Genomic Approaches to Skin Cancer Diagnosis FREE

Dermatopathologic analysis is the gold standard for the diagnosis of skin cancer. It is an accurate and cost-efficient method for classification of most cutaneous neoplasms. However, despite constantly refined morphological criteria, there remain areas in which histopathologic analysis does not permit an unequivocal diagnosis. Among these, lymphoproliferative and melanocytic lesions present the most frequent diagnostic problems. The predicaments of less-than-optimal agreement between observers and misdiagnosis of these disorders are well documented.^{1 - 7} Failure to recognize melanoma, as well as lymphoma, leads the list of malpractice claims among general pathologists, let alone dermatopathologists.⁸ The advent of therapeutic interventions for melanoma, some of which have serious adverse effects, makes the need for accurate, early diagnosis more important than ever.

Cancer is a genetic disease (ie, it is a disease of the genome), but not necessarily a hereditary one. Using genomic approaches for diagnosis therefore seems logical. Numerous attempts have been made to improve diagnosis by new methods such as DNA cytometry, in situ hybridization, mutation analysis, and others. For typing of primary extracutaneous lymphoproliferative disorders, genomic analyses have already entered clinical practice.

Unfortunately, little is known about the genetic events underlying cutaneous lymphomas. The same holds true for melanocytic tumors of the skin. Although several pathogenetically relevant genetic alterations have been detected in melanoma and nevi, our understanding is still too patchy to permit the development of diagnostic tests targeted to individual genes. The efforts of the Human Genome Project, with the human genomic sequence being almost complete, promise to dramatically speed up the identification of relevant skin cancer genes. However, an additional layer of complexity is added because there is not just a single genetic pathway for cancer; cancer cells can use several different routes to become malignant. Full malignant potential requires the acquisition of a series of characteristics such as immortality, resistance to apoptotic signals, invasive capacity, induction of angiogenesis, and others (see Hanahan and Weinberg⁹ for a review of these characteristics). The same features can be acquired by several different alterations of the respective pathways. Therefore, more comprehensive analyses encompassing many different genes may be required to obtain diagnostically useful information.

In this review several relatively novel techniques of molecular profiling that permit simultaneous analysis of thousands of markers will be introduced. Some of these focus on changes in the genome of the cancer cell, and others deal with changes in the expression levels of genes.

Genomic analysis is well established in clinical areas such as prenatal diagnosis, hereditary disorders, or diagnosis of certain cancers (eg, structural rearrangements of chromosomes and quantitative abnormalities of chromosomes are important diagnostic parameters for certain soft tissue tumors or hematopoietic malignancies).¹⁰ It is important to note that gross chromosomal alterations are a hallmark of solid tumors,¹¹ and only very few types of cancer do not show chromosomal alterations. In most cases, these chromosomal alterations are not simply adverse effects of cancer progression but represent an essential mechanism that contributes to the alterations required for cancer development.^{12 - 13} This raises the intriguing possibility that the presence or absence of chromosomal alterations could be used diagnostically.

Gross chromosomal aberrations can be studied by karyotyping. However, karyotyping requires viable tissue, and it is therefore of little use as a diagnostic tool for skin cancers, which typically are too small and already completely excised and formalin-fixed when diagnostic questions arise. Cytogenetic studies of melanoma have been performed mostly on melanoma metastases and cell lines and have shown a plethora of chromosomal aberrations with some consistent patterns.^{14 - 15} To circumvent the limitations of karyotyping, several techniques have been developed that permit the analysis of genomic DNA extracted from formalin-fixed, paraffin-embedded material.

Loss of heterozygosity or allelic imbalance can signal the presence of deletions or gains of specific alleles in paraffin-embedded material.¹⁶ For this approach, polymerase chain reaction (PCR) is used to amplify small genomic fragments that are likely present in 2 different variants (alleles) in an individual. Ideally these variants are of different size so that they can easily be detected on a gel. If one finds 2 alleles in the normal tissue of a patient, this marker is informative, and therefore one can assume imbalance if one detects only one allele in the tumor tissue. More detailed analysis can distinguish true losses.

Sites of recurrent losses are typically areas that harbor tumor suppressor genes. Loss of heterozygosity analysis for melanoma has shown recurrent losses of chromosomes 9p, 10q, 6q, and 18q, whereas nevi did not show frequent losses.¹⁷ The disadvantage of loss of heterozygosity analysis is that contamination of normal cells can make it difficult to decide whether an allele is lost. Another drawback is its inability to distinguish increases and decreases in DNA copy number. Also, one can only study 1 marker at a time, and some markers will not be informative because they will be homozygous at the locus.

Fluorescently labeled probes that are complementary to a stretch of genomic DNA are used in fluorescence in situ hybridization (FISH) analysis.¹⁸ The labeled probe and the target material (which can be metaphase spreads [Figure 1B], interphase nuclei [Figure 1C], or tissue sections [Figure 2]) are denatured and brought into contact for several hours to days. Given the right hybridization conditions, parts of the labeled probe will anneal with the corresponding sequence in the target.

The FISH analysis can be performed on sections of paraffin tissue. This is easiest if the probe is targeted to chromosomal regions rich in repetitive sequences such as the centromeres. In these regions the probe can hybridize multiple times and produce large, easy-to-detect hybridization signals. However, these regions typically do not contain any genes, and while increases in chromosome number are recognized, no direct information on the copy number of a specific cancer gene or locus can be obtained. De Wit and colleagues¹⁹ found that under FISH analysis, melanomas had a higher degree of aberrant signal numbers for the chromosome 1 centromere than Spitz nevi did.

However, human cancers, including melanoma, frequently have aberrations that involve only fragments of the chromosome. The detection of these types of aberrations requires probes targeted to unique (ie, nonrepetitive) sequences of DNA. Unique sequence probes give smaller hybridization signals and can be more difficult to detect. However, by using larger probe sizes (100-300 kilobases), detection of unique sequences is possible even in paraffin sections (Figure 2).

The advantage of FISH is that it can detect cells with aberrations in the presence of significant numbers of normal cells. For example we have been able to detect gene amplifications in single melanoma cells (Figure 2) in the basal melanocytes of histologically normal-appearing skin adjacent to acral lentiginous melanomas²⁰ and in regional lymph nodes. The disadvantage is that detection of heterozygous deletions is more difficult, especially if tumor cells cannot be reliably distinguished from contaminating normal cells. As most FISH studies use tissue sections cut at normal thicknesses for histological examination, only a single copy of a gene present in a nucleus may not represent a true loss; it may be an artifact of the sectioning procedure. Thus, analysis of large numbers of cells is required to detect a decrease in average copy number of one probe relative to a reference probe.

With the use of breakpoint-flanking probes, FISH can also detect translocations. This can be diagnostically helpful in certain hematopoietic malignancies with recurrent translocations, as in the case of large-cell anaplastic lymphoma.²¹ The disadvantage of FISH is that it can only look at a few loci at a time, and that analysis is time-consuming because signals in a large number of nuclei have to be counted.

As originally described,²² comparative genomic hybridization (CGH) detects and maps DNA sequence copy number variation throughout the entire genome onto a cytogenetic map supplied by metaphase chromosomes (Figure 3, left). It can be regarded as a variation of FISH in which the entire genome of a sample such as DNA from a skin tumor is used as a hybridization probe. The tumor is freed from contaminating normal cells by manual dissection; the DNA is then extracted and labeled with a fluorochrome (green by convention); and a reference probe of normal genomic DNA from a healthy donor is labeled with a different fluorochrome (red by convention). The green and red probes are mixed and hybridized onto spreads of normal human chromosomes. These metaphase spreads are also prepared from lymphocytes of a healthy donor to serve as a cytogenetic map. During the hybridization, the 2 DNA populations compete for binding to corresponding chromosomal regions. In the presence of deletions in the tumor genome, less green probe will be available to hybridize to the corresponding region of the chromosome, which will therefore appear redder. If an increased number of copies exists somewhere in the tumor genome, the corresponding regions of the chromosomes will appear more green (Figure 4). The ratio of red and green fluorescence intensity can be used to quantify the copy number change. A ratio of 1 indicates normal copy number; a ratio of less than 1 indicates a loss; and a ratio greater than 1 indicates a gain. Gains with a high ratio that affect only portions of a chromosomal arm are indicative of amplifications, which typically occur in chromosomal regions that contain oncogenes.

Recently, a new implementation of CGH was developed in which the metaphase chromosomes are replaced by arrays of genomic bacterial artificial chromosome clones (Figure 3, right).^{23 - 24} Relative copy number can then be measured at loci specified by the bacterial artificial chromosome clones by hybridization of fluorescently labeled test and reference DNAs as in conventional CGH.

The use of metaphase chromosomes as the hybridization target previously limited the resolution of CGH to 10 to 20 megabases, prohibited resolution of closely spaced aberrations, and only allowed linkage of CGH results to genomic information and resources with cytogenetic accuracy. However, in array CGH, the resolution is determined by the genomic spacing and/or length of the target clones, and the positions of the clones are accurately known on the human DNA sequences because each clone contains a sequence tag. Array CGH allows accurate quantification of DNA copy number variations over a wide dynamic range, including reliable detection of single-copy deletions and duplications.²⁴ Array CGH provides substantially improved resolution and sensitivity compared with conventional CGH in the analysis of tumor genomes.²⁴

It has been shown that most melanomas show chromosomal aberrations on CGH analysis, whereas benign tumors do not have any (or have only a very restricted set). Melanomas frequently have losses of chromosomes 9, 10, 6q, and 8p as well as gains involving chromosomes 7, 8q, 6p, 1q, 17, and 20.^{20 ,25} On the contrary, Spitz nevi have either no aberrations or an isolated gain of chromosome 11p. Acral melanomas differ from superficial spreading melanomas by a high prevalence of focused gene amplifications. In a study of 15 acral melanomas and 15 superficial spreading melanomas of comparable tumor thickness, we found that all acral melanomas had at least 1 (average 2.0) amplification, whereas only 2 superficial spreading melanomas had a single amplification each. We could show that the gene amplifications arise early in acral melanoma, and can already be detected at the in situ stage.

The clear differences in aberration patterns between benign and malignant melanocytic tumors suggest that they could be of diagnostic use in ambiguous cases. To date we have analyzed 185 primary tumors by CGH that were diagnosed unequivocally as either malignant (129) or benign (56) by conventional histopathologic techniques: 122 (95%) of the melanomas had chromosomal aberrations and 7 (5%) had no aberrations. On pathological review of the melanomas without aberrations we found that 4 cases had significant infiltration of lymphocytes, which possibly had diluted the tumor. One case had to be reclassified as Spitz nevus. Of the 57 benign nevi, 48 (84%) had no aberrations, and 6 (11%) had an isolated gain of the entire chromosome 11p (all of which were Spitz nevi). None of the 129 melanomas had a whole-arm gain of chromosome11p. One of the cases originally diagnosed as benign showed numerous aberrations, including several amplifications on chromosome 7. A reevaluation showed that the small punch-biopsy specimen containing bland-appearing dermal infiltrate of spindle cells originally diagnosed as blue nevus in fact was a large exophytic nodule on the toe. The clinician had not reported this information. A complete excision of the tumor showed an invasive acral lentiginous melanoma, which was suggested by the multiple amplifications found by CGH. Two (4%) of the benign cases showed aberrations of regions other than 11p.²⁶

A variety of techniques have been developed to determine the abundance of RNA molecules. Northern blots, PCR after reverse transcription of RNA, and nuclease protection assays are established techniques to determine abundance of individual RNA species. In situ hybridization is used to detect individual RNA species in tissue sections. Differential display,²⁷ subtractive hybridization,²⁸ complementary DNA (cDNA) fragment fingerprinting,²⁹ and serial analysis of gene expression³⁰ are techniques that permit the detection of differences in the amount of RNA species between samples. Subtractive hybridization has identified the melanoma metastasis suppressing genes MDA-7³¹ and KiSS-1.³² The KiSS-1 gene product that has been called metastin has recently been shown to bind to a G-protein–coupled receptor and promote adhesion of melanoma cells.³³ Differential display has been used to identify melastatin, a gene that was found at much reduced levels in highly metastatic as opposed to low metastatic melanoma cell lines.³⁴ Melastatin was also found to be differentially expressed in tissue sections of human melanocytic neoplasms by in situ hybridization. Benign nevi expressed high levels of melastatin, whereas primary melanomas showed variable melastatin expression. More recently, down-regulation of melastatin messenger RNA in primary cutaneous melanoma has been shown to be a prognostic marker that is independent of tumor thickness and other variables.³⁵

In recent years, array-based techniques have been developed that permit simultaneous analysis of gene expression on a scale of hundreds to thousands of genes (see Liotta and Petricoin³⁶ for review). These techniques reverse transcribe RNA extracted from tissues or cells into cDNA and labeled radioactively or fluorescently. The labeled probe is hybridized onto a nucleic acid array. These arrays are generally produced in one of two ways: (1) by robotic deposition of nucleic acids (PCR products, plasmids, or oligonucleotides) onto a glass slide or (2) by in situ synthesis (eg, using photolithography) of oligonucleotides.³⁷ Fluorescent labeling permits the simultaneous hybridization of a differentially labeled test and reference samples, as described for CGH. The concomitant hybridization of the test and a reference probe significantly improves the measurement precision because it adds an internal standard that allows adjustment for variation in performance of the individual targets of the array. Ideally, the reference RNA should express all RNAs represented on the array, and should be available in large quantities so that one source of reference RNA can be used for all experiments.

Preliminary studies indicate that large-scale expression analysis may be helpful for diagnostic classification. For example, by analyzing multiple samples obtained from patients with acute leukemia or diffuse large B-cell lymphoma, researchers discovered gene expression markers that could be used in the classification of these cancers.^{38 - 39} In these studies, a large number of genes (about 50) rather than single markers were required for classification. Expression screening has been used to determine differences in expression patterns of low- and high-metastatic subclones of the melanoma cell line A375.⁴⁰ The researchers⁴⁰ found that overexpression of the gene RhoC resulted in a metastatic phenotype in this cell line. Future studies are required to determine whether RhoC overexpression can be used diagnostically. Expression screening has also been used to attempt classification of human melanoma.⁴¹ However, the small cohort size of the study and the heterogeneous nature of the material precluded conclusions about its clinical applicability.

Expression screening as a diagnostic tool for dermatopathologic conditions is currently restricted by its need for relatively large quantities of unfixed material. At the writing of this manuscript no technique is available that permits the extraction RNA in the quantities and of the quality necessary for expression screening using microarrays. Changing current fixation protocols to permit RNA extraction for the small minority of cases in which diagnostic problems arise seems impracticable, at least at the moment. By contrast, DNA is much more stable than RNA and can be extracted from formalin-fixed tissue in quantities and qualities adequate for a variety of analyses.

Molecular techniques have the potential to play a decisive role in the differential diagnosis of histologically ambiguous tumors. While screening techniques such as CGH and expression analysis are largely confined to research settings because they are time-consuming and expensive, they can lead to refined variants such as FISH tests for certain genomic regions that may play an important adjunctive role in the diagnosis of difficult cases in the future.

Accepted for publication July 31, 2001.

Corresponding author and reprints: Boris C. Bastian, MD, Comprehensive Cancer Center, University of California, San Francisco, PO Box 0808, San Francisco, CA 94143 (e-mail: bastian@cc.ucsf.edu).