• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptNIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Cancer Res. Author manuscript; available in PMC Feb 13, 2008.
Published in final edited form as:
PMCID: PMC2241738

Prospective Molecular Profiling of Melanoma Metastases Suggests Classifiers of Immune Responsiveness


We amplified RNAs from 63 fine needle aspiration (FNA) samples from 37 s.c. melanoma metastases from 25 patients undergoing immunotherapy for hybridization to a 6108-gene human cDNA chip. By prospectively following the history of the lesions, we could correlate transcript patterns with clinical outcome. Cluster analysis revealed a tight relationship among autologous synchronously sampled tumors compared with unrelated lesions (average Pearson's r = 0.83 and 0.7, respectively, P < 0.0003). As reported previously, two subgroups of metastatic melanoma lesions were identified that, however, had no predictive correlation with clinical outcome. Ranking of gene expression data from pretreatment samples identified ~30 genes predictive of clinical response (P < 0.001). Analysis of their annotations denoted that approximately half of them were related to T-cell regulation, suggesting that immune responsiveness might be predetermined by a tumor microenvironment conducive to immune recognition.


Efforts aimed at the discovery of independent predictors of clinical outcome have identified molecular subsets of cancer based on mathematical analyses of their gene expression profiles (1-3). Subcategories of lymphomas (1) and breast carcinomas (3) with distinct prognostic and/or clinical behavior were recognized. Bittner et al. (2) suggested two biologically distinct molecular profiles of cutaneous melanoma lesions with divergent metastatic potential in vitro but unknown clinical relevance. In addition, one subclass was characterized by the enhanced expression of MART-1, a classic melanoma differentiation antigen. This triggered the hypothesis that perhaps two different disease taxonomies, both classified according to visual methods as melanoma, could be identified by total transcript analysis and might be characterized by different immune responsiveness. We, therefore, wondered whether direct ex vivo documentation of molecular portraits of metastatic melanoma lesions could prospectively identify taxonomically different entities of this disease or subsets related to its natural progression that could, in either case, be of clinical relevance (4). To account for experimental variance attributable to the intrinsic heterogeneity of different tumor deposits and/or the evolving genetic profiles of individual lesions with time, we directly correlated clinical information pertaining to each lesion with its biological profile. Thus, we collected serial FNAs2 (23-gauge) of individual lesions that allowed prospective documentation of their natural history and/or therapeutic outcome. Limitations attributable to the small amount of total RNA obtainable by FNA were circumvented by a recently validated amplification method (5). Because melanoma is a disease predisposed to different forms of immune modulation, in this exploratory study, emphasis was put on the intrinsic biology of individual lesions rather than the specificity of the immunotherapy administered. Thus, FNA samples obtained before conceptually similar but not identical forms of immunotherapy were studied with the goal of identifying candidate predictors of immune responsiveness based on the working hypothesis that common effector pathways may ultimately determine immune rejection of cancer.

Materials and Methods

FNAs were obtained before treatment in a period spanning from January 5, 1999 through December 16, 1999 from 37 metastases in 25 patients with metastatic melanoma referred to the Surgery Branch, National Cancer Institute for various immunotherapy treatments. The size of each metastasis was serially documented as two perpendicular diameters. The same operator (G. A. O.) performed the FNA by aspirating four quadrants of each lesion. To minimize RNA metabolism and degradation, the material was immediately placed at the bedside in ice-cold RPMI 1640 culture (Biofluids, Rockville, MD) without serum and carried on ice to the laboratory for processing. Over this period, ~300 FNAs of s.c. melanoma metastases were accrued. From those, thirty-seven s.c. lesions were selected from 25 patients based on quality of material available and relevant clinical outcome (Table 1). A second FNA was obtained, when possible after at least one course of treatment. Paired pre- and posttreatment samples were obtainable in 25 lesions (median follow-up, 11 weeks). Metastases were separated into groups: group 1, cr; group 2, pr (>50% reduction in the product of two perpendicular diameters); group 3, stable disease (<50% reduction and <25% increase); and group 4, no response (>25% increase). All responses lasted >30 days. Thirteen pretreatment FNAs were available from cr lesions. Of those, 4 (P3-a0-cr, P18-a0-cr, P18-b0-cr, and P21-a0-cr) regressed before follow-up FNAs could be obtained. Two pr, 8 sd, and 11 nr pretreatment FNAs were also available.

Table 1
Metastases that regressed completely in response to therapy


Total RNA extracted from FNA and control samples was transcribed in vitro into aRNA and reverse-transcribed into fluorescently labeled cDNA for hybridization to 6108 gene cDNA-based microarrays as described previously (5). Control samples consisted of NHEMs derived from neonatal foreskin and grown in melanocyte culture medium (Clonetics, San Diego, CA). The RCC, the P11-Mel melanoma cell line, and the fibroblast strain (FB) were expanded in standard RPMI 1640 containing 10% human AB serum from FNAs of metastatic renal cell (RCC) and a melanoma metastasis (P11-Mel and FB) cell lines. All three were used for the analysis before they reached the fifth passage in culture. Pooled peripheral blood mononuclear cells from six donors were used to prepare reference aRNA to be cohybridized in all experiments with test aRNA. cDNA targets were labeled with aRNA using Cy3 (green) for reference material and Cy5 (red) for test material. A 16 × 20 × 20 (6400-spot) human cDNA microarray printed at Advanced Technology Center, NCI has 6108 sequence-verified clones representing 5492 unique genes and 537 expressed sequence tag clusters.

Statistical Methods

All statistical analyses were performed with SPLUS package. The log10-based ratios were normalized by making the median value in an array equal to zero. Pearson correlation coefficients of log-ratios of two expression profiles were used to quantify the similarity between the samples. We visualized relationships among expression profiles by performing average linkage hierarchical clustering and multidimensional scaling analyses (6). In these analyses, we used one-correlation coefficient as the distance between pairs of samples. We performed these analyses both including all genes and including only genes showing high variation in log expression ratios across the entire set of samples. The results were similar, and the latter analyses are shown in this report. The variance of each gene across the entire set of samples was computed, and the median was determined. The genes with high variation were defined as those with variance significantly (P < 0.001 by χ2 test) greater than this median. Two-sample t-statistics and Wilcoxon rank statistics were used to identify the genes that are differentially expressed between two groups (such as cr versus other lesions, melanoma versus melanocytes, and others). We used relatively stringent cutoff levels for significance because of the number of genes being tested. To determine whether the number of differentially expressed genes is higher than expected because of chance, we randomly shuffled the phenotype labels (e.g., cr versus non-cr) and recomputed the two-sample t-statistics for each gene. This analysis was repeated 10,000 times, and the proportions of the random replications that resulted in as many significant genes as seen in the actual data were reported as the significance levels for the number of genes.

Paired value t-statistics and Wilcoxon rank statistics were used to identify the genes that have significant changes between pre- and posttreatment samples in each group. For the paired value t test, the labels of paired pre- and posttreatment samples were switched randomly, and such analysis was also repeated 1000 times to generate permutation-based statistical significance for the number of genes significantly changed between pre- and posttreatment samples (P < 0.001 by t test). To determine whether the association between pairs of synchronous lesions of the same patient is stronger than expected because of chance, we randomly shuffled the patient identifier labels for the synchronized pairs. The average correlation of log-ratios for synchronous pairs in the data set was compared with the distribution of average correlations based on 10,000 sets of randomly paired lesions.

Results and Discussion

Overview of Melanoma Metastases

The data set was globally screened by applying a high-stringency filter to minimize labeling or random hybridization bias (Ref. 5; Fig. 1). Clustering algorithms based on the resulting 4293 genes suggested a high degree of relatedness among simultaneously biopsied (synchronous) autologous lesions. Most metastases did not significantly change with time; however, exceptions were noted, as outlined by the color lines in Fig. 1. Two clusters were identified that included 49 and 14 samples, respectively. The larger cluster (cluster I) appeared closely associated with NHEMs. Correlation with clinical information did not segregate the two clusters into biologically distinct categories because metastases from the same patients biopsied at different time points were observed in either group. Instead, the smaller cluster (cluster II) appeared to portrait a late/progression expression profile because its members included an inordinate proportion of later FNAs (16 of 49 pretreatment samples in cluster I versus 11 of 14 in cluster II; Fisher test p2 = 0.003). Serially sampled lesions of some patients demonstrated a shift toward this subset with time, i.e., later samples of FNA pairs shifted in cluster II (P4-b1-pr, P5-b1-sd, P14-a1, cr, P14-b1-cr, and P25-b1-nr). Thus, global transcript analysis failed to identify subsets of melanoma metastases with predictive value with regard to immune responsiveness but rather suggested a progressive drift in time of melanoma metastases away from NHEMs.

Fig. 1
Evolving molecular portraits of metastatic melanoma. Eisen's hierarchical clustering dendrogram of all samples studied applied to 4293 genes allowed by high-stringency filtering (Cy5:Cy3 ratios with a 3-fold change, signal intensity >500 unless ...

Treatment-dependent Adaptation in Expression of Individual Genes

Because the previous analyses indicated that immune responsiveness could not be predicted by a specific subset of melanoma metastases, we turned our attention to individual gene expression. By comparing individual gene expression in paired pre- and posttreatment samples, we observed that the number of genes differentially expressed in response to immunotherapy was greater in cr lesions. In particular, 17 genes were differentially expressed with a high degree of statistical significance (t test, P < 0.001) in 9 pairs of cr lesions (Fig. 2A). Because when this large number of tests is performed a low P can simply be obtained by chance, we then performed a permutation analysis to address the frequency in which such a number of significant (P < 0.001) observations would occur by pure chance (see “Materials and Methods”). This was a significantly greater number than would be expected by chance as confirmed by permutation analysis (P < 0.015). This permutation test P gives a general estimate of the significance of the findings based on the complete data set and does not represent a correction for individual Ps. When a similar analysis was performed on 14 pairs of lesions that did not undergo clinical regression, only 6 genes were identified to such degree of significance, a number close to the one expected by chance (P < 0.2). This observation suggests that clinical regression is associated with significant alterations in the transcriptional profile of tumors, whereas lack of response is associated with an indolent intratumor microenvironment rather than a turbulent reaction to a brisk immune response through the adoption of escape mechanisms (7). Analysis of the functional annotations of the genes identified by this analysis depicted a general pattern associated with increased tissue metabolism because it could be expected during tissue destruction and reactive repair mechanism. However, IRF1 up-regulation implicated strong immune stimulatory conditions (see below).

Fig. 2
Putative predictors of immune responsiveness. A, highest ranking genes differentially expressed in posttreatment FNAs compared with pretreatment FNAs in lesions that regressed with treatment (top panel). The relative expression of gene markers of specific ...

Statistical Ranking of Individual Genes Differentially Expressed in Pretreatment Samples Suggests Classifiers of Immune Responsiveness

To avoid disturbances attributable to the effect of time and/or treatment, only pretreatment samples were analyzed further to test whether subsets with clinically divergent behavior could be identified. Lesions synchronously sampled before treatment from the same patients were significantly closer to each other than unrelated lesions (average R = 0.83 and 0.7, respectively; permutation test, P < 0.0001). However, clustering of pretreatment samples could not identify clear subsets of melanoma. Nonparametric comparison of individual gene expression between 13 pretreatment cr samples and all other pretreatment lesions (n = 21) identified 14 genes differentially expressed to a P < 0.001 level of significance (Wilcoxon test; Table 2; Fig. 2B). Similarly, parametric analysis identified 18 genes with levels of expression significantly different at P < 0.001 (t test), mostly overlapping those identified by the Wilcoxon test. Permutation analysis established that the number of differentially expressed genes was significantly greater than that expected by chance (P < 0.05). To further explore the significance of these findings, we compared differences between 13 cr versus 11 nr pretreatment FNAs characterized by most dramatically divergent clinical behavior. Nonparametric analysis identified 13 differentially expressed genes (Wilcoxon P < 0.001) and parametric analysis identified 9 (t test P < 0.001) several overlapping with the previous data set. In total, the combined analysis identified 33 genes of potential interest. Seven genes had no annotated function. Review of available literature suggested that 12 of the remaining 26 genes were associated with immune/inflammatory function (Table 2, boldface entries). This rate was significantly higher than the frequency of genes with known immune regulatory function present in the cDNA chip (10% of 100 randomly selected clones, Fisher test p2 < 0.0001). All of the statistical analyses described here (with the exception of the random sampling of 100 genes) were done a priori and independently of the characterization of the genes. Among the genes preferentially expressed in responding lesions, TIA-1 is a cytolytic granule component responsible for killing of CTL targets (8). EBI3 has a facilitator role in the secretion of IL-12 (9) and is strongly associated with activation of antigen-presenting cells (10). Other genes are associated with transforming growth factor-β-like function (MADH3, INHBA; Ref. 11), IFN (IRF-2 and IF127), or TNF regulation (FIP2). Similarly, JAK-1 and Txk, relatively suppressed in responding lesions, regulate T-cell signaling and differentiation. In addition, IRF-2 expression coincided with the expression of a casein kinase belonging to a family of kinases possible associated with IRFs transcriptional activity (12). Of particular interest was the enhanced expression of IRF2 in immune-responsive lesions. Both IRF1 (up-regulated in responding lesions) and IRF2 may act as functional agonists and can modulate genital wart immune responsiveness through the JAK/STAT pathway (13) and the responsiveness of carcinoid tumors (14) and chronic myeloid leukemia (15) to IFN-α treatment and, in general, immune-mediated tumor suppression (16). The differential expression of immune-regulatory genes between immune-responsive and resistant lesions was not associated with significant differences in density of different immune cell populations, as suggested by various differentiation markers present in the array (Table 2 and Fig. 2B). In addition, quantitative real-time PCR measurement demonstrated similar expression of lymphoid cell markers, such as T-cell receptor Vβ chain constant domain, CD3 and CD8 in cr versus nr lesions, while confirming the differential expression of immune-regulatory genes (data not shown).

Table 2
Classifiers of immune responsiveness

This explorative study was designed to test the potential of using FNA-derived material to follow, by global transcript analysis, the progression of events occurring in the tumor microenvironment. Our results show that FNA sampling can allow direct documentation of evolving biological processes marking the natural course of a disease or its response to treatment. Subsets of melanoma metastases were identified that could be best explained by temporal changes in the transcriptional pattern of the disease. Because the lesions consisted of a relatively homogeneous collection of s.c. metastases, broader molecular diversity predictive of metastatic potential and tissue localization suggested by others (2) might have been missed.

Contrary to other studies in which global transcript analysis identified clinically relevant subsets of lymphomas and breast carcinomas (1, 3), we did not observe subsets of melanoma predictive of clinical outcome. Individual gene analysis, however, suggested that immune responsiveness may be predetermined and not solely dependent upon the extent of the immune responses elicited by a given treatment. In the same set of data, we have observed that melanoma metastases express a heterogeneous array of cytokines, growth factors, and metalloproteinases (17, 18). Among them, several have chemotactic properties on nucleated blood cells such as BLC, eotaxin, IL-1, IL-8, IL-16, lymphotactin, MCP-1, MCP-3, MCP-4, and RANTES whereas others display potent inflammatory activity such as IL-6, MIP-1α, MIP-1β, MIP-2α (GRO 1/2), and TNF-γ. In addition, the expression profile was found to be surprisingly similar among various cytokines and correlated in most metastases with that of a subset of IFN-responsive elements including IRFs. Thus, it is possible that immune-stimulatory and/or inflammatory stimuli occurring at the tumor site may induce coordinate expression of several immune modulators in some tumors and predispose to immune rejection an otherwise dormant host's immune system tolerant of poorly immunogenic tumor cells. Our observations yield a novel hypothesis suggesting that responsiveness of melanoma metastases to immunotherapy is predetermined. Because the study was based on a patient population receiving conceptually similar yet heterogeneous therapy, these findings have only exploratory significance, and future studies should address their predictive significance in the context of different therapies.


We acknowledge Douglas E. Kesselring for the expert preparation of the visual graphics.


2The abbreviations used are: FNA, fine needle aspiration; cr, complete regression; pr, partial regression; sd, stable disease; nr, no response; aRNA, antisense RNA; NHEM, normal human epithelial melanocyte; RCC, renal cell carcinoma; TNF, tumor necrosis factor; IRF, IFN regulatory factor; IL, interleukin; MCP, monocyte chemotactic protein; MIP, macrophage inflammatory protein.


1. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Bedrick JC, Sabet H, Tran T, Xin Y, Powell JI, Yang L, Marti GE, Moore T, Hudson J, Jr., Lisheng L, Lewis DB, Tibshirani R, Sherlock G, Chan WC, Greiner TC, Weisenburger DD, Armitage JO, Warnke R, Levy R, Wilson W, Grever MR, Byrd JC, Botstein D, Brown PO, Staudt LM. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature (Lond.) 2000;403:467–578. [PubMed]
2. Bittner M, Meltzer P, Chen Y, Jiang E, Seftor E, Hendrix M, Radmacher M, Simon R, Yakhini Z, Ben-Dor A, Dougherty E, Wang E, Marincola FM, Gooden C, Lueders J, Glatfelter A, Pollock P, Gillanders E, Dietrich K, Alberts D, Sondak VK, Hayward N, Trent JM. Molecular classification of cutaneous malignant melanoma by gene expression: shifting from a continuous spectrum to distinct biologic entities. Nature (Lond.) 2000;406:536–540. [PubMed]
3. Perou CM, Sertle T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, al-Katib A, Fluge O, Pergamenschikov A, Williams C, Zhu SX, Lenning PE, Berresen-Dale A-L, Brown PO, Botstein D. Molecular portraits of human breast tumors. Nature (Lond.) 2000;406:747–752. [PubMed]
4. Wang E, Marincola FM. A natural history of melanoma: serial gene expression analysis. Immunol. Today. 2000;21:619–623. [PubMed]
5. Wang E, Miller L, Ohnmacht GA, Liu E, Marincola FM. High fidelity mRNA amplification for gene profiling using cDNA microarrays. Nat. Biotech. 2000;17:457–459. [PubMed]
6. Khan J, Simon R, Bittner M, Chen Y, Leighton SB, Pohida T, Smith PD, Jiang Y, Gooden GC, Trent JM, Meltzer PS. Gene expression profiling of alveolar rhabdomyosarcoma with cDNA microarrays. Cancer Res. 1998;58:5009–5013. [PubMed]
7. Marincola FM, Jaffe EM, Hicklin DJ, Ferrone S. Escape of human solid tumors from T cell recognition: molecular mechanisms and functional significance. Adv. Immunol. 2000;74:181–273. [PubMed]
8. Tian Q, Streuli M, Saito H, Schlossman SF, Anderson P. A polyadenylate binding protein localized to the granules of cytolytic lymphocytes induces DNA fragmentation in target cells. Cell. 1991;67:629–639. [PubMed]
9. Devergne O, Birkenbach M, Kieff E. Epstein-Barr virus-induced gene 3 and the p35 subunit of interleukin-12 form a novel heterodimeric hematopoietin. Proc. Natl. Acad. Sci. USA. 1997;94:12041–12046. [PMC free article] [PubMed]
10. Hashimoto S-I, Suzuki E, Nagai S, Yamashita T, Toyoda N, Matsushima K. Identification of genes specifically expressed in human activated and mature dendritic cells through serial analysis of gene expression. Blood. 2000;96:2206–2214. [PubMed]
11. Zhang Y, Feng X, Derynck R. Receptor-associated Mad homologues synergize as effectors of the TGF-β response. Nature (Lond.) 1996;383:168–172. [PubMed]
12. Lin R, Hiscott J. A role for casein kinase II phosphorylation in the regulation of IRF-1 transcriptional activity. Mol. Cell. Biol. 1999;191:169–180. [PubMed]
13. Arany I, Tyring SK, Brysk MM, Stanley MA, Tomai MA, Miller RL, Smith MH, McDermott DJ, Slade HB. Correlation between pretreatment levels of interferon response genes and clinical responses to an immune response modifier (Imiquimod) in genital warts. Antimicrob. Agents Chemother. 2000;44:1869–1873. [PMC free article] [PubMed]
14. Zhuo Y, Wang S, Gobl A, Oberg K. The interferon-α regulation of interferon regulatory factor 1 (IRF-1) and IRF-2 has therapeutic implications in carcinoid tumors. Ann. Oncol. 2000;11:707–714. [PubMed]
15. Hochhaus A, Yan XH, Willer A, Hehlmann R, Gordon MY, Goldman JM, Melo JV. Expression of interferon regulatory factor (IRF) genes and response to interferon-α in chronic myeloid leukemia. Leukemia (Baltimore) 1997;11:933–939. [PubMed]
16. Taniguchi T. Transcription factors IRF-1 and IRF-2: linking the immune responses and tumor suppression. J. Cell. Physiol. 1997;173:128–130. [PubMed]
17. Mocellin S, Wang E, Marincola FM. Cytokine and immune response in the tumor microenvironment. J. Immunother. 2001;24:392–407. [PubMed]
18. Marincola FM, Wang E, Herlyn M, Seliger B, Ferrone S. Tumors as elusive targets for immunotherapy. Lancet Oncol. 2002 in press.
19. Banerjee SS, Coyne JD, Menasce LP, Lobo CJ, Hirsch PJ. Diagnostic lessons of mucosal melanoma with osteocartilagenous differentiation. Histopathology. 1998;33:255–260. [PubMed]
20. Ross DT, Scherf U, Eisen MB, Perou CM, Rees CA, Spellman PT, Iyer V, Jeffrey SS, van de Rijn M, Waltham M, Pergamenschikov A, Lee JC, Lashkari D, Shalon D, Myers TG, Weinstein JN, Botstein D, Brown PO. Systematic variation in gene expression patterns in human cancer cell lines. Nat. Genet. 2000;24:227–235. [PubMed]
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...