Melanoma is mainly a disease of light-skinned populations. The chief pigmentary traits associated with melanoma risk are fair skin, blonde or red hair, blue eyes, a skin that sunburns easily, tans poorly and freckles. These are highly correlated traits, however, and are known to be partially determined by the melanocortin 1 receptor, MC1R [4,5], some variants of which result in a defect in activation of melanin production [6]. Almost all GEM participants (90%) had MC1R genotypes measured: cases were more likely than controls to carry multiple and higher-risk MC1R variants [7].

All participants completed a self-administered calendar and questionnaire before a one hour telephone interview, and gave a buccal DNA sample [10]. Lifetime residential histories were recorded in the calendar and used for assigning levels of ambient UV exposure. The questionnaire asked for skin, hair and eye color in standard categories. The interview included questions on ethnicity, ancestry, propensity to burn on first sun exposure in summer, tanning ability on repeated sun exposure and sun exposure in decade years from 10 years of age.

Ambient solar UV irradiance for each participant was estimated as the annual average erythemally weighted UV at their place of residence at birth and in each decade of age. These values were model-based and supplied to GEM by the US National Center for Atmospheric Research (Dr Julia M Lee Taylor). To calculate the lifetime average annual UV irradiance for each participant, the value for UV irradiance at birth (age 0) was assigned to each year from birth to age 4, that at age 10 to each year from age 5 to age 14, and so on, using irradiance in the last decade year up to the exact age at diagnosis and dividing the total by age. Early life ambient UV was calculated as the average of UV irradiance at birth and age 10.

Recreational sun exposure in beach and water activities from age 15 was obtained from direct questions about participation between 9am and 5pm on at least 10 days in any year since leaving school, the years started and stopped, the frequency, usual outdoor hours per day and the seasons; a lifetime sum of hours was calculated. Although vacations in a sunnier climate and sunburns were included initially in the present analyses because of their known associations with MPM [9], we excluded them from this report because they appeared to add little information beyond the contribution of beach and water activities and ambient UV.

DNA was isolated from buccal cells and MC1R sequenced for 2,202 controls and 1,099 cases in the GEM study (90% of participants), in whom 85 unique variants were identified [7]; 88% of participants in the centers included in this report had MC1R genotyped. The current analyses follow Kanetsky et al’s [7] grouping in which people with two copies of the consensus wild type (con/con) were the reference category and there were two additional categories, one formed by grouping people with any red-hair associated (R) variant (R151C, R160W, D294H, D84E) in an ‘any R’ category (con/R, r/R, R/R) and those having only other variants into an ‘r’ category (con/r; r/r). For most analyses we examined MC1R variants in 2 categories as no R (con/con, con/r; r/r) and any R variants (con/R, r/R, R/R). Carriage of synonymous variants did not affect group assignment. When we examined the association of each MC1R variant with red hair, ORs were 3.0 or higher for each of 7 variants: the 4 R variants R151C, R160W, D294H, D84E had p values <0.01 and Y152X, 86insA, and 537insC had p values <0.09 and had been previously identified as significantly associated with melanoma risk [11] or likely to produce a non-functional transmembrane receptor [12]. The latter 3 variants were included in the R genotype in sensitivity analyses.

We included only participants who reported exclusively European ancestry because of the known phenotypic differences of other ethnic groups and their small numbers in GEM. Included in these analyses are 1,018 cases and 1,875 controls (2,893 GEM participants) who had MC1R variants genotyped, of whom 75 participants developed their first primary in the control accession period and their second primary in the case accession period. Since epidemiologic theory clearly indicates that these participants should be included as both cases and controls in the analysis [13], they were included as both [10].

ORs for MPM by phenotype were lowest for poor tanning ability, tendency to burn and any childhood facial freckling, intermediate for red hair, relative to other colors, and highest for fair skin and for the most sun sensitive pigment score, Q4; all p values were <0.05 (Table 1). Relative to black hair, the OR for red hair was 2.05 (95% CI 1.26–3.34), that for blond or fair 1.58 (95% CI 1.02–2.45), for light brown hair 1.46 (95% CI 0.95–2.25) and that for dark brown 1.14 (0.74–1.78) (p for trend=0.001). Eye color had no apparent effect on MPM risk. All models for phenotype were adjusted for age, sex, center, ancestry, and age*sex interaction.

The strongest associations of sun exposure with MPM were for any beach and waterside activities and > median average annual lifetime ambient UV (Table 1). In a multivariable model, any beach and waterside activities (OR=1.58; 95% CI 1.25–1.99) and having any R genotype (OR=1.34; 95% CI 1.11–1.62) were both statistically significantly associated with MPM (p <0.05) with pigment score, age, sex, center, ancestry and age*sex interaction included as covariates. The p values were >0.05 for addition individually to this model of the variables for ambient lifetime and early life UV. The ORs changed very little when the R genotype included the 3 variants Y152X, 86insA, and 537insC in addition to R151C, R160W, D294H, D84E.

There was some evidence that MC1R genotype modified the effects of beach and water activities (Table 2). The OR for MPM was higher for any activities in the absence of R variants (OR=1.94) than in their presence (OR=1.39) (p for interaction 0.08) (Table 2). Adding the 3 variants Y152X, 86insA and 537insC to the 4 variants R151C, R160W, D294H, D84E changed the ORs only slightly. There was little evidence of any similar modification of the effects of lifetime or early life ambient UV (Table 2).

We further examined modification of the association of sun exposure with MPM by MC1R genotype in two body site categories. The ORs for ambient early life UV were higher in those with no R (OR=4.23) than in those with any R (OR=1.04) for melanomas on the head and neck (p=0.01 for 2 way interaction) and significantly different by body site (p=0.01 for 3 way interaction), with no evidence of a similar interaction for MPM at other body sites. Lifetime ambient UV had a similar pattern of ORs but all p values for interaction were high. The higher OR of MPM with beach and water activities in those with no R variants was mostly for MPM on body sites other than the head and neck but not significantly different by body site (p=0.89 for 3 way interaction) (Table 3).

Risk for MPM of the head and neck increased strongly with increasing lifetime ambient UV across 4 exposure categories, the OR for Q4 was 5.30 (95% CI 1.44–19.58; p for trend 0.01); for early life UV it was 4.19 (95% CI 1.83–9.63; p for trend <0.001). For other body sites, ORs for ambient UV rose only to 1.75 and showed no apparent trend. The increase in ORs for MPM on the head and neck with increasing lifetime ambient UV was evident only in the absence of MC1R R variants. In people with no R variants the ORs for Q4 were 14.61 for lifetime ambient UV, 7.8 for early life ambient UV and 2.75 for beach and water activities, compared with ORs <2.0 when R variants were present (see Figure 1). There was little evidence of these divergent patterns of increasing MPM risk with increasing ambient UV for melanomas of other body sites. For beach and water activities, MPM risk did not increase consistently beyond the second exposure quarter (OR ~2) and there was little evidence that the pattern of increase differed by body site or by MC1R genotype (Figure 1).

Like MC1R genotype (Table 3), phenotype also appeared to modify the effects of sun exposure on risk of MPM of the head and neck (Table 4). This effect was most evident for beach and water activities: the OR was 3.74 in those with a pigment score less than or equal to the median and 0.87 in those with a greater than median score (p=0.04 for 3 way interaction) (Table 4). The OR estimates in Table 4 changed little when MC1R genotype (no R, any R) was added to the model. The ORs in 4 quarters of beach and water activities for MPM of the head and neck rose with increasing exposure to Q3 in the less sun sensitive; ORs for the more sun sensitive were low (see Figure 1).

Our findings are consistent with a report on sun exposure modification of MC1R effects in which risk of melanoma for R variants was increased threefold for lower sun exposure but not all for higher exposure [16]. Other studies found no evidence of modification of MC1R effects by sun exposure [17–19] or did not report on sun exposure separately [20–22]. No study apart from our own has directly examined MC1R genotype modification of sun exposure effects.

In the few studies exploring how phenotype modifies the effects of sun exposure, melanoma risk with high sun exposure was greater with more sun sensitive (summary ORs 2.4 to 4.0) than less sun sensitive phenotypes (summary ORs 1.1 to 1.5) [23]. In our data, an increase in melanoma risk with higher sun exposure was generally more evident in the less sun sensitive phenotype, especially on the more continuously sun-exposed sites of the head and neck (Table 4), the opposite of the pattern in the earlier studies [23]. Our findings, however, are supported by more recent studies reporting a greater relative risk of melanoma with cumulative time outdoors in those who usually develop a deep tan than in those who do not [24], a stronger association between sunbed use and melanoma in people with dark than light or red hair [25], and that severe sunburn was associated with an increased risk of melanoma for a sun-resistant but not a sun-sensitive phenotype (Cust et al unpublished data).

That the effect of pigmentary phenotype appeared independent of MC1R genotype suggests that other genes that determine pigmentary or sun sensitivity phenotype also modify sun exposure effects on melanoma risk. Other gene variants consistently associated with pigmentary phenotype [26,27] and with melanoma [28–31] may have such a role. There are, however, no reports in which the effects of the interactions of other gene variants with sun exposure on melanoma risk have been studied.

If our findings were confirmed in future studies, it would suggest that MC1R or other sun sensitivity genotypes and pigmentary phenotype are modifiers of the effects of sun exposure on risk of melanoma, especially on more continuously sun-exposed skin. Further, it could suggest that the relative increase in sun exposure within the observed ranges on risk of melanoma is greater with lower than higher sun sensitivity. One potential explanation may be a ‘saturation’ effect in which genotype or phenotype modifies the risk associated with high doses of UV exposure. Under this model, more sun-sensitive people may reach a limit beyond which sun exposure does not increase their risk further while those less sensitive show a continuing rise in relative risk across the observed range of exposures. That effect modification was observable on the head and neck may suggest the effect was limited to one of the possible causal pathways to melanoma under the divergent pathway hypothesis, a pathway in which high levels of more continuous cumulative sun exposure drive the proliferation of epidermal melanocytes to cancer [32].

Differences in sun protection behaviour between more sensitive and less sensitive people are another possible explanation. People with sun-sensitive phenotypes adapt their behavior to make greater use of sun protection when exposed [33], an adaptive behaviour that might flatten the increase in melanoma risk with increasing sun exposure. It may occur before or after, and possibly as a consequence of, a first diagnosis of sun-related neoplasia or pre-neoplasia [23].

We also observed that the associations of red and other fairer hair colors with melanoma may be confined to people with MC1R variants. This observation was unexpected. Adding the red hair phenotype to its main genetic determinants should have had little additional effect on melanoma risk. It was consistent, however, with addition of risk of melanoma by the R151C MC1R variant (which is strongly correlated with red hair) in red haired women [20].

It is important to consider that some observed associations or interactions may have arisen by chance, given the number of gene-environment combinations that we explored and that the p values for interaction were not very low. On the other hand, that findings in other studies [24,25] [Cust et al unpublished data] were in the same direction as ours argues against it being due simply to chance. The GEM study relied on self-reported pigmentary phenotype and sun exposure history which could produce both non-differential and differential misclassification and lead to bias. Differential recall bias, however, would be minimised for the measurements of ambient UV irradiance at places of residence used in our analyses, since they offer a largely objective measure of potential for sun exposure at each decade of age. In addition, differential recall depending on a complex mix of case or control status, sun sensitivity and melanoma site would be required to produce the patterns we observed. Finally, studies on the reproducibility of standard sun exposure questions suggest reasonable reliability of participant responses [34,35] and agreement with histologic assessment of solar skin damage [36]. Other strengths in GEM were population-based ascertainment of participants, that participation was unlikely to be related to MC1R genotype since observed genotype frequencies were in the range of previous studies (see Kanetsky [7]), and that relative risk estimates in GEM for skin type, skin color, eye color and recreational sun exposure were consistent with a meta analysis [3,9,37,38].