Loss of Merlin induces metabolomic adaptation that engages dependence on Hedgehog signaling

The tumor suppressor protein Merlin is proteasomally degraded in breast cancer. We undertook an untargeted metabolomics approach to discern the global metabolomics profile impacted by Merlin in breast cancer cells. We discerned specific changes in glutathione metabolites that uncovered novel facets of Merlin in impacting the cancer cell metabolome. Concordantly, Merlin loss increased oxidative stress causing aberrant activation of Hedgehog signaling. Abrogation of GLI-mediated transcription activity compromised the aggressive phenotype of Merlin-deficient cells indicating a clear dependence of cells on Hedgehog signaling. In breast tumor tissues, GLI1 expression enhanced tissue identification and discriminatory power of Merlin, cumulatively presenting a powerful substantiation of the relationship between these two proteins. We have uncovered, for the first time, details of the tumor cell metabolomic portrait modulated by Merlin, leading to activation of Hedgehog signaling. Importantly, inhibition of Hedgehog signaling offers an avenue to target the vulnerability of tumor cells with loss of Merlin.

The NF2 gene encodes the tumor suppressor protein Merlin. Originally described as mutated in the context of neurofibromatosis and tumors of the nervous system, Merlin has more recently been demonstrated to function as a tumor suppressor in mesothelioma, melanoma, and breast cancer [1][2][3][4][5][6] . Multiple post-translational modifications and interactions with cytoskeletal proteins regulate and execute Merlin's tumor suppressor function. Mutations in the NF2 gene are rare in breast cancer; instead, Merlin protein levels decrease with disease progression and metastasis due to post-translational degradation 1 .
The molecular functions of Merlin have been described at the cortical membrane and in the nucleus. Merlin binds and interacts with the actin and microtubule cytoskeleton [7][8][9] . Merlin associates with α -catenin as an essential component of the formation of adherens junctions, exerts contact-dependent inhibition of proliferation by interacting with CD44 in response to hyaluronan, sequesters EGFR from signaling stimuli, and directly inhibits p21-activating kinase (PAK) [10][11][12][13][14][15] . Functionally, Merlin inhibits contact-dependent growth and suppresses invasiveness of tumor cells. In animal models, Merlin has been demonstrated to suppress xenograft formation of multiple cancer types. Loss of Merlin enables unrestricted growth of tumor cells characterized by the aberrant upregulation of several growth-promoting signaling events in the tumor cells 16 .
The process of tumorigenesis and tumor progression is associated with metabolic changes that enable tumor cells to evolve and progress. Recent advances have defined perspectives of metabolomic profiling of cancer that have defined treatment response, 'stemness' , oxidative stress, changes with respect to the estrogen receptor (ER) status of breast cancer etc. [17][18][19] . Our work on the loss of Merlin has revealed a vital role for Merlin in advanced breast cancer 1,20 . We adopted an untargeted approach to investigate metabolomic changes impacted in breast cancer cells by Merlin. Based on MS/MS fragmentation data, we discerned novel, specific changes in glutathione metabolites associated with Merlin. Following these leads, we determined that loss of Merlin increases oxidative   Loss of Merlin induces reactive oxygen species (ROS) concomitant with lower levels of reduced glutathione (GSH). As detailed above, the relative abundance of GSH-containing metabolites is significantly decreased in cells knocked down for Merlin (KD) ( Fig. 2A). In agreement with this, the intracellular levels of reduced glutathione are decreased in MCF7 and MCF10AT cells in conditions of (stable) Merlin deficiency (KD) (Fig. 2B). Recognizing the predominant role of reduced glutathione in keeping a check on cellular ROS levels (Supplementary Figure 5A), we evaluated ROS levels in cells silenced for Merlin. The steady state ROS levels show a modest but statistically significant (p < 0.05) increase in cells stably silenced for Merlin (Fig. 2C). Further increasing the cellular ROS levels with menadione resulted in significantly greater cell death of Merlin-silenced cells (Supplementary Figure 5B). This suggests that oxidative stress following Merlin deficiency is not sufficient MnTBAP was removed after 30 minutes and Menadione was added. At the end of one hour of initial treatment the medium was removed and fresh complete growth medium was added. The experiment was terminated after 8 hours. The luciferase promoter activity was normalized to total protein and represented as percentage of activity in comparison to the vehicle control. *indicates p < 0.05. (E) ROS induction in MCF10AT cells upregulates Hh signaling pathway. MCF10AT cells stably transfected with a non-targeting shRNA were treated with menadione to induce ROS production or MnTBAP to squelch ROS. After 24 hours of treatment, transcript expression of Hh receptor PTCH1 and transcription factors GLI1 and GLI2 was assessed by an qRT-PCR. C(t) values were normalized to GAPDH and the vehicle control (ethanol) group using ∆ ∆ Ct method. *indicates p < 0.05. (F) Menadione increases, whereas GANT61 inhibits, the expression of Hh transcription factors GLI1 and GLI2 and PTCH and SMOH. The expression status of these molecules was assessed by real-time quantitative RT-PCR and normalized to endorse control gene, β -Actin. *indicates p < 0.05. to confound cell survival; instead an additional increase in cellular ROS renders Merlin deficient cells more susceptible to cell death.
Cellular ROS has profound effects on intracellular signaling events and plays a protective role against apoptosis 21,22 . Notably, oxidative stress activates Hh signaling and protects cells against apoptotic death by invoking PI-3-kinase-AKT-Bcl-2 signaling 23,24 . We evaluated the effects of modulating cellular ROS on the transcription activity of GLI, the transcription factor of the Hh pathway. While increasing cellular ROS with menadione had a dose-dependent increase on the transcription activity of GLI, the ROS scavenger, MnTBaP attenuated the activity of GLI in MCF7 cells (Fig. 2D). The levels of bona fide GLI transcriptional targets GLI1, GLI2, and PTCH also increased concordantly with an increase in ROS levels. Importantly, increases in transcriptional targets with menadione treatment were significantly decreased with the GLI transcription inhibitor GANT61 (Fig. 2E,F), indicating that signaling initiated by elevated ROS is blunted upon blocking the activity of GLI. This evidence solidified the role of ROS in facilitating activation of Hh/GLI signaling. Increase in cellular ROS prompted by loss of Merlin results in activated Hh signaling. In order to attribute a direct role for Merlin in modulating Hh signaling, we assessed cells that were modulated for Merlin expression. Abrogating Merlin in MCF10AT cells (MCF10AT KD) and MCF7 cells (MCF7 KD) resulted in increased baseline activity of the Hh transcription factor GLI, measured as a luciferase readout (Fig. 3A). Conversely, SUM159 cells engineered to stably express Merlin showed a decrease in the GLI promoter activity (Fig. 3A). In concordance with the reporter activity, the steady state levels of the two important transcription factors GLI1 and GLI2 were significantly elevated in cells abrogated for Merlin (Fig. 3B). The levels of the GLI3 transcriptional repressor and SHH ligand remained unaltered. We also registered an increase in the nuclear accumulation of GLI1 and GLI2 in cells silenced stably for Merlin ( Fig. 3C; Supplementary Figure 6A) clearly demonstrating that not only were the levels of GLI increased, but also the nuclear accumulation of GLI transcription factors was greater when tumor cells are modulated for Merlin deficiency.
In order to determine the impact of elevated cellular ROS on Hh/GLI activity in the KD cells, we treated cells with the ROS scavenger, MnTBaP. This caused a significant decrease in the levels of the bona fide GLI transcriptional targets GLI1, GLI2, and PTCH, suggesting that elevated ROS levels regulate Hh/GLI activity. Co-treatment with menadione reversed this trend (Fig. 3D). Menadione alone could not be evaluated in this assay since it compromised cell viability, similar to that depicted in Supplementary Figure 5B. We assessed the effect of supplementing cells with GSH as a means to overcome Merlin deficiency. GSH supplementation significantly reduced GLI transcriptional activity with a concordant decrease in the transcriptional targets of GLI (Fig. 3E,F). Cumulatively the data clearly indicate that loss of Merlin causes dysregulated activation of Hh/GLI signaling. Elevated ROS in cells compromised for Merlin prompts activation of Hh/GLI signaling and the anti-oxidant GSH enables reduced Hh/GLI activity in cells with Merlin deficiency.
Targeting upregulated Hh/GLI activity is a strategy to reduce tumor burden in cells compromised for Merlin. Thus far, the data suggests that loss of Merlin causes aberrant activation of Hh signaling due to an increase in the levels and nuclear abundance of the GLI transcription factors. So we tested the impact of GANT61 on tumor growth in the perspective of Merlin deficiency in a pre-clinical system. We injected MCF10AT cells (control and silenced for Merlin) into the mammary fat pad of female athymic, nude mice. When tumors were palpable, mice were randomized and treated with GANT61 (50 mg/kg; i.p. 3 times per week). Cells silenced for Merlin formed robustly growing tumors compared to the MCF10AT NT control cells. Moreover, the tumors of MCF10AT silenced for Merlin were more responsive to GANT61 showing a notably decreased tumor growth rate (Fig. 3G).
We enriched the tumor cells from the mice treated with GANT61 and evaluated the effect of GANT61 on their viability in culture. Tumors from the MCF10AT KD cells showed a stark sensitivity to an increasing dose of GANT61 relative to the MCF10AT NT cells, which were resistant to GANT61 (Supplementary Figure 6B). This ex vivo data complements the in vivo findings and strongly attests targeting Hh/GLI signaling in cells that have lost Merlin expression. Thus, loss of Merlin enforces dependence on active Hh signaling for in vivo tumor growth.
Loss of Merlin manifests as enhanced mesenchymal phenotype that is sensitive to inhibition by GANT61. Hh signaling and oxidative stress stimulate invasiveness of tumor cells 25,26 . Cells knocked-down for Merlin showed a significant increase (p < 0.05) in their invasive potential (Fig. 4A, Supplementary Figure 6C). However, treatment with the small molecule GANT61, an inhibitor of GLI transcription activity, substantially decreased invasion (p < 0.01) in a dose-dependent manner (Fig. 4A). We also assessed the behavior of these cells in 3D culture. In contrast to the MCF7 NT and MCF10AT NT cells that form a distinctly circumscribed structure surrounded by intact basement as visualized with laminin V, the integrity of the laminin enclosure was breached in the cells silenced for Merlin. Importantly, treatment of the MCF7 KD and MCF10AT KD cells with GANT61 inhibited their invasive phenotype and restored the integrity of the laminin V-enclosed structure (Fig. 4B, Supplementary Figure 7A). This indicated that the invasive manifestation evoked by the loss of Merlin can be blunted by treatment with a Hh/GLI inhibitor.
Given the vital role of Hh activity in epithelial-mesenchymal plasticity that we and others have reported 27-30 , we queried the expression of proteins indicative of the epithelial or mesenchymal characteristics. Merlin deficiency in MCF10AT and MCF7 cells was associated with a gain in the expression of mesenchymal markers concomitant with a decrease in expression of epithelial markers ( Fig. 4C; Supplementary Figure 7B). In contrast, expression of Merlin in SUM159 cells was associated with blunted expression of mesenchymal markers (Fig. 4C). This was also strikingly visible in the MCF7 cells silenced for Merlin where the loss of Merlin resulted in increased expression of vimentin protein concomitant with loss of E-cadherin (Fig. 4D).
In the perspective of our findings that an increase in cellular ROS activates Hh signaling, we evaluated a role for cellular ROS in modulating epithelial-mesenchymal-plasticity. An increase in cellular ROS following menadione treatment caused a predominant increase in multiple mesenchymal markers. The GLI inhibitor, GANT61 blunted the effects of menadione on the expression of mesenchymal markers attesting that the mesenchymal attributes elicited by menadione-induced ROS were attributable, in part, to Hh/GLI activity (Fig. 4E). Collectively, the findings clearly indicate that the effects of loss of Merlin on manifestation of mesenchymal attributes are phenocopied by the increase in cellular ROS; inhibiting the activity of GLI attenuates the mesenchymal attributes. Collectively, the data suggest that loss of Merlin enables a mesenchymal phenotype that is sensitive to inhibition by GANT61. GLI1 enhances tissue identification and discriminatory power of Merlin. We evaluated primary breast tumors from patients for their expression of Merlin and GLI1 by immunohistochemistry. In contrast to Merlin expression in normal breast tissues, tumor specimens showed significantly decreased staining for Merlin (p < 0.0001) (Fig. 5A; Supplementary Figure 7C). GLI1 expression was notably increased (p < 0.043) in tumor tissues relative to normal tissues ( Fig. 5B; Supplementary Figure 7C). These relationships were independent of hormonal status, age, and tumor grade. To assess the discriminatory power of Merlin and GLI1, we applied a logistic regression model to a binary variable of normal and tumor tissue to our data. The Chi-square test for the ROC curve not being equal to chance alone (p < 0.001; ROC curve area = 0.9486; 95% CI for the ROC curve (0.8933, 1.0)) indicates that Merlin has a discriminatory power for distinguishing between normal and tumor tissues (Fig. 5C). The logistic regression also showed that GLI1 by itself has appreciable discriminatory power between normal and tumor tissues (p < 0.001; ROC curve area = 0.7730; 95% CI for the ROC curve (0.6687, 0.8773)) (Fig. 5D). Further, multiple logistic regression showed that GLI1 increases the discriminatory power of Merlin (p < 0.001; ROC curve area = 0.9946; 95% CI for the ROC curve (0.9822, 1.0) (Fig. 5E). Cumulatively, these data clearly present a compelling case for Merlin deficiency and gain of GLI1 expression.

Discussion
Loss of Merlin decreased levels of reduced glutathione in tumor cells. As a tumor suppressor, Merlin has been characterized for its role in mediating contact inhibition of cell growth while eliciting the involvement of receptor tyrosine kinase, PI3K/Akt, small GTPases, cell adhesion, mammalian target of rapamycin (mTOR), and Hippo pathways. Given the convergent role of Merlin in inhibiting cell proliferation, we undertook an unbiased, untargeted metabolomics approach to discern the global metabolomic profile impacted by Merlin in breast cancer cells. Of the more than 6500 metabolites detected via our untargeted LC-MS metabolomics approach, appropriate technical filters allowed for the discernment of 47 metabolites that were significantly changed upon modulation of Merlin expression in breast cancer. None of these metabolites matched the available MS/MS spectra with reference to publically accessible human metabolite databases, underscoring the novelty of these metabolites. We minimized this limitation by utilizing the mummichog functional grouping approach to pathway analysis 31 . Mummichog pathway analysis revealed that alterations in Merlin levels broadly affected metabolism of glutathione, amino acids, nucleotides, and vitamins. By analyzing MS/MS spectra, we were able to confirm that Merlin alters the levels of various metabolites containing amino acid derivatives and nucleotide derivatives.
In the present study, while the precise identification of metabolites was challenging due to their low abundance and unavailability of references, we identified that Merlin-modulates glutathione levels in breast cancer. The levels of reduced glutathione were significantly decreased in cells knocked down for Merlin as evident by the decreased abundance of the corresponding GSH-containing metabolites.

Increase in cellular ROS levels following Merlin deficiency resulted in increased activity of Hh/GLI
signaling. Glutathione plays a significant role in cellular progression, cell differentiation, and apoptosis.
Relevant to this study is the fact that GSH plays a main role in the maintenance of the intracellular redox balance, mainly by antioxidant defense against ROS and against lipid hydroperoxidases 32 . At low levels, ROS act as signaling molecules and activate pathways favoring cell survival and proliferation, while at moderate levels ROS induces DNA damage and promotes mutagenesis. Extreme levels of ROS levels exert oxidative stress on the cells that can lead to cell death or senescence 33 . The antioxidant function of GSH is determined by the redox-active thiol (− SH) of cysteine that is oxidized when GSH reduces its target proteins/molecules. In the process GSH is oxidized to GSSG. As such, the GSH:GSSG ratios are important to maintain the redox balance in cells. The modulation of GSH is viewed paradoxically. While enhancing the capacity of GSH protects cells from redox-related changes, the strategy of depleting GSH is aimed at sensitizing cancer cells to chemotherapy. The approach of inhibiting GSH in combination with standard-of-care chemotherapies is underway in clinical trials 34,35 . As such, our data indicate that by virtue of decreasing levels of reduced GSH, loss of Merlin enables an increase in the cellular levels of ROS. From the perspective that Merlin expression is severely reduced in advanced breast cancer and that increased ROS is associated with metastatic potential, our data uncover a novel role for Merlin in regulating tumor cell redox balance and ROS levels. ROS modulates multiple intracellular signaling events including PI3K/AKT, NF-κ B, STATs, Src, Wnt/β -catenin and others 21,36 . Our data clearly demonstrate that ROS causally increases the activity of Hh/GLI signaling characterized by upregulation of the transcriptional activity of the GLI transcription factors. In cells abrogated for Merlin, increased levels and nuclear accumulation of the Hh transcription factors GLI1 and GLI2, corroborated with greater GLI activity and manifestation of a mesenchymal phenotype characterized by loss of a distinct basement membrane structure. While Hh/GLI signaling is classically activated by specific Hh ligands, non-canonical activation of GLI (that bypasses the need for specific Hh ligands) is well-reported to occur through the activities of EGF, OPN, TGF-beta [37][38][39] . Activation of Hh signaling as a protective effort by neuronal cells in response to oxidative stress engages the cellular anti-apoptotic machinery 23,24 . Our data shed critical insight into the molecular mechanistic events in this process. Increase in cellular oxidative stress following loss of Merlin resulted in increased levels of the GLI1 and GLI2 transcription factors and their ensuing transcriptional activity.

Merlin deficiency sensitizes breast cancer cells to inhibition of Hh/GLI signaling.
Remarkably, cells abrogated for Merlin showed remarkable sensitivity to the GLI inhibitor GANT61. While the growth of cells silenced for Merlin was overall greater than control cells, cells abrogated for Merlin also showed the greatest response to GANT61, presenting evidence that cells compromised for Merlin acquire dependence on Hh/ GLI signaling. GANT61 also resulted in reversal of the invasive and mesenchymal phenotype. This effect was phenocopied in cells induced for cellular ROS with menadione, where GANT61 antagonized the effects of menadione-mediated ROS increase on the mesenchymal phenotype. Collectively, these observations underscored the fact that increased ROS following loss of Merlin sensitizes cells to inhibition of Hh/GLI signaling. In our cohort of breast tumor tissues, GLI1 enhanced tissue identification and discriminatory power of Merlin, cumulatively presenting as powerful substantiation of the relationship between these two proteins.

Conclusion
In conclusion, the approach of unbiased, untargeted metabolomics revealed a novel function of Merlin on glutathione regulation. The resultant increase in cellular ROS profoundly impacted activation of Hh/GLI signaling, likely as a survival mechanism. This is evident in the finding that increased GLI1 expression enhances the discriminatory power of Merlin in human breast tissues. The activation of Hh signaling in Merlin-deficient conditions also presents as the tumor cells' vulnerability since inhibiting Hh/GLI signaling in Merlin-deficient breast tumor cells limits their oncogenic growth. Notably, our evidence indicates that targeting Hh/GLI signaling may be a viable therapeutic strategy in breast tumors that are deficient in Merlin expression.

LC-MS/MS analysis.
An aliquot (5 μ L) of each sample was loaded onto a Nano cHiPLC 200 μ m ID × 0.5 mm ChromXP C18-CL 3 μ m 120 Å reverse-phase trap cartridge (Eksigent, Dublin,CA) at 2 μ L/min using an Eksigent autosampler. The bound peptides were flushed onto a Nano cHiPLC column 200 μ m ID × 15 cm ChromXP C18-CL 3 μ m 120 Å column (Eksigent) with a 15 min linear (5-95%) acetonitrile gradient in 0.1% formic acid at 1000 nL/min using an Eksigent 415 NanoLC system. The column was washed with 95% acetonitrile-0.1% formic acid for 5 min and then re-equilibrated with 5% acetonitrile-0.1% formic acid for 5 min. A 5600 Triple-Tof mass spectrometer (Sciex, Toronto, Canada) was used to analyze the metabolite profile. The IonSpray voltage for positive and negative modes were + /− 2300 V and the declustering potential was + /−80 V. Ionspray and curtain gases were set at 10 psi and 25 psi, respectively. The interface heater temperature was 120 °C. Eluted compounds were subjected to a time-of-flight MS survey scan from m/z 50-1000 to determine the top twenty most intense ions for MSMS analysis. Product ion time-of-flight scans at 50 msec using a collision energy spread of 15 eV with a set collision point of 35 eV were carried out to obtain the tandem mass spectra of the selected parent ions over the range from m/z 50-1000. Spectra were centroided and de-isotoped by Analyst software, version 1.6 TF (Sciex).

Data Analysis and Metabolite
Identification. LC-MS data were processed using XCMSonline (https:// xcmsonline.scripps.edu; Scripps, La Jolla). Each sample was normalized to total ion chromatogram. Peaks were manually filtered to retain median retention time between five and 25 min and mass tolerance was set to 0.01 Da. Excel.csv files were created containing the median mass and retention time for each peak and its peak area. These Scientific RepoRts | 7:40773 | DOI: 10.1038/srep40773 were uploaded for statistical analysis using MetaboAnalyst 3.0 (http://www.metaboanalyst.ca). Data were normalized by the total ion current of the filtered metabolite ions, mean centered and subjected to Pareto scaling. Using univariate analysis, a fold change less than 0.5 or greater than 2, and p-value < 0.05 (Student's t-test) was considered significant for further analysis. All peaks passing the initial filters were classified according to "real" or "not real" peaks based on visual confirmation of their ion chromatograms. Mass to charge ratios were searched via the METLIN database (Scripps) assuming an accuracy of 10 ppm for matches to previously identified metabolites with available MS/MS spectra. Similar searches were conducted on the Human Metabolite Database (HMDB) [41][42][43] Pathway analyses were conducted via the raw, unfiltered data using "Connections" feature of XCMSonline which utilizes the Mummichog metabolite functional grouping approach 31 . Mummichog analysis was also conducted at the command line level on the normalized data from the MetaboAnalyst procedure using Mummichog, version 1.0.5 (https://code.google.com/p/atcg/wiki/mummichog_for_metabolomics). The data files (in Excel.txt format) contained the m/z of all the metabolite ions, their retention times, p-values, and t-scores.
To identify metabolites that were significantly changed upon modulation of Merlin, three data analysis approaches were employed. First the parent or precursor ions and retention times (Rt) of all 47 metabolites altered in the context of Merlin were queried for potential matches in METLIN and the Human Metabolome Database. None of these metabolites could be identified by this method. Because the identity of these metabolites was not known, conventional pathway analysis was not possible. The mummichog analysis within XCMSOnline revealed likely changes in the metabolism of glutathione and a broad range of amino acids (Supplementary Table 2). As no structural information was obtained after exhaustive database search, accurate masses of selected precursor and product ions together with detailed interpretation of fragmentation patterns of precursor ions upon MS/MS enabled us to propose partial identification of the altered metabolites. Luciferase Assay. Cells (30,000) were transfected with an 8XGLI construct in pGL3 promoter plasmid as described previously (3). Empty pGL3 promoter vector was used as control. Readings were normalized to total protein content. Each parameter was studied in triplicate and the experiment repeated at least three times. To evaluate the effect of reduced glutathione (ROS scavenger) on GLI activity, MCF10AT KD cells were transiently transfected with the 8XGLI construct and the medium was supplemented with different concentrations of GSH (L-Glutathione reduced, Cat. No. 5219 TOCRIS). The experiment was terminated after 24 hours of GSH supplementation.

Western Blotting
Quantitative RT-PCR (qRT-PCR). qRT-PCR was done in triplicate and repeated at least once. TaqMan Gene Expression Assays (Applied Biosystems) were used to query the transcripts of interest. Transcript levels were normalized to endorse control gene GAPDH or β -actin (Δ CT) to calculate changes in expression (2 ΔΔ CT).

MTS Assay.
Cells were seeded in 96-well plates at a density of 10,000 cells/well. The effect of menadione (and ethanol as vehicle control) on cell viability was assessed by MTS assay (Promega, Madison, WI) after 24 hours of treatment.
Glutathione assay. Cells were seeded in 96-well plates overnight; 100 μ L of the GSH-Glo reagent (Promega, Madison, WI) was added to each well, mixed, and the plate was incubated at room temperature for 30 minutes. Luciferin Detection Reagent was added to each well and luminescence was recorded. Immunocytochemistry. Plated cells were fixed in 4% paraformaldehyde, permeabilized with 0.3% Triton X-100 followed by labeling with either anti-E-Cadherin (#3195, Cell Signaling) or anti-Vimentin (#5741, Cell Signaling) or anti-GLI1 primary antibody (#2643, Cell Signaling) overnight at 4 °C. Images were captured at 40X, and the acquisition time (1 second for E-Cadherin and 4 seconds for Vimentin and GLI1) was kept constant across compared groups. Invasion assay. Cells were treated for 24 hours with DMSO, GANT61 or were left untreated. BD Biocoat Matrigel invasion chambers were thawed and rehydrated prior to seeding 20,000 cells/chamber in serum-free Scientific RepoRts | 7:40773 | DOI: 10.1038/srep40773 medium containing DMSO or GANT61 concordant with cell treatment. The assay was performed in triplicate and repeated twice as previously described 44 . Animal Studies. The animal studies were approved by the IACUC at UAB. All methods were performed in accordance with relevant guidelines and regulations. Cells (5 × 10 6 ) were injected into the inguinal mammary fat pad of 6-week old female, athymic nude mice 3-5 days following implantation of an estrogen pellet (#SE-121:0.72 mg/pellet dose for 60 day release time; Innovative Research of America) in the sub-scapular region. GANT61 administration was initiated 8 days after cell injection and tumor measurements were recorded three times per week. The experiment was terminated at day 35. Tumor cells were harvested from xenografts and assessed for sensitivity to GANT61 by MTS assay.
Immunohistochemistry. Breast tumor tissue microarrays were procured from NCI Cooperative Breast Cancer Tissue Resource and stained immunohistochemically for Merlin (#A-19; Santa Cruz) and GLI1 (#ab7523; Abcam). Staining intensity was quantitated with computer-assisted image analysis in a Dako ACIS III Image Analysis System (Glostrup, Denmark) 45 .
Statistical analyses of in vitro, in vivo and patient data. Statistical differences between groups were assessed using the Mann-Whitney test, t-test or ANOVA, using GraphPad Prism 4 software or SAS software version 9.4 (SAS Inc.). All results with a p value less than 0.05 were considered statistically significant.
Associations between intensities of Merlin and GLI expressions and patient Breast Cancer status data were assessed using the Wilcoxon rank test. The univariate and multiple logistic regression models were fit to a binary variable normal versus tumor with Merlin and GLI1 as possible predictors. The possibility of developing a model using the relationship between GLI1 and Merlin was tested with a logistic regression model. The area under the receiver operating characteristic (ROC) curve and the corresponding 95% confidence interval was used to determine the predictive ability of models and in model selection. The Chi-square test was used to test the ROC curve not being equal to chance alone.