Defining recurrent regions of amplification and deletion

Further analysis of the focal alterations in the highly rearranged genomes of primary melanomas and metastases delimited the boundaries of informative Minimal Common Regions (MCRs) using a set of heuristically defined rules, including recurrence in 2 or more samples of a CNA spanning regions less than 2Mb in size with a peak log₂ ratio greater than 1.0 (see Methods). In the primary melanomas, this analysis defined 13 MCRs comprising 6 amplifications with a median size of 1.03 Mb (range 0.075–1.97 Mb) containing a total of 84 known genes, and 7 deletions with a median size of 0.32 Mb (range 0.098–0.94 Mb) containing 39 genes (Table 1). In comparison, analysis of the metastasis profiles defined 39 MCRs comprising 24 amplifications with a median size of 0.78 Mb (range 0.046–1.59 Mb) containing a total of 276 known genes, and 15 deletions with a median size of 0.53 Mb (range 0.035–1.7 Mb) encompassing 78 genes (Table 1). Although the cytological bands of MCRs in primary and metastatic melanoma do not entirely overlap, these do not necessarily represent unique events to one or the other melanoma type since they can be present as regions of larger amplifications or deletions or lower amplitude changes (and thus can be excluded from the list of informative MCRs due to the strict criteria used to define these events). The identification of regions of genomic alteration enriched in primary or metastatic melanoma is discussed below.

Of the genes residing within metastases MCRs boundaries (Table 1), many were linked to networks of relevance to carcinogenesis and metastasis. For example, a significant number of genes were involved in G1/S cell cycle transition and in p53-dependent apoptosis (Table 1; MetaCore™ analysis, p<0.01), including p14^ARF, p16^INK4A and p15^INK4B, which were deleted as part of the recurrent 9p21 locus deletion, as well as CDK4 and MDM2, both of which were recurrently amplified in metastatic melanoma (Supplemental Figure S3). Additionally, MetaCore™ analysis identified components of networks governing cell adhesion, motility and cell matrix assembly that were significantly represented among genes mapping to the metastases MCRs (p<0.01). For example, LIPRIN (PPFIA1), a gene known to enhance cell matrix interaction, and Contractin (CTTN), a gene implicated in squamous cell carcinoma migration and metastasis [19], [20], were both recurrently amplified in metastatic melanoma. Conversely, Fibulin 5 (FBLN5), shown to enhance cell adhesion [21], was recurrently deleted in the metastatic samples (Table 1).

IGC analysis of primary and metastastic melanoma genomes

Evolution from primary to metastatic disease is expected to be accompanied by the acquisition of, or selection for, genomic and genetic events that confer biologic capabilities necessary for mestastasis [22]. We thus hypothesized that CNAs observed in metastasis but not detected in primary disease would be more likely to represent potential drivers of metastasis. To define such events, we adopted an integrative genome comparison approach to define genes that were statistically different between primary and metastatic samples based on DNA and RNA data (Figure 2A). First, we employed a statistical test, Fisher-Exact, to delineate regions that were differentially altered in metastatic versus primary melanoma. Briefly, we collapsed all CBS-processed array-CGH profiles of primary or metastatic cohorts down to 2,907 reduced-segments (hereafter as “R-segments”) to generate two R-segment profiles corresponding to primary and metastatic melanoma genomes, respectively. For each R-segments above noise threshold (i.e. Log2 of +/−0.15), Fisher's Exact test p-values were calculated and corrected for multiple testing (see Methods) to define statistically significant events that were different between these two classes. At a false discovery rate (FDR) of 10% (q<=0.1), 300 R-segments spanning 32 contiguous regions of interest (ROIs) were found to be preferentially gained in metastatic relative to primary melanomas in a non-random fashion (Figure 2B). Many of these 32 ROIs clustered predominantly in several chromosomal regions, including 1q, 6p, 7, 17q and 22, and many of the regions were gained in poor-prognosis K3-1 and K3-2 subclasses (Figure 1C and 1D). Of note, no R-segments were found to be preferentially gained in primary relative to metastatic melanomas and no regions of loss were significantly different between primary and metastasis.

Open in a separate window

Figure 2

Integrative genomics identify high-confidence metastasis candidate melanoma genes.

(A) Flow chart of integrating copy number and expression analysis to compare primary and metastastic melanoma genomes. (B) Whole genome q-value profiles based on Fisher's Exact Test between primary and metastastic melanomas. X axis coordinates represent genomic map position and Y axis indicates q-value log10 of Fisher's Exact Test between primary and metastastic melanomas at each R-segment.

Next, we sought to determine whether genes resident in regions of genomic alterations exhibited a pattern of expression reflective of the underlying copy number aberrations. That is, metastasis candidate genes resident in regions that are preferentially gained in metastatic melanomas would be expected to show upregulation on mRNA level when compared to primary melanomas. Accordingly, we utilized the well-established SAM algorithm to identify those genes resident within the 32 Fisher-significant ROIs that exhibited overexpression patterns in metastastic melanomas relative to primary disease. Specifically, of the 1090 Affymetrix probe sets deemed expressed (see Methods) within these 32 ROIs, SAM analyses identified 676 probe sets that showed significant overexpression in metastases (FDR<=0.05). These 676 probes were furthered ranked by the relative fold change of expression to select the top 34 probes corresponding to 30 unique annotated genes exhibiting at least 2-fold overexpression in metastases (Table 2). A number of these genes mapped to chromosome 7, whose gain has been linked to metastasis and poor prognosis in patients with non-small cell lung cancer and peripheral nerve sheath tumors [23], [24]. Although a number of these candidates in Table 2 mapped to known regions of germline CNV, we did not exclude these from further consideration since well-validated cancer relevant genes have been known to reside within regions of germline CNV [25].

Metastasis candidates promote invasion in vitro

Among the 30 candidate metastasis genes is MET, a receptor tyrosine kinase (RTK) whose overexpression has been correlated with progression in multiple cancer types, including melanoma [26]. Indeed, in a Met-driven transgenic mouse model comprised of tyrosinase-driven rtTA and tet-Met transgenes on the Ink4a/Arf null background (Tyr-rtTA;Tet-Met;Ink4a/Arf^−/−, hereafter “iMet”), activation of Met signaling in melanocytes engendered a metastatic melanoma phenotype in vivo (Nogueira C and Chin L, unpublished). Consistent with such metastatic phenotype in vivo, derivative iMet melanoma cells showed robust invasion activity in response to HGF in Boyden chamber invasion assay in vitro (Supplemental Figure S4). Encouraged by this proof of concept validation of IGC, we next utilized this in vitro Boyden chamber invasion assay as a first step to examine the additional metastasis candidates in Table 2.

To this end, we selected 6 genes from the candidate list (ASPM, AKAP9, IMP3, PRKCA, RPA3, and SKAP2) based on literature support (see Discussion) to determine whether their knockdown would impact on the invasion of a human melanoma cell, 1205LU. As shown in Figure 3A, siRNA-mediated knockdown of these candidates resulted in a statistically significant inhibition of invasion in the Boyden Chamber assay compared to a non-targeting siRNA oligo pool (p<0.05, p<0.05, p<0.01, p<0.05, p<0.001, p<0.001, respectively). Correspondingly, we also demonstrated that overexpression of ASPM in WM3211, a weakly invasive human melanoma cell line, consistently increased invasion through matrix in the Boyden Chamber assay (p<0.05; Figure 3B). Similar results were obtained in a second melanoma cell line, WM115 (data not shown).

Open in a separate window

Figure 3

Functional and histopathologic characterization of high-confidence metastasis candidate genes.

(A) Knockdown of 6 candidate metastasis genes by siRNA inhibited 1205LU Boyden Chamber cell migration. Data represents the average of three replicates. Statistical significance was assessed using a Tukey-Kramer Multiple Comparisons Test, in which each target was compared to the effect of a non-targeting siRNA pool. *=p<0.05; **=p<0.01; ***=p<0.001. The level of target mRNA knockdown is shown in Supplemental Figure S5. (B) Exogenous expression of ASPM enhanced invasion through Matrigel compared to empty vector control on a modified Boyden Chamber assay. Representative images of Boyden chamber assays are shown on the right. Data represent three independent experiments. (C) Immunohistochemical survey of Survivin on a melanoma progression tissue microarray. Survivin expression was scored as 0–3+ (see Methods). Percent of TMA cores scored 0 to 3+ for major histopathlogical categories (benign nevi, thin and thick primary cutaneous melanomas, lymph node and visceral metastases) are plotted with p values calculated by χ² test shown in the table below. Representative cores are shown to demonstrate, from top to bottom, intensity of cytoplasmic Survivin expression scored as 0 for no staining, 1+ for mild stain intensity, 2+ for moderate stain intensity, and 3+ for intense stain intensity.

Melanoma samples and DNA extraction

The primary and metastatic melanoma samples analyzed in this study were obtained from three centers: The Medical University of Vienna, Austria (Supplemental Table S1), the Memorial Sloan Kettering Cancer Center of New York, NY and the Brigham and Women's Hospital, Boston, MA (Supplemental Table S2). Complete sample and clinical annotation can be found in Supplemental Table S1 and S2. Frozen tissue sections were prepared and manually macrodissected to obtain an enrichment of greater than 80% tumor cellularity. Genomic DNA from tissue and cell lines was extracted using DNeasy Tissue Kit (Qiagen, Valencia CA). All tumor sample DNA from the Vienna series were subjected to whole genome amplification (WGA) using the REPLI-g Kit (Qiagen) to obtain enough material for aCGH hybridization, while none of the Memorial Sloan Kettering Cancer Center samples and Brigham and Women's Hospital samples was subjected to WGA.

Array CGH profiling on oligonucleotide microarrays

Genomic DNA was fragmented and random-prime labeled as described previously [11] and hybridized to oligonucleotide arrays containing 22,500 elements designed for expression profiling (Human 1A V2, Agilent Technologies). All data is MIAME compliant, and the raw data has been deposited in to GEO under super-series accession #GSE7606. Using NCBI Build 35, 16,097 unique map positions were defined with a median interval between mapped elements of 54.8 Kb. Fluorescence ratios of scanned images were calculated as the average of two paired arrays (dye swap), and the raw profiles were processed to identify statistically significant transitions in copy number using Circular Binary Segmentation [11], [12]. Each segment was assigned a value that is the median of the log2 ratios of the spanned probes. The data were centered by the tallest mode in the distribution of the segment values. After mode-centering, we defined gains and losses as log2 ratios ≥0.15 or ≤−0.15 (±6 SD of the middle 75% quantile of data) and amplification and deletion as a ratio ≥0.4 or ≤−0.4 (representing 4 and 94% quantiles), respectively.

High-priority MCRs (see [11]) were chosen by requiring at least two samples to show a CNA event and at least one sample to show an extreme CNA event, defined by thresholds +1 and −1, and size of the MCRs was less than 2 MB. The MCRs were mapped to known regions of germline copy number variation (CNV), and CNV status was noted in Tables 1 and ​and2.2. Since well-validated cancer relevant genes have been known to harbor germline CNVs [25] we did not exclude candidates that are resident within these regions of known CNV.

gNMF and Fisher's Exact Test

Genomic NMF was applied to the current dataset as previously described [16]. Briefly, the segmented dataset was first dimension-reduced by eliminating redundant probe locations and then transformed to non-negative values. The resultant dataset was a non-negative matrix, which was subject to gNMF using a custom software package [17] and run in MATLAB (The MathWorks, Inc., Natick, MA). For each factor level two through six, gNMF was repeated 100 times to build a consensus matrix, and this was used to assign samples to clusters based on the most common consensus. The rank K=3 clustering was further tested for significance by permuting sample labels for secondary samples independently for each chromosome. One hundred permutations were subjected to Rank 3 NMF and the consensus matrix was assessed by cophenetic correlation.

Fisher's Exact Test was used to identify significantly different regional gains or losses between primary and metastastic melanoma. For each aCGH R-segment, each sample was classified as being copy number normal, gained or lost based on log₂ ratio thresholds of +/−0.15. Two-by-two contingency tables tested gained vs. normal and lost vs. normal between primary and secondary melanoma. Fisher's Exact Test p-values were corrected for multiple testing (q-value FDR 10%, “qvalue” package for R, http://cran.r-project.org).

Survivin immunohistochemistry and tissue microarrays

The melanocytic tumor progression TMA was as described previously [5]. TMA blocks were sectioned at ~4µm and antigen was unmasked in retrieval buffer (0.01M citrate buffer, pH6.0) using a pressure cooker at 125°C. Tissue sections were incubated with a 1/500 dilution of primary anti-Survivin polyclonoal antibody NB500-201 (Novus Biologicals, Littleton, CO) for 2 hours at room temperature followed by StreptAvidin- Biotin labeling. Signal was visualized using Alkaline Phosphatase with Permanent Red substrate (DAKO, Carpinteria, CA). L.M.D. and R.M.N. scored each core by visual microscopic inspection as follows: 0+ for no staining and no background; 1+ for weak blush of cytoplasmic staining; 2+ for moderately intense granular cytoplasmic staining; 3+ for markedly intense granular cytoplasmic staining. Most of the cores showed expression in more than 75% of the tumor cells; therefore the staining was graded on intensity rather than % of positive tumor. Statistical comparisons of Survivin IHC staining were performed using a Chi Square test corrected for multiple testing.

We thank Dr. R.A. DePinho for critical reading of the manuscript and members of the laboratories for helpful discussion. We thank Dr. David Hill of Human ORFeome at DFCI for the ASPM expression clone.