Article Text

Download PDFPDF

CDKN2A mutations and melanoma risk in the Icelandic population
  1. A M Goldstein1,
  2. S N Stacey2,
  3. J H Olafsson3,
  4. G F Jonsson2,
  5. A Helgason2,
  6. P Sulem2,
  7. B Sigurgeirsson3,
  8. K R Benediktsdottir4,
  9. K Thorisdottir3,5,
  10. R Ragnarsson5,
  11. J Kjartansson5,
  12. J Kostic2,
  13. G Masson2,
  14. K Kristjansson2,
  15. J R Gulcher2,
  16. A Kong2,
  17. U Thorsteinsdottir2,
  18. T Rafnar2,
  19. M A Tucker1,
  20. K Stefansson2
  1. 1
    Genetic Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, Bethesda, Maryland, USA
  2. 2
    deCODE Genetics, Reykjavik, Iceland
  3. 3
    Department of Dermatology, Landspitali-University Hospital, Reykjavik, Iceland
  4. 4
    Department of Pathology, Landspitali-University Hospital, Reykjavik, Iceland
  5. 5
    Department of Plastic Surgery, Landspitali-University Hospital, Reykjavik, Iceland
  1. Dr A M Goldstein, Genetic Epidemiology Branch/NCI/NIH/DHHS, Executive Plaza South, Room 7004, 6120 Executive Blvd MSC 7236, Bethesda, MD 20892-7236, USA; goldstea{at}


Background: Germline CDKN2A mutations have been observed in 20–40% of high risk, melanoma prone families; however, little is known about their prevalence in population based series of melanoma cases and controls.

Methods: We resequenced the CDKN2A gene, including the p14ARF variant and promoter regions, in approximately 703 registry ascertained melanoma cases and 691 population based controls from Iceland, a country in which the incidence of melanoma has increased rapidly.

Results: We identified a novel germline variant, G89D, that was strongly associated with increased melanoma risk and appeared to be an Icelandic founder mutation. The G89D variant was present in about 2% of Icelandic invasive cutaneous malignant melanoma cases. Relatives of affected G89D carriers were at significantly increased risk of melanoma, head and neck cancers, and pancreatic carcinoma compared to relatives of other melanoma patients. Nineteen other germline variants were identified, but none conferred an unequivocal risk of melanoma.

Conclusions: This population based study of Icelandic melanoma cases and controls showed a frequency of disease related CDKN2A mutant alleles ranging from 0.7% to 1.0%, thus expanding our knowledge about the frequency of CDKN2A mutations in different populations. In contrast to North America and Australia where a broad spectrum of mutations was observed at a similar frequency, in Iceland, functional CDKN2A mutations consist of only one or two different variants. Additional genetic and/or environmental factors are likely critical for explaining the high incidence rates for melanoma in Iceland. This study adds to the geographic regions for which population based estimates of CDKN2A mutation frequencies are available.

View Full Text

Statistics from

The CDKN2A (MIM 600160) gene is the major gene conferring high risk of cutaneous malignant melanoma. CDKN2A encodes several cell cycle inhibitors translated from alternatively spliced transcripts. The isoform1 (α) transcript comprising exons 1α, 2 and 3 encodes the cyclin dependent kinase inhibitor p16 (INK4a).1 Isoform4 (β) comprises an alternative first exon, 1β, spliced to exons 2 and 3 in common with isoform1, leading to a shifted reading frame relative to isoform1; the alternative reading frame protein is designated p14ARF.2 p14ARF acts via the p53 pathway to induce cell cycle arrest or apoptosis.3 4 Isoform3, whose expression has only been seen in the pancreas, utilises an alternative splice donor at the end of exon 1 resulting in a 12 kD protein that exhibits retinoblastoma independent cell growth inhibition.5

The frequency of mutations in CDKN2A has been investigated extensively using family based approaches but rarely using population based approaches. Germline CDKN2A mutations have been observed in 20–40% of melanoma prone families. The mutation frequency varies by geographic region with higher frequencies in regions with lower incidence.6 Outside of high risk families, the prevalence of mutations is much lower. The two major population based studies assessing CDKN2A mutation frequencies were based primarily on melanoma patients from Australia and North America and showed prevalences of functional CDKN2A mutations ranging from 0.2–2%.79

Melanoma incidence has increased rapidly in Iceland in the last decades. During 2000–2004, the age standardised incidence was 10.9/100 000 for males and 19/100 000 for females and had tripled from the 1980 incidence.1012 No population based estimates of disease related CDKN2A mutations (outside of families) are yet available from Northern Europe. The goal of this study was to estimate the frequency of CDKN2A mutations in population based Icelandic melanoma cases and controls, assess the risk of melanoma in identified variants, and examine evidence for associations with other cancers (pancreatic cancer (PC)), breast cancer (BC), neural system tumours (NST), and head and neck cancers) previously associated with CDKN2A mutations.13


Patients and design

Study approval was granted by the Icelandic National Bioethics Committee, the Icelandic Data Protection Authority, and the National Cancer Institute Institutional Review Board. Records of melanoma diagnoses, all histologically confirmed, from 1955–2005 were obtained from the Icelandic Cancer Registry (ICR). Invasive cutaneous malignant melanoma (CMM) was identified through ICD10 code C43. Diagnoses of melanoma-in-situ (MIS) from 1980–2003 were identified by ICD10 code D03. Ocular melanoma (OM) was identified by code C69 in combination with a SNOMED morphology code indicating melanoma. Melanomas arising at mucosal sites (primarily oral, nasal, and genital) were identified by listing non-C43 codes that were associated with a SNOMED morphology code indicating melanoma, then manually inspected to remove misclassification errors; verified entries associated with a mucosal site ICD10 code were designated mucosal melanoma. Metastatic melanoma was identified by a SNOMED morphology code indicating melanoma with a 6 suffix, regardless of the ICD10 code. Breast cancers were identified by ICD10 code C50. Head and neck tumours were identified by codes C00–C09, C11, C13, C14, C30–C33. NSTs were identified by codes C70–C72 or C47. Pancreatic tumours were identified by code C25.

All melanoma patients identified through the ICR were invited to a recruitment centre where they signed an informed consent form, provided a blood sample and answered a questionnaire with the aid of a study nurse. Study consent was obtained from 776 melanoma patients comprising 527 invasive CMM, 217 MIS, 17 OM, 13 mucosal melanoma and two metastatic melanoma cases. The overall recruitment success rate was approximately 80%.

The population based control group comprised 827 individuals, unrelated to each other at three meioses, selected randomly from the Icelandic Genealogical Database with all individuals aged 18–70 years. Medical records of the controls were not investigated and melanoma patients were not excluded if they occurred among the controls. Patients who were selected both as melanoma cases and controls (four individuals) were treated statistically as two separate individuals. Further details of the selection and recruitment of the control group have been described previously.14


CDKN2A exons 1α, 1β, 2, and 3 and about 1 kilobase of the promoter regions upstream of exons 1α and 1β from approximately 703 melanoma patients and 691 controls were successfully sequenced. This corresponds to sequencing yields of 90.5% and 83.6% for cases and controls, respectively. Procedures for sequencing are described elsewhere.14 The common IVS2-105A>G mutation15 was sequenced; no mutations were detected. All identified CDKN2A variants were confirmed by manual inspection of primary signal traces. Twenty-one variants were detected. The call rate for the IVS2-278C>T variant was <66% in cases and controls, so this variant was excluded from further evaluation. For the remaining variants, the call rate was 94% for cases and 98% for controls.

Statistical analysis

Differences in CDKN2A variant frequencies between cases and controls were evaluated by Fisher’s exact or Wilcoxon Mann–Whitney (WMW) test using StatXact (version 4.0). Point estimates and 95% confidence intervals (CI) of unadjusted odds ratios (OR) were calculated and used as the measure of association between melanoma risk and specific individual CDKN2A variants.

High penetrance mutations in CDKN2A would be expected to result in a higher risk of melanoma in relatives of affected mutation carriers. Therefore, we determined a familial relative risk (fRR1°–3°) parameter which compares, for each CDKN2A variant, the risk of melanoma in first through third degree (1°–3°) relatives of melanoma affected variant carriers with the risk of melanoma in 1°–3° relatives of the overall group of (genotyped) melanoma patients. First through third degree relatives of each melanoma patient were identified through the Icelandic Genealogical Database.16 Recorded diagnoses of melanomas and other cancers within this circle of relatives were obtained from the ICR. The familial relative risk (fRR) method has been described previously.17 Briefly, for each CDKN2A variant, the fRR was determined as:

Embedded Image

where r is the number of relatives of melanoma-affected carriers of the CDKN2A variant (counting individuals multiple times who are related to more than one CDKN2A variant carrier18) and a is the number of relatives of the CDKN2A variant carriers who are themselves affected with the disease being queried (that is, melanoma, BC, head and neck cancer, NSTs, or PC), again counting individuals multiple times who are related to more than one carrier. In the denominator, n is the total number of relatives of the genotyped melanoma patients, and x is the number of those relatives who are affected with the disease being queried. fRR was calculated for relative circles at each degree of relatedness from 1°–3°. A combined fRR for 1°–3° relatives (fRR1°–3°) was then derived using a weighting scheme as described previously.17 Also, as in the previous study, the fRR for each degree of relation is a weighted average over subpopulations, where the population is stratified according to sex and in 5 year birth periods.

We assessed the significance of fRR1°–3° by simulation. The observed fRR1°–3° value was compared to a set of empirical fRR1°–3° values derived from 1000 groups of matched controls, each group being the same size as the number of carriers for each CDKN2A variant, drawn at random from among the genotyped melanoma patients. The p value is the proportion of the 1000 control groups that gave a fRR1°–3° as large as or larger than the fRR1°–3° for the variant carriers. This fRR approach compares the risk to relatives of variant carrying melanoma patients to the risk to relatives of the overall group of melanoma patients studied but not to the general population risk. Icelandic population based fRR assessments for melanoma have been published previously.17

The predicted biochemical severity of the coding variants was evaluated using the BLOSUM62 matrix.19 Using BLOSUM62, missense mutations were categorised as non-conservative or conservative.


Twenty variants were reliably identified, 16 in CDKN2A and four in p14ARF. Table 1 presents the variants and their frequencies in all melanoma cases combined, invasive CMM, MIS, mucosal, metastatic and OM patients.

Table 1 Allele frequencies for CDKN2A variants identified in population based Icelandic melanoma cases and controls

The major known variants, including the -981G>T, -735G>A, and -493A>T promoter variants, c.-191G>A 5′ UTR variant, and Nt500C>G and Nt540C>T 3′ UTR variants, were all detected. Coding CDKN2A variants were A57V, G89D, and A148T. A57V and G89D occur in the overlapping coding sequence of p14ARF; however, both result in synonymous p14ARF variants, designated R112R and G143G, respectively.

Because some observed variants were rare, we assessed their potential for affecting disease risk using three complementary methods. First, we looked for association with disease incidence. Second, we examined whether relatives of patients who carried mutations were at increased risk of melanoma or several other cancers. Third, we examined whether the non-synonymous coding variants made amino acid changes that were predicted to be non-conservative.

The only variant that showed a significant association with melanoma was the missense variant G89D (p = 0.011) (table 1). The G89D association strengthened when the analysis was restricted to invasive CMM patients (p = 0.0015). G89D was observed in nine invasive CMM patients and one control (OR 13.8, 95% CI 1.9 to 606). Two rare non-coding CDKN2A variants (IVS1-122G>C and IVS1+37G>C) were observed in one and two melanoma cases, respectively, but not in any controls. The previously described IVS1+37G>C variant, believed to be a neutral polymorphism,9 20 generates a non-synonymous G63R mutation in exon 1 of the CDKN2A isoform3 (p12) protein. None of the known neutral polymorphic or novel rare non-coding variants were significantly associated with melanoma (table 1). Further, no disease related non-synonymous variants were detected in any ocular or mucosal melanoma patients.

We used the familial relative risk (rRR1°–3°) approach to examine whether relatives of the patients who carried the four detected missense variants (A57V, G89D, A148T, and IVS1+37G>C (Iso3 G63R)) were at increased risk of melanoma. Patients who carried G89D had a significantly increased risk of melanoma among their 1°–3° relatives compared to the overall melanoma group (table 2). There was no detectable increase in melanoma risk among 1°–3° relatives of melanoma patients who carried A148T or IVS1+37G>C (Iso3 G63R). For A57V, the point estimate of fRR was over threefold; however, this estimate was not significant (p = 0.066).

Table 2 Familial relative risk (fRR) for melanoma and other cancers in 1°–3° relatives of melanoma affected carriers of selected CDKN2A variants

The fRR method was also used to compare the risks of PC, head and neck cancers, NSTs and BC among relatives of variant carrying melanoma patients compared to relatives of the overall group of melanoma patients. Relatives of melanoma patients who carried G89D had significantly greater risks of head and neck cancers and pancreatic carcinomas than relatives of the general group of melanoma patients (table 2). A148T exhibited a non-significant fRR of 1.83 for head and neck cancers (p = 0.059). There was no evidence that any of the CDKN2A variants conferred high penetrance cross risk for either BC or NSTs.

In addition, because of the purported relationship between p14ARF and NSTs, we evaluated fRR for the four exon 1β-specific variants TS-229A>T, TS-102C>T, TS-89C>T and c.316-121T>C. No increased familial risk for melanoma, NSTs, head and neck cancers, pancreatic carcinoma or BC was detected (data not shown).

We used the BLOSUM62 matrix to evaluate the biochemical severity of the three CDKN2A missense variants. According to BLOSUM62, G89D was classified as a non-conservative amino acid change. In contrast, A57V and A148T were classified as conservative amino acid replacements using the BLOSUM62 matrix. In summary, the observations are fully consistent with the notion that G89D is a highly penetrant melanoma risk allele.

The germline G89D variant has not been reported outside Iceland and may therefore be a unique Icelandic founder mutation. All carriers of G89D shared a SNP haplotype background for the CDKN2A region (data not shown). Reference to the Icelandic Genealogical Database identified a most recent common ancestor linking all G89D carriers (case and control) at 10 generations. The common ancestor was a female who lived from approximately 1605–1665 in Hunavatnssysla county in northern Iceland.

Table 3 shows the frequency of potentially disease related CDKN2A variants in Icelandic melanoma patients and population based controls. The allelic frequency of CDKN2A disease related variants in melanoma patients is 0.7% (95% CI 0.3% to 1.3%) for G89D alone and 1.0% (95% CI 0.55% to 1.7%) for G89D and A57V together. The population based allelic frequency for these variants in controls is 0.08% (95% CI 0.002% to 0.43%) for G89D alone and 0.38% (95% CI 0.13% to 0.90%) for G89D and A57V combined. These estimates are consistent with those reported by the population based Genes and Environment in Melanoma (GEM) study,8 whether G89D is considered alone (p = 0.62) or both G89D and A57V are classified as disease related variants (p = 0.75).

Table 3 Frequency of G89D and A57V variants in the Icelandic population


This population based study of Icelandic melanoma cases and controls showed a frequency of disease related CDKN2A mutant alleles ranging from 0.7–1.0% (combined 95% CI 0.3% to 1.7%) in melanoma cases and 0.08–0.38% (combined 95% CI 0.002% to 0.9%) in population based controls, thus expanding our knowledge about the frequency of CDKN2A mutations in different populations. Since these are allelic frequencies, the carrier frequencies will be approximately double these values. The frequency of CDKN2A mutations in melanoma patients observed in this study is thus consistent with previous population based estimates from predominantly North American and Australian populations.79 However, in North America and Australia, a broad spectrum of mutations is observed whereas in Iceland a similar frequency of functional mutations consists of only one or two different variants. This pattern is similar to what is seen in BRCA2 in Iceland, where the frequency of mutation carriers is as high or higher than in other countries, yet mutation carriers all harbour a single founder mutation, BRCA2 999del5.21 22 The predominance of a single CDKN2A founder mutation is similar to what has been observed in Sweden and the Netherlands.6 2326

This study identified G89D as the major high penetrance CDKN2A mutation in Iceland. G89D had the strongest, most consistent association with melanoma and showed a ninefold increase in risk of any melanoma and a 13-fold increase for invasive CMM. Further, based on the fRR evaluation, G89D conferred a high penetrance cross-risk for head and neck cancers and pancreatic carcinoma. In addition, similar to most CDKN2A mutations, G89D may be a founder mutation, with all carriers stemming from a common ancestor who resided in northern Iceland in the 1600s. Finally, G89D was estimated to be carried by 1.4% of Icelandic melanoma patients and over 2% of invasive CMM patients.

The evidence for A57V being a disease related mutation is less clear. A57V was not significantly associated with melanoma, being present in four cases and four controls. Further investigation revealed that one of the controls had a family history of melanoma (with a parent and an aunt with CMM), while the other three controls did not. In addition, the patients who carried A57V showed an increased non-significant fRR estimate for melanoma compared to other patients. Moreover, evaluation of the biochemical severity showed that the amino acid change was conservative. Also, alanine 57 is not conserved across 14 animal species previously examined. Further, valine is at position 57 in the rat, mouse, and opossum species considered further, suggesting that A57V does not result in a functional defect in the p16 protein.6 27 This variant has, however, been observed in a pancreatic tumour,28 in a melanoma prone family,29 and two multiple primary melanoma patients from GEM.8 Thus, A57V may moderately increase melanoma risk; however, additional studies will be required to determine its relationship with melanoma completely.

The A148T variant is a frequently reported polymorphism. In the current study, its allelic frequency was 2.5% in cases and 2.7% in controls; there was no evidence for A148T conferring increased melanoma risk. While most reports have classified it as a neutral polymorphism, Debniak et al30 reported that the variant conferred a 2.5-fold increased CMM risk in Polish subjects. The variant also appeared to confer cross-risk of breast and other cancers in the Polish study.30 No such cross-risk of BC, PC, or NSTs was observed in the Icelandic sample. Debniak et al30 speculated that the risk might be due to A148T being in linkage disequilibrium with a promoter polymorphism -493A>T. This promoter variant was observed in 32 cases and 36 controls in the current study, was in strong linkage disequilibrium with A148T, and showed no association with melanoma risk (OR 0.88, 95% CI 0.54 to 1.44). Taken together, our association data support the view that both A148T and -493A>T are neutral polymorphisms in the Icelandic population.

The CDKN2A IVS1+37G>C variant generates a G63R coding mutation in the isoform3 p12 protein.5 It was observed in two melanoma cases and no controls. Neither case exhibited an increased family history of melanoma, pancreatic or other cancers. Although this variant would be classified as a non-conservative amino acid using the BLOSUM62 matrix, there was no evidence for an association to melanoma in this study. In addition, this variant was originally observed in a non-carrier melanoma case from an Italian multiplex melanoma family harbouring a R24P CDKN2A mutation; the variant was predicted to have no effect on mRNA processing using the Splice View program.20 The variant was also classified by the GEM study as a non-functional variant and observed in 0.58% of single primary melanoma and 0.52% of multiple primary melanoma patients.9 Thus, this variant does not appear to be disease related.

Accession numbers

CDKN2A isoform1 (p16): NM_000077

CDKN2A isoform3 (p12): NM_058197

CDKN2A isoform4 (p14ARF): NM_058195

A number of cancers including BC, NSTs and PC have previously been associated with CDKN2A mutations.13 The strongest and most consistent association has been observed between specific CDKN2A variants and PC.6 13 31 NSTs have also been reported to be associated with large deletions and/or mutations that alter p14ARF, but this relationship has not been fully resolved.6 13 In addition, a significantly increased risk of BC was reported in CDKN2A melanoma prone families from Sweden, mainly with the predominant Swedish founder mutation (R112_L113insR),24 but this cross-risk has not been observed for other CDKN2A mutations. There are also reports of a relationship between head and neck cancers and germline CDKN2A mutations.13 32 33 We used the fRR approach to examine these non-melanoma cancers and specific CDKN2A variants. The major Icelandic melanoma related variant G89D showed significant fRR for both pancreatic and head and neck cancers, suggesting a high penetrance cross-risk for these tumours. Indeed, carriers of G89D had more frequent familial connections to head and neck cancer cases than to melanoma cases (frequencies 0.0122 and 0.0100, respectively), suggesting that the absolute risks for both cancers might be similar in these families. None of the exon 1β-specific variants or missense variants showed an increased familial risk for NSTs. Similarly, no cross-risk was detected for BC, suggesting that this risk may be limited to the R112_L113insR mutation.

In conclusion, this study of CDKN2A mutations from Iceland expands the geographic regions for which there are population based estimates of prevalences of CDKN2A variants in melanoma patients and controls. This study also identified a unique Icelandic founder mutation G89D that confers increased risk of melanoma and familial risks of pancreatic carcinoma and head and neck cancers. Further, G89D is the key high penetrance CDKN2A variant in the Icelandic population. Thus, although specific CDKN2A mutations substantially increase risk for melanoma, based on the mutation frequency estimated in this study, the overall impact of CDKN2A mutations on Icelandic melanoma patients and the general Icelandic population is expected to be low. Additional genetic and/or environmental factors are likely critical for explaining the high incidence rates for melanoma in Iceland.


We thank all study participants for taking part in this research. We thank S Sveinsdottir for assistance in the recruitment of patients. This research was supported in part by the Intramural Research Program of the National Institutes of Health, NCI, DCEG; by a subcontract agreement to UVS (subsidiary of DeCode Genetics) under Westat contract N02-CP-91026 to the National Cancer Institute.

Copyright: Title 17 U.S.C § 105 states that copyright protection is not available for any work of the United States Government. Since my authorship contribution was done as part of my official duties as a National Institutes of Health (NIH) employee, my work is a work of the United States Government and as such is in the public domain. If Publisher intends to disseminate the work in foreign countries, Publisher may secure copyright to the extent authorized under the domestic laws of those foreign countries. The copyright will be subject to a paid-up, nonexclusive, irrevocable worldwide license to the United States in the manuscript of such copyrighted work to reproduce, prepare derivative works, distribute copies to the public and perform publicly and display publicly the work, and to permit others to do so. Publisher should not pay royalty income for work done by Federal employees as part of their official duties. This agreement shall be governed and construed in accordance with Federal law as interpreted by the Federal courts in the District of Columbia.


View Abstract


  • Competing interests: None declared.

  • Ethics approval: Study approval was granted by the Icelandic National Bioethics Committee, the Icelandic Data Protection Authority, and the National Cancer Institute Institutional Review Board.

  • Patient consent:All melanoma patients identified through the Icelandic Cancer Registry were invited to a recruitment centre where they signed an informed consent form.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.