs preferentially between high versus low TAZ mRNA (Supplementary Table S2).
Taxol (paclitaxel), a microtubule poison, is used to treat NSCLC (50). Interestingly, inspection of a drug sensitivity dataset (33) revealed that cell viability upon Taxol treatment was negatively correlated with levels of YAP, but not TAZ mRNA (Fig. 6A). Further analysis of the largest publicly available cancer cell line drug sensitivity collection (GDSC dataset; ref. 35) confirmed that Taxol sensitivity was more significantly correlated with YAP protein levels (R = 0.22, P-value <0.02; Supplementary Fig. S12) than with TAZ protein levels (R = 0.10, P-value = 0.23) in lung cancer cell lines.
To compare directly the impact of YAP versus TAZ on Taxol sensitivity, we treated H1299 cells with Taxol after knockdown of either YAP or TAZ. As seen in Fig. 6B, YAP silencing indeed greatly compromised the cytotoxicity of Taxol, whereas TAZ silencing had only a marginal effect. Conversely, transient overexpression of YAP, but not TAZ, increased Taxol sensitivity (Fig. 6C). Thus, YAP and TAZ also affect differentially the sensitivity to specific anticancer agents.
Involvement of TEAD in expression of YAP- and TAZ-regulated genes
YAP and TAZ do not bind directly to DNA, and much of their transcriptional activity is dependent on “piggybacking” to TEA domain proteins 1–4 (TEAD1–4) DNA binding proteins (10). To assess whether YAP/TAZ distinct transcriptional effects depend on TEAD, we overexpressed mutant forms of either YAP or TAZ defective in TEAD binding, and monitored their impact on relevant genes (overexpression validation in Supplementary Fig. S13A). As seen in Supplementary Figs. S13B and S13C, loss of TEAD binding attenuated the induction of common YAP/TAZ-regulated genes and of YAP-regulated genes. Interestingly, while the upregulation of CLDN5 by TAZ was compromised by loss of TEAD binding, this was not so for three TAZ-repressed genes. Hence, the role of TEAD in the differential transcriptional effects of YAP and TAZ is gene and mechanism dependent.
TAZ binds preferentially to chromatin regions encompassing TAZ-regulated migration-related genes
To further explore the mechanisms underlying YAP/TAZ differential transcriptional programs, we performed ChIP sequencing (ChIP-seq) analysis of H1299 cells with antibodies against either YAP or TAZ. As expected, there was approximately 50% overlap between YAP and TAZ binding sites (Supplementary Fig. S14A). Furthermore, consistent with previously published data (6), most peaks were within 10 to 100 kb from the nearest transcription start site (TSS; Supplementary Fig. S14B).
We then explored the peaks (approximately 1,200) displaying a significant differential binding of YAP or TAZ (|fold change| ≥ 2, FDR ≤ 0.05; Fig. 7A), hereafter referred to as YAP-differential or TAZ-differential peaks. Remarkably, whereas 61% of TAZ-differential peaks were located less than 10 kb from the nearest TSS (yellow and green in Fig. 7B), the majority of YAP-differential peaks were located over 100 kb away from any annotated gene (red in Fig. 7B), suggesting that differential YAP/TAZ binding sites might control gene regulation via distinct regulatory mechanisms.
We next assigned TAZ-differential peaks to the nearest gene, considering only peaks residing within 10 kb from the nearest TSS [owing to the negligible number of YAP-differential peaks meeting this condition (yellow and green in Fig. 7B), these were not further analyzed]. As expected, genes assigned to TAZ-differential peaks were differentially expressed upon siTAZ, but not siYAP, compared with control (Fig. 7C). Interestingly, several of these genes are known to regulate cancer cell migration (51, 52). For instance, CLDN5 promotes cell migration, whereas SYNPO inhibits migration (51, 52). As shown in Fig. 7D, TAZ bound more than YAP to regulatory regions of CLDN5 and SYNPO. Hence, the preferential effects of TAZ on migration-related genes may be partly due to preferential binding to their regulatory regions.
Discussion
In this study, we compared the functional impact of YAP versus TAZ in NSCLC-derived cells. We found that although both paralogs modulated to a similar extent a common subset of genes, distinctly larger subsets of genes were preferentially affected in a paralog-specific manner. A pronounced difference in their biological impact was revealed: whereas YAP preferentially affected cell-cycle progression, TAZ preferentially modulated cell migration. This finding was apparent not only in cell lines, but also in data from human lung tumors. Thus, there exists a division of labor between the two paralogs in lung cancer cells and tumors. Moreover, this division of labor is associated with differential responses to anticancer drugs, such as Taxol.
Of note, many YAP-regulated genes were also mildly affected by TAZ depletion, and vice versa (Fig. 1C and D), suggesting that the relative contribution of YAP versus TAZ to the expression of such genes may depend on the relative abundance of each paralog in a given cell. A similar idea was recently proposed for protein–protein interactions: two paralogs may share potential binding partners, but the actual interaction profile of each paralog will be determined by its relative abundance and relative affinities for the different interaction partners (53). Indeed, TAZ silencing caused a mild decrease in cell proliferation in some but not all cell lines (Fig. 3A): likewise, YAP slightly affected cell migration in some experiments (Supplementary Figs. S9A, S9C, and S9D), in agreement with earlier reports (17, 54, 55). Therefore, the division of labor between YAP and TAZ is preferential rather than absolute, and may differ between different types of cancer and perhaps even between individual cases of the same cancer type. Interestingly, Sun and colleagues (56) have recently shown that although YAP and TAZ similarly promote the proliferation of nondifferentiated myoblasts, they exert opposing effects when such cells are induced to undergo myogenic differentiation. Furthermore, Plouffe and colleagues (57) found that inactivation of YAP in human embryonic kidney cells had greater effects than TAZ inactivation, supporting the notion that one paralog may dominate over the other, in a context-dependent manner, in regulating specific biological functions.
Many gene duplication events play major roles in shaping the genomes of different species. Paralogs originating from a gene duplication event can diverge and acquire changes that contribute to organismal diversity. Invertebrates have only a single YAP/TAZ orthologue; in Drosophila this orthologue, Yorkie (Yki), is required for both cell proliferation and cell migration in the course of multiple biological processes (58–62). In vertebrates, YAP and TAZ regulate both proliferation and migration, but with different efficiencies, enabling refined regulation of complex biological processes. We propose that the gene duplication and the consequent division of labor between the two paralogs might explain their retention throughout vertebrate evolution.
Our analysis demonstrates that YAP and TAZ display both shared and differential genome-wide chromatin binding patterns. In the case of TAZ, this may explain in part its ability to preferentially regulate a subset of migration-related genes. However, a comparable mechanistic explanation for YAP-regulated genes is still lacking. Thus, more work is needed to fully explain the molecular basis for the differential transcriptional effects of the two paralogs. Plausibly, YAP and TAZ may vary in their relative affinities for particular interacting proteins, which may lead to different transcriptional preferences. This might be subject to further regulation, for example, by posttranslational modifications, providing an additional layer of context-dependent diversity. Interestingly, it was recently demonstrated (21) that TAZ, but not YAP, readily forms nuclear condensates via liquid–liquid phase separation (LLPS), whereas YAP was shown to form LLPS condensates under other conditions (63). Notably, we revealed hundreds of genomic sites located far away from any annotated gene, which were bound selectively by YAP and not TAZ. This raises the intriguing possibility that YAP may have an additional function, not shared with TAZ, perhaps associated with long-range chromatin interactions or high-order genome organization.
In sum, our findings suggest that tumorigenesis may take advantage of the division of labor between YAP and TAZ to orchestrate complementary oncogenic functions. These findings may enable a better stratification of lung cancer tumors, and perhaps reveal cancer vulnerabilities that might be exploited toward more effective therapies.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
M. Shreberk-Shaked: Conceptualization, resources, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. B. Dassa: Data curation, formal analysis, investigation, assisted in analyzing RNA-seq and ChIP-seq data. S. Sinha: Formal analysis, investigation, synthetic lethality and differential drug analyses and scientific contributions regarding synthetic lethality. S. Di Agostino: Performing experiments. I. Azuri: Formal analysis, assisted in analyzing TCGA data. S. Mukherjee: Performing experiments. Y. Aylon: Scientific contributions regarding Hippo pathway biology. G. Blandino: Scientific contributions regarding Hippo pathway biology. E. Ruppin: Synthetic lethality and differential drug analyses and scientific contributions regarding synthetic lethality. M. Oren: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing.
Acknowledgments
We thank all members of the Oren lab for fruitful discussions, Yosef Shaul and Kunxin Luo for their generous gift of plasmids, Yarden Nuriel for technical assistance, Itay Tirosh and Benny Geiger for scientific advice, Uri Alon and Tamar Danon for access to their research facilities, and scientists of the Weizmann Institute Department of Life Sciences core facilities for help and guidance. This work was supported in part by grants from the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation and the DKFZ-MOST Cooperation in Cancer Research, a grant from Anat and Amnon Shashua, and the Moross Integrated Cancer Center. M. Oren is incumbent of the Andre Lwoff chair in molecular biology.
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.