Abstract
Systematic analyses of cancer genomes promise to unveil patterns of genetic alterations linked to the genesis and spread of human cancers. High-density single-nucleotide polymorphism (SNP) arrays enable detailed and genome-wide identification of both loss-of-heterozygosity events and copy-number alterations in cancer1, 2, 3, 4, 5. Here, by integrating SNP array-based genetic maps with gene expression signatures derived from NCI60 cell lines, we identified the melanocyte master regulator MITF (microphthalmia-associated transcription factor) as the target of a novel melanoma amplification. We found that MITF amplification was more prevalent in metastatic disease and correlated with decreased overall patient survival. BRAF mutation and p16 inactivation accompanied MITF amplification in melanoma cell lines. Ectopic MITF expression in conjunction with the BRAF(V600E) mutant transformed primary human melanocytes, and thus MITF can function as a melanoma oncogene. Reduction of MITF activity sensitizes melanoma cells to chemotherapeutic agents. Targeting MITF in combination with BRAF or cyclin-dependent kinase inhibitors may offer a rational therapeutic avenue into melanoma, a highly chemotherapy-resistant neoplasm. Together, these data suggest that MITF represents a distinct class of 'lineage survival' or 'lineage addiction' oncogenes required for both tissue-specific cancer development and tumour progression.
To begin to organize human cancers on the basis of large-scale chromosomal alterations, we first evaluated the genomes of NCI60 cell lines6 using pre-release 100K SNP arrays (Affymetrix). These arrays interrogate over 124,000 SNP alleles spaced with a median intermarker distance of 8.5 kilobases (kb). The NCI60 cell lines represent tumours from nine different tissue types and are annotated by multiple large-scale data sets7, 8, 9, 10. Thus, these cell lines offer a platform for testing integrated and orthogonal analytic approaches (Fig. 1a). The complete NCI60 SNP array data are available at http://dtp.nci.nih.gov/mtargets/mt_index.html and at http://www.ncbi.nlm.nih.gov/geo (accession number GSE2520).
Figure 1: Increased MITF expression associated with chromosome 3p amplification in melanoma cell lines.
a, Schematic of the integrated genomic approach (see text). b, Hierarchical clustering of raw copy-number data from NCI60 cell lines and normal diploid controls shows cell line subclusters (coloured rectangles, top) and the corresponding chromosome specific SNP clusters (right). CNS, central nervous system tumour; NSCLC, non-small-cell lung cancer. c, Integration of copy-number and gene-expression data. Shown are colourgrams of SNP copy number at 3p14–3p12 (top) and the significant gene-expression correlates from chromosome 3 following supervised analysis (bottom). d, Chromosome 3 copy-number values (top), SNP signal intensity colourgram (middle), cytoband map (bottom), significant genes (arrows), and the MALME-3M core amplicon (red box) are shown.
High resolution image and legend (90K)To determine whether patterns of copy number alterations identified distinct genetic subgroups, hierarchical clustering11 was used to organize 58 NCI60 cell lines based on copy-number alterations defined by SNP array analysis (see Methods). The resulting dendrogram contained subclusters in which the samples segregated largely according to tissue of origin (Fig. 1b, Supplementary Fig. 1). One such group consisted of lung cancer lines, another mainly of colon tumour lines, and another of cell lines derived from malignant melanoma (Fig. 1b). For each of these cell line clusters, there were also associated SNP clusters derived from contiguous chromosomal regions (Fig. 1b and Supplementary Fig. 1).
The tissue-based organization of the samples raised the possibility that the associated chromosomal aberrations might harbour lineage-specific cancer genes. To investigate this, we focused on a region of copy gain on chromosome 3 at 3p13–3p14 that defined the melanoma subcluster (Fig. 1b,c). Here, supervised analysis12 (see Supplemental Methods) was performed using available NCI60 gene expression data, looking for gene expression correlates of the class distinction '3p amplified' (six melanoma cell lines) versus '3p non-amplified' (remaining NCI60 lines; Fig. 1c). 583 transcripts demonstrated significantly increased expression in the 3p-amplified class after Bonferroni correction, with P < 10-6 by t-statistic. This expression signature was influenced significantly by lineage-related differences in transcript profiles between melanomas and the other NCI60 tissue types, as noted by others7. Nonetheless, only one highly expressed gene, MITF, was located within the amplified region (t-ratio = 17.36, P-value = 2.97 
10-12, adjusted threshold for significance = 3.96 10-6; Figs 1c, d). In a similar analysis restricted to just the eight NCI60 melanoma cell lines, the mean MITF expression ratio between the 3p-amplified (six samples) and non-amplified (two samples) classes exceeded that of nearly all randomly permuted class distinctions (data not shown). MITF encodes a basic helix–loop–helix/leucine zipper transcription factor required for development of the melanocyte lineage13, but not previously known to be the target of an acquired somatic mutation. These data raised the possibility that MITF might also function as a lineage-specific oncogene.
We then investigated MITF gene dosage in human tumours by performing quantitative polymerase chain reaction (PCR) on DNA derived from a series of melanocytic nevi, primary cutaneous melanomas, and melanoma metastases. MITF amplification was observed in 3 of 30 (10%) primary cutaneous and 7 of 32 (21%) metastatic tumours, but not in the ten benign nevi tested (Fig. 2a). MITF was the most consistently amplified gene in the 'core' amplified region defined by MALME-3M cells (Fig. 1d) and, when amplified to high levels, appeared to represent the epicentre of the amplicon (see, for example, samples IM3 and MM5; Fig. 2c). Together, these data delineated MITF as the probable oncogene targeted by this genetic alteration.
Figure 2: MITF maps to the epicentre of an amplicon present in a subset of malignant melanomas.
a, Quantitative PCR analysis of the MITF locus across 62 benign nevi, primary and metastatic melanomas. Cases with MITF amplification are shaded black. b, Transcript map of the core 3p13–3p14 amplicon present within the MALME-3M cell line. The FOXP1 gene resides outside the core amplicon. c, Quantitative PCR analysis of several genes within the amplicon (see Methods for details). MALME-3M and SK-MEL-5 are NCI60 cell lines with and without the amplicon, respectively. IM1 and IM3 are primary melanoma samples; MM4, MM5 and nr 99/2 are metastatic melanoma samples. TMF1, UBE1C, GRSP1, LOC391543 and FOXP1 are genes flanking MITF.
High resolution image and legend (78K)Next, we examined nearly 200 tissue specimens derived from primary and metastatic melanomas by fluorescence in situ hybridization (FISH) using bacterial artificial chromosome (BAC) probes spanning the MITF locus (to detect extra-chromosomal amplification events). MITF amplification (range = 4–13 copies per cell; Fig. 3b and Supplementary Table 3) was detected in 2 of 19 (10.5%) cutaneous tumours and 27 of 160 (15.2%) metastatic samples. Once again, no amplification was detected in nine melanocytic nevi analysed. In metastatic melanoma, MITF amplification was associated with decreased 5-year survival (Fig. 3c; Kaplan–Meier log rank P value = 0.024), but not with other clinicopathologic parameters (Supplementary Figs 2 and 3). MITF protein levels were also analysed in the melanoma tissue microarray using automated quantitative analysis (AQUA) technology14. MITF amplification correlated with a significantly increased mean MITF protein expression in metastatic disease (P = 0.019; Figs 3d–f). This association remained significant when expression was measured as a continuous variable in relation to MITF copy number (data not shown). Together, these observations implicate MITF amplification in the progression and lethality of a subset of human melanomas.
Figure 3: FISH, Kaplan–Meier and AQUA analysis of MITF in human melanoma samples.
![Figure 3 : FISH, Kaplan|[ndash]|Meier and AQUA analysis of MITF in human melanoma samples. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v436/n7047/images/nature03664-f3.0.jpg)
Green digoxin- and red SpectrumOrange (from Vysis)-labelled BAC probes detected the MITF locus and chromosome 3 centromere, respectively, in a melanoma tissue microarray. More than 50 nuclei were scored per amplified sample. a, b, A diploid case (a) and a case of MITF amplification (b) are shown. c, Kaplan–Meier survival analysis in metastatic melanoma patients with or without MITF amplification (log rank test, P = 0.024). d, e, AQUA images for MITF protein levels in non-amplified and highly amplified samples (
10 copies per cell). f, Mean and standard error of AQUA scores for MITF protein in non-amplified (left) and amplified (right) metastatic melanoma samples.
In melanocytes, activated MAP kinase triggers MITF phosphorylation at serine-73 (ref. 15), recruiting the transcriptional coactivator p300 (ref. 16) while simultaneously targeting MITF for ubiquitin-dependent proteolysis17. Normal melanocyte growth/differentiation may also require the coordinated actions of MITF and cyclin-dependent kinase (CDK) inhibitors such as p16 or p21 (refs 18, 19). On the other hand, aberrant MAP kinase signalling through activating BRAF or NRAS mutations may underlie a substantial percentage of melanomas20, 21. Interestingly, all NCI60 cell lines harbouring MITF copy gain also contained both the BRAF(V600E) mutation and p16 pathway inactivation22 (Supplementary Table 2). Thus, genetic amplification of MITF might promote tumour formation and/or survival in the setting of cell cycle deregulation and excess MAP kinase pathway activation.
To test this hypothesis in vitro, we sought to overexpress MITF in genetically modified human melanocytes. In these cells, p53 and p16/CDK4/RB (where RB is the retinoblastoma protein) pathways were inactivated in conjunction with telomerase (hTERT) expression (hTERT/CDK4(R24C)/p53DD melanocytes; see Supplementary Methods). Though characterized by a markedly extended proliferative capacity (> 1.5 years in culture), these cells require both TPA (12-O-tetradecanoylphorbol-13-acetate) and cyclic (c)AMP agonists for survival (a hallmark of non-transformed melanocytes).
Next, these modified melanocytes were transduced with either empty retrovirus or retroviruses directing the expression of BRAF(V600E) or MITF (Figs 4a, b). In the presence of TPA and dibutyryl cAMP (dbcAMP), MITF did not alter the growth of these melanocytes, although BRAF(V600E) expression was incompatible with these factors (data not shown). In the absence of TPA and cAMP agonists, neither vector control nor MITF expression alone had any effect on melanocyte growth factor requirements (Fig. 4a). Notably, ectopic BRAF(V600E) expression was associated with loss of MITF protein (Fig. 4b), and enabled minimal factor-independent growth (Fig. 4a). Both primary and hTERT/CDK4(R24C)/p53DD melanocytes showed identical MITF and wild-type BRAF protein levels (not shown). In contrast, the expression of MITF together with activated BRAF conferred robust factor-independence (Fig. 4a). BRAF(V600E) and haemagglutinin (HA)-tagged MITF co-expression was associated with enrichment of the upper band near 60 kilodaltons (kDa) (Fig. 4b) shown previously to represent the serine-73 phosphorylated variant associated with MITF activation15, 23. When the same cells were suspended in soft agar, only cells expressing both BRAF(V600E) and HA-MITF formed anchorage-independent colonies (Figs 4c, d). Thus, deregulated MITF expression cooperated with BRAF(V600E) to transform human melanocytes.
Figure 4: A role for deregulated MITF in melanoma tumorigenesis and survival.
a, MITF and BRAF(V600E) co-expression confers factor-independent growth in hTERT/CDK4(R24C)/p53DD melanocytes. Absorbance (mean and standard error) following crystal violet staining, and photographs at 3 weeks are shown. b, Extracts from hTERT/CDK4(R24C)/p53DD melanocytes expressing BRAF(V600E) 
HA-MITF were immunoblotted using antibodies against BRAF, MITF, HA-tag or 
-tubulin. p53DD, dominant-negative p53. c, d, Growth of cells expressing BRAF(V600E) HA-MITF in soft agar. Colonies (mean and standard error) and photographs taken at 8 weeks are shown (magnification 40 ). e, Growth of melanoma cell lines expressing empty vector or dominant-negative MITF (dn) at 48 hours relative to uninfected controls. Means and standard deviations are shown; MITF copy number is indicated. f, Cell growth (mean and standard deviation, 72 h) of MALME-3M cells expressing dnMITF compared to uninfected controls, in the presence or absence of docetaxel or cisplatin (*P < 0.05, **P < 0.01).
Genetic amplification of MITF also suggests that the melanocyte lineage dependency on MITF might be maintained in melanoma tumour cells23. To test this, an adenovirus expressing a dominant-negative MITF mutant23, 24 (Ad-dnMITF; see Methods) was introduced into NCI60 melanoma cell lines that exhibited varying levels of MITF amplification (Supplementary Fig. 4c). Growth of three NCI60 melanoma cell lines was inhibited following introduction of Ad-dnMITF; however, those with MITF copy gain were more refractory to the dnMITF effect (Fig. 4e). Similar results were also observed following short hairpin (sh)RNA-mediated MITF knockdown (data not shown). Inhibition of MITF function in melanoma cells may trigger CDK2-mediated growth arrest24 or apoptosis though BCL2 downregulation23. Thus, these data suggest that deregulation of MITF through amplification or other mechanisms may preserve a critical lineage survival function in melanoma.
Melanocyte survival mechanisms that are modified by MITF may also contribute to melanoma chemoresistance23, 25. To investigate whether MITF copy gain correlated with drug resistance, we performed a supervised analysis of the available NCI60 pharmacological data, where chemical sensitivity to a large number of small molecules and natural products is known (see Supplementary Methods). 3p copy gain was associated with a significantly increased mean GI50 in 270 compounds (median 35 39.7 compounds expected at random). This degree of chemoresistance was statistically significant within the overall NCI60 data set (P = 0.016; Supplementary Fig. 5a, b). An analysis restricted to just the eight NCI60 melanoma cell lines revealed a similar trend, though this did not reach statistical significance with the limited available permutations (Supplementary Fig. 5c). To examine the specific effect of MITF on melanoma drug resistance, MALME-3M melanoma cells (which contain the MITF amplicon) were infected with Ad-dnMITF at sublethal multiplicity of infection (MOI) and cultured in the presence of commonly used chemotherapeutic agents. MALME 3M cells infected with empty adenovirus showed pharmacologic profiles similar to wild-type controls (Supplementary Fig. 5d). However, cells expressing dnMITF were significantly more susceptible to inhibition by 20 
M of both cisplatin and docetaxel after 72 h (Fig. 4 h). Drug titration curves also showed a four- to fivefold decrease in the cellular growth inhibitory concentration (GI50) of each agent in the presence of dnMITF (data not shown). Thus, reduction of MITF activity may sensitize melanomas to conventional chemotherapeutics.
Somatic alteration through gene amplification suggests that MITF may be a member of a newly recognized 'lineage survival' or 'lineage addiction' oncogene subclass. Unlike oncogene addiction, where inappropriate gene-activation events render cells hyperdependent on particular pathways, dependency in this case might be pre-established during development and maintained in tumours through genetic aberrations. The androgen receptor (AR) is a prototype of this oncogene class that now includes MITF: both AR and MITF are required for the development and survival of their respective (prostate and melanocyte) lineages, maintained in cancers of these lineages, and amplified (or activated by mutation) in association with advanced disease. AR gene amplification typically occurs after androgen withdrawal therapy in prostate cancer patients. Although the cause of MITF amplification is not known, we speculate that increased MITF gene dosage (or deregulation by other means) may preserve its essential survival function in settings where the normal cues preserving MITF activity are lost. BRAF mutation and metastasis may constitute two such settings.
Although AR induces differentiation and growth arrest in non-transformed prostate epithelia in the presence of androgen, AR activation in the setting of p53 and Rb inactivation causes tumours in mice26. Similarly, MITF induces both differentiation and growth arrest in non-transformed cells18, 19. Our results suggest that inactivation of p53 and Rb (the latter through loss of p16 activity) likewise enables MITF to transform human melanocytes in cooperation with mutated BRAF.
AR inhibition remains the most effective therapeutic intervention in prostate cancer, and emerging data suggests a continued dependency on AR in hormone-refractory disease27. Extending the lineage survival oncogene analogy, patient selection based on MITF, BRAF and p16 pathway mutation status might identify responders or non-responders to the BRAF or CDK inhibitors currently in clinical development. Future combined genomic and functional studies may clarify tumour-survival mechanisms across many tumour types, thereby increasing biological understanding and providing new therapeutic possibilities.
Methods
SNP array hybridization
Genomic DNA from 58 NCI60 cell lines was obtained from R. Camalier and D. Scudiero of the National Cancer Institute Developmental Therapeutics programme (NCI DTP). All DNAs were quantified by Picogreen (Molecular Probes) before SNP array analysis. SNP array data collection, normalization and copy-number determination are described in the Supplementary Information.
Hierarchical clustering using raw copy number data
Raw copy number data from CentXba arrays was filtered to reduce invariant SNPs using dChipSNP software (available at http://www.dchip.org) by adjusting parameters for the values of standard deviation/mean. Filtered SNPs were subjected to hierarchical clustering using the Pearson coefficient and average linkage method11. Two-dimensional dendrograms (one for cell lines and one for SNPs) were generated (the full two-dimensional dendrogram is shown in the Supplementary Fig. S1). Nearly identical dendrograms were obtained for both CentHind and CentXba SNP data (not shown).
Supervised analysis of SNP array and gene expression data
The NCI60 cell lines were grouped into two classes based on the presence or absence of amplification at chromosome 3p14-3p13. We then performed supervised analyses using in-triplicate NCI60 gene expression data generated by Novartis on the Affymetrix U95v2 array platform (http://dtp.nci.nih.gov/mtargets/madownload.html). An extended description of this approach is presented in the Supplementary Information. To identify genes whose means differed significantly between classes, we performed 500 permutations of class labels using the t-test metric. t scores and corresponding P values were also calculated by standard methods, and P-values were adjusted to account for multiple hypotheses, as described in the Supplementary Information.
Quantitative, real-time PCR
Quantitative PCR was performed using a SYBR Green kit (Applied Biosystems) and either a PRISM 7500 sequence detector or the PRISM 7300 384-well sequence detector (Applied BioSystems) as described in the Supplementary Information.
FISH
Tissue samples used for interphase FISH analysis were derived from a melanoma tissue microarray described elsewhere28. BAC clones RP11-584A6 and RP11-444P10 spanning the MITF locus were obtained from the Children's Hospital of Oakland Research Institute (CHORI). BAC DNA was prepared with the NucloeBond Plasmid Mini Kit (Clontech). Probe labelling with Digoxin-dUTP (Roche) was performed as described29. Centromere probes were obtained from Vysis. The integrity and purity of all probes were verified by hybridization to metaphase spreads before tissue analysis. Tissue hybridization, washing, and bicolour detection were performed as described previously29. MITF amplification was defined as the presence of four or more gene copies compared to two copies of the chromosome-3 centromere probe in at least 50 nuclei.
Kaplan–Meier analysis
Survival curves for 160 metastatic melanoma patients (analysed by FISH as described above) were calculated using Kaplan–Meier analysis with assessment of statistical significance using the Mantel–Cox log-rank test. The Cox proportional-hazards model was used for multivariate analysis to determine relative risk and independent significance. Analyses were performed with Statview 5.0.1 (SAS Institute). Patients were deemed uncensored if they died of melanoma within 5 years of their initial date of diagnosis. Survival was measured from the time of initial diagnosis. Clinicopathologic correlations were determined as described in the Supplementary Information.
AQUA of MITF protein expression
AQUA image acquisition and analysis14 was performed as described in the Supplementary Information.
Soft agar assays
Cells (10,000 per well) were seeded in 0.5% low-melting-point agarose in DMEM with 10% FBS, layered onto 0.8% agarose in DMEM/10% FBS. Colonies were stained with crystal violet, enumerated and photographed at 4 magnification (40 microscopy) eight weeks after seeding.
Pharmacologic data analysis
Pharmacologic data (- log10[GI50]) for 42,796 compounds was downloaded from the NCI website and analysed by supervised learning methods as described in the Supplementary Information. GI50 is defined as the drug concentration achieving 50% growth inhibition relative to untreated cells.
Drug sensitivity and cell viability assays
MALME-3M cells were plated at 5,000 cells per well in a 96-well plate and infected at 500 MOI with adenovirus containing dominant-negative MITF (as described above). After 12 h, infected cells and controls were incubated with RPMI containing either 20 M docetaxel or 20 M cisplatin for 72 h. Cell viability following drug treatment was measured using the MTS-based CellTiter 96-cell proliferation assay (Promega).
