A comprehensive genome-wide analysis of melanoma Breslow thickness identifies interaction between CDC42 and SCIN genetic variants

Abstract

Breslow thickness (BT) is a major prognostic factor of cutaneous melanoma (CM), the most fatal skin cancer. The genetic component of BT has only been explored by candidate gene studies with inconsistent results. Our objective was to uncover the genetic factors underlying BT using an hypothesis-free genome-wide approach. Our analysis strategy integrated a genome-wide association study (GWAS) of single nucleotide polymorphisms (SNPs) for BT followed by pathway analysis of GWAS outcomes using the gene-set enrichment analysis (GSEA) method and epistasis analysis within BT-associated pathways. This strategy was applied to two large CM datasets with Hapmap3-imputed SNP data: the French MELARISK study for discovery (966 cases) and the MD Anderson Cancer Center study (1,546 cases) for replication. While no marginal effect of individual SNPs was revealed through GWAS, three pathways, defined by gene ontology (GO) categories were significantly enriched in genes associated with BT (false discovery rate ≤5% in both studies): hormone activity, cytokine activity and myeloid cell differentiation. Epistasis analysis, within each significant GO, identified a statistically significant interaction between CDC42 and SCIN SNPs (p_meta-int =2.2x10⁻⁶, which met the overall multiple-testing corrected threshold of 2.5x10⁻⁶). These two SNPs (and proxies) are strongly associated with CDC42 and SCIN gene expression levels and map to regulatory elements in skin cells. This interaction has important biological relevance since CDC42 and SCIN proteins have opposite effects in actin cytoskeleton organization and dynamics, a key mechanism underlying melanoma cell migration and invasion.

Introduction

Cutaneous melanoma (CM) is the most aggressive form of skin cancer and has shown a dramatic rise in incidence in fair skinned populations over the last decades.¹ Mortality rates have also been rising between 1955 and 1985 but have stabilized in recent years. However, mortality from melanoma continues to increase among men aged 65 and older.² Breslow thickness (BT), which is a measure of primary melanoma depth, represents the most important prognostic factor of this cancer.³ The 10-year survival rate was found to be 92% in patients with thin melanomas (BT < 1 mm) and to decrease to 50% in patients with melanomas more than 4 mm thick.³ BT has been reported to be associated with host factors that are also associated with melanoma risk. A greater BT was found to be consistently associated with older age at diagnosis and male sex and to be inversely associated with number of nevi.^{4, 5} To date, genetic studies of Breslow thickness have been limited to candidate genes, including genes associated with melanoma risk, and, for the most part, have led to marginal and/or inconsistent associations.^5–7 Thus, the genetic factors influencing Breslow thickness are mostly unknown and deserve to be discovered using hypothesis-free genome-wide approaches.

Genome-wide association studies (GWASs) have been very successful in identifying thousands of loci associated with many diseases or traits,⁸ including 20 loci for risk of melanoma.⁹ These GWASs typically focus on the analysis of individual SNPs and, because of the large number of SNPs analyzed, require stringent thresholds to declare an association as genome-wide significant. Consequently, GWASs are underpowered to detect genetic variants which have small marginal effect but rather act jointly or interact with each other in disease or trait variability. Complementary approaches to GWAS, such as GWAS pathway analysis and epistasis analysis, have been proposed to overcome these issues. Pathway analysis of GWAS outcomes is based on the premise that genes do not work in isolation; instead, genes that belong to biological and functional units (pathways or gene sets) can harbor markers which might be detectable when examined jointly. This approach identifies pathways enriched in genes associated with the trait and can also provide novel insight into the trait biology. Another way to increase power of detecting association of markers with a trait is to conduct statistical analysis that considers gene-gene interactions; this is especially true for low marginal-effect SNPs. Pairwise SNP-SNP interactions can be investigated either at the genome-wide level or by using statistical and/or biological filtering procedures (such as pathway analysis) to limit the search for interactions among a subset of genetic markers¹⁰ and, thus, to increase statistical power by reducing the multiple testing burden. Therefore, the use of both GWAS and post-GWAS approaches can maximize power to uncover genes underlying complex phenotypes. Indeed, the integration of pathway and epistasis analyses has recently been successful in identifying two novel genes interacting in melanoma risk¹¹ that were not detected by a large-scale meta-analysis of melanoma GWASs.⁹

The objective of this study is to identify genetic factors influencing Breslow thickness by applying a combination of GWAS and post-GWAS approaches to two large studies of melanoma with genome-wide SNP data: the French MELARISK study (966 cases), that served as the discovery dataset, and the North-American MD Anderson Cancer Center (MDACC) study (1,546 cases), that was used as the replication dataset. For that purpose, we first conducted a genome-wide single-SNP association analysis to identify marginal effects of individual genetic variants associated with BT. This was followed by a pathway analysis of GWAS outcomes to identify gene sets enriched in genes associated with BT and the most promising candidates among genes driving the pathways through literature mining. Finally, we carried out an epistasis analysis within BT-associated pathways to identify interactions between genetic variants influencing BT. To the best of our knowledge, this is, to date, the most comprehensive assessment of the genetic factors influencing Breslow thickness.

Materials and methods

Study populations

Protocols of both MELARISK and MDACC studies have been approved by the ethical committees of Paris-Saint-Louis, Paris-Necker and Ile-de-France II (Paris, France) for the MELARISK study and by the MD Anderson Cancer Center Institutional Review Board (Houston, Texas) for the MDACC study. All subjects participating to these studies gave their written informed consent.

The MELARISK study comprised of 1,244 European-ancestry melanoma cases who were recruited through a nationwide network of French Dermatology Departments and Oncogenetic clinics between 1992 and 2011. The protocol of case recruitment and data collection has been described in detail elsewhere.¹² Briefly, confirmation of each reported CM case was sought through review of medical records, review of pathological material, and/or from pathological reports. Information on Breslow thickness of the primary tumor was retrieved from medical and/or pathological records and was obtained together with age at diagnosis of melanoma and sex in 1,011 confirmed CM patients. The MDACC study has been previously described¹³ and consisted of 1,804 non-Hispanic patients with newly diagnosed CM who were recruited from The University of Texas MD Anderson Cancer Center between March 1998 and August 2008. Information on Breslow thickness was retrieved from patient records and maintained in the MD Anderson Melanoma Informatics, Tissue Resource, and Pathology Core and was obtained together with age at diagnosis of melanoma and sex in 1,804 CM patients. In both studies, for patients that had multiple primary melanomas, the Breslow thickness of the first melanoma was used.

Genotyping and quality control

MELARISK subjects were genotyped in three stages, as part of previous GWASs of melanoma risk.^{9, 14, 15} Genotyping was based on Illumina HumanHap300 Beadchip version 2 duo array for stage 1 (481 CM cases with known BT) and on Illumina Human610-Quad array for stages 2 and 3 (530 CM cases with known BT). MDACC samples were genotyped with the Illumina HumanOmni1-Quad_v1-0_B array.¹³ Standard QC measures were applied to samples and SNPs, as described previously for MELARISK^{9, 14, 15} and MDACC.¹³ After quality control (QC) of genotypic data, the number of melanoma patients kept for analysis was 966 in MELARISK and 1,546 in MDACC. To get the same set of SNPs across samples, genotypic imputations were carried out in each dataset (and by genotyping stage in MELARISK) using MaCH¹⁶ and Hapmap3 CEU population as reference panel. Only SNPs that had imputation quality score (rsq) ≥ 0.8 and minor allele frequency (MAF) ≥ 0.05 were retained for analysis. There were 1,032,745 SNPs that passed QC in MELARISK and 1,067,258 SNPs in MDACC.

Statistical analysis

Descriptive statistics of sex, age at diagnosis and Breslow thickness and the effects of age and sex on BT were assessed in each dataset using Stata V12 (distributed by Stata Corporation, College Station, Texas, USA). Since BT is right-skewed, we used the logarithm to base 10 of BT (log₁₀BT) for the latter and subsequent analyses. The workflow of our genome-wide analysis strategy is presented in Figure 1.

Fig. 1. Workflow of analysis strategy.

Our analysis strategy included: - a genome-wide association study (GWAS) of individual SNPs with Breslow thickness (BT) (first-stage in MELARISK and second-stage in MDACC); - a pathway analysis of MELARISK and MDACC GWAS outcomes to identify pathways (defined by gene ontology categories) enriched in genes associated with BT; a SNP-SNP interaction analysis within BT-associated pathways to identify the interactive effect of genetic variants influencing BT.

Genome-wide single SNP association analysis

To assess the effect of single SNPs on BT, we performed a two-stage GWAS. In the first stage conducted in MELARISK, residual BT (after adjustment of log₁₀BT on significant sex and age effects) was regressed onto SNP allele dosage (assuming an additive model) while adjusting for stage of genotyping, using Stata V12. Test of SNP effect was based on the Wald-test. All SNPs reaching a critical threshold of 10⁻⁵ were followed-up in MDACC at stage 2. The association analysis in MDACC was performed using ProbABEL¹⁷ assuming an additive model for SNP effect and adjusting for sex and age at diagnosis. The results of MELARISK and MDACC were combined using a fixed effects meta-analysis and a threshold (p_meta) of 5x10⁻⁸ was used to declare genome-wide significance.

Genome-wide pathway analysis

Pathway analysis was applied to the outcomes of MELARISK and MDACC GWASs. For that purpose, we used genome-wide single SNP statistics obtained from an MDACC GWAS of BT.¹⁸ Pathway analysis was based on the gene-set enrichment analysis (GSEA) approach implemented in the GenGen package.¹⁹ GSEA computes an enrichment score to detect gene-sets significantly enriched in genes associated with the trait compared to the whole genome. Imputed SNPs from BT GWAS were first mapped to genes (between the start site and 3’-untranslated region of each gene) using dbSNP Build 132 and human Genome Build 37.1. The GWAS-single SNP association statistic for the most significant SNP in each gene was used to represent the gene. To map genes to pathways, we used Gene Ontology (GO) categories as provided by GenGen (i.e. biological process and molecular function level-4 GO categories to minimize the number of overlapping GOs). Our analysis was also restricted to GOs that contained at least 20 and at most 200 genes to avoid testing overly narrow or broad functional categories. The enrichment score of a pathway was based on the weighted Kolmogorov-Smirnov-like running sum statistic.¹⁹ Statistical significance of pathway enrichment scores was determined by 100,000 permutations of SNP statistics. These permutations allow not only computing empirical p-values and false discovery rates (FDRs) to correct for multiple testing but also normalizing the observed enrichment scores based on the null distribution of these scores, hence adjusting for variation in gene length and pathway size. We used a FDR 5% as a stringent criterion for statistical significance in the discovery dataset and a FDR 5% in the replication dataset as a criterion for a validated result. We identified the genes driving the enrichment score of each significant pathway (also called “leading edge genes”). To further prioritize the leading edge genes shared by MELARISK and MDACC as relevant candidates for BT, we used text mining based on the domain knowledge score tool (DKS).²⁰ DKS performs automated PubMED searches with each gene and any combination of modifier terms. The modifier terms were “melanoma” and either “Breslow thickness” or any one of cancer survival or progression-related terms ("metastasis", "progression", "recurrence", "survival", "prognosis", "invasion", "invasiveness", "mortality"). The DKS that each gene gets is the number of PubMed abstracts and/or titles retrieved with the querying terms. We selected as BT candidates the leading edge genes belonging to the top 10% of DKS distribution.

SNP-SNP interaction analysis within BT-associated pathways

In MELARISK, we analyzed all cross-gene SNP-SNP interactions for leading edge genes shared by MELARISK and MDACC within each BT-associated pathway. Pairwise SNP-SNP interactions were evaluated by regression analysis of residual BT assuming an additive model and adjusting for genotyping stage using Stata V12. The additive effect of a SNP was represented by a variable that was coded 1, 0, and -1 for homozygote for the minor (effect) allele, heterozygote, and homozygote for the major (reference) allele, respectively. The interaction term was the multiplication of these variables between SNPs. Since we had imputed SNPs, each SNP was coded as the weighted average of these three genotypic values with weight being the probability of each genotype. Test of interaction was performed by comparing the full model, which included the additive effects of the two SNPs plus the interaction term to the restricted model without interaction, using a likelihood-ratio test which follows a χ² distribution with one degree of freedom.

For each gene, we examined all SNPs lying from 50kb upstream to 50kb downstream of the gene (Build 37.1). We discarded strongly correlated SNPs (r² ≥ 0.8) and all SNP pairs for which one or more of the 9 genotype-by-genotype combinations appeared in fewer than five subjects. All pairs of SNPs showing suggestive evidence for interaction (p_int ≤ 10⁻⁴) in MELARISK were subjected to replication in MDACC. SNP pairs that replicated in MDACC at the nominal 5% level were further meta-analyzed from the two datasets using a fixed-effects model; the meta-analyzed interaction effect was tested using a Wald test (p_meta-int). To correct for multiple testing, we used the same hierarchical method as previously described.¹¹ We first computed, for each gene pair, the effective number of independent SNP-SNP interaction tests using an extension of Li and Ji’s method.²¹ The effective number of independent tests in a pathway was estimated by the sum of the effective numbers of independent tests for a gene pair over all gene pairs tested within that pathway in the discovery dataset. The corrected critical threshold for the number of tests in a pathway (T_pathway) was thus equal to the 5% type I error divided by the effective number of independent tests in that pathway. Finally, to correct for the number of BT-associated pathways tested, we applied a Bonferroni correction to the pathway-corrected threshold (T_pathway) to get the overall critical threshold (T_overall).

Results

The characteristics of 966 MELARISK and 1,546 MDACC melanoma patients are shown in Supplementary Table S1. The sex distribution differed between the two datasets (p<10⁻⁴) with a lower proportion of males (45%) in MELARISK as compared to MDACC (58%). The age at diagnosis of melanoma was younger in MELARISK (46.5 ± 17.7) than in MDACC (52.2 ± 14.5; p<10⁻⁴). The mean Breslow thickness of primary melanomas was also lower in MELARISK than in MDACC (1.46 ± 2.04 mm vs 1.94 ± 1.56 mm; p <10⁻⁴). Assessment of the effect of sex and age at melanoma diagnosis on log₁₀BT showed significant effects of sex, age and age squared in MELARISK and sex and age in MDACC. There was no interaction between sex and age in either dataset. After adjustment on these covariates, the residual BT values had similar distribution in the two datasets (p=0.74).

Genome-wide single SNP association analysis

In the first stage GWAS conducted in MELARISK, no evidence of any systematic bias was observed (see quantile-quantile (QQ) plot in Supplementary Figure S1; the genomic inflation factor (λ) was equal to 1.01). A total of five loci (1p34.1, 5q34, 12q23.3, 13q12.3, 18q11.2) showed associations with BT exceeding the stage-1 screening threshold of p ≤ 10⁻⁵ (p-values ranging between 4x10⁻⁷ and 8x10⁻⁶ for the lead SNPs; Supplementary Table S2 and Supplementary Figure S2). However, none of these SNPs was even nominally significant in MDACC and their meta-analyzed effect did not reach genome-wide significance (Supplementary Table S2).

Genome-wide pathway analysis

Pathway analysis was applied to MELARISK and MDACC GWAS outcomes to detect gene sets enriched in genes associated with BT. As for MELARISK, the MDACC GWAS outcomes did not show any particular bias (λ=1.0) and did not reveal any signal reaching genome-wide significance.¹⁸ In MELARISK, 459,637 out of 1,032,745 Hapmap3-imputed SNPs were mapped to 21,810 genes. Of these, 6,873 genes were assigned to 316 level-4 GO categories. In MDACC, 475,093 out of 1,067,258 SNPs were mapped to 22,096 genes. Of these, 6,909 genes were assigned to 319 level-4 GO categories. A total of 28 GOs reached the threshold for statistical significance (FDR ≤ 0.05) in MELARISK. Three of these GOs were successfully replicated in MDACC (FDR ≤0.05) and were the following: hormone activity (GO:0005179), cytokine activity (GO:0005125) and myeloid cell differentiation (GO:0030099) (Table 1). A total of 63 leading edge genes were driving the pathway enrichment scores in both MELARISK and MDACC (Table 1) and were distributed almost uniquely among the three significant GOs. Text mining of PubMed abstracts applied to these leading edge genes pinpointed five candidates having a Domain Knowledge Score (DKS) in the top 10% of the pathway-specific DKS distribution (Supplementary Figure S3). These genes were POMC for hormone activity, VEGFA, IL1B, TNFSF10 for cytokine activity and CDC42 for myeloid cell differentiation.

Table 1.

Pathways associated with Breslow thickness in MELARISK and MDACC datasets

Pathway (GO class)	MELARISK			MDACC			No. of leading edge genes shared by the two datasets3
	No. of genes1	p2	FDR2	No. of genes1	p2	FDR2
Hormone activity (GO:0005179)	85	4.0x10−5	2.0x10−3	87	4.6x10−4	0.01	16
Cytokine activity (GO:0005125)	183	1.3x10−3	0.03	188	7.0x10−5	6.0x10−3	31
Myeloid cell differentiation (GO:0030099)	81	1.8x10−3	0.04	83	9.5x10−4	0.02	16

¹Number of genes in a pathway having at least one GWAS SNP mapped to a gene

²Empirical p-value and False Discovery Rate (FDR) of a pathway estimated by 100,000 permutations of SNP statistics

³The leading edge genes are the genes which drive the enrichment score of the pathway

SNP-SNP interaction analysis within BT-associated pathways

In the MELARISK discovery dataset, we examined 120, 465 and 120 gene pairs (corresponding to 20,213, 167,041 and 28,495 SNP pairs tested) in the hormone activity, cytokine activity and myeloid cell differentiation GOs, respectively. There were three SNP pairs harbored by three gene pairs that had p_int ≤ 10⁻⁴ in MELARISK and replicated at the nominal 5% level in MDACC (Table 2). Two of these SNP pairs, related to BMP6/NRG1 and CYFIP2/PRL gene pairs and belonging to the cytokine activity GO, had an inverse direction of interaction effect in MDACC as compared to MELARISK and were therefore not significant in the meta-analysis of the two datasets (p_meta-int>0.20). However, one SNP pair related to CDC42 and SCIN loci in the myeloid cell differentiation GO, had the same direction of interaction effect in the two datasets and showed increased evidence for interaction in the meta-analysis of MELARISK and MDACC (p_int=5.6x10⁻⁵ in MELARISK, p_int=7.1x10⁻³ in MDACC, and p_meta-int=2.2x10⁻⁶ in meta-analysis, which is less than both multiple testing corrected thresholds, T_pathway = 7.5x10⁻⁶ and T_overall = 2.5x10⁻⁶, Supplementary Table S3). This statistically significant SNP pair showed a pattern of interaction in which the effect associated with genotype GG (or AA) of rs7798406 (SCIN) had an inverse effect on BT depending on the genotype, GG (or AA), at rs3117048 (CDC42) (Figure 2). We also noted that for this SNP pair, no single SNP had a nominally significant marginal effect and, in presence of interaction in the model, the SNP main effects were also not nominally significant in either MELARISK or MDACC except for rs7798406 (SCIN) in MELARISK (p=0.02) (Supplementary Table S4). Further analysis of SNPs, that were in strong linkage disequilibrium (LD; r² ≥ 0.80) with rs3117048 and/or rs7798406 and were thus discarded from our initial analysis (see methods), showed additional interaction effects involving two other SNPs CDC42 rs2501299 (r²=0.89 with rs3117048) and SCIN rs991317 (r²=0.88 with rs7798406). These interactions reached (or almost reached) the multiple-testing corrected thresholds, which strengthens our finding (Supplementary Table S3 and S5).

Table 2.

SNP pairs showing evidence for interaction effect on Breslow Thickness

Pathway	SNP	Chromosome	Gene1	Alleles E/R2	EAF2	MELARISK		MDACC		Meta-analysis		Multiple-testing corrected thresholds6
						Interaction effect		Interaction effect		Interaction effect
						βint3 (seint)	pint4	βint3 (seint)	pint4	βmeta-int5 (semeta-int)	pmeta-int5	Tpathway	Toverall
Myeloid cell differentiation
	rs3117048	1p36.1	34kU CDC42	A/G	0.31	0.11	5.6x10−5	0.07	7.1x10−3	0.09	2.2x10−6	7.5x10−6	2.5x10−6
	rs7798406	7p21.3	SCIN (intron)	A/G	0.46	(0.03)		(0.03)		(0.02)
Cytokine activity
	rs9505270	6p24-p23	BMP6 (intron)	A/G	0.27	0.12	2.1x10−5	−0.05	0.05	0.02	0.22	1.5x10−6	5.0x10−7
	rs4276645	8p12	NRG1 (intron)	A/G	0.33	(0.03)		(0.02)		(0.02)
	rs3734034	5q33.3	CYFIP2 (exon)	T/C	0.35	0.10	3.7x10−5	−0.04	0.05	0.02	0.28
	rs2066266	6p22.3	17kD PRL	G/A	0.44	(0.03)		(0.02)		(0.02)

¹Gene symbol; k indicates distance to gene in kilobases (Build 37.1); U indicates "upstream of"; D indicates "downstream of ". Upstream and downstream respectively refer to the 5’ and 3’ ends of the coding strand of each gene.

²E/R=Effect allele/Reference allele; EAF=Effect allele frequency.

³β_int is the regression coefficient of interaction effect between SNPs on Breslow thickness; se_int is the standard error of β_int

⁴p_int is the p-value of the likelihood-ratio test for interaction (one degree of freedom test)

⁵β_meta-int and se_meta-int are the regression coefficient and standard error of interaction effect between SNPs in the meta-analysis of MELARISK and MDACC; p_meta-int is the p-value associated with the Wald test of meta-analysed interaction effect

⁶Tpathway and Toverall are respectively the pathway-corrected and overall-corrected critical thresholds to take into account multiple testing.

Fig. 2. Two-locus effects and 95% confidence intervals on Breslow thickness for each genotypic combination of rs3117048 (CDC42) and rs7798406 (SCIN) SNPs.

The x and y axes represent genotypic combinations of rs3117048 and rs7798406 SNPs and the effect of these genotypes on BT, respectively. The genotype-specific effects were computed using the estimates of SNP main effects and interaction obtained from the combined analysis of MELARISK and MDACC. The effect of a SNP was coded 1, 0, and -1 for homozygote for the minor (effect) allele, heterozygote, and homozygote for the major (reference) allele, respectively and the interaction was represented by the multiplication of these variables between SNPs.

Functional annotation of CDC42 and SCIN SNPs

None of the SNPs that have a significant interaction effect on BT were located within a coding sequence. Using data on expression Quantitative Trait Loci (eQTLs) generated in blood, lymphoblastoïd cell lines, skin and adipose tissue^22–24 and data on regulatory elements from the ENCODE and NIH ROADMAP Epigenomics projects,^{25, 26} we investigated whether the interacting SNPs influence CDC42 and SCIN gene expression levels (cis-eQTLs) and/or map to functionally important regulatory regions (Table 3). In fact, rs3117048, located 34kb upstream of CDC42, is a strong cis-eQTL for CDC42 in blood (p=1.5x10⁻⁵⁵) and maps to enhancer histone marks in keratinocytes and binding sites for transcription factors (Table 3). The SNP rs7798406, located in the fourth intron of SCIN, is a cis-eQTL in sub-cutaneous adipose tissue (p=1.9x10⁻⁵) and maps to binding sites of transcription factors. Moreover, by interrogating not only the Hapmap3-imputed SNP data of the current study but also the 1,000 Genomes project database, we found proxies of rs3117048 and rs7798406 that are also cis-eQTLs for CDC42 and SCIN respectively and map within histone activating regions, DNAse I hypersensitivity sites and/or transcription factor binding sites (Table 3). Notably, rs2501299 (33kb upstream of CDC42, r²=0.89 with rs3117048) is a strong cis-eQTL for CDC42 in blood (p=8.2x10⁻⁶⁵) and maps to enhancer histone marks in keratinocytes and melanocytes and rs6956491 (within SCIN intron 5; r² = 0.90 with rs7798406) maps to a DNAse I hypersensitivity site in skin fibroblasts.

Table 3.

Functional annotations of CDC42 and SCIN SNPs

Gene	SNP	Chromosome: Position	Variant Location1	r2 with rs3117048 or rs7798406	Cis-eQTL data		Regulatory elements
					p2	Tissue	Promoter histones marks	Enhancer histones marks	DNase I hyper sensitivity site	Transcription factor binding site
CDC42
	rs7537414	1:22,325,016	54kU	0.97	8.8x10−8	LCL	Yes	Yes (melanocytes)	Yes	TCF3, LMO2, RAD21, ZEB1
	rs3117048	1:22,345,093	34kU	-	1.5x10−55	Blood	No	Yes (keratinocytes)	No	HIF1A, MAF
	rs2501299	1:22,345,647	33kU	0.89	8.2x10−65	Blood	Yes (keratinocytes)	Yes (keratinocytes,melanocytes)	Yes (keratinocytes,melanocytes)	ELF1, EGR1, EHF, NR3C1,SPI1,RAD21
SCIN
	rs7798406	7:12,644,033	Intron 4	-	1.9x10−5	Adipose tissue	No	No	No	CDX2, HOXA10, HOXC9
	rs6956491	7:12,650,851	Intron 5	0.90	4.5x10−6	Adipose tissue	Yes	Yes	Yes (skin fibroblasts)	SP1, BPTF,ARID5B
	rs11763533	7:12,657,746	Intron 5	0.90	4.6x10−6	Adipose tissue	No	Yes	Yes	CTCF

The SNPs rs3117048 (CDC42) and rs7798406 (SCIN), which were identified as having a statistically significant interaction effect on Breslow thickness were queried for correlated proxies (r² ≥ 0.8) using the Hapmap3-imputed SNP data of the current study and the 1,000 Genomes project database. All correlated SNPs were tested for their association with CDC42 or SCIN gene expression levels using eQTL data generated in blood,²⁴ lymphoblastoïd cell lines (LCL),^22,23 sub-cutaneous adipose tissue, and skin.²² These SNPs were also mapped to regulatory elements using ENCODE and ROADMAP data and the HaploReg tool (http://www.broadinstitute.org/mammals/haploreg/haploreg.php).

¹k indicates distance to gene in kilobases (Build 37.1); U indicates "upstream of".

²p-value of association between SNP and gene expression levels provided by eQTL databases.

Discussion

For the first time, we investigated genetic factors associated with Breslow thickness using an hypothesis-free strategy that combines genome-wide single-SNP analysis, pathway analysis and epistasis analysis within BT-associated pathways. While no marginal effect of individual SNPs on BT was revealed through GWAS, pathway analysis, which allows testing for association on the basis of functional units such as gene ontology categories, identified three gene sets that were significantly enriched in genes influencing BT. Epistasis analysis within these BT-associated pathways showed significant evidence for interaction between potentially functional genetic variants at CDC42 and SCIN loci.

While the GWAS identified SNPs modestly associated with BT in the discovery MELARISK dataset, none of these SNPs showed any consistent association in MDACC and therefore any improvement of association signals when the results of the two studies were combined. We checked that our two-stage study had enough power (≥70%) to detect individual SNPs accounting for at least 2.5% of BT variance. This was approximately the proportion of BT variance individually accounted for by SNPs detected in MELARISK but which were not replicated in MDACC. Although these two datasets showed differences according to sex and age at diagnosis of melanoma, the difference in the distribution of BT was no longer significant after adjusting for these covariates. This indicates that the genetic variants underlying BT are likely to have a small marginal effect and/or to be involved in complex mechanisms, which makes their detection difficult by single SNP analysis.

The three GO categories, hormone activity, cytokine activity and myeloid cell differentiation, that were consistently identified by pathway-based analysis in the two melanoma datasets, were driven by genes modestly associated with BT (all had best SNP p-values greater than 10⁻⁴). To further assess whether these three pathways were specific to BT, we repeated pathway analysis for ulceration, another melanoma prognostic factor, that was available in the majority of MELARISK and MDACC melanoma cases. While enrichment of hormone activity in ulceration association signals had FDR that reached 5% in MELARISK but only 10% in MDACC, the other two pathways, cytokine activity and myeloid cell differentiation, did not show significant enrichment (results not shown). This indicates that the genetic factors underlying BT and ulceration are likely to be distinct for the most part. The identification of the hormone activity pathway fits well the observed sex disparities in melanoma outcomes.²⁷ Indeed, among the biological mechanisms that have been proposed to explain the prognostic advantage of female compared with male patients with melanoma, stand the differences in sex hormones levels and estrogen and androgen receptor expression.²⁷ Our finding of an enrichment of hormone activity GO in genes associated with BT strengthens the potential role of sex hormones in melanoma outcomes. The association of the cytokine activity pathway with BT has biological relevance given the key role of immune-related mechanisms in melanoma progression.²⁸ A number of cytokines have been found to be highly expressed in human melanomas, both in primary tumors and metastases.²⁹ Interestingly, the cytokine activity GO was previously found associated with melanoma risk using the same two datasets.¹¹ This supports the importance of this pathway in both melanoma occurrence and progression as recently evidenced by the identification of immune signatures associated with survival in large-scale genomic data of cutaneous melanomas³⁰ and the growing development of immunotherapy regimens.³¹ The third identified GO, myeloid cell differentiation, is also relevant to melanoma outcomes since myeloid-derived suppressor cells (MDSCs) are one of the drivers of tumor-mediated immune evasion.³² Interestingly, the pro-tumoral properties of intra-tumoral myeloid cells have been evidenced by a mouse model of spontaneous melanoma³³ and the level of circulating MDSCs was found to have a prognostic value in advanced melanoma.³⁴

A text-mining approach applied to genes driving the significant pathways enabled to pinpoint five BT candidate genes. Two of these genes, VEGFA and POMC, have been previously reported to be related to BT. Vascular endothelial growth factor A (VEGFA) polymorphisms were found associated with BT³⁵ and VEGFA expression was associated with BT in melanoma tumors.³⁶ The POMC (pro-opiomelanocortin) gene encodes the precursor of α-MSH (melanocyte-stimulating hormone) that binds to the melanocortin receptor 1 encoded by MC1R, a gene found consistently associated with melanoma risk³⁷ and potentially with BT.⁷

Epistasis analysis within the three BT-associated pathways identified a statistically significant interaction between two CDC42 and SCIN SNPs that satisfied the multiple-testing correction. Interactions were further found with proxies of these SNPs, which strengthens our finding. This is remarkable since very few of the increasing number of studies focusing on gene-gene interactions have been successful in detecting statistically significant interactions and/or showing replication. Although our analysis was based on imputed SNPs, we maximized imputation accuracy by using stringent QC criteria to retain SNPs for analysis. For those interacting SNPs that were genotyped, the correlation between genotyped and imputed SNPs was at least 0.95. We assumed an additive genetic model but further analysis, under a more general model that included additive and dominance terms for main effects and interaction, did not show any departure from additivity (results not shown). Exploration of functional annotations of the interacting SNPs showed that these SNPs and proxies are potentially functional since they are strongly associated with CDC42 and SCIN gene expression levels and map within regulatory elements in melanocytes, keratinocytes and/or skin fibroblasts. The proteins encoded by CDC42 (cell division cycle 42) and SCIN (scinderin) have particularly relevant biological functions regarding cancer invasion. The cell division cycle 42 (CDC42) protein belongs to the family of Rho GTPases that play pivotal roles in the control of cell proliferation, cytoskeletal reorganization and cell migration.³⁸ It is overexpressed in a number of cancers and influences oncogenic transformation and invasion.³⁹ CDC42 was found to be involved in melanoma invasiveness⁴⁰ and to have a melanocytic expression positively correlated with Breslow thickness in patients with fatal outcome.⁴¹ Moreover, proteomic analysis of melanocytes from dysplastic nevus and normal skin demonstrated that CDC42 has increased expression in dysplastic nevi.⁴² The scinderin protein (also known as adseverin), encoded by SCIN, is a calcium-dependent actin severing and capping protein, which belongs to the gelsolin superfamily. Proteins of this superfamily are regulators of actin cytoskeleton dynamics and are involved in many actin-related processes.⁴³ SCIN is highly expressed in cancers and its expression was related to cell proliferation.⁴⁴ Notably, CDC42 and scinderin have opposite effects in actin dynamics which drives cell shape changes and motility and therefore plays a key role in cancer cell migration and invasion including melanoma cells.^{45, 46} CDC42 stimulates actin filament assembly while scinderin severs the actin filaments. Both CDC42 and scinderin activities are related to PIP2 (phosphatidylinositol 4,5-bisphosphate): CDC42 mediates PIP2-induced actin polymerization⁴⁷ while PIP2 inhibits the actin-severing activity of scinderin.⁴³ Another player of interest in actin and cytoskeleton related processes and which promotes cell proliferation and invasion is AFAP1L2 (actin filament associated protein 1-like 2 also known as XB130)⁴⁸ encoded by AFAP1L2 gene that was recently found to interact with TERF1, one of the telomere maintenance genes, in melanoma risk.¹¹ All these observations provide strong support for our finding and outline the importance of actin dynamics-related genes in melanoma occurrence and progression. Further experimental studies are of course needed to provide better insight into the mechanisms involved.

In conclusion, this study shows that examination of the joint effects of multiple genetic variants and their interactions represents a powerful approach to disentangle the mechanisms underlying complex phenotypes. Indeed, it allowed not only identifying pathways enriched in genes associated with BT but also pointing out strong candidates underlying BT. Of particular interest is the significant interaction found between CDC42 and SCIN, which has important biological relevance given the key role of the proteins encoded by these genes in controlling cytoskeleton organization and actin dynamics and, consequently, in regulating melanoma invasion.

What’s new?

Breslow thickness (BT) is a major prognostic factor for melanoma. In this study, the authors present a comprehensive analysis strategy to identify the genetic factors influencing BT. By integrating genome-wide association, pathway and epistasis analyses, they identified three pathways associated with BT and showed significant interaction between variants at CDC42 and SCIN loci. This finding is biologically relevant, given the opposite effects of CDC42 and SCIN proteins in actin dynamics, a key mechanism underlying melanoma cell invasion.

Abbreviations

AFAP1L2

actin filament associated protein 1-like 2

BMP6

bone morphogenetic protein 6

BT

Breslow thickness

CDC42

cell division cycle 42

CEU

Northern Europeans from Utah

CM

cutaneous melanoma

CYFIP2

cytoplasmic FMR1 interacting protein 2

DKS

Domain Knowledge Score

IL1B

interleukin 1, beta

MC1R

melanocortin 1 receptor (alpha melanocyte stimulating hormone receptor)

MDSCs

myeloid-derived suppressor cells

NRG1

neuregulin 1

PIP2

phosphatidylinositol 4,5-bisphosphate

POMC

pro-opiomelanocortin

PRL

prolactin

SCIN

scinderin

TERF1

telomeric repeat binding factor (NIMA-interacting) 1

TNFSF10

tumor necrosis factor (ligand) superfamily, member 10

VEGFA

vascular endothelial growth factor A

α-MSH

melanocyte-stimulating hormone