Metastatic disease is the primary cause of death in cutaneous malignant melanoma (CMM) patients. To understand the mechanisms of CMM metastasis and identify potential predictive markers, we analyzed gene-expression profiles of 34 vertical growth phase melanoma cases using cDNA microarrays. All patients had a minimum follow-up of 36 months. Twenty-one cases developed nodal metastatic disease and 13 did not. Comparison of gene expression profiling of metastatic and nonmetastatic melanoma cases identified 243 genes with a >2-fold differential expression ratio and a false discovery rate of <0.2 (206 up-regulated and 37 down-regulated). This set of genes included molecules involved in cell cycle and apoptosis regulation, epithelial-mesenchymal transition (EMT), signal transduction, nucleic acid binding and transcription, protein synthesis and degradation, metabolism, and a specific group of melanoma- and neural-related proteins. Validation of these expression data in an independent series of melanomas using tissue microarrays confirmed that the expression of a set of proteins included in the EMT group (N-cadherin, osteopontin, and SPARC/osteonectin) were significantly associated with metastasis development. Our results suggest that EMT-related genes contribute to the promotion of the metastatic phenotype in primary CMM by supporting specific adhesive, invasive, and migratory properties. These data give a better understanding of the biology of this aggressive tumor and may provide new prognostic and patient stratification markers in addition to potential therapeutic targets. [Cancer Res 2007;67(7):3450–60]

In primary cutaneous malignant melanoma (CMM) patients, it is essential to determine the molecular changes associated with metastasis and to apply this knowledge to the fields of outcome prediction and targeted treatment. This information will lead to a better understanding of the biology of this tumor, and will probably provide prognostic information for defining subgroups of patients with a less favorable prognosis as potential candidates for adjuvant or novel therapies.

Currently, the prognosis of primary CMM is mainly based on histopathologic criteria. The most important of these is the Breslow index, although it is merely a measure of tumor depth. New molecular markers that correlate with melanoma genesis and/or progression are continuously being identified but, to date, most of them have been obtained in experimental models and have not yet been confirmed in series of human samples.

The development of high-throughput screening techniques in genomics and proteomics has enabled the analysis of the expression of multiple genes and proteins in large series of tumor samples (1) and may contribute to the resolution of some specific issues of clinical relevance, such as identifying the steps for melanoma progression and metastasis. Until now, most studies characterizing CMM have been done using cell lines (2, 3), mouse models (4), or metastatic tumor samples (2, 5). However, data obtained from primary CMM samples are difficult to obtain, due in part to the lack of retrospective collections of frozen primary melanomas samples or adequate follow-up (57).

Molecular changes associated with the acquisition of metastatic capacity in vertical growth phase melanoma are still to be fully described. We might expect a repertoire of differentially expressed genes defining the metastatic phenotype for primary CMM cases. The present study aimed to investigate the metastasis signature in primary CMM. To this end, we compared the gene expression profile in a series of vertical growth phase primary CMMs that had developed metastatic disease with vertical growth phase CMM without metastatic disease by the end of follow-up. Our results confirm the multifactorial genesis of melanoma metastasis and identify the epithelial-mesenchymal transition (EMT) and the relation with the extracellular matrix as key steps in melanoma progression.

Patients and tissue samples. This study featured a series of 34 primary CMMs provided by the Hospital San Cecilio (Granada, Spain) and the Hospital 12 de Octubre (Madrid, Spain). Patients were recruited between 1990 and 2002 to ensure a minimum follow-up of 36 months. Samples were collected and frozen according to standard protocols, and their histology was reviewed by two pathologists (S.R.A. and J.L.R-P.) to confirm that they contained at least 50% tumor cells. All selected cases corresponded to consecutive melanoma tumors with potential metastatic capacity (i.e., vertical growth phase cases with a Breslow index >1 mm). The cutoff was selected taking into account the criteria of the American Joint Committee on Cancer by which thin melanomas (≤1 mm) have an excellent prognosis (>90% survival at 5 years) compared with melanomas with a penetration depth of >1 mm (8). Histologic review of the cases was done on paraffin-embedded tissue, whereas frozen sections were examined for assessing the proportion of the tumoral cells and the character of the neoplastic infiltrate.

For tissue microarray (TMA) analysis, we used a retrospective cohort of patients representing 127 primary vertical growth phase melanoma cases (formalin-fixed and paraffin-embedded tissue) collected from 1980 to 2000. These cases were obtained from the Hospital 12 de Octubre (Madrid) and were included in six separate TMAs (1.5-mm core diameter), with two representative duplicate cores for each case (9), and constructed with a manual tissue arrayer (Beecher Instruments, Sun Prairie, WI) using a standard method (10). Patient medical records were reviewed to gather information on age, gender, localization, tumor thickness, distant invasion (lymph node or skin), and follow-up of at least 36 months. Patients were not treated before the development of metastasis. The work was conducted in accordance with the Declaration of Helsinki Principles and under the supervision of the Hospital 12 de Octubre Ethics Committee.

RNA isolation and amplification. Total RNA was extracted from frozen samples of primary CMM using TRIzol reagent (Invitrogen, Carlsbad, CA) followed by purification with the RNeasy Mini kit (Qiagen Inc., Valencia, CA) and digestion with RNase free DNase I according to the manufacturer's instructions. RNA quality and integrity were verified using the Bioanalyzer system (Agilent Technologies, Palo Alto, CA). Double-stranded cDNA was synthesized from 5 μg of total RNA using the Superscript Choice System for cDNA synthesis (Invitrogen) with an oligo-dT primer containing a T7 RNA polymerase promoter. In vitro transcription was carried out using the T7 Megascript in vitro transcription kit (Ambion, Austin, TX) as previously described (11, 12). The quality of the amplified RNA produced was checked by electrophoresis and its concentration was measured.

Microarray procedures: preparation/synthesis of fluorescent cDNA and hybridization. For each melanoma sample, 5 μg of amplified RNA (aRNA) were directly labeled with cyanine 3-conjugated dUTP (Cy3), whereas 5 μg of aRNA from the Universal Human Reference RNA (Stratagene, La Jolla, CA) were labeled with cyanine 5–conjugated dUTP (Cy5). The Centro Nacional de Investigaciones Oncológicas (CNIO) OncoChip platform (v.1.4) was used in all cases to perform the cDNA microarray procedure. Basically, the CNIO OncoChip is a cDNA microarray especially designed for the analysis of genes involved in cancer. It includes 2,489 cancer-relevant genes in addition to genes involved in drug response, tissue-specific genes, and control genes. The platform used has a total of 6,386 genes represented by 7,237 human clones purchased from Research Genetics (Huntsville, AL). A list of these genes can be found online.8

Sample hybridizations were done as described elsewhere (12). After washing, the two fluorescent signals on the slides were scanned with a standard two-color microarray scanner (Scanarray 5000XL, GSI Lumonics, Kanata, Ontario, Canada). Images were analyzed with the GenePix 4.1 software (Axon Instruments, Inc., Union City, CA). The clone sequences of all the genes included in the OncoChip and the reproducibility of the expression data of multiple genes have been previously verified (12).

Data extraction and analysis. Data from each hybridization were maintained in a database for analysis. Fluorescence intensity measurements were subjected to automatic background subtraction. The Cy3/Cy5 ratios were normalized to the value of the median ratio of all spots in the array. The sum of the median background for each channel was calculated, and spots with total intensities less than the calculated sum of median backgrounds were discarded. Additionally, spots with background-subtracted signal intensities <500 fluorescence units (sum of the two channels) and bad spots were excluded from the analysis. All ratio values were log-transformed (base 2), and duplicated spots in the array were averaged. Inconsistent duplicates were discarded and all consistent duplicate spots and genes were averaged. In addition, genes for which fewer than 70% of the potential data were available were excluded from further analysis. The median expression of each gene was calculated for each patient group (metastatic and nonmetastatic primary CMM). Differences in expression between the two groups were analyzed by the Wilcoxon test. To account for the effect of multiple hypothesis testing on the identification of significance, adjusted P values were also computed using the method proposed by Hochberg and Benjamini (13) for controlling the false-discovery rate (FDR). A single value for each gene was obtained. For our purposes, a gene was deemed to be up-regulated or down-regulated if there was an at least 2-fold difference in expression and if the FDR value was <0.2. The regulated genes were functionally classified manually on the basis of exhaustive searches in PubMed, the Genecards database, and Gene Ontology.

TMA immunohistochemistry. We undertook immunohistochemical analysis to identify at the protein level a selection of the statistically and biologically significant genes obtained in the cDNA microarray study. This was done in a TMA series of 127 vertical growth phase melanoma cases with known follow-up. The proteins selected for validation featured the group that proved to be biologically relevant in melanoma metastasis development, including glypican 3 (polyclonal, Santa Cruz Biotechnology, Santa Cruz, CA), N-cadherin (3B9, Zymed, San Francisco, CA), osteonectin/SPARC (15G12, Novocastra, Newcastle, United Kingdom), osteopontin (polyclonal, Abcam, Novus Biologicals, Littleton, CO), and protein kinase Cα (PKCα; H7, Santa Cruz Biotechnology, Santa Cruz, CA). A heat-induced, epitope-retrieval step was done in a solution of sodium citrate buffer (pH 6.5) in the case of osteonectin/SPARC, osteopontin, and PKCα, and with EDTA buffer in the case of glypican 3 and N-cadherin. The slides were then heated for 2 min in a conventional pressure cooker and rinsed in cool running water for 5 min. They were then quickly washed in TBS (pH 7.4), and incubated with the selected primary antibodies glypican 3 (1:25), N-cadherin (1:10), osteonectin/SPARC (1:25), osteopontin (1:1500), and PKCα (1:25). Immunodetection was done with biotinylated secondary antibodies, followed by peroxidase-labeled streptavidin biotin (Dakocytomation, Glostrup, Denmark) visualization in the case of glypican, osteopontin, and PKCα and Envision (Dakocytomation) for N-cadherin and SPARC. Diaminobenzidine chromogen was used as peroxidase substrate. All immunostaining was done in a TechMate 500 automatic immunostaining device (DAKO, Glostrup, Denmark). Incubations omitting the specific antibody were used as a control of the technique. Scoring of the results and selection of the thresholds, internal controls for the antibody reactivity, and tissue controls for the series were done according previously published methods (9).

Scoring systems. Immunostaining results were evaluated by two different pathologists (S.R.A. and J.L.R-P.) and scored using clear cutoff criteria to facilitate the reproducibility of the method. Discrepancies were resolved by simultaneous reevaluation. Briefly, the result was recorded as positive or negative and high versus low expression, taking into account the expression in tumoral cells and the specific cutoff for each marker (see Table 1 for description of thresholds). As a general criterion, the cutoffs were selected to facilitate reproducibility and, when possible, to translate biological events. Scoring for the sample replicates was highly reproducible in this series (96%).

Table 1.

Antibodies used in the study indicating clone, source, dilution, visualization method, scoring, threshold, and positive controls

ProteinCloneSourceDilutionVisualization systemScoringThresholdPositive control
Glypican 3 Polyclonal Santa Cruz Biotechnology 1:25 LSAB/DAB Pos/neg >10% positive cells Hepatocarcinoma 
N-cadherin 3B9 Zymed 1:10 LSAB/DAB Pos/neg ≥5% positive cells, membranous expression Ovarian carcinoma, cardiac muscle 
SPARC/osteonectin 15G12 Novocastra 1:25 LSAB/DAB Pos/neg >10% positive cells, cytoplasm Endothelial cells in malignant tumors 
Osteopontin Polyclonal Abcam 1:1,500 LSAB/DAB Pos/neg >10% positive cells, cytoplasm Stromal cells in normal skin 
PKCα H7 Santa Cruz Biotechnology 1:25 LSAB/DAB High/low ≥50% positive cells, cytoplasm Small lymphocytes, nevus 
ProteinCloneSourceDilutionVisualization systemScoringThresholdPositive control
Glypican 3 Polyclonal Santa Cruz Biotechnology 1:25 LSAB/DAB Pos/neg >10% positive cells Hepatocarcinoma 
N-cadherin 3B9 Zymed 1:10 LSAB/DAB Pos/neg ≥5% positive cells, membranous expression Ovarian carcinoma, cardiac muscle 
SPARC/osteonectin 15G12 Novocastra 1:25 LSAB/DAB Pos/neg >10% positive cells, cytoplasm Endothelial cells in malignant tumors 
Osteopontin Polyclonal Abcam 1:1,500 LSAB/DAB Pos/neg >10% positive cells, cytoplasm Stromal cells in normal skin 
PKCα H7 Santa Cruz Biotechnology 1:25 LSAB/DAB High/low ≥50% positive cells, cytoplasm Small lymphocytes, nevus 

Abbreviations: Pos/neg, positive/negative; LSAB, peroxidase-labeled streptavidin biotin; DAB, diaminobenzidine.

Statistical analysis. To validate protein expression in the series analyzed by TMA, the relationship between marker expression in the patient samples and disease-free survival (DFS) curves were derived by the Kaplan-Meier method. Statistical significance of associations between individual variables and DFS was determined using the log-rank test. Cox univariate proportional hazard analysis was also done independently for each variable. Significance was concluded for values of P < 0.05. All statistical tests were two-sided. A multivariate model including the Breslow index as a continuous variable was also developed by backward elimination. All statistical analyses were carried out using the Stata statistical program (StataCorp 2001, release 9.0).

Clinical features. The study included two sets of patients. The training set was composed of 34 patients (14 males and 20 females) with primary vertical growth phase CMM and >1 mm (Breslow index). After a median follow-up of 67.3 months (range 16–166 months), 21 developed metastasis (median 9.4 months, range 0.73–139).

The validation set of patients, analyzed by TMAs, included 127 patients (57 males and 70 females) with primary vertical growth phase CMM and a median age at diagnosis of 60 years (range 21–91 years). The median follow-up was 116.8 months (range 1–276.5 months), during which 73 patients developed metastatic disease and 54 did not (median 52.7 months, range 1–222). Figure 1 shows an outline of the study.

Figure 1.

Representative diagram of the study profile. Metastatic disease was considered as end point for DFS. VGP, vertical growth phase; CMM, cutaneous malignant melanoma; EMT, epithelial mesenchymal transition; FDR, false discovery rate; DFS, disease-free survival; TMA, tissue microarray. K-M, Kaplan-Meier.

Figure 1.

Representative diagram of the study profile. Metastatic disease was considered as end point for DFS. VGP, vertical growth phase; CMM, cutaneous malignant melanoma; EMT, epithelial mesenchymal transition; FDR, false discovery rate; DFS, disease-free survival; TMA, tissue microarray. K-M, Kaplan-Meier.

Close modal

Global changes in gene expression between metastatic and nonmetastatic primary vertical growth phase CMM cases. To identify genes that might be potential markers for melanoma metastases, cDNA microarray studies were done to compare primary invasive CMM cases with and without metastatic disease.

From the 5,253 clones, corresponding to 2,945 known genes and 1,984 expressed-sequence tags that were suitable for analysis after filtering steps, 243 genes were differentially expressed (>2-fold, FDR<0.2) in primary vertical growth phase melanomas with metastases compared with primary nonmetastatic vertical growth phase melanomas. Of these, 206 genes were up-regulated and 37 genes were down-regulated (see Supplementary Table S1 for details). These genes were categorized according to their main biological function and/or mechanism of action. As seen in Table 2, six biological process classes (cell cycle and apoptosis, EMT, immune modulation, metabolism, nucleic acid binding and transcription, protein synthesis and degradation, and signal transduction) were identified as relevant in our metastatic gene signature. A group of genes with unknown function was also identified (see Supplementary Data for details). Figure 2 shows the expression profile of the significant genes in the training series classified by these gene-functional categories.

Table 2.

Functional categories of differentially expressed genes

Functional categoriesGene name
Cell cycle and apoptosis MAPRE1, SKP1A, FAIM, PMP22, SET, CDKN1A, CDC5L, WEE1, CASP5, DAD1, APEX1, VDAC1, BCCIP, CKS1B, CYC1, HBXIP, TP53I3, KIF11, BCL2L10, TOB2 
EMT LUM, SDCBP, MFAP1, CTSB, PRKCA, RAB1A, H2-ALPHA, RRAGA, TUBB3, CLIC4, SPARC, ANLN, RAN, ENC1, EMP1, DSG2, SEPP1, ITGAV, TUBA2, HMMR, MASA, TUBA3, CX3CR1, CDH2, EDNRB, CSPG2, CD63, SMARCA1, KIT, KRT8, ANXA5, FGG, PFN1, SPA17, FZD1, APLP2, SERPINA3, SEMA3B, MMP2, TUBA1, EDN2, PCDH9, CDH10, WNT2, PAPPA 
Immune modulation CXCL12, MAGEA6, IFITM2, CCBP2 
Metabolism PRDX5,VKORC1, ACADM, ALDH1A2, GLO1, HIBADH, PMM1, NDUFAB1, NDUFB11, ENO1, MDH1, ACSL3, ATP6V0B, MGC4172, OSBP, PFKM, SLC39A1, ATP6V0D1, YWHAQ, LRPAP1, ADK, SLC30A6, GGH, GSTA4, DCK, ATP5L2, LIPA, COX8A, CA9, AKR7A2, SCD, AK3L1, TXNRD1, SLC25A17, NPC2, NDUFB3, PAICS, COX6B1, ADSL, CYP7B1, CYP19A1, IDS 
Nucleic acid binding and transcription factors NCOA4, HIST1H4C, RBM4, MAP4K3, NOLA2, BZW1, ARID5B, CBFB, POLR3K, TRIP3, RDBP, ASH2L, CDYL, XRCC5, HIST1H4B, KLF10, TCEB1, HIST1H3D, GTF2H2, FOXO1A, REST, OLIG2 
Protein synthesis and degradation HSPA8, MRPL15, CCT5, EIF5, UBE2D2, MRPL37, MRPS10, RPL10A, VBP1, RPS22, UBL5, CCT6A, GC20, EIF3S3, RPL35A, SECISBP2 
Signal transduction CAV1, NME2, SRI, PPP1R3C, S100A10, WBP5, GABARAP, IGFBP1, PPP1R8, C11orf15, PRKAR1A, PGR1_GPR153, SEC5L1, CGI-141, MYB, SPRY2, PPP2CB, STK24, GNG12, GRB2, MAL, SHB, DUSP12, TGFBR2, PC4, IDE, CGA, IL1RL1, PPP2R5C, TIMM10, PDE5A, RABAC1, GIMAP6, TXK, EPS15, NTF5, AKAP13 
Functional categoriesGene name
Cell cycle and apoptosis MAPRE1, SKP1A, FAIM, PMP22, SET, CDKN1A, CDC5L, WEE1, CASP5, DAD1, APEX1, VDAC1, BCCIP, CKS1B, CYC1, HBXIP, TP53I3, KIF11, BCL2L10, TOB2 
EMT LUM, SDCBP, MFAP1, CTSB, PRKCA, RAB1A, H2-ALPHA, RRAGA, TUBB3, CLIC4, SPARC, ANLN, RAN, ENC1, EMP1, DSG2, SEPP1, ITGAV, TUBA2, HMMR, MASA, TUBA3, CX3CR1, CDH2, EDNRB, CSPG2, CD63, SMARCA1, KIT, KRT8, ANXA5, FGG, PFN1, SPA17, FZD1, APLP2, SERPINA3, SEMA3B, MMP2, TUBA1, EDN2, PCDH9, CDH10, WNT2, PAPPA 
Immune modulation CXCL12, MAGEA6, IFITM2, CCBP2 
Metabolism PRDX5,VKORC1, ACADM, ALDH1A2, GLO1, HIBADH, PMM1, NDUFAB1, NDUFB11, ENO1, MDH1, ACSL3, ATP6V0B, MGC4172, OSBP, PFKM, SLC39A1, ATP6V0D1, YWHAQ, LRPAP1, ADK, SLC30A6, GGH, GSTA4, DCK, ATP5L2, LIPA, COX8A, CA9, AKR7A2, SCD, AK3L1, TXNRD1, SLC25A17, NPC2, NDUFB3, PAICS, COX6B1, ADSL, CYP7B1, CYP19A1, IDS 
Nucleic acid binding and transcription factors NCOA4, HIST1H4C, RBM4, MAP4K3, NOLA2, BZW1, ARID5B, CBFB, POLR3K, TRIP3, RDBP, ASH2L, CDYL, XRCC5, HIST1H4B, KLF10, TCEB1, HIST1H3D, GTF2H2, FOXO1A, REST, OLIG2 
Protein synthesis and degradation HSPA8, MRPL15, CCT5, EIF5, UBE2D2, MRPL37, MRPS10, RPL10A, VBP1, RPS22, UBL5, CCT6A, GC20, EIF3S3, RPL35A, SECISBP2 
Signal transduction CAV1, NME2, SRI, PPP1R3C, S100A10, WBP5, GABARAP, IGFBP1, PPP1R8, C11orf15, PRKAR1A, PGR1_GPR153, SEC5L1, CGI-141, MYB, SPRY2, PPP2CB, STK24, GNG12, GRB2, MAL, SHB, DUSP12, TGFBR2, PC4, IDE, CGA, IL1RL1, PPP2R5C, TIMM10, PDE5A, RABAC1, GIMAP6, TXK, EPS15, NTF5, AKAP13 

NOTE: Categories are based on Gene Cards, National Center for Biotechnology Information, HUGO gene nomenclature committee, and PubMed. EMT is a category that includes genes associated with cell adhesion, cell motility, migration, and extracellular matrix organization.

Figure 2.

Results of global expression profiling between primary CMM with metastasis at follow-up (red squares) and primary CMM without metastasis (yellow squares). Expression data for each gene in each profiled sample are presented in the form of a heat map (green to red scale) of log2-transformed ratios. Names on the right represent the main functional category for each gene. Up-regulated genes are at the top and down-regulated genes are at the bottom.

Figure 2.

Results of global expression profiling between primary CMM with metastasis at follow-up (red squares) and primary CMM without metastasis (yellow squares). Expression data for each gene in each profiled sample are presented in the form of a heat map (green to red scale) of log2-transformed ratios. Names on the right represent the main functional category for each gene. Up-regulated genes are at the top and down-regulated genes are at the bottom.

Close modal

Interestingly, one of the most important functional clusters recognized here as markers of metastatic melanoma included genes involved in EMT, a category that includes genes associated with cell adhesion, cell motility, migration, and extracellular matrix interaction and organization (Table 3). This cDNA microarray screening identified several genes involved in the EMT phenomenon (N-cadherin, SPARC, and WNT receptor frizzled) and the physical link between the actin cytoskeleton, the extracellular matrix (mainly mediated by receptors of the integrin family, such as integrin αV), and supporting vessels (endothelin receptor B). Interestingly, the increase in the expression of N-cadherin is associated with the loss of the type II cadherin 10 (CDH10).

Table 3.

EMT-related genes

Gene nameGene descriptionn-fold changeLog2 change
LUM Lumican 12.278 3.618 
SDCBP Syndecan binding protein (syntenin, melanoma differentiation-associated protein 9-MDA; pro-TGF-α cytoplasmic domain-interacting protein 18; scaffold protein Pbp1) 6.753 2.756 
MFAP1 Microfibrillar-associated protein 1 4.147 2.052 
CTSB Cathepsin B 3.487 1.802 
PRKCA Protein kinase Cα 3.207 1.681 
RAB1A RAB1A member RAS oncogene family 2.986 1.578 
H2-ALPHA α-Tubulin isotype H2-α 2.934 1.553 
RRAGA Ras-related GTP binding A 2.917 1.545 
TUBB3 Tubulin β3 (other designation: tubulin β4) 2.874 1.523 
CLIC4 Chloride intracellular channel 4 2.870 1.521 
SPARC Secreted protein, acidic, cysteine-rich (osteonectin) 2.840 1.506 
ANLN Anillin, actin binding protein 2.764 1.467 
RAN RAN, member RAS oncogene family 2.745 1.457 
ENC1 Ectodermal-neural cortex (with BTB-like domain). Other designations: nuclear restricted protein, BTB domain-like (brain); tumor protein p53 inducible protein 10 2.660 1.412 
EMP1 Epithelial membrane protein 1 2.645 1.403 
DSG2 Desmoglein 2 2.629 1.395 
SEPP1 Selenoprotein P, plasma, 1 2.628 1.394 
ITGAV Integrin αV (vitronectin receptor, α polypeptide, antigen CD51) antigen identified by monoclonal antibody L230 2.563 1.358 
TUBA2 Tubulin α2 2.457 1.297 
HMMR Hyaluronan-mediated motility receptor (RHAMM) intracellular hyaluronic acid binding protein 2.439 1.286 
MASA L1 neuronal cell adhesion molecule, MASA (mental retardation, aphasia, shuffling gait and adducted thumbs) 2.425 1.278 
TUBA3 Tubulin α3 2.400 1.263 
CX3CR1 Chemokine (C-X3-C motif) receptor 1 2.380 1.251 
CDH2 Cadherin 2, type 1, N-cadherin [neuronal N-cadherin 1; cadherin 2, N-cadherin (neuronal); cadherin 2, type 1; calcium-dependent adhesion protein, neural cadherin] 2.375 1.248 
EDNRB Endothelin receptor type B 2.315 1.211 
CSPG2 Versican (chondroitin sulfate proteoglycan 2) 2.261 1.177 
CD63 Melanoma 1 antigen 2.205 1.141 
SMARCA1 SWI/SNF-related matrix-associated, actin-dependent regulator of chromatin, subfamily a, member 1 2.199 1.137 
FGG Fibrinogen γ chain 2.149 1.104 
PFN1 Profilin 1 2.142 1.099 
SPA17 Sperm autoantigenic protein 17 2.129 1.090 
FZD1 Frizzled homologue 1 (Drosophila2.121 1.085 
APLP2 Amyloid β (A4) precursor-like protein 2 2.098 1.069 
SERPINA3 Serine (or cysteine) proteinase inhibitor, clade A (α-1 antiproteinase, antitrypsin), member 3 2.076 1.054 
MMP2 Matrix metallopeptidase 2 (gelatinase A, 72 kDa gelatinase, 72 kDa type IV collagenase) 2.046 1.033 
TUBA1 Tubulin α 1 (testis-specific) 2.018 1.013 
EDN2 Endothelin 2 −2.279 −1.189 
PCDH9 Protocadherin 9 −2.345 −1.230 
CDH10 Cadherin 10 type 2 (T2-cadherin) −2.397 −1.262 
WNT2 Wingless-type MMTV integration site family member 2 −3.274 −1.711 
PAPPA Pregnancy-associated plasma protein A, pappalysin 1 −3.323 −1.733 
Gene nameGene descriptionn-fold changeLog2 change
LUM Lumican 12.278 3.618 
SDCBP Syndecan binding protein (syntenin, melanoma differentiation-associated protein 9-MDA; pro-TGF-α cytoplasmic domain-interacting protein 18; scaffold protein Pbp1) 6.753 2.756 
MFAP1 Microfibrillar-associated protein 1 4.147 2.052 
CTSB Cathepsin B 3.487 1.802 
PRKCA Protein kinase Cα 3.207 1.681 
RAB1A RAB1A member RAS oncogene family 2.986 1.578 
H2-ALPHA α-Tubulin isotype H2-α 2.934 1.553 
RRAGA Ras-related GTP binding A 2.917 1.545 
TUBB3 Tubulin β3 (other designation: tubulin β4) 2.874 1.523 
CLIC4 Chloride intracellular channel 4 2.870 1.521 
SPARC Secreted protein, acidic, cysteine-rich (osteonectin) 2.840 1.506 
ANLN Anillin, actin binding protein 2.764 1.467 
RAN RAN, member RAS oncogene family 2.745 1.457 
ENC1 Ectodermal-neural cortex (with BTB-like domain). Other designations: nuclear restricted protein, BTB domain-like (brain); tumor protein p53 inducible protein 10 2.660 1.412 
EMP1 Epithelial membrane protein 1 2.645 1.403 
DSG2 Desmoglein 2 2.629 1.395 
SEPP1 Selenoprotein P, plasma, 1 2.628 1.394 
ITGAV Integrin αV (vitronectin receptor, α polypeptide, antigen CD51) antigen identified by monoclonal antibody L230 2.563 1.358 
TUBA2 Tubulin α2 2.457 1.297 
HMMR Hyaluronan-mediated motility receptor (RHAMM) intracellular hyaluronic acid binding protein 2.439 1.286 
MASA L1 neuronal cell adhesion molecule, MASA (mental retardation, aphasia, shuffling gait and adducted thumbs) 2.425 1.278 
TUBA3 Tubulin α3 2.400 1.263 
CX3CR1 Chemokine (C-X3-C motif) receptor 1 2.380 1.251 
CDH2 Cadherin 2, type 1, N-cadherin [neuronal N-cadherin 1; cadherin 2, N-cadherin (neuronal); cadherin 2, type 1; calcium-dependent adhesion protein, neural cadherin] 2.375 1.248 
EDNRB Endothelin receptor type B 2.315 1.211 
CSPG2 Versican (chondroitin sulfate proteoglycan 2) 2.261 1.177 
CD63 Melanoma 1 antigen 2.205 1.141 
SMARCA1 SWI/SNF-related matrix-associated, actin-dependent regulator of chromatin, subfamily a, member 1 2.199 1.137 
FGG Fibrinogen γ chain 2.149 1.104 
PFN1 Profilin 1 2.142 1.099 
SPA17 Sperm autoantigenic protein 17 2.129 1.090 
FZD1 Frizzled homologue 1 (Drosophila2.121 1.085 
APLP2 Amyloid β (A4) precursor-like protein 2 2.098 1.069 
SERPINA3 Serine (or cysteine) proteinase inhibitor, clade A (α-1 antiproteinase, antitrypsin), member 3 2.076 1.054 
MMP2 Matrix metallopeptidase 2 (gelatinase A, 72 kDa gelatinase, 72 kDa type IV collagenase) 2.046 1.033 
TUBA1 Tubulin α 1 (testis-specific) 2.018 1.013 
EDN2 Endothelin 2 −2.279 −1.189 
PCDH9 Protocadherin 9 −2.345 −1.230 
CDH10 Cadherin 10 type 2 (T2-cadherin) −2.397 −1.262 
WNT2 Wingless-type MMTV integration site family member 2 −3.274 −1.711 
PAPPA Pregnancy-associated plasma protein A, pappalysin 1 −3.323 −1.733 

A substantial number of the genes in the metastatic signature could be included in a generic category named melanoma- and neural-related genes, which was created consistently with the neural origin of the melanocytes (Table 4).

Table 4.

Melanoma and neural-related genes

Gene nameGene descriptionn-fold changeLog2 change
SDCBP Syndecan binding protein (syntenin, melanoma differentiation-associated protein 9-MDA; pro-TGF-α cytoplasmic domain-interacting protein 18; scaffold protein Pbp1) 2.7555 6.753 
CHN1 Chimerin 1 (GTPase-activating protein, ρ, 2); n-chimerin 1.523 2.874 
PMP22 Peripheral Myelin Protein 22 (growth arrest-specific 3) 1.505 2.838 
S100A10 S100 calcium binding protein A10 [Annexin II ligand, calpactin I, light polypeptide (p11)] 1.491 2.811 
ENC1 Ectodermal-neural cortex (with BTB-like domain). Other designations: nuclear restricted protein, BTB domain-like (brain); tumor protein p53 inducible protein 10 1.4115 2.66 
EMP1 Epithelial membrane protein 1 1.403 2.645 
SEPP1 Selenoprotein P, plasma 1 1.394 2.628 
MASA L1 neuronal cell adhesion molecule, MASA (mental retardation, aphasia, shuffling gait and adducted thumbs) 1.278 2.425 
PTPRZ1 Protein tyrosine phosphatase, receptor-type, Z polypeptide 1 1.255 2.387 
MITF Microphthalmia-associated transcription factor 1.2505 2.379 
MAGEA6 Melanoma antigen family A, 6 1.2075 2.309 
AKT1S1 AKT1 substrate 1 (proline-rich) 1.1705 2.251 
CD63 Melanoma 1 antigen 1.141 2.205 
MAL Mal, T-cell differentiation protein 1.138 2.201 
KIT V-kit Hardy-Zuckerman 4 feline sarcoma viral oncogene homologue, C-Kit, CD117, PBT, SCFR 1.123 2.178 
APLP2 Amyloid β (A4) precursor-like protein 2 1.069 2.098 
PDGFRA Platelet-derived growth factor receptor, α polypeptide 1.055 2.078 
ALS4 Amyotrophic lateral sclerosis 4 1.053 2.075 
PCDH9 Protocadherin 9 −1.2295 −2.345 
CDH10 Cadherin 10 type 2 (T2-cadherin) −1.2615 −2.397 
NTF5 Neurotrophin 5 (neurotrophin 4/5) −1.791 −3.461 
OLIG2 Oligodendrocyte lineage transcription factor 2 −1.9555 −3.879 
Gene nameGene descriptionn-fold changeLog2 change
SDCBP Syndecan binding protein (syntenin, melanoma differentiation-associated protein 9-MDA; pro-TGF-α cytoplasmic domain-interacting protein 18; scaffold protein Pbp1) 2.7555 6.753 
CHN1 Chimerin 1 (GTPase-activating protein, ρ, 2); n-chimerin 1.523 2.874 
PMP22 Peripheral Myelin Protein 22 (growth arrest-specific 3) 1.505 2.838 
S100A10 S100 calcium binding protein A10 [Annexin II ligand, calpactin I, light polypeptide (p11)] 1.491 2.811 
ENC1 Ectodermal-neural cortex (with BTB-like domain). Other designations: nuclear restricted protein, BTB domain-like (brain); tumor protein p53 inducible protein 10 1.4115 2.66 
EMP1 Epithelial membrane protein 1 1.403 2.645 
SEPP1 Selenoprotein P, plasma 1 1.394 2.628 
MASA L1 neuronal cell adhesion molecule, MASA (mental retardation, aphasia, shuffling gait and adducted thumbs) 1.278 2.425 
PTPRZ1 Protein tyrosine phosphatase, receptor-type, Z polypeptide 1 1.255 2.387 
MITF Microphthalmia-associated transcription factor 1.2505 2.379 
MAGEA6 Melanoma antigen family A, 6 1.2075 2.309 
AKT1S1 AKT1 substrate 1 (proline-rich) 1.1705 2.251 
CD63 Melanoma 1 antigen 1.141 2.205 
MAL Mal, T-cell differentiation protein 1.138 2.201 
KIT V-kit Hardy-Zuckerman 4 feline sarcoma viral oncogene homologue, C-Kit, CD117, PBT, SCFR 1.123 2.178 
APLP2 Amyloid β (A4) precursor-like protein 2 1.069 2.098 
PDGFRA Platelet-derived growth factor receptor, α polypeptide 1.055 2.078 
ALS4 Amyotrophic lateral sclerosis 4 1.053 2.075 
PCDH9 Protocadherin 9 −1.2295 −2.345 
CDH10 Cadherin 10 type 2 (T2-cadherin) −1.2615 −2.397 
NTF5 Neurotrophin 5 (neurotrophin 4/5) −1.791 −3.461 
OLIG2 Oligodendrocyte lineage transcription factor 2 −1.9555 −3.879 

This series included two primary cases with metastatic disease samples whose expression profiles somewhat resembled those identified in the nonmetastatic group, in spite of not exhibiting differences in the Clark level or presence of ulceration, when compared with the rest of nonmetastatic cases. These two cases had in common the expression of the transforming growth factor β receptor II (TGFBR2), a signal transduction protein whose mutations have been associated with tumor progression in some solid tumors (14). Moreover, case mel-08 had altered cell cycle and apoptosis regulator genes with a demonstrated role in melanoma progression, such as overexpression of p21 (CDKN1A; ref. 9) or down-regulation of BCL2L10. The mel-09 case showed up-regulation of ENC1 (ectodermal neural cortex 1), a gene associated related with the Wingless (WNT) pathway and whose expression has been linked with colorectal carcinogenesis (15), and meningioma progression (16).

Validation of gene expression results by immunohistochemistry on TMAs. To corroborate the gene expression data and to identify in situ the cells expressing the various markers, immunohistochemical studies of an independent series of primary vertical growth phase CMMs were done using TMAs. The relevance and intensity in the changes of expression in genes linked with EMT in the training series led us to concentrate on validating some significantly up-regulated proteins (>2-fold change and FDR<0.2) in this category (N-cadherin, osteonectin/SPARC, and PKCα). Prompted by a literature search and the nearly significant results reported here, we also studied glypican 3, which has recently been described as a novel tumor marker for melanoma (17), and osteopontin, a protein belonging to this EMT group that has a crucial role in melanoma (18, 19) and seems to be diagnostically useful in the detection of mesothelioma in serum obtained from peripheral blood (20).

Figure 3 shows the results of the univariate analyses of selected proteins and representative immunohistochemistry for N-cadherin, osteonectin/SPARC, and osteopontin, which were the statistically significant markers. These results are closely consistent with the cDNA microarray data.

Figure 3.

A, Kaplan-Meier analysis of DFS among the validation set of 127 vertical growth phase melanoma cases included in the TMA. The genes evaluated were N-cadherin, osteonectin (SPARC), osteopontin, glypican 3, and PKCα. Kaplan-Meier plots show the curves for significant proteins. 95% CI, confidence interval. B, representative immunohistochemical staining of significantly up-regulated EMT candidate genes identified from the microarray data. Brown staining denotes a positive signal. Magnification, ×4 (inset ×20).

Figure 3.

A, Kaplan-Meier analysis of DFS among the validation set of 127 vertical growth phase melanoma cases included in the TMA. The genes evaluated were N-cadherin, osteonectin (SPARC), osteopontin, glypican 3, and PKCα. Kaplan-Meier plots show the curves for significant proteins. 95% CI, confidence interval. B, representative immunohistochemical staining of significantly up-regulated EMT candidate genes identified from the microarray data. Brown staining denotes a positive signal. Magnification, ×4 (inset ×20).

Close modal

Immunostaining for these proteins confirmed that they were expressed by tumoral melanoma cells. The adhesion molecule N-cadherin was found with a membrane and cytoplasmic pattern. Osteonectin (SPARC) was recognized in tumoral melanocytes and in endothelial vessels close to the tumor, as described in other neoplasms (21, 22). Osteopontin is an acidic matrix protein (bone syaloprotein) commonly localized within normal elastic fibers of the skin and secreted by many transformed cells. Immunolocalization of this protein in our melanoma cases was found in tumoral cells and also in the stroma (see Fig. 3B). As shown in the table in Fig. 3A, the expression of these three proteins was significantly associated with an increased incidence of melanoma metastases [P = 0.013 for N-cadherin; P = 0.006 for osteonectin (SPARC); P = 0.05 for osteopontin], thereby confirming its potential prognostic value in CMM.

PKCα, a protein involved in melanoma progression (23), was also expressed in the cytoplasm of tumoral cells, as previously described (24), but the univariate study failed to show any significant association with tumor progression in this series. However, the multivariate analysis including N-cadherin, osteopontin, osteonectin, PKCα, and Breslow index showed that only PKCα was independent of the Breslow index (Table 5).

Table 5.

Kaplan-Meier, log-rank test, and Cox regression univariate and multivariate analyses of DFS

Univariate and Breslow-adjusted analyses
MarkerNo. casesCases with metastatic diseaseHazard ratio (95% CI)PAdjusted by Breslow
Hazard ratio (95% CI)P
N-cadherin 108 60 1.95 (1.15–3.31) 0.013 1.49 (0.86–2.58) 0.155 
Osteonectin 112 66 1.99 (1.21–3.25) 0.006 1.48 (0.89–2.46) 0.126 
Osteopontin 102 61 1.88 (1.00–3.55) 0.05 1.72 (0.91–3.23) 0.095 
Glypican 3 107 64 1.51 (0.80–2.83) 0.199 1.19 (0.62–2.26) 0.603 
PKCα 119 69 1.21 (0.75–1.95) 0.429 2.03 (1.19–3.44) 0.009 
Multivariate analysis
 
      
Marker
 
Hazard ratio
 
SE
 
z
 
P > z
 
95% CI
 

 
PKCα 2.313 0.7188 2.89 0.004 1.3092–4.0870  
Breslow 1.415 0.0655 7.50 0.000 1.2925–1.5497  
Univariate and Breslow-adjusted analyses
MarkerNo. casesCases with metastatic diseaseHazard ratio (95% CI)PAdjusted by Breslow
Hazard ratio (95% CI)P
N-cadherin 108 60 1.95 (1.15–3.31) 0.013 1.49 (0.86–2.58) 0.155 
Osteonectin 112 66 1.99 (1.21–3.25) 0.006 1.48 (0.89–2.46) 0.126 
Osteopontin 102 61 1.88 (1.00–3.55) 0.05 1.72 (0.91–3.23) 0.095 
Glypican 3 107 64 1.51 (0.80–2.83) 0.199 1.19 (0.62–2.26) 0.603 
PKCα 119 69 1.21 (0.75–1.95) 0.429 2.03 (1.19–3.44) 0.009 
Multivariate analysis
 
      
Marker
 
Hazard ratio
 
SE
 
z
 
P > z
 
95% CI
 

 
PKCα 2.313 0.7188 2.89 0.004 1.3092–4.0870  
Breslow 1.415 0.0655 7.50 0.000 1.2925–1.5497  

NOTE: Data were measured from the time of primary vertical growth phase CMM to clinical or histopathologic metastatic disease. The P values are shown for the univariate analysis of the validation TMA set before and after introducing the Breslow index (bivariate Cox model). Significant markers (P < 0.05) are shown in bold.

Abbreviation: 95% CI, 95% confidence interval.

The findings reported here confirm microarray data previously obtained in cell lines (e.g., lumican; ref. 3) or in other experimental assays (e.g., osteopontin; ref. 19), ENDRB (25), PKCα (26), and seem to imply that the metastatic melanoma signature in human samples is recognized by the increased expression of clusters of genes involved in the control of EMT, cell cycle and apoptosis, immune response, metabolism, transcription regulation, protein synthesis and degradation, and signal transduction.

In accordance with the results obtained, we have focused our attention on the validation step in the EMT because a large and highly significant group of up-regulated genes were members of this functional cluster. EMT is the process by which an epithelial cell suffers transitory changes in cell structure and becomes a more motile mesenchymal cell with migratory and invasive properties (27). As a result, cell-cell junctions are altered; cells lose polarity, express mesenchymal markers, and the actin cytoskeleton becomes reorganized. Consequently, tumoral cells lose contact with neighboring cells, become motile, and interact with extracellular matrix, invading surrounding territories and acquiring capacity for metastasis. The relevance of the EMT signature to the origin of metastasis is not restricted to melanoma, and has been identified as a determinant of local invasion and metastasis in other tumor types, such as breast (28), gastrointestinal, or prostate carcinomas (29), among others.

The present results suggest that migration and invasion in melanoma are the result of a specific interaction between tumoral and stromal cells (30, 31) associated with the expression of a set of molecules involved in EMT (31). Thus, the study reveals N-cadherin, lumican, glypican, osteonectin (SPARC), osteopontin, metalloproteinases, and integrins, among others, as being involved in the melanoma metastatic process. Although it is difficult to verify EMT experimentally in vivo due to the reversible nature of the process, expression profiling identifies a molecular signature linked with this phenomenon. Moreover, the immunohistochemical studies of N-cadherin, osteopontin, glypican 3, osteonectin (SPARC), and PKCα validate the data presented and confirm that at least these components of the EMT signature are expressed by the melanoma cells themselves.

One of most remarkable findings of this study is the presence of what has been called the switch of the cadherin class: loss of epithelial cadherins (E-cadherin) with gain of neural cadherins (N-cadherin). Cadherins are calcium-dependent, cell-adhesion molecules that are critical for the development and maintenance of epithelial architecture. As described in carcinoma models, N-cadherin is a crucial molecule in the EMT event, acting as an oncogene in many tumors by promoting tumor invasiveness and progression (32). In our study, increased N-cadherin was matched by loss of cadherin 10 expression, a type II cadherin with near-equivalent binding strength to that of E-cadherin (33) as described previously in cell line models (34, 35) and in desmoplastic melanoma (36). Although melanoma cells are not epithelial in nature, the EMT for this tumor is well recognized and the relevance of the cadherin switch has been previously supported by experimental approaches, demonstrating that melanoma cell lines transfected with N-cadherin are morphologically transformed from an epithelial-like shape to a fibroblast-like shape; by contrast, adenoviral reexpression of E-cadherin in melanoma cells down-regulates endogenous N-cadherin and reduces their malignant potential (37). The present study not only confirms the role of N-cadherin in melanoma progression but also shows for the first time how gain of N-cadherin and loss of class II type cadherin can facilitate metastatic dissemination in human primary CMM samples.

Osteonectin (SPARC or BM40) is a secreted extracellular matrix glycoprotein involved in tumor cell migration, invasion, and angiogenesis (38). Data presented here are consistent with previous reports of SPARC overexpression as a marker of aggressiveness or local invasion in CMM (39, 40).

Osteopontin is a secreted phosphoprotein with a crucial role in tumor progression and metastasis in many kinds of tumors, including melanoma (19). It promotes antiapoptotic signaling and angiogenesis and induces matrix metalloproteinases (MMP; ref. 18). Osteopontin expression may be acquired in the early stages of melanoma invasion, with moderate to high levels in primary invasive and metastatic melanoma samples and null or low levels of expression in benign nevi, dysplastic nevi, or in situ melanoma (19). Here, cytoplasmic osteopontin expression in invasive melanomas was associated with metastatic disease, mimicking previous experimental models in which osteopontin allowed murine melanocytes to adhere, spread, and survive in three-dimensional collagen gels (41).

The serine/threonine kinase PKCα is a protein with a well-recognized role in regulating cell growth and progression (42). Melanocytes and melanoma cells express several PKC isoforms, with proposed roles in melanoma genesis, invasion, and metastasis through the WNT signaling pathway or by regulating integrin molecule expression of integrin. The specific role of PKCα in CMM has been a controversial issue, with some results suggesting that PKCα and PKCδ isoforms suppress cell growth, whereas others imply its involvement in cell motility (42, 43) or in metastasis promotion (44). Expression profiling data in this series identifies PKCα as one of the genes up-regulated in melanoma with metastasis, and the multivariate analysis showed that it is Breslow independent, implying that there is a relationship between tumor growth and proliferation. This awaits further investigation.

Lumican, the most highly up-regulated gene in melanoma cases with metastasis, is a proteoglycan of the extracellular matrix involved in collagen-fiber organization, epithelial-cell migration, and tissue repair. Lumican is an established EMT marker (45) whose protein expression level is associated with tumorigenesis and progression in a variety of tumors (46, 47). Data presented here are also consistent with those obtained from the comparison of normal melanocytes with melanoma cell strains from advanced lesions (3).

Cell-matrix interactions are closely related to EMT category. The integrins are a family of cell surface adhesion molecules that coordinate cell-cell and cell-matrix interactions. This study identified up-regulation of integrin αV (ITGAV) as a marker of melanoma metastases, confirming a previous report (48). Closely related are the MMP family of proteins, which are involved in the breakdown of the extracellular matrix in normal physiologic processes and in cancer invasion and metastasis. Here, MMP2 (which degrades type IV collagen) and ADAM9 (disintegrin and metalloproteinase domain 9) are associated with melanoma progression, confirming previous findings regarding invasion (49).

Two of the metastatic cases analyzed here (mel-08 and mel-09) displayed a nonmetastatic phenotype, with the exception of the increased expression of TGFBR2, p21 (CDKN1A), or ENC1. It is known that melanoma and endothelial cells express type 2 TGF-β receptor (50), which can enhance adhesion of melanoma cells to endothelium and favor invasion through activation of the TGF-β pathway. In the same way, p21 overexpression has been proposed as a marker of melanoma progression (9). The specific relevance of each functional cluster of genes, as identified here, requires further study, both in human samples and experimental models.

Interestingly, the multivariate study showed that three of the protein markers analyzed in TMA were associated with local invasion and metastasis, thereby linking the two phenomena. Overall, our results lead us to propose that metastasis in melanoma is determined by the interaction of sets of molecules involved in the regulation of EMT immune response, cell metabolism, nucleic acid binding and transcription, protein synthesis and degradation, and signal transduction. Moreover, validation of protein expression in TMA confirms how some of the up-regulated genes are predictors of metastasis development. Taking all our results into consideration, it seems that the control of the EMT plays a pivotal role in the metastatic process in melanoma. This confirms and builds on findings linking EMT genes with melanoma progression obtained in experimental models and human samples studies, such as cadherins, MMPs, integrins, lumican, osteopontin, EDNRB, Snail, and others (6, 49).

Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).

Grant support: FIS grant PI 040641, MMA 2005-085, and European Commission (Translational and Functional Onco-Genomics) grant LSHC-CT-2004-503438. S.R. Alonso was supported in part by FIS grant CM03/00034 and by the CNIO (Madrid, Spain). L. Tracey received support though grants from the CNIO and the Higher Education Authority of Ireland through the Department of Haematology/Institute of Molecular Medicine, St. James Hospital, Dublin, Ireland.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank the Tumour Bank of the Hospital San Cecilio; the Spanish National Tumour Bank Network, CNIO; Laura Cereceda (CNIO) and Alicia Maroto (CNIO, Hospital 12 de Octubre) for their valuable help in providing specific tissue samples for this study; Pilar Sandoval (CNIO, Hospital Gregorio Marañón) for her time with gridding; Dr. David Hardisson and Raquel Marcos (CNIO, Hospital Universitario La Paz) for helping with the SPARC immunostaining; Ramón Diaz for helping with the Wilcoxon test; Amancio Carnero for their valuable guidance with gene classification; and Phil Mason for his help with the English edition of the final version of the manuscript.

1
Zhang L, Zhou W, Velculescu VE, et al. Gene expression profiles in normal and cancer cells.
Science
1997
;
276
:
1268
–72.
2
Bittner M, Meltzer P, Chen Y, et al. Molecular classification of cutaneous malignant melanoma by gene expression profiling.
Nature
2000
;
406
:
536
–40.
3
Hoek K, Rimm DL, Williams KR, et al. Expression profiling reveals novel pathways in the transformation of melanocytes to melanomas.
Cancer Res
2004
;
64
:
5270
–82.
4
McGill GG, Horstmann M, Widlund HR, et al. Bcl2 regulation by the melanocyte master regulator Mitf modulates lineage survival and melanoma cell viability.
Cell
2002
;
109
:
707
–18.
5
Haqq C, Nosrati M, Sudilovsky D, et al. The gene expression signatures of melanoma progression.
Proc Natl Acad Sci U S A
2005
;
102
:
6092
–7.
6
Winnepenninckx V, Lazar V, Michiels S, et al. Gene expression profiling of primary cutaneous melanoma and clinical outcome.
J Natl Cancer Inst
2006
;
98
:
472
–82.
7
Talantov D, Mazumder A, Yu JX, et al. Novel genes associated with malignant melanoma but not benign melanocytic lesions.
Clin Cancer Res
2005
;
11
:
7234
–42.
8
Retsas S, Henry K, Mohammed MQ, et al. Prognostic factors of cutaneous melanoma and a new staging system proposed by the American Joint Committee on Cancer (AJCC): validation in a cohort of 1284 patients.
Eur J Cancer
2002
;
38
:
511
–6.
9
Alonso SR, Ortiz P, Pollan M, et al. Progression in cutaneous malignant melanoma is associated with distinct expression profiles: a tissue microarray-based study.
Am J Pathol
2004
;
164
:
193
–203.
10
Rimm DL, Camp RL, Charette LA, et al. Tissue microarray: a new technology for amplification of tissue resources.
Cancer J
2001
;
7
:
24
–31.
11
Eberwine J. Amplification of mRNA populations using aRNA generated from immobilized oligo(dT)-T7 primed cDNA.
Biotechniques
1996
;
20
:
584
–91.
12
Tracey L, Villuendas R, Ortiz P, et al. Identification of genes involved in resistance to interferon-α in cutaneous T-cell lymphoma.
Am J Pathol
2002
;
161
:
1825
–37.
13
Benjamini Y, Yekutieli D. Quantitative trait Loci analysis using the false discovery rate.
Genetics
2005
;
171
:
783
–90.
14
Fernandez-Peralta AM, Nejda N, Oliart S, et al. Significance of mutations in TGFBR2 and BAX in neoplastic progression and patient outcome in sporadic colorectal tumors with high-frequency microsatellite instability.
Cancer Genet Cytogenet
2005
;
157
:
18
–24.
15
Fujita M, Furukawa Y, Tsunoda T, et al. Up-regulation of the ectodermal-neural cortex 1 (ENC1) gene, a downstream target of the β-catenin/T-cell factor complex, in colorectal carcinomas.
Cancer Res
2001
;
61
:
7722
–6.
16
Wrobel G, Roerig P, Kokocinski F, et al. Microarray-based gene expression profiling of benign, atypical and anaplastic meningiomas identifies novel genes associated with meningioma progression.
Int J Cancer
2005
;
114
:
249
–56.
17
Nakatsura T, Kageshita T, Ito S, et al. Identification of glypican-3 as a novel tumor marker for melanoma.
Clin Cancer Res
2004
;
10
:
6612
–21.
18
Samanna V, Wei H, Ego-Osuala D, et al. α-V-dependent outside-in signaling is required for the regulation of CD44 surface expression, MMP-2 secretion, and cell migration by osteopontin in human melanoma cells.
Exp Cell Res
2006
;
312
:
2214
–30.
19
Zhou Y, Dai DL, Martinka M, et al. Osteopontin expression correlates with melanoma invasion.
J Invest Dermatol
2005
;
124
:
1044
–52.
20
Pass HI, Lott D, Lonardo F, et al. Asbestos exposure, pleural mesothelioma, and serum osteopontin levels.
N Engl J Med
2005
;
353
:
1564
–73.
21
Chen X, Higgins J, Cheung ST, et al. Novel endothelial cell markers in hepatocellular carcinoma.
Mod Pathol
2004
;
17
:
1198
–210.
22
Kato Y, Lewalle JM, Baba Y, et al. Induction of SPARC by VEGF in human vascular endothelial cells.
Biochem Biophys Res Commun
2001
;
287
:
422
–6.
23
Gruber JR, Desai S, Blusztajn JK, et al. Retinoic acid specifically increases nuclear PKC α and stimulates AP-1 transcriptional activity in B16 mouse melanoma cells.
Exp Cell Res
1995
;
221
:
377
–84.
24
Timar J, Liu B, Bazaz R, et al. Association of protein kinase-C-α with cytoplasmic vesicles in melanoma cells.
J Histochem Cytochem
1996
;
44
:
177
–82.
25
Lahav R, Endothelin receptor B is required for the expansion of melanocyte precursors and malignant melanoma.
Int J Dev Biol
2005
;
49
:
173
–80.
26
Lahn MM, Sundell KL. The role of protein kinase C-α (PKC-α) in melanoma.
Melanoma Res
2004
;
14
:
85
–9.
27
Grunert S, Jechlinger M, Beug H. Diverse cellular and molecular mechanisms contribute to epithelial plasticity and metastasis.
Nat Rev Mol Cell Biol
2003
;
4
:
657
–65.
28
Blanco MJ, Moreno-Bueno G, Sarrio D, et al. Correlation of Snail expression with histological grade and lymph node status in breast carcinomas.
Oncogene
2002
;
21
:
3241
–6.
29
Tomita K, van Bokhoven A, van Leenders GJ, et al. Cadherin switching in human prostate cancer progression.
Cancer Res
2000
;
60
:
3650
–4.
30
Hendrix MJ, Seftor EA, Kirschmann DA, et al. Remodeling of the microenvironment by aggressive melanoma tumor cells.
Ann N Y Acad Sci
2003
;
995
:
151
–61.
31
Johnson JP. Cell adhesion molecules in the development and progression of malignant melanoma.
Cancer Metastasis Rev
1999
;
18
:
345
–57.
32
Hazan RB, Phillips GR, Qiao RF, et al. Exogenous expression of N-cadherin in breast cancer cells induces cell migration, invasion, and metastasis.
J Cell Biol
2000
;
148
:
779
–90.
33
Shimoyama Y, Tsujimoto G, Kitajima M, et al. Identification of three human type-II classic cadherins and frequent heterophilic interactions between different subclasses of type-II classic cadherins.
Biochem J
2000
;
349
:
159
–67.
34
Hsu MY, Meier FE, Nesbit M, et al. E-cadherin expression in melanoma cells restores keratinocyte-mediated growth control and down-regulates expression of invasion-related adhesion receptors.
Am J Pathol
2000
;
156
:
1515
–25.
35
Hsu MY, Wheelock MJ, Johnson KR, et al. Shifts in cadherin profiles between human normal melanocytes and melanomas.
J Investig Dermatol Symp Proc
1996
;
1
:
188
–94.
36
Gallego Attis M, Burchette JL, Angelica Selim M, et al. Differential expression of N-cadherin distinguishes aggressive spindle cell melanomas from desmoplastic melanomas.
Am J Dermatopathol
2006
;
28
:
232
.
37
Krengel S, Groteluschen F, Bartsch S, et al. Cadherin expression pattern in melanocytic tumors more likely depends on the melanocyte environment than on tumor cell progression.
J Cutan Pathol
2004
;
31
:
1
–7.
38
Jendraschak E, Sage EH. Regulation of angiogenesis by SPARC and angiostatin: implications for tumor cell biology.
Semin Cancer Biol
1996
;
7
:
139
–46.
39
Rumpler G, Becker B, Hafner C, et al. Identification of differentially expressed genes in models of melanoma progression by cDNA array analysis: SPARC, MIF and a novel cathepsin protease characterize aggressive phenotypes.
Exp Dermatol
2003
;
12
:
761
–71.
40
Ledda F, Bravo AI, Adris S, et al. The expression of the secreted protein acidic and rich in cysteine (SPARC) is associated with the neoplastic progression of human melanoma.
J Invest Dermatol
1997
;
108
:
210
–4.
41
Geissinger E, Weisser C, Fischer P, et al. Autocrine stimulation by osteopontin contributes to antiapoptotic signalling of melanocytes in dermal collagen.
Cancer Res
2002
;
62
:
4820
–8.
42
Oka M, Kikkawa U. Protein kinase C in melanoma.
Cancer Metastasis Rev
2005
;
24
:
287
–300.
43
Szalay J, Bruno P, Bhati R, et al. Associations of PKC isoforms with the cytoskeleton of B16F10 melanoma cells.
J Histochem Cytochem
2001
;
49
:
49
–66.
44
Dennis JU, Dean NM, Bennett CF, et al. Human melanoma metastasis is inhibited following ex vivo treatment with an antisense oligonucleotide to protein kinase C-α.
Cancer Lett
1998
;
128
:
65
–70.
45
Saika S, Kono-Saika S, Ohnishi Y, et al. Smad3 signaling is required for epithelial-mesenchymal transition of lens epithelium after injury.
Am J Pathol
2004
;
164
:
651
–63.
46
Ping Lu Y, Ishiwata T, Asano G. Lumican expression in α cells of islets in pancreas and pancreatic cancer cells.
J Pathol
2002
;
196
:
324
–30.
47
Naito Z, Ishiwata T, Kurban G, et al. Expression and accumulation of lumican protein in uterine cervical cancer cells at the periphery of cancer nests.
Int J Oncol
2002
;
20
:
943
–8.
48
Kageshita T, Hamby CV, Hirai S, et al. α(v)β3 expression on blood vessels and melanoma cells in primary lesions: differential association with tumor progression and clinical prognosis.
Cancer Immunol Immunother
2000
;
49
:
314
–8.
49
Redondo P, Lloret P, Idoate M, et al. Expression and serum levels of MMP-2 and MMP-9 during human melanoma progression.
Clin Exp Dermatol
2005
;
30
:
541
–5.
50
Teti A, De Giorgi A, Spinella MT, et al. Transforming growth factor-β enhances adhesion of melanoma cells to the endothelium in vitro.
Int J Cancer
1997
;
72
:
1013
–20.

Supplementary data