MAX Inactivation in Small Cell Lung Cancer Disrupts MYC–SWI/SNF Programs and Is Synthetic Lethal with BRG1

Small cell lung cancer (SCLC) accounts for about 20% of lung cancer diagnoses and is a highly aggressive malignancy. However, the genetics underlying its development are still largely unknown. The genes most widely known to be frequently altered in SCLC are TP53, RB1, and those of the MYC family (1). PTEN, PIK3CA, and BRG1 (also called SMARCA4) are less frequently altered (1, 2). Novel high-throughput sequencing screening approaches, such as exome sequencing, have been performed on SCLCs, revealing alterations at other genes, including the chromatin modifiers CREBBP and EP300 (2).

Amplifications of L-MYC, N-MYC, and C-MYC are some of the best established gene alterations in lung cancer; L-MYC and N-MYC are more commonly amplified in SCLC than in non–small cell lung cancer (NSCLC; ref. 1). In fact, the pattern of gene alterations in SCLC is rather specific to this tumor type, probably reflecting the different cell of origin of the distinct classes of lung cancers. For this reason, it has been suspected for some time that some SCLC subtype arises from neuroendocrine cells in the lung, which are commonly found in clusters known as neuroendocrine bodies (3). Recent observations, using mouse models for targeted Trp53 and Rb1 inactivation in distinct cell types of the adult lung, support the explanation of the neuroendocrine origin of at least some SCLCs (4). Either because of its possible neuroendocrine origin or because of a specific neural tumor differentiation, SCLCs are enriched in transcripts that are related to neural tissues (5, 6). The neural origin of some SCLCs may serve to explain why some of the genes mutated in this type of lung cancer are also altered in other neural-related tumors. Such is the case of N-MYC and RB1, which are commonly altered in neuroblastomas and retinoblastomas, respectively (7, 8).

Recently, germline-inactivating mutations at MAX, the MYC-associated factor X gene, were found to be responsible for hereditary pheochromocytoma, a tumor with neuroendocrine features (9). Homozygous inactivation of MAX in rat adrenal pheochromocytoma PC12 cells had previously been reported (10). Inasmuch as the genes of the MYC family are commonly activated in SCLC and that, similar to pheochromocytomas, SCLCs have neuroendocrinal features (11), we decided to test for MAX inactivation in lung cancer.

We sequenced the entire coding region and the intron–exon boundaries of MAX in lung cancer cell lines (Supplementary Table S1) and found MAX intragenic homozygous deletions, which caused the complete loss of MAX protein, in H1417, Lu134, Lu165, and COR-L95 cells, all of which are of the SCLC type (Fig. 1A and B). We also sequenced MAX in primary SCLCs and performed multiplex ligation-dependent probe amplification (MLPA; ref. 12) to test for intragenic deletions (Fig. 1C and D). We tested the tumor xenograft directly derived from the same primary tumor as the Lu134 cells and confirmed the presence of an identical MAX alteration, which ruled out the possibility that the observation was a cell culture artifact (Supplementary Fig. S1). Overall, we found homozygous and tumor-specific MAX-inactivating alterations in about 6% of the 98 SCLCs tested (Table 1), a prevalence similar to that of the recently identified CREBBP and EP300 tumor-suppressor genes (2). All of the tumor specimens were surgically resected before treatment, so that the possibility can be ruled out that alterations of MAX are secondary alterations due to chemotherapy or radiotherapy.

The MAX protein contains a basic helix–loop–helix zipper domain, which is required to form heterodimers with MYC, and which binds to hexameric E-box motifs (CACGTG) in the DNA to activate transcription. The MYC–MAX heterodimers also indirectly repress the expression of other genes (13, 14). In keeping with the current view that proteins acting in a common biologic pathway are not simultaneously altered in the same tumor specimen (1), we found that the alterations in MAX and amplification of the MYC genes were mutually exclusive. Moreover, none of the MAX-mutant cells carried concomitant mutations of BRG1 (Supplementary Table S1), which are also known to be mutually exclusive with amplification of the MYC genes in lung cancer (15).

The functional relationship of MYC with both MAX and BRG1 is well established (13, 14, 16). Here, we aimed to elucidate the mechanistic interaction between MAX and BRG1 and to explore the role of MAX inactivation in cancer. To this end, we took advantage of three lung cancer cell lines lacking MAX and studied the effects of restoring MAX activity and depleting BRG1 in these cells (Fig. 2A). Reconstitution of MAX significantly reduced cell growth in the three cancer cell lines (Fig. 2B and Supplementary Fig. S2A), which is consistent with a previous observation in PC12 cells (10) and provides evidence of the tumor-suppressor function of MAX. Likewise, depletion of BRG1 gave rise to a highly significant reduction in cell viability.

We observed that the 5′-untranslated region (5′-UTR) of MAX contains putative glucocorticoid response elements. Glucocorticoids are critical in normal lung differentiation, and BRG1 is required to mediate the transcriptional activities of the glucocorticoid receptor (3, 16). Therefore, we examined whether MAX was responsive to, or involved in, mediating the response to treatment with glucocorticoids. For this purpose, we cloned MAX with the 5′-UTR (henceforth 5′-UTR–MAX; Fig. 2C). Ectopic levels of MAX were low in cells infected with 5′-UTR–MAX, in a hormone-free environment, and became upregulated after treatment with glucocorticoids. In contrast, cells carrying MAX devoid of the 5′-UTR exhibited high basal levels of MAX, which increased moderately upon treatment with glucocorticoids (Fig. 2D). This indicates that the 5′-UTR contains regulatory elements that modulate the expression of MAX in response to glucocorticoids.

Unexpectedly, we found a severe reduction in the levels of ectopic MAX after depleting BRG1 in the glucocorticoid-treated 5′-UTR–MAX cells, implying that the regulation of MAX expression through its 5′-UTR is strongly dependent on BRG1 (Fig. 2E). The requirement of BRG1 for transactivation of endogenous MAX was verified in several lung cancer cell lines and in neuroblastoma-derived SHSY-5Y cells. These exhibited a strong reduction of MAX following depletion of BRG1 (Fig. 2F and Supplementary Fig. S2B). In addition, the findings were reproduced in cells derived from the H1299 lung cancer cells, which are BRG1 deficient. These were the isogenic cells H1299tr-BRG1wt and H1299tr-BRG1mut, which express the wild-type and a mutant version of BRG1, respectively (16). The level of expression of ectopic MAX from the 5′-UTR-MAX was very low in the BRG1-mutant cells, in a hormone-free environment, and increased slightly upon the addition of glucocorticoids. In contrast, neither the status of BRG1 nor the presence of glucocorticoids impinged on the levels of MAX from the construct lacking the 5′-UTR (Supplementary Fig. S2C).

We then examined whether BRG1 is recruited to the MAX promoter. Chromatin from the H1299tr-BRG1wt and H1299tr-BRG1mut cells was precipitated [chromatin immunoprecipitation (ChIP)] using a BRG1 antibody. DNA enrichment was measured by quantitative PCR (qPCR) using primer sets flanking a region of about 3,000 bp. We observed enrichment of BRG1 in a region of about 700 bp within the 5′-UTR of MAX (Fig. 2G). These findings demonstrate, for the first time, a direct functional connection between these two tumor suppressors, in virtue of which BRG1, as part of the switch/sucrose nonfermentable (SWI/SNF) complex, facilitates the access of the glucocorticoid receptor to the MAX promoter, thereby activating its expression.

Depletion of BRG1 dramatically impaired (by >95%) cell viability in the MAX-deficient cells (Fig. 2B and Supplementary Fig. S2A). To test whether this behavior also took place in cancer cells with wild-type MAX, we depleted BRG1 in a panel of six lung cancer cell lines with amplification of either MYC, MYCL, or MYCN. Only a moderate decrease (<25%) in cell growth was found in some cells, implying that the depletion of BRG1 was preferentially toxic in MAX-deficient cells (Fig. 3A). It is of particular note that the MAX–shBRG1 cells, from parental Lu134 and Lu165 cells, were more viable than either the MAX-reconstituted or shBRG1 cells (Fig. 3B and Fig. 2B). This suggests that MAX restores, to some extent, the cell viability that is undermined by shBRG1, or vice versa. Taken together, these results imply the existence of a synthetic lethal type of interaction between MAX and BRG1, and raise the possibility of developing a therapeutic strategy for patients with MAX-deficient tumors.

To elucidate the functional relationship between MAX and BRG1, we performed global gene expression analysis in the ectopic MAX and shBRG1 cell models. On the basis of their Gene Ontology (GO) function, reconstitution of MAX showed a significant enrichment of transcripts related to neural differentiation and to glycolysis/carbohydrate metabolism (Supplementary Fig. S3). Some of these transcripts are well-established targets of MYC (Fig. 4A; refs. 17, 18). Because SCLCs have neuroendocrine features (6, 11), the upregulation of neural-related genes upon MAX reconstitution possibly reflects the activation of prodifferentiation programs. Reconstitution of MAX also upregulated genes encoding glycolysis enzymes (e.g., LDHA, HK2, PDK1, PKM2, and PGK1), which are targets of MYC. Conversely, the levels of these glycolysis-related genes were lower in the shBRG1 cells than in the control cells (Fig. 4A), whereas the MAX–shBRG1 cells showed a reversion in the expression of glycolysis- and neural-related genes toward the profile of the control cells (Fig. 4B).

Most of the MYC targets that were upregulated by MAX were inversely associated with the gene expression profile of shBRG1 cells (Fig. 4C). The MAX signature was also negatively correlated with that of embryonic lungs from mice overexpressing Nmyc and Cmyc, whereas there was a direct association with the expression profile after BRG1 reconstitution in lung cancer cells (Fig. 4C; ref. 16). These observations imply that BRG1, MAX, and MYC orchestrate the transcriptional regulation of a common set of genes, and suggest that MYC represses cell differentiation–related transcripts in a MAX-independent manner.

MAX interacts not only with MYC, but also with other BHLHZ-containing proteins (i.e., MXD1, MXI1, MXD3, MXD4, MNT, and MGA), which antagonize the transcriptional control of MYC, at the same E-box target DNA sequences (14). Some of these MAX-binding proteins promote differentiation in vivo, block cellular growth and MYC-induced transformation, and suppress the development of cancer (19). Furthermore, apart from BRG1, genes coding for other components of the SWI/SNF complex are altered in cancers, including lung cancer (20, 21). We have exhaustively searched public databases for gene alterations in partners of MAX in all types of lung cancer and found inactivation of the MAX dimerization protein, MGA, in lung cancer cell lines (Fig. 5A) and in lung primary tumors (http://www.cbioportal.org/) of both the NSCLC and SCLC type. We have tested a panel of these cell lines and confirmed the alterations, most of which are homozygous and predictive of truncated proteins (Fig. 5B). Moreover, inactivation of MGA and alterations at different members of the SWI/SNF complex at MAX and MYC were mutually exclusive in lung cancer (Fig. 5C).

We report the discovery of the recurrent inactivation of MAX in SCLC. The alterations were tumor-specific and homozygous, leaving little doubt that MAX constitutes a bona fide tumor-suppressor gene. MAX-inactivating alterations occurred in about 6% of the SCLCs, a prevalence similar to that of the recently identified CREBBP and EP300 tumor-suppressor genes (2). Most alterations of MAX were intragenic deletions, which may explain why inactivation of MAX has not been picked up in recent exome or whole-genome sequencing of SCLCs (2, 22). Germline-inactivating mutations at MAX were found to be responsible for hereditary pheochromocytoma, a tumor with neuroendocrine features (9). This is interesting because MAX inactivation was found only in SCLC and not in the NSCLC type, the latter comprising about 80% of the lung cancers. Taking into account that, unlike NSCLCs, SCLCs have neuroendocrine features, it can be hypothesized that MAX is preferentially mutated in neuroendocrine-related malignancies.

None of the MAX-mutant cells carried concomitant amplification at the MYC family of genes or mutations at BRG1. In lung cancer, genetic inactivation of BRG1 is mutually exclusive with amplification of MYC genes, which is consistent with the biologic connection between these two cancer proteins (15). BRG1 encodes one ATPase of the SWI/SNF chromatin remodeling complex and is involved in the transcriptional control of various cell processes, such as embryonic development and cell differentiation (23, 24). Some of these activities also require the complex to interact with nuclear receptors (25). The functional relationship of MYC with BRG1 and with the SWI/SNF complex is well established. For example, MYC physically interacts with the SWI/SNF component, SMARCB1 (26), and BRG1 is required to regulate the expression of MYC and MYC target genes (16, 27). In tumors carrying BRG1 mutations, this regulation is abolished, thereby preventing cell differentiation and promoting cell growth (16). Here, we provide evidence that BRG1 also regulates the levels of MAX, stimulated by the presence of glucocorticoids. In virtue of this regulation, and as part of the SWI/SNF complex, BRG1 would be directly recruited to the 5′-UTR of MAX to facilitate the access of the glucocorticoid receptor to the MAX promoter, thereby activating its expression. This constitutes the first demonstration of a direct functional connection between these two tumor suppressors. In addition to glucocorticoid receptor, the SWI/SNF complex interacts with other nuclear receptors, such as estrogens, retinoic acid, and vitamin D3 receptors (25, 28, 29). Therefore, we cannot rule out the possibility that the activation of these other receptors also influences the expression levels of MAX in an SWI/SNF–dependent manner.

We also noted that depletion of BRG1 triggered a strong inhibition of lung cancer cell growth. The effect was very specific to the MAX-deficient cells, as growth was not affected after depleting BRG1 in a panel of lung cancer cells with amplified MYC genes. Regardless of the mechanism underlying this specific and dramatic impairment in the growth of MAX-deficient cells after BRG1 depletion, it is important to highlight its potential clinical relevance. This information suggests therapeutic possibilities for the treatment of patients with SCLC, and possibly for patients with pheochromocytoma, with MAX-deficient tumors.

The reconstitution of MAX in SCLC cells triggered the expression of neural-related and glycolysis/carbohydrate metabolism-related transcripts, some of which are well-established targets of MYC (17, 18). Because it has been speculated that some SCLCs arise from neuroendocrine cells (4), the upregulation of neural-related genes upon MAX reconstitution may reflect the activation of prodifferentiation programs. Genes encoding glycolysis enzymes (e.g., LDHA, HK2, PDK1, PKM2, and PGK1) are targets of MYC and are commonly overexpressed in cancer cells (18). Conversely, the levels of these glycolysis-related genes were lower in the shBRG1 cells than in the control cells. It is intriguing that the reconstitution of a tumor suppressor, MAX, activates the expression of genes that are typically upregulated in cancer cells. Although additional research is likely to provide an explanation for these observations, we can speculate in the meantime that a substantial increase (i.e., after reconstitution of MAX) or decrease (i.e., after depletion of BRG1) of glycolysis enzymes, in a predominantly tumor-genetic context, triggers an energetic imbalance, with dire consequences for cell viability. In support of this concept, the rescue of cell viability observed in MAX–shBRG1 cells but not MAX-reconstituted or BRG1-depleted cells is accompanied by a reversion in the expression of glycolysis- and neural-related genes toward the profile of the control cells.

MAX was believed to be essential to the oncogenic function of MYC (30). However, in Drosophila, an Myc mutant lacking the Max-interaction domain still retained partial activity (31). These observations, coupled with the existence of MAX inactivation in pheochromocytomas (9) and now in SCLC, imply that some MYC activities are independent of its heterodimerization with MAX. Here, we found that most of the MYC targets that were upregulated by MAX were inversely associated with the gene expression signatures of shBRG1 cells and of embryonic lungs from mice overexpressing Nmyc and Cmyc. In contrast, there was a direct association with the expression profile after BRG1 reconstitution in lung cancer cells (16). These findings imply that BRG1, MAX, and MYC orchestrate the transcriptional regulation of a common set of genes and suggest that MYC represses cell differentiation–related transcripts in a MAX-independent manner. Figure 6 depicts a model that speculates about the different scenarios that could take place in the context of MAX, MYC, or BRG1 genetic alterations in a cancer cell (Fig. 6).

Finally, through a search of public cancer databases, we discovered recurrent inactivation at the MAX dimerization protein, MGA, which was mutually exclusive with alterations of members of the SWI/SNF complex and with MAX and MYC. Thus, MGA is a tumor-suppressor gene in lung cancer and is probably another cog on the SWI/SNF–MYC functional axis. In this regard, it is important to highlight that in recent RNA interference screening, Brg1, Max, and Mga were identified as repressors of germ-cell gene expression in murine embryonic stem cells. This is additional evidence of a functional connection of these proteins with the control of cell differentiation (32).

In conclusion, our current genetic and molecular findings provide powerful evidence that MAX is a tumor-suppressor gene involved in SCLC development. We propose that the abnormal function of the SWI/SNF–MYC axis, arising from genetic alterations of any of its components, prevents cell differentiation and constitutes an acquired ability of the cancer cells that should be added to the Hanahan and Weinberg model (33).

Lung cancer cell lines representative of the most common lung histopathologies were included (Supplementary Table S1). Cell lines were obtained from the American Type Culture Collection, grown under recommended conditions, and maintained at 37°C in a humidified atmosphere of 5% CO₂/95% air. All cells tested negative for Mycoplasma infection. The cells lines were authenticated by genotyping for TP53 and other known mutations (e.g., BRG1/SMARCA4 and STK11). The genotyping of the mutations was done before starting the experiments or, in some particular cases, also during the experimental part of the study. The genotyping was done using direct Sanger sequencing of PCR products, as indicated previously (1). The mutations found were in agreement with those provided in public databases, as indicated in Supplementary Tables. Most samples were from the National Cancer Center Biobank at the National Cancer Center Hospital (Tokyo, Japan). Four samples were obtained from the Biobanco del SSPA (Sistema Sanitario Público de Andalucía, Spain). Genomic DNA and total RNA were extracted by standard protocols. The study was approved by the relevant Institutional Review Boards and ethics committees.

To extract genomic DNA, freshly frozen tissue from tumors was meticulously dissected to ensure enriched material containing at least 40% tumor cells. Approximately 10- to 20-μm sections were collected and placed in 1% SDS/proteinase K (10 mg/mL) at 58°C overnight. Digested tissue was then subjected to phenol–chloroform extraction and ethanol precipitation following standard protocols. For mutation screening of the MAX gene, exons 1 to 5 (NM_002382.3) and MGA (exons 1–23; NM_001164273.1) were amplified from 30 ng of genomic DNA extracted from all tumors. Cycle sequencing of PCR products was carried out using Big Dye Terminator chemistry (Applied Biosystems) with an ABI PRISM 3700 DNA Analyzer (PerkinElmer Life Sciences, Inc.). All of the variants identified in the study were confirmed by resequencing of independent PCR products. MLPA technique was used to determine the presence of intragenic homozygous deletions at MAX in the lung primary tumors. The MLPA protocol used here has been described elsewhere (12). In short, 100 ng of genomic DNA was denatured for 5 minutes at 98°C, after which 3 μL of the probe mixture was added. The sample DNA and P429-A1 MLPA probe mixture was heated at 95°C for 1 minute and then incubated at 60°C for 16 hours, followed by a ligation step (Ligase-65 enzyme; MRC-Holland). Subsequently, a 10-μL mixture was added containing deoxynucleotide triphosphates (dNTP), Taq polymerase, and one unlabeled and one 6-carboxyfluorescein amidite–labeled (FAM) universal PCR primer in each reaction. PCR was carried out for 35 cycles, and the fragments were analyzed with an ABI model 3130XL capillary sequencer (Applied Biosystems) using GeneScan 500 LIZ size standard (Applied Biosystems). MLPA fragment analysis and comparative analysis were performed using CoffalyserNET software (34).

Two main forms of MAX have been described: the short and long forms, containing 151 and 160 amino acids, respectively (35). These arise by alternate mRNA splicing that inserts nine amino acids between residues 12 and 13 of the short form of MAX. Here, we cloned the short form (NM_145112.2) because it was the more abundant transcript in the tumors and lung cancer cell lines tested and was functionally equivalent to the long form. Human wild-type MAX was cloned into the EcoRI and NotI restriction sites of the vector pLVX–IRES–ZsGreen1 (Clontech Laboratories, Inc.). All constructs were verified by automatic sequencing. Short hairpin RNAs (shRNA) were purchased from SIGMA-MISSION (LentiExpress Technology; Sigma-Aldrich) as a glycerol stock of 5 pLKO plasmids carrying BRG1-specific shRNA sequences. Two of these shRNAs had previously been shown to deplete BRG1 expression efficiently and specifically (depleted BRG1 but not BRM expression; ref. 16). A scramble shRNA (Sigma MISSION shRNA non-mammalian control SHC002) was used as a control. The lentiviruses were generated within the 293T packaging cells. Lentiviruses carrying MAX constructs were generated using Lenti-X (LentiExpress Technology; Clontech Laboratories), following the manufacturer's recommendations. The lentivirus carrying the shBRG1 and scrambled shRNAs, were cotransfected with each construct and the packaging plasmids psPAX and pMD2.6 (Sigma-Aldrich). After 48 hours, 293T filtered supernatants were collected and subconfluent cells were infected with harvested virus and selected with puromycin for 72/96 hours.

ChIP assays were performed as previously described (16). Briefly, preliminary fixation experiments were performed over a predetermined period. Cells were then fixed in 1% formaldehyde for 10 minutes and final conditions were chosen that yielded the best combination of in vivo fixed chromatin, high DNA recovery, and small average size of chromatin fragments (an average length of 0.25–1.00 kb). Three independent ChIP experiments were performed. qPCRs were performed using SYBR Green Master Mix (Applied Biosystems). Relative enrichment was determined from a standard curve of serial dilutions of input samples. For semi-qPCR, amplifications were performed with 30 cycles in a total volume of 25 μL and run on 2% agarose gels. qPCR was performed using Power SYBR Green Master Mix (Applied Biosystems). The sequences of primer sets used in each case are available upon request.

The following primary antibodies were used for Western blot analyses: polyclonal anti-MAX, sc-197 antibody (Santa Cruz Biotechnology), polyclonal anti-BRG1, H88 (1:1,000; Santa Cruz Biotechnology), anti–C-MYC 9E10 (1:500; Santa Cruz Biotechnology), anti-tubulin (T6199 mouse; Sigma-Aldrich), and anti–β-actin (13854; Sigma-Aldrich). For Western blot analyses, whole-cell lysates were collected in a buffer containing 2% SDS, 50 mmol/L Tris–HCl (pH 7.4), 10% glycerol, and protease inhibitor cocktail (Roche Applied Science). Protein concentrations were determined using a Bio-Rad DC Protein Assay kit (Life Science Research). Equal amounts of lysates (20 μg) were separated by SDS-PAGE and transferred to a polyvinylidene difluoride (PVDF) membrane that was blocked with 5% nonfat dry milk. Membranes were incubated with the primary antibody overnight at 4°C, and then washed before incubation with species-appropriate horseradish peroxidase (HRP)–conjugated secondary antibodies for 1 hour at room temperature.

Dexamethasone was used for glucocorticoid treatment. Cells were first depleted of FBS and subjected to a hormone-free medium by transfer to 10% charcoal-dextran–treated, heat-inactivated FBS for 24 hours before hormone treatment (16). Cells were then treated for 24 to 72 hours with the indicated concentrations of dexamethasone before harvesting.

For cell viability assays, 10 μL of a solution of 5 mg/mL MTT (Sigma Chemical Co.) was added. After incubation for 3 hours at 37°C, the medium was discarded, the formazan crystals that had formed were dissolved in 100 μL dimethyl sulfoxide (DMSO), and absorbance was measured at 596 nm. Results are presented as the median of at least two independent experiments performed in triplicate for each cell line and for each condition.

RNA (100 ng) was used for the gene expression microarray analysis. RNA integrity values were in the range of 9.0 to 10.0 when assayed by Lab-chip technology in an Agilent 2100 Bioanalyzer. For labeling, we used a commercial “One-Color Microarray-Based Gene Expression Analysis” version 5.5 kit and followed the manufacturer's instructions (Agilent manual G4140-90050, February 2007). Hybridization was performed on the Human Gene Expression v2 microarray 8 × 60K (Agilent microarray design ID 014850, P/N G4112F). For scanning, we used a G2505B DNA microarray scanner. Images were quantified using Agilent Feature Extraction Software (v. 9.5). The cells used for microarray gene expression analysis were as follows: (i) control cells: H1417, Lu134, and Lu165 containing the control vector; (ii) H1417, Lu134, and Lu165, each containing 5′-UTR–MAX; (iii) H1417, Lu134, and Lu165, each expressing shBRG1; and (iv) Lu134, expressing both 5′-UTR–MAX and shBRG1. To generate the lists of upregulated and downregulated transcripts for each condition, we chose the following criteria: (i) transcripts induced or repressed by a factor of at least 1.5 under each MAX or shBRG1 condition with respect to their matched cell line carrying the empty vector, and (ii) statistical significance (see below). The genes are listed in Supplementary Tables S2 and S3.

Data were analyzed using a χ² test or a two-tailed Student unpaired-samples t test. Group differences were presented as mean and SDs. Differences were considered statistically significant for any value of P < 0.05. Expression data were analyzed using the R program. Raw data were normalized by the robust multiarray average (RMA) algorithm available in Bioconductor's affy package (36). We used limma package to obtain limma-moderated t statistics to derive the signatures of differentially expressed genes. Test P values were adjusted with respect to the false discovery rate (FDR). Test statistics with an associated value of P < 0.05 were considered to be statistically significant. The lists of genes (ranked by the n-fold values of change) were subjected to gene set enrichment analysis (GSEA; ref. 37) using the indicated gene expression signatures as the gene set. After Kolmogorov–Smirnov testing, our gene sets were considered significantly enriched between comparison classes for values of FDR < 0.25, a widely accepted cutoff for the identification of biologically relevant gene sets (38).

S. Savola is employed by MRC-Holland, manufacturer of commercially available MLPA probemixes. No potential conflicts of interest were disclosed by the other authors.

Conception and design: O.A. Romero, M. Sanchez-Cespedes

Development of methodology: O.A. Romero, M. Torres-Diz, E. Pros, S. Savola, A. Villanueva

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): O.A. Romero, M. Torres-Diz, C. Saez, L.M. Montuenga, T. Kohno

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O.A. Romero, M. Torres-Diz, E. Pros, A. Gomez, S. Moran, A. Villanueva, M. Sanchez-Cespedes

Writing, review, and/or revision of the manuscript: O.A. Romero, M. Torres-Diz, S. Savola, C. Saez, L.M. Montuenga, M. Sanchez-Cespedes

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O.A. Romero, M. Torres-Diz, E. Pros, S. Savola, R. Iwakawa, J. Yokota, M. Sanchez-Cespedes

Study supervision: O.A. Romero, M. Sanchez-Cespedes

The authors thank Patricia Cabral and Sara Verdura (Genes and Cancer Group) at IDIBELL for technical assistance and HaciAli Yigittop for performing the MLPA experiments.

This work has been supported by the Spanish Ministry of Economy and Competitiveness (grant SAF2011-22897 to M. Sanchez-Cespedes), the Red Temática de Investigación del Cáncer-RTICCs (RD12/0036/0045 to M. Sanchez-Cespedes and RD12/0036/0040 to L.M. Montuenga), Red de Biobancos (RD09/0076/00085) and the European Community's Seventh Framework Programme (FP7/2007-13 to M. Sanchez-Cespedes, L.M. Montuenga, and S. Savola), under grant agreement no. HEALTH-F2-2010-258677–CURELUNG. Additional support came from a Grant-in-Aid from the Ministry of Health, Labor and Welfare for the Third-term Comprehensive 10-year Strategy for Cancer Control, Japan, to T. Kohno and J. Yokota. M. Torres-Diz is the recipient of a Fellowship (FPI) from the Spanish Ministry of Economy and Competitiveness.