A comparison of CDKN2A mutation detection within the Melanoma Genetics Consortium (GenoMEL)

Abstract

CDKN2A is the major melanoma susceptibility gene so far identified, but only 40% of three or more case families have identified mutations. A comparison of mutation detection rates was carried out by “blind” exchange of samples across GenoMEL, the Melanoma Genetics Consortium, to establish the false negative detection rates. Denaturing high performance liquid chromatography (DHPLC) screening results from 451 samples were compared to screening data from nine research groups in which the initial mutation screen had been done predominantly by sequencing. Three samples with mutations identified at local centres were not detected by the DHPLC screen. No additional mutations were detected by DHPLC. Mutation detection across groups within GenoMEL is carried out to a consistently high standard. The relatively low rate of CDKN2A mutation detection is not due to failure to detect mutations and implies the existence of other high penetrance melanoma susceptibility genes.

1. Introduction

GenoMEL (the Melanoma Genetics Consortium; http://www.genomel.org) is comprised of groups on four continents working on high penetrance genes in multiple case melanoma families. The major susceptibility locus for melanoma is CDKN2A on chromosome 9p. The majority of causal mutations at this locus, many of which are single base pair substitutions in exons 1α and 2, affect the function of the protein p16INK4a. Some of the mutations in exon 2 also impact on the alternative splice product of the locus, p14ARF. Since p16INK4a was first described as a melanoma susceptibility gene,^1–3 increasing numbers of mutations at the locus have been described.

Less common types of germline mutation have been reported, including a promoter variant that creates an alternative initiation codon,^4,5 and a deep intronic mutation common in England.⁶ A comprehensive screen of the intronic regions of CDKN2A identified two additional putative intronic mutations. However, in English pedigrees at least, these do not appear to explain predisposition to melanoma in a significant proportion of families.⁷

Recently, rare causal mutations have been identified in exon 1β; these mutations impact p14ARF alone. Specifically, a germline deletion not affecting p16INK4a was reported in 2001,⁸ a 16 base pair insertion in exon 1β was detected in a Spanish melanoma family,⁹ and a number of pedigrees with exon 1β splice site variants have been described.^10,11 Finally, a recent screen of 146 English melanoma families identified a small number of pedigrees with germline deletions at the 9p21 locus.¹²

Within GenoMEL the overall proportion of families with identifiable mutations is relatively low and there is considerable variation between centres.^13,14 In a study from Italy, 33% of pedigrees with 2 or more cases of melanoma had mutations,¹⁵ whereas a Spanish study showed that 17% of melanoma families had CDKN2A mutations.¹⁶ In Australia lower percentages have been reported, e.g. 8.4% of 2 or more case families.¹⁷ The variation between centres may result from founder effects and the variable presence of other as yet unidentified susceptibility genes such as the putative gene at 1p22.¹⁸ There may also be an effect of the environment. Clustering in families in areas of high sun exposure such as Australia may result from enhanced contribution of lower penetrance susceptibility genes such as MC1R. Indeed, compared to Europe, there is almost doubling of the penetrance of CDKN2A mutations in Australia thought to be due to higher ultraviolet radiation flux.¹⁹

Another possibility however, is that groups had failed to identify significant numbers of mutations at the CDKN2A locus, particularly since early mutation detection studies often used the single stranded conformational polymorphism (SSCP) analysis rather than sequencing. GenoMEL, therefore, designed an audit to evaluate the overall quality of mutation detection across the entire CDKN2A locus. We also investigated the utility of denaturing high performance liquid chromatography (DHPLC) as a screening approach to be used by GenoMEL in large numbers of samples. Samples that had initially been genotyped at the centre of origin by sequencing (eight centres) or by SSCP (one centre) were sent to Leeds, UK, for screening with DHPLC. The study also therefore provides a comparison of sequencing with DHPLC.

2. Materials and Methods

2.1. Samples and general organisation

The core groups within GenoMEL agreed to send samples to the Division of Epidemiology and Biostatistics of the Cancer Research UK Clinical Centre at Leeds, UK. The participating groups were from Barcelona, Spain (BCN), Leiden University Medical Center, the Netherlands (LUMC), Queensland Institute of Medical Research, Australia (QIMR). Massachusetts General Hospital, Boston, USA (MGH), the National Cancer Institute, Washington, USA (NCI/USA), an NCI group collaborating with Emilia-Romagna, Italy (NCI/Italy), the University of Genoa, Italy (U Genoa), the University of Pennsylvania, Philadelphia USA (U Penn) and Westmead Institute for Cancer Research, New South Wales, Australia (WICR). The samples were labelled by study number alone, and therefore the Leeds group were blind to the mutation status of the sample.

All groups provided DNA from two melanoma cases from families with three or more melanoma patients that had been screened by that group, whether a mutation had been detected or not. In each case the initial mutation detection screen carried out at the centre of origin was by sequencing, with the exception of WICR, where the primary screen was by SSCP for CDKN2A exon 1 and by sequencing for exon 2.

The samples were processed by the Leeds group and results sent to the NCI in Washington Bethesda, MD where DHPLC audit results were pooled with the original groups’ results. Only coding mutations were assessed; polymorphisms were not considered in this analysis.

2.3. Statistical analysis

The DHPLC results from Leeds were compared to the results from the original centres using two units of evaluation: “sample” and “exon”. Sample summarized the results over the five different exons evaluated. Exon separately examined CDKN2A exons 1α, 1β, 2 and 3, and CDK4 exon 2. Two measures of evaluation were used: failure and discrepancy. Failure was defined as the percentage of samples or exons that failed the DHPLC assay. Discrepancy was the proportion of inconsistencies between DHPLC and the original centre’s results. To confirm discrepancies and eliminate any sample handling errors at any point in the process, all samples with initial evidence for discrepant results were sequenced at the University of Toronto (D. Hogg).

2.4. PCR amplification

The four exons of CDKN2A (exons 1α, 1β, 2 and 3) and CDK4 exon 2 were amplified from genomic DNA by PCR, using previously described primers (Table 1).^2,3,20 PCR was carried out in a total volume of 25 μl, using 25ng genomic DNA, 0.2mM dNTPs, 50μM each primer, 5% (v/v) DMSO, 1.5mM MgCl2, and 1 unit of AmpliTaq Gold DNA polymerase (Applied Biosystems, Warrington, UK), in the reaction buffer supplied by the manufacturer. PCR amplification conditions were as follows: An initial denaturation at 94°C for 10 min; followed by 30 cycles of denaturing at 94°C (30s), annealing at 55°C (30s), and extension at 72°C (30s), with a final 7 minute extension at 72°C. PCR fragments were isolated by agarose gel electrophoresis and purified prior to sequencing using the QIAquick Gel Extraction Kit (Qiagen, Paisley, UK).

Table 1.

PCR primers and DHPLC conditions

Exon	Primers	Fragment Size	DHPLC Temperature
CDKN2A Exon 1α	F-CAGCACCGGAGGAAGAAAG R- GCGCTACCTGATTCCAATTC	351 bp	65, 68, (69)
CDKN2A Exon 2	F- GGAAATTGGAAACTGGAAGC R- GGAAGCTCTCAGGGTACAAATTC	499 bp	60, 65, 70
CDKN2A Exon 3	F- CCATTGCGAGAACTTTATCC R- TGGACATTTACGGTAGTGGG	329 bp	56, 62
CDKN2A Exon 1β	F- CACCTCTGGTGCCAAAGGGC R- CCTAGCCTGGGCTAGAGACG	351 bp	61, 65, 69
CDK4 Exon 2	F- GCTGCAGGTCATACCATCCT R- ATCATCACACCCCACCTATAGG	371 bp	62

2.5. DHPLC analysis

The DNA samples were screened for sequence changes in CDKN2A exons 1α, 1β, 2, and 3, and CDK4 exon 2, by DHPLC analysis. The DHPLC system had previously been optimised using the Leeds family samples. Each of 22 separate variants observed in the Leeds melanoma samples^21,22 could be clearly detected by DHPLC using the conditions described below.

Temperatures for mutation detection were calculated using the DHPLC Melt program available at http://insertion.stanford.edu/melt.html.²³ Melting temperatures were also determined empirically by running a wild type sample at progressively increasing temperatures until a reduction in retention time of 1 minute was observed. For CDKN2A exons 1α, 1β and 2, analysis was carried out at the temperatures determined by DHPLC Melt. For exon 3, one of the two temperatures used was 1°C higher than that predicted by DHPLC Melt (Table 1).

DHPLC was carried out using a Transgenomic WAVE Nucleic Acid Fragment Analysis system and DNASep column (Transgenomic, Crewe, UK). The composition of buffer A was 0.1M triethylammonium acetate (TEAA); buffer B contained 0.1M TEAA and 25% (v/v) acetonitrile. Analysis was carried out at a flow rate of 0.9 ml/min and a buffer B gradient increase of 2%/min for 4 minutes. Start and end concentrations of buffer B were determined empirically for each fragment.

PCR products were prepared for DHPLC by denaturing at 95°C for 5 minutes and then cooling to 65°C to allow the formation of heteroduplexes. Data analysis was by visual inspection of chromatograms by two independent observers.

2.6. Sequencing analysis of DHPLC ‘positive’ samples

All PCR fragments which displayed an aberrant DHPLC chromatogram were sequenced to identify the underlying nucleotide change. Sequencing reactions were carried out using the ABI PRISM BigDye v2 Terminator Cycle Sequencing Kit. Data collection was performed using a 3100 Genetic Analyser (Applied Biosystems, Warrington, UK) running Applied Biosystems Data Collection Software (version 1). Data analysis was carried out by visual inspection of electropherograms, and using Applied Biosystems Sequence Navigator analysis software (version 1.0.1). DNA sequencing was performed in both directions, initiated from the forward and reverse primers used in the initial PCR amplification of each fragment.

3. Results

A total of 537 samples were screened by DHPLC for sequence variation in five exons (CDKN2A exons 1α, 1β, 2 and 3, and CDK4 exon 2). Of the 2685 products processed by DHPLC 106 (4%) failed the assay. DNA that had been extracted from buccal samples showed a much higher failure rate than DNA extracted from blood; 18/40 (45%) of DHPLC assays on buccal-derived DNA failed, compared to 88/2645 (3%) of assays on blood-derived samples.

The DHPLC audit identified 40 different mutations (138 in total) and 10 different polymorphisms (149 in total) in the five exons investigated. A total of 37/1343 (3%) of DHPLC assays were found to give a false positive screening result in that the DHPLC traces were judged to be atypical, but subsequent sequencing showed that the samples were wild type (Table 2). Where comprehensive local sequencing data were available, a comparison was made between the DHPLC results and the sequencing data. The DHPLC audit results could be compared to the local screening data for (at least) CDKN2A exons 1α and 2, in a total of 451 samples.

Table 2.

Comparison of audit DHPLC results with primary screening results

Exon	Number of exons screeneda	DHPLC positiveb	Causal mutations detected	Polymorphisms detected	False positivesc	Mutations missedd
CDKN2A exon 1α	348	47	37	3	7	2 (Trp15OPA)
CDKN2A exon 2	310	132	83	37	12	1 (Gly101Trp)
CDKN2A exon 3	244	112	7	104	1	0
CDKN2A exon 1β	179	13	6	0	7	0
CDK4 exon 2	262	20	5	5	10	0
Total	1343	324	138	149	37 (3%)	3 (0.2%)

^aThe number of samples where local primary screening data was available for comparison with audit DHPLC results for each exon analysed.

^bThe number of samples displaying an aberrant DHPLC trace.

^cThe number of aberrant DHPLC traces which on sequencing appeared to be wild type.

^dThe number of mutations identified in the local primary screen that were not detected by the audit DHPLC.

Three mutations, identified by the local centre’s primary screen, were not detected by the audit DHPLC (3/1343 = 0.2% by exon, or 3/451 = 0.7% by sample). A Gly101Trp mutation, in CDKN2A exon 2, was not detected by DHPLC in a sample originating from the National Cancer Institute, USA (NCI/USA), despite this mutation being clearly detected in several other samples in this investigation. Sequencing confirmed the presence of the Gly101Trp mutation in this sample. The CDKN2A exon 1 variant Trp15OPA was not identified in two samples originating from the Westmead Institute for Cancer Research, Australia (WICR), where this variant had been detected by SSCP in the primary screen. Sequencing confirmed the presence of the Trp15OPA mutation in these samples, however repeated DHPLC analysis demonstrated that this variant was not detectable using the DHPLC conditions employed. The application of an additional melting temperature (69°C) for the exon 1 DHPLC subsequently enabled the detection of the Trp15OPA variant.

The centralised DHPLC audit did not identify any additional mutations that had not been reported by individual groups after the local primary mutation screen.

4. Discussion

We carried out an assessment of mutation detection by blind exchange of samples from nine melanoma research groups within GenoMEL. The assessment was carried out with the intention of 1) standardising mutation detection across a range of research groups worldwide, 2) identifying mutations that may have been missed in the primary mutation screen, and 3) validating DHPLC as a screening technique.

The DHPLC technique is sensitive, rapid and relatively inexpensive,^24–27 and was therefore considered to be well suited to an audit of this nature. Different screening techniques are known to have different limitations in identifying mutations.^28–30 The use of DHPLC for the audit screen has the further advantage of complementing a primary screen using complete sequencing, the screening technique employed at most GenoMEL centres.

The DHPLC failure rate was low, only 3% of assays on blood-derived DNA samples could not be analysed. However, the failure rate for buccal-derived samples was considerably higher, at 45%. The overwhelming majority of DHPLC failures were in fact failures of the initial PCR reaction, resulting in the absence of PCR product for DHPLC analysis. DHPLC is to some extent sensitive to the quality of DNA used in the initial PCR reaction. Because the number of buccal samples were so few (n=8), we did not attempt to reoptimise exon-specific PCR conditions.

A small number (3%) of DHPLC assays were found to give a false positive screening result. The majority of these were found to be the result of over-cautious scoring of DHPLC traces, as samples that gave a weak or atypical DHPLC trace were typically recorded as positive, as is appropriate for a screening technique. Although unlikely, there is the possibility that these DHPLC positive traces represent a true mutation that is undetectable by sequencing.

Our observed high concordance between primary and audit screens (99.8%) was based on a large sub-set of genotyping data from across GenoMEL. In some cases a direct comparison could not be made between primary and audit screens, as a number of research groups had not routinely screened the whole of CDK4 exon 2 or CDKN2A exon 1β, due to the low reported frequency of mutations in these genes.^20,31–33 Also, some samples had been screened by a specific test for a mutation already identified in an additional family member, hence the primary screen did not cover the entire exon.

Although the DHPLC assay can be optimised to detect known variants in a particular exon, there is always the possibility that the technique could be insensitive to a previously unidentified variant. Employing as broad a set of variants as possible to optimise the DHPLC assay can reduce this risk. One of the benefits of undertaking a mutation detection audit across a number of research groups is that it enables the pooling of many different CDKN2A variants, and therefore increases confidence in the ability of the technique to identify variants in future studies.

The DHPLC analysis of CDKN2A exon 3 was somewhat problematic. Approximately 50% of all exon 3 DHPLC traces were positive, but only 7/244 (3%) carried a causal mutation. The presence of two common 3’UTR polymorphisms (500 c>g and 540 c>t) substantially increased the cost of subsequent sequencing to identify the underlying mutation. In addition, the frequency of the polymorphisms in the population (15% and 20% respectively) is such that they are occasionally found as homozygous variants, which are not detected by DHPLC.³⁴ Therefore, DHPLC is not ideally suited to the analysis of this exon. Perhaps an improved approach to high throughput screening of CDKN2A exon 3 and its flanking regions would be the use of mutation specific tests for the three known variants in this part of the gene (the two 3’UTR polymorphisms in the exon and the causal melanoma associated intronic variant IVS2-105 a>g). No other germline causal variant has been identified in exon 3, and indeed the size of the translated portion (4 amino acids) and demonstrated lack of function of exon 3³⁵ make the existence of additional mutations unlikely.

DHPLC was evaluated as an alternative technique for mutation detection. This study has shown that the technique can be used to carry out primary CDKN2A mutation screening for future large-scale studies where issues of speed and expense are critical. DHPLC has been shown to have only a slightly lower rate of mutation detection than direct sequencing, which is generally regarded to be the most sensitive screening technique.^36,37 Out of a total of 50 different variants identified in this study, a single variant (Trp15OPA) was initially opaque to the audit DHPLC. This represents a sensitivity of 98% compared to sequencing. Following further optimisation the sensitivity was subsequently increased to 100%. The technique’s relative low cost (approximately 8 times cheaper than sequencing) and high throughput makes it ideal for screening large numbers of samples in which the expected mutation frequency is low, for example a population based screen for CDKN2A mutations.

This audit also has shown that the variation in mutation detection frequencies between the different groups is not a result of variation in sequencing approaches; in fact the standard of screening across groups is uniformly high. Rather the variation in mutation frequency between groups reflects differences between populations, either through differing genetic backgrounds or environmental contributions. GenoMEL continues to address this issue to better understand the mutation frequency variation across populations.