We first identified genes differentially expressed between 69 melanoma samples and 87 samples obtained from available primary or metastatic solid tumors (Table I). RCC samples were excluded from the statistical comparison because this tumors share immune responsiveness with metastatic melanoma and, therefore, were considered separately from non-immunogenic tumors. Differential expression was defined significant at a p₂-value ≤ 0.001 (unpaired two-tailed Student t test). This test identified 4,658 cDNA clones differentially expressed between melanoma metastases and tumors of other histology (see Additional file 1). Non parametric Wilcoxon test yielded comparable results in terms of number and identity of differentially expressed genes (data not shown). Permutation analysis strongly supported the significance of these findings. Approximately half of the differentially expressed clones (2,044) were up-regulated in melanoma metastases relative to other tumors and the remaining 2,614 clones were down-regulated. Up-regulation was defined as a positive value after subtracting the average ratio of other tumors from that of melanoma lesions (Figure 1). Down-regulation was considered a negative value resulting from the same formula. A large proportion of the genes down-regulated in melanoma relative to other tumor were lineage specific and reflected its unique ontogeny from the neuroectoderm while the tumors studied were mostly of epithelial origin. We have previously described the weight that ontogeny may play in balancing the transcriptional profile of RCC 11. Unfortunately, for this type of analysis to be conclusive availability of matched normal tissues is required which is not as readily achievable in the case of melanoma due to the scattered distribution of normal epithelial melanocytes within the skin layers. The complete list of the 4,658 genes differentially expressed by melanomas is available at http://www.societymelanomaresearch.org

Several signatures representing genes preferentially expressed by melanomas were identified that could be partially linked to specific gene functions. Those signatures were segregated according to unsupervised gene rearrangement based on the Eisen's clustering method. The first cluster (cluster a, Figure 1) included 76 clones of which 63 were named corresponding to 53 genes. A subset of genes in this cluster were commonly up-regulated in melanoma and RCC including enolase 2 (neuronal γ-enolase) which is a previously described serum marker of RCC also associated with renal carcinogenesis 111213. Differential expression of enolase-2 between melanoma and other cancers with the exception of RCC reached a significance of 5 × 10^-7 and 1 × 10^-6 for two clones representing this gene (Student's t-test p₂ value). Overall, this cluster was enriched of genes associated with active cellular metabolism and included only few genes of previous known relevance to melanoma with the exception of a member of the melanoma antigen family D (MAGED2, Table 2 and Figure 2). Cluster b included 91 clones (72 named representing 65 distinct genes) predominantly associated with growth regulation and apoptosis. Among the genes included in this cluster was BNIP3L (BCL2/Adenovirus E1B interacting protein like-3, t-test p₂-value = 2 × 10^-7 for both cDNA clones representing this gene) that we have previously reported to be associated with the immune responsiveness of melanoma metastases 10. Two large and related clusters (cluster c and d) included 262 and 613 clones, respectively (112 and 296 named corresponding respectively to 110 and 289 genes). These clusters were characterized by a high density of unnamed clones and by relatively low Cy5/Cy3 ratios. However, it should be noted that these clusters may be of particular interest because the gene expression profile was similar between melanoma and RCC tumors suggesting that some of these genes may conceal the enigma of immune responsiveness. In particular, a relatively sizable sub-cluster was noted with genes predominantly up-regulated in RCC but also expressed by melanoma lesions compared with other tumors (Blue arrow and double vertical bar, Figure 1). Genes concomitantly up-regulated in melanoma and RCC will be separately discussed later, however, it is important to note that this cluster included JAK-1 (t-test p₂-value = 2 × 10^-5) that was previously also reported in association with melanoma immune responsiveness to interleukin-2-based immunotherapy 10. JAK-1 was recently linked to the apoptotic role that interleukin-24 (melanoma differentiation associated gene-7: MDA 7) may exert on melanoma cells 14. The following cluster (cluster e) included 208 clones (151 named representing 143 different genes) predominantly associated with immune function. This immune signature was underlined by the high prevalence of expression of these genes in samples obtained from lymph node metastases whether from melanoma or colorectal primaries. Cluster f integrated 129 clones (91 named representing 87 distinct genes) including a mixture of genes with disparate functions difficult to categorize into a predominant pattern. This group also included APPBP1 (amyloid β precursor protein; t-test p₂-value = 1 × 10^-4) which was previously reported in association with melanoma immune responsiveness 10. APPBP1 is a recently discovered epidermal growth factor that regulates dendrite motility and melanin release in epidermal melanocytes and melanoma cells 15. It is possible that some of its functions may have an indirect effect in modulating the immunological profile of subcutaneous metastases. Cluster g included genes preferentially up-regulated in subcutaneous melanoma metastases known to be more responsive to immunotherapy with interleukin-2 16. This cluster included 201 clones (142 named representing 132 genes). Among the genes representative of this cluster were two classic melanoma associated genes (PRAME and tyrosine-related protein-1; TRP-1). In a small proportion, this cluster included a group of genes only over-expressed in fine needle aspirates (FNA) and generally expressed by circulating cells revealing blood contamination of FNA material (red arrow, Figure 1). Cluster h included 131 clones (102 named representing 99 genes). Cluster I included 222 clones (171 named identifying 155 genes) with most of the melanoma differentiation antigens (MDA) clustering in close proximity with the exception of the TRP-1 already discussed in cluster g. Interestingly, this cluster was also highly enriched of genes associated with ribosomal function and active translation. Furthermore, it included the melanocyte master regulator MITF (t-test p₂-value for two respective cDNA clones = 2 × 10^-15 and 8 × 10^-14) which has been shown to modulate lineage survival and melanoma cell viability through interaction with the anti-apoptotic protein BCL2 17. MITF was coordinately expressed with several genes associated with calcium and other solute metabolism including cytochrome p450 (t-test p₂-value for two respective cDNA clones = 3 × 10^-12 and 7 × 10^-13) solute carrier family 7 (t-test p₂-value for two respective cDNA clones = 2 × 10^-10 and 2 × 10^-6), G protein coupled receptor 56 (t-test p₂-value = 2 × 10^-7) and calpain 3 (t-test p₂-value = 6 × 10^-15), a calcium-regulated gene found to be highly expressed in melanoma cells 18. Finally, cluster J included 88 clones of which the 62 named identified 58 genes.

Genes previously described to be preferentially expressed by melanoma lesions were confirmed to be so at a very high level of significance (Figure 2). Exceptions included AIM-1, CXCL-1 (GRO-α), D2S448 and MAGEA1 which are all significantly more expressed by tumors other than melanomas. Interestingly, different types of melanoma associated genes displayed a different pattern of expression with MDA (tyrosinase, gp100/PMel17 and MART-1/MelanA) being co-coordinately expressed in close proximity to each other in cluster I and MAGE family genes preferentially expressed in cluster d (Table II). Cluster g included a number of genes whose expression had been previously associated with melanoma including preferentially expressed antigen in melanoma (PRAME) and the tyrosine-related protein-1 (TRP-1). When the melanoma associated genes were studied alone, PRAME clustered close to the other MDA believe to be involved in the pigmentation process (tyrosinase, MART-1/Melan and gp100/PMel17). This is of particular interest because PRAME has been also reported to be highly expressed in other cancers of ectodermic origin such as medulloblastoma and neuroblastoma suggesting a link between ectoderm and pigmentation 1920. The coordinated expression of MDA suggests that their down-regulation or loss of expression during melanoma progression may be related to a central regulatory pathway not as yet identified. Indeed in previous studies 2122232425, we noted that loss of expression of MART-1/Melan A paralleled that of gp100/PMel17 (SILV) in melanoma metastases while genes of the MAGE family manifested an independent behavior 25. This finding may have important repercussions in the design of antigen-specific immunization protocols and at the same time may complicate the interpretation of tumor antigen loss variant analysis by broadening loss of expression to antigens other than those targeted by a given therapy.

The large majority of genes associated with immune function were included in cluster e. These genes appeared up-regulated in lymph node metastases of melanoma as well as those from colorectal primaries suggesting that their expression results from lymphoid cell infiltration (Figure 3). The same genes were up-regulated in a significant proportion of subcutaneous melanoma metastases suggesting that a strong and active infiltrate of immune cells is present in these tissues. In fact, most of the genes included in cluster e were significantly up-regulated in 10 cutaneous/subcutaneous melanoma lesions compared to 70 primary cancers of other histology (Table III shows a selection of the most significantly up-regulated genes in cutaneous/subcutaneous melanomas). Of interest is the observation that several of the genes up-regulated in these lesions are clustered in specific chromosomal locations with a high predominance of genes located in position 6p21.3, 11p11.2, 19p13 and 19q13. Among the immunologically-related genes specifically up-regulated in subcutaneous melanoma metastases, some are of particular interest because of their known relationship with effector T cell function. In particular, we find interesting that NK4, an anti-angiogenic factor released by natural killer cells 26, was constitutively expressed by cutaneous melanomas. We found this gene to be associated with regression of a melanoma metastasis during interleukin-2 therapy 27 and to be one of the genes most frequently up-regulated during activation of antigen-specific T cells in vitro 2829. Of interest was also the constitutive expression of CD27 a co-stimulatory member of the TNF receptor family strongly associated with cell activation 3031. The expression of CX3CR1 a gene constitutively expressed by natural killer cells that makes them sensitive to chemo-attraction by CXCL12 and CXC3L1 32 may be an explanation for a preferential localization of these effector cells in melanoma lesions. In particular, this finding suggests that the microenvironment of melanoma metastases is rich of fractalkine (CX3CL1) which is a potent chemo-attractant released by endothelial cells stimulated by interferons 33. Overall, the presence of these and other (KLRG1; killer cell lectin like receptor subfamily G, member 1 and KLRK1; killer cell lectin like receptor subfamily K, member 1 and the interleukin-21 receptor) natural killer cell-related genes suggests potent chemo-attraction toward natural killer cells by the tumor micro-environment of subcutaneous and cutaneous melanoma metastases. This is also emphasized by the high expression of interleukin-21 receptor which is usually expressed by natural killer cells and stimulates their cytolytic activity upon ligation with interleukin-21 produced by activated T cells 3435. The constitutive expression of interferon regulatory factor (IRF)-7 implicated in the amplification of the innate immune response 36 through interactions with the NF-κB pathway 37 may lead to the activation of various types of type I interferons 38. More puzzling is the constitutive expression of interleukin-16, a pleiotropic cytokine with predominant chemo-attractant activity for CD4+ T cells 39 and CD4+ eosinophils 40; a relationship between this cytokine and melanoma metastases has never been observed before. In summary, the immunological signature portrayed by subcutaneous melanoma metastases is that of an active innate immune response centered on natural killer cells. More broadly, the preferential expression of genes with immune function in melanoma lesions compared with other tumors suggests that this cancer is constitutively immunologically active and this status may predispose metastatic melanoma to respond to general or antigen-specific immune manipulation.

A second and smaller group of immunologically-related genes was identified and included genes that segregated separately in clusters f to j. These genes had an expression profile opposite to the immune-related genes seen included in cluster e and appeared over-expressed in subcutaneous compared to lymph node metastases. Two granzyme-related genes were found strongly up-regulated in cluster e including granzyme A and M. This observation contrasted with the increased expression of cathepsin F and L in cluster f to j suggesting an opposite regulation of these genes involved in cell death or survival.

It has been reported that subcutaneous metastases of melanoma are more responsive to immunotherapy with interleukin-2 than lymph nodal and visceral metastases 16. Therefore, we identified genes differentially expressed in the former compared with the latter. Since no material from visceral metastases was available, we limited the comparison to subcutaneous versus lymph nodal metastases. Overall, clusters g - j appeared to demonstrate a preferential expression of genes in subcutaneous metastases independent of the technique used for biopsy (excision versus FNA). In particular, cluster g contained a small node of 47 clones highly expressed in subcutaneous metastases that included PRAME and TRP-1. This cluster also included the renal tumor antigen RAGE which has been previously shown to be highly expressed by melanomas 41 and melanophilin and the s100 protein often associated with clinical parameters in melanoma 42. Interestingly, closely linked to PRAME was the pattern of expression of the serine/threonine-specific protein kinase B-RAF. This gene is mutated in approximately 70 % of melanomas and it is often over-expressed 43. Although several of these genes had been associated with melanoma their co-ordinate expression has never been previously appreciated. Overall, the identity of the genes over-expressed in subcutaneous metastases did not offer an obvious explanation for the increased immune responsiveness of these lesions and more extensive understanding of their relationship will be necessary in the future.

We then pooled together melanoma and RCC samples data to identify genes commonly expressed by these tumors and not by tumors of other histology. Significance was assessed by a two-tailed unpaired student's t test and identified 4,221 genes at a cut off p₂-value of ≤ 0.001. The data set was then filtered using the Cluster Program (Stanford, CA) selecting genes that were expressed in at least 80 % of the experiments and for which a Cy5/Cy3 log₂ ratio ≥ 2 was present in at least one experiment. Two-thousand eight hundred and forty-three genes resulted from this filter. Because of the predominance of melanoma lesions (69 melanoma lesions compared to 14 RCC) the genes identified strongly represent differences between melanomas and other tumors. Therefore, we identified among them those that were not differentially expressed between melanoma and RCC samples to identify those genes that are truly uniquely expressed by the two immune responsive cancers. Two-thousand three hundred and fifty-eight genes were expressed similarly between the two types of cancer (at a t test p₂-value > 0.05). A significant number of genes commonly expressed by melanoma and RCC and not by other tumors had no known function (681 genes). The remaining 1, 677 genes were further analyzed by separating those up-regulated from those down-regulated in melanoma and RCC compared with other tumors. The genes up-regulated in RCC and melanoma were considered those with a median Log_Ratio above 0.3 in either RCC or melanomas (a selection of these genes is shown in Table IV). This analysis selected 199 genes. A proportion of genes appeared to be specifically related to lymph nodal and immune infiltration as they were particularly up-regulated in melanoma metastases to lymph nodes and in lymph node metastases obtained from patients with CRC (orange vertical bar, Figure 4). These genes included annotations related to immunological function. A set of genes was specifically expressed in cutaneous metastases of melanoma and in RCC and not other tumor samples (dark blue vertical bar, Figure 4). These genes included microphtalmia transcription factor MITF 1744. that has been shown previously to exert a central role in the regulation of transcriptional activity of melanoma cells. Similarly, enolase-2 (previously known to be up-regulated in RCC) 11 was found to be over-expressed in common between the two histologies. This is somewhat surprising since immunohistochemical analysis has used lack of staining for enolase-2 as a reliable method to differentiate malignant melanoma (enolase-2 negative) from Merkel cell carcinoma 45. It is possible, that although identifiable at the transcriptional level, enolase-2 is not processed into a protein in melanomas. On the other hand, enolase-2 has been shown to be expressed in approximately 90 % of canine oral melanomas 46. In addition, the macrophage migration inhibiting factor (MIF) which is a modulator of cell cycle progression and angiogenesis in melanoma 4748. was found co-expressed by melanoma and RCC lesions. MIF has modulatory properties on natural killer cell mediated lysis of cancer cells contributing, therefore, to an immune privileged microenvironment in uveal melanoma 49. Two genes coding for adhesion molecules; L1 cell adhesion molecule (LCAM) and melanoma cell adhesion molecule (MCAM) were also up-regulated in both lesions and may play an important role in mediating migration of immune cells to the tumor deposits 50. Finally, it is remarkable that serologically defined colon cancer antigen 8 was specifically expressed by melanoma and RCC while was completely absent in colon cancers underlying the need for a better nomenclature of newly identified genes.

This is a descriptive study where genes specifically expressed by melanoma metastases were identified comparing a large collection of samples from patients with metastatic cutaneous melanoma with other primary tumors and lymph nodal metastases. A limitation of the study is the lack of other samples including visceral metastases of melanoma and metastases from tumors of other histology. Nevertheless, we considered useful to compile a list of genes characteristically expressed by subcutaneous and lymph nodal lesions of melanoma for reference purposes and we are willing to provide full information about these genes upon request. In spite of the limitations of this study, few general conclusions could be drawn.

Fourteen primary renal cell carcinoma (RCC) specimens were collected at the Department of Urology of The Johannes Gutenberg-University, Mainz, Germany; one primary melanoma, three in transit metastases, seven cutaneous metastases, thirty-five lymph nodal metastases and two visceral metastases of cutaneous melanoma were collected at the Department of Surgical Sciences, University of Padua, Italy; twenty-one fine needle aspirates of cutaneous melanoma metastases were obtained at the Surgery Branch, National Cancer Institute, National Institutes of Health, Bethesda, MD; seventeen primary epithelial ovarian cancer (EOC) specimens were obtained at the Department of Gynecologic Oncology, MD Anderson Cancer Center, TX; three primary sarcoma, one primary endometrial cancer, one primary laryngeal cancer, two primary breast cancers and one primary colon adeno-carcinoma were obtained from the Tissue Network (Philadelphia, PA); twelve primary carcinomas of the esophageal junction were obtained from the NCI (Division of Cancer Treatment and Diagnosis); thirty-five primary, 16 lymph node metastases and one hepatic metastasis from colorectal adeno-carcinomas were obtained from the Department of Pathology of the University of Pisa, Italy. Specimens were collected as the result of routine operative procedures and portions were frozen for subsequent analysis while the remnant tissue was used for pathological confirmation. Tissue procurement followed standard ethical procedure according to institutional policy. A summary of the specimens studied is presented in Table 1 with their order reflecting their distribution in figures where supervised analyses are shown.

Total RNA was extracted from frozen material using Trizol reagent according to manufacturer's instructions (Invitrogen, CA) and amplified into anti-sense RNA (aRNA) as previously described 10275152. Although the quantity of starting total RNA was in most cases sufficient for cDNA array hybridization, we have shown repeatedly that the fidelity of aRNA hybridization is at least equal and likely superior to total RNA for transcriptional profiling due to lack of contaminant ribosomal and transfer RNA 5153. Therefore, we used aRNA to increase consistency of results particularly when low quality total RNA was documented by Agilent Bioanalyzer 2000 (Agilent Technologies, Palo Alto, CA). After amplification the quality of aRNA was tested with the Agilent Bioanalyzer as previously described 52.

Total RNA from peripheral blood mononuclear cells pooled from six normal donor was extracted and amplified to serve as constant reference as previously described 10275152. Test and reference RNA were labeled with Cy5 (red) and Cy3 (green) and co-hybridized to a costum-made17.5 K cDNA micro-array http://nciarray.nci.nih.gov/gal_files/index.shtml. Micro-arrays were printed at the Immunogenetics Section, DTM, CC, NIH with a configuration of 32 × 24 × 23 and contained 17,500 elements. Clones used for printing included a combination of the Research Genetics RG_HsKG_031901 8 k clone set and 9,000 clones selected from the RG_Hs_seq_ver_070700 40 k clone set. The 17,500 spots included 12,072 uniquely named genes, 875 duplicated genes and about 4,000 expression sequence tags.

All statistical analyses were performed using the log₂-based ratios normalizing the medial log₂ ratio value across the array equal to zero. Validation and reproducibility were performed using our internal reference concordance system as previously described 54. Unsupervised clustering was performed according to the Eisen's Pearson correlation method 55 and visualized with Tree-View software (Stanford University, CA). Genomic portraits were depicted according to the central method for display using a normalization factor as suggested by Ross et al. 56. Details about different tests are discussed in the respective results section. Identification of tumor-specific genes was performed using un-paired 2-tailed Student's t test. The same analyses were performed using un-paired Wilcoxon's non-parametric assessment and provided the same conclusions (not shown). Details of each analysis are presented in the results section.