INTRODUCTION

Melanocortin 1 Receptor (MC1R) is a G-protein coupled receptor (GPCR) belonging to a family of five highly related melanocortin receptors [1, 2]. MC1R is expressed in the cutaneous melanocytes, located in the basal layer of the epithelium [3, 4]. Melanocytes synthesize two types of melanin: black/brown eumelanin and yellow/red pheomelanin, the balance and amount of which determine skin and hair color [5,6,7,8]. When stimulated by its agonist α-melanocyte-stimulating hormone (α-MSH) in response to UV radiation, MC1R stimulates the production of cAMP and results in the production of eumelanin, which has been shown to be photoprotective [9, 10].

MC1R consists of 317 amino acids forming 7 transmembrane domains, an extracellular N-terminus, and an intracellular C-terminus (Supplemental Fig. 1). The C-terminus of MC1R is rather short and has a cysteine at position 315 that is palmitoylated [11]. Hundreds of protein-altering genetic variants in MC1R have been reported [12,13,14]. Some MC1R variants are associated with pale skin and red hair [15, 16] and an increased risk of melanoma and other skin cancers [17,18,19,20,21]. MC1R variants have been designated as low penetrance “r” or high penetrance “R” for red hair color (RHC) [22] and those with “R” designated variants are believed to be at higher risk of skin cancers [18, 23]. Among variants with MAF > 0.005, Val60Leu, Val92Met and Arg163Gln are designated as “r” alleles and Asp84Glu, Arg142His, Arg151Cys, Ile155Thr, Arg160Trp, and Asp294His are designated as “R” alleles showing variable penetrance [22], but Ile155Thr was later deemed a possible “r” allele [22, 24]. Individuals heterozygous or homozygous for “r” and “R” variants are all at an increased risk of cutaneous melanoma, nonmelanoma skin cancers or actinic keratosis independent of pigmentation, showing variable expressivity [25,26,27]. Both “r” and “R” variants of MC1R show a spectrum of functional defects in vitro with either reduced cell surface expression and cAMP production, normal expression but reduced cAMP response, or normal to elevated cAMP production compared to wild-type MC1R. Furthermore, some MC1R variants were found to act in a dominant-negative manner by reducing cell surface receptor expression and intracellular cAMP signaling (Asp84Glu, Arg151Cys, Ile155Thr, and Arg160Trp) or by only reducing cAMP signaling of coexpressed wild-type MC1R (Asp294His) [28].

Numerous studies have shown the association of MC1R variants with various cancers of the skin. However, many studies are limited by small case–control size [16, 26, 28], or unadjusted significance reporting (e.g., using p < 0.05 without adjusting for multiple comparisons) for larger genome-wide association pooled analysis studies [23] or only examining a single clinical phenotype [18]. Even larger genome-wide association studies (GWAS) have relatively small numbers (e.g., discovery population of 4,336 control and 1,650 cases and two replication cohorts of 964 case and 1,149 control and 903 case and 1,163 control) [29]. MC1R GWAS studies with a large number of participants have evaluated common genetic variant associations with different hair colors and do not evaluate for cancers of any type [30, 31]. Therefore, the spectrum of clinical phenotypes associated with MC1R variants remains largely unexplored.

Using exome sequencing (ES) data from the 135,947 participants of Geisinger-Regeneron DiscovEHR collaboration, we performed a phenome-wide association scan (PheWAS) in a discovery cohort of 38,155 unrelated individuals and replicated these findings in a cohort of 51,712 unrelated individuals for whom we had an average of 14 years of longitudinal clinical data in a well-maintained electronic health record (EHR) system. The longitudinal data combined with ES allowed an unbiased approach using all phenotypes captured in EHR mapped to 1,866 PheCodes (see “Materials and Methods”). We determined the phenotypes associated with both missense (amino acids altering) and nonsense (predicted loss of function [pLOF] due to start-loss, early termination or frameshift) MC1R variants, as well as in individuals with a copy-number variant (CNV) having either one or three copies of MC1R. We found associations with missense MC1R variants and PheCodes only in the dermatologic and neoplasm categories which correlate with the levels of functional defect in each variant. Remarkably, we found that nonsense variants are only weakly associated with milder skin phenotypes.

MATERIALS AND METHODS

Study population, clinical variables, and exome sequencing

The research protocol was approved by the Geisinger Clinic Institutional Review Board and included 135,947 participants in the MyCode Health Initiative who have ES data obtained as part of the Geisinger-Regeneron DiscovEHR collaboration. Patients are consented to participate in MyCode and DiscovEHR from all clinics throughout the health system. All clinics share a uniform EHR that has been in place for over 20 years. Basic demographic information for participants in this study can be found in Supplemental Table 1. All participants provided written informed consent, and all experiments were performed in accordance with relevant guidelines and regulations. The authors did not have access to any identifying information for the participants. The human phenotype and genotype data in this study were all de-identified by a data broker who was not involved in the study before any analysis was performed. De-identified clinical data were obtained from EHRs. Genomic DNA was isolated from patients’ blood or saliva. ES was performed in collaboration with Regeneron Genetics Center as previously described [32]. Probes from NimbleGen (VCRome, referred to as VCR henceforth) or a modified version of the xGEN probe from Integrated DNA Technologies (IDT) were used for target sequence capture [33, 34]. Sequencing was performed by paired end 75-bp reads on either an Illumina HiSeq2500 or NovaSeq. Coverage depth for all exome sites was sufficient to provide more than 20% coverage over 85% of the targeted bases in 96% of the VCR samples and 90% coverage for 99% of IDT samples. For MC1R, VCR samples had an average coverage of 30.6 at 93.5% of all sites in exon 3 (the coding exon) and IDT samples had an average coverage of 31.5 at 92.7% of all sites in exon 3. Alignments and variant calling were based on GRCh38 human genome reference sequence. Average read depth for sites with genetic variants (ref + alt) used in the analysis was 44.95 (range 39.8–52.8). The average allele balance (defined as alt / [ref + alt]) for variant sites was 49.8% (range 48.2–52.0). Nonsense or predicted loss-of-function (pLOF) variants are defined in this study as variants that cause a start-loss, frameshift, or early termination/stop-gain of the encoded protein.

Clinical traits, phenotype, and PheCode definitions

International Classification of Diseases Ninth (ICD-9) and Tenth (ICD-10) revision disease diagnosis codes were extracted from patients EHR. ICD codes were mapped to PheCodes using PheCodes Map 1.26 (https://phewascatalog.org/phecodes). For each individual, duplicate PheCode occurrences on the same date were dropped such that only one occurrence per date for a given PheCode remained. To ensure that individuals in the study were adequately assessed for clinical history during clinical care, we restricted the analyses to individuals who were cases for at least one phenotype, which was defined as patients who were diagnosed with that phenotype on at least three distinct clinical encounters. Patients with zero diagnoses were deemed controls, whereas patients with one or two diagnoses were excluded from analysis of the phenotype, i.e., they were neither case nor control.

PheWAS analysis

PheWAS was performed to evaluate the effects of nonsynonymous variants in MC1R with phenotypes encoded in EHR. First, second, and high-confidence third-degree relationships were removed using IBD estimates from Primus to obtain a maximal set of unrelated individuals as previously described [35]. Removing related subjects resulted in a discovery cohort of 38,155 individuals (sequenced by VCR) and a replication cohort of 51,712 individuals (sequenced by IDT) in the final analyses. We used a threshold of at least 0.1% cases per code for each population (51 cases for IDT-sequenced and 38 for VCR-sequenced) to be included in the model. Associations were calculated using Firth logistic regression adjusted for age, sex, and ancestry using the first three principal components. Accounting for ten principal components provided almost identical results. The analyses were performed assuming an additive genetic model; that is, we assume the risk due to an alternate allele is increased by r for heterozygotes and 2r for homozygotes. A circular plot of associations across the analyses for all variants was generated using Circos.

SKAT-O

For rare nonsense variants where single locus PheWAS was not feasible, we performed Optimized Sequence Kernel Association Test (SKAT-O) analysis [36]. SKAT-O was used to examine all nonsense variants (early terminations, frameshifts, and start-loss) except Asn29LysfsTer14, which was analyzed using PheWAS since there were sufficient number of subjects with this variant. Analysis was performed using the Robust SKAT-O method from SKAT version 2.0.0 in R. The analyses were performed assuming an additive genetic model. as described above.

Molecular biology

Untagged human MC1R or N-terminal 3× HA tagged MC1R in pcDNA3.1+ were purchased from cDNA.org. Individual amino acid substitutions were made with the Quickchange site-directed mutagenesis kit (Stratagene). All constructs were confirmed by sequencing of the full-length clone.

Cell culture and transfection

HEK293 cells (ATCC, Manassas, VA, USA) were cultured in MEM with 10% FBS at 37 °C and 5% CO2. For transient transfections, cells were transfected with plasmids described above by Xtremegene (Roche, Indianapolis, IN, USA) and used two days post transfection. For the cAMP pGlo and enzyme-linked immunosorbent assay (ELISA) assays, cells were transfected in one batch and then split for use in each assay. HEK293 cells stably expressing pGloSensor-20F cAMP plasmid (Promega) under Hygromycin selection were transfected with HA-MC1R, or HA-MC1R harboring one of the variants, in wells of a 6-well dish. One day post transfection, approximately 10,000 cells per well were added to white-bottom (cAMP pGlo Assay) or clear poly-L-lysine coated (ELISA) 96-well dishes.

cAMP pGlo assay

Two days post transfection, the media was carefully removed from the white-bottom plate and replaced with media containing 2% GloSensor cAMP reagent (Promega) and incubated at 37 °C for 2 hours. Cells were stimulated with MC1R agonist α-MSH (0.01–300 nM) or with 100 µM L-850851, a water soluble forskolin analog (to determine maximum cAMP), for 10 minutes and then the luminescence was read on a Spectramax M3 plate reader (Molecular Devices). Basal cAMP luminescence was subtracted, and cAMP values plotted as a percentage of the maximum cAMP measured for cells transfected with each MC1R variant. Data shown are from three independent experiments (mean ± SEM). Significant differences from wild type were determined using one-way analysis of variance (ANOVA) with Dunnett’s post hoc.

ELISA

Two days post transfection, cells plated on clear poly-L-lysine coated 96-well plates were washed with PBS and fixed with either methanol (for total expression) or 4% paraformaldehyde (for surface expression). Cells were then blocked with 1% milk and incubated in peroxidase conjugated anti-HA antibody. The plate was washed with TBS-T three times and then incubated with 100 µL 3,3′,5,5′-Tetramethylbenzidine Liquid Substrate (Sigma, St. Louis, MO, USA) for 30 minutes. Then, 100 μL of 1 mol/L sulfuric acid was added to each well to stop the reaction. Absorbance was then read at 450 nm on a Spectramax M3 plate reader (Molecular Devices). The absorbance from untransfected cells was subtracted and then the cell surface labeled signal was plotted as a percentage of total signal (calculated as the nonpermeabilized signal divided by the permeabilized signal × 100) and plotted. Total expression as a percentage of wild-type HA-MC1R for each experiment was also plotted. Data shown are from three independent experiments (mean ± SEM). Significant differences from wild type were determined using one-way ANOVA with Dunnett’s post hoc.

RESULTS

We have performed ES on 135,947 individuals using two different capture platforms: VCR and IDT. Patient DNA samples were collected from clinics within our integrated health system. However, all patient clinical data were captured in the same EHR. Among those with ES, we identified 89,867 unrelated individuals of whom 38,155 were sequenced with the VCR platform and 51,712 were sequenced with the IDT platform. We used the 38,155 unrelated individuals (VCR) as discovery and the subsequent 51,712 unrelated individuals (IDT) as a replication cohort for this study.

In the discovery cohort we found 158 nonsynonymous variants of MC1R consisting of 14 nonsense variants and 144 missense variants. In the replication cohort we found 199 nonsynonymous variants of MC1R, consisting of 20 nonsense variants and 179 missense variants. In totality, the 239 nonsynonymous variants of MC1R consisted of 24 nonsense and 215 missense variants (Supplemental Fig. 1 and Supplemental Table 2), of which 40 were novel variants (Supplemental Table 2) [12,13,14]. Ten variants (9 missense: Val60Leu, Asp84Glu, Val92Met, Arg142His, Arg151Cys, Ile155Thr, Arg160Trp, Arg163Gln, Asp294His and 1 nonsense: Asn29LysfsTer14) had minor allele frequency [MAF] > 0.005 and MAF > 0.004 respectively, sufficient for single variant PheWAS analysis where the reference (wild-type) alleles were ref, and each variant (separately) was the risk allele.

PheWAS analysis showed strong association of MC1R variants to PheCodes related almost exclusively to dermatologic and neoplasm categories (Fig. 1, Supplemental Fig. 2, Supplemental Tables 2–5) in both the discovery and replication cohort. Fig 1 shows a Circos plot of all PheWAS data highlighting significant associations solely to the dermatological and neoplasm categories, except for one significant association between the Ile155Thr variant and an endocrine/metabolic PheCode. More granular associations between MC1R missense variants and dermatologic and neoplasm PheCodes can be seen in Fig. 1b, c. Odds ratios calculated for phenotypes identified in PheWAS are shown in the heat map in Fig. 2 and Supplemental Tables 4 and 5. The strongest associations were found in the PheCode descriptions: melanomas of the skin, basal cell carcinoma, skin cancer, other nonepithelial cancer of skin, neoplasm of uncertain behavior of the skin, squamous cell carcinoma, actinic keratosis, fibrosis of skin, and degenerative skin disorders and scar conditions. PheCode for degenerative skin disorders includes condition within ICD709.3, which encompasses a range of skin disorders including calcinosis, colloid milium, skin degeneration, skin deposits, senile dermatosis, and subcutaneous calcification, it may also include patients with scars from previous procedures including skin cancer surgery. Of the 10 variants examined by PheWAS, only Ile155Thr was not significantly associated with any PheCodes in the discovery or replication cohorts (Figs. 1 and 2, Supplemental Fig. 2 and Supplemental Tables 4 and 5). Arg151Cys was significantly associated with all skin neoplasm and dermatologic PheCode descriptions listed above in both the discovery and replication cohorts (Figs. 1 and 2, Supplemental Fig. 2 and Supplemental Tables 4 and 5). All remaining variants tested were significantly associated with a number of PheCodes in both the discovery and replication cohorts.

Fig. 1: Phenome-wide association scan (PheWAS) for 10 common variants of MC1R, 9 missense variants, and 1 nonsense MC1R variant.
figure 1

(a) Circle graph showing associations for 1886 PheCodes mapped from the electronic health record (EHR). The dotted line represents the –log10 Bonferroni corrected p value for each variant (range is 4.2–4.28 for IDT-sequenced group and 4.29–4.55 for VCR sequenced group). Significant association were almost exclusively observed among dermatologic and neoplasm PheCode classes in the PheWAS analysis. (b, c) Manhattan plot for association of dermatologic and neoplasms of skin PheCodes with MC1R variants.

Full size image
Fig. 2: Heat map of odds ratios from phenome-wide association scan (PheWAS) of common MC1R variants shows the association of MC1R common variants with neoplasms of skin or dermatologic phenotypes.
figure 2

(a) PheWAS-VCR. (b) PheWAS-IDT. X = minimum number of observations not met (38 for VCR and 51 for IDT).

Full size image

To understand why some of the variants were significantly associated with these phenotypes and some were not, we decided to functionally assess each variant in vitro. Cells expressing MC1R, a Gαs coupled receptor, respond to α-MSH by producing cAMP. We evaluated MC1R variants for the ability to produce cAMP in response to α-MSH and for cell surface and total expression. Dose–response curves for cAMP response to α-MSH showed a variety of variant effects: for example, Val92Met was not different from wild type, whereas Asp84Glu had a right-shifted dose response and a lower maximum response. Fig 3a, b and Supplemental Fig. 3 show data for all nine amino acid substitutions tested. Calculated EC50 for α–MSH showed that the EC50s for Val60Leu, Asp84Glu, Arg142His, Arg151Cys, Ile155Thr, and Asp294His were significantly different compared to wild type MC1R (Fig. 3a). α–MSH induced-maximum cAMP was significantly lower than wild type for Asp84Glu, Arg142His, Arg151Cys, Ile155Thr, Arg160Trp, Arg163Gln and Asp294His (Fig. 3b). Importantly, in cells from the same transfection as those used in the cAMP assay, the cell surface and total expression of all variants tested were similar to wild-type MC1R (Fig. 3c, d). We plotted the odds ratios from the associated neoplasm and dermatologic phenotypes identified in PheWAS versus both the EC50s and maximum cAMP levels for the discovery (○) and replication cohorts (●) (Figs. 4 and 5). Higher EC50, i.e., reduced potency, corresponded to higher odds for these phenotypes (Figs. 4 and 5 and Supplemental Fig. 4), while lower maximum cAMP, i.e., reduced efficacy, corresponded to a higher odds ratio for these phenotypes (Figs. 4 and 5 and Supplemental Fig. 5).

Fig. 3: In vitro functional data for missense variants with MAF > 0.005.
figure 3

(a) Calculated EC50 for α-MSH for missense variants. The EC50s for Val60Leu, Asp84Glu, Arg142His, Arg151Cys, Ile155Thr, and Asp294His variants are significantly different compared to wild-type MC1R. (b) Calculated α –MSH induced-maximum cAMP (% of L-858051) for missense variants. The maximum cAMP for Asp84Glu, Arg142His, Arg151Cys, Ile155Thr, Arg160Trp, Arg163Gln and Asp294His variants are significantly lower than wild-type MC1R. (c) Cell surface and (d) total expression of common missense variants are similar to wild-type MC1R.

Full size image
Fig. 4: Relationship of neoplasms and functional consequences of common MC1R missense variants.
figure 4

(a) Plots of the odds ratio for various skin neoplasms versus EC50 for each common variant. The dotted line represents the EC50 for wild-type MC1R. (b) Plots of the odds ratio for various skin neoplasms versus maximum α-MSH induced cAMP for each common variant. Error bars represent 95% CI. The dotted line represents wild-type MC1R max cAMP (%L-858051). Symbols are color coded to match bars in Fig. 3.

Full size image
Fig. 5: Relationship of dermatologic clinical traits and functional consequences of common MC1R missense variants.
figure 5

(a) Plots of the odds ratio for dermatologic phenotypes versus EC50 for each common variant. The dotted line represents the EC50 for wild-type MC1R. (b) Plots of the odds ratio for dermatologic phenotypes versus maximum α-MSH induced cAMP for each common variant. Error bars represent 95% CI. The dotted line represents wild-type MC1R max cAMP (%L-858051). Symbols are color coded to match bars in Fig. 3.

Full size image

In addition to the missense variants in MC1R found in the discovery and replication cohort, we found nonsense variants in MC1R. We recently showed that for MC1R family member melanocortin 4 receptor, truncation before the s-acylated Cys318 results in a nonfunctional receptor [37]. All nonsense MC1R variants identified in our cohort occur before the palmitoylation site [11]; we therefore considered all of the nonsense variants true loss-of-function variants and included them in the analysis. Asn29LysfsTer14 was the only nonsense variant where the number of heterozygous and homozygous individuals was sufficient for PheWAS analysis. Asn29LysfsTer14 was significantly associated with actinic keratosis and marginally with skin cancer in both the discovery and replication cohorts (Figs. 1 and 2, Supplemental Fig. 2, Supplemental Tables 3–5). Since there was not a sufficient number of individuals with each of the other nonsense variant to perform PheWAS, we examined the remaining nonsense variants by robust SKAT-O, or optimized sequence kernel association test (Supplemental Table 6) [36]. Robust SKAT-O uses efficient resampling and saddle point approximation and aggregates the adjusted statistics to control for errors due to unbalanced case–control ratios. This analysis showed that nonsense variants were significantly associated with actinic keratosis (Supplemental Table 7) but not with neoplasms.

ES also revealed 21 individuals with CNVs of MC1R: 5 individuals with 1 copy of MC1R and 16 individuals with 3 copies of MC1R. Subjects with CNVs had few neoplasm or dermatologic phenotypes in their EHR (Supplemental Table 7). Interestingly, one of the subjects with an MC1R CNV deletion had an Arg151Cys variant in the remaining copy. This individual had neoplasm of uncertain behavior of skin. Seven subjects with MC1R duplication also harbored other variants in MC1R: 5 with Arg160Trp, 1 with Arg163Gln, and 1 with Met128Thr. Only one of the individuals carrying an MC1R duplication and an Arg160Trp variant had a neoplasm phenotype.

DISCUSSION

We have conducted, to our knowledge, the largest and most comprehensive study of MC1R genotype/phenotype relationship to date. We used exome sequencing and longitudinal clinical data from two cohorts with a total of >135,000 participants, with almost 90,000 unrelated participants used for association analyses, combined with in vitro assessment for function and expression of variants found in the sequencing data, to conclusively establish the scope and spectrum of effects for common MC1R variants in skin disorders and neoplasms. We provide novel data on loss-of-function and copy-number deletion variants to show a lack of strong association with severe phenotypes, as well as strong association of many relatively common missense variants with skin disorders and neoplasms. These data establish that individuals with missense MC1R variants that impair receptor function are at highest risk for the more severe neoplasms associated with skin.

Knowledge of MC1R genetic effects on diseases of the skin has primarily relied on case–control studies with relatively small number of individuals. Additionally, family studies could artificially enrich the influence of a particular genetic makeup on a phenotype. GWAS have identified some MC1R variants associated with melanoma [29] but are not well suited to determine associations with a wide spectrum of phenotypes. MC1R variants have previously been designated as low penetrance “r” or high penetrance “R” for red hair color with many studies grouping variants based on the R/r designation in their analyses. In this study we looked at the association of clinical traits with individual MC1R variants, not with R/r classifications of variants. Grouping variants whose encoded proteins function differently would not accurately capture the individual differences among these variants and their associated clinical phenotypes even if they have all been associated with the red hair color phenotype. Additionally, MC1R related skin cancers are independent from hair color [26, 27]. For example, each of the missense variants maintain some function while early frameshifts and terminations do not. Grouping variants into “R” and “r” groups does not allow for the nuanced associations we found when we examined these variants individually. Even among missense variants with similar cAMP responses and expression, Ile155Thr and Asp294His, we found significant differences in associations of each variant with phenotypes. Rare, loss-of-function variants however can be grouped together for analysis because they fail to express or function.

We established discovery and replication cohorts of 38,155 and 51,712 unrelated individuals for these analyses, eliminating the inherent bias that can occur in family studies. These individuals were not selected based on RHC phenotype or history of skin cancer as in other studies [13,14,15]. In totality we found 239 nonsynonymous variants in MC1R consisting of 24 nonsense and 215 missense variants including 40 previously unreported variants [38,39,40] (Supplemental Fig. 1 and Supplemental Table 2). Ten variants (9 missense: Val60Leu, Asp84Glu, Val92Met, Arg142His, Arg151Cys, Ile155Thr, Arg160Trp, Arg163Gln, Asp294His and 1 nonsense: Asn29LysfsTer14) had MAF > 0.004, sufficient for PheWas analysis. The remaining nonsense variants were grouped and evaluated by SKAT-O analysis due to the small number of heterozygotes for each variant.

In the discovery and replication cohorts MC1R variants Val60Leu, Asp84Glu, Val92Met, Arg142His, Arg151Cys, Arg160Trp, Arg163Gln, and Asp294His were significantly associated with actinic keratosis and skin cancer (odds ratios >1 and p values <4.46 ×10-5 − 6.31 ×10–5). Other nonepithelial cancer of the skin and degenerative skin conditions and other dermatoses were significantly associated with the variants Asp84Glu, Val92Met, Arg142His, Arg151Cys, Arg160Trp, Arg163Gln, and Asp294His. Ile155Thr has been categorized as an “R” allele and then later as a possible “r” allele [22, 24], but regardless of the categorization Ile155Thr has reportedly been associated with melanomas [17, 23]. Interestingly Ile155Thr was not significantly associated with any skin cancer or dermatologic condition in either cohort, contradicting earlier reports. In addition to skin cancer, other nonepithelial cancer of the skin, actinic keratosis, and degenerative skin conditions, those having the Arg151Cys variant were significantly associated with neoplasm of uncertain behavior of skin, melanomas of skin, squamous cell carcinoma, and basal cell carcinoma.

Functionally there were a variety of effects observed for these missense variants from almost normal functionality (Val92Met) to significantly impaired (significantly shifted EC50 and significantly reduced maximum cAMP) (Asp84Glu, Arg142His, Arg151Cys, Ile155Thr, and Asp294His). Interestingly, Ile155Thr significantly altered both the EC50 and maximum cAMP. When we examined odds ratios versus the EC50 or maximum cAMP for skin cancer phenotypes particularly for skin cancer or melanomas we observed that higher odds ratio correlated well with higher EC50 and lower maximum cAMP response.

The Asn29LysfsTer14 variant was significantly associated with actinic keratosis and skin cancer in both the discovery and replication cohorts, though the association with skin cancer was only marginally significant (Fig. 1b). SKAT-O analysis revealed that the individuals with other MC1R loss-of-function variants were not at increased risk of any skin cancers or diseases of the skin aside from actinic keratosis. Additionally, only two of five individuals with MC1R CNV deletion had neoplasm in their EHR, but one also had an Arg151Cys variant in the remaining copy, which is most likely driving the phenotype given the strong association of Arg151Cys variant with multiple skin and neoplasm phenotypes (Supplemental Table 8). Combined data from protein truncating and copy-number variants strongly suggest that a single functioning copy of MC1R is sufficient to protect from the more severe skin disorders associated with the MC1R missense variants that impair receptor function.

We have combined genetic data from a very large cohort with a phenotype agnostic approach to establish the scope and spectrum of MC1R genotype/phenotype relationship. We could not replicate previously reported association of Ile155Thr with various skin phenotypes, while all other missense variants with MAF > 0.005 were strongly associated with multiple skin and neoplasm phenotypes. Most importantly, we had 1,781 individuals with total loss-of-function variants (early terminations, frameshifts) and 5 individuals with copy-number deletion of MC1R due to large chromosomal deletions. These individuals effectively only have one copy of MC1R but showed surprisingly weak associations with skin phenotypes and almost no associations with neoplasms. Additionally, one individual who had a CNV deletion combined with a missense variant of MC1R had neoplasm of uncertain behavior of skin.

These data strongly suggest that a single functional copy of MC1R, in the absence of a missense variant in the other copy, is necessary and sufficient to produce enough cAMP to be photoprotective due to production of eumelanin. Additionally, heterozygous, and homozygous individuals for missense variants Asp84Glu, Val92Met, Arg142His, Arg151Cys, Arg160Trp, Arg163Gln, Asp294His, and to some extent Val60Leu are at a greater risk of skin neoplasms and other dermatoses than reference, but those with nonsense or CNV variants of MC1R are not. Our findings provide new generalizable guidelines for use of MC1R genetics in assessing risk of skin disorders, including skin cancer, independently of the red hair phenotype.