Molecular profiling and computational network analysis of TAZ-mediated mammary tumorigenesis identifies actionable therapeutic targets.

Triple-negative breast cancer (TNBC) accounts for approximately 15–20% of all breast cancer (BC) cases and contributes disproportionately to BC mortality. TAZ, a key transducer of the Hippo pathway, has recently been demonstrated to confer breast cancer stem cell (CSC) traits. However, TAZ target genes and the underlying transcriptional regulatory pathways responsible for the CSC phenomenon remain unknown. Here, we demonstrate that the oncogenic activity of TAZ is essential for propagation of the malignant phenotype. We further show that constitutively active TAZ tumor-derived cells exhibit unique tumor-initiating properties, including increased self-renewal and metastatic seeding potential, acquired chemotherapy resistance and the ability to efficiently regenerate tumor formation in vivo. Combined digital RNA expression analysis and computational network approaches identify several signaling pathways that distinguish breast cancer tumor-initiating cells (T-ICs) from bulk tumor cells. We demonstrate the utility of this approach by repositioning the small molecule tyrosine kinase inhibitor, Dasatinib, which selectively targets T-ICs and inhibits TNBC growth in vivo.


INTRODUCTION
The Hippo pathway is an evolutionary conserved regulator of tissue growth and cell fate during mammalian development and regeneration, while deregulation of the Hippo pathway caused by gene mutation or anomalous protein expression has been linked to human diseases including cancer [1,2]. Currently, over 20 regulators have been identified that intersect with core Hippo pathway components and potentially contribute to carcinogenesis. For example, loss of Neurofibromin2 (NF2) tumor suppressor function has been associated with the development of colorectal, hepatocellular and thyroid carcinomas, as well as in melanoma [3]. Central to the Hippo pathway are two key downstream effector proteins, YAP (Yes-associated protein) and TAZ (transcriptional co-activator with PDZ-binding motif) that are tightly regulated by a number of upstream signaling molecules, such as Mst1/2, Lats1/2 and RASSF family proteins [4,5]. Inactivation of YAP/TAZ by the Hippo pathway is mediated primarily via cytoplasmic sequestration from 14-3-3 binding and protein degradation, respectively [6][7][8][9]. In addition, Hippo pathway-independent restriction of YAP/TAZ mediated by angiomotin, scribble, PTPN14, α-catenin and other junction protein sequestration mechanisms has been reported [10][11][12][13][14][15].
TAZ does not contain an intrinsic DNA-binding domain, but instead actively recruited to target genes by interacting with multiple transcription factors and mediates diverse transcriptional programs in a contextdependent manner [16]. Interestingly, loss of the TAZ protein resulted in uncontrolled differentiation of www.impactjournals.com/oncotarget human embryonic stem cells (hESCs) as well as loss of self-renewal of hESCs [17]. Furthermore, TAZ was recently shown to sustain self-renewal potential and tumorinitiation capacities of breast CSCs [11]; however, TAZ target genes and the underlying transcriptional regulatory pathways responsible for the CSC phenomenon remain poorly characterized. Nonetheless, therapeutic modulation of TAZ could improve current cancer treatment strategies.
Unfortunately, direct pharmacological inhibition of TAZ is challenging because it has no known catalytic activity [18,19]. Accordingly, in the current study we employ a strategy that exploits the functional interconnectivity of intracellular signaling networks to unambiguously identify disease-specific 'druggable' targets, located downstream of TAZ. Moreover, we clearly demonstrate that expression of constitutively active TAZ experimentally confers BC T-IC properties and metastatic colonization capacity to non-transformed human basal-like mammary epithelial cells. Notably, we present a novel differential network-based framework to detect biologically meaningful cancer-related genes and subsequently prioritize/rank genes as potential drug targets. Finally, we demonstrate that Dasatinib, a Src family kinase and receptor tyrosine kinase (RTK) inhibitor, selectively targets TAZ-induced T-ICs and provides novel insight to how modulation of TAZ-driven pro-tumorigenic transcriptional programs could help guide future BC treatment strategies.

In vivo-derived TAZ-induced tumor initiating cells (T-ICs)
TAZ has been previously indicated to serve as a key breast CSC determinant [11]. Accordingly, to further characterize TAZ-induced CSC-like traits during mammary tumorigenesis, we orthotopically injected TAZ-4SA (constitutively activated form) transduced MCF10A cells into the mammary fat pad of SCID mice and generated primary tumors. We explanted the TAZ-induced mammary tumors and derived cell lines (herein denoted TAZ-M#1-6) that could be stably propagated in tissue culture (Fig. 1A). We confirmed ectopic TAZ expression in all tumor-derived cells as well as in the parental cell line by immunoblot (Fig. 1B). We previously showed that overexpression of constitutively active TAZ promoted cell migration [20]. To independently test whether tumor-derived cell lines maintained such capacity, we performed transwell cell migration assays and found that they all migrated aggressively (Fig. 1C).
A subpopulation CD44 high /CD24 low of breast cancer cells has been reported to have stem/progenitor cell properties [21]. To estimate the proportion of CD44 high / CD24 low cells, we characterized tumor-derived cell lines by flow cytometry for surface expression of CD44 and CD24, respectively. As shown in Figure 1D, this population was indeed enriched in the majority of tumor-derived cell lines tested but phenotypically reverted to the CD44 high / CD24 high population after extended culture (Supplemental Fig. 1A). Interestingly, the CD44 high /CD24 low sub population displayed a lower in vitro proliferation rate compared to parental cells and resistance to chemotherapy (Fig. 1E, F). However, they exhibited selective sensitivity to Salinomycin [22] and Bortezomib [23] (Supplemental Fig. 1B, C), consistent with the recently reported proteasome addiction of basal-like TNBC cells [23].
Next, to determine whether the tumor-derived cells contained self-renewal capacity, we used an in vitro mammosphere formation assay and found that majority of the cell lines generated mammospheres of increased size and number ( Fig. 2A). The acquisition of CSC traits has been previously associated with the epithelial to mesenchymal transition (EMT) [24]. Therefore, we further analyzed the alterations of EMT markers in the tumor derived cells. The alterations of EMT-associated markers, such as: loss of epithelial genes CDH1, CDH3 and gain of mesenchymal genes CDH2 and FN1, as well as the adoption of a migratory mesenchymal phenotype were maintained in all the tumor-derived cell lines ( Fig. 1C and 2B). Interestingly, the TAZ-M#5 cell line underwent EMT but lost the stem-like features (Fig. 1C&D, 2 A&B) and tumor initiation potential (data not shown).
To explore whether tumor-derived CD44 high / CD24 low cells acquired tumor-initiating capacity and to accurately enumerate their frequency, we used in vivo clonal tumor initiating assays to functionally identify CSCs rather than on the basis of immunophenotype or in vitro mammosphere assay. Consequently, we estimated a > 1000-fold increase in the frequency of self-renewing T-ICs in the TAZ-M#1 cell line relative to the parental TAZ-4SA-expressing MCF10A cell line ( Fig. 2C and data not shown). The likelihood of nodal metastases is increased in BC patients whose tumors have breast CSCs [21]. Furthermore, emerging evidence indicates that breast CSCs and EMT co-operate to generate circulating tumor cells (CTCs) that are highly competent for metastasis [25]. To further explore this possibility and specifically determine whether the tumor-derived cells promote metastasis, both mammary fat pad and tail-vein injections were performed in NSG mice. In contrast to control mice, both metastasis to the lungs and spontaneous metastasis were observed in NSG mice injected with tumor-derived TAZ-induced cells (TAZ-M#1), strongly suggesting that TAZ increased metastatic potential and tumor-seeding ability (Fig. 2D).

Network-based comparison of TAZ-mediated gene expression patterns
To examine the TAZ-mediated pro-tumorigenic transcriptional program, we performed RNA-seq www.impactjournals.com/oncotarget    Table 1). www.impactjournals.com/oncotarget As shown in Figure 3B, five data subtypes were identified by consensus NMF clustering using an abundance matrix containing differentially changed genes (> 2.0 fold change, with p < 0.05 in all expression profiles).
Using DAVID (database for annotation, visualization and discovery), we assessed whether any GO biological or molecular processes were statistically over-or underrepresented. Over-represented pathways (FDR < 0.001) included EMT related pathways, along with integrin signaling and focal adhesion, chromatin remodeling, PIK3 signaling, growth hormone and nuclear receptor co-activators, WNT/cadherin signaling and ATM/Rb related pathways (Supplemental Table 2-4). In addition, Gene Set Enrichment Analysis (GSEA) of differentially regulated transcripts in the TAZ-M#1 gene cluster/signature revealed an overrepresentation of genes associated with several published invasive, metastatic BC patient and breast CSC transcriptome studies (Supplemental Table 5).
To expand on these analyses, we constructed a functional interaction (FI) network and used network community analysis to identify sub-networks that contain modules involved in annotated biological processes and canonical pathways [26]. Specifically, two interaction networks were independently generated from the TAZ-M#1 and TAZ-M#5 tumor data sets. Network subtraction was subsequently used to obtain a differentialnetwork model and to identify both overlapping and unique clusters ( Fig. 3C and Supplemental Table 6). Network comparison reflected differential functional interactions unique to TAZ-M#1, a breast tumor that   had acquired stem-like traits. PDGFR-β, WNT and NF-kappa B signaling pathways were highly deregulated in TAZ-M#1 (FDR < 0.0001), and may represent key mediators and/or regulators of biological properties associated with CSCs [27].
Next, to reduce the list of candidate genes to those that are most likely associated with signaling pathways involved in self-renewal and activated in CSCs, as opposed to general cell proliferation and pro-survival mechanisms, we identified all network bottleneck proteins. These proteins are known to control the flow of biological information within a network and maintain network integrity. Briefly, we define network bottlenecks as nodes (proteins) with the highest betweenness centrality scores and estimated by calculating the number of shortest paths that cross a given node. In addition, highly connected nodes (hubs) are also known to play an integral role in the propagation of signals across a network and were also calculated ( Fig. 3D & Supplemental Methods). As summarized in Fig. 3D, top-ranked bottleneck genes are highly deregulated in BC patients and several genes, such as PDGFR-β, FYN and NF-kappaB components.

Dasatinib targets the TAZ-induced T-ICs and TNBC Cells
Data presented thus far identify TAZ as a potential new drug target to treat aggressive types of BC. Unfortunately, no pharmacological agents directly targeting active TAZ have been reported. Therefore, we rationalized that modulating 'druggable' targets downstream of TAZ, rather than the oncogene itself, could be therapeutically exploited. Therefore, we used a subtractive integrated network approach and generated a TAZ-specific bipartite graph in which every link connected all FDA-approved drugs to a protein, if the protein is a known target of the drug (i.e. drug-target network, DTN) (Supplemental Table 7). As summarized in Figure 4A, one FI sub-cluster of particular interest was highly enriched with the PDGFR-β signaling pathway components. Furthermore, PDGFR-β, FYN and Src genes included in this module were independently identified as key bottleneck network genes (Fig. 3D).
Dasatinib is a potent, multi-target kinase inhibitor predicted from our DTN analyses (Supplemental Table 7). It was initially developed as an inhibitor of the Src family kinases (SFKs, such as Fyn, Yes, Src and Lyk) but is now known to also inhibit BCR-ABL, EphA2, platelet derived growth factor receptor (PDGFR) and c-KIT [28]. Accordingly, we rationalized that Dasatinib would represent an apposite and novel multi-target repositioning agent to pharmacologically modulate TAZ-mediated pro-tumorigenic transcriptional programs. To our great interest, we found that Dasatinib inhibited the anchorageindependent growth of TAZ-M#1 cells in soft-agar assay and reduced self-renewal as measured by mammosphere formation (Fig. 4B, C). Remarkably, FACS analysis of Dasatinib-treated TAZ-M#1 cells had an almost completely depleted CD44 high /CD24 low subpopulation after drug treatment for 24 hours and recovery of 72 hours; and only the CD44 high /CD24 high BC subpopulation remained viable (Fig. 4D). Collectively, these results indicate that Dasatinib selectively killed the TAZ driven BC T-ICs.
TNBC frequently has enriched breast CSCs that are thought to drive tumorigenesis and contribute to maligancy, therapeutic resistance and clinical relapse [29]. To further explore whether Dasatinib specifically targets TNBC cells, we used a panel of human TNBC and luminal cell lines to further assess differential sensitivity and response to Dasatinib. First, we examined the halfmaximal inhibitory concentration (IC50) values for Dasatinib and found that TNBC cell lines were more sensitive to Dasatinib than luminal cell lines (Fig. 4E). In addition, we confirmed the effect of Dasatinib on TNBC cell line proliferation in vitro by colony formation assay (Fig. 4F). Finally, to evaluate the effect of Dasatinib on mammary tumor-forming potential in vivo, we injected either vehicle control or 24-hour low-dose (300 nM) Dasatinib treated MDA-MB-231 cells into the mammary fat pad of SCID mice and clearly demonstrated that Dasatinib decreased tumor formation capacity (Fig. 4G).

DISCUSSION
The CSC hypothesis is based on the observation that many cancers are driven by a subpopulation of tumor-initiating cells (popularly known as CSCs) and promote tumor growth. In addition, CSCs are resistant to chemotherapy and radiation treatments, which potentially explain the limitations of these agents in curing human malignancies [30][31][32]. Despite considerable research efforts in recent years, it remains unclear what fraction of cancers follow the CSC model as well as the clinical behaviors explained by the model. Nonetheless, delineation of essential pathways that can distinguish T-ICs from their normal counterparts could provide novel opportunities for therapeutic intervention.
Multi-omics approaches allow the characterization of biological systems in various disease states to be measured from an increasing number of biochemical and molecular perspectives. Unfortunately, conventional functional enrichment analyses are limited in their efficacy because they do not incorporate known interdependencies among genes within a pathway. Furthermore, they treat all gene alterations as equal, which is not expected to be valid for many biological systems. Conversely, network-based pathway analysis identifies markers not as individual entities but as sub-networks and the resulting sub-networks provides an effective way to visualize genes and their interactions/relationships [33]. Here, employing www.impactjournals.com/oncotarget digital RNA sequencing and computational approaches, we present a differential network-based framework to detect biologically meaningful cancer-related genes. Notably, we identify several developmental pathways currently under investigation as potential therapeutic targets in CSCs, such as the Hedgehog, WNT, NOTCH and Nodal/Activin pathways [34]. However, at present, it is unclear of which these pathways are direct or indirect targets of a TAZ-mediated pro-tumorigenic transcriptional program. Nevertheless, our network analyses identified a novel PDGFR-β centered FI subcluster. The role of the PDGFR-β in different tumor cell types and in metastasis remains poorly characterized. For example, PDGFR-β signaling contributes to diverse tumorassociated processes, such as autocrine growth simulation of malignant cells, paracrine-stimulation of blood vessel formation and stimulation of fibroblasts and pericytes in the tumor stroma [35]. It was recently reported that mutant p53-induced up-regulation of PDGFRβ drives pancreatic cancer invasion and metastasis [36]. Of the intra-cellular pathways that operate downstream of PDGFR-β, the activation of the Ras and ERK MAP kinase pathway is important for cell proliferation. Actin reorganization and cell migration is dependent on PI3-kinase mediated activation of Rac, whereas PI3-kinase activation of AKT is important for cell survival. Several of these pathways are up-regulated in TAZ-expressing tumors as determined by high-density antibody microarrays (data not shown). However, direct inhibition of PDGFR-β alone through either RNAi or PDGFR-specific inhibitors, had no effect on BC cell viability in vitro or TAZ-mediated tumorigenicity in vivo (data not shown).
Recent advances in computational biology have identified a network structure and phenotypic robustness that strongly suggests the higher clinical efficacy of multi target drugs as compared with individual drug targets. Adopting this basic tenet, we rationalized that combinatorial gene inhibition by the tyrosine kinase inhibitor such as Dasatinib could modulate TAZ-mediated tumorigenesis. Correspondingly, we clearly show that Dasatinib selectively targets the CD44 high /CD24 low subpopulation in vitro and inhibits TNBC tumor growth in vivo. In a recent study, PDGFR-β was shown to lie downstream of FOXC2 in BC cells induced to undergo EMT. Furthermore, the PDGFR-β inhibitor sunitinib reduced tumor growth and metastasis of FOXC2expressing tumor cells [37], but given our findings it is most likely the effects mediated by sunitinib in these studies are a consequence of complex RTK inhibition and not direct PDGFR-β inhibition.
Current failure in TNBC treatment usually is not due to a lack of primary response or initial induction of remission but instead relapse after chemotherapy, in which CSCs are believed to play an essential role. If the CSC hypothesis is correct, then the most appropriate cells for investigation of genes involved in TNBC may be CSCs.
Here, we demonstrate how the multi-target inhibitor, Dasatinib, selectively targets CSCs and inhibits basallike and TNBC tumor growth. Similar to other targeted cancer therapies it will be important to determine which BC patients are most likely to benefit from this treatment. Furthermore, future studies are required in defining optimal treatment strategies for aggressive TNBC, both in the advanced as well as adjuvant setting.

Plasmid construction
The human TAZ-4SA expression constructs were described previously [20].

Flow cytometry analysis
Cells were passed through a 35-μm filter, pelleted, washed in 1X phosphate buffered saline (PBS) + 0.5% fetal calf serum (FBS) and counted. One million cells were suspended in 1X PBS + 0.5% FBS and stained with anti-CD44-APC conjugate and anti-CD24-PE conjugate (BD Biosciences) for 30 min at 4°C. Cells were washed 3 times and then analyzed by flow cytometry.

Mammosphere formation assay
Mammosphere formation assay was performed by plating 5X10 4 cells in serum-free DMEM/F12 1:1 media (Gibco) supplemented with EGF (20 ng/mL) and B27 (2%) into ultra-low attachment 6-well plates (Corning). Mammospheres were allowed to grow or in presence of drugs for 7 days. Total mammospheres greater than 100 μm in diameter were counted. Each experimental group was done in triplicate and the same experiments were repeated at least three times.

Cell proliferation (MTT) assay
Two thousand to four thousand cells were plated into 96 wells plate. Plates were harvested daily. Twenty microliters of 5mg/ml MTT (3-(4, 5-Dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide, a tetrazole) was added to each well, incubated for 3.5 hours in 37°C, and carefully removed without disturbing the cells. One hundred and fifty microliters of MTT solvent (4 mM HCl and 0.1% Nondet P-40 (NP40) in isopropanol) was added into each well and plate shaken at room temperature for 15 min. Absorbance was read at 590 nm with a reference filter of 620nm.

Anchorage-independent growth
Colony formation in soft agar assay was performed as described previously [10]. Briefly, cells were suspended in a growth medium mixed with 0.4% agar and seeded into 6-well plate containing a base of 0.5% agar, at a concentration of 5X10 4 cells/well. Cells were incubated and colonies were read after approximately 2 weeks.

In vivo tumor growth and metastasis assays
For TAZ tumor formation, 1 ´ 10 6 TAZ-4SA transduced MCF10A were injected into the mammary fat pad of female NOD/SCID mice of 6-8 weeks old. For serial dilution experiments, vector control or TAZ-4SA transduced MCF10A cells, TAZ-M#1 or TAZ-M#5 cells in exponential growth phase were harvested and suspended in PBS (50% matrigel), and 1 × 10 6 (in 0.1 mL) vector or 1 × 10 6 , 1 × 10 5 , 1 × 10 4 , 1 × 10 3 TAZ-4SA transduced MCF10A, TAZ-M#1 and TAZ-M#5 were injected into the mammary fat pad of female NOD/SCID mice of 6-8 weeks old. The SCID mice were generated at the Roswell Park Cancer Institute. Tumor sizes were measured twice a week using calipers. Mammary tumor formations were also detected by the In Vivo Luminescence Imaging System. For tail vein injection, 1 × 10 6 (in 0.1 mL) TAZ-M#1 cells were injected into the tail vein of female NSG mice of 6-8 weeks old. Lung tumor formations were detected by the In Vivo Luminescence Imaging System. The care and use of animals was approved by the Institutional Animal Care and Use Committee of the Roswell Park Cancer Institute (Buffalo, NY).

Cultivation of explanted tumor cells
An explant method was used for culturing mouse tumor cells adopted from previous description [39]. Briefly, the primary mouse mammary tumors were excised, washed with growth medium (MCF10A growth medium) and then minced with scalpels to obtain pieces approximately 2 mm in diameter. Twelve to fifteen of these pieces were then evenly dispersed over the bottom of each 6-cm dish (Falcon Plastics) and placed at room temperature for approximately 15 minutes to allow attachment. The tissue was then covered with 5 ml MCF10A growth medium. Cultures were incubated at 37ºC and observed daily for evidence of growth. During the first week, the culture medium was changed only if a large amount of cellular debris was in the flasks or if the pH became noticeably alkaline or acidic. Care was taken to avoid dislodging the tissue explants. After the first week or after significant growth, the medium was changed at least twice a week. When cultures were confluent, cells were conservatively sub-cultured.

Integrative informatics and pathway analysis
For network and pathway enrichment analysis, we used the Reactome FI and Cytoscape plugin. Significantly enriched pathway gene sets were exported and analyzed using Enrichment Map to determine relationships between pathways [40]. To investigate functional network and gene ontology relationships, candidate genes were first categorized using the GO analysis and were performed with the DAVID software [41]. The reference population was defined by our gene study set (see above) and an adjusted P value of 0.05 (Benjamini and Hochberg correction was used as a threshold for the identification of significant GO terms.