Molecular effects of indoor tanning

Abstract

Tanning bed users have a significantly increased risk of melanoma, but it remains unclear how indoor tanning drives melanomagenesis. To better understand the etiology of melanoma associated with tanning bed use, we described the patterns of melanoma in patients with quantifiable tanning bed usage and performed exome sequencing of melanocytes from normal skin of a subset of these patients. Tanning bed users were more likely to have melanoma on body sites with low cumulative levels of sun damage and to have multiple melanomas. The melanocytes in skin from tanning bed users had higher mutation burdens and higher proportions of cells with pathogenic mutations—these differences were most prominent over body sites that experience comparatively less exposure to natural sunlight. We conclude that tanning bed radiation induces melanoma by increasing the mutation burden of melanocytes and by mutagenizing a broader field of melanocytes than are typically exposed to natural sunlight.

The skin cells of tanning bed users have distinct mutational features as compared to people who only experience natural sunlight.

INTRODUCTION

Melanoma is responsible for an estimated 11,000 deaths annually in the United States (1). The main cause is exposure to ultraviolet (UV) radiation, which generates mutations in melanocytes, driving their transformation to melanoma. UV radiation naturally comes from sunlight but can also be delivered from artificial sources, such as tanning beds. Individuals with a history of tanning bed use are at higher lifetime risk of melanoma (2), and their melanomas occur at an earlier age (2, 3). The incidence of melanoma has been rising for decades, likely in part from increased screening (4). However, the rising incidence of melanoma has also coincided with an uptick in tanning bed usage and has disproportionately affected young women, the main clients of the tanning industry. Population-based studies have shown a spiked incidence of melanoma in young women following periods of increased tanning bed usage nationally, suggesting that some of the increase in melanoma among this demographic is attributable to indoor tanning (5, 6). On the basis of the available evidence, the American Academy of Dermatology opposes indoor tanning (7), and the World Health Organization (WHO) classifies tanning beds as a group 1 human carcinogen, similar to asbestos or cigarette smoke (8). Despite these warnings, 30 million people, including 2.3 million adolescents, use indoor tanning annually in the United States (9).

Popularity and support of indoor tanning is bolstered by our incomplete understanding of the effects of artificial UV radiation, providing opportunities for the tanning bed industry to market their product despite its link to skin cancer. For instance, the UV radiation that reaches the earth’s surface is 95% UVA and 5% UVB (10), while tanning beds typically have even higher proportions of UVA relative to UVB (9). The tanning bed industry argues that indoor tanning is safer than natural sunlight because UVB is more mutagenic than UVA (11); however, the absolute irradiance of tanning beds and outdoor sunlight is similar in the UVB range and 10 to 15 times higher in the UVA range than outdoor sunlight, counteracting their argument (12). Moreover, the tanning industry has marketed the prevacation tan as a safe way to photoadapt skin in anticipation of recreational exposure (13), supported by observations that the relative risk of melanoma is higher in people who have a history of intermittent, blistering sunburns than in outdoor workers, and experiencing daily UV radiation (8). However, people who burn easily tend to avoid occupations with intense sun exposure, and most outdoor workers wear protective clothing over their trunk and upper extremities, where melanoma is most common.

Considering the high stakes of tanning bed exposure, research is needed to illuminate the effects of tanning bed radiation on skin cells to better understand how tanning bed usage drives melanomagenesis. Toward this goal, we performed a case/control analysis of patients with/without a history of tanning bed use and a molecular analysis of skin cells from a subset of these patients.

RESULTS

Associations between tanning bed usage and melanoma

To better understand the relationship between melanoma and indoor tanning, we interrogated the medical records of 32,315 patients seen by the Dermatology service at Northwestern University (Fig. 1A). Among these patients, 7474 had a self-reported history of tanning bed usage. The extent of tanning bed usage was quantifiable for 2932 patients, forming our “epidemiological case” cohort. Among 24,841 patients with no history of tanning bed usage, we randomly selected a subset of 2925 patients, age-matched to the tanning bed cohort, to serve as the “epidemiological control” cohort. The tanning cohort was more likely to be female (87.7% tanning cohort versus 57% nontanning bed cohort) and has a history of sunburn (71.2% versus 50.3%) and a family history of melanoma (18.1% versus 16.8%) (table S1). The incidence of melanoma in our tanning bed cohort was 5.1% compared to 2.1% in the nontanning bed cohort (P < 0.0001, chi-square test). Multiple logistic regression analysis showed that tanning bed use was associated with an increased risk for melanoma (odds ratio of 2.85; 95% confidence interval 2.0, 3.99), after adjusting for age, sex, family history of melanoma, and sunburn history. Further, there was a dose-response relationship between the number of tanning bed exposures and relative risk for melanoma (P < 0.0001) (Fig. 1B).

Fig. 1. Tanning bed users are more likely than nonusers to have multiple melanomas on body sites with low cumulative sun damage.

(A) A case/control cohort was generated from patients seen at a high-risk skin cancer clinic at Northwestern University. Only patients with quantifiable tanning bed use were included in the case cohort, and an equal number of age-matched controls were randomly selected. (B) Unadjusted and adjusted odds ratio and 95% confidence interval (CI) of melanoma from logistic regression models evaluating associations with tanning bed (TB) usage, number of TB sessions, history (Hx) of sun burn, and family history of melanoma. Separate multivariable logistic regression models were fitted for adjusted TB usage and TB sessions. CIs that do not include 1 indicate statistical significance (P < 0.05). (C) The anatomic distribution of melanomas on body sites with low- or high-cumulative sun damage. T and C indicate the tanning and control cohorts. Asterisk indicates a P value less than 0.05 (Student’s t test). (D) The likelihood of previous tanning bed usage in patients, stratified by the number of melanomas that they have had. Asterisk indicates significant deviance from relative risk (RR) value of 1 (chi-square test).

Melanomas in tanning bed patients had a different anatomic distribution than melanomas from the control cohort. The WHO distinguishes melanomas that arise on skin with high cumulative sun damage versus melanomas on skin with low cumulative sun damage (14). Melanoma was more common on body sites with low cumulative sun damage in tanning bed users compared to nonusers (76.1% versus 61.2%; Fig. 1C). A previous study observed a similar anatomic distribution of melanoma in tanning bed users, although their case-control cohort was not matched for age (15). Multiple primary melanomas were also more common in tanners than nontanners (Fig. 1D).

On the basis of these observations, we hypothesized that indoor tanning increases the risk of melanoma in two ways: first, by increasing the mutation burden of melanocytes, and second, by mutagenizing a larger field of melanocytes, beyond the body sites that are typically exposed to natural sunlight, creating a broader field effect. To test this hypothesis, we compared the mutational landscapes of melanocytes from normal skin samples of tanning bed users to nonusers.

Molecular consequences of tanning bed usage

Shave biopsies of normal skin were collected from the lower backs or upper backs of 11 tanning bed users. Tanning bed users filled out a questionnaire, modeled after the UK Biobank, to assess past histories of sun exposure and other risk factors for skin cancer (16). They had self-reported histories of extreme tanning bed usage—ranging from 50 to more than 750 lifetime sessions, among other risk factors summarized in table S2.

Normal skin samples were also collected from two different control groups (Fig. 2A) for comparison. In the first control group, nine patients were recruited from the same high-risk skin cancer clinic as the tanning cohort and matched for age, sex, and risk profiles of the tanning cohort, based on responses to the UK Biobank questionnaire. The tanning cohort was more likely to have been diagnosed with melanoma (100% in the tanning cohort compared to 33.3% in the control cohort and 0.4% in the UK Biobank cohort) and to have a history of tanning bed usage, but no other risk factors for skin cancer were significantly different between the groups (table S2 and fig. S1A). While the tanning cohort and first control group were matched for common risk factors, when they were compared to the participants in the UK Biobank, both were more likely to have a history of melanoma (70% versus 0.4%), red or blonde hair (45.0% versus 15.1%), a history of sunburn (with 45.0% reporting more than five sunburn before the age of 15 versus 7.3%), and poor tanning ability (75% versus 17.3%) (fig. S1B), likely because they were recruited from the same high-risk skin cancer clinic.

Fig. 2. High mutation burdens in melanocytes from tanning bed users.

(A) An overview of the tanning and control cohorts. In (B) to (E), each data point corresponds to the mutation burden [measured in mutations per megabase (mut/Mb)] of an individual melanocyte with black bars indicating median mutation burdens. Asterisks denote P values less than 0.05 (Wilcoxon rank sum test). (B) A comparison of all melanocytes from tanning bed donors to all melanocytes from control donors. (C) A comparison of melanocytes from tanning bed donors to control donors, separately for each anatomic site. (D) A comparison of biopsy mutation burdens from tanning bed donors to control donors on the lower back. The mutation burden of each biopsy was calculated from the median mutation burden of its constituent melanocytes. (E) A comparison of biopsy mutation burdens from tanning bed donors to control donors on the upper back. NS, not significant.

For a second control group, we collected normal skin from six cadavers through the University of California, San Francisco (UCSF) Willed Body Program (Fig. 2A and table S2). We assumed that the cadaver tissue, which included two donors of dark skin tone, would be more representative of the general population than the donors recruited from a high-risk skin cancer clinic. A limitation to this control group is that donors were nearly twice the age of the tanning cohort (78.3 versus 43.6 years of age on average), and we were unable to interrogate the past histories of tanning bed usage from the deceased donors, although extreme tanning bed usage (>50 sessions) is rare in the general population.

After collecting normal skin samples, we measured somatic mutations at single-cell resolution from a total of 182 melanocytes derived across these donors (table S3). It is challenging to comprehensively and accurately detect mutations in an individual cell. Therefore, we developed a protocol to overcome this obstacle, as previously described (17). Briefly, we established melanocytes in tissue culture from epidermal biopsies and clonally expanded individual melanocytes to form small colonies, with a median of 213 cells per colony. DNA and RNA were extracted from each colony and further amplified in vitro. Exome and transcriptome sequencing was performed on the amplified DNA/RNA from each colony. We developed a bioinformatic workflow to root out amplification artifacts, permitting the detection of somatic mutations at high specificity. Mutation calls were internally benchmarked for each cell, in part, by assessing agreement in mutation support from DNA and RNA of the same cell within highly expressed genes, showing an average detection of 95.2% sensitivity and 97.1% specificity (table S3).

The mutation burden of melanocytes in tanning bed users was nearly twofold that of melanocytes from control donors [median mutations per megabase (mut/Mb) of 5.69 versus 2.86] (Fig. 2B). This trend remained significant when we separately compared melanocytes by anatomic site (median mut/Mb of 8.82 versus 4.60 in upper back and 5.19 versus 2.05 in lower back) (Fig. 2C). The mutation burdens of cells from both control groups were less than the cells in the tanning cohort (fig. S2A).

We developed a linear mixed-effects (LME) model to determine whether tanning bed users had higher mutation burdens after adjusting for anatomic site and outliers (18). Specifically, we compared the tanning and control cohorts at the cell level while introducing a variable for anatomic site and including a random effect for different donors (fig. S3). In this analysis, the cells from tanning bed users had a significantly higher mutation burden (P = 1.3 × 10⁻²). However, we noted that cells from one donor (donor 59) had an extremely high mutation burden. To ensure that this donor was not entirely responsible for the differences observed, the analyses were repeated with a robust linear mixed-effects (RLME) model, which is less sensitive to outliers, heavy-tailed distributions, and model violations (19). In this analysis, the cells from tanning bed users continued to have a significantly higher mutation burden (P = 6.1 × 10⁻⁵).

We also compared mutation burdens at the biopsy level. We defined the mutation burden of a biopsy as the median mutation burden of its constituent melanocytes. Our statistical power was more limited in these comparisons because there were fewer independent biopsies than individual cells. On the lower back, tanning bed biopsies had significantly higher mutation burdens than control biopsies (Fig. 2D and fig. S2B). The difference was not statistically significant on the upper back (Fig. 2E and fig. S2C), although cell-level differences were significant at this site (Fig. 2C). The upper back is the most common site of sunburn (20) and melanoma (21), indicating that natural sunlight can induce high mutation burdens at this site, even without the additional influence of tanning beds. Since tanning bed–induced mutations would be expected to contribute to a smaller proportion of mutations in the underlying skin cells from the upper back, we would need a larger sample size to detect a statistically significant difference at the biopsy level.

Given that tanning bed users experience higher doses and different blends of UV radiation than typically encountered from natural sunlight, we performed mutational signature analyses (22–24) to interrogate the types of mutations in their melanocytes. These previously published signatures were identified in independent datasets from the Pan-Cancer Analysis of Whole Genomes consortium and The Cancer Genome Atlas (24). The dominant mutational signature in cells from both tanning bed users and nonusers was signature 7 (95.1% versus 90.7%). Signature 7 is characterized by cytosine to thymine (C>T) mutations with a pyrimidine upstream of the mutant base pair (fig. S4A), and it has been attributed to UV radiation–induced damage (24). Cells from tanning bed users had higher proportions of mutations associated with signature 7, but the difference was not significant (Fig. 3B).

Fig. 3. Distinct types of mutations between melanocytes in tanning bed donors versus control donors.

(A) The top depicts the mutation burden (mut/Mb) of melanocytes with each column representing a single melanocyte, arranged in descending order for each cohort. The bottom shows the fractions of different mutation signatures for each melanocyte. (B) The left scatterplot shows proportions of signature 7 (SBS7) in melanocytes from the tanning cohort (T) and control cohort (C) (P value: 0.609, Wilcoxon rank sum test). The right bar graph compares the fraction of melanocytes with any detectable signature 7 in each cohort (P value: 0.6867, Poisson test). (C) Data are plotted as described in the previous panel but for signature 11 (SBS11) (P value: 0.0405, Wilcoxon rank sum test on the left; and P value: 0.00813, Poisson test on the right).

SBS11 (single base substitution signature 11) was the only signature to reach a statistically significant difference between the cohorts (14.7% versus 5.5%) (Fig. 3C). The presence of SBS11 was spread out relatively evenly among the tanning bed donors, whereas it was more concentrated among the control donors. In total, 8 of 11 tanning bed donors had at least one cell with SBS11. By contrast, 3 of 15 control donors had at least one cell with SBS11. SBS11 shows similarities to signature SBS7 in that it is characterized by C>T mutations, but in contrast to SBS7, there is a pyrimidine downstream of the mutant base pair (rather than upstream; fig. S4B). Tanning bed melanocytes had a higher, albeit modest, proportion of C>T mutations with a downstream pyrimidine (fig. S4C). In the current catalog of mutational signatures [Catalogue of Somatic Mutations In Cancer (COSMIC) v. 3.4; fig. S4D], melanoma is the most common tumor subtype associated with SBS11. The tanning history of melanomas included in the catalog of mutational signatures is unknown, but it is likely that a subset arose from patients with a history of indoor tanning. Signature SBS11 was originally attributed to temozolomide treatment, based on an anecdotal association with glioblastoma pretreated with temozolomide (23); however, in vitro studies have found that temozolomide induces a different mutational profile (25). Now, the etiology of SBS11 is unknown.

We also performed de novo mutational signature analysis (fig. S5). Two signatures were found. They were most similar to SBS7 (UV radiation) and SBS1/5/40a (aging), corroborating the main signatures observed in the deconvolution analyses. These analyses did not reveal an undiscovered signature associated with tanning bed usage.

Together, the types of mutations are similar in indoor tanners versus nontanners; however, there is a subtle difference in trinucleotide contexts of C>T mutations. Further studies will be needed to validate these subtle differences and determine whether they are associated with the unique blends of UV radiation experience by indoor tanners.

Building on these findings, we next examined specific pathogenic mutations present in individual melanocytes to assess potential drivers of melanoma. Melanocytes from physiologically normal skin can harbor these mutations. These cells are potential precursors to melanoma but require additional genetic alterations and/or microenvironmental stimuli to grow out and form clinically detectable neoplasms. We found 40 pathogenic mutations in 23 unique melanocytes (Fig. 4A). Most driver mutations, observed here, were predicted to activate the mitogen-activated protein kinase signaling pathway, in which loss-of-function mutations affecting the NF1 tumor suppressor gene were especially common. While we were underpowered to perform cancer gene discovery analyses on these cells by themselves, we did note signatures of positive selection—for example, hotspot mutations in oncogenes and a skew toward damaging mutations for NF1 (fig. S6). Melanocytes from tanning bed users were more likely to have a pathogenic mutation than melanocytes from control donors (23.0% in the tanning bed cohort versus 7.3% in the control cohort), and melanocytes from the upper back were more likely to have a pathogenic mutation than those from the lower back (22.4% in upper back versus 4.1% in lower back) (Fig. 4, B and C).

Fig. 4. Pathogenic mutations are common in melanocytes from tanning bed users.

(A) A list of pathogenic mutations observed, broken down by cohort, body site, donor, and cell. (B) The fraction of cells with pathogenic mutations in tanning bed users (T) or control donors (C). Asterisk indicates a P value less than 0.05 (Poisson test) with error bars showing 95% confidence intervals. (C) Data are plotted as in (B), but here, cells are broken down by anatomic site. (D) Four biopsies had two or more melanocytes with shared subsets of mutations. In these schematics, each dot represents an individual melanocyte, and phylogenetically related melanocytes are circled. Melanocytes with pathogenic mutations are noted in red with their driver mutations labeled. The true spatial localization of each cell, within biopsies, is unknown, but these schematics attempt to illustrate field sizes and clonal distribution of skin biopsies.

We also examined the trinucleotide contexts of pathogenic mutations and observed that 40% of unique pathogenic mutations in the tanning bed cohort were C>T (C/T) substitutions, compared to 25% in the control cohort (fig. S4E). Although the current sample size is too small to draw definitive conclusions, this trend suggests that a greater proportion of pathogenic mutations in the tanning bed cohort may exhibit a trinucleotide context generally associated with mutational signature 11.

Some melanocytes shared subsets of mutations, indicating a phylogenetic relationship (Fig. 4D). These melanocytes likely descended from fields of clonally related cells present in the skin. We did not directly detect a relationship between tanning bed usage and clonal structure; however, biopsies with high mutation burdens were more likely to have a field of melanocytes (13.1% cells in tanning bed cohort versus 5.8% cells in control cohort) (Fig. 4D and fig. S7).

In addition to mutational profiling, we examined whether tanning bed exposure was associated with changes in the melanocyte transcriptome. To minimize cell-level noise while preserving biological variability, sequencing data from individual colonies of melanocytes were aggregated to produce pseudobulk gene expression profiles of melanocytes from each biopsy. We identified 19 genes up-regulated in samples from the tanning bed cohort and 26 genes up-regulated in the control cohort. Many of the differentially expressed genes were long noncoding RNAs or pseudogenes of unknown significance (fig. S8). APOBEC3B, which is commonly active in cancers, was up-regulated in tanning bed melanocytes. However, more work will be needed to confirm whether tanning bed usage induces permanent changes to the transcriptional states of melanocytes.

DISCUSSION

Our study was motivated by a clinical presentation recurrently seen in dermatology clinics where tanning bed usage is especially high. Young tanning bed users, without a family history of melanoma, periodically develop multiple melanomas on body sites that receive low cumulative sun damage. The presentation of tanning bed–induced melanomas is reminiscent of familial melanoma (26). In families with a strong predisposition to melanoma, each person inherits one “hit” in their germline DNA (e.g., a CDKN2A mutation) bringing all their melanocytes one step closer to melanoma. Like tanning bed users, patients with familial melanoma also develop disease at a young age and are more likely to get multiple melanomas, and a larger field of melanocytes is at risk for transformation (26). We hypothesized that tanning beds mimic these circumstances by mutagenizing a large surface area of the body, beyond the sites typically exposed to natural sunlight. Body sites that typically receive low cumulative sun damage are intensely exposed during tanning sessions. This constitutes nearly 1.5× the body surface area of heavily sun-exposed skin (fig. S9) (27, 28), significantly increasing the skin surface at high risk of melanoma in tanning bed users (Fig. 5).

Fig. 5. High proportion of melanocytes in tanning bed users had pathogenic mutations, increasing risk of melanoma, particularly over body sites with low cumulative sun damage.

Young patients, with an extreme history of tanning bed usage, had more mutations in their cutaneous melanocytes than donors who were nearly twice their age from the UCSF Willed Body Program. Tanning bed users also had more mutations than demographically matched donors at similarly high risk of skin cancer. The difference in mutation burdens was most glaring on the lower back—a body site that receives comparatively less damage from natural sunlight but is exposed in tanning beds. In addition to having higher numbers of mutations, melanocytes from tanning bed users were more likely to have a pathogenic mutation than melanocytes from either control group (Fig. 5). Having a high fraction of melanocytes with pathogenic mutations would be expected to increase one’s risk of melanoma and potentially multiple primary melanomas by generating a large pool of precursor cells one step closer to transformation, akin to patients with heritable forms of melanoma. In our experience, among patients with multiple primary melanomas, tanning bed use is a more common etiologic factor than high penetrant germline mutations (29).

There were subtle differences in the types of mutations in melanocytes between tanning bed users and control donors. Tanning bed users were more likely to have a pyrimidine immediately downstream of the mutant site, indicated by a higher proportion of cells with signature 11 mutations. These differences may result from the unique blends of radiation experienced by tanning bed users, with much higher levels of UVA radiation than natural sunlight. However, more work is needed to understand the source of these mutations. One signature was notably absent. UVA radiation can induce a reactive oxygen species (ROS) signature (30), but these signatures were not observed here. It is possible that ROS-induced mutations are difficult to detect amid the large numbers of UV radiation–induced mutations. Alternatively, physiological levels of ROS may not reach mutagenic thresholds in vivo.

Few studies have sequenced melanomas from patients with documented tanning bed histories (15, 31). The limited data available suggest that BRAF^V600E mutations are more common in melanomas arising in tanning bed users. However, these analyses did not control for age. In these studies, the tanning bed users were younger, and younger patients are more likely to have BRAF^V600E-driven tumors (32). Conversely, in our study, we found that melanocytes from normal skin of tanning bed users most frequently harbor NF1 mutations, a driver typically associated with melanomas in older individuals (33). However, our finding is biased, too, because we focused on normal skin; had we instead sampled melanocytic neoplasms, we likely would have detected BRAF^V600E mutations. Together, these findings suggest that tanning bed radiation induces a broad spectrum of driver mutations. The earliest neoplasms to grow out are likely BRAF^V600E-driven, but melanocytes carrying other oncogenic alterations persist and can evolve into melanoma later in life.

Our work provides key insights to inform public health guidance on indoor tanning. Given the high levels of mutational damage in skin cells from tanning bed users, it is difficult to justify marketing claims that the spectra of UV radiation in tanning beds are safer than natural sunlight. Another popular claim by tanning advocates is that a prevacation tan can photoadapt skin in anticipation of recreational sun exposure. However, we see that tanning bed usage raises the mutation burden and risk of melanoma particularly in skin cells that receive low cumulative sun damage. In closing, tanning bed exposures are often thought of as a substitute for natural UV radiation despite differences in the maximum doses, UV content, body sites exposed, and patterns of melanoma that arise. Our work highlights unique ways in which tanning beds shape the mutational landscapes of skin cells, helping to explain the distinctive presentations of melanoma in this patient population.

MATERIALS AND METHODS

Epidemiological analyses

Under Northwestern University Institutional Review Board protocol STU00211546, patient records collected from the Department of Dermatology at Northwestern Medicine were evaluated using an EDW (electronic data warehouse) consolidated database based on the following inclusion criteria: aged 18 to 70 before 2019, a quantifiable history of tanning bed use, at least one dermatology visit before 2019, and at least one additional visit 3 or more years before or after the first visit. A total of 2932 patients met these criteria and were sorted into the positive tanning bed cohort. The control cohort included a random selection of 2925 age-matched patients with no prior history of tanning bed use after duplicate patient records were excluded (Fig. 1A). Several risk factors for melanoma were scored, as described below, based on self-reported data from each patient (tables S1 and S6).

For family history, we used binary variables in which any family history of melanoma in a first-degree relative was scored as a “yes,” and those patients with no first-degree family members with a history of melanoma were scored as a “no.” For sun burn, if the patient had any history of sun burns, this was scored as a yes, whereas those patients with zero history of sun burns were scored as a no. Sunburn was included in our study because a meta-analysis from 51 independent studies shows an increased risk of melanoma for “ever” sun burned (34).

For tanning bed use, any patients with a history of 10 or more tanning bed sessions were scored as a yes, while those patients with no history of tanning bed use were scored as a no. Patients with uncertain numbers of tanning bed use or reporting a number greater than 0 but less than 10 were eliminated from the analysis. For analyses shown in Fig. 1B, we also performed a separate dose-dependent assessment looking at the risk for melanoma with various numbers of sessions such as 10 to 50, 51 to 100, 101 to 200, and greater than 200 sessions.

We considered coding overall sun exposure as a risk factor. However, a significant number of patients had missing data regarding sun exposure history. Moreover, other studies have shown that sun exposure has a poor recall (35, 36).

Once all the risk factors were coded, a chi-square test was conducted to compare the prevalence of melanoma between individuals who used tanning beds and those who did not. To examine associations between all risk factors and melanoma, unadjusted logistic regression models were fitted separately for each risk factor including tanning bed use, number of tanning bed sessions, history of sunburns, family history of melanoma, age, and sex. Separate multivariate logistic regression models were then fitted for tanning bed use (yes/no) with other risk factors and tanning bed sessions (quantified) with other risk factors to assess the independent effects of these factors. SAS version 9.4 was used to perform logistic regression analysis described above. P < 0.05 was considered statistically significant and denoted by * (Fig. 1, C and D).

Skin biopsy collection for molecular analyses

Donors were recruited from the Northwestern Dermatology high-risk skin cancer clinic for the tanning cohort and first control cohort (see Fig. 2A and table S2 for an overview of all sample cohorts used in molecular analyses). For the tanning cohort, 11 donors were recruited, and each had greater than 50 self-reported tanning sessions. No other inclusion criteria were put in place, and since tanning usage is most common in young women, the cohort skewed toward this demographic. We also asked the donors in the tanning cohort to fill out a questionnaire related to their previous exposures to UV radiation and risk factors for skin cancer. To generate the questionnaire, we identified the subset of questions from the UK Biobank that related to skin cancer risk and asked our donors to answer them as well. This strategy allowed us to put their responses in relation to the population in the UK Biobank. Donors in control cohort 1 were recruited from the same patient pool at Northwestern and selected to match the sex, age, and risk profiles of the tanning cohort with the exception of tanning bed usage. All donors in control cohort 1 self-reported no tanning bed usage. Both the tanning and control cohorts underwent additional scrutiny to contextualize them within the general population. This was achieved by comparing their responses to risk factor questionnaires with publicly available responses from participants of the UK Biobank survey (fig. S1). Informed consent was obtained from all donors recruited at Northwestern. This study was approved by the Institutional Review Board protocol STU00009443.

Donors from control cohort 2 were collected from the UCSF Willed Body Program. The UCSF Willed Body Program was established to receive the remains of individuals who choose to donate their body for medical research. All donors consented, as part of their living will, before their death. There were no inclusion criteria for Willed Body donors, although most donors through this program are of advanced age. All biopsies analyzed in this study were obtained from the upper back (including shoulders) and lower back (including mid-back) regions.

Skin biopsy preparation for single-cell genotyping

Shave biopsies, ranging from 3 to 5 mm in their longest dimension, were collected from living donors or cadaver tissue, placed in saline, put on ice, and transported to the Shain laboratory at UCSF for molecular analyses. Upon receipt, skin biopsies were treated with dispase (10 mg/ml) for 16 hours, breaking up collagens connecting the dermis to the epidermis. After incubation, the epidermis was physically peeled from the dermis, minced with a scalpel, trypsinized (0.05% trypsin for 3 min), and vortexed (every 5 to 10 s for 3 min) to establish a suspension of single cells. Suspended cells were placed in tissue culture and grown in CNT40 media (CELLnTEC) + 5% antibiotic-antimycotic. After 1 week, the cultures contain a mixture of melanocytes and keratinocytes. To separate these cell types, trypsinization (0.05%) was performed for 3 min, detaching melanocytes while leaving keratinocytes adherent to the plate.

After achieving stability in bulk cell culture, individual cells were seeded into 96-well plates using serial dilution. These plates underwent immediate screening to eliminate wells containing multiple or no cells. The single-cell cultures were maintained until they ceased expanding, typically resulting in colony sizes of 200 cells. Colonies were harvested in RNA lysis (RLT) buffer (QIAGEN, 79216).

We further amplified the genomic DNA and mRNA from each harvested colony using the single cell genome and transcriptome sequencing (G&T) sequencing protocol (37, 38). The G&T-seq protocol describes how to separate genomic DNA and mRNA. Once separated, the genomic DNA was amplified in vitro via multiple displacement amplification (MDA) (QIAGEN REPLI-g Single Cell Kit, 150345) or primary template amplification (PTA) (BioSkryb ResolveDNA Whole Genome Amplification Kit, 100136). We switched to PTA when the technology became available because it has higher fidelity amplification than MDA (39). The mRNA was amplified with SMART-Seq2 (Switch Mechanism at the 5′ End of RNA Templates 2) (40). Collectively, the clonal expansion and in vitro amplification of nucleic acids produced sufficient genomic material to call somatic mutations from individual melanocytes.

A reference source of normal DNA was collected from each donor. For the living donors recruited at Northwestern, a buccal swab was collected. For the cadaver tissue at UCSF, we biopsied a distant skin sample. DNA from buccal swabs was isolated with prepIT.L2P (DNA Genotek, PT-L2P-5), whereas DNA from skin samples was isolated using the DNeasy Blood & Tissue Kit (QIAGEN, 69504).

Sequencing and somatic alteration calls

Nucleic acids were fragmented to a target size of 350 bp using Covaris LE220, followed by end-repair and ligation to Integrated DNA Technologies (IDT) 8 or 10 dual index adaptors. Amplification was carried out using the KAPA HyperPrep Kit (KK8504). Subsequently, the genomic DNA libraries were enriched for the exome through hybridization with NimbleGen SeqCap EZ Exome + UTR (06740294001), KAPA HyperExome v1 (09062556001), or KAPA HyperExome v2 (9718648001) probes using a KAPA HyperCapture Reagent kit (09075828001). Paired-end sequencing (either 100 or 150 bp) was conducted on Illumina HiSeq 2500 or NovaSeq 6000 instruments.

Sequencing data from genomic DNA were aligned to the hg19 version of the genome with Burrows–Wheeler Aligner (BWA, v2.0.5) (41) and deduplicated with Picard (v2.1.1). Subsequently, the reads underwent additional curation to realign indels and recalibrate base quality using Genome Analysis Toolkit (GATK, v4.1.2.0). For RNA sequencing data, alignment to both the genome and transcriptome was conducted with the Spliced Transcripts Alignment to a Reference (STAR) aligner (v2.1.0) (42). The reads were then deduplicated using Picard (v2.1.1), and gene-level read counts were quantified using RNA-Seq by Expectation-Maximization (RSEM, v1.2.0) (43).

Copy number alterations were inferred from DNA and RNA sequencing data using CNVkit (v.0.9.6.2) (44, 45). A candidate set of germline heterozygous single-nucleotide polymorphisms (SNPs) was called with FreeBayes (v.1.3.1) (46) and further filtered to only include SNPs observed in the 1000 Genomes Project and between 40 and 60% allelic frequency mapping to each allele. A candidate set of short insertions and deletions was called with Pindel (47). Candidates were filtered to remove those with fewer than four supporting reads and which were present in the reference bam. Remaining candidates were manually screened to remove likely alignment artifacts.

Somatic point mutations were called as previously described (17). Briefly, a candidate list of point mutations was called with MuTect2 by comparing the bam file of each clonal expansion to the reference bam file from the same patient. The MuTect2 calls included both somatic point mutations and amplification artifacts that arose during MDA or PTA. To remove amplification artifacts, we leveraged patterns in the sequencing data. We searched for supporting reads in the RNA sequencing data from each clonal expansion, allowing us to validate or invalidate candidates in highly expressed genes.

We also interrogated phasing patterns for variants near germline heterozygous SNPs. True mutations occur in complete linkage with at least one parental allele, whereas artifacts tend not to show this pattern. For the remaining variants, which were not in highly expressed genes or near germline heterozygous SNPs, we inferred the likelihood that they were a true somatic mutation based on their allele frequency.

Mutation burden and mutational signature analyses

Mutation burdens were calculated as mutations per megabase (Fig. 2 and fig. S2). We counted the number of mutations as described above. To determine the footprint of genome with sufficient coverage in each clonal expansion, we ran the footprints software (17) to count the precise number of base pairs with 10× coverage or greater. For the analysis of mutational signatures, we compiled somatic mutations across all cells from both cohorts (table S4) and established trinucleotide contexts for single base substitutions using the Bioconductor library BSgenome.Hsapiens.UCSC.hg19 (v1.4.3). This analysis was restricted to cells harboring a minimum of 10 mutations, as profiles with fewer mutations are statistically unreliable. DeconstructSigs R package (v1.9.0) (22) was used to generate the mutation signature profile for each cell using predefined 78 COSMIC (v3.4) signatures. These signatures were previously delineated by SigProfiler (48) and listed in the COSMIC database (https://cancer.sanger.ac.uk/signatures/sbs/).

A stacked bar graph was constructed to illustrate the prevalence of signatures 7a, 7b, 7c, 7d, and the top five additional signatures for each individual cell (Fig. 3A). Remaining signatures were grouped under an “others” category. Specifically, we also compared signatures 7 and 11 between the tanning bed and control cohorts, examining the fraction of these signatures per cell. We used the Wilcoxon rank sum test to assess the significance of any observed differences. We further scrutinized signatures 7 and 11 by calculating the fraction of cells positive for each signature in both cohorts. Confidence intervals and P values were computed using the Poisson test.

De novo mutational signature analysis was performed using the SigProfilerExtractor package (v1.2.2) in Python (v3.13.3) (48). The matrix of mutation frequencies, stratified in their 96-trinucleotide context (generated for DeconstructSigs) was used as an input. Signature extraction was run in matrix mode, with the GRCh37 reference genome specified as both the opportunity genome and the reference genome. The solution space is set to a minimum of 1 and a maximum of 10 signatures. Nonnegative matrix factorization was performed with 100 independent replicates. The algorithm evaluates candidate solutions for stability and reconstruction error (fig. S5A). More specifically, a maximum mean sample cosine distance of 0.2 and a minimum average stability of 0.8 are set for de novo signatures. Output included the extracted de novo signatures, their contributions to each sample, and visualization files. Extracted signatures were subsequently compared to known COSMIC v3 signatures using cosine similarity to identify potential matches and to interpret the underlying mutational processes (fig. S5, B to D).

Annotation of pathogenic mutations

Somatic mutations were manually scrutinized by the authors to annotate mutations known to be under selection in melanoma. The full list of mutations is available in table S4, and our pathogenic annotations are shown in column AA of that table. The final list of pathogenic mutations is available in Fig. 4A.

Clonality analyses

To visually depict the phylogenetic relationship (fig. S7) among melanocytes sharing somatic mutations, we created clonality plots, as shown in Fig. 4D. In these plots, the size of each square corresponds to the surface area of the skin biopsy at the indicated scales. To simplify the visualization, the schematics depict skin biopsies as squares, although their shapes varied. For punch or shave biopsies, the area was calculated on the basis of their known diameters. In cases of larger biopsies with irregular shapes, images were captured with a scale and analyzed using ImageJ (49) to determine the area. Melanocytes were represented as randomly positioned dots, with clonally related ones enclosed within circles.

The total surface area represented by each melanocyte was computed by dividing the biopsy area by the total number of sequenced melanocytes. Consequently, the circle’s size is proportionate to the area covered by the enclosed cells. Because of geometric disparities between squares and circles, cases with multiple subclones might result in circles extending beyond the square boundaries. In such scenarios, the sizes of all circles within the square were proportionally reduced to fit within the square outline. Melanocytes carrying pathogenic mutations were marked in red, along with the outline of their enclosing circle. In instances where multiple cells shared somatic mutations, the largest circle represented the trunk of the phylogenetic tree with the subsequent circles within depicting the branches.

Mixed-effects modeling

Mutation burden was compared between the tanning cohort and the combined data from two control cohorts at the level of individual cells. In univariate analyses, we noted that the mutation burden correlated with anatomic site and tanning bed history and varied from donor to donor. Therefore, to determine whether tanning bed usage was associated with mutation burden, on top of the effects from anatomic site and donor-specific effects, we developed a mixed-effects linear model (LME).

Our model included fixed effects for cohort (tanning versus nontanning) and anatomical site (lower back versus upper back) as well as a random effect for subject, which accounts for repeated measures within individuals. Because the mutation burden (mut/Mb) exhibited a non-Gaussian distribution, it was normalized using the transformation log[(mut/Mb) + 1]. The normalization yielded an approximately Gaussian distribution and can be found in column J of table S6 (Fig. 2 and tables S2 and S3). The models compared were as follows:

Null model: mut_norm ~ 1 + Anatomical_site + (1 | Person)

Alternate model: mut_norm ~ 1 + Cohort + Anatomical_site + (1 | Person)

Here, mut_norm denotes the normalized mutation burden, 1 is the intercept (baseline level), Anatomical_site and Cohort are fixed effects controlling for site- and cohort-specific differences, and (1 | Person) is the random intercept for each subject (fig.S3). The cohort effect was assessed using a likelihood ratio test using the lme4 R package (18), testing the null hypothesis that the cohort coefficient is equal to zero, with statistical significance defined as P < 0.05.

Because melanocytes from donor 59 exhibited a relatively high mutation burden, the analysis was repeated with the RLME R package (19). This approach extends the standard LME framework to reduce sensitivity to outliers, heavy-tailed distributions, and model violations. The fixed and random effects were specified identically to those used in the original linear model, with significance again defined as P < 0.05.

Estimation of body surface area

The body surface area of various anatomical sites was approximated using the Lund and Browder chart (28). For epidemiological investigations (Fig. 1C and fig. S9), we categorized the face/head/neck, arms below the elbows, and lower extremities below the knees as high cumulative sun damage regions, as they are more likely to remain uncovered and thus are chronically exposed to sunlight. Conversely, regions typically shielded by clothing, such as the upper arms, front and back of the torso, and thighs, were defined as low cumulative sun damage sites. The total body surface area for high- and low-cumulative sun damage regions was calculated by summing the anatomical sites mentioned above.

Analysis of gene expression profile between the tanning bed and control cohorts

Reads from all melanocytes of the same biopsy were aggregated to generate pseudobulk RNA sequencing data. A small number of patients had two or more biopsies, and in these cases, the biopsies were treated as independent samples—i.e., we combined sequencing data of melanocytes from the same biopsy but not all melanocytes from the same patient. This resulted in 11 pseudobulk samples for the tanning bed cohort and 20 pseudobulk samples for the control cohort.

Genes with fewer than 10 total reads across all pseudobulk samples were removed before analysis. Differential expression was assessed using DESeq2 (v1.46.0) (50) in R (v4.4.3). Counts were normalized using the DESeq2’s default estimateSizeFactors method and variance-stabilizing transformation. Genes with adjusted P < 0.1 (Benjamini-Hochberg) and log₂(fold change) > 0 were considered significant (table S5).

Significant genes were visualized in a volcano plot (fig. S8A) generated with ggplot2 (v4.0.0) (51) and a heatmap (fig. S8B) using pheatmap R package (v1.0.13) (52) without clustering. The heatmap displays genes up-regulated in the tanning bed cohort versus the control cohort.

Supplementary Materials

The PDF file includes:

Figs. S1 to S9

Legends for tables S1 to S6

Other Supplementary Material for this manuscript includes the following:

Tables S1 to S6