Differentiation of Malignant Melanoma From Benign Nevus Using a Novel Genomic Microarray With Low Specimen Requirements

This study was approved by the University of Utah (Salt Lake City) Institutional Review Board, and institutional guidelines regarding the use of archived specimens and patient chart review were followed. Cases of benign nevus, malignant melanoma, and MLUMP were identified by natural language search of digitized diagnostic reports from 1991 to 2009 at the University of Utah Department of Dermatology. All diagnoses were rendered by board-certified dermatopathologists. Terms used to identify MLUMPs were “atypical AND Spitz” as well as “favor AND melanoma.” Malignant melanomas were identified using the term “melanoma AND Clark IV,” targeting large lesions with a high likelihood of malignant behavior. The hematoxylin-eosin stain slides of candidate cases were reviewed (by W.M.C.) and cases were excluded if there was a dense lymphocytic infiltrate, if the specimen was judged to be too small (qualitative judgment), or if the percentage of melanocytic nuclei was judged to be below 40%. After selection, digital microscopic images of the samples were analyzed. The smallest specimen had a surface area of 6.2 mm² and the mean surface area of all specimens was approximately 24 mm²—equivalent to a 6 × 2-mm shave biopsy bisected. Sixty-four melanocytic lesions were included in this study: 23 benign nevi, 27 primary malignant melanomas, 3 metastatic melanomas (all from different body sites on the same patient) and 11 MLUMPs (Table). Clinical features of patients including outcome data were not used to select cases. In the later stages of the study, samples from benign nevi with less than 250 ng of DNA (in 45 µL of solution) were excluded, as were samples of malignant melanoma and MLUMP with less than 150 ng of DNA, although some samples in the earlier stages had only 50 ng of DNA and still produced robust results.

After cases were identified, the University of Utah's electronic medical record was searched to identify the surgical treatments received and clinical outcomes related to the melanocytic lesions. When electronic records were unavailable or the last follow-up appointment was remote, paper records from the University of Utah Department of Dermatology were searched.

Eight sections, each 10 µm thick, were taken of the archived FFPE tissue blocks used for clinical diagnosis. There was no microdissection of the FFPE specimens. DNA was extracted and purified using the Ambion RecoverAll Total Nucleic Acid Isolation kit (Applied Biosystems, Carlsbad, California) following the manufacturer's recommended protocol, with the exception of extending the protease digestion time up to 90 hours for larger specimens. Briefly, the procedure entails deparaffinization with xylene, protease digestion, ethanol, and filter cartridge–based DNA isolation followed by an on-filter RNase treatment and elution. Excluding the time required for protease digestion, the time to complete the remaining steps of DNA isolation is 1 hour and 15 minutes.

Extracted DNA was quantified using Quant-iT PicoGreen ds DNA HS reagent and Qubit fluorometer (Invitrogen, Carlsbad, California) following the manufacturer's package insert procedure. Briefly, 1 µL of sample was mixed with 199 µL of Quant-iT working solution, vortexed for 3 seconds, incubated at room temperature for 2 minutes, and assayed on the Qubit fluorometer.

Genomic microarray hybridization was performed by Affymetrix Research Services Laboratory through the OncoScan FFPE Express Service (Affymetrix, Santa Clara, California). Technical documentation is available on the Affymetrix website.10 Briefly, this microarray uses 330 000 molecular inversion probes targeting SNPs spanning the genome to identify changes in copy number and loss of heterozygosity. The OncoScan FFPE Express assay has been designed to work with DNA from FFPE tissue and involves 2 rounds of PCR amplification. The design and performance of the molecular inversion probe GMA has been described extensively by other authors.11–13 Formalin-fixed, paraffin-embedded particle clots from normal bone marrow aspirates without disease (individuals unrelated to this study cohort) were used to generate control probe signal values.

Data generated by Affymetrix (probe signal intensity and genome location) were analyzed at the University of Utah using Nexus Copy Number v5.1 (BioDiscovery, El Segunda, California). Initial processing of the data was done with the SNP-FASST2 segmentation algorithm, a hidden Markov model–based approach, in conjunction with quadratic wave correction. A minimum of 3 probes were required to create a segment. Study authors (W.M.C., L.R.R., M.S.J., J.D.S., and S.T.S.) familiar with the literature concerning genomic changes in malignant melanoma and the performance of the microarray reviewed the genome-view dot plots blinded to histologic diagnosis to render a GMA-based diagnosis of malignant melanoma, benign melanocytic lesion, or uncertain. In cases where there was not unanimity, a majority diagnosis was used.

Of 23 benign nevi, 1 had an abnormal SNP-GMA pattern, with an isolated loss of genetic material on the proximal long arm of chromosome region 12q13-q14 of 21 Mb, a pattern not reported in malignant melanomas. There were no benign nevi (Figure 1) with an SNP-GMA pattern consistent with malignant melanoma (100% specificity). All of the patients with benign nevi had an uneventful clinical course.

Of the 27 primary malignant melanoma samples analyzed, 24 showed gains or losses characteristic of malignant melanoma (89% sensitivity; Figure 2) and 1 showed changes that were indeterminate by consensus (Figures 3 and 4). Chromosomal gains characteristic of malignant melanoma frequently involved segments of the following chromosomal arms: 6p (67%), 7q (48%), 7p (33%), 1q (15%), and 8q (15%). Chromosomal losses characteristic of malignant melanoma frequently involved segments of the following chromosomal arms: 9p (59%), 9q (56%), 10q (37%), 10p (33%), 6q (26%), 5q (26%), 14q (22%), 3q (19%), 11q (19%), 16q (19%), 1p (15%), and 11p (15%). Two melanoma samples displayed copy number–neutral loss of heterozygosity, involving 9p in one sample (Figure 5) and 10p and 10q in the other. Ten of the 27 patients with malignant melanoma had a more aggressive melanoma characterized by the development of metastatic disease in the form of a positive lymph node (including sentinel lymph node [SLN]) or in-transit or distant metastases. Nine of 10 (90%) of these more aggressive melanomas were identified by SNP-GMA.

Of the 3 metastatic melanoma samples from the same patient, 2 samples showed a similar pattern of gains and losses characteristic of malignant melanoma. The third remaining metastatic melanoma sample showed a pattern of low-amplitude gains and losses, which blinded consensus did not initially designate as melanoma; however, when viewed in light of the patient's other metastatic samples, the faint pattern of gains and losses was similar and was sufficient to designate this third sample as metastatic melanoma arising from the same clone.

Of the 11 MLUMPs analyzed, 4 showed gains and losses consistent with malignant melanoma. None of these 4 patients had any clinical evidence of malignancy during a mean follow-up period of 3.5 years (range, 0–16.4 years). One MLUMP displayed malignant behavior in the form of a positive SLN. This clinically malignant MLUMP had an SNP-GMA pattern consistent with a benign nevus (Figure 6, A through C). This patient was lost to follow-up immediately after the positive SLN diagnosis.

This SNP-based molecular inversion probe GMA is a sensitive and highly specific method for detecting melanoma (89% sensitivity, 100% specificity) in this cohort of cases with a definitive histologic diagnosis. This SNP-GMA platform's performance in predicting the malignant potential of cases with an ambiguous histologic diagnosis was less impressive, identifying the 1 MLUMP with malignant clinical behavior (a positive SLN) as a benign nevus and 4 MLUMPs with no further evidence of disease during follow-up as having a genomic pattern similar to that observed in malignant melanoma.

A limitation of this study is the relatively short clinical follow-up period (median, 3.1 years) for the 4 MLUMPs with an SNP-GMA pattern consistent with melanoma. It is possible that some of these lesions will eventually display clinically malignant behavior. It is also possible that these types of lesions have a low malignant potential that is effectively treated by narrow excision, or that these lesions have no malignant potential but still possess genomic changes that resemble melanoma as does their histology.

In addition to demonstrating this platform's ability to differentiate malignant melanoma from benign nevi, this study has accomplished a significant technologic achievement by forgoing microdissection of the paraffin block before microarray analysis of selected clinical biopsy specimens. Figures 7 and 8 show that 3 copies of chromosome 2, present in approximately 20% of the amplified sample, can be clearly differentiated from the adjacent chromosome 1q with a normal copy number of 2.14 This finding implies that a chromosomal gain or loss present in as few as 20% of the biopsy cells can be detected against a normal background. Rigorous dilution studies should be performed for confirmation. If these findings are upheld, microdissection would be unnecessary for many compound melanocytic lesions, but would still be required for most junctional melanocytic proliferations or lesions with a dense inflammatory infiltrate. Removing the nonautomatable step of microdissection streamlines the process of SNP-GMA, making it simpler and even more practical for research and diagnostic purposes.

A second major technical accomplishment of this study is the demonstration that a very small quantity of DNA can provide high-quality SNP-GMA results for melanocytic lesions. Using the OncoScan FFPE Express Array, we were able to detect clear gains and losses consistent with the diagnosis of melanoma from a shave biopsy from which we extracted 50 ng of DNA (Figure 9). The current recommended specimen requirement for the Agilent 180K microarray testing for melanoma at University of California San Francisco is 500 ng. Therefore, with this novel Affymetrix platform, SNP-GMA testing could become an ancillary diagnostic option for even smaller FFPE biopsy specimens.

Recently, fluorescence in situ hybridization (FISH) has been developed as an ancillary molecular test for malignant melanoma. The benefit of FISH testing is that it requires less tissue for analysis (about 30 malignant cells)9 than standard genomic array technology and can be used effectively without microdissection in situations where a relatively small melanocytic population is surrounded by an abundance of normal cells.8 We expect FISH testing to remain an effective diagnostic tool for the differentiation of nodal nevus versus melanoma metastatic to a lymph node and for benign proliferative nodule versus melanoma developing in a large nevus. However, our study has effectively dropped the minimum specimen requirement by a factor of 10, from 500 to 50 ng of DNA, making microdissection coupled with SNP-GMA analysis a viable option for some of these lesions.

A major advantage of SNP-GMA compared with FISH is the breadth and quantifiable detail of the results. A commercially available FISH panel for the diagnosis of melanoma contains 4 probes directed against 2 chromosomes (6p25 RREB1, 6q23 MYB, 6centromere, 11q13 CCND1). This FISH panel was first described by Gerami et al9 in 2009 with an 86.7% sensitivity, and later with a 77% sensitivity after further testing and development by NeoGenomics.15 A flurry of recent publications has demonstrated excellent sensitivity of FISH for a number of melanoma subtypes, with desmoplastic melanoma being the exception.16–20 However, a small study recently suggested that the sensitivity of this FISH panel for the detection of melanocytic lesions that will have a malignant clinical behavior is much lower at 60%.21 Our results suggest that this FISH panel would have missed 6 melanomas in our cohort that SNP-GMA detected, one of which proved to be lethal to the patient (Figures 10 and 11). This assumed sensitivity is based upon the fact that we did not detect gains or losses in the regions targeted by the FISH probes or the copy number changes the probes were designed to detect (loss versus gain). It is possible that gains or losses could have been detected in a subpopulation if they were actually analyzed by FISH. These 6 melanomas detected by SNP-GMA that we expect would have been missed by the FISH panel all had a gain or loss that involved a portion of either chromosome 7, 8, or 9. The original work of Bastian et al6,7 described changes to these chromosomes as some of the most common in malignant melanoma.

The wealth and specificity of information available from an SNP-GMA will make this technology cost-effective as mutation-specific therapies for malignant melanoma become standard. Although the SNP-GMA used in this study was not optimized for melanoma, the modularity of the molecular inversion probe design means clinically relevant SNPs such as the BRAF V600E could potentially be included as one of the 330 000 SNPs detected. In fact, the most recent version of the OncoScan FFPE Express from Affymetrix (version 2.0, released May 2011) has been designed and optimized to detect the BRAF V600E and other somatic mutations. This implies a single SNP-GMA test could be used to confirm the diagnosis of malignant melanoma, stratify the risk of recurrence, and direct pharmacologic interventions.

Although the sensitivity and specificity of SNP-GMA for detecting malignant melanoma is good, our small series highlights one case where the failure of SNP-GMA is concerning. This case involved the only histologically ambiguous melanocytic lesion that had a malignant clinical course, in the form of a positive SLN (Figure 6). The patient had no additional clinical follow-up at our facility following a positive SLN biopsy, so additional outcome data are not available. The possibility that the melanocytes in the SLN represent a nodal nevus or benign metastasizing nevus is small, given the histologic findings, which included a mitotic figure in the parenchymal nodal metastasis and the patient's postpubertal age of 19 years. It is possible that the false-negative SNP-GMA result occurred because of a small malignant clone that metastasized early, but was below our threshold of detection in the primary lesion. It is also possible that this melanoma developed the ability to metastasize through point mutations or translocations that did not result in gains or losses of chromosomal material and were therefore undetectable by our SNP-GMA. A final possibility is a technical failure in the performance of the assay, as we were unable to perform repeat testing on this sample.

In summary, we have demonstrated that an SNP-GMA can detect gains and losses and loss of heterozygosity of chromosomal material characteristic of malignant melanoma in archived FFPE biopsy specimens with as little as 50 ng of DNA. The sensitivity of melanoma detection was good in the entire study sample (including both histologically malignant and ambiguous lesions with malignant clinical behavior) at 86% (n  =  24 of 28). The sensitivity for detecting melanomas with an aggressive clinical course (positive SLN and/or metastatic disease) was similar at 82% (n  =  9 of 11). Our results show that SNP-based GMA will have a role in the diagnosis and potentially in the risk stratification and treatment of malignant melanoma.

This study was supported by a $10 000 Resident Research Support Grant from the University of Utah Department of Pathology (Dr Chandler) and $33 300 in University of Utah ARUP Laboratories Research and Development Funds (Dr South).