Additional Material in This Document;

Davies et al Supplementary Information

Additional material in this document;

1)Supplementary methods (with references)

2)Supplementary Table S1

3)Supplementary Table S2

Supplementary methods

Genotyping of MC1R

The coding sequence of the single exon MC1R gene together with approximately 100bp flanking sequence was sequenced bi-directionally using four overlapping amplicons. The primers used were: ex1F 5’-GAGGCCTCCAACGACTCCTT; ex1BF 5’-CGCTACCACAGCATCGTGAC; ex1BR 5’-GTCACGATGCTGTGGTAGCG; and ex1R 5’-GGGAGGTGGTGATATTGTGTGG.

Derivation of explanatory covariates

To derive a proxy measure for sun-sensitivity, factor analysis was applied to six correlated variables: hair color, eye color, self-reported freckling as a child, propensity to burn, ability to tan and skin color on the inside upper arm as described previously[1]. Of the 954 cases and 513 controls used in factor analysis, 92 cases and 19 controls had at least one of the six variables missing, and multivariate imputation was used to impute incomplete data for the factor analysis. The estimated first factor scores were averaged over the five imputation sets produced and this average was used as a proxy for sun sensitivity. The median score was used as the cut-off point to partition participants into sun-sensitive and non-sun-sensitive phenotypes.

Sun exposure measures were derived from the questionnaire data using the most recent reported record of sun exposure frequency and duration provided at interview. These data were then divided into tertiles based upon their distribution in the population controls. The percentage coverage of freckling on the shoulders, arms and face (taken as an overall measure of freckling), the percentage coverage of freckling on the shoulders only, and the total count of nevi were classified into thirds based on the distribution in the control population. Data on self-reported significant sunburns (defined as causing pain for two or more days) were dichotomised as ever/never reporting sunburn, both before the age of 20 years and after the age of 20 years. Recent sunscreen usage is reported as not used at all (None), usage of a sunscreen with a Sun Protection Factor (SPF) less than 10 (SPF Low) or usage of a sunscreen with an SPF higher than 10 (SPF High). For each participant, the body mass index (BMI) was derived from self-reported height and weight, using the formula kg/m2.

Nonsynonymous mutations in the melanocortin 1 receptor gene (MC1R) were classified as “R” alleles (D84E, R151C, R160W and D294H) or “r” alleles (V60L, V92M, R142H and R163Q) depending on the strength of their association with the red hair phenotype as described in Duffy et al.[2].

Statistical methodology

To investigate the association of sun exposure with vitamin D levels, aggregated sun exposure variables were classified into thirds (as above) and the mean vitamin D level calculated for each group. Estimates were adjusted as above for cases and controls and separately for cases alone. A linear test for trend was performed on the adjusted values in both analyses.

Adjusted estimates in the multiple linear least squares regression models used as a baseline the estimated vitamin D level of a 54 year old woman case with a BMI score of 25, living in an area with a Townsend score of 0 (neither deprived nor affluent), with blood sampled in winter and no vitamin D supplementation (where appropriate). In the control only group the baseline was calculated using the estimated vitamin D level of a population control instead of a case. For multiple linear regression modelling the aggregated sun exposure variables were first grouped into tertiles and then collapsed into a single variable (0,1,2).

The distribution of vitamin D levels was slightly positively skewed. We investigated the effect of removing outliers to see if these data have undue influence on the results, but found that the conclusions are broadly similar.

To investigate the effect of sun exposure, sun sensitivity and supplementation on vitamin D levels, raw vitamin D levels were plotted against average weekend sun exposure for: i) cases and controls ii) cases without supplementation and iii) cases with supplementation. Separate LOESS curves were fitted for the sun-sensitive and non-sun sensitive groups using the “lowess” routine in R 2.10.1.

For comparisons between cases, population controls and sibling controls, the Kruskal-Wallis test was used to perform a non-parametric test for significant differences in vitamin D levels and vitamin D levels stratified by season. Chi squared tests were used to test for differences in gender, sun sensitivity, BMI, Townsend score and recent sunscreen usage between the three groups. For age of diagnosis, a Chi squared test was not appropriate as small numbers were observed in one cell of the contingency table and so Fisher’s exact test was used instead. The Kruskal-Wallis test, chi-squared test and Fisher’s exact test were performed using the “kruskal.test”, “chisq.test” and “fisher.test” routines in R.

To investigate differences in vitamin D levels attributable to the rs2282679 and rs7944926 SNPs and possible interactions with sun exposure, LOESS curves were fitted to scatter plots of average weekend sun exposure against vitamin D levels of the cases. Two sets of LOESS curves were plotted for each SNP, firstly grouping subjects under a dominant model (0 rare alleles v 1+ SNP alleles), secondly grouping subjects by the three distinct genotypes (0 rare alleles v 1 rare allele v 2 rare alleles). Vitamin D levels were adjusted for BMI, age, season the sample was taken, sex, case control status and Townsend score. Pairwise LD measures D’ and r2 between rs3829251 and rs7944926 were calculated using the “pwld” routine in Stata version 10.

The coefficient of determination (r2) was calculated to measure the percentage of variance explained by each covariate levels of vitamin D. The explained variation was calculated separately for the cases, the controls and in the cases and controls combined. In each group, r2was calculated both for seasonally adjusted vitamin D levels and for levels of vitamin D adjusted for sex, age, BMI, case control status (where appropriate) as well as season. Supplementation data was available for the cases only; a separate multiple linear regression model was fitted in this subset that additionally adjusted for vitamin D supplementation.

References

1.Newton-Bishop J, Chang Y-M, Elliott F, Chan M, Leake S, Karpavicius B, et al. (2011) Relationship between sun exposure and melanoma risk for tumors in different body sites in a large case-control study in a temperate climate. Eur J Cancer. 47: 732-41.

2.Duffy DL, Box NF, Chen W, Palmer JS, Montgomery GW et al. (2004) Interactive effects of MC1R and OCA2 on melanoma risk pheontypes. Hum Mol Genet. 13:447-61.

Supplementary Table S1 Associations between nurses’ assessed freckling scores and serum vitamin D level at recruitment and patterns of sun exposure

Serum vitamin D level and sun exposure / Freckling scores
Face / Arms / Shoulders
Correlation§ / Partial correlation^ / Correlation§ / Partial correlation^ / Correlation§ / Partial correlation^
Serum vitamin D level (nmol/L) / 0.11 (0.0003) / 0.12 (0.0006) / 0.11 (0.0004) / 0.13 (0.0002) / 0.15 (<0.0001) / 0.17 (<0.0001)
Average weekday exposure at cooler months (hours/day) / -0.06 (0.03) / -0.01 (0.73) / -0.06 (0.04) / -0.03 (0.41) / 0.00 (0.99) / 0.01 (0.84)
Average weekday exposure at warmer months (hours/day) / -0.00 (0.92) / 0.02 (0.53) / -0.02 (0.42) / 0.02 (0.62) / 0.04 (0.18) / 0.06 (0.10)
Average weekday exposure (hours/day) / -0.03 (0.35) / 0.01 (0.79) / -0.04 (0.18) / -0.00 (0.97) / 0.02 (0.47) / 0.04 (0.26)
Average weekend exposure at cooler months (hours/day) / -0.05 (0.07) / 0.00 (0.99) / -0.00 (0.95) / 0.02 (0.49) / 0.09 (0.001) / 0.08 (0.029)
Average weekend exposure at warmer months (hours/day) / 0.01 (0.73) / 0.06 (0.10) / 0.03 (0.35) / 0.04 (0.23) / 0.12 (<0.0001) / 0.11 (0.0018)
Average weekend exposure (hours/day) / -0.02 (0.38) / 0.03 (0.40) / 0.01 (0.64) / 0.04 (0.27) / 0.11 (<0.0001) / 0.10 (0.0036)
Average daily exposure (hours/day) / -0.03 (0.32) / 0.02 (0.57) / -0.02 (0.46) / 0.02 (0.56) / 0.06 (0.04) / 0.07 (0.046)

Supplementary TableS1(Continued)

Serum vitamin D level and sun exposure / Freckling scores
Face / Arms / Shoulders
Correlation§ / Partial correlation^ / Correlation§ / Partial correlation^ / Correlation§ / Partial correlation^
Average holiday exposure (hours/year) § / 0.12 (<0.0001) / 0.11 (0.0015) / 0.08 (0.008) / 0.06 (0.10) / 0.13 (<0.0001) / 0.12 (0.0005)
Average holiday exposure from 10am to 2pm (hours/year) / 0.11 (0.0004) / 0.11 (0.0014) / 0.06 (0.04) / 0.05 (0.17) / 0.11 (0.0004) / 0.10 (0.0071)
Average holiday exposure below 45N (hours/year) / 0.11 (<0.0001) / 0.10 (0.0043) / 0.06 (0.04) / 0.04 (0.26) / 0.18 (<0.0001) / 0.17 (<0.0001)
Average holiday exposure below 45N from
10am to 2pm (hours/year) / 0.11 (0.0002) / 0.11 (0.0023) / 0.05 (0.12) / 0.04 (0.33) / 0.17 (<0.0001) / 0.16 (<0.0001)
Regression coefficient / Covariate adjusted^ / Regression coefficient / Covariate adjusted^ / Regression coefficient / Covariate adjusted^
Sunburn before age 20* / 6.66 (<0.0001) / 6.30 (<0.0001) / 6.44 (0.0017) / 5.71 (0.0044) / 7.36 (0.0003) / 6.81 (0.0008)
Sunburn after age 20* / 3.46 (0.01) / 2.73 (0.0402) / 6.48 (0.0014) / 5.24 (0.0080) / 7.03 (0.0005) / 6.32 (0.0016)

§Spearman’s correlation (p-value) between serum vitamin D levels or self-reported sun exposure patterns and freckling score.

^Spearman’s partial correlation (p-value) or regression coefficient of freckling score on sunburn (p-value) after adjusting for hair color in the model.

*Regression coefficient (p-value) of freckling score on sunburn.

Supplementary table S2Association between sun exposure and serum vitamin D levels at recruitment.

n / Crude mean (SD)
Cases, sibling controls & population controls / n / Adjusted mean* (SE)
Cases, sibling controls & population controls / P trend§ / n / Adjusted mean** (SE)
Cases only / P trend§
Sunburn before age 20
Never / 742 / 54.8 (21.4) / 726 / 46.3 (0.74) / 0.8 / 488 / 41.6 (0.89) / 0.8
At least once / 363 / 53.5 (22.7) / 347 / 45.9 (1.11) / 249 / 41.2 (1.21)
Sunburn after age 20
Never / 666 / 55.4 (22.3) / 652 / 47.0 (0.82) / 0.4 / 440 / 42.2 (0.96) / 0.6
At least once / 451 / 53.9 (21.5) / 435 / 45.9 (0.94) / 313 / 41.5 (1.04)
Weekday exposure in cooler months (hours/day)
0 / 442 / 55.3(22.0) / 433 / 45.8(1.00) / 0.08 / 294 / 41.1(1.13) / 0.1
0-1.2 / 374 / 52.5(21.6) / 366 / 44.5(1.04) / 258 / 40.1(1.10)
>1.2 / 345 / 55.8(22.0) / 334 / 48.7(1.10) / 228 / 44.1(1.35)
Weekday exposure in warmer months(hours/day)
0 / 408 / 56.0 (22.0) / 397 / 46.0(1.03) / 0.1 / 265 / 41.2(1.13) / 0.2
0-2 / 479 / 52.4 (21.9) / 465 / 45.0(0.94) / 319 / 40.1(1.05)
>2 / 274 / 56.2 (21.6) / 271 / 49.0(1.22) / 196 / 44.1(1.44)
Weekend exposure (hours/day)
≤ 2.5 / 452 / 50.6 (21.6) / 444 / 42.8 (0.97) / <0.0001 / 322 / 39.5(1.21) / 0.0002
2.5-4 / 447 / 55.6 (21.1) / 433 / 47.3 (0.93) / 307 / 40.3(1.10)
>4 / 271 / 59.8 (22.7) / 264 / 50.5 (1.27) / 159 / 44.4(1.26)
Weekend exposure in cooler months (hours/day)
≤ 2 / 650 / 52.1(21.5) / 635 / 44.3(0.80) / <0.0001 / 461 / 40.0(0.87) / 0.002
2-3 / 216 / 57.1(22.1) / 208 / 47.4(1.36) / 145 / 41.0(1.61)
>3 / 305 / 58.4(22.0) / 299 / 49.9(1.18) / 152 / 45.6(1.49)
Weekend exposure in warmer months (hours/day)
≤ 3 / 494 / 50.4(21.2) / 485 / 42.5(0.89) / <0.0001 / 349 / 38.2(0.97) / <0.0001
3-5 / 443 / 56.8(21.5) / 430 / 48.5(0.98) / 287 / 43.8(1.18)
>5 / 233 / 59.4(22.8) / 226 / 50.4(1.35) / 152 / 44.7(1.55)
Daily exposure(hours/day)
≤1.14 / 366 / 52.8(22.0) / 361 / 43.9(1.08) / <0.0001 / 250 / 39.5(1.21) / 0.004
1.14-1.96 / 414 / 54.3(22.1) / 397 / 45.8 (1.01) / 274 / 40.3(1.10)
>1.96 / 378 / 56.5(21.6) / 372 / 49.0 (1.04) / 255 / 44.4(1.26)
Holiday exposure (hours/year)
≤44 / 442 / 51.5(22.6) / 429 / 43.6(1.03) / <0.0001 / 299 / 39.2(1.17) / <0.0001
44-98 / 420 / 53.8(20.5) / 413 / 45.8(0.90) / 296 / 40.7(0.98)
>98 / 316 / 60.2(22.0) / 304 / 51.1(1.18) / 196 / 46.5(1.46)

Supplementary Table S2(Continued)

N / Crude mean (SD)
Cases, sibling controls & population controls / N / Adjusted mean* (SE)
Cases, sibling controls & population controls / P trend§ / N / Adjusted mean** (SE)
Cases only / P trend§
Holiday exposure from 10am to 2pm below 45N(hours/year)
0 / 566 / 51.4(21.2) / 551 / 43.1(0.84) / <0.0001 / 383 / 38.5(0.94) / <0.0001
0-44.8 / 336 / 55.3(21.8) / 330 / 47.6(1.11) / 231 / 42.3(1.30)
>44.8 / 275 / 60.6(22.3) / 264 / 51.9(1.26) / 176 / 47.6(1.44)
Holiday exposure from 10am to 2pm (hours/year)
≤21.9 / 449 / 50.9(22.0) / 436 / 43.0(0.98) / <0.0001 / 308 / 38.3(1.05) / <0.0001
21.9-56 / 550 / 55.6(21.5) / 537 / 47.4(0.85) / 375 / 42.8(0.99)
>56 / 178 / 61.3(21.6) / 172 / 51.8(1.55) / 107 / 47.0(1.98)
Holiday exposure below 45N (hours/year)
0 / 552 / 51.0(21.0) / 538 / 42.7(0.84) / <0.0001 / 375 / 38.2(0.94) / <0.0001
0-87.5 / 321 / 55.9(21.7) / 314 / 47.5(1.13) / 219 / 42.2(1.29)
>87.5 / 305 / 60.0(22.7) / 294 / 51.8(1.21) / 197 / 47.4(1.41)

*Adjusted means of serum vitamin D levels were corrected for age, sex, month sampled, BMI, case-control status and Townsend score. Adjusted estimates used as a baseline the estimated vitamin D level of a 54 year old woman case with a BMI score of 25, living in an area with a Townsend score of 0 (neither deprived nor affluent), whose blood was sampled in winter.

** Adjusted means of serum vitamin D levels were corrected for age, sex, month sampled, vitamin D supplementation, BMI and Townsend score. Adjusted estimates used as a baseline the estimated vitamin D level of a 54 year old woman case with a BMI score of 25, living in an area with a Townsend score of 0 (neither deprived nor affluent), whose blood was sampled in winter and who has not taken vitamin D supplementation.

§ P-values from linear trend test, values in bold if P < 0.05.