The metastatic microenvironment: Brain-residing melanoma metastasis and dormant micrometastasis



Brain metastasis occurs frequently in melanoma patients with advanced disease whereby the prognosis is dismal. The underlying mechanisms of melanoma brain metastasis development are not well understood. We generated a reproducible melanoma brain metastasis model, consisting of brain-metastasizing variants and local, subdermal variants that originate from the same melanomas thus sharing a common genetic background. The brain-metastasizing variants were obtained by intracardiac inoculation. Brain metastasis variants when inoculated subdermally yielded spontaneous brain dormant micrometastasis. Cultured cells from the spontaneous brain micrometastasis grew very well in vitro and generated subdermal tumors after an orthotopic inoculation. Expression analysis assays indicated that the brain metastasis and micrometastasis cells expressed higher levels of angiopoietin-like 4, prostaglandin-synthesizing enzyme cyclooxygenase-2, matrix metalloproteinase-1 and preferentially expressed antigen in melanoma and lower levels of claudin-1 and cysteine-rich protein 61 than the corresponding cutaneous variants. The reproducible models of human melanoma metastasizing experimentally and spontaneously to the brain will facilitate the identification of novel biomarkers and targets for therapy and contribute to the deciphering of mechanisms underlying melanoma metastasis.

Brain metastasis represents a significant cause of death in melanoma patients, and its frequency is increasing,1 possibly as a result of new therapies prolonging patient survival.2 Of all human solid tumors, malignant cutaneous melanoma has one of the highest risks to develop brain metastasis. More than 40% of advance stage melanoma patients are treated for complications due to brain metastasis.1, 3

Treatment options for melanoma patients with cerebral brain metastasis are limited and not effective to date.4 Tumor cells with the potential to metastasize and colonize the brain may express distinctive molecular determinants that promote metastasis formation in this organ. They may also be able to respond to brain-derived growth factors or to deliver signals that alter the brain microenvironment, making it more supportive to metastasis development.3 Prevention strategies for brain metastasis could be used if cells expressing such molecules could be identified in the primary melanoma. Currently, such molecular biomarkers are unknown.

Human to mouse melanoma xenograft models that recapitulate the phenotypes seen in the clinic provide a valuable resource of cells for translational research and can accelerate drug discovery processes for this disease.5 Current human melanoma brain metastasis models consist of xenografted cells inoculated into immune-deficient mice mainly by intracarotid or intracardiac administration.6 Although these types of injections bypass the initial steps of brain metastasis (separation from the primary neoplasm, invasion and release into blood vessels or lymphatics), all subsequent steps in the process (arrest in brain capillary bed, extravasation, growth and angiogenesis) must occur for metastasis to succeed.7 Attempts to isolate and culture human melanoma cells capable of spontaneously metastasizing from primary tumors to the brain were mostly unsuccessful. A recent report described the generation of a spontaneous xenograft model of brain metastasis of human melanoma. Severe combined immunodeficiency (SCID) mice subdermally inoculated with cells derived from xenografted melanoma lung metastasis and treated with metronomic chemotherapy developed spontaneous brain metastasis.8, 9

Ongoing studies in our laboratory deal with site-specific metastasis of melanoma to the brain and the crosstalk between these tumor cells and components of their microenvironment and also the downstream effects of such interactions. To facilitate such studies, we set out to develop an orthotopic mouse model for human melanoma metastasis.

In our study, we characterize novel xenograft human melanoma models, composed of cutaneous and brain-metastasizing variants that originated in three different melanomas. Three cycles of in vivo passage of cells from one of these human melanomas in the brain of nude mice yielded a variant, which when injected subdermally into nude mice disseminated spontaneously to the brain and produced dormant micrometastasis.

The xenograft melanoma models were used for the discovery of genes that are distinctively overexpressed or underexpressed in brain metastasis, as compared to cutaneous melanoma from which they have developed. The biologic validation and functions of these genes would advance the deciphering of the mechanisms underlying melanoma brain metastasis and may lead to novel diagnostic and therapeutic approaches. For example, we recently suggested that the chemokine (C-C motif) receptor 4 (CCR4)–chemokine (C-C motif) ligand 22 (CCL22) axis may be involved in the attraction of melanoma cells to the brain.10 Such genes may promote or suppress brain metastasis.


ANGPTL4: angiopoietin-like 4; BDMC: brain disseminated metastasizing cells; BMMC: brain-metastasizing melanoma cells; CLDN1: claudin-1; CYR61: cysteine-rich protein 61; FC: fold change; GFAP: glial fibrillary acidic protein; GO: gene ontology; H&E: hematoxylin and eosin; IC: intracardiac; IP: intraperitoneally; MMP: matrix metalloproteinase; PRAME: preferentially expressed antigen in melanoma; PTGS2: prostaglandin-endoperoxide synthase 2; RS9: ribosomal protein S9; RT-qPCR: quantitative real-time PCR; SC: subcutaneous

Material and Methods

Cell Culture

Primary tumor-derived human melanoma cell lines RALL and RKTJ and human melanoma brain metastasis-derived cell line YDFR were kindly provided by Prof. Michael Micksche (Department of Applied and Experimental Oncology, Vienna University, Austria) and described by Berger et al.11 All human melanoma cells were maintained as described by Izraely et al.,10 and immortalized human brain microvascular endothelial cell line (hCMEC/D3) human brain endothelial cells were maintained as described by Weksler et al.12 The cultures were tested and determined to be free of Mycoplasma.


Male athymic nude mice (Balb/c background), 7–10 weeks of age were purchased from Harlan Laboratories (Jerusalem, Israel). The mice were housed and maintained in laminar flow cabinets under specific pathogen-free conditions in the animal quarters of Tel Aviv University and in accordance with current regulations and standards of the Tel Aviv University Institutional Animal Care and Use Committee.

Orthotopic and experimental inoculation of tumor cells

For in vivo inoculation, cells were harvested by trypsinization and transferred into roswell park memorial institute medium (RPMI)-1640 medium supplemented with 5% fetal calf serum (FCS).

Subdermal inoculation

To generate subcutaneous (SC) tumors, 1 × 106 cells in 100 μl of 5% FCS RPMI-1640 medium were inoculated subdermally into the right thigh of mice.

Intracardiac inoculation

Prior to intracardiac inoculation, nude mice were anesthetized by ketamine (100 mg/kg body mass) and xylazine (10 mg/kg body mass; Kepro Deventer, The Netherlands) administered intraperitoneally.13 Using a small animal ultrasound device (Vevo 770 High-Resolution In Vivo Micro-Imaging System; VisualSonics, Toronto, Canada), 5 × 105 cells in 50 μl of 5% FCS RPMI-1640 medium were inoculated into the left ventricle of the heart, using a 29-gauge needle.

Necropsy procedure and histopathologic studies

Mice were sacrificed; local SC tumor, brain and lungs were harvested. The organs were immediately divided into three parts: one part was minced and put in culture, the second was frozen and stored at −70°C until used for RNA extraction and the third was fixed in 4% buffered formalin (Bio-Lab, Jerusalem, Israel) and processed for routine histology. Sections were stained with hematoxylin/eosin solution (H&E) and evaluated by light microscopy for the presence of metastases.

Magnetic resonance imaging

Mice were anesthetized with isoflurane (1–3%) in 1 l/min oxygen 98% and restrained in a prone position on the magnetic resonance imaging (MRI) mouse head-holder (Bruker, Rheinstetten, Germany). Respiration rate was monitored and maintained at 30–60 breath/min throughout the experimental period. Mice were scanned in a 7T/30 spectrometer (Bruker) using a 10-mm quadrature surface coil dedicated to the mouse head and 400 mT/m gradient system. Contrast-enhanced T1-weighted images (spin echo, repetition time (TR)/echo time (TE) 600/11 msec and two or four averages) were acquired 10 min after intraperitoneal injection of the contrast agent (150 μl of 0.5 mmol/kg gadopentetate dimeglumine (Gd) diluted in saline [Magnetol, Soreq, Israel]). In all experiments, the field of view was 2 cm, and 12 axial contiguous slices of 1-mm thickness were used with matrix dimensions of 256 × 128 (zero filled to 256 × 256).


The following antibodies were used for flow cytometry: anti-human leukocyte antigen (HLA)-A, -B and -C mAb (W6/32)14 and anti-H-2 mAb (20-8-4S)15 that were kindly provided by Dr. R. Ehrlich (Department of Cell Research and Immunology, Tel Aviv University) and were used at the dilutions of 1:500 and 1:3,000, respectively. Monoclonal anti-human MelanA (Abcam, Cambridge, UK) was used at a dilution of 1:30. Anti-human matrix metalloproteinase 2 (MMP-2) mAb (EMD Chemicals, San Diego, CA) at a concentration of 1 μg/ml was used for Western blotting. Polyclonal rabbit anti-glial fibrillary acidic protein (GFAP; Dako, Glostrup, Denmark) was used for immunohistochemistry staining. Fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse IgG and goat anti-rat IgG and horseradish peroxidase (HRP)-conjugated goat anti-mouse were used according to the manufacturer's instructions (Jackson ImmunoResearch Laboratories, West Grove, PA).

Flow cytometry

Flow cytometry was performed as previously described.10 For intracellular staining of MelanA, cells were incubated with absolute methanol at −20°C for 30 min and rinsed twice with fluorescence-activated cell sorting (FACS) medium (RPMI-1640 supplemented with 5% FCS and 0.01% sodium azide), before the staining procedure. Antigen expression was determined using Becton Dickinson FACSort and CellQuest software. Baseline staining was obtained by labeling the cells with secondary antibodies alone.


Formalin-fixed, paraffin-embedded 5-μm-thick sections were deparaffinized with xylene. The sections were treated with 10 mM citrate buffer, pH 6.0 at 120°C for 20 min, cooled to room temperature and then blocked with Protein Block Serum-Free (DakoCytomation, Carpinteria, CA) to block nonspecific staining. The sections were incubated with a relevant antibody overnight at 4°C. The sections were then labeled with a secondary Link-Streptavidin HRP solution (DakoCytomation), developed with 3,3′-Diaminobenzidine (DAB)+ chromogen (DakoCytomation) and counterstained with hematoxylin and examined at ×400 magnification. Control slides were incubated with secondary antibodies alone or with matching isotype control antibodies.

In vivo tumorigenicity assay

To test the tumorigenic properties of derived cell lines, 1 × 106 cells in 100 μl of 5% FCS RPMI-1640 medium were injected subcutaneously into nude mice. Local SC tumors were measured once a week using a caliper. Tumor volume was evaluated by the ellipsoid volume calculation formula 0.5 × (length × width2).16, 17

RNA preparation and reverse transcription polymerase chain reaction

For quantitative real-time polymerase chain reaction (RT-qPCR), total cellular RNA was extracted using EZ-RNA Total RNA Isolation Kit (Biological Industries). RNA concentrations were determined by the absorbance at 260 nm. RNA samples were used for first-strand cDNA synthesis using the Moloney Murine Leukemia Virus (M-MLV) Reverse Transcriptase (Ambion, Austin, TX).

For Affymetrix microarray assay, total RNA was isolated using the Qiagen RNeasy Mini Kit (Qiagen, Valencia, CA). Samples were treated using DNase Set (Qiagen). RNA concentrations were determined by the absorbance at 260 nm, and quality control standards were A260/A280 = 1.8–2.0.

Quantitative real-time PCR

Quantification of cDNA targets was performed on Rotor-gene 6000TM (Corbett life science, Australia) using Rotor gene 6000 series software. Reactions were run two to three times, in duplicates. Transcripts were detected using SYBR Green I (Thermo Fisher Scientific, ABgene, Hamburg, Germany).

For the detection of human melanoma cells, in each reaction two pairs of specific primers were used for: (i) human β2 microglobulin (β2M; does not react with mouse cDNA) and (ii) normalizing gene, ribosomal protein S9 (RS9; reacts with human and mouse cDNA to yield one product). This method enabled us to identify the presence of human cells (human mRNA copies) in brains or lungs of mice inoculated with the YDFR.C or YDFR.CB3 variants. Murine DA3 cells were used as negative control. PCR amplification was performed over 40 cycles (95°C for 15 sec, 59°C for 20 sec and 72° for 15 sec). Melting curves for each primer set had a single product, and no template controls were negative after 40 cycles, indicating the absence of a nonspecific product. Primer sequences are described in Supporting Information Table 1.

Cell growth

A total of 2 × 105 cells/well were plated in a six-well tissue culture plate in growth medium. Cells were counted after 24, 48 and 72 hr.

Gelatin substrate zymography and Western blotting analysis

Melanoma cells were plated in growth medium. After an overnight incubation, growth medium was removed and replaced by serum-free RPMI-1640 medium for an additional 24 hr. Medium was collected for further testing. Activity of MMPs in the conditioned medium was determined as previously described.18

For the detection of active MMP-2, conditioned medium was collected and centrifuged to discard cell debris. After collecting the supernatants and adding Laemmli sample buffer, samples were resolved on sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto nitrocellulose membrane. The membrane was blocked at room temperature with 3% bovine serum albumin (BSA) diluted in Tris-buffered saline (TBS)-Tween for 1 hr. The target proteins were detected by using the relevant primary antibodies and suitable HRP-conjugated secondary antibodies. The bands were visualized by chemoluminescence-enhanced chemiluminescence (ECL) reaction (Amersham, Buckinghamshire, United Kingdom) and autoradiographed by exposure to Fuji film.

Adhesion to hCME/D3 cells

At 37°C, 96-well culture plates were coated with 100 μg/ml Collagen type I for 1 hr. After one wash with Phosphate buffered saline (PBS) 1×, 5 × 104 hCMEC/D3 cells were cultured for 24 hr to form a confluent monolayer. Adhesion of live melanoma cells fluorescently tagged with carboxyfluorescein diacetate (CFDA—a fluorescent tag of live cells) onto activated hCMEC/D3 cells was performed as previously described.19

The total fluorescent signal of all labeled cells added to the well, before removing the nonadherent cells, was measured by a fluorescent ELISA reader (BioTek FL500, Biotek instruments, Winooski, VT) at wavelength of 490/530 (OD490/530). To estimate the number of adherent cells, wells were washed with PBS 1× to remove the nonadherent cells, and fluorescence of the adhering cells was measured as above. To obtain the percentage of adherent cell, the ratio between the optical density (OD) of the adherent cells and that of the total cells plated has been calculated. The results are presented as percent of adherent cells.

Microarray analysis

cDNA of cutaneous or brain metastasis variants was hybridized to GeneChip Human Exon 1.0 ST arrays (Affymetrix, CA). Partek® Genomics Suite™ software (version 6.5; Partek, St. Louis, MO; was used for microarray analysis. Raw data (CEL files) were normalized at the transcript level using robust multiaverage method.20 Median summarization of transcript expressions was calculated. Gene-level data were then filtered to include only those probe sets that are in the “core” meta-probe list, which represent RefSeq genes and full-length GenBank mRNAs. Analysis of variance (ANOVA)21 for p values was used to identify differentially expressed genes. Tissue type (cutaneous versus brain metastasis variants) was used as the candidate variable in the ANOVA model to obtain metastasis-specific genes. Differentially expressed gene lists with p < 0.05 and fold change (FC) >1.25 or <−1.25 were obtained for the various conditions.

Gene expression analysis and gene ontology

Gene and protein names, symbols and accession numbers were identified by Partek® Genomics Suite™ and by using various Bioinformatics databases (National Center for Biotechnology Information [NCBI] RefSeq: http:// and UniProt:

The identification of genes differentially expressed in all three melanoma models was performed using the Venny tool ( Gene Ontology (GO) annotations were analyzed with Panther Protein Classification System ( to identify functional annotations that were significantly enriched (p < 0.05) in a given gene set compared to the entire human genome annotations.

Statistical analysis

Paired or unpaired Student's t-test was used to compare in vitro results and tumor size. Survival analysis was computed by the Kaplan–Meier method.


Generation of cutaneous and brain metastasis variants of human melanoma

Melanoma cells from the RALL, RKTJ and YDFR cell lines were inoculated subdermally. The local tumors were harvested and established in culture and referred to as cutaneous (C) variants (Fig. 1a).

Figure 1.

Generation of cutaneous and brain metastasis variants of human melanoma with an identical genetic background. (a) Generation of three brain metastasis models composed of cutaneous and brain metastasis variants (as described in Material and Methods section). (b) Identification of brain metastatic lesions in the brains of RALL.C (upper row) and YDFR.C (lower row) intracardially injected mice. (I) A T1-weighted MR image of melanoma-inoculated mouse brain after 1:5 gadopentetate dimeglumine-diethylene triamine pentaacetic acid (Gd-DTPA) IP injection (the lesion in circled). Image of excised brain, showing dark melanotic lesions (arrowheads). Brain sections were stained with H&E and evaluated by light microscopy for the presence of metastasis. Magnification: 400×, Scale bar: 100 μm. (c) Immunohistochemistry staining of GFAP (brown) in a brain section of a mouse inoculated with YDFR.CB1. The brain metastatic lesion (Met, right image) is densely surrounded with reactive astrocytes (arrowheads). The left image shows a nonmetastatic area in the same brain. Magnification: 400×, Scale bars: 100 μm. [Color figure can be viewed in the online issue, which is available at]

Cells from these cultures were inoculated into the left ventricle of the heart of athymic nude mice. Brain metastasis was confirmed by MRI. Animals were sacrificed when reduced body mass occurred. Visible melanotic lesions observed in the brain and H&E staining confirmed the presence of tumor cells (Fig. 1b).

To increase the brain-seeking capacity of the brain-metastasizing variants, we used an approach undertaken previously.18, 23–27 Cells were passaged in the brain for two or three successive passages yielding brain metastasis variants (CB1, CB2 or CB3—the numerals represent the number of selection cycles). All these variants were established as cell lines.

All the melanoma variants strongly expressed human HLA class I antigens and the human melanoma differentiation antigen MART-1 and did not express mouse H-2 class I antigens indicating their identity as human melanoma cells (data not shown).

As seen in Figures 1bIII and 1c, the brain metastasis was located in the brain parenchyma.

Brain sections stained for GFAP, which is particularly high in motile, hypertrophic reactive astrocytes in injured brain tissue,28 demonstrated astrogliosis in the metastatic loci (Fig. 1c), a common feature in brain metastasis patients.28 The infiltration of astrocytes into the inner tumor mass may indicate contact-dependent interactions between melanoma cells and astrocytes in the brain or may reflect cytokines/chemokines secretion (such as stromal cell-derived factor-1 (SDF1α)) by the tumor cells attracting astrocytes. Assessment of the nonmetastatic area in the same brain did not show the same pattern of astrocytes organization and density.

Orthotopically inoculated melanoma cells form micrometastasis in brain and lungs

We next determined the metastatic potential of orthotopically (subdermally) inoculated YDFR.C and YDFR.CB3 variants. The inoculated nude mice were killed when reduced body mass occurred (91–197 days postinoculation), and local tumors, brains and lungs were harvested.

Neither lungs nor brain exhibited macroscopic metastasis (seen with the naked eye). H&E staining of histological brain sections and microscopic examination did not reveal the presence of melanoma cells in these brains.

Portions of the brain and lungs of the inoculated mice were minced and cultured at standard conditions. About 2 weeks later, spreading cells appeared in cultures established from the brains of three of four mice inoculated with the brain metastasis variant but in none of the cultures from brains of mice injected with the cutaneous variant (Table 1). Both the cutaneous variant as well as the brain-metastasizing variant disseminated equally to the lung (Table 1).

Table 1. Metastatic potential of YDFR.C and YDFR.CB3 cells inoculated into mice
inline image

Flow cytometry analysis was used to immunophenotype the brain- and lung-disseminated cells. All the variants strongly expressed the human melanoma MART-1 antigen and human HLA class I antigens and did not express mouse H-2 class I antigens (data not shown). These results confirmed the identity of the cells as human melanoma cells. Cultures from these cells were designated as brain micrometastatic variants (SB1, SB2 and SB3) and lung micrometastatic variant (SL3). Cultures of local tumors that developed in mice inoculated with YDFR.CB3 cells were designated as CB3C1, CB3C2 and CB3C3. These results obtained thus far only with the YDFR tumor indicated that human melanoma cells inoculated orthotopically into nude mice may disseminate spontaneously to the brain and lungs of such mice.

A RT-qPCR assay developed in our laboratory for the detection of xenografted human cells in nude mice (see Material and Methods section) was used to confirm these results. Table 1 demonstrates that brains of three of four mice inoculated with the brain-metastasizing YDFR.CB3 variant contained human cells, and that the brain of one of five mice inoculated with the cutaneous variant contained human cells. The results of a preliminary experiment indicated that another brain-metastasizing variant, YDFR.CB2, also yielded spontaneous brain micrometastasis after a subdermal inoculation. The RT-qPCR assay also confirmed that both the cutaneous as well as the brain-metastasizing variants disseminated equally to lungs.

Tumorigenicity profile of cutaneous and brain-metastasizing melanoma variants

To determine the tumorigenicity profile of the cutaneous YDFR.C, the brain-metastasizing YDFR.CB3 and the brain-micrometastasizing YDFR.SB3 variants, cells from these variants were each injected into the orthotopic site (i.e., subdermally) of 6–11 athymic nude Balb/c mice.

Both the cutaneous as well as the brain-metastasizing variants formed large visible SC tumors. After 23 days of inoculation, the tumors formed by the YDFR.CB3 variant were eightfold larger than those formed by the YDFR.C variants (p < 0.05; Fig. 2a). After 16–45 days of inoculation, the tumors formed by the YDFR.SB3 variant were 1.4–8-fold larger than those formed by the YDFR.C variants (p < 0.05).

Figure 2.

Tumorigenicity profile of cutaneous and brain metastasis variants of human melanoma. Cells (1 × 106) were injected subdermally into the right thigh of athymic nude Balb/c mice (n = 6–11 per group). Mice were monitored weekly after the inoculation. The mice were killed when moribund and necropsied. (a) Tumor volume throughout 62 days after implantation of YDFR.C, YDFR.CB3 and YDFR.SB3 cells into nude mice. Tumor dimensions were measured using a caliper and volume was obtained by the ellipsoid volume calculation formula 0.5 × (length × width2). YDFR.CB3 versus YDFR.C, p < 0.05 (t = 23 days); YDFR.SB3 versus YDFR.C, p < 0.05 (t = 16-45 days). (b) Kaplan–Meier survival curve of tumor bearing mice (not significant).

The Kaplan–Meier survival curves of cutaneous and brain-metastasizing variants (Fig. 2b) demonstrate that most of the YDFR.CB3-inoculated mice died earlier than the YDFR.C-inoculated mice. Median survival of YDFR.C- and YDFR.CB3-inoculated mice was 175 and 91 days, respectively. These results demonstrate that the brain-metastasizing variants are more aggressive than the corresponding cutaneous variants.

In vitro proliferation of cutaneous and brain- metastasizing melanoma variants

The in vitro growth curves of the cutaneous variants and the brain-metastasizing variants of RALL, RKTJ and YDFR showed no significant differences in proliferative capacity of the cutaneous and the brain metastasis variants under normal culture conditions (results not shown). This indicates that although these cells express different malignancy phenotypes in vivo, their intrinsic capacity to proliferate is identical.

Melanoma brain micrometastases are dormant

When inoculated subdermally to nude mice, the brain-metastasizing variant YDFR.CB3 (that originated in a human melanoma brain metastasis11) disseminated spontaneously to brain. The brain micrometastatic melanoma cells did not develop into macroscopic metastasis but remained as single cells or small clusters in the brain, hardly visible by histology. When transferred to in vitro culture conditions, the brain micrometastatic melanoma cells proliferated well.

To determine whether the brain micrometastatic variants can form cutaneous tumors when reinjected into athymic nude mice, YDFR.SB2 and YDFR.SB3 cells were subdermally inoculated into such mice. All the mice (n = 4/group) formed SC tumors. Median survival time of mice inoculated with YDFR.SB2 and YDFR.SB3 cells was 126 and 136 days, respectively (data not shown).

Taken together, these results indicate that the micrometastatic melanoma cells are dormant while residing in the brain but are capable of propagation when transferred to another microenvironment or to in vitro conditions. It is not unlikely that factors in the brain microenvironment inhibit proliferation of these cells. These results also indicate that the brain-metastasizing melanoma variants (generated by repeated in vivo selection cycles in the brain) contained a subset of cells which can spontaneously and selectively disseminate to the brain from orthotopic sites.

Secretion of MMP-2 and MMP-9 from cutaneous and brain metastasis variants

The secretion of MMPs is a phenotypic characteristic of malignant cancer cells.29–31 Gelatin zymography was used to determine whether the metastatic potential of the various melanoma variants correlated with their ability to secrete MMP-2 and MMP-9.

Only the MMP-2 gelatin-degrading enzyme (66 kDa) was secreted by the tested cells. The brain metastasis variants of two melanomas displayed a higher activity of MMP-2 than the corresponding cutaneous variants (Fig. 3a). Using Western blot, we confirmed that the proteolytic activity was mediated by the active form of MMP-2 (data not shown). These results suggest that the brain-metastasizing variants have a greater ability to degrade extracellular matrix (ECM), a phenotype that enhances cell motility, invasion and metastasis.30

Figure 3.

In vitro characterization of cutaneous and brain metastasis variants. (a) Gelatin zymography of cutaneous and brain metastasis variants supernatants. A representative image of one out of three experiments is demonstrated. (b) Adhesion of CFDA-stained cutaneous and brain metastasis variants of human melanoma to human brain endothelial cells (hCMEC/D3) was measured. A total of six replicates were performed in each experiment. The graphs represent the average % adherent cells ± SD of three independent experiments. * p < 0.05, **p < 0.005. In both assays, RALL cutaneous and brain metastasis variants did not exhibit significant differences.

Adhesion of melanoma variants to human brain endothelial cells

To form brain metastasis, melanoma cells have to transmigrate through the brain endothelium. An initial step in this process is the adhesion of cancer cells to the endothelium.1

In this series of experiments, we compared the adhesion ability of CFDA-stained RALL, RKTJ and YDFR cutaneous and brain metastasis variants to the human brain endothelial hCMEC/D3 cells. Figure 3b demonstrates that in two melanomas the brain-metastasizing variants adhered significantly better to brain endothelial cells than the corresponding cutaneous variants.

Brain-metastasizing melanoma variants highly express genes that mediate breast cancer brain metastasis

Prostaglandin-synthesizing enzyme cyclooxygenase-2 (PTGS2), collagenase-1 (MMP1) and angiopoietin-like 4 (ANGPTL4) are upregulated in brain metastasis of breast cancer.32 RT-qPCR analyses indicated that PTGS2, ANGPTL4 and MMP1 genes are also upregulated in melanoma brain metastasis as compared to the corresponding cutaneous variants of all three melanoma models (Fig. 4a).

Figure 4.

Genes that are differentially expressed in cutaneous and brain metastasis melanoma variants. (a) PTGS2, MMP1 and ANGPTL4 mRNA expression in cutaneous and brain metastasis human melanoma variants of all three melanoma models (RALL, RKTJ and YDFR) was analyzed using RT-qPCR. The obtained values were normalized to RS9. (b) PTGS2 and ANGPTL4 mRNA expression in cutaneous variants, spontaneous brain metastasis variants and spontaneous lung metastasis variant were analyzed using RT-qPCR. The obtained values were normalized to RS9. (c) CLDN1 and CYR61 mRNA expression in cutaneous and brain metastasis variants of all three melanoma models and PRAME mRNA expression in cutaneous and brain micrometastssis and macrometastasis variants of the YDFR model were analyzed using RT-qPCR. The obtained values were normalized to RS9.

PTGS2 and ANGPTL4 were also upregulated in the brain micrometastatic melanoma cells (Fig. 4b). The expression of ANGPTL4 was higher in the brain micrometastatic cells than in the lung metastasis-derived cells. This may imply that the upregulation of this gene is specific to brain metastasis.

Gene expression of melanoma cells metastasizing to brain

The identification of differentially expressed molecules in primary tumors and in corresponding brain metastases contributes to the understanding of the metastatic process to this organ.33 Moreover, the expression of genes that characterize melanoma brain metastasis by cells in primary cutaneous melanoma might indicate that these cells are likely to spread specifically to the brain and could thus serve as specific targets for antimetastasis therapy.

Using GeneChip Human Exon 1.0 ST arrays (Affymetrix), we generated gene-expression profiles of cutaneous variants, the corresponding first cycle brain-metastasizing variants and a second cycle brain-metastasizing variant.

The data were statistically analyzed by calculating a two-way ANOVA (p < 0.05), using Partek® Genomics Suite™ software. Only genes that passed the threshold cut-off of FC < −1.25 or FC > 1.25 were considered downregulated or upregulated (respectively) in the brain metastasis variants compared to the cutaneous variants. We collected lists of downregulated or upregulated genes between: RALL.CB1 vs. RALL.C, RKTJ.CB1 vs. RKTJ.C, YDFR.CB1 vs. YDFR.C and YDFR.CB2 vs. YDFR.C. Venn diagram analysis was performed to provide one common list of differentially expressed genes, showing that 353 genes were differentially expressed in the four lists (data not shown).

To focus on genes that were differentially expressed in the same pattern in all three melanoma models, we screened for those genes whose expression was either downregulated or upregulated in all the brain metastasis variants. We found that 35 genes followed these criteria: 16 genes were downregulated and 19 were upregulated in the brain metastasis variants (Table 2).

Table 2. Differentially expressed genes in cutaneous and brain-metastasizing variants of three human melanoma xenograft models (RALL, RKTJ and YDFR), as obtained in Affymetrix microarray analysis
inline image

Biological properties of the differentially expressed genes

To classify the 353 differentially expressed genes into biological categories, we analyzed GO annotations of these genes with PANTER. Data were compared to the entire NCBI reference list of the human genome. The 353 differentially expressed genes were classified into biological processes, molecular functions and pathways (Supporting Information Table 2). Significantly enriched functional categories were: biological processes such as cell adhesion, cell cycle and signal transduction; molecular functions such as pyrophosphatase activity and transmembrane cytoskeletal protein binding and pathways such as p53 pathway.

Genes differentially expressed in cutaneous and brain-metastasizing melanoma variants

In the next series of experiments, we compared the expression of four molecules known to be associated with the metastatic process, in the cutaneous and brain-metastasizing melanoma variants.


Claudin-1 (CLDN1), a member of the claudin family, is an integral membrane protein and a component of tight junction strands. Loss of tight junction function may lead to loss of cell–cell adhesion and promotes metastasis.34

The Affymetrix GeneChip Exon array (see above) demonstrated that the brain-metastasizing variants of all three melanoma models expressed lower levels of the CLDN1 gene than the corresponding cutaneous variants (Table 2). To validate this result, cutaneous and brain metastasis variants were analyzed for CLDN1 mRNA expression by RT-qPCR. The RT-qPCR results were consistent with the microarray findings (Fig. 4c).


It has been demonstrated recently, that overexpression of the cysteine-rich protein 61 (CYR61) decreases melanoma cell motility, invasion and MMP-2 activity and inhibits tumor growth and metastasis, identifying CYR61 as a tumor suppressor gene in human melanoma cells.35

RT-qPCR results indicated (Fig. 4c) that CYR61 expression was lower in the brain-metastasizing variants than in the corresponding cutaneous variants of all three melanoma models, supporting the notion that CYR61 may function as a melanoma suppressor gene.


Preferentially expressed antigen in melanoma (PRAME) encodes a melanoma-associated antigen recognized by cytotoxic lymphocytes. In melanoma cell lines, it inhibits retinoic acid-induced differentiation, growth arrest and apoptosis.36 We compared the expression of PRAME in the cutaneous variant of YDFR and in the corresponding brain metastatic and micrometastatic variants using RT-qPCR. The results indicated that PRAME expression was higher in the brain metastasis and micrometastasis variants than in the cutaneous variant (Fig. 4c). These results suggest that PRAME may serve as a molecular marker for melanoma brain metastasis.


The intracardiac inoculation of human melanoma cells to nude mice resulted in the formation of brain metastasis. We have at our disposal three sets of local (subdermal) and brain-metastasizing variants (designated hereafter brain-metastasizing melanoma cells [BMMCs]) that originated in three different human melanomas. Each set of variants originated in the same human melanoma thereby ensuring a common genetic background. Molecular differences between the local and the brain-metastasizing variants can thus be ascribed to their different metastatic phenotype. Similar models exist for breast cancer, lung adenocarcinoma and fibrosarcoma.5, 9

The orthotopic inoculation of BMMCs of one melanoma cell line produced spontaneous micrometastasis in the brains of nude mice (designated hereafter brain micrometastasis [BM-M]). Such an observation is not unique to melanoma. Formation of spontaneous micrometastasis in organ sites that also harbor macrometastasis was recently reported by us. Nude mice inoculated orthotopically (into the adrenal gland) with human neuroblastoma cells presented with dormant micrometastasis in the lungs, an organ site that also harbored neuroblastoma metastasis.37

As the brain microenvironment supported the proliferation of BMMC but not that of the BM-M and in view of the fact that the latter cells form tumors when inoculated subdermally and proliferate in vitro, we concluded that the BM-M cells are dormant in the brain microenvironment. It is possible that the brain microenvironment expresses growth-restraining molecules that specifically target the BM-M, which proliferate and die in comparable rates and thus will not grow progressively and remain dormant.38

At some point, the micrometastatic cells may be stimulated by signals from the brain microenvironment and acquire properties that will drive their progression to fatal macrometastasis.

Cruz-Munoz et al. previously described the generation of brain metastasis from melanoma in SCID mice. Our model differs from that of Cruz-Munoz by several parameters, in particular their model shows full blown metastasis whereas we developed a xenograft melanoma model that contains both spontaneous dormant BM-M as well as macrometastasis with an identical genetic background. Moreover, metastatic dormancy of melanoma was investigated thus far only with regards to lymph node micrometastasis.36, 39

Melanoma cells “pre-adapted” to the brain microenvironment generate spontaneous BM-M more efficiently than “virgin” melanoma cells which are newcomers to this microenvironment. However, both pre-adapted as well as virgin melanoma cells have an equal capacity to generate lung micrometastasis. Therefore, we conclude that the spontaneous dissemination of melanoma cells to the brain is a nonrandom event enhanced by previous exposure of the cells to factors present in the brain microenvironment and that local and brain-metastasizing melanoma cells interact differently with the brain and the lung microenvironments.

Alterations in gene expression patterns occurring during tumor progression control the ability to respond to microenvironment-derived signals.3 Such alterations could confer on a minor subset of tumor cells the ability to disseminate from the primary tumor to secondary organ sites and subsequently form metastasis.40

The profiling experiments summarized above demonstrated differences in gene expression patterns between BMMC and the corresponding local variants from the three different melanomas analyzed in our study. The array results demonstrated that 35 genes were differentially expressed in BMCC and in the local, subdermal variants. Differences between the expression of other genes, some of which detected by the GeneChip Exon array and linked to the metastasis process were also detected in BMMC and the local subdermal variants.

Obviously, some of the effects of the brain microenvironment on gene expression of the melanoma cells that were previously exposed to this microenvironment would fade away after their explantation to culture conditions. However, it is established, in several biologic systems, that the downstream effects of exogenous signals can endure for extended periods of time or even be permanent.41–43

The gene expression pattern of BM-M was in general similar to that of BMMC. Based on the biological differences between these two types of brain-localizing melanoma cells, a thorough comparative analysis between these cells is warranted. Differentially expressed molecules in these two types of brain-localizing melanoma cells may serve as biomarkers and therapy targets for melanoma brain metastasis and micrometastasis.

The functional significance of the genes differentially expressed in BMMC and the subdermal melanoma cells, to brain metastasis is investigated at present and is outside the scope of our study.

We asked if genes that mediate breast cancer metastasis to the brain32 are differentially expressed in BMMC and in local subdermal melanoma variants. Several genes which are highly expressed in breast cancer brain metastasis are also more highly expressed in BMMCs than in the corresponding cutaneous variants, suggesting the existence of a molecular signature of brain metastasis common to several types of cancer.

The melanoma micrometastasis and macrometastasis models may lead to a better understanding of the biology of melanoma dormancy, to solve possible mechanisms underlying the transition of melanoma micrometastasis into macrometastasis and to find ways to induce or prolong melanoma dormancy. This in turn may lead to the improvement of the quality of life of melanoma patients and to a reduction in morbidity and mortality from this disease.

The spontaneous human melanoma BM-M model has been described thus far for a single human melanoma xenograft. This finding has, obviously, to be confirmed using additional melanoma tumors. If established as a general phenomenon, such models would offer an unlimited source for biologic material for melanoma metastasis research needed for the development of effective melanoma treatment modalities.


Our study was supported by The Dr. Miriam and Sheldon G. Adelson Medical Research Foundation (Needham, MA) (I.P.W. and D.S.B.H.).