MYC-driven epigenetic reprogramming favors the onset of tumorigenesis by inducing a stem cell-like state

Breast cancer consists of highly heterogeneous tumors, whose cell of origin and driver oncogenes are difficult to be uniquely defined. Here we report that MYC acts as tumor reprogramming factor in mammary epithelial cells by inducing an alternative epigenetic program, which triggers loss of cell identity and activation of oncogenic pathways. Overexpression of MYC induces transcriptional repression of lineage-specifying transcription factors, causing decommissioning of luminal-specific enhancers. MYC-driven dedifferentiation supports the onset of a stem cell-like state by inducing the activation of de novo enhancers, which drive the transcriptional activation of oncogenic pathways. Furthermore, we demonstrate that the MYC-driven epigenetic reprogramming favors the formation and maintenance of tumor-initiating cells endowed with metastatic capacity. This study supports the notion that MYC-driven tumor initiation relies on cell reprogramming, which is mediated by the activation of MYC-dependent oncogenic enhancers, thus establishing a therapeutic rational for treating basal-like breast cancers.


Introduction
Tumorigenesis can be ascribed to a succession of genetic and epigenetic alterations which turn in heritable changes in gene expression programs, ultimately leading to the formation of a cell population characterized by functional and phenotypic heterogeneity 1 2 . Cell transformation frequently involves activation of developmental signaling programs, which endow cells with unlimited self-renewal potential and aberrant differentiation capability 3 . Somatic stem cells have been considered putative candidates for targets of transformation because of their inherent self-renewing capacity and their longevity, which would allow the acquisition of the combination of genetic and epigenetic aberrations sufficient for cell transformation 4 . Nevertheless, recent studies demonstrated that, upon oncogenic alterations, progenitors or committed cells can serve as tumor initiating cells (TICs) by dedifferentiating and re-acquiring stem cell-like traits 5 6 7 . In the context of mammary gland tumorigenesis, it has been demonstrated that the BRCA1 basal-like breast cancer subtype may arise from luminal progenitor cells 8 9 . More recently, it has been shown that expression of oncogenic PIK3CA H1047R in oncogene-driven normal lineage-restricted mouse mammary cells causes cell dedifferentiation and development of multi-lineage mammary tumors 10 11 . Although these findings highlighted a functional role for oncogene-driven cell dedifferentiation in tumor initiation, the molecular mechanisms underlying cell reprogramming are incompletely understood.
Cell reprogramming requires overcoming those epigenetic barriers which are involved in maintaining cell-specific transcriptional programs, thereby preserving cell identity 12 13 14 . The activation of a specific repertoire of cis-regulatory elements -enhancers-is critical for cell specification. Cooperative binding of lineage-determining (LDTF) and signal-dependent (SDTF) transcription factors dictates the spatio-temporal pattern of gene expression 15 . Enhancers are characterized by accessible chromatin, marked by the deposition of H3K4me1, and their activation is associated with an increment of H3K27 acetylation 16 . Given their pivotal role in the determination of cell identity, decommissioning of active enhancers represents a critical step towards cell reprogramming 17 . Of importance, evidence indicates that dis-regulation of chromatin players responsible for enhancer regulation could favor tumorigenesis by driving the aberrant activation of oncogenic transcriptional programs 18 19 20 21 22 .
Among the transcription factors (TFs) with a documented function in somatic cell reprogramming 23 , the proto-oncogene MYC has a pivotal role in growth control, differentiation and apoptosis and its expression level is tightly regulated in physiological conditions 24 . In breast cancer, MYC deregulation has been associated with up to 40% of tumors, and its hyper activation is a hallmark of the basal-like subtype 25 26 24 . Despite MYC proven oncogenic potential and its known function in the maintenance of selfrenewing capacity and pluripotency 27 28 , a causal link between MYC role as reprogramming factor and its tumorigenic effects has not been investigated.
Here we demonstrate that MYC acts as an oncogenic reprogramming factor by inducing cell plasticity that predisposes mammary luminal epithelial cells to acquire basal/stem cell-like properties and to onset of tumorigenesis, giving rise to tumor initiating cells (TICs) endowed with long-term tumorigenic capacity and metastatic potential.

MYC alters luminal epithelial cell identity by affecting their transcriptional program
In order to evaluate the role of MYC in perturbing cell identity of somatic cells, we transduced hTERT-immortalized human mammary epithelial cells (thereafter named IMEC) with a retroviral vector expressing low levels of the exogenous c-Myc (Fig. 1a).
MYC over-expression induced alteration of the epithelial morphology with cells loosing polarity and adhesion, growing in semi-adherent condition and forming spheroids (Fig.   1b). Importantly, upon MYC over-expression we observed a similar phenotype in the luminal breast cancer cell lines MCF7, T47D and ZR751 (Supplementary Fig. 1a-b). Of note, the observed phenotype could not solely rely on induction of EMT, as we did not detect induction of EMT-related TFs ( Supplementary Fig. 1c). In addition, the modest level of MYC over-expression did not cause major changes in the cell cycle profile or

ML-specific TFs are down-regulated in response to MYC over-expression
On the basis of these results, we asked whether the expression of lineage-specific transcription factors (LSTFs) was perturbed in consequence of MYC overexpression. We found that IMEC-MYC down-regulated ML-specific TFs while they did not show a global and consistent modulation of the expression level of LP-specific regulators (Fig. 1d) 31 32 33 . Importantly, genes whose expression is dependent on luminal-specific TFs binding on their cognate enhancers 34 resulted down-regulated in IMEC-MYC (Fig. 1e). We therefore focused on GATA3 and ESR1 TFs, two master regulators of mammary gland morphogenesis and luminal cell differentiation 35 36 . We confirmed that their transcriptional down-regulation was not restricted to IMEC, (Figure 1f) as the same pattern was induced by MYC over-expression in different luminal breast cancer cell lines ( Supplementary Fig. 2b). Moreover, knocking-down the exogenous MYC was sufficient to revert the observed down-modulation of these master regulators of mammary epithelial cells ( Supplementary Fig. 2c-e). We then asked whether GATA3 and ESR1 downregulation could be mediated by MYC binding to their cis-regulatory elements. Upon MYC over-expression, we measured a concomitant increase of MYC association and reduction of active histone marks on GATA3 and ESR1 regulatory elements (Fig. 1g).
Considering that often MYC deregulation causes transcription repression of its targets by antagonizing the transcriptional activity of MIZ1 37 38 , we determined whether the MYC-dependent down-regulation of GATA3 and ESR1 could be mediated by MIZ1 binding. ChIP assay showed that MIZ1 associated on the analyzed cis-regulatory elements (Fig. 1g) and its knock-down reverted the MYC-driven transcription repression of GATA3 and ESR1 (Supplementary Fig. 2f-g).
To establish the pathological relevance of the anti-correlation between MYC overexpression and ESR1/GATA3 down-regulation, we assessed the expression level of these ML-specific TFs in large cohorts of breast cancer samples. Analysis of different datasets of breast cancer patients 25 39 showed that MYC over-expression anti-correlated with ESR1 and GATA3 transcript levels ( Supplementary Fig. 2h-j). Moreover, querying the proteome of genome-associated TCGA tumor samples 40 showed that the protein abundance of both ESR1 and GATA3 decreased in those breast cancers with augmented level of MYC (Fig. 1h). Together, these data indicate that MYC over-expression induced dedifferentiation of luminal cells by down-regulating the expression of lineage-specific TFs, thereby supporting the reprogramming towards a progenitor-like state (Fig. 1i).

Sustained MYC over-expression confers stem cell-like traits
On the basis of the observed MYC-induced cell reprogramming, we asked whether MYC over-expression could enrich for cells with functional stem cell-like properties. We therefore measured the ability of IMEC WT and -MYC to grow for subsequent passages in low adherence conditions as mammospheres 41 . While WT cells formed mammospheres with low efficiency and did not proliferate beyond the second passage, cells over-expressing MYC showed enhanced sphere formation efficiency (SFE) (Fig. 2ac). A similar increment in mammospheres formation was measured in luminal breast cancer cell lines upon MYC over-expression ( Supplementary Fig. 3a-b). MYC sustained the propagation of mammospheres for several passages, indicating acquisition of longterm self-renewal capacity (Fig. 2b-c). Furthermore, IMEC-MYC mammospheres showed enrichment for cells expressing ALDH1, a distinctive marker of mammary stem cells 42 (Fig. 2d). Of importance the observed phenotype was a MYC-dependent effect, as IMEC expressing other oncogenic hits showed reduced long-term capacity to propagate as mammospheres ( Supplementary Fig. 3c-f). To quantify the relative enrichment for cells endowed with self-renewing capacity, we performed single cell clonogenic assay. The obtained result indicated that IMEC WT could not give rise to any single cell-derived clone, while MYC over-expression was associated with the highest clonogenic potential (Fig. 2e). Moreover, single cell-derived primary spheres (named M1) were further enriched in cells with self-renewing capacity, showing higher SFE in respect to the parental heterogeneous population (Fig. 2f). Of importance, the measured enrichment of SFE was not due to clonal selection as independent single-cell isolated clones gave rise to similar increment in mammospheres formation ( Supplementary Fig. 3g). Accordingly, knock-down of the exogenous MYC in mammospheres impaired the measured clonogenic potential ( Supplementary Fig. 3h-i). Finally, we showed that under differentiation conditions, single cell-derived mammopsheres expressed luminal (CK8 and ER-α) and myoepithelial (CK14 and α-SMA) markers, indicative of enrichment for stem cell-like cells endowed with multipotency (Fig. 2g).
To investigate whether MYC supported the activation of a stem cell-like transcriptional program, we profiled gene expression of single cell-derived secondary mammospheres (clone M2 #1, thereafter named M2), determining differentially expressed genes in respect to IMEC-MYC (Fig. 2h). GO analyses showed that mammospheres were characterized by further up-regulation of genes involved in metabolic pathways and down-regulation of genes involved in developmental processes (Fig. 2h- Fig. 4f-h), thus arguing against clonal selection. Collectively, the above data suggest that MYC over-expression in luminal cells favor the onset of stem cell-like traits, such as sustained self-renewing capacity and re-activation of a pluripotencyassociated transcriptional program (Fig. 2k).

MYC induces an alternative epigenetic program in mammary epithelial cells
To gain insights into the mechanisms through which MYC induces cellular reprogramming, we performed ChIP-seq analyses to profile chromatin modifications and the binding of MYC in IMEC WT, -MYC and mammospheres (Fig. 3). Considering that nearly 50% of MYC binding sites localized at promoters (Fig. 3a), we analyzed the transcriptional effects of increasing the MYC levels on these loci, in response to its overexpression. By ranking MYC-bound genes on the basis of their gene expression pattern, we defined two distinct subsets of targets whose expression augmented or decreased in response to MYC association, respectively ( Fig. 3b and Supplementary Fig. 5a-b).
Comparative analyses between these two subsets showed that in the steady state (IMEC WT) MYC occupancy was higher among the up-regulated genes and it further increased in response to MYC over-expression ( Fig. 3c and Supplementary Fig. 5c). Of note, the different MYC occupancy correlated with a specific enrichment for canonical E-box among the induced genes ( Supplementary Fig. 5d), in agreement with previous reports 37 38 . Importantly, by analyzing previously published datasets 37 38 , we confirmed that these two distinct subsets of MYC targets were induces and repressed accordingly, upon MYC activation in other two independent cell lines ( Supplementary Fig. 5e-f).
Considering that the transcriptional response to MYC overexpression has been correlated with the ratio between MYC and MIZ1 binding 37 , we quantified their relative occupancy at the promoters of the two subsets ( Supplementary Fig. 5g-j). We showed that in both analyzed dataset the down-regulated genes had a lower MYC/MIZ1 ratio, supporting the notion that these targets are directly repressed by MYC in conjunction with MIZ1 binding, while high MYC/MIZ1 ratio characterizes the up-regulated genes, indicating a direct MYC-mediated transcriptional activation. Importantly, GO analyses showed that these two subsets were enriched for genes belonging to different functional categories (Fig. 3d). Supra-physiological expression of MYC has been associated to invasion of almost all active regulatory elements in the genome 37 30 46 29 . The specificity of the differential binding affinity and its association with transcription modulation was further supported by the ChIP-seq data analyses showing that more than 18.000 active promoters marked by H3K4me3 were not bound by MYC ( Supplementary Fig. 6a). These results indicated that in this biological context MYC activation did not caused chromatin invasion of active regulatory elements 37 30 46 29 . Together these analyses showed that MYC occupancy on promoters determined the transcriptional outcomes of MYC-target genes.
Considering that MYC also associated to introns and intergenic regions (Fig. 3a), we investigated whether it occupied and modulated the activation of enhancers. By profiling the distribution of H3K4me1 in IMEC WT, -MYC and M2, we mapped all the putative distal cis-regulatory elements (Fig. 3e, upper panel). Thereafter, we defined the active enhancers by profiling the relative enrichment for H3K27ac at these loci. Overall, the cellular reprogramming was mirrored by a highly dynamic modulation of the defined cis-regulatory elements giving rise to different enhancer states (Fig. 3e, lower panel).
The comparative analyses showed that a subset of enhancers resulted repressed in the MYC-over-expressing cells as they showed a consistent reduction of the H3K27ac level ( Fig. 3f). Using a criterion of proximity to assign each enhancer to its regulated gene 47 , we observed that enhancer decommissioning determined the down regulation of their related-genes ( Fig. 3f and Supplementary Fig. 6b). Among these genes we identified TFs involved in establishing the transcriptional regulatory network of luminal cells, such as TFAP2C, TBX3 and ZNF217 ( Fig. 3f and Supplementary Fig. 6c). In addition, these repressed enhancers were enriched for binding sites of luminal-specific TFs (Fig. 3g), in accordance with the down-regulation of genes associated to ML-specific enhancers (Fig.   1e). Of note, GO analyses highlighted that the WT-specific activated enhancers were mainly related to genes involved in the integrin, EGF and PI3K signaling pathways (Fig.   3h).
By focusing on the chromatin modulations occurring in the mammospheres, we identified a subset of enhancers, which were specifically activated in M2 (Fig. 3e). These de novo enhancers were defined as distal genomic regions, which did not carry H3K4me1 and H3K27ac in IMEC and gained these histone modifications upon transition to a stem cell-like state (Fig. 3f). In addition, the activation of de novo enhancers was associated with an increment of MYC binding and with an overall increased expression of the related genes ( Fig. 3f and Supplementary Fig. 6b, d-e). We found that stem cellassociated TFs and genes involved in activating the Wnt signaling were strongly enriched in this subset of enhancers ( Fig. 3f-h). Taken together, these results indicated that the MYC-induced alteration of the luminal-specific transcriptional program associates with the repression of those enhancers that modulate the expression of the luminal lineage-specific TFs. In addition, the acquisition of a stem cell-like fate is associated with the activation of de novo enhancers that control the expression of TFs and signaling pathways which are frequently activated in both somatic and cancer stem cells 48 5 49 50 51 .

Activation of de novo enhancers drives oncogenic pathways
We further characterized the de novo enhancer by ranking their related genes according to their expression level in mammospheres and we observed a positive correlation between over-expressed genes and increased MYC recruitment at their enhancers ( Fig.   4a and Supplementary Fig. 7a). GO analyses showed that the enhancer-dependent regulated genes were associated with the modulation of Wnt pathways (Fig. 4b).
Specifically, we identified genes coding for oncogenic TFs as well as genes involved in regulating both the canonical and non-canonical Wnt pathways, which are often deregulated in breast cancer ( Fig. 4c and Supplementary Fig. 7b-c) 52 53 54 . By defining the set of genes whose de novo enhancers were bound by MYC and induced in mammospheres, we showed that MYC associated with one third of the 289 regulated genes ( Fig. 4c and Supplementary Fig. 7d). Of importance, the increment of expression of this subset of genes correlated with augmented MYC occupancy at the relative enhancers ( Fig. 4d and Supplementary Fig. 7e-f). In addition the knock-down of the exogenous MYC, which cause a 50% reduction of total MYC protein ( Supplementary Fig.   3h), impaired the transcriptional activation of its targets ( Supplementary Fig. 8a). Next, we investigated the direct contribution of MYC binding to the chromatin state of the de novo enhancers by measuring the relative enrichment for H3K4me1, H3K27ac and MYC at these loci (Fig. 4e). These analyses showed that the M2-induced enhancers are characterized by a large distribution of both H3K27 and K4me1 marks, spanning as average regions over 3.1 kb (Fig. 4e). Of note, the distribution of these histone marks is similar to the pattern of the stretch-and super-enhancers which compromise dense transcription factors binding sites, forming cluster of enhancers that regulate the expression of lineage-specifying genes 55 20 . In addition, on MYC-target de novo enhancers we found that MYC binding peaked at the center of the H3K27ac-enriched region, suggesting a direct contribution to the deposition of this active histone mark ( Fig. 4e). Moreover, by performing motif discovery analysis we found the highest enrichment for FOX-and SOX-family members, as well as ETS1 motifs (Fig. 4f).
Importantly, among the MYC-target de novo enhancers we found a specific enrichment for a non-canonical E-box 29 , indicating that MYC association is mediated by its direct binding to the chromatin. In summary, these data strongly support the notion that de novo enhancers modulate the transcriptional activation of oncogenic pathways. In addition we characterized a subset of de novo enhancers, which are enriched for MYC binding at their epicenter, suggesting a modulatory function in their activation.

Reactivation of WNT pathway supports MYC-induced stem cell features
To establish whether this enhancer-mediated regulation determined the overall hyperactivation of the Wnt pathway in mammospheres, we verified the transcriptional up-regulation of Wnt pathway-related genes, including the FZD1 and FZD8 receptors and LRP6 co-receptor (Fig. 5a). In addition, the two major inhibitors of the pathway, DKK1 and SFRP1, were strongly down-regulated in cells over-expressing MYC (Fig. 5a).
Next, to detect Wnt responsive cells, we transduced IMEC-MYC with a lentiviral vector containing a 7xTCF-eGFP reporter cassette (7TGP). FACS analyses showed that the Wnt pathway was activated in mammospheres and not in IMEC-MYC (Fig. 5b). In order to determine whether WNT signaling activation could have a functional role in MYCinduced stem cell features, we discerned between IMEC-MYC with the highest (GFP high ) and the lowest (GFP low ) signal for Wnt pathway activation. Dye retention assay showed that Wnt responsive cells (GFP high ) were enriched for slow-dividing cells, which retained higher level of the cell tracer, suggesting enrichment for stem cells (Fig. 5c). Given the cellular heterogeneity within the mammospheres population, we performed single cell sorting of GFP high and GFP low cells ( Fig. 5d and Supplementary Fig 8c). On average, by analyzing independent clones we determined that the GFP high sub-population showed enrichment for cells with self-renewing capacity (Fig. 5e, left panel). We further characterized the GFP high -derived primary spheres (GFP high -derived M1) in respect to the relative enrichment for Wnt pathway activation. The obtained results showed a concomitant increment of Wnt signaling in GFP high -in respect to the GFP low -derived M1 cells (Fig. 5e, left panel and Supplementary Fig 8c). Furthermore, by performing serial clonogenic assay of both GFP high and GFP low cells derived from independent clones, we observed that the Wnt responsive population was further endowed with self-renewing capacity, giving rise to clones characterized by enhanced activation of the pathway (Fig.   5e, right panel and Supplementary Fig 8c). Gene expression profiling of sorted GFP high and GFP low cells showed that Wnt-responsive cells were enriched for a mammary stem cell transcriptional program and correlated with metastatic transcriptional signatures ( Fig. 5f-g). These results suggest a correlation between the reactivation of Wnt pathway and acquisition of a stem cell-like transcriptional program, which has been associated with increased risk of developing recurrent cancer 42 56 57 .

MYC-induced reprogramming favors the onset of TICs
To determine whether MYC-induced reprogramming favors the onset of TICs in vivo, we challenged IMEC-MYC with an additional oncogenic insult by over-expressing PIK3CA H1047R , which caused hyper-activation of PI3K pathway ( Supplementary Fig. 9a). forming highly proliferative and heterogeneous tumors expressing both luminal (CK8/18) and basal (CK5/6 and p63) markers ( Fig. 6b and c). In addition, tumors were negative for ER-α and PR and did not show over-expression of HER2 (Fig. 6c), recapitulating the histo-pathological features of basal-like breast cancer. To determine long-term tumorigenic potential, xenograft-derived (XD) cells obtained from primary tumors were re-injected in the mammary gland of secondary recipient mice. Serial transplantations showed that the XD cells maintained tumorigenicity, forming tumors with features resembling the primary one (Fig. 6d). Notably, XD cells showed considerable migration and metastatic seeding capacity as, after chirurgical resection of secondary tumors, all treated animals developed macro-metastasis in liver, lung and spleen (Fig. 6e). Taken together, these results suggest that IMEC-MYC-PIK3CA H1047R were endowed with long-term tumorigenic capacity and metastatic potential.

Activation of MYC-driven oncogenic signature in basal-like breast cancers
We next asked whether the MYC-dependent oncogenic signature, activated in M2 by de novo enhancer modulation, could be associated with IMEC-MYC-PIK3CA H1047R tumorigenicity. We measured a significant transcriptional up-regulation of these oncogenes in both primary and secondary tumors, in respect to parental cells ( Fig. 7a and Supplementary Fig. 9g). Concomitantly, we determined activation of the respective enhancers in XD cells, suggesting that the same molecular mechanism driven by MYC in M2 cells was responsible for their up-regulation in tumorigenic cells ( Fig. 7b and Supplementary Fig. 9h). To assess whether our findings are clinically relevant, we investigated the expression of MYC-dependent oncogenic signature in a database of breast cancer patients. The average expression of MYC-induced oncogenes is strongly up-regulated in basal-like breast cancers and predictive of a worst prognosis for this specific molecular subtype (Fig. 7c-e). Among the MYC targets, those that were specifically overrepresented within the basal-like breast cancers, which include modulators of kynurenine, prostaglandin and Wnt pathways, are frequently deregulated in human cancers 60 61 62 . The pathological relevance of the identified MYC signature was further corroborated by the observation that the expression of these genes correlated with reduced metastatic-free survival in patients affected by high-grade breast cancer ( Supplementary Fig. 9i). Taken together these data demonstrated that MYC-modulated enhancers activate oncogenic pathways, which are associated with basal-like breast cancer in patients with a poor prognosis.

Discussion
In this work, we report the central role of MYC in initiating and sustaining a stepwise cell reprogramming process of mammary epithelial cells toward a stem cell-like condition, favoring tumor initiation and progression. Specifically, we show that MYC induces dedifferentiation toward a progenitor-like state achieved through down-regulation of lineage-specifying TFs, resulting in decommissioning of luminal-specific enhancers. The oncogene-triggered loss of cell identity favors the acquisition of stem cell traits, which is mirrored by activation of de novo oncogenic enhancers. Importantly, the herein deciphered epigenetic reprogramming supports tumorigenesis as the oncogenic enhancers are reactivated in the transformed cell counterpart. We further show that MYC directly binds to a subset of de novo enhancers, suggesting that it participates in activating oncogenic pathways involved in the formation and maintenance of TICs.
Overall, we established a key role of MYC as tumor reprogramming factor by guiding the acquisition of stem cell-like traits, thereby increasing the likelihood of neoplastic transformation upon further oncogenic insults.
In oncogenic setting, supraphyisiological activation of MYC promotes tumorigenesis by conferring selective cell growth advantage 63  It has been suggested that oncogene-driven cell plasticity in which, following an appropriate oncogenic insult, most cell in a tissue has the potential to acquire stem celllike properties, participates in determining tumor heterogeneity 68

Cell lines
All experiments were performed in following cell lines and derivatives: human IMEC, T47D, MCF7 and ZR751.

Animal studies
Animal

Mammospheres Culture and Related Assays
Mammospheres culture was performed as previously described 41  and mammospheres area (μm 2 ) were measured using the NIS Element software (Nikon).
Single cell clonogenic assay was performed in 96 well plates, in at least 3 biological replicates. Single cells were sorted with a BD FACS Aria III sorter (BD Bioscieces), one cell/well and formed mammospheres were counted after 3 weeks by microscope observation (time window required for primary spheres formation).

Immunofluorescence
For mammospheres differentiation assay, cells were grown in mammospheres culture

Flow Cytometry Analysis (FACS)
ALDH activity was assessed with the Aldefluor kit (Stemcell Technologies #1700) on IMEC WT and IMEC-MYC culture as mammospheres for one passage.
Dye retention assay was performed with CellTrace Violet Cell Proliferation Kit (molecular probes #C34557) on IMEC-MYC-7TGP cultured as mammospheres for one passage. After the staining, cells were re-plated in the same conditions and acquired to FACS after 6 days.
Following digestion, tumor cell suspensions were pelleted and then suspended in 0.25% trypsin for two minutes.

Immunohistochemical analysis
Immunohistochemical

Soft agar assay
Colony forming assay was carried out using Noble agar (Sigma-Aldrich #A5431). For the lower layer, agar was mixed with IMEC medium, reaching a final concentration of 0,6% and plated on 6 well plates. 4x10 5 cells were plated on top of it, in 0,3% agar. Colony formation was monitored up to 21 days by microscope observation.

Protein Extraction and Western Blot Analysis
Total protein extracts were obtained as follows. Cells were washed twice with cold PBS, harvested by scrapping in 1 ml cold PBS and centrifuged for 5 minutes at 1500 rpm. Primary antibodies used are as follows: β-Actin (Sigma-Aldrich #A5441), c-Myc (Cell Signaling #5605). Relative optical density was quantified with ImgeJ Software. fold change threshold in all the analyses performed.

Computational Analysis of Gene Expression Data
Scatter plots, correlation heatmaps and PCA analysis of gene expression data were performed in R (http://www.R-project.org/). Differentially expressed genes were checked for biological and functional enrichment using the GO based online tool PANTHER Classification System. Geneset Enrichement Analysis (GSEA) was performed with genesets retrieved from both public available databases and indicated papers.

Chromatin Immunoprecipitation (ChIP) Assay
Each ChIP experiment was performed in at least three independent biological samples. Collection and processing of breast cancer gene expression data.
Breast cancer gene expression data have been obtained from a collection of 4,640 samples from 27 major data sets comprising microarray data of breast tumors annotated with pathological information and clinical outcome (Supplementary Table 2).
All data were measured on Affymetrix arrays and have been downloaded from NCBI Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) and EMBL-EBI ArrayExpress (http://www.ebi.ac.uk/ arrayexpress/). Prior to analysis, all datasets have been ro-organized as described in 71 . Since raw data (.CEL files) were available for all samples, integration, normalization, and quantification of gene expression levels has been obtained with the procedure described in 72 . The type and content of pathological and clinical annotations have been standardized, among the various datasets, as described in (Cordenonsi et al., 2011). This resulted in a compendium (meta-data set) comprising 3,661 unique samples from 25 independent cohorts (Supplementary Table   3

Survival analysis.
To identify two groups of tumors with either high or low MYC direct target signature we used the classifier described in 74  involving animals were conducted randomly and not blinding.

Data resources
Raw and quantile normalized data files for the microarray analysis have been deposited in the NCBI Gene Expression Omnibus under accession number GSE86407.
Raw data and genomic regions of ChIP-seq peaks have been deposited in the NCBI Gene Expression Omnibus under accession number GSE86412.
The following link has been created to allow review of record GSE86416 while it remains in private status: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=qpynswmgnrgzbcz&acc=GSE86