Biallelic Deletions in INK4 in Cutaneous Melanoma Are Common and Associated with Decreased Survival

Purpose: Both the retinoblastoma and p53 pathways are often genetically altered in human cancers and their complex regulation is in part mediated by the three gene products p16, p14^ARF, and p15 of the INK4 locus on chromosome 9p21. Partial or complete biallelic deletions of the INK4 locus have been recognized in a variety of malignant tumors, including malignant melanoma. We have in the present study measured the frequency of INK4 deletions in a large number of melanoma metastases and determined their association with clinicopathologic variables and survival data.

Experimental Design: Quantitative real-time PCR, as well as fluorescence-based fragment analysis, has been used to perform measurements of the relative allelic concentrations of the INK4 genes in 112 human melanoma tumor samples from 86 patients.

Results: Thirty-eight of 86 melanoma patients (44%) had metastases with biallelic losses in INK4. Ten of 20 patients with multiple metastases showed similar deletion patterns in all analyzed tumors. There was no significant association between any of the clinicopathologic variables and loss of INK4. However, loss of INK4 had an adverse effect on median survival from time of diagnosis. Patients with tumors with diploid INK4 had a median survival of 142 months, whereas those with monoallelic or biallelic loss in INK4 had a median survival of only 47 months (P = 0.006).

Conclusions: Our results point to homozygous deletions in the INK4 region as being one of the most common genetic alterations in malignant cutaneous melanoma. INK4 deletions are associated with an adverse prognosis.

The chromosome 9p21 region harbors the INK4 gene cluster consisting of the CDKN2A and CDKN2B genes. Due to the presence of two alternative exons 1, exon 1α and exon 1β, the CDKN2A gene encodes two different proteins. Using exon 1α, CDKN2A encodes the p16 protein that inhibits cdk4/cyclin D–mediated phosphorylation of the retinoblastoma protein and thereby plays a central role in regulation of the G₁ checkpoint of the cell cycle. In addition, CDKN2A exon 1β, located ∼12 kb centromeric to exon 1α, can be spliced onto CDKN2A exon 2 and translated in an alternative reading frame resulting in the nonhomologous protein p14^ARF that inhibits mdm2-mediated degradation of p53 and thus may enhance p53-related functions. The CDKN2B gene, located 21 kb centromeric to CDKN2A, encodes the p15 protein which is a second cdk4/cyclin D inhibitor. All three known gene products of the INK4 locus are thus of importance for regulation of two major pathways commonly involved in cancer development: the retinoblastoma and p53 pathways (1). Biallelic deletions of the whole or part of this gene region have been reported in cells from a variety of human malignancies, including cutaneous malignant melanoma. Studies on cutaneous melanomas using Southern blot techniques revealed an 11% to 25% biallelic deletion rate in the CDKN2A region (2, 3) and a somewhat higher deletion rate has been observed in melanoma cell lines (4). Up to 35% of sporadic human cutaneous melanomas contain NRAS codon 61 mutations (5–7) and an additional 38% to 66% have been reported to harbor mutations in BRAF (mainly in codon 600, previously denoted as 599; refs. 8–13). Thus, as many as 90% of cutaneous melanomas may have mutations activating the Ras/Raf/mitogen-activated protein kinase kinase/extracellular signal transduction pathway (12). In correspondence to a mouse melanoma model, activation of this pathway may thus cooperate with CDKN2A deficiency in the development of human cutaneous melanoma (14).

Loss of p16 expression has been associated with a shorter overall survival in different tumor types (15–18). Measurement of allelic concentrations of the INK4 sequences and mutational analysis of NRAS and BRAF in genomic DNA from tumors of melanoma patients may thus be a way to provide a molecular genetic profile characteristic of cutaneous malignant melanoma tumors and may also be of possible prognostic value and of use in the choice of treatment strategy.

We used quantitative real-time PCR to measure the relative allelic concentrations of the three INK4 genes in frozen or formalin-fixed melanoma tumor samples. This approach has enabled us to do an extensive screening of INK4 deletions in a large number of melanoma metastases. Allelic losses were confirmed independently by fluorescence-based fragment analysis. We now report on analyses done on 112 melanoma tumor samples from 86 patients. The results of analyses of INK4 gene dosage were compared with previously obtained data on NRAS and BRAF mutations as well as to clinicopathologic information and data on patient survival.

Patients and tumor biopsies. A total of 112 metastatic melanoma tumor biopsies from 86 patients who underwent surgery at the Karolinska University Hospital Solna were included in this study. The investigation was approved by the Ethics Committee of the Karolinska Institute. Tumor biopsies were obtained from 49 (57%) male and 37 (43%) female patients with a median age of 58 years (range, 21-92 years). At the time of the tumor biopsy (or the first biopsy in patients where multiple samples were obtained) 63 of the patients (73%) were in clinical stage III and 23 in stage IV (27%). The sites of metastases in this study included lymph nodes (74 tumors: 60 regional and 14 distant), skin (34 tumors: 15 regional and 19 distant), breast (3 tumors), and brain (1 tumor). From 20 patients, multiple metastases were analyzed: two metastases in 15 cases, three metastases in four cases, and four metastases in one case. Clinicopathologic data on the corresponding primary melanomas, including site, histopathologic tumor type, level of tumor invasion according to Clark, tumor thickness according to Breslow, and ulceration were obtained. Overall survival data from time of diagnosis of the primary melanoma were also obtained for all patients except one.

Cells and extraction of DNA. Human genomic DNA CEPH 1347-02 (PE Applied Biosystems, Foster City, CA) was used as wild-type reference, assuming a diploid representation of the gene sequences under investigation. Genomic reference DNA was also prepared from the human melanoma cell lines 224, which is diploid for INK4, and 397, which has a homozygous INK4 deletion. Thin slices from fresh frozen or formalin-fixed paraffin embedded melanoma metastases were used for genomic DNA extraction using a QIAamp DNA-Mini-kit according to the manufacturer's instructions (Qiagen GmbH, Hilden, Germany). In some cases, genomic DNA was also extracted from homogeneous tumor cell populations, manually dissected from 20-μm sections of formalin-fixed biopsies of cutaneous malignant melanoma metastases, as previously described (19). In a few cases, laser capture microdissection was done using a PixCell II LCM system (Arcturus Engineering, Mountain View, CA) and DNA was extracted from the captured tumor cell populations using a buffer containing 10 mmol/L Tris (pH 8.0), 1 mmol/L EDTA, 1% Tween 20, and 1 mg/mL proteinase K (Boehringer, Mannheim, Germany) at 56°C overnight. The proteinase K was then inactivated at 95°C for 10 minutes. Aliquots of the extracts were then used directly for PCR analysis.

Conventional PCR amplification. End-point PCR was done on a Perkin-Elmer 9600 thermal cycler (PE Applied Biosystems). Primer pairs chosen from intron 1 and exon 1β of CDKN2A, CDKN2B exon 1, NRAS exon 2, and aldolase B exon 3 sequences were used for detection of the corresponding allelic regions (Table 1). PCR was done on 25 ng of genomic DNA in a final reaction volume of 10 μL containing 1 μL 10× PE standard buffer (Perkin-Elmer/Roche, Branchburg, NJ). PCR reactions were carried out in the presence of 5% DMSO for 30 cycles including denaturation at 94°C for 30 seconds, annealing at 58°C to 60°C for 30 seconds, and elongation at 72°C for 30 seconds. Genomic control DNA CEPH 1347-02 and genomic DNA from melanoma cell lines 224 and 397 were used as templates in end-point PCR reactions using the PCR primer pairs mentioned above (Table 1). All five gene regions were successfully amplified from the human control DNA and from cell line 224 DNA, whereas amplification reactions using melanoma cell line 397 DNA (with a homozygous INK4 deletion) as template resulted in products only for the aldolase B exon 3 and NRAS exon 2 amplifications, as expected (data not shown).

Real-time PCR amplification. The same PCR primers were used for real-time PCR analyses on an ABI Prism 7700 sequence detector (PE Applied Biosystems) in combination with five specific R-Q probes directed against unique CDKN2A, intron 1 and exon 1β, CDKN2B, NRAS exon 2, and aldolase B exon 3 sequences (Table 1). The reporter dye was 6-carboxyfluorescein attached to the 5′ terminal nucleotide and the quencher dye was an internally incorporated dabcyl-deoxythymidine in all cases. The PCR reaction volumes were 50 μL containing 100 ng genomic DNA, 5 μL standard PE Taqman buffer (Perkin-Elmer/Roche), 2.0 mmol/L Mg²⁺. PCR conditions were 50°C for 2 minutes followed by 95°C for 10 minutes followed by 40 cycles consisting of 15 seconds at 95°C and 60 seconds at 60°C. Each sample was analyzed at least in duplicate. All PCR primers and R-Q probes were synthesized by Scandinavian Gene Synthesis AB (Köping, Sweden). Fluorescence signals were monitored in the ABI Prism 7700 Sequencer and the fluorescence signals were plotted against cycle number. Rn, the normalized reporter signal is the fluorescence signal divided by the signal of the passive internal reference dye rox. The baseline fluorescence signal was read from the first 15 cycles and a threshold level was set above the mean baseline level. The cycle number at which the fluorescence signal exceeded the threshold level defined the actual cT value. Standard curves using dilution series of genomic control DNA were constructed for all five primer probe combinations. Human genomic control DNA CEPH 1347-02 (PE Applied Biosystems) was used to obtain the calibration standard curves for the real-time PCR amplification reactions. Amplification plots together with the standard curves for CDKN2A intron 1, CDKN2A exon 1β, CDKN2B exon 1, NRAS exon 2 or aldolase B exon 3, were obtained from dilution series of the normal control DNA CEPH 1347-02 (2-fold dilution steps ranging 100-0.78 ng of template DNA, data not shown). The melanoma cell lines 224 and 397 were used as known reference samples. The amplification plots from the corresponding real-time PCR analyses of human control DNA and from the melanoma cell line 224, using the five R-Q probes, resulted in efficient amplification of all five gene regions, whereas the corresponding amplification reactions on DNA from cell line 397 yielded exclusively products of the aldolase B and NRAS sequences, as expected (data not shown). The percentage of INK4 target sequences present in tumor cell populations were expressed as the δδcT values, which were calculated by subtraction of the δcT values obtained from aldolase B exon 3 or NRAS exon 2 and each of the three INK4 target sequences in tumor DNA from the corresponding δcT values obtained from control DNA according to the formula in User Bullentin #2: ABI PRISM 7700 Sequence detection system (20). The δδcT values obtained from human control DNA were taken as representing the diploid 100% level. δδcT shifts of 1 unit were assumed to represent a 50% change in target sequence concentration in the actual tumor cell population.

Fluorescence-based quantitative fragment analysis. The PCR primers had the same sequence as those used in real-time PCR, but the 5′ primers were fluorescently labeled with either 6-carboxyfluorescein (aldolase B) or VIC (CDKN2A intron 1, CDKN2A exon 1β, and CDKN2B exon 1). The PCR was carried out in a 15-μL reaction containing genomic template DNA, fluorescently labeled 5′ primer, unlabeled 3′ primer, 1× PE standard buffer, 2.5 mmol/L MgCl₂, 200 μmol/L deoxynucleotide triphosphate, and 0.75 unit Taq gold. PCR reactions were carried out in the presence of 5% DMSO under the following amplification conditions: denaturation at 95°C for 10 minutes followed by 26 to 28 cycles denaturation at 94°C for 30 seconds, annealing at 55°C for 30 seconds, extension at 72°C for 30 seconds followed by a 30-minute final extension at 72°C. The amplified PCR products were analyzed using the ABI Prism 310 genetic analyzer (PE Applied Biosystems). The CDKN2A (intron 1), CDKN2A (exon 1β), and CDKN2B (exon 1) fluorescence signals, obtained from control DNA CEPH 1347-02, were divided by the corresponding aldolase B (exon 3) signal. The similarly obtained ratio between the actual INK4 target sequence and aldolase B in tumor samples was divided by the corresponding control DNA ratio to obtain information on relative target concentration.

Statistical analyses. To evaluate the possible relationship between INK4 deletions and various clinicopathologic variables, either one-way ANOVA or the χ² test were used. Patients in whom multiple metastases were investigated were assigned to the group corresponding to the tumor with the lowest INK4 target level in these analyses. The effect on overall survival was determined with Wilcoxon-Gehan's statistics. P ≤ 0.05 was regarded as statistically significant.

DNA samples extracted from 112 metastatic melanoma lesions in 86 patients were analyzed by quantitative real-time PCR to measure the relative allelic concentrations of the INK4 gene regions, encoding the p16, p14^ARF, and p15 proteins, respectively. Percentage figures for the relative allelic concentrations of the targeted INK4 sequences obtained from the tumor samples are summarized in Table 2. The samples are divided in three groups, according to the lowest remaining level of the three analyzed targets. Group I (diploid tumors) comprise samples with remaining INK4 target corresponding to 75% to 100% of diploid levels; group II (tumors with monoallelic loss) consist of samples with remaining target ranging between 26% and 74% of diploid levels; and group III (tumors with biallelic loss) of samples with remaining target between 0% and 25% of diploid levels. In cases where multiple metastases were analyzed from the same individual, the patients are assigned to the group corresponding to the tumor with the lowest INK4 target level.

Table 3 summarizes the INK4 genotypes in tumors with respect to biallelic losses. Forty-eight of the 56 tumors belonging to the 38 patients in group III showed biallelic loss of the INK4 region, which was partial in 28 tumors (58%) and complete in 20 tumors (42%), respectively. Forty-two of these tumors showed biallelic loss of CDKN2A intron 1 (88%), 32 of CDKN2A exon 1β (67%), and 24 of CDKN2B exon 1 (50%).

Fluorescence-based fragment analysis was carried out on 36 of the 112 metastases for the purpose of independent confirmation of the results of quantitative real-time PCR. In 10 of the cases, quantification of all three sequences, CDKN2A intron 1, CDKN2A exon 1β, and CDKN2B exon 1, was done, and in the remaining 26 samples one or two of three target sequences were analyzed. Figure 1 exemplifies results obtained from three metastases analyzed with both methods. In total, 78 analyses were done (24 in group I, 25 in group II, and 29 in group III) and an overall concordance of 78% with respect to group assignment of tumors was found between the two independent methods (group I, 83%; group II, 68%; and group III, 83%). Moreover, among the discordant samples a further 9% yielded results that were 5% above or below the chosen cutoff levels, corresponding to an overall similarity of estimates in 87% of tumors. Thus, the results of real-time PCR were generally confirmed by fluorescence-based fragment analysis.

To compare the INK4 deletion pattern in multiple metastases derived from the same primary tumors, analyses were done on DNA extracted from between two and four separate metastases from each of 20 patients (Table 4). NRAS and BRAF mutation data from previously done analyses of the same metastases (7, 12)¹

show identical patterns among multiple metastases in each patient, illustrating their clonal origin (data not shown). In seven patients (shown in bold face in Table 4), the deletion patterns for all three INK4 target sequences were identical. In another three patients (patients 7, 9, and 10), the numerical results of analyses differed not more than 20% for all three sequences. However, because some of the results were clustered around the cutoff values, the tumors were assigned to different groups for one or more of the INK4 target sequences. Thus, deletion patterns for all three INK4 target sequences were very similar for multiple metastases in 10 of the 20 patients, supporting the hypothesis that the INK4 gene deletions not infrequently occur already in the primary melanomas. In 6 of 10 remaining cases, either CDKN2A intron 1 or exon 1β sequences, or both, exhibited similar target levels in multiple metastases, indicating that alterations of p16 and p14^ARF are frequently early events which occur in the primary tumors. However, we also compared INK4 deletion status in regional metastases in stage III patients and distant metastases in stage IV patients. Diploid INK4 levels were more common in regional metastases (11 of 75, 15%) compared with distant metastases (3 of 37, 8%), indicating that some of the deletions occur later in the metastatic process when patients develop disseminated stage IV disease.

None of the clinicopathologic variables analyzed showed a significant difference between patients with tumors that had an intact INK4 locus and those with tumors with monoallelic or biallelic loss of INK4 locus (data not shown). However, a tendency of a lower frequency of nodular melanoma relative to superficial spreading melanoma was observed in group I (1 of 9) compared with groups II (13 of 14) and III (11 of 14).

We did univariate and multivariate analyses of overall survival from time of diagnosis of primary melanoma tumors in relation to patient characteristics, such as gender, tumor site, histopathologic type, Clark level of invasion, ulceration, tumor thickness, age, and INK4 status. As expected, the clinicopathologic variables that showed a significant effect on overall survival in the univariate analysis were the previously well documented prognostic factors: tumor thickness, Clark level of invasion, ulceration, age, and histopathologic type. In addition, INK4 status had a significant effect on overall survival (Fig. 2). Thus, patients with diploid INK4 levels (group I) had a median overall survival of 142 months, compared with 47 months for patients with monoallelic or biallelic INK4 loss (groups II-III; P = 0.006). Patients with tumors with partial biallelic loss of INK4 showed similar overall survival as those with tumors with complete biallelic loss of the INK4 gene. Tumor thickness was the only factor that showed a significant effect on overall survival in the multivariate analysis.

The present report describes the use of a 5′ nuclease real-time PCR assay using three INK4 gene–specific R-Q probes to estimate the relative gene dosage of CDKN2A intron 1 (encoding p16), CDKN2A exon 1β (encoding p14^ARF), and CDKN2B exon1 (encoding p15) in defined tumor cell populations, as well as an independent confirmation of these results by fluorescence-based fragment analysis. The investigation represents an application of these techniques in a large-scale screening of human melanoma metastases.

The quantification of data on genetic losses in histologically heterogeneous tumor biopsies is generally complicated by the fact that they usually consist of a mixture of tumor and contaminating normal cells. Estimates of allelic loss in such tumor material will thus frequently be attenuated by the presence of varying amounts of diploid nonmalignant cells. To allow for such attenuation, our results were accordingly evaluated after subdivision into three groups representing different ranges of remaining INK4 target levels. Group I includes 14 of 112 tumors (12%), showing remaining target levels of 75% to 100% of the diploid amount, which is considered to indicate an intact diploid genotype for the targeted sequences. Group II includes 50 of 112 samples (45%) with remaining target levels of 26% to 74% and represents tumors with an indication of monoallelic losses, whereas group III includes 48 of 112 (43%) samples with remaining target levels of 0% to 25% indicating biallelic losses of INK4 sequences. Thus, our results imply that, as a minimum estimate, as many as 38 of 86 (43%) of the patients had at least one metastasis with biallelic losses in INK4 (Table 2). In tumors with monoallelic loss, inactivation of the remaining allele may occur by other mechanisms, such as point mutations or hypermethylation.

The conclusion that allelic loss of INK4 is frequent and of biological significance is further supported by recent findings in a study of gene expression profiling of melanoma metastases.²

We have investigated gene expression using Affymetrix HG-U133A chips in melanoma metastases characterized for INK4 allelic loss, which partially overlap with the tumor material in the present report. CDKN2A was among the top genes that were significantly differentially expressed in tumors with different INK4 allelic status. Thus, tumors with monoallelic INK4 loss had a relative CDKN2A mRNA expression of 43% to 62% (depending on probe set) and those with biallelic loss of 20% to 23%, compared with diploid metastases. This shows that the genomic INK4 alterations are reflected in differences in gene expression levels.

The relative importance of loss of p16 and p14^ARF for tumor development is still uncertain. In our melanoma tumors, the most common biallelic loss was deletion of CDKN2A intron 1 but three (6%) of the tumors showed isolated loss of the exon 1β sequence, indicating that specific inactivation of p14^ARF may contribute to melanoma development in a subset of cases (Table 3). The selective loss of exon 1β has also previously been reported in melanoma cell lines (21). Recent findings from murine melanoma models in which INK4a/ARF deficiencies accelerated induction of melanoma tumors are of interest in this connection (22). p14^ARF deficiency may lead to deregulated proliferation via the p53 pathway (23) and also via up-regulation of E2F-dependent transcription (24).

Previous reports indicate that p16 loss correlates with advanced stages of melanoma (25, 26), which may contribute to the high frequency of monoallelic and biallelic losses of INK4 observed in our material of metastatic samples. However, remarkably similar deletion patterns for all three INK4 target sequences were seen in separate tumors from 10 of 20 patients in whom multiple metastases were analyzed. In another six patients, either CDKN2A intron 1 or exon 1β sequences, or both, exhibited similar target levels in multiple metastases. These patient-specific deletion patterns point to an origin of the deletion events before metastasis and thus to a clonal relationship between the metastases in these cases (Table 4). The remaining four patients with multiple metastases showed differences in INK4 deletion patterns for all three INK4 target sequences among their tumors, which probably indicates that INK4 deletions may also occur later during the metastatic process, or that multiple metastases may derive from different cell clones within a primary tumor. An alternative possibility is that discrepant results may be a result of contaminating DNA from normal cells present in the extracts, because DNA extracts from undissected tissue sections were used for the analyses. The same BRAF and NRAS mutation patterns were detected in samples from the cases with multiple metastases which confirm the clonal relationship of the separate metastases (data not shown). A further indication that INK4 deletions sometimes occur at a late stage during tumor progression is our finding of a lower proportion of tumors with diploid INK4 levels among distant metastases in stage IV patients compared with regionally metastatic tumors in stage III patients.

Both the results from previously done mutational analyses of NRAS and BRAF, which also included analysis of the corresponding primary tumors, and the presently presented INK4 deletion patterns thus point to an early occurrence of altered signal transduction in combination with cell cycle dysregulation as possible driving forces of clonal expansion in human cutaneous melanoma. A comparison with mutational analyses revealed that 35 of 75 (47%) patients with melanoma metastases with INK4 deletions also had BRAF codon 600 mutations, whereas 23 of 75 (31%) of the patients had metastases with NRAS codon 61 mutations and 2 of 75 (3%) had both BRAF and NRAS mutations, (7, 12).² The frequency of BRAF mutations was numerically higher in group III (51%) than group I (27%), whereas the frequency of NRAS mutations showed an opposite trend, being less frequent in group III (23%) compared with group I (55%). Although these differences were not statistically significant (P = 0.249), they may indicate that BRAF mutated tumors to a higher extent need INK4 loss for tumor progression than those with NRAS mutation. However, recently an inverse association between NRAS/BRAF mutations and the frequency of loss of heterozygosity on 9p21 at the marker D9S942 located within the CDKN2A gene has been reported in melanomas, suggesting that NRAS/BRAF mutations may to some extent compensate for the requirement of allelic loss on 9p21(27).

Despite the fact that all patients in our study had metastatic disease and therefore generally a poor prognosis, we found that patients with tumors with INK4 loss had a significantly shorter median survival compared with those with an intact INK4 locus. Our findings thus indicate that INK4 is a prognostic marker, which is in agreement with previous reports obtained at the protein level. Straume et al. thus reported that p16 expression analyzed by immunohistochemistry in vertical growth phase melanomas was an independent favorable prognostic factor in multivariate analysis (15). Furthermore, tissue microarrays consisting of a panel of 39 different antibodies was used to analyze melanoma biopsies corresponding to different histologic progression phases. A predictor model consisting of a combination of four markers (Ki67, p16, p21^CIP1, and bcl-6) was associated with shorter survival in patients with vertical growth phase melanoma. This predictor model was validated using an independent series of vertical growth phase melanomas (28).

In summary, deletions in the INK4 region represent, thus far, the most common genetic alterations detected in malignant cutaneous melanoma, beside activation of Ras/Raf/mitogen-activated protein kinase kinase/extracellular signal regulated kinase transduction, and they are significantly associated with a decreased overall survival.

We thank Sune Kvist and Scandinavian Gene Synthesis AB (Köping, Sweden) for design and generous supply of primers and R-Q probes and Bo Nilsson for statistical analyses.