OPEN Oncogene (2017) 36, 2737–2749 www.nature.com/onc ORIGINAL ARTICLE Stem cell-like transcriptional reprogramming mediates metastatic resistance to mTOR inhibition F Mateo1,49, EJ Arenas2,49, H Aguilar1,49, J Serra-Musach1, G Ruiz de Garibay1, J Boni1, M Maicas3, S Du4, F Iorio5,6, C Herranz-Ors1, A Islam7, X Prado1, A Llorente1, A Petit8, A Vidal8, I Català8, T Soler8, G Venturas8, A Rojo-Sebastian9, H Serra10, D Cuadras11, I Blanco12, J Lozano13, F Canals14, AM Sieuwerts15, V de Weerd15, MP Look15, S Puertas16, N García1, AS Perkins17, N Bonifaci1, M Skowron1, L Gómez-Baldó1, V Hernández18, A Martínez-Aranda18, M Martínez-Iniesta16, X Serrat19, J Cerón19, J Brunet20, MP Barretina21, M Gil22, C Falo22, A Fernández22, I Morilla22, S Pernas22, MJ Plà23, X Andreu24, MA Seguí25, R Ballester26, E Castellà27, M Nellist28, S Morales29, J Valls29, A Velasco29, X Matias-Guiu29, A Figueras10, JV Sánchez-Mut30, M Sánchez-Céspedes30, A Cordero30, J Gómez-Miragaya30, L Palomero1, A Gómez30, TF Gajewski31, EEW Cohen32, M Jesiotr33, L Bodnar34, M Quintela-Fandino35, N López-Bigas36,37, R Valdés-Mas38, XS Puente38, F Viñals10, O Casanovas10, M Graupera10, J Hernández-Losa39, S Ramón y Cajal39, L García-Alonso5, J Saez-Rodriguez5, M Esteller30,37,40, A Sierra41, N Martín-Martín42, A Matheu43,44, A Carracedo42,44,45, E González-Suárez30, M Nanjundan46, J Cortés47, C Lázaro12, MD Odero3, JWM Martens15, G Moreno-Bueno48, MH Barcellos-Hoff4, A Villanueva16, RR Gomis2,37 and MA Pujana1 Inhibitors of the mechanistic target of rapamycin (mTOR) are currently used to treat advanced metastatic breast cancer. However, whether an aggressive phenotype is sustained through adaptation or resistance to mTOR inhibition remains unknown. Here, complementary studies in human tumors, cancer models and cell lines reveal transcriptional reprogramming that supports metastasis in response to mTOR inhibition. This cancer feature is driven by EVI1 and SOX9. EVI1 functionally cooperates with and positively regulates SOX9, and promotes the transcriptional upregulation of key mTOR pathway components (REHB and RAPTOR) and of lung metastasis mediators (FSCN1 and SPARC). The expression of EVI1 and SOX9 is associated with stem cell-like and metastasis signatures, and their depletion impairs the metastatic potential of breast cancer cells. These results establish the mechanistic link between resistance to mTOR inhibition and cancer metastatic potential, thus enhancing our understanding of mTOR targeting failure. Oncogene (2017) 36, 2737–2749; doi:10.1038/onc.2016.427; published online 19 December 2016 1 Breast Cancer and Systems Biology Laboratory, Program Against Cancer Therapeutic Resistance (ProCURE), Catalan Institute of Oncology (ICO), Bellvitge Institute for Biomedical Research (IDIBELL), L’Hospitalet del Llobregat, Barcelona, Spain; 2Oncology Program, Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology, Barcelona, Spain; 3Centre for Applied Medical Research (CIMA) and Department of Biochemistry and Genetics, University of Navarra, Pamplona, Spain; 4Department of Radiation Oncology, New York University School of Medicine, New York, NY, USA; 5European Molecular Biology Laboratory-European Bioinformatics Institute (EMBL-EBI), Wellcome Trust Genome Campus, Cambridge, UK; 6Cancer Genome Project, Wellcome Trust Sanger Institute, Hinxton, UK; 7Department of Genetic Engineering and Biotechnology, University of Dhaka, Dhaka, Bangladesh; 8 Department of Pathology, University Hospital of Bellvitge, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 9Department of Pathology, MD Anderson Cancer Center, Madrid, Spain; 10 Angiogenesis Research Group, ProCURE, ICO, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 11Statistics Unit, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 12Hereditary Cancer Programme, ICO, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 13Department of Molecular Biology and Biochemistry, Málaga University, and Molecular Oncology Laboratory, Mediterranean Institute for the Advance of Biotechnology and Health Research (IBIMA), University Hospital Virgen de la Victoria, Málaga, Spain; 14ProteoRed-Instituto de Salud Carlos III, Proteomic Laboratory, Vall d'Hebron Institute of Oncology (VHIO), Vall d’Hebron University Hospital, Barcelona, Spain; 15Department of Medical Oncology, Erasmus University Medical Center, Daniel den Hoed Cancer Center, Cancer Genomics Centre, Rotterdam, The Netherlands; 16Chemoresistance and Predictive Factors Laboratory, ProCURE, ICO, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 17University of Rochester Medical Center, School of Medicine and Dentistry, Rochester, NY, USA; 18Biological Clues of the Invasive and Metastatic Phenotype Laboratory, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 19Cancer and Human Molecular Genetics, IDIBELL, Hospitalet de Llobregat, Barcelona, Spain; 20Hereditary Cancer Programme, ICO, Girona Biomedical Research Institute (IDIBGI), Girona, Spain; 21Department of Medical Oncology, ICO, IDIBGI, Girona, Spain; 22Department of Medical Oncology, ICO, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 23Department of Gynecology, University Hospital of Bellvitge, IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 24Department of Pathology, Parc Taulí Hospital Consortium, Sabadell, Barcelona, Spain; 25Medical Oncology Service, Parc Taulí Hospital Consortium, Sabadell, Barcelona, Spain; 26Department of Radiation Oncology, University Hospital Germans Trias i Pujol, ICO, Germans Trias i Pujol Research Institute (IGTP), Badalona, Barcelona, Spain; 27Department of Pathology, University Hospital Germans Trias i Pujol, ICO, IGTP, Badalona, Barcelona, Spain; 28Department of Clinical Genetics, Erasmus Medical Centre, Rotterdam, The Netherlands; 29Hospital Arnau de Vilanova, University of Lleida, Biomedical Research Institute of Lleida (IRB Lleida), Lleida, Spain; 30Cancer Epigenetics and Biology Program (PEBC), IDIBELL, L’Hospitalet del Llobregat, Barcelona, Spain; 31Departments of Pathology and Medicine, University of Chicago, Chicago, IL, USA; 32Moores Cancer Center, University of California San Diego, La Jolla, CA, USA; 33Department of Pathology, Military Institute of Medicine, Warsaw, Poland; 34Department of Oncology, Military Institute of Medicine, Warsaw, Poland; 35Breast Cancer Clinical Research Unit, Spanish National Cancer Research Center (CNIO), Madrid, Spain; 36Department of Experimental and Health Sciences, Barcelona Biomedical Research Park, Pompeu Fabra University (UPF), Barcelona, Spain; 37Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain; 38Department of Biochemistry and Molecular Biology, University Institute of Oncology of Asturias, University of Oviedo, Oviedo, Spain; 39 Department of Pathology, Vall d’Hebron University Hospital, Barcelona, Spain; 40Department of Physiological Sciences II, School of Medicine, University of Barcelona, Barcelona, Spain; 41 Molecular and Translational Oncology Laboratory, Biomedical Research Center CELLEX-CRBC, Biomedical Research Institute ‘August Pi i Sunyer’ (IDIBAPS), and Systems Biology Department, Faculty of Science and Technology, University of Vic, Central University of Catalonia, Barcelona, Spain; 42Center for Cooperative Research in Biosciences (CIC bioGUNE), Derio, Spain; 43Neuro- Oncology Section, Oncology Department, Biodonostia Research Institute, San Sebastian, Spain; 44Ikerbasque, Basque Foundation for Science, Bilbao, Spain; 45Department of Biochemistry and Molecular Biology, University of the Basque Country (UPV/EHU), Bilbao, Spain; 46Department of Cell Biology, Microbiology, and Molecular Biology, University of South Florida, Tampa, FL, USA; 47Department of Medical Oncology, VHIO, Vall d’Hebron University Hospital, Barcelona, Spain and 48Department of Biochemistry, Autonomous University of Madrid (UAM), Biomedical Research Institute ‘Alberto Sols’ (Spanish National Research Council (CSIC)-UAM), Translational Research Laboratory, Hospital La Paz Institute for Health Research (IdiPAZ), and MD Anderson International Foundation, Madrid, Spain. Correspondence: Dr RR Gomis, Oncology Program, Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology, Baldiri Reixac 10, Barcelona 08028, Spain or Dr MA Pujana, ProCURE, Catalan Institute of Oncology, IDIBELL, Hospital Duran i Reynals, Gran via 199, L'Hospitalet del Llobregat, Barcelona 08028, Spain. E-mail: roger.gomis@irbbarcelona.org or mapujana@iconcologia.net 49 These authors contributed equally to this work. Received 10 March 2016; revised 31 August 2016; accepted 10 October 2016; published online 19 December 2016 Resistance to mTOR inhibition F Mateo et al 2738 INTRODUCTION Metastasis dependence on mTORC1 signaling The mechanistic target of rapamycin (mTOR) kinase integrates To test the contribution of mTOR signaling to metastasis, we used cues from nutrients and growth factors and is thus a master the well-defined MDA-MB-231 breast cancer cell line, including its regulator of cell growth and metabolism.1 As such, mTOR is parental poorly metastatic population and the lung metastatic activated in most cancer types and is frequently associated with derivatives LM1 and LM2.16 Western blot analyses showed poor prognosis.2 Moreover, oncogenic mTOR signaling has a increased levels in LM2 cells of several components of the direct role in promoting cancer progression by inducing a pro- mTORC1 signaling pathway, and particularly of RAPTOR and RHEB invasion translational program.3 This program includes the across the sub-populations (Figure 1b). The enhanced signaling in downregulation of the tuberous sclerosis complex 2 (TSC2) gene, LM2 cells compared with the poorly metastatic parental popula- whose product, in a heterodimer with the TSC1 product, serves as tion was confirmed by quantification of immunohistochemical a negative regulator of mTOR complex 1 (mTORC1).4 Conse- staining of pS6 in the lung metastases that developed the cells quently, loss of Tsc2 in mice promotes breast cancer progression upon tail vein injection (Figure 1c). Expanding on these and metastasis.5 Collectively, current knowledge supports the observations, analysis of TCGA data showed negative correlations notion that mTOR signaling has a key role in cancer initiation, between TSC1/2 and an upregulated gene set whose expression progression and metastasis. was clinically and experimentally associated with breast cancer As mTOR is a key factor in cancer biology, therapies based on its metastasis to lung (lung metastasis signature (LMS)-up; Pearson’s inhibition have been widely studied6 and are central to the correlation coefficients (PCCs) o − 0.25; P-values o10 − 8). Nota- treatment of advanced metastatic breast cancer.7 However, the bly, this set was derived from the study of LM2 cells.16 success of monotherapy assays has been limited. Critically, within Next, we tested the causal role of mTOR activity in the a relatively short term, allosteric mTOR inhibition concomitantly experimental model of lung metastasis. The capacity of LM2 cells induces upstream receptor kinase signaling, which mediates to colonize the lung was assessed in the presence or absence of therapeutic resistance.8 Thus, therapies that combine allosteric an allosteric mTOR inhibitor. LM2 cells stably expressing green inhibitors (rapamycin (sirolimus) and rapalogs) with inhibitors of fluorescent protein (GFP) and luciferase were injected into the growth factor signaling have been extensively evaluated.9 lateral tail vein of immunocompromised mice, which were then Intriguingly, recent studies have further linked mTOR activity to randomly allocated to a group treated with dimethyl sulfoxide a stem cell-like cancer phenotype that mediates breast cancer (DMSO) or a group treated with everolimus, both for 38 days. metastasis10,11 and, using triple-negative (TN) breast cancer cell A significant reduction of lung colonization (and, as expected, of lines, have described that mTORC1/2 inhibition spares a cell pS6 intensity) was observed in the latter group, both by population with stem cell-like properties and enhanced NOTCH measurements of in vivo photon flux and the relative lung activity.12 These results are consistent with previous observations metastasis area ex vivo by histology (Figure 1d). Collectively, these concerning the required activation of mTOR signaling in breast data suggest that mTORC1 signaling is associated with breast cancer stem-like viability and maintenance,13 the enhancement of cancer metastatic potential and that inhibition of mTOR prevents NOTCH signaling in poorly differentiated breast tumors14 and the lung metastasis. However, it is unclear whether this association increase of tumor-initiating capacities with mTOR inhibition in persists in settings of resistance to mTOR inhibitors. liver cancer.15 In this scenario, a fundamental question emerges as to whether relative long-term adaptation or resistance to mTOR Metastatic resistance to mTOR inhibition inhibition is functionally linked to tumor-initiating properties and, To evaluate the mechanisms responsible for resistance to mTOR eventually, metastasis. inhibitors, we used two independent metastatic tumor models, Here, we explored the hypothesis that mTOR signaling supports namely a human TN BRCA1-mutated breast tumor orthotopically metastasis and remains active in therapeutic resistance in engrafted in nude mice (hereafter ortho-xenograft; Supplementary metastatic breast cancer. We found that abnormal mTOR signaling Figure 1) and the TN 4T1 murine breast carcinoma cell line enhances tumor-initiating properties and metastatic potential. engrafted in syngeneic background mice. Cells from both tumor This activity is dependent on EVI1, which in cooperation with models showed substantial mTORC1 signaling activity, particularly SOX9 sustains a transcriptional reprogramming response. at the tumor invasive front (Supplementary Figure 1).17 Unexpect- edly, although systemic treatment with sirolimus or everolimus blunted primary tumor growth in each model, it did not reduce RESULTS the number or size of lung metastases (Figure 2a). In addition, and Active mTORC1 signaling associates with distant metastasis contrary to expectations, the intensity of pS6 staining at the mTORC1 is the target of one of the latest drugs approved for the invasive tumor fronts of the primary lesion and in the lung treatment of breast cancer in the advanced metastatic setting,7 metastases of the sirolimus-treated ortho-xenografts was sig- which suggests that this protein complex has a potential role in nificantly higher than in the control animals (Figure 2b). Similarly, supporting metastasis and aggressive features. To study this a key factor in cancer metastasis initially identified in LM2 cells relationship, a tissue microarray of primary breast tumors was and human data analyses (thus included in LMS-up), FSCN1,16,18 assessed for mTORC1 activity by means of immunohistochemical was found to be significantly overexpressed in both experi- determination of phospho-Ser235/236-ribosomal protein S6 (pS6), mental models exposed to mTOR inhibitors (Figure 2c). a well-established downstream target of mTORC1.1 An association Subsequent gene expression analysis of the treated tumors between pS6 positivity and the basal-like tumor phenotype or CK5 revealed coordinated changes concurrent with mTOR inhibition positivity was observed (Figure 1a; Mann–Whitney test Po 0.01). that were associated with LMS activation (Supplementary Most importantly, an association was also detected between Figure 2). These changes included overexpression of LMS-up in medium-high pS6 positivity and the development of distant the sirolimus-treated ortho-xenografts and underexpression metastases (Fisher’s exact test P = 0.02; odds ratio (OR) = 2.64, 95% of LMS-down in the everolimus-treated 4T1 tumors (as measured confidence interval (CI) 0.95–7.35). Intriguingly, whereas the by the gene set expression analysis (GSEA), P-values o 0.05; analyses by tumor subtypes were underpowered, both estrogen Supplementary Figure 2). receptor (ER)-positive and ER-negative cases suggested a trend To further study resistance to allosteric mTOR inhibition, we toward increased metastatic risk (ORs = 4.44 and 1.96, respec- subjected MCF7 ER-positive and HCC1937 TN cells to long-term tively). Thus, enhanced mTOR activity and breast cancer meta- exposure to 50 and 150 nM of everolimus, respectively. After a static potential appear to be linked. period of sensitivity defined by undetectable or very low levels of Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2739 Figure 1. mTORC1 activity concurrent with enhanced metastatic potential. (a) Left panels, representative immunohistochemical scores (0, negative, to 3, highest expression) of pS6 staining in the tissue microarray (TMA) of primary breast tumors. Right panel, results for the association between pS6 staining and distant metastasis. (b) Increased expression of mTORC1 pathway components with enhanced metastatic potential of MDA-MB-231 cells. The loading control (α-tubulin, TUBA) is shown. Bottom panel, graph showing quantifications of protein levels relative to parental and TUBA (per sample). (c) Increased pS6 expression in lung metastases developed by LM2 cells. The arrows mark magnified fields. Right panel, box-and-whisker plots for the quantification (pixels/area, p/a) of pS6 intensity; three mice and three similar lung metastases were analyzed in each setting. The P-value of the two-tailed Mann–Whitney test is shown. (d) Left panel, graph showing the in vivo photon flux quantification in mice injected with LM2 and treated with DMSO or everolimus. Representative images from bioluminescence in lungs from DMSO- or everolimus-treated mice are shown. The scale bar depicts the range of photon flux values as a pseudo-color display, with red and blue representing high and low values, respectively. Right top panels, quantification of lung colonization (total metastasis area normalized per total lung area, based on HE). Right bottom panels, representative immunohistochemical results for pS6 and quantification of normalized intensities. pS6, both cell lines recovered canonical mTORC1 signaling in metastatic and tumor initiation properties. However, the precise 90–120 days (Figure 2d, top panels). Similarly to the in vivo regulators of this aggressive reprogramming remain to be observations, FSCN1 increased concurrently with adaptation to determined. everolimus in both cell settings (Figure 2d, bottom panels). Subsequently, transcriptome analyses showed a significant TSC1/2 expression correlates negatively with tumor-initiating change of the LMS in both cell lines, and particularly of the features LMS-up in HCC1937 cells (Supplementary Figure 3). Given that differences in colony formation assays and tumor Interestingly, both everolimus-adapted cell models showed initiation properties were observed in mTOR inhibitor-resistant cell significantly higher colony-forming capacity, with the higher relative difference found in HCC1937 cells (Figure 2e). Accordingly, populations, we then explored the association between mTOR fluorescence-activated cell sorting revealed an increase of CD49f+ signaling and cancer cell initiation features in gene expression and of CD44+/CD24 − cells in everolimus-adapted MCF7 and profiles from patient samples. To this end, we computed the HCC1937 cultures, respectively (Figure 2f). Although MCF7 did not expression correlations between TSC1 or TSC2 (TSC1/2) and 20 show an increase in CD44+/CD24 − , CD49f positivity has been previously defined gene expression signatures using breast linked to cancer stem cell-like properties.19 In addition, quantita- cancer data from The Cancer Genome Atlas (TCGA).21 The tive gene expression analysis revealed a significant increase of signatures (full annotation is provided in Supplementary Table 1) SOX2 in everolimus-adapted MCF7 cells and, in turn, an increase of include a consensus set derived from the study of embryonic stem NANOG and OCT4 (but not SOX2) in everolimus-adapted HCC1937 cell-like cells (sESCs),22 a consensus set of correlated master cells (Supplementary Figure 4). Notably, an increase in SOX2, but regulators of breast cancer stemness-like functions (hereafter not the two additional stem cell-like markers, has also been sMRS),23 and a MYC-centered regulatory network (sMYC);24 described in MCF7 cells resistant to tamoxifen.20 Therefore, by importantly, these signatures were originally associated with poor combining in vivo and in vitro models of breast cancer, we reveal prognosis and/or metastatic potential of ER-negative breast that exposure to allosteric mTOR inhibitors consistently promotes cancer and other types of cancer.22–24 The expression profiles of Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2740 Figure 2. Metastatic resistance to mTOR inhibition. (a) Left panels, graphs showing the average and standard deviation of micro- and macro- metastases observed in the lungs of the DMSO- and sirolimus- or everolimus-treated orthoxenografts and 4T1 tumors, respectively. The results correspond to the last day of treatment, and micro- versus macro-metastases were defined using a 2 mm width threshold, and by examining at least three tissue levels separated by 4 20 μM. Right panels, growth rates of the DMSO- and sirolimus- or everolimus-treated tumors. (b) Representative immunohistochemical results for pS6 at the invasive tumor fronts (magnifications; top right panels) and the lung metastases (right panels) of DMSO- or sirolimus/everolimus-treated mice. The middle panels show quantifications, which correspond to three tumors, three equal front areas, and three metastases in each case. (c) Representative immunohistochemical tumor results for FSCN1/Fscn1 in DMSO- or sirolimus/everolimus-treated mice; quantifications are shown in right panels. (d) Recovered pS6 signal with concurrent FSCN1 overexpression through adaptation to everolimus in MCF7 and HCC1937 cells. Days of treatment are shown. (e) Left panel, graph showing the quantification of colonies from untreated and everolimus-adapted cells (12 culture fields were analyzed). The one-tailed t-test P-values are shown. Representative images of cell cultures are shown in right panels. (f) Flow cytometry results showing the cell counts for CD49f/EpCAM and of CD44/CD24 positivity in untreated or everolimus-adapted MCF7 and HCC1937 cells, respectively. TSC1/2 were found to be negatively correlated (PCCs o − 0.10; P- changes in the cell line models, but most of the regulators of sMRS values o 0.05) with most of the signatures (Figure 3a). In turn, were found to be overexpressed in HCC1937 cells (Figure 3e). positive correlations were observed with the downregulated These results were confirmed by western blot analysis of HMGA1 genes that characterize mammary epithelial basal and luminal (Figure 3f), which has been associated with poor prognosis and progenitor cells (Figure 3a).25 Moreover, the expression of TSC1 metastatic breast cancer.28 The observed differences between the was positively correlated with a downregulated gene expression in vitro and in vivo expression changes may be due to the signature associated with oncogenic PI3KCA activity in breast molecular specificity and/or biological conditions involved in each cancer.26,27 Collectively, these results confirmed that mTOR activity setting. Globally, however, inhibition of mTOR appears to be is associated with stem cell-like gene expression profiles in breast coupled to the transcriptional reprogramming that sustains cancer. metastatic and tumor initiation features. Next, we observed that the sESC, sMRS and sMYC largely distinguished the expression profiles of mTOR inhibitor-treated EVI1 couples mTOR signaling to metastasis tumors from those treated with DMSO (Figure 3b). The regulators Feedback activation of known mediators of resistance to rapalogs of the sMRS (originally defined as Core-9)23 were commonly was not observed in the sirolimus-treated ortho-xenografts, but an overexpressed upon mTOR inhibition (Figure 3c). Furthermore, a increase in phospho-Thr202/Tyr204 ERK (pERK) was detected in strong overexpression of the three signatures was detected when 4T1 tumors treated with everolimus (Supplementary Figure 5). the ortho-xenografts were allowed to re-grow following treatment Exome sequence comparison between one DMSO- and one with sirolimus (relative to the DMSO-treated re-growth, GSEA P- sirolimus-treated ortho-xenograft did not identify acquired muta- values o0.001; Figure 3d and Supplementary Table 2). Analysis of tions affecting components of the canonical TSC/mTOR pathway the three signatures defined above did not reveal significant (Supplementary Table 3). In vitro, only modest time-dependent Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2741 Figure 3. Co-expression analysis and stem cell-like signatures. (a) TCGA network of significant co-expression levels (PCC P-values o 0.05) between TSC1 or TSC2 and signatures derived from stem cell-like cell studies (Supplementary Table 1). The nodes represent TSC1/2 and the signatures, and the edges positive (red) or negative (green) correlations. Edge width is proportional to the corresponding PCC value. (b) Clustering correlation of sESC, sMRS and sMYC. The ortho-xenografts are differentially clustered relative to the treatment, and a similar trend is observed for 4T1 tumors. (c) The master regulators of the sMRS (that is, Core-9) are found to be relatively overexpressed upon mTOR inhibition. (d) Significant overexpression of sESC, sMRS and sMYC in regrown ortho-xenografts after sirolimus treatment. The GSEA ESs and the nominal P-values are shown. (e) Most of the regulators of sMRS are relatively overexpressed in everolimus-adapted HCC1937 cells. (f) HMGA1, which is encoded in Core-9, is upregulated upon adaptation to mTOR inhibition, particularly in HCC1937 cells. changes of phospho-S473 AKT (pAKT) and phospho-Y703 STAT3 was observed in the gene expression profiles from ortho- (pSTAT3) were observed (Supplementary Figure 6). Although the xenograft samples and a target gene set of the ecotropic viral lack of detection of known factors of resistance in our in vitro and integration site-1 (EVI1) proto-oncogene (Transfac V$EVI1_02; in vivo analyses may be due to long-term treatments, we next Supplementary Figure 7). Similar associations for predicted EVI1 sought to analyze a different mechanism that could be common target sets were observed using data from the 4T1 tumors and to all models. MCF7 cells (Supplementary Figure 7). Importantly, expression Given the transcriptomic changes observed across the in vivo analysis using TCGA data revealed positive correlations between and in vitro models, the data were analyzed to identify alternative EVI1 and stem cell-like signatures (sMYC and from mammary regulators. A significant association (false discovery rate o 5%) stem and progenitor cells),25 in addition to metastatic signatures, Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2742 including LMS-up,16 a signature from low-burden breast cancer Figure 4c), whereas GFP-EVI1 overexpression conferred higher metastatic cells,10 and of breast cancer multipotency promoted by cellular viability in response to everolimus (Figure 4d). Of note, the oncogenic PI3KCA26,27 (Figure 4a). Moreover, a set of 79 levels of pS6 in the GFP-EVI1 overexpression assays were relatively commonly overexpressed genes (40.25 log2; Supplementary low (Figure 4d), which could be due to the lack of full adaptation Table 4) across the in vivo and in vitro models of mTOR inhibitor and/or the need for EVI1 co-factors. resistance showed significant positive co-expression with EVI1 To further validate the direct transcriptional role of EVI1, (Figure 4b). Of note, this set included LEF1, which regulates stem chromatin immunoprecipitation (ChIP) assays of predicted tran- cell maintenance in different contexts and is functionally scriptional targets were performed. This analysis revealed connected to SOXs.29 In addition, this set showed over- increased (one-tailed P-values o0.01) EVI1 binding at the follow- representation of gene products involved in actin-cytoskeleton ing loci with adaptation to everolimus in at least one cell model: remodeling (Supplementary Table 4). FSCN1 and SPARC (from the LMS-up); SCUBE3 and TCF4 (from V EVI1 is essential for hematopoietic stem cell self-renewal30 and $EVI1_02); and RHEB, RPS6KA1 and RAPTOR (from the mTOR its overexpression has been associated with worse recurrence- pathway) (Figure 4e). In addition, RAPTOR and RHEB were found to free, overall and distant metastasis-free survival of ER-negative be overexpressed as a function of everolimus adaptation in MCF7 breast cancer.31 In vitro depletion of EVI1 reduced the levels of and HCC1937 cells, respectively (Figure 4f; RAPTOR showed pS6, particularly in everolimus-adapted HCC1937 cells (60% transitory underexpression in HCC1937), and EVI1 depletion reduction, and MCF7 showed a reduction of 10% in any condition; reduced the expression of both proteins in the everolimus- Figure 4. EVI1 couples stemness, metastatic potential and resistance to mTOR inhibition. (a) TCGA network of significant co-expression (PCC P-values o0.05) between EVI1 and signatures derived from stem cell-like cells and/or metastatic settings (Supplementary Table 1). (b) Distributions of PCCs between EVI1 and the commonly overexpressed 79 genes across the studied models or the complete microarray gene list as background control. The P-value of the Mann–Whitney test for the comparison of the distributions is shown. (c) Reduced pS6 levels with EVI1 depletion in cell models. The quantification of pS6/S6 signal ratios is show at the bottom (relative to siControl). (d) Ectopic overexpression of GFP-EVI1 in MCF7 (left panels) and HCC1937 (right panels) cells provides higher viability upon exposure to everolimus, relative to GFP-only overexpression. Also shown are the western blot results for defined markers across the drug-exposed cell cultures. The quantification of pS6/S6 signal ratios is show at the bottom (relative to TUBA per sample). (e) Increased EVI1 binding at predicted target promoters/gene loci with adaptation to everolimus. The fold changes are relative to the immunoglobulin control and the promoter gene targets are shown in the X axis. (f) Relative overexpression of RAPTOR and/or RHEB with adaptation to everolimus in MCF7 and HCC1937 cells. The quantification is show at the bottom (relative to untreated and TUBA per sample). (g) Relative reduction of RAPTOR and RHEB expression following EVI1 depletion, in particular in the everolimus-adapted setting. Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2743 adapted settings (Figure 4g). However, the expression of FSCN1 functions. The V$EVI1_02 gene set and several other stem cell- was not significantly reduced with EVI1 depletion (Figure 4g) and, like cells and/or metastasis-associated gene signatures were in turn, EVI1 was found to be overexpressed with FSCN1 depletion positively co-expressed with a key regulator of these functions, (Supplementary Figure 8). Therefore, these results may reflect the SOX9 (Figure 5a).32,33 Subsequently, whole-genome EVI1 ChIP data initial response towards mTOR inhibition; in fact, a similar effect corroborated the positive correlation between EVI1 and SOX9 was observed for RAPTOR and RHEB when EVI1 was depleted in binding sites in HCC1937 cells (Supplementary Figure 9). Interest- parental HCC1937 cells (Figure 4g, right panels). ingly, EVI1 ChIP data also showed a positive correlation with SLUG targets34 in both cell models, and with SNAIL targets34 in HCC1937 (Supplementary Figure 9). Notably, MCF7 differentiation to a basal- EVI1 cooperates with SOX9 like phenotype requires SLUG activity.35 Based on the strong association between EVI1 and stem cell-like/ Expanding on the above results, sirolimus-treated ortho- tumor initiation gene expression signatures, we searched for xenografts and everolimus-adapted cells showed increased potential EVI1-transcriptional target genes mediating such SOX9 expression compared with their corresponding controls Figure 5. EVI1 cooperates with SOX9 and regulates its expression. (a) TCGA network of significant co-expression (PCC P-values o 0.05) between EVI1 or SOX9 and signatures derived from stem cell-like cells and/or metastatic settings (Supplementary Table 1). (b) Increased SOX9 expression in ortho-xenograft tumor fronts of mice treated with sirolimus; the results correspond to at least three ortho-xenografts of each group. (c) Increased SOX9 and ALDH1 expression in everolimus-adapted cells. (d) Graph showing the results from the analysis of the complete drug panel for the correlation between IC50 profiles and the expression of the V$EVI1_02 gene set; drugs are ranked according to PCC log P-values. Negative and positive PCCs are indicated with different colors, and the mTOR inhibitors in the panel are denoted. (e) Left panel, unsupervised clustering and correlation analysis of the difference in EVI1 ChIP results at the SOX9 locus between everolimus-adapted and untreated cells. Right panels, results of ChIP assays targeting a predicted EVI1-binding site in the SOX9 promoter (Supplementary Table 5); the input, control immunoglobulin immunoprecipitation (IP), and EVI1-IP results are shown. The control results for the binding site in FSCN1 are also shown. (f) Depletion of EVI1 leads to a reduction of SOX9 expression in three cell conditions (the results correspond to Figure 4c; the ratios are relative to siControl and TUBA per sample). (g) Depletion of EVI1 leads to a reduction of colony-forming capacity. The results of the one-tailed t-test are shown. (h) Depletion of Evi1 impairs the tumorigenic potential of 4T1 cells. The log-rank P-value is shown for the comparison between the shControl and short hairpin RNA (shRNA)-EVI1 #1; note that transduction with shRNA-EVI1 #2 completely impaired tumor formation so a P-value could not be computed (n.a.). Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2744 (Figures 5b and c, respectively; the same antibody did not everolimus (Figures 6e and f), which fully suppressed pS6 signal recognize mouse Sox9). The expression of ALDH1—canonical (Supplementary Figure 11). Collectively, these results indicate that stem/progenitor marker in normal breast tissue and tumors, and transcriptional reprogramming mediated by EVI1-SOX9 is one of also associated with poor prognosis36—was also detected to be the key factors in metastatic resistance to mTOR inhibition. increased with adaptation to everolimus in both MCF7 and HCC1937 cells (Figure 5c). The higher expression at the tumor invasive front is consistent with previous observations of invasive DISCUSSION leader cells showing positivity for basal/stem cell-like markers.10,37 We provide evidence of the association between EVI1-SOX9 In addition to the results from the models, the analysis of data function, mTOR inhibition resistance and metastasis in breast from hundreds of cell lines38 revealed significant positive cancer. We also show that EVI1-SOX9-mediated transcriptional correlations between EVI1 and SOX9 expression, and with EVI1 reprogramming drives the molecular processes that support locus copy number (PCCs 40.20, P-values o 10 − 4). Moreover, in breast cancer tumor initiation features and metastatic potential this data set, both EVI1 and SOX9 expression correlated positively in therapeutic resistance (Figure 7). These data are coherent and (that is, linked to resistance) with the half maximal inhibitory expand on the concept that cancer stem cell-like cell populations concentration (IC50) of temsirolimus (PCCs = 0.22 and 0.24, have high tumor-initiating capacity and are frequently the source respectively, P-values o10 − 4). In fact, a ranking-based analysis of therapy resistance and metastasis.39 Data from hundreds of cell using the V$EVI1_02 gene set as a surrogate of EVI1 activity lines38 suggest that the proposed mechanism is relevant in showed equivalent results, and the second and fourth most settings beyond breast cancer. Importantly, EVI1 maps in a correlated drugs, respectively, were BX-795 (inhibitor of PDK1 genomic region (including PI3KCA and SOX2) whose amplification activity) and rapamycin (Figure 5d). To further assess these is an independent predictor of breast cancer recurrence.40 This findings, an ER-positive and HER2-positive breast cancer cell region is frequently found to be amplified in basal-like and BRCA1- model, BT-474, was exposed to 150 nM of everolimus for mutated breast cancer,21 as well as in non-small cell lung and approximately 100 days and subsequently profiled for gene ovarian cancers.41 In addition, EVI1 amplification is independent of expression changes; the results showed significant associations PIK3CA mutations,41 which further reinforces the link with basal- with EVI1 and SOX9 targets, and with the commonly over- like breast cancer and is consistent with a role in resistance to expressed genes detected across the above in vitro and in vivo allosteric mTOR inhibition. It remains to be determined whether models (Supplementary Figure 10). EVI1 expression is upregulated through genomic amplification Further analysis of the predicted functional relationship and/or whether its function is enhanced by biochemical between EVI1 and SOX9 revealed a positive correlation in the modifications by casein kinase II42 and/or ERK signaling as seen differential (between control and everolimus-adapted) binding of to be activated in our 4T1 assays. EVI1 at the SOX9 locus (Figure 5e, left panel). Next, targeted ChIP Our preclinical findings and mechanistic model are consistent assays confirmed that EVI1 binds at the SOX9 locus in untreated with and expand on recent observations across different HCC1937 and both everolimus-adapted cell models (Figure 5e, neoplastic settings. EVI1 contributes to epithelial-to-mesen- right panel). Thus, depletion of EVI1 reduced the expression of chymal transition (EMT) and invasion of acute myeloid SOX9 in three out of four conditions; however, it remains to be leukemia,43 and influences EMT in ovarian cancer cells,44 whereas determined which co-factor(s) may maintain SOX9 expression at EMT mediates resistance to rapamycin.45 In the latter study, MCF7 normal levels in everolimus-adapted HCC1939 cells (Figure 5f). In cells transfected with constitute active SNAIL showed increased parallel, depletion of EVI1/Evi1 reduced the colony-forming ERK signaling and decreased sensitivity to rapamycin. Thus, our capacity of both models of everolimus adaptation (Figure 5g) study provides a mechanistic explanation for these observations. and impaired in vivo tumorigenic potential of 4T1 cells (Figure 5h). In addition, it has recently been shown that SOX2 and SOX9 Collectively, ChIP, gene/protein expression analyses and in vivo mediated the maintenance of latent metastatic stem cell-like functional assays depict a link between EVI1 and SOX9 in the cells,46 and it was previously demonstrated that SLUG and SOX9 regulation of stem cell-like and tumor initiation features. cooperatively determine mammary stem cell state,32 and that SOX9 function links tumor initiation and invasion.33 Moreover, a stem cell-like cancer phenotype that mediates breast cancer In vivo evaluation of the EVI1, SOX9 and mTOR relationship metastasis is predicted to exhibit abnormal mTORC1 signaling10 The functional cooperation between EVI1 and SOX9 was evaluated and, in turn, mTORC1/2 inhibition promotes stem cell-like in vivo using LM2 and 4T1 cells transduced with a short-hairpin properties and enhanced NOTCH1 activity in TN breast cancer cell RNA scrambled control or directed against EVI1, and with or lines.12 Thus, enhanced mTOR signaling impairs cell differentiation without concomitant overexpression of Sox9 (mouse protein). by potentiating NOTCH1 activity, and this signaling is found to be Thus, EVI1 depletion significantly reduced the capacity to colonize increased in poorly differentiated breast tumors.14 Intriguingly, the lungs, and concurrent Sox9 overexpression partially rescued NOTCH may also regulate SOX9 expression,47 which, in turn, is a metastatic potential (Figures 6a and b). As shown in everolimus- master regulator of stem and progenitor cells.48 In parallel, adapted cell lines, EVI1/Evi1 depletion caused a significant allosteric mTOR inhibition increases the number of tumor- decrease of both SOX9 and pS6 expression in the corresponding initiating cells in a model of liver cancer15 and the metastatic metastasis (Supplementary Figure 11). Concurrent Sox9 over- potential in a model of pancreatic neuroendocrine cancer.49 In this expression recovered pS6 signal in 4T1 but not in LM2 cells scenario, our study proposes that EVI1 and SOX9 functionally (Supplementary Figure 11), which suggest differences in the cooperate to sustain mTORC1 activity, EMT and metastatic precise regulation of mTOR activity between the models. potential, thereby providing new insights into therapeutic In addition to EVI1, depletion of SOX9 in LM2 cells led to a resistance. significant decrease in lung colonization capacity and, conversely, Our findings—particularly those from MCF7 cell assays—may Sox9 overexpression increased this capacity (Figure 6c and have clinical implications for the established use of mTOR Supplementary Figure 12). Depletion of Sox9 in 4T1 cells and inhibitors in endocrine-resistant ER-positive advanced metastatic everolimus treatment of both cell models also reduced lung breast cancer,7 in which resistance to treatment is eventually colonization (Figures 6c and d). In addition, depletion of FSCN1 in reported. Our results indicate that exposure to allosteric mTOR both models also led to a substantial impairment of lung inhibition selects a stem cell-like cancer cell population with colonization (Figures 6e and f). Moreover, a greater effect was metastatic capacity, which may therefore promote disease observed when the animals were simultaneously treated with progression. Although in vivo assays may be warranted to further Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2745 Figure 6. In vivo assessment of the role of EVI1 and SOX9. (a, b) Depletion of EVI1/Evi11 expression (using two different short hairpin RNAs (shRNAs)) in LM2 and 4T1 cells reduced lung colonization, and Sox9 overexpression partially recovered this potential, left panels. (a) Right panels show representative images of lungs and their respective HE staining. (c) Depletion of SOX9 and overexpression of Sox9 reduced and increased, respectively, lung colonization of LM2 cells. Treatment with everolimus of shControl LM2 cells also reduced lung colonization. (d) Depletion of Sox9 or treatment with everolimus of 4T1 cells reduced lung colonization. (e, f) Depletion of FSCN1/Fscn1 expression in LM2 and 4T1 cells reduced lung colonization, and concurrent treatment with everolimus further impaired this potential. assess this observation, and while we cannot rule out that the clinical status). Selection of the highest value for a given case, blindly to its specific population may arise through the acquisition of new status, solved discordant scores. The study was approved by the ethics mutations, it is noteworthy that the Breast Cancer Trials of Oral committee of the MD Anderson Cancer Center and written informed Everolimus-2 (BOLERO-2) study for the efficacy of everolimus plus consent was obtained from all patients. exemestane in endocrine resistance showed similar benefits for patients with or without visceral metastases.50 Nevertheless, full Gene expression analyses and durable pathway inhibition—such as obtained that by the Pre-processed and normalized data of human breast cancer were taken next generation of targeted drugs51—may fully impair metastatic from the corresponding publication16 and from the TCGA repository resistance. (http://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp).52 RNA samples were extracted using TRIzol Reagent (Invitrogen, Karlsruhe, Germany) and RNeasy Kit (Qiagen, Venlo, Netherlands), and quality was evaluated in an MATERIALS AND METHODS Agilent Bioanalyzer (Foster City, CA, USA) 2100. The RNAs were amplified Tissue microarray using the Ribo-SPIA system (NuGEN Technologies Inc., San Carlos, CA, USA) The tissue microarray included 138 infiltrating ductal breast carcinoma and subsequently hybridized on the Human Genome U219 microarray tumors collected at the Department of Pathology of the MD Anderson platform (Affymetrix, Santa Clara, CA, USA; IRB Core Facility, Barcelona, Cancer Center, Madrid (Spain). The patients underwent surgery between Spain). Gene expression data from the ortho-xenograft, and MCF7 and 2003 and 2004, and all tumors were classified as grade 3. According to the HCC1937 cell lines have been deposited under the GEO reference TNM system, 45 tumors belonged to stage I, 48 to stage II and 45 to stage GSE39694. Gene expression data from the 4T1 tumors and BT-474 cells III-IV. The linked data included ER (n = 104), progesterone receptor (n = 127) have been deposited under the GEO references GSE50712 and GSE85801, and epidermal growth factor receptor 2 (HER2; n = 125) status, and CK5 respectively. The GSEA and DAVID (for functional term analyses) tools were expression (n = 128), absence/presence of lymph node metastasis (n = 124), used with standard parameters.53,54 The signature correlations were and absence/presence of distant metastasis (n = 127). The tissue microarray computed by selecting genes with s.d. 41.0 and using the average Z- contained duplicated cases and normal tissue, and the immunohisto- score value per gene set. The quantification of NANOG, OCT4 and SOX2 chemical results were scored independently and blindly (to molecular and gene expression was performed as previously described.55 Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2746 Cell culture The LM2 cell derivative is a lung metastatic sub-line originated from MDA- MB-231 breast cancer cells from the laboratory of Professor Massagué.16 The 4T1 cells derived from a spontaneous BALB/c mouse breast cancer tumor56 and were obtained from the ATCC (Rockville, MD, USA). The LM2 cells were cultured in Dulbecco's modified Eagle's medium (GIBCO, Karlsruhe, Germany) supplemented with 10% fetal bovine serum (GIBCO), 1x L-glutamine (Biowest, Nuaillé, France) and 1% penicillin/streptomycin (Biowest). The 4T1 cells were cultured in Roswell Park Memorial Institute (RPMI)-1640 medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum, 1x L-glutamine and 1% penicillin/streptomycin. All MDA-MB-231 cells/sub-lines were stably transfected with a thymidine kinase and GFP luciferase construct and sorted for GFP expression. The MCF7 and HCC1937 cell lines were obtained from ATCC and cultured in supplemen- ted Dulbecco's modified Eagle's medium and RPMI-1640 medium, respectively. The Matrigel (BD Biosciences) colony formation assays were performed using standard protocols with 5% fetal bovine serum. Everolimus was purchased from Selleck Chemicals (Houston, TX, USA) and LC Laboratories (Woburn, MA, USA). Fluorescence-activated cell sorting was performed using FACS Canto (Becton Dickinson, Franklin Figure 7. Proposed mechanistic model. In untreated cancer, low Lakes, NJ, USA) and Diva software (Becton Dickinson) package, and TSC1/2 expression is associated with enhanced mTORC1 activity and, antibody-based cell labeling was performed as previously described.55 therefore, with a primary metastatic and stemness phenotype. In cancer treated for mTOR inhibition, EVI1-SOX9 become activated (in Western blotting part by overexpression) and positively sustain the following To analyze extracts, cells were lysed in standard 150 mM NaCl buffer features: mTOR signaling (through upregulation of RHEB and supplemented with protease inhibitor cocktail (Roche Molecular Biochem- RAPTOR), metastatic potential (through LMS-up and other signa- icals, Mannheim, Germany) and, in some instances, a phosphatase inhibitor tures) and stemness (through at least SOX9). was added (1 mM NaF, Sigma-Aldrich). Lysates were clarified twice by centrifugation at 13 000 × g and protein concentration was measured using the Bradford method (Bio-Rad, Solna, Sweden). Lysates were Antibodies resolved in sodium dodecyl sulfate–polyacrylamide gel electrophoresis Anti-total and pAKT (#9272 and #9271, respectively, Cell Signaling gels and transferred to Immobilon-P (Merck Millipore, Billerica, MA, USA) or Technology, Danvers, MA, USA; #4060 for immunohistochemistry assays), PVDF membranes (Roche Molecular Biochemicals). Target proteins were identified by detection of horseradish peroxidase-labeled antibody anti-ALDH1 (#611194, BD Biosciences, Oxford, UK), anti-4EBP1 (#9452, Cell complexes with chemiluminescence using the ECL Western Blotting Signaling Technology), anti-p4EBP1 (#2855 and #9451, Cell Signaling Detection Kit (GE Healthcare, Amersham, UK). Technology), anti-ER (#IR151, Dako, Glostrup, Denmark), anti-total and phospho-Thr202/Tyr204 ERK (#4695 and #4376, respectively, Cell Signaling Technology), anti-EVI1 (#2265 and #2593, Cell Signaling Technology; and Lung colonization assays #A301-691A, Bethyl Laboratories, Montgomery, TX, USA), anti-FSCN1 The Animal Care and Use Committee of IRB Barcelona approved the (#SC-56531, Santa Cruz Biotechnology, Dallas, TX, USA), anti-GFP ChIP following animal studies. Female BALB/c nude (MDA-MB- 231 cells) or grade (#ab290, Abcam, Cambridge, UK), anti-GLUT1 (#652, Abcam), anti- BALB/c wild-type mice (4T1 cells) were used. For tail vein injections, cells HER2 (#790-100, Ventana, Tucson, AZ, USA), anti-HMGA1 (#129153, were suspended in 1x phosphate-buffered saline (GIBCO; 200 μl per Abcam), anti-pIGF1R (#39398, Abcam), anti-IRS1 (#2382 and #9451, Cell mouse) and injected into the lateral tail vein of mice using a 26G needle, as Signaling Technology), anti-KI67 (#IR626, Dako), anti-CK19 (#IR615, Dako), previously described.57 Before the injection of cells, mice were anesthe- anti-PR (#IR168, Dako), anti-RAPTOR (#SC-81537, Santa Cruz Biotechnology), tized with ketamine (100 mg/kg body weight) and xylazine (10 mg/kg anti-RHEB (#SC-6341, Santa Cruz Biotechnology), anti-S6 (#SC-74459, Santa body weight), and immediately after injection they were imaged for Cruz Biotechnology), anti-pS6 (#4858, Cell Signaling Technology), anti-S6K luciferase activity by injecting 50 μl of beetle luciferin potassium salt (#9202, Cell Signaling Technology), anti-pS6K (#9205, Cell Signaling (Promega, Madison, WI, USA) at 15 mg/ml. To induce the expression of Technology), anti-SOX9 (#5535, Abcam), anti-total and pSTAT3 (#9132 short hairpin RNA in vivo, doxycycline (1 mg/ml, Sigma-Aldrich) was and #9145, respectively, Cell Signaling Technology), anti-TUBA (#44928, administered ad libitum in drinking water containing 25 mg/ml sucrose Abcam) and anti-VCL (V9131, Sigma-Aldrich, St Louis, MO, USA). The (Sigma-Aldrich). When indicated, DMSO solution (at the same concentra- antibodies used for fluorescence-activated cell sorting were anti-CD24-PE, tion as for the compound test, 5%) or everolimus (5 mg/kg; SC-218452, anti-CD44-APC, anti-CD49f-Alexa-647, and anti-EPCAM-FITC (#555428, Santa Cruz Biotechnology) was administered daily by intraperitoneal 559942, 562473, and 347197, respectively; BD Biosciences). injection. Mice were monitored weekly using IVIS imaging, unless otherwise indicated. Lung tumor development was followed up once a week by bioluminescence imaging of the upper dorsal region that Immunohistochemistry corresponds to lung position. Bioluminescent images were quantified with The assays were performed on serial paraffin sections (3–4 μm thick) using Living Image 2.60.1 software (Perkin-Elmer, Waltham, MA, USA). All values the EnVision (Dako) or Ultraview (Ventana) systems. Antigen retrieval was were normalized to those obtained at day 0. The HE staining of lung performed using citrate- or EDTA-based buffers. Endogenous peroxidase sections scored the lung colonization capacity of 4T1 cells 3 weeks post- was blocked by pre-incubation in a solution of 3% H2O2 and blocking was inoculation. Five sections, separated by 50 μm, per mouse lung were performed in 1X phosphate-buffered saline with 5% goat serum or 1% counted. The average of the total metastatic area normalized to total lung bovine serum albumin and 0.1% Tween 20 (Sigma-Aldrich). In all area was measured. The average total lung metastasis area for all mice was experiments, equivalent sections were processed without incubation with then plotted. The tissue was dissected, fixed in 10% buffered formalin the primary antibody, which did not reveal immunostaining in any case. (Sigma-Aldrich), and embedded in paraffin. Sections (3 μm thick) were Sections were hematoxylin and eosin (HE)-counterstained and examined stained with HE. To analyze the metastatic area, images were taken with a with an Olympus BX51 (Tokyo, Japan) microscope. The immunohisto- scanner, and the area of each metastatic lesion was quantified with the chemistry microscopic images were color deconvoluted and quantitated ImageJ software. Five images per section/animal were evaluated, and the using the regions of interest methodology in ImageJ (http://rsb.info.nih. average area was plotted. The Fiji Trainable Weka segmentation, an ImageJ gov/ij). Quantification of tumor fronts was based on rectangular areas of plugin based on the Weka58 Java machine learning library, was used to 25 μM × 50–300 μM. When quantifying the results from lung metastases, the classify images on the basis of local colorimetric, textural and structural complete metastatic area was considered because the fronts were often features in the neighborhood of each pixel. Images were processed with a difficult to outline histologically. custom macro created at the Microscopy Core Facility of IRB Barcelona. Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2747 The two-tailed non-parametric Mann–Whitney test was used to assess EVI1-binding sites and primers designed for the assays are detailed in significance of the immunohistochemical staining results. Supplementary Table 5. Whole-genome ChIP data were obtained by hybridization to SurePrint G3 Human Promoter 1x1M microarrays (IRB Core Facility) and analyzed by MACS (version 2.0.9).60 The data Ortho-xenograft have been deposited under the GEO reference GSE50905. The complete The patient was a 33-year-old woman with a pathological germline BRCA1 ranking of differential EVI1 binding between adapted and sensitive mutation and diagnosed with breast cancer shortly after pregnancy. At MCF7 or HCC1937 cells was used as input for the GSEA of transcription diagnosis she presented a locally advanced TN ductal infiltrating factor targets. carcinoma of the breast (T4) with involvement of ipsilateral nodes (N2) and lung metastasis. Primary systemic chemotherapy was initiated with TAC regimen for four cycles, followed by mastectomy to prevent local Gene expression alterations complications because of extensive breast involvement. Following surgery, Stable LM2 and 4T1 cell lines expressing short hairpin RNAs were the patient received further chemotherapy with the same regimen. The generated as described previously.57 The shFSCN1/Fscn1 (that is, targeting patient was diagnosed with brain metastases shortly after and died both human and mouse gene expression) #1 and shEVI1/Evi1 #2 were 8 months post-diagnosis as a result of disease progression. Mutational encoded in lentiviral vectors (inducible pTRIPZ lentiviral short hairpin RNAs, analysis of BRCA1 was carried out by the Molecular Diagnostics Unit GE Dharmacon, Lafayette, CO, USA). The short hairpins were induced by (Catalan Institute of Oncology, Barcelona, Spain) following standards for 1 μg/ml doxycycline for 72 h. The shControl and shEVI1/Evi1 #1 were genetic testing and pathological determination. The patient provided encoded in a retrovirus pGFP-V-RS (OriGene, Rockville, MD, USA). The written informed consent, and the study was approved by the IDIBELL shFSCN1/Fscn1 #1 was encoded in a pSUPER (Addgene, Cambridge, MA, Ethics Committee. Female athymic (nu/nu) mice (Harlan, Harlan Labora- USA) vector. The shSox9 (against mouse gene sequence) and shSOX9 tories, Barcelona, Spain) between 4 and 6 weeks of age were used for (human) were obtained from the MISSION library (SHCLND-NM_011448 engraftment. The orthotopic model developed histologically detectable and SHCLND-NM_000346, respectively; Sigma-Aldrich). An additional lung metastases in a period of approximately 50 days after engraftment. shSOX9 was obtained from Addgene, catalog #40644. For Sox9 over- The protocol was reviewed and approved by the IDIBELL Animal Care and expression, the corresponding coding sequence was cloned into a Use Committee. A daily oral treatment with sirolimus (Rapamune) or lentiviral pWXL vector. Stable cell lines expressing the various constructs control solution (DMSO, Sigma-Aldrich) was applied. described above were generated under puromycin selection for 48 h. The siRNA against EVI1 expression was an ON-TARGETplus SMARTpool (L-006530-02-0010, Dharmacon). The following primer sequences were Exome analysis used to assess gene expression changes in real-time (using SYBR Green, The National Centre for Genomic Analysis (CNAG) carried out exome Applied Biosystems) polymerase chain reaction assays: EVI1, 5′-CATT sequencing. Sequence capture and amplification was performed using GGGAACAGCAACCAT-3′ and 5′-GGTCACCAAAGCCTTTTCAT-3′; Evi1, 5′-CAC Agilent (Agilent Technologies, Palo Alto, CA, USA) SureSelect Human All AGAAAGTCCAAATCACAGG-3′ and 5′-GCCACACGTTGGAGGAAC-3′; Sox9, Exon kit (Agilent) according to the manufacturer’s instructions. Paired-end 5′-GTACCCGCATCTGCACAAC-3′ and 5′-CTCCTCCACGAAGGGTCTCT-3′; ACTB, sequencing was performed on a HiSeq2000 instrument (Illumina, San 5′-GGAGTGGGTGGAGGCAG-3′ and 5′-AACTAAGGTGTGCACTTTTGTTC-3′; and Diego, CA, USA) using 76-base reads. Reads were aligned to the reference mL32, 5′-GAAACTGGCGGAAACCCA-3′ and 5′-GGATCTGGCCCTTGAACCTT-3′. genome (GRCh37) and BAM files were generated using SAMtools. Duplicates were removed using SAMtools and custom scripts, and single-nucleotide variant calling was performed using a combination of Genomics of drug sensitivity data analyses SAMtools and Sidrón algorithms as described previously.59 The reads were For the correlation analysis between the basal expression of EVI1, its first aligned to mouse genome (mm9), and those read-pairs that did not predicted target genes and the drug responses across cancer cell lines, align to mouse were then aligned to the human genome following the data were downloaded from the GDS project (web-release April 2012).38 same pipeline as above. Only mismatch variants were taken into account This data set included IC50 values for 131 drugs that were assessed in a and small insertions and deletions were not counted. Common variants, panel of 638 human cancer cell lines. The basal gene expression data were defined as those present in dbSNP135 with a minor allele frequency 4 1%, downloaded from ArrayExpress reference E-MTAB-783. Non-annotated were filtered out. probes were removed and expression values were averaged when multiple probes mapped to the same gene. Correlation scores and P-values were computed using the PCC. The EVI1 target set included 20 genes that were 4T1 tumors represented by at least one microarray probe (Supplementary Table 6). The The animal studies were conducted using protocols that had undergone extent of the basal expression of the EVI1 targets was quantified using an appropriate review and approval at the New York University School of enrichment score (ES) computed with a Matlab implementation of the Medicine. Balb/C mice were injected subcutaneously with 5 × 104 4T1 cells, GSEA algorithm. To estimate ES significance, a null model was created by measured for tumor size at day 10, and randomly organized in two generating 10 000 random gene sets (of the same size as the EVI1 target equivalent groups that were treated with DMSO solution (the same set) and used to query the data set through GSEA. Next, two inverse concentration as for the compound test) or everolimus (5 mg/kg; Gaussian distributions (for positive and negative ES values) were fitted on SC-218452, Santa Cruz Biotechnology) daily by intraperitoneal injection. the resulting empirical distribution and used to compute P-values. The Tumors were excised at day 23 and processed. Half of each tissue sample correlations between IC50 profiles and ESs were computed by considering was used for immunohistochemistry and half for gene expression only cell lines whose basal expression profile yielded a significant ES microarray analysis. For the tumorigenicity assays, 250 000 4T1 cells were (Po 0.05), according to the null model. The enrichment P-values of mTOR injected at the orthotopic site, mixed with growth factor-reduced Matrigel inhibitors among drugs whose IC50 profile was anti-correlated with the (BD Biosciences) before inoculation (1:1). Once palpable, tumors were EVI1 target ES were computed using Fisher’s exact test and considering the measured with a digital caliper, and the tumor volume was calculated. The total set of 131 drugs. ethics committee of the CIC bioGUNE approved these assays. ChIP assays CONFLICT OF INTEREST Assays were prepared using 107 cells of each cell line per condition. The authors declare no conflict of interest. Chromatin was fragmented by sonication (Bioruptor, Diagenode, Denville, NJ, USA) for 30 min (30-s pulses, 30-s pauses) and assays were carried out following the manufacturer’s protocol (kch-mahigh-A16, HighCell# ChIP Kit, ACKNOWLEDGEMENTS Diagenode), using anti-EVI1 (#2593, Cell Signaling Technology) or an equal We thank Ander Urruticoechea, Gabriel Capellá and George Thomas for helpful amount of IgG isotype as negative control (#2729, Cell Signaling discussion; Ana Isabel Extremera, Antoni Xaubet and Julio Ancochea for adminis- Technology). The amount of DNA was analyzed by real-time polymerase trative support; and Sebastien Tosi from the Advanced Digital Microscopy Facility at chain reactions using SYBR Green-based assays (Applied Biosystems, Life IRB Barcelona for image analysis. We also thank Xiyun Deng, Yanna Cao, Tien C Ko, Technologies, Foster City, CA, USA). The results were calculated using the Yi Zhang (The University of Texas Health Science Centre at Houston) and Adrian W ΔCt method. A genomic region of the GAPDH gene was used as negative Moore (RIKEN Brain Science Institute) for their support in evaluating antibodies control (Diagenode). The corresponding human genome coordinates, against EVI1. This study was supported by the following bodies and grants: Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2748 the Scientific Foundation ‘Asociación Española Contra el Cáncer’ (AECC, Stable 15 Yang Z, Zhang L, Ma A, Liu L, Li J, Gu J et al. Transient mTOR inhibition facilitates Coordinated Group, Hereditary Cancer); the BBVA Foundation; the Eugenio Rodríguez continuous growth of liver tumors by modulating the maintenance of CD133+ Pascual Foundation grant 2012; Generalitat de Catalunya AGAUR SGR 2012 grants cell populations. PLoS One 2011; 6: e28405. 283, 290 and 312, and SGR 2014 grants 364, 530, and 535; Spanish Ministry of Health 16 Minn AJ, Gupta GP, Siegel PM, Bos PD, Shu W, Giri DD et al. Genes that mediate ISCIII FIS grants PI10/00057, PI10/00222, PI10/01422, PI12/01528, PI13/00132, and breast cancer metastasis to lung. Nature 2005; 436: 518–524. PI14/00336. ISCIII RTICC grants RD06/0020/1051, RD12/0036/0007, RD12/0036/0008 17 Issa A, Gill JW, Heideman MR, Sahin O, Wiemann S, Dey JH et al. Combinatorial and RD12/0036/0063; Spanish Ministry of Science and Innovation, ‘Fondo Europeo de targeting of FGF and ErbB receptors blocks growth and metastatic spread of Desarrollo Regional (FEDER), una manera de hacer Europa’, MINECO grants SAF2010- breast cancer models. Breast Cancer Res 2013; 15: R8. 20203 and SAF2013-46196; and the Telemaraton 2014 ‘Todos Somos Raros, Todos 18 Chen L, Yang S, Jakoncic J, Zhang JJ, Huang XY. Migrastatin analogues target Somos Únicos’ grant P35. EJA was supported by ‘la Caixa’ PhD fellowship program, F fascin to block tumour metastasis. Nature 2010; 464: 1062–1066. Iorio was supported by a fellowship from the EMBL-EBI and the Wellcome Trust 19 Cariati M, Naderi A, Brown JP, Smalley MJ, Pinder SE, Caldas C et al. Alpha-6 Sanger Institute Postdoctoral (ESPOD) program, and NL-B, ME and RRG were integrin is necessary for the tumourigenicity of a stem cell-like subpopulation supported by ICREA. No funding bodies had any role in study design, data collection within the MCF7 breast cancer cell line. Int J Cancer 2008; 122: 298–304. and analysis, decision to publish, or preparation of the manuscript, either in the 20 Piva M, Domenici G, Iriondo O, Rabano M, Simoes BM, Comaills V et al. Sox2 submission form or the text of the manuscript. promotes tamoxifen resistance in breast cancer cells. EMBO Mol Med 2014; 6: 66–79. 21 TCGA. Comprehensive molecular portraits of human breast tumours. Nature 2012; 490: 61–70. AUTHOR CONTRIBUTIONS 22 Wong DJ, Liu H, Ridky TW, Cassarino D, Segal E, Chang HY. Module map of stem RRG and MAP: performed the design and interpretation of experiments, and cell genes guides creation of epithelial cancer stem cells. Cell Stem Cell 2008; 2: wrote the manuscript; FM, EJA, HA, GRG, J Boni, XP, CH, AL, HS, MS, LG-B, XS, 333–344. J Cerón and JG-M: molecular and cell biology experiments; FM, EJA, HA, SD, 23 Ben-Porath I, Thomson MW, Carey VJ, Ge R, Bell GW, Regev A et al. An embryonic MHB-H and A Villanueva: in vivo studies; FM, HA, MM and MDO: chromatin stem cell-like gene expression signature in poorly differentiated aggressive immunoprecipitation assays; JS-M, FI, AI, DC, NB, LP, AG, NL-B, LG-A and JS-R: human tumors. Nat Genet 2008; 40: 499–507. 24 Kim J, Woo AJ, Chu J, Snow JW, Fujiwara Y, Kim CG et al. A Myc network accounts bioinformatic analyses; FM, HA, J Boni, S Puertas, NG, VH, MM-I and AF: for similarities between embryonic stem and cancer cell transcription programs. immunohistochemistry experiments; AR-S, A Martínez, MPB, M Gil, CF, AF, IM, Cell 2010; 143: 313–324. S Pernas, MJP, XA, MAS, RB, EC, SM, JV, A Velasco, XM-G, MAQ, AS and GM-B: 25 Lim E, Wu D, Pal B, Bouras T, Asselin-Labat ML, Vaillant F et al. Transcriptome tissue microarray studies; RV-M and XSP: exome analyses; A Petit, A Vidal, IC, analyses of mouse and human mammary cell subpopulations reveal multiple TS and GV: pathological evaluations; AMS, VW, MPL, M Nellist, JVS-M, ME, MJ, conserved genes and pathways. Breast Cancer Res 2010; 12: R21. LB and JWMM: genetic analyses; FM, HA, J Boni, EG-S and A Cordero: three- 26 Koren S, Reavie L, Couto JP, De Silva D, Stadler MB, Roloff T et al. PIK3CA(H1047R) dimensional cell culture experiments; JH-L, SRC and J Cortés: breast cancer induces multipotency and multi-lineage mammary tumours. Nature 2015; 525: 114–118. tumor analyses; JL, FC, IB, A Perkins, J Brunet, FV, OC, M Graupera, NM-M, 27 Van Keymeulen A, Lee MY, Ousset M, Brohee S, Rorive S, Giraddi RR et al. Reac- A Matheu, A Carracedo, TFG, EEWC, MS-C, M Nanjundan and CL: experimental tivation of multipotency by oncogenic PIK3CA induces breast tumour hetero- design and interpretation. geneity. Nature 2015; 525: 119–123. 28 Huang R, Huang D, Dai W, Yang F. Overexpression of HMGA1 correlates with the malignant status and prognosis of breast cancer. Mol Cell Biochem 2015; 404: REFERENCES 251–257. 1 Zoncu R, Efeyan A, Sabatini DM. mTOR: from growth signal integration to cancer, 29 Kormish JD, Sinner D, Zorn AM. Interactions between SOX factors and Wnt/beta- diabetes and ageing. Nat Rev Mol Cell Biol 2011; 12: 21–35. catenin signaling in development and disease. Dev Dyn 2010; 239: 56–68. 2 Courtney KD, Corcoran RB, Engelman JA. The PI3K pathway as drug target in 30 Kataoka K, Sato T, Yoshimi A, Goyama S, Tsuruta T, Kobayashi H et al. Evi1 is human cancer. J Clin Oncol 2010; 28: 1075–1083. essential for hematopoietic stem cell self-renewal, and its expression marks 3 Hsieh AC, Liu Y, Edlind MP, Ingolia NT, Janes MR, Sher A et al. The translational hematopoietic cells with long-term multilineage repopulating activity. J Exp Med landscape of mTOR signalling steers cancer initiation and metastasis. Nature 2012; 2011; 208: 2403–2416. 485: 55–61. 31 Patel JB, Appaiah HN, Burnett RM, Bhat-Nakshatri P, Wang G, Mehta R et al. 4 Crino PB, Nathanson KL, Henske EP. The tuberous sclerosis complex. N Engl J Med Control of EVI-1 oncogene expression in metastatic breast cancer cells through 2006; 355: 1345–1356. microRNA miR-22. Oncogene 2011; 30: 1290–1301. 5 Nasr Z, Robert F, Porco Jr JA, Muller WJ, Pelletier J. eIF4F suppression in breast 32 Guo W, Keckesova Z, Donaher JL, Shibue T, Tischler V, Reinhardt F et al. Slug and cancer affects maintenance and progression. Oncogene 2013; 32: 861–871. Sox9 cooperatively determine the mammary stem cell state. Cell 2012; 148: 6 Fruman DA, Rommel C. PI3K and cancer: lessons, challenges and opportunities. 1015–1028. Nat Rev Drug Discov 2014; 13: 140–156. 33 Larsimont JC, Youssef KK, Sanchez-Danes A, Sukumaran V, Defrance M, Delatte B 7 Baselga J, Campone M, Piccart M, Burris HA 3rd, Rugo HS, Sahmoud T et al. et al. Sox9 controls self-renewal of oncogene targeted cells and links tumor Everolimus in postmenopausal hormone-receptor-positive advanced breast cancer. initiation and invasion. Cell Stem Cell 2015; 17: 60–73. N Engl J Med 2012; 366: 520–529. 34 Ye X, Tam WL, Shibue T, Kaygusuz Y, Reinhardt F, Ng Eaton E et al. Distinct EMT 8 Chandarlapaty S. Negative feedback and adaptive resistance to the targeted programs control normal mammary stem cells and tumour-initiating cells. Nature therapy of cancer. Cancer Discov 2012; 2: 311–319. 2015; 525: 256–260. 9 Markman B, Dienstmann R, Tabernero J. Targeting the PI3K/Akt/mTOR pathway-- 35 Sflomos G, Dormoy V, Metsalu T, Jeitziner R, Battista L, Scabia V et al. A preclinical beyond rapalogs. Oncotarget 2010; 1: 530–543. model for ERalpha-positive breast cancer points to the epithelial microenviron- 10 Lawson DA, Bhakta NR, Kessenbrock K, Prummel KD, Yu Y, Takai K et al. Single-cell ment as determinant of luminal phenotype and hormone response. Cancer Cell analysis reveals a stem-cell program in human metastatic breast cancer cells. 2016; 29: 407–422. Nature 2015; 526: 131–135. 36 Ginestier C, Hur MH, Charafe-Jauffret E, Monville F, Dutcher J, Brown M et al. 11 Ruiz de Garibay G, Herranz C, Llorente A, Boni J, Serra-Musach J, Mateo F et al. ALDH1 is a marker of normal and malignant human mammary stem cells and a Lymphangioleiomyomatosis biomarkers linked to lung metastatic potential and predictor of poor clinical outcome. Cell Stem Cell 2007; 1: 555–567. cell stemness. PLoS One 2015; 10: e0132546. 37 Cheung KJ, Gabrielson E, Werb Z, Ewald AJ. Collective invasion in breast cancer 12 Bhola NE, Jansen VM, Koch JP, Li H, Formisano L, Williams JA et al. Treatment of requires a conserved basal epithelial program. Cell 2013; 155: 1639–1651. triple-negative breast cancer with TORC1/2 inhibitors sustains a drug-resistant 38 Garnett MJ, Edelman EJ, Heidorn SJ, Greenman CD, Dastur A, Lau KW et al. and Notch-dependent cancer stem cell population. Cancer Res 2016; 76: Systematic identification of genomic markers of drug sensitivity in cancer cells. 440–452. Nature 2012; 483: 570–575. 13 Zhou J, Wulfkuhle J, Zhang H, Gu P, Yang Y, Deng J et al. Activation of the 39 Clevers H. The cancer stem cell: premises, promises and challenges. Nat Med 2011; PTEN/mTOR/STAT3 pathway in breast cancer stem-like cells is required for 17: 313–319. viability and maintenance. Proc Natl Acad Sci USA 2007; 104: 16158–16163. 40 Janssen EA, Baak JP, Guervos MA, van Diest PJ, Jiwa M, Hermsen MA. In lymph 14 Ma J, Meng Y, Kwiatkowski DJ, Chen X, Peng H, Sun Q et al. Mammalian target of node-negative invasive breast carcinomas, specific chromosomal aberrations are rapamycin regulates murine and human cell differentiation through STAT3/p63/ strongly associated with high mitotic activity and predict outcome more accurately Jagged/Notch cascade. J Clin Invest 2010; 120: 103–114. than grade, tumour diameter, and oestrogen receptor. J Pathol 2003; 201: 555–561. Oncogene (2017) 2737 – 2749 Resistance to mTOR inhibition F Mateo et al 2749 41 Hagerstrand D, Tong A, Schumacher SE, Ilic N, Shen RR, Cheung HW et al. 53 Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA Systematic interrogation of 3q26 identifies TLOC1 and SKIL as cancer drivers. et al. Gene set enrichment analysis: a knowledge-based approach for inter- Cancer Discov 2013; 3: 1044–1057. preting genome-wide expression profiles. Proc Natl Acad Sci USA 2005; 102: 42 Bard-Chapeau EA, Gunaratne J, Kumar P, Chua BQ, Muller J, Bard FA et al. EVI1 15545–15550. oncoprotein interacts with a large and complex network of proteins and inte- 54 Huang DW, Sherman BT, Tan Q, Collins JR, Alvord WG, Roayaei J et al. The DAVID grates signals through protein phosphorylation. Proc Natl Acad Sci USA 2013; 110: gene functional classification tool: a novel biological module-centric algorithm to E2885–E2894. functionally analyze large gene lists. Genome Biol 2007; 8: R183. 43 Stavropoulou V, Kaspar S, Brault L, Sanders MA, Juge S, Morettini S et al. MLL-AF9 55 Palafox M, Ferrer I, Pellegrini P, Vila S, Hernandez-Ortega S, Urruticoechea A et al. expression in hematopoietic stem cells drives a highly invasive AML expressing RANK induces epithelial-mesenchymal transition and stemness in human mam- EMT-related genes linked to poor outcome. Cancer Cell 2016; 30: 43–58. mary epithelial cells and promotes tumorigenesis and metastasis. Cancer Res 44 Dutta P, Bui T, Bauckman KA, Keyomarsi K, Mills GB, Nanjundan M. EVI1 splice 2012; 72: 2879–2888. variants modulate functional responses in ovarian cancer cells. Mol Oncol 2013; 7: 56 Aslakson CJ, Miller FR. Selective events in the metastatic process defined by 647–668. analysis of the sequential dissemination of subpopulations of a mouse 45 Holder AM, Akcakanat A, Adkins F, Evans K, Chen H, Wei C et al. Epithelial to mammary tumor. Cancer Res 1992; 52: 1399–1405. mesenchymal transition is associated with rapamycin resistance. Oncotarget 2015; 57 Morales M, Arenas EJ, Urosevic J, Guiu M, Fernandez E, Planet E et al. RARRES3 6: 19500–19513. suppresses breast cancer lung metastasis by regulating adhesion and differ- entiation. EMBO Mol Med 2014; 6: 865–881. 46 Malladi S, Macalinao DG, Jin X, He L, Basnet H, Zou Y et al. Metastatic latency and 58 Witten IH, Frank E, Trigg L, Hall M, Holmes G, Cunningham SJ. Weka: practical immune evasion through autocrine inhibition of WNT. Cell 2016; 165: 45–60. machine learning tools and techniques with Java implementations. Proceedings of 47 Chao CH, Chang CC, Wu MJ, Ko HW, Wang D, Hung MC et al. MicroRNA-205 the ICONIP/ANZIIS/ANNES’99 Workshop on Emerging Knowledge Engineering and signaling regulates mammary stem cell fate and tumorigenesis. J Clin Invest 2014; Connectionist-Based Information Systems 1999, pp 192–196. 124: 3093–3106. 59 Puente XS, Pinyol M, Quesada V, Conde L, Ordonez GR, Villamor N et al. Whole- 48 Malhotra GK, Zhao X, Edwards E, Kopp JL, Naramura M, Sander M et al. The role of genome sequencing identifies recurrent mutations in chronic lymphocytic leu- Sox9 in mouse mammary gland development and maintenance of mammary kaemia. Nature 2011; 475: 101–105. stem and luminal progenitor cells. BMC Dev Biol 2014; 14: 47. 60 Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE et al. Model- 49 Pool SE, Bison S, Koelewijn SJ, van der Graaf LM, Melis M, Krenning EP et al. based analysis of ChIP-Seq (MACS). Genome Biol 2008; 9: R137. mTOR inhibitor RAD001 promotes metastasis in a rat model of pancreatic neuroendocrine cancer. Cancer Res 2013; 73: 12–18. 50 Campone M, Bachelot T, Gnant M, Deleu I, Rugo HS, Pistilli B et al. Effect of visceral This work is licensed under a Creative Commons Attribution- metastases on the efficacy and safety of everolimus in postmenopausal women NonCommercial-NoDerivs 4.0 International License. The images or with advanced breast cancer: subgroup analysis from the BOLERO-2 study. Eur J other third party material in this article are included in the article’s Creative Commons Cancer 2013; 49: 2621–2632. license, unless indicated otherwise in the credit line; if the material is not included under 51 Rodrik-Outmezguine VS, Okaniwa M, Yao Z, Novotny CJ, McWhirter C, Banaji A the Creative Commons license, users will need to obtain permission from the license et al. Overcoming mTOR resistance mutations with a new-generation mTOR holder to reproduce the material. To view a copy of this license, visit http:// inhibitor. Nature 2016; 534: 272–276. creativecommons.org/licenses/by-nc-nd/4.0/ 52 Koboldt DC, Fulton RS, McLellan MD, Schmidt H, Kalicki-Veizer J, McMichael JF et al. Comprehensive molecular portraits of human breast tumours. Nature 2012; 490: 61–70. © The Author(s) 2017 Supplementary Information accompanies this paper on the Oncogene website (http://www.nature.com/onc) Oncogene (2017) 2737 – 2749