If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Dermatology, Massachusetts General Hospital, Boston, MassachusettsDepartment of Population Medicine, Harvard Pilgrim Healthcare Institute, Boston, Massachusetts
The incidence of basal cell carcinoma (BCC) is higher among men than women. Susceptibility loci for BCC have been identified through genome-wide association studies, and two previous studies have found polygenic risk scores (PRS) to be significantly associated with the risk of BCC. However, to our knowledge, sex-stratified PRS analyses examining the genetic contribution to BCC risk among men and women have not been previously reported. To quantify the contribution of genetic variability on the BCC risk by sex, we derived a polygenic risk score and estimated the genetic relative risk distribution for men and women. Using 29 published single nucleotide polymorphisms, we found that the estimated relative risk of BCC increases with higher percentiles of the polygenic risk score. For men, the estimated risk of BCC is twice the average population risk at the 88th percentile, while for women, this occurs at the 99th percentile. Our findings indicate that there is a significant impact of genetic variation on the risk of developing BCC and that this impact may be greater for men than for women. Polygenic risk scores may be clinically useful tools for risk stratification, particularly in combination with other known risk factors for BCC development.
, 2008). Because individual variants typically have small effect sizes, combining variants into polygenic risk scores (PRS) is potentially more informative for identifying high- and low-risk individuals, especially in conjunction with clinical risk factors (
), although neither stratified by sex. Since the impact of inherited genetic variation on differential BCC risk by sex has not been established, we created a PRS using published GWAS loci to estimate the sex-specific relative risk distribution owing to known risk alleles, with the goal of identifying high-risk men and women.
Results
We identified 29 single nucleeotide polymorphisms (SNPs) at 27 loci meeting our inclusion criteria (Table 1). The estimated relative risk of BCC by the percentiles of the 29-SNP PRS for men and women is shown in Figure 1. A two-fold increase in relative risk has been suggested as a benchmark for clinical significance in studies of genetic risk (
). In our analysis, the estimated relative risk was approximately two at the 88th percentile for men and the 99th percentiles for women, indicating that our PRS identifies 12% of men and 1% of women at two-fold increase in risk of BCC relative to the average population risk.
Table 1SNPs Included in the Polygenic Risk Score for BCC
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
Abbreviations: BCC, basal cell carcinoma; BP, physical position; N/A, the positions of these variants were not reported in the publication; OR, odds ratio; SNP, single nucleotide polymorphism.
1 These variants were reported to have an inverse association with the BCC risk (OR < 1); therefore, the risk allele is the major allele, and the major allele frequency and inverse OR are presented.
Figure 1Relative risk of BCC by percentile of polygenic risk score for men and women. Percentile of PRS is plotted against the relative risk of BCC. Arrows point to the PRS percentile at which the relative risk of BCC is 2 (men: 92nd percentile; women: 98th percentile). BCC, basal cell carcinoma; PRS, polygenic risk score; SNP, single nucleotide polymorphism.
Our findings suggest that our PRS has potential clinical utility for identifying individuals at increased risk of developing BCC and classifies different proportions of high-risk men and women. We included putative pigmentation genes (MC1R, SCL45A2, BNC2, IRF4, OCA2, and RALY) and the tumor suppressor gene TP53, oncogene MYCN, and ras oncogene homolog RHOU. Additional tumor progression loci include FOXP1, CASC15, CUX1, and the microRNA MIR3939. Other loci function in immune regulation (HLA-DQA2, LPP), epidermal development (TGM3, KRT5, GATA3, ALSCR12), telomere maintenance (OBFC1, CLPTM1L), and cell cycle progression (CDKN2B, RCC2).
Several potential mechanisms of action may explain the sex disparities we observed. Hormonal factors, such as the use of hormone replacement therapy, has been suggested as a risk factor for BCC, and hormone replacement therapy users may represent a high-risk subgroup (
). Differential UV exposure between men and women may also explain some of the disparity, and there is evidence that sun protection behaviors differ between men and women as well (
. The clinical implementation of PRS for the risk prediction of these cancers suggests that PRS may also be useful tools for risk stratification in keratinocyte carcinomas. Our estimates indicate that 12% of men and 1% of women at two-fold increased risk of developing BCC can be identified using a PRS. As an annual examination for suspicious skin lesions is the only screening modality available, it is possible that PRS could be used to determine high-risk individuals who require more frequent monitoring, and conversely, low-risk individuals who are unlikely to require annual screening. Although the proportion of high-risk men and women classified in our study is small, given the prevalence of BCC, identifying individuals at greather than two-fold increased risk may still prove useful for large healthcare delivery systems. For example, within Kaiser Permanente, which has a membership of 4.2 million, hundreds of thousands of high-risk individuals, who may potentially benefit from more intensive screening, may be identified. This risk stratification approach has the potential to reduce false positive screenings and the concomitant expenses associated with pathologic confirmation, as well as to alleviate patient anxiety. However, low-risk does not necessarily equate to risk-free, and the probabilistic nature of genetic risk profiles would need to be clearly communicated to patients to prevent the abandonment of traditional skin cancer prevention practices. While the use of PRS as tools for screening shows promise in preliminary data, further study and validation is needed before clinical use is feasible.
Previous studies have examined the associations between PRS and BCC risk (
). In the Michigan Genomics Initiative, a 19-SNP PRS was significantly associated with BCC risk (odds ratio [OR] = 2.7, 95% confidence interval 2.2–3.2; top versus bottom quartile, adjusted for age, sex, ancestry, and genotyping array) (
). Among renal transplant recipients, a three-fold increased risk of BCC (OR = 3.03, 95% confidence interval 1.78–5.16), per one standard deviation increase in the normalized PRS, was observed (
). Sex was not significantly associated with the BCC risk in this study, which may be population-specific, as sex is differentially associated with the BCC risk in the general population. The risk prediction of any non-melanoma skin cancer was also explored; adding the PRS to a model containing the recruitment site and transplant age and era resulted in an increase of 0.02 to the area under the receiver operator characteristic curve, which is a graphic representation of the true and false positive proportions (
). However, this study did not include other risk factors, such as sun exposure and immunosuppressive treatment, and these estimates may not be generalizable to other populations.
We incorporated variants from published two-stage GWAS that were replicated and reached genome-wide significance in meta-analysis, using statistical methods similar to previous studies to create our 29-SNP PRS. In contrast, one previously published study (
) used previously reported variants from the National Human Genome Research Institute-European Bioinformatics Institute GWAS Catalog. While these studies used individual-level genotyping data to investigate the associations between PRS and BCC risk, we used summary OR from all the published GWAS meeting our criteria to investigate the BCC relative risk distribution owing to the variants included in our PRS. Our estimates of the distribution of polygenic risk in the population depend on our modeling assumptions, including the log-additive odds model, the accuracy of estimated per-allele ORs, and the independence of genotypes across the risk SNPs. Whereas these assumptions have been empirically validated for other cancers (
Additive interactions between susceptibility single-nucleotide polymorphisms identified in genome-wide association studies and breast cancer risk factors in the Breast and Prostate Cancer Cohort Consortium.
), the model-based BCC risks should be validated in an independent study sample. In particular, because the log-additive model assumes that risk increases exponentially, the predicted risks in the tails of the distribution should be empirically calibrated. The behavior of the curves for men and women at the higher end of the distribution—for example, the increasing risk difference between the mean and women at higher values of the PRS—is a consequence of the log-additive model. Our model assumes that the effect of sex on the relative risk scale is independent of the genetic risk (i.e., no effect measure modification by sex). However, this assumption requires confirmation in an independent study population.
Another potential limitation is the inclusion of data from studies relying on self-reported BCC diagnoses, rather than histopathologically confirmed cases, which could have potentially resulted in misclassification within these studies. The validity of self-reported BCC diagnoses has previously been established in the 23andMe (
) cohorts. Furthermore, for cutaneous squamous cell carcinoma, variants reaching genome-wide significance in a cohort with histologically-proven cases (
). This suggests that for keratinocyte carcinomas, reliance on self-reported diagnosis is a valid approach. Moreover, because we did not use individual-level data, our model could not include other known BCC risk factors, such as sun exposure and skin pigmentation. Finally, the GWAS we included were conducted exclusively in non-Hispanic white populations, where the risk of BCC is greatest, which limits the generalizability to other racial- ethnic groups.
Our findings raise questions about the potential genetic contribution to the earlier onset of BCC in men, the greater number of BCCs in men compared with women, and the more aggressive disease course observed in men compared with women. Our study, using summary data, was not designed to address questions such as these. Future studies of genetic risk, incorporating individual-level data and detailed information on the course of disease, will be needed to investigate these questions.
Although the methodologies for creating PRS differed, our results are consistent with previous studies and support the hypothesis that genetic variability influences BCC development. To our knowledge, this is the first study to examine the differential stratification of BCC risk by sex using PRS. Further studies incorporating PRS and known BCC risk factors, such as sun and ionizing radiation exposure, will be necessary to facilitate translation of polygenic scores to clinical practice.
Materials & Methods
We compiled data from two-stage GWAS of BCC risk, published through November 1, 2018, identifying studies from PubMed, Embase, and the National Human Genome Research Institute-European Bioinformatics Institute GWAS Catalog, using the search terms “basal cell carcinoma”, “BCC”, “keratinocyte carcinoma”, “skin cancer”, “genome wide association study”, and “GWAS.” Bibliographies of relevant articles were also searched. We included seven studies in our analysis (
, 2008). Hospital- and cancer registry-based BCC cases were histopathologically confirmed; 23andMe , Nurses’ Health Study, and Health Professionals Follow-up Study included self-reported BCC diagnoses. Controls without BCC were drawn from 23andMe , Nurses’ Health Study and Health Professionals Follow-up Study, deCODE Genetics (
, 2008). All participants were of European ancestry.
We abstracted minor allele frequencies, OR, and P-values from the discovery and replication phases and meta-analysis. Biallelic SNPs were included in our analysis if they reached P < 0.05 in the replication stage and genome-wide significance (P < 5x10-8) in the meta-analysis. For variants in multiple studies and those in linkage disequilibrium (r2 > 0.3), we used data from the larger study population.
The methodology used to derive the PRS and estimate the population distribution of the risk score has previously been published (
). Briefly, we used the additive OR and risk allele frequency for each SNP meeting our inclusion criteria to construct the PRS, by summing the number of risk alleles across SNPs and assuming a log-additive model for joint effects. For variants with OR < 1, we used the inverse OR and 1 – the reported effect allele frequency, so that the association was in the same direction for all the SNPs. We then calculated the standard deviation of the PRS using the previously published method (
). We estimated the sex-stratified population distribution of the BCC risk owing to the 29 variants in the PRS, relative to the overall population average, using the standard deviation of the PRS and a published estimate of the BCC relative risk for men versus women (incidence rate ratio = 1.65) to stratify by sex (
). For men, the estimated distribution is essentially shifted by a factor of 1.65 to represent the relative risk associated with the exposed group (men). We normalized the distributions for men and women using the proportion of men and women ≥ 65 years old (46.3% and 53.7%, respectively) in the United States population. Although there are currently no clear recommendations as to when screening for BCC should begin (
). We therefore used age 65 as a cutoff to define the target at-risk population. Census data were used to calculate the proportion of men and women at age 65 in the 2017 United States population (
). We plotted percentiles of the PRS against the estimated BCC risk distribution to aid interpretation. Analyses were performed using Microsoft Excel 2016 and R version 3.2.0.
Data Availability Statement
The datasets related to this article can be found in the referenced published studies.
This study was funded by NIAMS K24 AR069760 (MMA). The funding agency had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript, or decision to submit the manuscript for publication.
Additive interactions between susceptibility single-nucleotide polymorphisms identified in genome-wide association studies and breast cancer risk factors in the Breast and Prostate Cancer Cohort Consortium.
33378-0&id=gr1.jpg){kind=link}