One Signal, Multiple Pathways: Diversity Comes from the Receptor

Type “Wnt” into Google Scholar, and you'll get nearly 72,000 hits, revealing the pivotal role this widely conserved family of signaling proteins plays in development and disease. Wnt proteins trigger complex signaling cascades that regulate cell growth, migration, differentiation, and many other aspects of development with the help of numerous interacting components. In the best-understood, “canonical” pathway, Wnt signaling molecules (called ligands) bind simultaneously to two coreceptors on the cell surface (Frizzled and LRP), allowing β-catenin proteins to stabilize (avoid destruction), enter the nucleus, associate with the transcription factor complex TCF/LEF, and activate genes involved in cell survival, proliferation, or differentiation. Inappropriate activation of β-catenin has been linked to several types of cancer.


INTRODUCTION
The incidence of melanoma is increasing at one of the highest rates for any form of cancer in the United States [1]. At present, there are no systemic agents available that significantly extend the lifespan of patients with advanced disease, and the key to improved survival in all affected individuals remains early diagnosis and treatment. While early stage disease may result in occasional deaths, there are no available tests to predict which early stage tumors have a high likelihood of progression and therefore a worse prognosis. Thus, an urgent need exists for the identification of molecular signatures of melanoma progression which can be used to develop accurate prognostic markers and effective targeted therapies. High-throughput gene expression profiling technologies offer an opportunity to uncover critical molecular events in the development and progression of human melanoma and can be used to design improved prognostic testing and effective treatment strategies. Previous transcriptome analyses in other malignancies have provided valuable information for the assessment of patient group classifications such as subgroups of patients that are likely to respond to a particular therapy [2]. Expression profiling of metastatic melanomas was able to identify previously unrecognized subtypes of disease and predict phenotypic characteristics which may be of importance to melanoma progression [3]. Further studies using serial analysis of gene expression (SAGE) and cDNA arrays have yielded the identification of additional novel molecules and pathways which may be involved in melanoma development [4][5][6][7]. Such studies have been limited in utility due to the lack of concordance from one study to the next suggesting tumor heterogeneity [8]. In addition, the limited availability of primary tissue from early stages of disease has hindered the ability to identify serial molecular events that lead to melanoma onset. This shortcoming of tissue availability has largely restricted gene expression profiling studies in melanoma to the use of small numbers of established tumor cell lines and cases of metastatic disease.
Previous studies of primary human melanomas have identified gene signatures associated with tumor progression [9][10][11]. These signatures included upregulation of cell cycle regulatory proteins, mitotic checkpoint genes, genes involved in DNA replication and repair, and cellular stress response genes in addition to loss of genes promoting apoptosis; however, few of the genes identified in these studies were concordant suggesting limitations due to tumor variability. Since better knowledge of gene expression signatures associated with melanoma progression may identify improved screening tools and therapeutic strategies, we used high density cDNA microarrays for gene expression profiling of genetically well-defined melanoma cell lines isolated from distinctive stages of tumor progression. Novel data reduction algorithms were used to identify gene signatures associated with tumor invasion and metastasis. Here we report unique sets of gene expression signatures that are associated with melanoma progression. Many of these pathways have previously been implicated in melanoma progression; however, the specific signature genes identified are novel. These particular progression-associated genes may reflect the underlying molecular mechanisms of the various phases in the known tumor progression pathways of melanoma. As such, the pathways and molecules identified in this study have the potential to be utilized as therapeutic targets for melanoma as well as novel molecular markers for melanoma progression. Moreover, these studies support the use of renewable sources of tumor cells, such as informative tumor cell lines, for the early identification of genes associated with malignant progression which can be subsequently validated using more precious primary tissue specimens.

RESULTS
In order to define gene expression patterns during the course of melanoma development and progression, we evaluated a series of primary and metastatic melanoma cells derived from lesions of discrete phases of melanoma progression as well as pools of primary human melanocytes. Tumor cell lines derived from three radial growth phase (RGP) melanomas (WM35, SBC12, and WM1552C), four vertical growth phase (VGP) melanomas (WM902B, WM278, WM983A, and WM793), and three metastatic melanomas (WM852, WM983B, 1205Lu) were evaluated. These cell lines possess a notable ability to recapitulate the clinical stages of disease from which they were derived [12,13] and have been characterized with respect to tumorigenicity and metastasis [14][15][16][17]; cellular growth characteristics including life span, growth factor dependency, anchorage-independent growth [12]; and pigmentation and morphology [18]. In addition, cytogenetic analyses in these cell lines including non-random abnormalities such as deletions, translocations, and amplifications have been well-documented and suggest high relevance to the primary tumor of origin (reviewed in [12]).
Global gene expression patterns were obtained using Affymetrix gene chips and comparison of gene expression profiles was performed using hierarchical clustering analysis. This clustering analysis identified two distinct groups of melanoma cell lines based on the similarity of their expression patterns, separating radial growth phase (RGP) and metastatic melanomas (MM) ( Figure 1); however, vertical growth phase (VGP) melanomas failed to form a distinctive cluster ( Figure 1A). The first group, which we characterized as ''less-aggressive'' primary melanomas (designated as Group1), included all three RGP melanomas (WM35, Sbcl2, and WM1552C) and two VGP melanomas (WM902B and WM278). The second group, which we characterized as ''moreaggressive'' melanomas (designated as Group2), included all three metastatic melanomas (WM853, WM983B, and 1205Lu) and two VGP melanomas (WM983B and WM793). Additional cluster analyses with different linkage matrices produced similar results (data not shown). Of note, only 2 of the 10 cell lines (Sbcl2 and WM853) were found to be wildtype for BRAF kinase and this genotype failed to demonstrate a notable cluster in our hierarchical analyses.
In order to identify a cohort of genes differentially expressed between our defined groups of melanomas, the gene expression array dataset was subjected to the microarray data analysis program Significance Analysis of Microarray (SAM) [19]. This analysis resulted in 142 differentially expressed probesets with a 3.5 % false discovery rate ( Figure 1B and C). In total, 89 probe sets representing 65 well-defined genes were found to be upregulated in more-aggressive (group 2) versus less-aggressive (group 1) melanomas and 53 probe sets representing 37 well-defined genes were found to be downregulated (Table S1). When more stringent criteria were applied (well-characterized genes which are differentially expressed greater than 4-fold) to this signature, we identified 21 upregulated and 5 downregulated genes in the more-aggressive melanoma cells ( Figure 1C). Of note, the set of genes highly expressed in more-aggressive melanomas includes many novel genes with reported functional roles in cell cycle regulation and proliferation such as ZWINT, CDCA2, NCAPH, NCAPG, NCAPG2, PBK, NUSAP1, BIRC5, ESCO2, HELLS, MELK, and CDKN2C [20][21][22][23][24][25][26][27][28][29][30][31] as well as genes that are involved DNA replication and repair processes including GINS1, GINS4, RAD54L, TYMS, and DHFR [32][33][34] (Table 1). Differential expression of these genes was validated by quantitative real-time RT-PCR ( Figure 1D).
Since we were interested in defining melanoma progression signatures, and all melanomas are initiated in primary human melanocytes, we evaluated our expression profiling data in the context of cultured neonatal primary human melanocytes ( Figure 2). Surprisingly, when two pools of short-term cultured primary human melanocytes (HPM1 and HPM2) were included in the previously employed hierarchical clustering protocol, the global gene expression pattern of the normal melanocytes was found to be more similar to that of the more-aggressive melanomas (Group 2) than the less-aggressive melanomas (Group 1) (Figure 2A). Since early cultures of primary human melanocytes derived from neonatal foreskins divide rapidly yet possess a normal differentiation program, we reasoned that the similarities of these cells to more aggressive tumors was likely due to their proliferative potential. In order to test this hypothesis we compared gene expression profiles of more-aggressive melanoma cells (Group 2) to those of short-term cultured primary human melanocytes. Expression profiles were subjected to SAM analysis which identified a cohort of differentially expressed genes with a 0.85% false discovery rate. Remarkably, all differentially expressed genes were found to be down-regulated genes in aggressive melanoma cells versus primary human melanocytes suggesting that loss of specific gene signatures may be a key event in the development of advanced melanomas ( Figure 2B). Further assessment of all melanoma expression profiles using TreeView revealed that the majority of these melanoma-associated genes are also downregulated in the less-aggressive primary melanomas (Group1) ( Figure 2C). This gene signature is comprised of critical mediators of cellular adhesion and melanocyte development and differenti-ation and includes: CDH3, CDH1, c-KIT, PAX3, CITED1/MSG-1, TYR, MELANA, MC1R, and OCA2 [35][36][37][38][39][40][41][42] (Table 2). While such a loss of cellular adhesion by E-cadherin and P-cadherin has been extensively documented in melanoma (reviewed in [43]), and the loss of differentiation-associated genes is not wholly surprising, this signature does notably identify specific defects in the intrinsic melanocyte development program that may contribute to melanoma development. In addition, genes with tumor suppressor and metastasis suppressor functions (DPP4, SYK) are included in this melanoma signature [44,45]. Significant down-regulation of these genes in the aggressive metastatic melanoma cells was validated using semi-quantitative duplex RT-PCR ( Figure 2D). Furthermore, this differentially expressed ''melanoma signature'' contains many genes whose functional roles in melanoma progression have not been well characterized and may provide novel insights into the early development of melanoma from primary melanocytes.
A current melanoma progression model suggests the sequential evolution of primary in situ tumors and minimally invasive tumors which are termed ''radial growth phase'', followed by a subsequent conversion to a more aggressive ''vertical growth phase'', in which tumor cells are programmed to cross the epidermal basement membrane and invade vertically into the dermis. It has been postulated that the VGP is the critical stage in which a tumor gains metastatic capacity. We therefore compared the gene expression profiles of RGP and VGP melanomas using a uniquely designed data reduction algorithm in order to identify genes that are likely to be relevant to this critical invasive phenotype ( Figure 3A). Our . B) SAM plot sheet illustrating a signature for differentially expressed genes in aggressive melanomas. Gene expression profiles from the two groups of melanomas were compared (Group1 vs. Group 2) and a differentially expressed gene signature was identified by SAM. Red and green dots represent gene probesets upregulated and downregulated respectively in Group 2. C) The melanoma gene signature was visualized using Java TreeView. Genes over four-fold differentially expressed are indicated on the right side of the image. D) Validation of select differentially expressed genes by real-time RT-PCR. Three genes upregulated in aggressive melanomas (Group 2) were selected for analysis and their differential expression was verified. 3.0 mg of total RNA was subjected to cDNA synthesis reaction as described in the materials and methods. 1.0 ml of the final cDNA samples (100 ml) were used for realtime Q-PCR reaction. For the measurement of gene transcript level, standard curves were generated for each gene using known amount of PCR amplified product from the corresponding genes. Error bars are SD of three independent experiments. doi:10.1371/journal.pone.0000594.g001 melanoma invasion-specific signature is notably characterized by the inclusion of several genes involved in chemotaxis and the inflammatory response (CXCL1, CXCL2, IL8, and IL6), cell adhesion (HNT, ITGA4, ITGB8, CSPG2, ZP4, and FLRT3), and extracellular matrix organization (MMP1, COL4A1, COL4A2, and COL5A2) ( Table 3). These genes and their relative expression profiles are depicted in Figure 3B. These cellular processes have previously been implicated in tumor progression for a wide variety of malignancies including melanoma and are felt to be essential components of tumor invasion and metastasis. In addition, many of these invasion-specific signature genes are also upregulated in metastatic melanomas ( Figure 3B). The differential expressions were validated on the selected genes using semi-quantitative duplex-PCR analysis ( Figure 3C).
Since the melanoma invasion-specific signature was associated with common functions of matrix invasion/inflammation/cell migration we sought to determine whether a common upstream regulatory pathway might link these signature genes. The eight most highly up-regulated genes from our melanoma invasionspecific signature were selected for further evaluation and gene promoter sequences were analyzed to identify transcription factor binding cis elements. This promoter analysis yielded a profile of transcription factors with common sequence elements in the signature genes. The most ubiquitous cis elements among the gene promoters evaluated were E12, E47, GCN4, GR, HES-1, IL-6, MEF-2, NF-KB, N-Oct-3, PU.1, RAR-alpha1, SRF, and the basal gene transcriptional complex components of TFIID, TBP, and TBF1. We notably identified the NF-KB binding sequence in 7 out of 8 of the most upregulated invasion-specific signature genes ( Figure 3D). We were particularly interested in the NF-KB pathway as a mediator of melanoma invasion since our most highly upregulated invasion-specific genes, CXCL-1 and IL-8, have previously been reported to be activated by NF-KB and had previously been implicated in melanoma progression (reviewed in [46]). Given the consequences of NF-KB activation in a cell, NF-KB function is highly regulated by specific cytosolic inhibitory activities which prevent inappropriate NF-KB activation and shuttling to the nucleus. Thus, only nuclear NF-KB is considered to be functionally activated. In order to evaluate NF-KB function in our tumor cell lines, we used NF-KB cellular localization as a surrogate marker for NF-KB activity. We find that invasive (VGP) melanomas posses both cytosolic (inactive) and nuclear (active) localization of NF-KB, while non-invasive (RGP) melanomas possess NF-KB confined to the cytosolic compartment suggesting specific activation of NF-KB during melanoma progression ( Figure 3E).

DISCUSSION
Molecular profiling studies of melanoma to date have been variably successful and often inconsistent. Much of this inconsistency has been attributed to the heterogeneous nature of this malignancy and the lack of significant sources of meaningful archived tissue specimens for analysis. In addition, variable sample preparation techniques are also likely to lead to disparate results between investigators. Here we have used a series of welldefined melanoma cell lines from varying stages of malignant progression to assess molecular signatures associated with disease progression. These cell lines have undergone extensive characterization of their tumorigenic potential and invasion capacity [14,47], and have been shown to possess a remarkable ability to recapitulate the clinical stages of disease from which they were derived. We show that unsupervised hierarchical clustering of global gene expression profiles of melanoma cell lines allows for the classification of tumor cells into 2 groups ( Figure 1A) that we have defined as less-aggressive (Group 1) and more-aggressive (Group 2) melanomas. While all radial growth phase melanomas clustered in Group 1, and all metastatic melanomas clustered in Group 2, vertical growth phase melanomas failed to form a distinctive cluster suggesting that vertical growth phase melanomas may be considered to be a transient or transition phase within the current melanoma progression model [48]. Aggressive (Group 2) melanomas were characterized by upregulation of genes associated with cell cycle progression, DNA replication and repair, and altered expression of apoptosisrelated genes including upregulation of the antiapoptotic gene BIRC5/survivin [49] and downregulation of the novel stressassociated apoptosis inducer TRIB3 [50] (Table 1). These signature genes for melanoma progression are remarkably similar to those obtained from recent large-scale studies using primary human melanomas and suggest high correlation with alterations seen in primary tumor specimens [10]. Notably, we did not identify a dominant signature associated with BRAF kinase mutations which may be reflective of the relative infrequency of wildtype BRAF in these cell lines. As a whole, this gene signature suggests a series of molecular alterations occur in aggressive melanomas that promote melanoma cell growth, survival and apoptotic resistance which contribute to the unresponsiveness of melanomas to traditional chemotherapeutic agents [51].  While gene signatures associated with aggressive melanomas provide insights into molecular pathways important for tumor progression, further analysis of these tumor cell lines in conjunction with expression profiles from primary human melanocytes using SAM analysis revealed a striking signature characterized exclusively by gene loss in melanomas and primarily by loss of cellular adhesion and melanocyte differentiation-associated genes ( Figures 2B, 2C). We suggest that this melanoma-associated signature defines critical molecular mechanisms involved in melanocyte development and differentiation which distinguish these tumor cells from their primary cell of origin. In fact, the identification of several genes in this signature with established functional roles in melanocyte differentiation and melanin biosynthesis such as CDH3, CDH1, c-KIT, PAX3, CITED1/MSG-1, TYR, MELANA, MC1R, and OCA2 [35][36][37][38][39][40][41][42], supports this notion. Moreover, this signature has identified two tumor suppressor genes, DPP4 and SYK, whose downregulation has previously been implicated in melanoma development [52][53][54].
Finally, evaluation of an invasion-specific signature for melanoma identified dominant gene activation by the transcription factor, NF-KB. Constitutive activation of NF-KB and an inflammatory response is an emerging hallmark of various tumor types [55]. In addition, NF-KB has specifically been implicated in the development of invasive aggressive melanomas through autocrine and paracrine mechanisms (reviewed in [46]). Our melanoma invasion signature is associated with upregulation of critical NF-KB effectors including CXCL1, FMN2, MMP1, IL-8, IGFBP3 which have been implicated in the regulation of tumor cell proliferation, motility, migration, and/or invasion ( Figure 3D). In addition, the putative NF-KB target gene GAGE7B which we identified in our melanoma invasion-specific signature, has been associated with apoptotic resistance and worse prognosis in other tumors [56]. In addition, several of our invasion-specific signature genes are chemokines including CXCL1, CXCL2, and IL-8 which have been implicated in the promotion of tumor-associated angiogenesis, a critical feature of invasive tumors [57].
In summary, our gene expression profiling studies of melanoma cell lines from varying stages of malignant progression and primary human melanocytes have identified several important melanoma signatures including: 1) Aggressive melanomas are characterized by upregulation of genes associated with cell cycle progression, DNA replication and repair and apoptotic resistance as well as loss of genes associated with apoptotic susceptibility, 2) Melanomas notably differ from their cell of origin, primary human melanocytes, due to a loss of cellular adhesion and differentiationassociated genes, and 3) Invasive melanomas are characterized by a signature indicative of global activation of NF-KB and downstream effector genes associated with tumor cell migration, invasion, chemotaxis, and proliferation. Since pathways associated with tumor progression may have clinical utility as prognostic tumor markers and therapeutic targets, we expect novel melanoma signature genes identified in this study will be further developed for such translational endpoints. Moreover, the important information regarding melanoma biology gleaned from these studies on renewable cell resources cannot be understated. A major roadblock to advances in melanoma therapy has been the relative paucity of informative tissue specimens available for analysis in profiling studies as well as the notoriously heterogeneous nature of this malignancy. The use of surrogate tissue resources including tumor cell lines for the early discovery phases in melanoma, as used in this study, will undoubtedly allow for the conservation of precious tissue specimens for use in more advanced validation studies. It is expected that the novel melanoma progressionassociated genes identified in this study will provide new insights into the molecular defects associated with this malignancy and ultimately pave the way for the development of new melanoma biomarkers and novel targeted therapies.

Gene Expression Profiling
Total RNA was isolated from exponentially growing melanoma cell lines using RNeasy column purification per manufacturer's protocol (Qiagen). Two sets of short-term cultured (2 to 3 passage numbers) normal human melanocytes were prepared from neonatal foreskins. In order to minimize genetic variability melanocytes from 4-5 individuals were pooled for each culture. Total RNA from normal melanocytes was extracted and purified by a combination of phase extraction and chromatography using TRIzol reagent (Invitrogen Life Technologies Inc.) and RNeasy columns (Qiagen) in order to remove melanin. In brief, exponentially growing melanocytes were lysed with TRIzol reagent and lysate was incubated at 65uC for 2 minutes to inactivate melanin. Lysate was then subjected to phase extraction and RNeasy column purification. RNA quality checks, double strand complementary DNA synthesis, hybridization with Human Genome U133 Plus 2.0 Array Chips (Affymetrix Inc. Santa Clara, CA), and initial data extraction were performed at The Gene Array Core Facility in the Malaria Research Institute (JHMRI) at The Johns Hopkins Bloomberg School of Public Health (http:// malaria.jhsph.edu/jhmri/resources_education/gene_array_core).

Data Extraction and Statistical Analysis
SAM [19], Gene Cluster 3.0 [58] and TreeView (http://bonsai.ims. u-tokyo.ac.jp/,mdehoon/software/cluster/index.html), Access, and Excel (Microsoft, Seattle, WA) programs were used. For all of the statistical analysis beyond the initial description of datasets, microarray data were normalized (Dataset S1) and a subset of the 12 microarray data (10 from melanoma cell lines and 2 from normal human melanocytes) was obtained by filtering to require each gene probe to have at least one observation in the expression intensity resulting in a 'present' call from all 12 samples. This produced a subset of data containing a total of 32,632 affymetrix gene probes (Dataset S2). This filtered subset of data was used for all of the additional analysis. Cluster Analysis. Unsupervised hierarchical clustering analysis was performed on the subset of data (without log transformation) with Gene Cluster 3.0 (http://bonsai.ims.utokyo.ac.jp/,mdehoon/software/cluster/index.html) by using the correlation (uncentered) similarity metric and centeroid linkage clustering method. The resulting tree-images were visualized using Java TreeView. Statistical Analysis of Microarray (SAM). SAM was performed on the subset of array data without log transformation using SAM software package. Groups are defined based on the hierarchical clustering; for example, group 1 = less-aggressive primary melanomas (RGP melanomas: WM35, Sbcl2, WM1552C and VGP melanomas: WM902B and WM278), and group 2 = aggressive metastatic melanomas (Metastatic melanomas: WM852, WM983B, and 1205Lu; and VGP melanomas: WM983A and WM793) as seen Figure 1. Delta was chosen to limit the output gene list so that minimum predicted false-positives would be included. Three-step Data Reduction Algorithm. In order to identify melanoma invasion-specific gene signature, uniquely designed three-step data reduction algorithm was applied to the subset of expression data. First step is that the proveset should be called as 'present' in three samples out of four VGP melanoma cells and two samples out of three RGP melanoma cells. Second step is that the candidate proveset should be expressed five folds or more in VGP melanoma cell lines than RGP melanoma cell lines. The last step is that the gene probesets were retained only when the expression level is greater than three folds in VGP melanoma cell lines compared to that of primary human melanocytes. The last step is implemented because the candidate gene expression level should be higher if the gene products have certain degree of functional roles in the invasion processes of malignant melanoma. The probesets, those that pass through three-step filtration criteria are subjected to probeset to gene mapping using NetAffx, a web interface program from Affymetrix Inc. Gene annotation for the gene mane, gene symbol, and GO Biological Procession Analysis also performed by the NetAffx.  to ensure single amplified product. All samples were performed in triplicate. Amplification data were analyzed with an Applied Biosystems Prism Sequencer Detection Software Version 2.3 (Applied Biosystems, Forster City, CA). Human GAPDH gene was used as endogenous control. To normalize the relative expression of the genes of interest to the GAPDH control, standard curves were prepared for each gene and GAPDH in each experiment.

Semi-quantitative Duplex PCR
Semi-quantitative duplex RT-PCR was performed by an MJ Research Programmable Thermal Controller (PTC-100, Inc., Watertown, MA) and the amplified products were separated on an agarose gel. Our duplex PCR utilized 20 bp oligonucleotides to amplify regions of 300-400 bp from the genes of interest. Intitial optimization experiments were conducted to establish the most favorable primer concentrations between the genes of interest and internal control GAPDH, yielding 0.8 mM and 0.04 mM, respectively. The PCR was carried out in a total volume of 25 mL, containing 2.5 mL of 10X PCR Buffer (containing 15 mM MgCl 2 ), 0.2 mM dNTPs, and 0.3 ul AmpliTaq Gold DNA Polymerase (Applied Biosystems, Foster City, CA). Thirty to thirty-five amplification cycles were performed by an MJ Research Programmable Thermal Controller (PTC-100, Inc., Watertown, MA), using a denaturing temperature of 95uC for 25 seconds, an annealing temperature varying between 55uC260uC (depending on gene) for 30 seconds, and primer extension at 72uC for 30 seconds. Each amplification experiment also included two negative PCR controls, a no-RNA control from reverse transcription procedures and a no-cDNA water control. Following amplification, 25 mL of the samples were separated via electrophoresis on a 3% agarose gel. The primer sequences were designed by using Primer3, primer analysis software (http://frodo. wi.mit.edu/cgi-bin/primer3/primer3_www.cgi), yielded only one amplified product and had the following sequences:

Immunofluorescence Labeling
RGP and VGP Cells were plated on glass slides and cultured in melanocyte growth media overnight without any stimulation. Cells were fixed at room temperature for 15 minutes using 3.5% paraformaldehyde solution. Cells were washed briefly with PBS and then permeabilized with either 0.5% Triton X-100 for 10 minutes or 220uC cooled methanol for 15 minutes. Slides were blocked with 16% normal goat serum (Santa Cruz Biotech., Santa Cruz, CA) for 1 hour and then incubated with rabbit polyclonal IgG p65 antibody (Santa Cruz Biotech., Santa Cruz, CA) at 1:100 dilution. Subsequent to overnight incubation at 4uC, the slides were washed with PBS and incubated with goat anti-rabbit IgG-Alexa 594 (Molecular Probes Eugene, OR) at 1:200 dilution at room temperature for 1 hour. Stained slides were washed with PBS and viewed under a fluorescence microscope (Eclipse TS100, Nikon, Tokyo, Japan).

Gene Transcription Promoter Analysis
Transcription factor binding cis element sequence profiling in the selected gene promoter was performed by using a web tool known as TESS (Transcription Element Search System, http://www.cbil. upenn.edu/tess). Each genes promoter sequences from the transcription start site up to 2.0 kb of upstream of the genes were subjected to the TESS and screened by TRANSFAC database to identify matched consensus sequences of known DNA binding transcription factors.

SUPPORTING INFORMATION
Dataset S1 A combined and normalized raw dataset from the 12 sets of microarray data (two sets of primary human melanocytes and ten melanoma cell lines).

ACKNOWLEDGMENTS
We thank M. Herlyn (Wistar Institute) for generously providing melanoma cell lines used in these studies. We thank G. Parmigiani (Johns Hopkins University School of Medicine) for his statistical support and members of the Alani Lab for their careful review of this manuscript and helpful comments.