MC1R variants, melanoma and red hair color phenotype: A meta-analysis



Melanocortin-1-receptor (MC1R) is one of the major genes that determine skin pigmentation. MC1R variants were suggested to be associated with red hair, fair skin, and an increased risk of melanoma. We performed a meta-analysis on the association between the 9 most studied MC1R variants (p.V60L, p.D84E, p.V92M, p.R142H, p.R151C, p.I155T, p.R160W, p.R163Q and p.D294H) and melanoma and/or red hair, fair skin phenotype. Eleven studies on MC1R and melanoma, and 9 on MC1R and phenotype were included in the analysis. The 7 variants p.D84E, p.R142H, p.R151C, p.I155T, p.R160W, p. R163Q and p.D294H were significantly associated with melanoma development, with ORs (95%CI) ranging from 1.42 (1.09–1.85) for p.R163Q to 2.45 (1.32–4.55) for p.I155T. The MC1R variants p.R160W and p.D294H were associated both with red hair and fair skin, while p.D84E, p.R142H, and p.R151C were strongly associated with red hair only- ORs (95%CI) ranged from 2.99 (1.51–5.91) for p.D84E to 8.10 (5.82–11.28) for p.R151C. No association with melanoma or phenotype was found for p.V60L and p.V92M variants. In conclusion this meta-analysis provided evidence that some MC1R variants are associated both with melanoma and phenotype, while other are only associated with melanoma development. These results suggest that MC1R variants could play a role in melanoma development both via pigmentary and non-pigmentary pathways. © 2008 Wiley-Liss, Inc.

Melanoma incidence has increased over the last decades in almost all Western countries,1, 2 and metastatic melanoma has a 5-year survival rate of only 11%.3 Risk factors for melanoma include environmental UV radiation exposure as well as genetic/host factors such as light skin, family history, number of melanocytic nevi, eye, and hair pigmentation.2, 4, 5

Melanocortin-1-receptor gene (MC1R, MIM no. 155555) is responsible for constitutive pigment variation in humans and has been shown to be a risk factor for melanoma. It is located on chromosome 16q24.3 and encodes for a 7-pass transmembrane G-protein coupled-receptor of 317 amino acids, which has a high affinity for α-melanocyte-stimulating hormone (α-MSH) and adrenocorticotropin.6, 7 The binding of α-MSH to the functional MC1R on melanocytes stimulates the synthesis of eumelanin,8 which determines black/brown pigment. MC1R therefore contributes to determine pigmentation by regulating the relative proportion of eumelanin and phaeomelanin (red/yellow pigment). Eumelanin has been shown to reduce the accumulation of DNA photoproducts, while phaeomelanin may contribute to cancer risk because it generates free radicals following UV exposure.9–11

MC1R is highly polymorphic in the Caucasian population: more than 80 variants have been recently described.12 Some of these variants result in partial loss of the receptor's signaling ability, since they are unable to stimulate cyclic adenosine monophosphate (cAMP) production as strongly as the wild-type receptor in response to α-MSH stimulation.13, 14 It results in a quantitative shift of melanine synthesis from eumelanin to phaeomelanin, which is associated with the “red hair color” (RHC) phenotype. RHC phenotype is characterized by fair pigmentation (fair skin, red hair and freckles), and by sun sensitivity (poor tanning response and solar lentigines).

Genetic association studies have found that the MC1R variants p.D84E, p.R151C, p.R160W and p.D294H, defined as “R” alleles,12, 14 were strongly associated with the RHC phenotype.15–21 Other 2 less frequent variants (p.R142H and p.I155T) have also been classified as R alleles14 based on findings of strong familial association with RHC phenotype.18, 22 The p.V60L, p.V92M and p.R163Q variants seem to have a relatively weak association with RHC phenotype and are designated as “r” alleles.14, 22

Several studies in different populations have reported that the risk of melanoma is higher among individuals who carry MC1R variant alleles than among individuals who are wild type for MC1R.17, 20, 23–29 Melanoma risk attributable to MC1R may arise through the determination of the tanning response of skin to UV light, which can then either ameliorate or exacerbate the genotoxic effects of sunlight.30 However, the relationship between some MC1R variants and melanoma also in darkly-pigmented Caucasian populations suggests that MC1R signaling pathway may have an additional role in skin carcinogenesis beyond the UV-filtering differences between dark and fair skin.12

We review here the studies that have considered the association between the 6 R variants p.D84E, p.R142H, p.R151C, p.I155T, p.R160W, p.D294H and the 3 r variants p.V60L, V92M, p.R163Q and melanoma and/or RHC phenotype, and performed a meta-analysis of the available data. Meta-analysis is a useful approach because the pooled dataset has greater power than each individual study and could therefore lead to more conclusive results on this issue. This systematic revision also provides some clues towards the epidemiology of melanoma by looking extensively at inconsistencies and variability in the estimates of the association with MC1R. Meta-analysis permitted questions to be debated on whether the association of melanoma with MC1R may depend on the composition of the population under study, the country or the methodological features of the studies.


AR, attributable risk; cAMP, cyclic adenosine monophosphate; CI, confidence interval; MC1R, melanocortin-1-receptor; α-MSH, α-melanocyte-stimulating hormone; OR, odds-ratio; RHC phenotype, red hair color phenotype.

Material and methods

A literature search was performed to June 2007 using the following databases: PubMed, ISI Web of Science (Science Citation Index Expanded) and Embase. We identified published papers and abstracts on melanoma and any of the 9 most studied MC1R variants (p.V60L, p.D84E, p.V92M, p.R142H, p.R151C, p.I155T, p.R160W, p.R163Q and p.D294H), along with reports studying at least one of the MC1R variants and hair color and/or skin phenotype. The publications were retrieved using the keywords “MC1R” and “melanocortin 1 receptor” alone or in combination with “melanoma,” “hair color,” “skin color” and “skin type.” The computer search was supplemented by consulting the bibliographies of the articles and reviews. The search was limited to human studies but no language or time restrictions were applied.

An initial screening provided 30 papers on MC1R variants and melanoma and/or phenotype, but among them only 22 included separate data on allele or genotype frequencies for at least one of the 9 MC1R variants considered. Eligible papers for the present analysis were genotype-based studies that reported the frequency for MC1R variants or the odds ratio (OR), with 95% confidence intervals (CI), for MC1R variants and melanoma (single primary) and/or phenotype. Genotype-based studies were preferred to allele-based studies because the allele frequency is low for some variants, and several studies presented data on heterozygous and variant homozygous as a single group. Both hospital and population-based case-control studies, and both familial and sporadic melanoma cases were included in the analysis. Two26, 31 of the 22 selected papers were excluded from the main analysis because they presented allele frequencies of MC1R variants without providing information on genotypes. However, they were included in a separate analysis (not presented here) using allele instead of genotype frequencies. Another study32 was excluded because it analyzed a very selective population characterized by individuals carrying a CDKN2A mutation. Since this population is selected for a genetic factor, we believed that it was not comparable with the other studies. A large, international study33 including data on more than 3,000 individuals with a single primary melanoma or a second or higher-order primary melanoma had to be excluded because, according to the study design, no data on free-of-disease controls were provided. Since it is not known the degree to which this design may affect association related to MC1R variants, we think it could not be comparable with the other case-control studies. Two papers based on the same population, provided separate information on MC1R genotype in melanoma cases and controls,20 and on MC1R and red hair.19 Two papers15, 22 presented overlapped data on red hair phenotype, therefore we considered only the most recent study, which was based on a larger number of subjects.22 Since the smaller study15 included further data on MC1R and skin color, it was not excluded, but it contributed only to the analysis on skin color. In total, 18 articles were considered for the meta-analysis: 1120, 23, 24, 27–29, 34–38 reporting on the association between MC1R variants and melanoma, and 915, 16, 18, 19, 21, 22, 24, 34, 39 on MC1R and phenotype (2 papers24, 34 reported data on MC1R and both melanoma and phenotype). Two papers18, 27 included data from 2 independent family-based and population-based studies. The 2 independent samples were treated as different studies in the meta-analysis.

For each study the following information was retrieved:

  • study characteristics: publication year, study design, study location, percentages of subjects with history of familial melanoma;

  • laboratory methods to detect MC1R variants;

  • study population: number and source of cases and controls, history of familial melanoma or other cancers.

Statistical analysis

The departure of frequencies of each MC1R variant from expectation under Hardy–Weinberg (H–W) equilibrium was assessed by χ2 test in controls. For each variant the ORs were calculated from crude data for subjects carrying at least 1 variant allele. The reference category for each variant included all the subjects who did not carry that variant. An estimation of the overall association of MC1R variants with melanoma and phenotype was provided by random effects models, using the DerSimonian–Laird method.40 Heterogeneity across studies was evaluated by the Q statistic and by I2,41 which represents the percentage of total variation across studies that is attributable to heterogeneity rather than to chance. Since the χ2 test has limited power, we considered that statistically significant heterogeneity existed when the p-value was ≤0.10.42 Between-study heterogeneity was explored through subgroup analyses and meta-regression.43 Four variables were included in the meta-regression to explore the heterogeneity in melanoma studies: the percent of familial melanoma cases, European versus non-European study, healthy versus hospital control source and year of publication. For phenotype studies, 1 further variable was included in the meta-regression, beside year of publication and geographic location: the assessment of phenotype only on healthy subjects or on both healthy subjects and melanoma cases. Within European studies, meta-regression was also performed to evaluate heterogeneity between northern and southern countries, to take into account the role of fairer and darker pigmentation. Greece, Italy, and Spain were considered as southern European countries in this analysis. Publication bias was graphically represented by funnel plot and it was assessed by Egger's test.44

Sensitivity analyses were conducted by excluding studies that deviated from H–W equilibrium, and non-European studies. Among melanoma case-control studies, subgroup analyses were also performed only on papers with sporadic cases or including no more than 10% of familial cases, and separately on studies including hospitalized or healthy controls. Sensitivity analyses for the association between MC1R variants and phenotype were also performed, by excluding papers that assessed phenotype in both healthy individuals and melanoma patients.

When more than 1 risk estimate were provided (i.e. for familial and population) in a single study, we used random effects models with the 2 sources of variation (within and between studies) in sensitivity analysis. Summary ORs were estimated pooling the study-specific estimates by random effects models fitted using SAS (Proc Mixed) with maximum likelihood estimate.

The attributable risk in the population was computed with Miettinen's formula (OR − 1/OR × proportion of exposed cases).45

The statistical analysis was performed using STATA and SAS software.


Departure from the H–W equilibrium was observed for p.V60L and p.R163Q in the same study, (familial melanoma sample)27 for p.D84E in 1 study,20 for p.V92M and p.I155T in the same study,22 for p.D294H in 1 study,21 for p.R142H in 2 studies,19, 21 and for p.R151C in 2 studies.34, 35 Deviation from H–W equilibrium was not assessed in 3 studies23, 28, 37 since data on heterozygous and variant homozygous subjects were not presented as separated groups but were collapsed together.

MC1R variants were identified by sequencing analysis in 9 studies.15, 18, 19, 20, 22, 27–29, 39 Five studies genotyped subjects for specific single nucleotide polymorphisms (SNPs) using Taqman,35, 36 restriction fragment length polymorphisms (RFLP)24 or SNaPshot.21, 34 The remaining 4 studies used a combination of sequencing and RFLP analysis: the latter was applied on a subset of sample16, 23 and/or to identify specific variants.37, 38

MC1R variants and melanoma

A description of the studies included in the meta-analysis of MC1R and melanoma is presented in Table I. All but 2 studies34, 36 were conducted in European countries on Caucasian subjects. The study from USA included more than 95% of Caucasians; fewer than 5% of subjects had missing or other ethnicity. The study from Australia was performed on northern European ancestry individuals. Four studies20, 27, 28, 35 included only sporadic melanoma patients, while 1 study (familial sample)27 included only familial melanoma cases. There were 4 hospital-based studies and 5 population-based studies; 3 papers included both hospitalized and healthy controls.

Table I. List of Studies Included in the Meta-Analysis
  • 1

    Heterozygous and variant homozygous combined, no information on alleles.

  • 2

    >95% Caucasians.

  • 3

    Adolescent twins and relatives, (1) population study and (2) family study.

  • 4

    In all but two studies (15, 34) fair skin was defined as skin type I–II, dark skin as skin type III–IV.

Melanoma case-control studies
AuthorYearCountryN casesN controlsN familial cases (%)Source of controls
Valverde et al.11996UK39441 (3%)Hospital
Ichi-Jones et al.1998UK306190Not statedHospital
Kennedy et al.2001The Netherlands1233850Hospital
Dwyer et al.2004Australia164291Not statedPopulation
Landi et al. (1)2005Italy1651710Mixed
Landi et al. (2)2005Italy8420384 (100%)Mixed
Fargnoli et al.2006Italy1651650Hospital
Han et al.22006USA21987311 (5%)Population
Stratigos et al.12006Greece1231550Mixed
Debniak et al.12006Poland47442158 (12%)Population
Mossner et al.2007Germany32234718 (6%)Population
Fernandez et al.2007Spain116188Not statedPopulation
All  2,3003,433  
Phenotype studies
AuthorYearCountryN red hairN non red hairHealthy or diseased subjects
Valverde et al.1995UK and Ireland2436Healthy
Ichii-Jones et al.1998UK1181Healthy
Smith et al.11998Ireland1358Healthy
Flanagan et al. (1)2000UK18149Healthy
Flanagan et al. (2)2000UK74100Healthy
Bastiaens et al.2001The Netherlands71889Diseased
Duffy et al.32004Australia901,633Healthy
Branicki et al.2007Poland76108Healthy
All  3773,054 
AuthorYearCountryN fair skin4N dark skin4Healthy or diseased subjects
Ichii-Jones et al.1998UK6474Healthy
Valverde et al.1995UK and Ireland7758Healthy
Box et al.31997Australia834Healthy
Smith et al.11998Ireland4526Healthy
Dwyer et al.2004Australia132135Healthy
All  401297 

The pooled ORs for each MC1R variant and melanoma are presented in Table II, along with allele frequency of each variant in controls, and with measure of heterogeneity among studies. All but 2 variants (p.V60L and p.V92M) were significantly associated with melanoma development, with ORs (95%CI) ranging from 1.42 (1.09–1.85) for p.R163Q to 2.45 (1.32–4.55) for p.I155T.

Table II. Allele Frequency, Odds Ratios (OR) With 95% Confidence Intervals (CI) for the Association Between MC1R Variants and Melanoma, and Heterogeneity Estimates
VariantPercent (95%CI) of allele frequency in controlsN studiesN casesN controlsOR (95%CI)Q test p-valueI2 for heterogeneity (%)
p.V60L12.3 (11.7–12.9)101,9033,1621.15 (0.92–1.43)0.0158
p.D84E1.3 (1.1–1.6)81,2711,7732.40 (1.50–3.84)0.470
p.V92M8.8 (8.3–9.4)101,6352,6311.22 (0.99–1.50)0.3114
p.R142H1.2 (1.0–1.4)71,0981,6141.66 (1.01–2.75)0.460
p.R151C9.1 (8.6–9.7)101,9053,1421.78 (1.45–2.20)0.2521
p.I155T0.9 (0.7–1.2)61,0211,9292.45 (1.32–4.55)0.377
p.R160W8.0 (7.5–8.5)101,9003,1641.43 (1.20–1.70)0.640
p.R163Q4.1 (3.7–4.5)81,6172,7301.42 (1.09–1.85)0.369
p.D294H2.6 (2.3–2.9)101,6572,8161.77 (1.17–2.69)0.0841

A significant heterogeneity was found among studies on p.V60L and p.D294H variants (Q test p-values/ I2 = 0.02/58% and 0.08/41%, respectively). The forest plots for these variants are reported (Fig. 1). The forest plots for the other variants are presented in Supplementary Figure 1. None of the variables included in the meta-regression model seems to significantly explain the heterogeneity for p.V60L. However, it seems mainly due to 2 studies from the Mediterranean area28, 29 with the highest ORs. By excluding these studies from the analysis, the hypothesis of homogeneity could be accepted (Q test p-value = 0.18, I2 = 31%) and the OR (95%CI) became 1.03 (0.85–1.23). For p.D294H a significant heterogeneity was observed for northern versus southern European countries (p-value = 0.04). Darker-pigmented Caucasians presented higher risk of melanoma associated with p.D294H (OR (95%CI) = 2.75 (1.62–4.68)) than northern populations (OR (95%CI) = 1.27 (0.76–2.11).

Figure 1.

Study-specific and meta-odds ratio (OR) with 95% confidence intervals (CI) for the association between MC1R p.V60L (a) and p.D294H (b) variants and melanoma in case-control studies.

No publication bias was observed for the analyses of MC1R variants and melanoma. The sensitivity analysis did not provide any significant difference between the main OR and the ORs in subgroups of studies with similar characteristics.

Attributable risk (AR) for the 7 MC1R variants significantly associated with melanoma is presented in Figure 2. The highest AR was observed for the 2 common variants p.R151C (7.48%) and p.R160W (4.54%).

Figure 2.

Attributable risks* in the population for melanoma according to each MC1R variant significantly associated with melanoma (percentages).

MC1R variants and phenotype

Table I presents a description of the studies included in the meta-analysis of MC1R and phenotype. Only 1 study for the analysis on red hair and 2 studies for the analysis on skin color were conducted in Australia; all other studies were performed in Europe. Only 1 study19 presented phenotype data on healthy and melanoma subjects together.

The pooled ORs and heterogeneity assessment for each MC1R variant and phenotype are presented in Table III. The MC1R variants p.D84E, p.R142H, p.R151C, p.R160W and p.D294H were found to be strongly associated with red hair, with ORs (95%CI) ranging from 2.99 (1.51–5.91) for p.D84E to 8.10 (5.82–11.28) for p.R151C. Among the remaining 4 MC1R variants, p.V92M and p.I155T were not significantly associated with red hair, while p.V60L and p.R163Q presented an inverse, significant association. Two of the five MC1R variants associated with red hair were also found to increase the probability of having fair skin: ORs (95%CI) for subjects carrying at least 1 p.R160W or 1 p.D294H allele were 2.81 (1.43–5.50) and 5.10 (1.23–21.23), respectively. p.D84E and p.R151C, as well as p.V92M, presented OR higher than 1.00, but they did not reach the statistical significance. No association was found for p.V60L variant, while ORs were not calculated for the last 3 variants, due to the small number of studies.

Table III. Odds Ratios (OR) and 95% Confidence Intervals (CI) for the Association Between MC1R Variants and Phenotype, and Heterogeneity Estimates
VariantN studiesN red hair/fair skinN non red hair/dark skinOR (95%CI)Q test p-valueI2 for heterogeneity (%)
  1. nc = not calculated.

Red hair
p.V60L53292,8770.50 (0.35–0.72)0.460
p.D84E73632,9902.99 (1.51–5.91)0.2919
p.V92M73632,9900.42 (0.16–1.09)0.00766
p.R142H63422,9354.96 (1.73–14.23)0.0457
p.R151C63422,9358.10 (5.82–11.28)0.2327
p.I155T42581,9901.20 (0.36–4.08)0.440
p.R160W63422,9355.00 (3.43–7.28)0.1439
p.R163Q53292,8770.35 (0.17–0.70)0.990
p.D294H83773,0835.92 (4.07–8.61)0.480
Fair skin
p.V60L32601650.97 (0.60–1.58)0.720
p.D84E43532681.84 (0.54–6.22)0.640
p.V92M32201353.62 (0.62–21.11)0.0762
p.R151C32591632.24 (0.90–5.56)0.2527
p.R160W32601652.81 (1.43–5.50)0.650
p.D294H43562715.10 (1.23–21.23)0.0855

A significant heterogeneity was found among the studies included in the analysis on p.V92M and p.R142H with red hair (Q test p-values/ I2 = 0.007/66% and 0.04/57%, respectively). By meta-regression, we found no significant difference among study characteristics for both variants, probably because of the small number of studies. However the heterogeneity for p.V92M seemed mainly due to the oldest study39 with the highest OR, and to the family-based study18 with the lowest OR (Fig. 3a). By excluding these 2 studies, the OR (95%CI) became 0.44 (0.25–0.80) and the tests for heterogeneity became not significant (p-value for the Q-test = 0.24, I2 = 27%). For p.R142H (Fig. 3b), all the heterogeneity seemed to be explained by the (family-based studies18, 22, both with low OR). By excluding these studies the heterogeneity completely disappeared (Q-test p-value = 0.99, I2 = 0) and the ORs (95%CI) became 10.70 (5.11–22.37). The forest plots for the other variants with red hair color are presented in Supplementary Figure 2. p.V92M and p.D294H presented a significant heterogeneity for analysis on skin color (Q test p-values/ I2 = 0.07/62% and 0.08/55%, respectively). The forest plots of these variants are presented in Figure 4, while those of the other variants in Supplementary Figure 3. As for the analysis on red hair, the oldest study for p.V92M presented a significantly higher OR (meta-regression p-value for publication year = 0.03). The heterogeneity for p.D294H seemed to be due to the geographic location (meta-regression p-value for European versus non-European countries = 0.04): the 2 European studies found higher OR than the 2 Australian papers (OR (95%CI) = 15.24 (2.71–85.82) vs. 2.47 (0.97–6.28)).

Figure 3.

Study-specific and meta-odds ratio (OR) with 95% confidence intervals (CI) for the association between MC1R p.V92M (a) and p.R142H (b) variants and red hair.

Figure 4.

Study-specific and meta-odds ratio (OR) with 95% confidence intervals (CI) for the association between MC1R p.V92M (a) and p.D294H (b) variants and fair skin.

No publication bias was found for MC1R variants and phenotype.


This meta-analysis helps elucidating the role of the 9 most studied MC1R variants in melanoma development and in RHC phenotype determination.

The 5 variants p.D84E, p.R142H, p.R151C, p.R160W and p.D294H were associated both with melanoma development and RHC phenotype (Table IV). Earlier studies have already suggested that the 3 most frequent R variants p.R151C, p.R160W and p.D294H were significantly associated with red hair and fair skin15–20, 22 and with melanoma risk.17, 20, 26, 27, 35–37 We also found a significant association with melanoma and RHC phenotype for the less frequent R variants p.D84E and p.R142H. The association of p.D84E and p.R142H with melanoma was controversial: in some studies these variants were found significantly associated with melanoma risk (3 studies for p.D84E20, 23, 38 and 3 for p.R142H17, 20, 46) but other studies (2 for p.D84E17, 26 and 1 for p.R142H28) did not show a particular relationship between these variants and melanoma. The controversial results could be explained with the fact that these variants are less common and therefore large sample sizes are necessary to reach powerful results. In vitro expression studies on p.D84E, p.R151C and p.R160W receptors revealed that they have reduced cell surface expression and a corresponding impairment in cAMP coupling. The p.R142H and p.D294H variants showed normal cell surface expression, but had reduced functional responses.14 This could explain why these mutations are common in individuals with red hair and fair skin. Therefore we can at least partly explain the association of these 5 MC1R variants with melanoma by pigmentary pathways: red hair and fair skin individuals are unable to increase melanin levels in the skin in response to high exposure to UV light and therefore increase the levels of pheomelanin, which is mutagenic and cytotoxic.47, 48 However, MC1R variants have been previously associated with melanoma risk independently of the pigmentary characteristics17, 19, 20, 23, 26, 27, 36 and in Caucasian populations with dark skin,26–28 suggesting that other nonpigmentary pathways could act in addition to the effects of MC1R on eumelanin synthesis.

Table IV. MC1R Variants According with their Association with Melanoma and Red Hair Phenotype
 Positive association with red hair
  • 1

    Also positively associated with fair skin.

Positive association with melanoma NoYes

The 2 variants p.I155T and p.R163Q were found to be positively associated with melanoma risk but not with red hair nor with fair skin (Table IV). These results suggest that, for these variants, melanoma risk could be mainly increased via nonpigmentary pathways. It is well documented that α-MSH has immunomodulatory and antiinflammatory functions,49–52 therefore the association between MC1R and skin cancer could be a result of inflammatory or immune mechanisms influencing tumorigenesis. Modulation of melanocyte growth, development and differentiation, and increased DNA damage possibly associated with production of reactive oxygen species are other hypothesized mechanisms contributing to MC1R carcinogenesis.11, 53, 54 Borderline association of p.I155T and p.R163Q and melanoma was previously observed,36, 37 while no association between p.R163Q and red hair was observed in previous studies.18–20, 22 We found a surprisingly inverse association between p.R163Q variants and red hair color. This could be explained with the fact that our analysis compared, for each variant, carrier with noncarrier subjects. This means that all the individuals carrying at least 1 R common variant but no p.R163Q variant were included in the reference group, and the inverse association could therefore reflect the higher association with red hair of R common variants compared with p.R163Q variant.

We found no positive association between p.V60L and p.V92M with melanoma or RHC phenotype (Table IV). As for p.R163Q, a significantly inverse association was observed for p.V60L, which could as well reflect a weaker association of this variant with RHC phenotype compared with the other common R variants. Previous studies on p.V60L mutation have found an association with blond/light brown hair15, 16 and a less clear association with red hair than that seen for some of the other variants,20, 34 although in vitro study showed that p.V60L had a decrease in functional ability to stimulate cAMP levels.14 p.V92M polymorphism showed no association with any hair or skin classification,15, 18, 24 in agreement with in vitro studies showing that this variant had a functional cAMP response similar to or higher than wild type.14 Conflicting data are available on the association of the p.V60L and p.V92M alleles with melanoma risk: a positive association was demonstrated in some studies (4 for p.V60L20, 26, 28, 29 and 2 for p.V92M36, 29) whereas no association was shown in other papers (2 for p.V60L17, 36 and 3 for p.V92M24, 28, 38).

For p.V60L and p.D294H variants and melanoma, we found a significant between-study heterogeneity, which seemed mainly due to higher ORs in southern than in northern European countries. A similar pattern was observed for p.I155T and p.R163Q variants (Supplementary Figure 1). A possible explanation could be the minor frequency of the common p.R151C and p.R160W variants in Mediterranean European countries than in northern ones12: while these 2 variants could be strongly associated with melanoma in fairer-pigmented populations, other MC1R variants, as p.V60L and p.D294H, could play a stronger role in those countries where they are less frequent. However, this source of heterogeneity needs to be further explored in future studies involving a larger number of fairly and darkly-pigmented Caucasians.

The highest AR in the population was observed for p.R151C variant, which was the MC1R variant with the strongest association with melanoma in previous studies.26, 37, 46 High attributable risk (AR = 4.54%) was observed also for p.R160W, while low risks were found for the 2 variants p.I155T and p.D84E (AR = 1.1 and 1.5%, respectively). Given the rarity of these 2 variants relative to others, they cannot explain a high proportion of melanoma cases, even if their association with melanoma seemed the strongest one (OR (95%CI) = 2.45 (1.32–4.55) and 2.40 (1.50–3.84), respectively). Attributable risks calculated in a previous study for p.R151C, p.R160W and p.D294H,26 were very similar to our findings.

To our knowledge, this is the first study that review published literature on MC1R and melanoma, taking into account also the relationship between MC1R and RHC phenotype, and try to obtain definitive conclusion on the role of the most common MC1R variants. The powerful approach of the meta-analysis let found significant association also for less common variants, as p.D84E and p.R142H that up to now were not found to be definitively associated with melanoma risk, probably due to the small sample sizes. Our results seemed to be robust according both to the type of genotype data and to the statistical methods used. The analysis performed on allele instead of genotype data provided similar results. Moreover, the analysis including the within and between studies variation and using random effects models fitted with maximum likelihood estimate provided the same punctual estimates. The 95% CI, however, were larger due to the fact that the maximum likelihood estimate method is more conservative than the DerSimonian–Laird method used in our main analysis.

A possible limitation of this study is the heterogeneous presentation of results in different studies, not always showing both allele and genotype frequencies. This heterogeneity creates problems in including all the identified studies in the same analyses and forces us to exclude allele-based studies from the main analysis. However, as previously explained, a sensitivity analysis with allele data was performed as well. The number of study subjects was low for most studies, and individuals who carried no MC1R variant were presented only in a few studies: this did not allow us to compare subjects carrying each MC1R variant with subjects carrying no variant. The association of each variant with melanoma or phenotype could be therefore underestimated due to the inclusion of other “at risk” variants in the reference group. On the contrary, the fact that unpublished data were not included in the study could overestimate the association if positive results had more chance to be published than negative ones. Although this could be a possibility, we found no evidence of publication bias for none of the studied variants. Finally, using only published data, we could not take into account the possible interaction of MC1R with other genes that could modify its effect, and it was not possible to study the independent effect of MC1R variants on melanoma stratifying by RHC phenotype. A pooled analysis of both published and unpublished data could be useful to clarify these issues.

In conclusion, the association of the 3 most common R variants p.R151C, p.R160W and p.D294H with melanoma and phenotype was confirmed by our analysis, and evidence of a significant association with melanoma and phenotype was also found for the 2 less frequent variants p.D84E and p.R142H. A possible role in melanoma development via nonpigmentary pathways was suggested for p.I155T and p.R163H variants. It would be worthwhile to further study MC1R variants and melanoma stratifying by hair color and skin type, to identify the specific contribution of each MC1R variants in melanoma development due to pigmentary and nonpigmentary pathways.