{"paper_id":"04d36dff-7dfe-458b-b861-8650795734b9","body_text":"Functional characterization of a multi-cancer risk locus on chromosome band 2q33.1 1 \nnear CASP8 2 \n 3 \nHyunkyung Kong1, Jiyeon Choi1, Tongwu Zhang2, Cathrin Gräwe3, Mai Xu1, Rohit 4 \nThakur1, Hayley Sowards1, Rebecca C Hennessey1, Andrew Vu1, Jianxin Shi4, D Timothy 5 \nBishop5, Julia Newton-Bishop6, Jeremie Nsengimana7, Mark M. Iles6,8, Maria Teresa 6 \nLandi2, Michiel Vermeulen3,9, Matthew H. Law10,11,12, Laufey T. Amundadottir1, Melanoma Meta-7 \nAnalysis Consortium13, Kevin M Brown1 8 \n1Laboratory of Translational Genomics, Division of Cancer Epidemiology and Genetics, National 9 \nCancer Institute, Bethesda, MD, 20892, USA 10 \n2Integrative Tumor Epidemiology Branch, Division of Cancer Epidemiology and Genetics, 11 \nNational Cancer Institute, Bethesda, MD, 20892, USA 12 \n3Department of Molecular Biology, Radboud Institute for Molecular Life Sciences, Radboud 13 \nUniversity Nijmegen, Nijmegen, 6525 XZ, Netherlands 14 \n4Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD, 15 \n20892, USA 16 \n5Leeds Institute of Medical Research, School of Medicine, University of Leeds, Leeds, UK 17 \n6Leeds Institute for Data Analytics, University of Leeds, Leeds, UK 18 \n7Population Health Sciences Institute, Faculty of Medical Sciences, Newcastle University and 19 \nNIHR Biomedical Research Centre, Newcastle upon Tyne, UK 20 \n8NIHR Leeds Biomedical Research Centre, Leeds Teaching Hospitals NHS Trust, Leeds, UK 21 \n9Division of Molecular Genetics, Netherlands Cancer Institute, Amsterdam, Netherlands 22 \n10Population Health Department, QIMR Berghofer Medical Research Institute, Herston, QLD, 23 \nAustralia 24 \n11School of Biomedical Sciences, Faculty of Health, Queensland University of Technology, 25 \nBrisbane, QLD, Australia 26 \n12School of Biomedical Sciences, University of Queensland, Brisbane, QLD, Australia 27 \n 28 \nCorresponding author: kevin.brown3@nih.gov  29 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nAbstract 30 \nGenome-wide association studies (GWAS) of melanoma have identified numerous susceptibility 31 \nloci. However, causal genes and variants underlying risk have yet to be established for most. It 32 \nis becoming apparent that many functional variants underlying complex traits act via cis-33 \nregulation that may be context-specific, dependent on availability of specific transcription 34 \nfactors/complexes in specific cell types and cell-states. To characterize a risk locus on 35 \nchromosome band 2q33.1 associated with melanoma, breast cancer, and keratinocyte cancers, 36 \nwe integrated fine-mapping, cell-type specific expression quantitative trait locus (eQTL) analysis, 37 \na massively parallel reporter assay, individual luciferase assays, and SNP-based proteomics. 38 \nIntegrated analysis implicates the presence of multiple functional variants lying primarily within a 39 \npromoter for CASP8. A haplotype containing rs3769823 appeared have the largest effect on 40 \nexpression. Strikingly, both tumor/normal context and this risk-associated haplotype play critical 41 \nroles in mediating allelic cis-regulatory activity. Quantitative mass spectrometry for rs3769823 42 \nidentified both E4F1, a transcriptional repressor, and IRF2, a transcriptional activator, as binding 43 \npreferentially to risk-associated rs3969823-A. The binding of these transcription factors was 44 \nvalidated via EMSA, supershift, and chromatin immunoprecipitation (ChIP) assays. The relative 45 \nlevels of E4F1 and IRF2 differ by cell-type and play a role in mediating transcriptional activity in 46 \na cell-type specific manner. Our results indicate that the top credible causal set variant 47 \nrs3769823 likely influences expression of CASP8 and FLACC1 in a cell-type specific manner 48 \nand may be a relevant functional variant for multiple cancers associated with this locus. 49 \n 50 \nKeywords 51 \nMelanoma; genome-wide association study; GWAS; expression quantitative trait locus; eQTL; 52 \nfine-mapping; caspase 8; CASP8  53 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nIntroduction 54 \nOver the past decade, genome-wide association studies (GWAS) have identified 54 genome-55 \nwide significant susceptibility loci for cutaneous malignant melanoma (CMM) in populations of 56 \nEuropean ancestry 1-10. Of these, Barrett and colleagues initially identified a genome-wide 57 \nsignificant locus for melanoma on chromosome band 2q33.1 adjacent to the Caspase 8 gene 58 \n(CASP8) 5. Multiple subsequent melanoma GWAS meta-analyses since have replicated this 59 \nfinding, including the most recent meta-analysis of 36,760 melanoma cases (rs10931936-T, P = 60 \n2.12 × 10-12, OR = 1.08)10. This region is a multi-cancer risk locus, with highly-correlated risk 61 \nalleles for melanoma and other cancers sharing a common haplotype, including signals for 62 \nbreast cancer (rs3769821, P = 3.97 x 10-18, r2 with rs10931936 = 0.63) 11, keratinocyte cancers 63 \nincluding both cutaneous basal cell carcinoma (BCC, rs6714430, P = 5.42 x 10-55, r2 with 64 \nrs10931936 = 0.99), squamous cell carcinoma (SCC; rs10931936, P = 1.03 x 10-7), or both 65 \ncombined (rs6743068, P = 1.50 x 10-48, r2 with 10931936 = 0.99)12,13, and non-small cell lung 66 \ncancer (rs3769821, P = 4.45 x 10-8, r2 with rs10931936 = 0.63)14. Additionally, the region also 67 \nharbors a highly-correlated signal for prostate cancer (rs59308963, P = 2.41 x 10-8, r2 with 68 \nrs10931936 = 0.93)15, where the protective allele is highly correlated with risk alleles for 69 \nmelanoma and the other cancers.  However, whether causal variants and genes are shared 70 \nacross multiple cancers remains under-explored. 71 \nThe topologically-associated domain (TAD) at 2q33.1 harbors multiple strong a priori candidate 72 \ngenes with well-established roles in apoptosis, including Caspase-8 (CASP8), Caspase 10 73 \n(CASP10), and the CASP8 and FADD-like apoptosis regulator (CFLAR). CASP8 is an apical 74 \ncaspase that initiates apoptosis through the extrinsic pathway via a subset of proteins in the 75 \ntumor necrosis factor superfamily (TNFRSF) that includes TNFR1 (tumor necrosis factor 76 \nreceptor-1), CD95/Fas, and TRAIL-R (TNF-related apoptosis-inducing ligand receptor) 16,17. 77 \nCASP8 is composed of a catalytic domain, a protease domain and two death domains 78 \ncomprising the prodomain, which is important for forming the death-inducing signaling complex 79 \n(DISC), death receptor-FADD-CASP8 18,19. Like all caspases, CASP8 is present as a 80 \nmonomeric procaspase zymogen20. Binding of CASP8 to the DISC facilitates zymogen 81 \nmaturation by dimerization and proteolytic cleavage of the caspase itself 21,22. Fully processed 82 \nactive CASP8 is released from DISC and activates downstream effector caspases. Notably, two 83 \nproteins encoded by adjacent genes at this locus, CASP10 and CFLAR, are highly similar to the 84 \nN-terminus of CASP8 and play a different role in the apoptotic pathway by interacting with 85 \nFADD and CASP8 23,24. While CASP10 is an apical caspase like CASP8, CFLAR splice variants 86 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\ninhibit different processing steps of CASP8 activation through heterodimerization 24-26. Given 87 \nthat most common trait loci are thought to function via cis-gene regulation, often at long 88 \ndistances, other genes within this TAD cannot be ruled out as candidate causal genes. 89 \nIn this study, we fine-mapped multiple cancer GWAS datasets and integrate tissue-based and 90 \ncell-type specific quantitative trait locus (QTL) data in order to identify likely causal variants and 91 \ngenes. We identified multiple highly-correlated potentially cis-regulatory variants, and establish 92 \nrs3769823 as a key functional variant, likely acting via altered binding of the IRF2 and E4F1 93 \ntranscription factors. We establish cis-regulation of CASP8 as a likely causal mechanism 94 \nunderlying melanoma and breast cancer risk at this locus, and given the strong LD between 95 \nsignals, this mechanism may potentially play a role in risk of other associated cancers. 96 \n 97 \nResults 98 \nComparison of the 2q33.1 melanoma risk signal with GWAS of susceptibility to other 99 \ncancers 100 \nThe most recent GWAS meta-analyses of melanoma identified rs10931936 as the sentinel 101 \nvariant at the 2q33.1 risk locus (rs10931936-T, P = 2.12 × 10-12, OR = 1.08)10. Conditional and 102 \njoint analysis of summary GWAS meta-analysis data27,28 identified no additional genome-wide 103 \nsignificant association signals at this locus10. The next most significant signal within 1 Mb in 104 \neither direction of rs10931936 is rs563855920 (Pconditional = 3.05 x 10-4; Figure S1; Table S1) 105 \nmore than 300 kb away, suggesting that there are no additional major signals.  106 \nThis region has been identified as a genome-wide significant risk locus via GWAS of multiple 107 \nother cancer types including breast cancer (122,977 cases, rs3769821-C, P = 3.97 × 10-18, 1000 108 \nGenomes EUR r2 with rs10931936 = 0.63)11 and keratinocyte cancers (47,742 cases, 109 \nrs6743068-A, PMTAG = 1.50 × 10-48, r2\nrs10931936 = 0.99)13, including cutaneous BCC (31,787 cases, 110 \nrs6714430-C, P = 5.42 × 10-55, r2\nrs10931936\n = 0.99)13 and SCC (9,674 cases, rs10931936-T, P = 111 \n1.93 × 10-7)13 where the genotypes of these risk alleles are correlated. To formally assess 112 \nwhether these traits may share common causal variants with melanoma risk, we compared 113 \nmelanoma fine-mapping data to other cancer datasets using both HyPrColoc29 and eCAVIAR30 114 \nand summary statistic data for the set of SNPs +/- 100 kb centered over the melanoma GWAS 115 \nlead SNP. Colocalization analyses suggest a likely shared causal between melanoma risk and 116 \nbreast cancer (HyPrColoc PP = 0.93; maximum eCAVIAR CLPP = 0.04). Across the larger 117 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nregion, HyPrColoc finds the strongest evidence for shared causals for rs3769821 (HyPrColoc 118 \nSNP score = 0.52, r2 = 0.63 to rs10931936) and rs3769823 (SNP score = 0.11, r2 = 0.77 to 119 \nrs10931936) (Figure S2, Tables S2 and S3). Colocalization analyses of keratinocyte cancers 120 \n(both cutaneous BCC and SCC combined; PP = 0.99, maximum CLPP = 0.09), cutaneous BCC 121 \n(PP = 0.98; maximum CLPP = 0.02), and SCC (PP = 0.96; maximum CLPP = 0.02) all show 122 \nevidence for colocalization with melanoma risk, primarily showing the strongest evidence for 123 \nmelanoma lead SNP rs10931936 (SNP score is 0.42, 0.22 and 0.29, respectively for all 124 \nkeratinocyte cancers, BCC, and SCC) (Tables S2 and S3). LocusCompare31 plots of regional 125 \nsummary data for melanoma compared to keratinocyte cancers or BCC suggest that the signals 126 \nfor these non-melanoma skin cancers may be complex and involve multiple potential signals 127 \n(Figures S3-S5). Overall, these results suggest that a risk signal tagged by rs10931936 is 128 \ncommon to all tested cancers and nominate several potential shared causal variants. 129 \nFine-mapping of melanoma risk-associated variants at 2q33.1 130 \nTo comprehensively identify candidate causal variants within the 2q33.1 melanoma risk locus, 131 \nwe performed fine-mapping using multiple complementary approaches. Firstly, using GWAS 132 \nsummary data from Landi and colleagues10, we identified a set of 24 variants with a log-133 \nlikelihood ratio (LLR) of up to 1:1000 relative to the lead SNP at the locus (rs10931936; Figure 134 \n1, Table S4). We also performed Bayesian fine-mapping using DAP-G32,33, identifying two 135 \nclusters of potential candidate variants. The first cluster consisted of a 95% credible set of 15 136 \nvariants (set 1; marked by rs10931936, PIP = 0.16; Figure 1, Tables S6 and S7). A second 137 \ncluster consisted of a set of six variants with cumulatively low posterior inclusion probabilities 138 \n(PIP; set 2; led by rs77292590, PIP = 0.01; Tables S5 and S6) consistent with conditional 139 \nanalysis of the melanoma GWAS that suggested no major second signal at this locus. 140 \nConsidering the possibility of a cis-regulatory function for this locus, we also fine-mapped using 141 \nPAINTOR34,35, weighting the priors using melanocyte-specific epigenomic annotations from the 142 \nRoadMap Epigenome Project and locations of a set of genes specifically expressed in 143 \nmelanocytes36 (Figure 1, Table S7), identifying a 95% credible set of seven SNPs. Of these, 144 \nrs3769823, which was also fine-mapped by the other methods, had by far the highest posterior 145 \nprobability (0.789) of any variant. PAINTOR identified only one variant not in the DAP-G credible 146 \nset 1 (rs74574949, r2 = 0.02 with rs10931936, DAP-G credible set 3, PIP = 0.0002; Table S5). 147 \nLastly, we assessed whether other variants not successfully imputed in the meta-analysis, and 148 \nthus not directly fine-mapped, could also be credible causal variants. Here, we chose an LD-149 \nbased cutoff of r2 = 0.625 to the lead variant, identifying an additional nine potential candidate 150 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\ncausal variants. Combining variants from these analyses, we identified a total set of 27 credible 151 \ncausal variants for the melanoma risk signal (Figure 1, Table S8). 152 \n  153 \n 154 \nFigure 1. Fine-mapping of multiple melanoma GWAS signals on chromosome band 155 \n2q33.1. (Top) A view of the 2q33.1 locus (hg19) including all fine-mapped candidate causal 156 \nvariants nominated by log-likelihood ratio (LLR; red), Bayesian fine-mapping either unweighted 157 \n(DAP-G, green) or weighted using melanocyte-specific epigenomic annotations (PAINTOR, 158 \nbrown), or LD to the melanoma lead SNP (rs10931936, r2 ≥ 0.625, orange). Also shown are 159 \ntracks representing NCBI RefSeq genes and imputed chromatin state (ChromHMM) from two 160 \nprimary melanocyte cultures, two primary keratinocyte cultures, and two breast mammary 161 \ncultures generated by the Roadmap Epigenome Project (Red: Active TSS; Orange-Red: 162 \nFlanking Active TSS; Yellow: Enhancers; Green-Yellow: Genic enhancers; Green: Strong 163 \ntranscription, Dark Green: Weak transcription). (Bottom) A zoomed-in view of the same 164 \nannotations for a region encompassing select candidate causal variants. Data and rendering 165 \nwas performed using the UCSC Genome Browser. 166 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n 167 \nPotential molecular mechanisms underlying the 2q33.1 locus 168 \nOf these fine-mapped variants, only one, rs3769823, is protein-coding, being a missense variant 169 \nof codon 14 (K14R) of a single CASP8 isoform (isoform G; ENST00000358485.4; RefSeq 170 \nNM_001080125.2; Figures 1 and S6), where the allele encoding K14 is associated with 171 \nincreased risk. We assessed potential functions via Jpred4, which suggested that CASP8K14R 172 \nmay be located within a helical structure (Figure S7). To assess potential impacts on protein 173 \nfunction, we used five different in silico prediction tools via PredictSNP2 37. Of these, the 174 \nmajority (4/5) predicted rs3769823 to be benign; FunSeq2 however suggested that this variant 175 \nmay be deleterious (Table S9). While we cannot rule out a protein-coding function for 176 \nrs3769823, these data suggest rs3769823 is benign. 177 \nWe also assessed splicing QTLs (sQTL) from primary human melanocytes38. Notably, a 178 \nsignificant risk-associated SNP, rs10804111 (PGWAS = 6.79 × 10-10, LD to rs10931936 r2 = 0.62 179 \nand D’ = 1; Table S1), is a significant sQTL for alternative splicing of the CASP8 exon 8 to 9 180 \njunction (PsQTL = 7.33 × 10-12; Figure S8). Specifically, this generates an alternative CASP8 181 \ntranscript (isoform H; ENST00000339403.6, RefSeq NR_111983.2; Figures 1 and S6), 182 \nretaining part of intron 8 and leading to frameshift and premature termination of the CASP8 183 \ntranscript, which may act as a dominant-negative inhibitor of the caspase cascade39. Here, 184 \nrs10804111-C is associated with higher melanoma risk and lower levels of the alternative 185 \nisoform H (Figure S8B). To test the potential of rs10804111 to mediate alternative splicing, we 186 \ngenerated C- or T- allele-specific mini-genes (Figure S9A, see material and methods). Using 187 \nqRT-PCR in melanoma cells, we confirmed the association between rs10804111-C and lower 188 \nusage of the exon junction for isoform H (exon 8-9 junction, P = 7.23 x 10-5; exon 8L-9 junction, 189 \nP = 1.78 x 10-3; Figure S9B). However, we note fine-mapping of the melanoma GWAS signal 190 \ndoes not identify rs10804111 as a member of any credible causal set at this locus, and formal 191 \ncolocalization between the melanoma risk signal and melanocyte sQTL signal did not provide 192 \nsignificant evidence of colocalization. 193 \nGiven the lack of strong support for protein-coding or splice variants explaining the 2q33.1 194 \nsignal, as well as a clustering of fine-mapped variants within regions marked as promoter or 195 \nenhancer in melanocytes (Figure 1, Table S8), we turned to eQTL analysis to explore the 196 \npotential of this signal being explained by cis-gene regulation. We first investigated previously-197 \npublished cell-type specific eQTL data generated from primary cultures of human melanocytes 198 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n(n=106; Table S10)36. While multiple genes within the TAD harboring the association signal had 199 \nnominally significant eQTL, CASP8 is the only significant eQTL gene for the lead melanoma 200 \nGWAS SNP (rs10931936; PCASP8 eQTL = 1.2 x 10-9, slope = 0.67, where lower CASP8 expression 201 \nassociated with the T-risk allele; Figure 2A, Table S10)36. Conditional analysis of melanocyte 202 \nCASP8 eQTL identified multiple variants comprising only a single eQTL signal within this locus 203 \n(Table S11). Colocalization analysis of the CASP8 eQTL and melanoma GWAS summary data 204 \nusing both HyPrColoc29 and eCAVIAR30 suggest CASP8 as a strong candidate gene 205 \n(HyPrColoc PP = 0.97, eCAVIAR colocalization posterior probability/CLPP = 0.09 for rs3769823) 206 \nwith the strongest evidence of shared causal variants for rs3769823 (HyPrColoc SNP score = 207 \n0.86, eCAVIAR CLPP = 0.09, r2 to rs10931936 = 0.77) and rs3769821 (HyPrColoc SNP score = 208 \n0.14, eCAVIAR CLPP = 0.03, r2 to rs10931936 = 0.63) (Figure 3, Table S12 and S13). 209 \nCollectively these results suggest a shared common causal variant between melanoma risk and 210 \nmelanocyte-specific cis-regulation of CASP8. While no other genes had nominally significant 211 \nmelanocyte eQTLs for the sentinel melanoma risk SNP rs10931936, other strongly risk-212 \nassociated SNPs at this locus were, notably for the CASP8-adjacent gene FLACC1 (rs3769821, 213 \nP = 2.42 x 10-5, r2 to rs10931936 = 0.63; lead FLACC1 eQTL SNP rs796181752, P = 9.01 x 10-6; 214 \nTable S10), where the direction of effect relative to risk allele is opposite to that for CASP8. 215 \nColocalization using HyPrColoc29 and eCAVIAR30 give discordant results, with eCAVIAR 216 \nsuggesting potential eQTL colocalization (CLPP = 0.04 for rs3769821; Table S13) while 217 \nHyPrColoc does not (HyPrColoc PP < 0.5; Table S12). A LocusCompare31 plot suggests cis-218 \nregulation of FLACC1 in melanocytes may be complex and involve multiple potential signals 219 \npotentially explaining the discordant results (Figure S10). Collectively, these data suggest that 220 \nthe melanoma GWAS signal may be explained by cis-regulation of CASP8 and FLACC1, with 221 \nopposite directions of effect for the two genes. 222 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n 223 \nFigure 2. The melanoma risk-associated T allele of rs10931936 is correlated with lower 224 \nCASP8 expression. eQTL analyses were performed for rs10931936 combining genotype and 225 \nexpression level data derived from (A) 106 human primary melanocyte cultures (TT, n=7; TC, 226 \nn=39; CC, n=60), (B) 349 melanoma tumors from TCGA-SKCM (TT, n=42; TC, n=126; CC, 227 \nn=181), and (C) a panel of 59 early-passage melanoma cell lines (TT, n=8; TC, n=23; CC, 228 \nn=25), where the risk-T allele is labeled in red. A significant eQTL effect with higher expression 229 \ndriven from the C-protective allele was observed for CASP8 in both melanocytes and melanoma 230 \ntumors (P = 1.2 x 10-9 and P = 6.8 x 10-3, respectively), while the result was marginal but in the 231 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nsame direction in melanoma cell lines (P = 0.07). Significance determined by linear regression; 232 \nmean with SEM are plotted along with individual data values.  233 \n 234 \n 235 \nFigure 3. CASP8 eQTL in human primary melanocyte cultures is colocalized with the 236 \nmelanoma GWAS. (A) LocusZoom plots present –log10 P-values for melanoma GWAS (upper) 237 \nand melanocyte CASP8 eQTL (lower) for a 1 Mb region encompassing rs10931936. The 238 \nmelanoma risk lead SNP rs10931936 is labeled and highlighted in purple in both panels, and LD 239 \n(r2 based on 1000G EUR) of all other SNPs to the melanoma GWAS lead SNP is color-coded. 240 \n(B) A LocusCompare plot compares P-values between melanoma GWAS (x-axis) and 241 \nmelanocyte CASP8 eQTL (y axis) for the same region. Genomic coordinates are based on hg19. 242 \n 243 \nThese data are consistent with results from a transcriptome wide association study (TWAS) 244 \nstudy we previously performed using the same GWAS summary statistics and melanocyte 245 \neQTL data10, where only CASP8 is significant and FLACC1 is marginal (PCASP8 = 5.57 x 10-12; 246 \nPFLACC1 = 3.83 x 10-5; Table S14). Assessing models trained on each of 47 tissue types in the 247 \nGTEx v.7; GTEx portal (https://gtexportal.org)40 identified CASP8 as a significant TWAS gene in 248 \n30 out of 47 GTEx tissues, strikingly with the direction of effect for six tissues being opposite to 249 \nthat observed for melanocytes, including multiple brain tissues and pituitary (Table S15)10. 250 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nTWAS across GTEx tissues likewise found a significant positive association between imputed 251 \nFLACC1 levels and risk in 37 tissues (Table S16). We also assessed eQTLs in both 349 252 \nmelanoma tumors from TCGA-SKCM36,41,42 as well as 59 early-passage melanoma cell lines 253 \nusing expression microarray data43. In TCGA melanomas, we observed an FDR-significant 254 \neQTL for FLACC1 (P = 6.57 x 10-4), as well marginal eQTLs for CASP8 (P = 6.82 x 10-3; Figure 255 \n2B) and TRAK2 (P = 0.03) (Table S17). In melanoma cultures, the rs10931936 risk-T allele is 256 \nmarginally associated with multiple transcripts, including CASP8 (P = 0.07; Figure 2C, Table 257 \nS18). Both CASP8 eQTLs were in the same direction as observed in melanocytes. 258 \nFinally, given potential colocalization of melanoma risk with signals for breast cancer and 259 \ncutaneous SCC, we also assessed candidate genes specifically in mammary and skin tissues 260 \nprofiled by GTEx v.740. Colocalization analyses with the melanoma risk signal suggests 261 \nmelanoma risk may share common causal variants with eQTLs for CASP8 (eCAVIAR CLPP = 262 \n0.012 for rs3769821) and FLACC1 (eCAVIAR CLPP = 0.05 for rs3769823) in mammary tissues, 263 \nand FLACC1 in skin-not-sun-exposed (HyPrColoc PP = 0.97; SNP score = 0.62 and CLPP = 264 \n0.013 for rs3769823) and skin-sun-exposed tissues (HyPrColoc PP = 0.79; SNP score = 0.15 265 \nand CLPP = 0.012 for rs3769823) (Tables S19 and S20). TWAS of melanoma risk using 266 \nmammary and skin tissues found both CASP8 and FLACC1 to be significant TWAS genes with 267 \nthe same direction of effect observed in melanocytes (Tables S15 and S16). These results 268 \nsuggest that the 2q33.1 melanoma risk locus may not only share common causal variants with 269 \nbreast and other cancers, but also with cis-regulatory variants influencing CASP8 and FLACC1 270 \nlevels in breast and skin tissues, suggesting a potential shared cis-regulatory etiology for 271 \nmultiple cancers. 272 \nPrioritization of potential cis-regulatory variants using epigenomic data 273 \nGiven the strong evidence for the 2q33.1 GWAS signals for melanoma as well as perhaps other 274 \ncancers being explained by cis-regulation of CASP8 or potentially FLACC1, we next prioritized 275 \ncredible set variants for assessment of allelic cis-regulatory activity. We prioritized for further 276 \nstudy eight variants with active imputed ChromHMM chromatin states consistent with 277 \nmelanocyte gene regulatory regions as annotated by Regulome DB and HaploReg (Figure 1, 278 \nTable S8; individual melanocyte regulatory marks are shown in  Figure S11)44. Most of these 279 \neight variants also overlapped H3K27ac histone marks and or open chromatin (FAIRE-seq) 280 \nfrom a panel 11 melanoma cultures45 (Figure S12). Given the shared GWAS signal between 281 \nmelanoma, breast cancer, and keratinocyte cancers, we additionally prioritized fine-mapped 282 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nvariants that overlap regulatory imputed ChromHMM states in primary keratinocytes and 283 \nmammary epithelial tissues (Table S8); all such variants were already prioritized from 284 \nmelanocyte ChromHMM data. Thus, we proceeded to move forward with eight fine-mapped 285 \nvariants for functional assessment. Of these eight, three variants in close proximity (rs3769823, 286 \nrs3769821, rs59308963, within 641 bp), are located in a single continuous region of active 287 \ntranscription start site (TSS)-proximal promoter state (TssA, Table S8)44 in melanocytes, 288 \nkeratinocytes, and mammary epithelial cells, overlapping the first exon and first intron of two 289 \nannotated CASP8 isoforms, isoform G and H (Figures S11 and S12), potentially representing a 290 \ncis-regulatory haplotype. 291 \nIdentification of functional cis-regulatory fine-mapped variants 292 \nTo identify potentially functional cis-regulatory variants from these eight candidates, we 293 \nperformed individual reporter assays for each in the context of ~140 bp of DNA sequence 294 \nsurrounding the variant cloned in both forward and reverse orientations and tested in two 295 \ndifferent melanoma cell lines (summarized in Table S21). Only one variant, rs3769823, showed 296 \na significant allelic difference in both forward and reverse orientations in both cell lines (Figures 297 \n4A and S13, Table S21), albeit with the opposite direction of effect relative to the melanocyte 298 \nCASP8 eQTL. rs3769821 also had a significant albeit weaker allelic effect in the reverse 299 \ndirection in both cell lines with the direction of effect matching the CASP8 eQTL, while in one 300 \ncell line in the forward direction there we noted a reproducible allelic difference in the opposite 301 \ndirection (Figure S14, Table S21). Lastly, two other variants, rs59308963 and rs1861270 302 \n(Table S21), showed significant but weak allelic difference across cell lines only in the reverse 303 \norientation (Figure S15, Table S21), with the effect direction for only rs59308963 matching the 304 \nCASP8 eQTL. We also examined two previously published melanoma MPRA studies which 305 \nassayed five of these variants in human melanoma cells as well as HEK293T kidney cells 306 \n(MPRA v.146; Figure S16), or six of these variants in both melanoma cells and human primary 307 \nmelanocytes (MPRA v.2)47 with both studies testing a similar insert size (Table S21). Of the 308 \neight variants tested in either study, only rs3769823 showed an FDR-significant (FDR < 0.01) 309 \nallelic effect in melanoma cells (PMPRA v.1 = 1.86 x 10-12, PMPRA v.2 = 1.09 x 10-52; Table S21), 310 \nwhere the direction of effect was consistent with data from individual reporter assays and where 311 \nthe allelic effect is opposite to the CASP8 eQTL. Likewise, in the second MPRA study, of the six 312 \nvariants tested, only rs3769823 was FDR-significant in primary melanocytes (PMPRA v.2 = 3.96 x 313 \n10-9), albeit with a smaller effect size (beta = -0.11) as compared to that observed in melanoma 314 \ncells (betaMPRA v.1 = -0.65; betaMPRA v.2 = -0.45) (Table S21). Of the remaining variants, 315 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nrs3769821 was significant only in a joint analysis of data from both melanoma and HEK293T 316 \nkidney cells (PMPRA v.1 = 3.11 x 10-4; Table S21) with direction of effect matching that observed in 317 \nindividual luciferase assays (reverse orientation) and consistent with the CASP8 eQTL. Given 318 \nthe potential of shared signal in this region with breast cancer, we also tested three variants, 319 \nrs3769823, rs3769821, rs59308963, in multiple breast cancer cell lines replicating the 320 \nsignificant allelic effect and direction observed in melanoma for rs3769823, but with inconsistent 321 \nresults for the other two variants (Figures S13-S15). Collectively, these data suggest the 322 \npotential for multiple cis-regulatory variants, with risk-associated allelic effects in differing 323 \ndirections and in some cases possibly dependent on cellular context, with the strongest 324 \nevidence for rs3769823. 325 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n 326 \nFigure 4. The melanoma-associated rs3769823 is a functional cis-regulatory variant and 327 \ndisplays allele-preferential binding by E4F1 and IRF2. (A) Individual luciferase reporter 328 \nactivity assays for rs3769823 were conducted using the melanoma cell line UACC903. 138 bp 329 \nsequences encompassing rs3769823 construct were cloned 5’ of the pGL4.23 minimal TATA 330 \npromoter and transfected. Luciferase activity was measured 24 h after transfection and was 331 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nnormalized against Renilla luciferase activity. One representative set from three biological 332 \nreplicate experiments is shown. Mean with SEM are plotted. Individual P values are shown for 333 \nA-risk allele versus G-protective allele for the replicate shown (two-tailed, unpaired t-test 334 \nassuming unequal variances). TATA, minimal promoter control; A, risk allele construct marked 335 \nin red; G, protective allele construct. (B) EMSAs were performed using 21 bp biotin-labeled 336 \ndouble-stranded oligonucleotides for the A-risk (red) or G-protective (black) alleles of rs3769823 337 \nand nuclear extract from UACC903 melanoma cells. 2x, 5x, or 10x molar excess of unlabeled 338 \ncompetitor was added in specified lanes. One representative set from three replicate 339 \nexperiments is shown. (C) Allele-specific rs3769823 binding proteins were identified by 340 \nquantitative mass spectrometry using UACC903 melanoma nuclear extracts and 21 bp 341 \nbiotinylated double-stranded DNA probed with A-risk or G-protective alleles. The dimethyl-342 \nlabeling ratios of proteins bound to A/G or G/A probes are plotted on the x and y axes for label-343 \nswapping experiments. Red-filled circles highlight proteins enriched above the background in 344 \nboth experiments. (D) Antibody super-shift assay of rs3769823 EMSA in UACC903 melanoma 345 \ncells using anti-E4F1 or anti-IRF2 antibody is shown, where the A-risk specific shifted band 346 \n(arrows) is diminished. One representative set from three independent experiments is shown. (E) 347 \nChIP was performed using antibody to anti-E4F1, anti-IRF2 or anti-IgG and chromatin prepared 348 \nfrom UACC903 melanoma cells, followed by qPCR. DNA quantity was normalized to input DNA 349 \nfor each immunoprecipitation. Mean of qPCR triplicates with SEM are plotted and P-values 350 \nshown for one representative experiment from three biological replicates. 351 \n 352 \nThree of these variants are in close proximity and located within a common melanocyte 353 \nenhancer element, and individual luciferase assays for all three reflect enhancer activity well 354 \nabove empty vector controls. We speculated that these may influence enhancer function in the 355 \ncontext of a haplotype and thus performed reporter assays for a larger 641 bp region harboring 356 \nall three variants in the context of the four most common haplotypes observed in the 1000 357 \nGenomes EUR sample set. Here, the haplotype harboring risk alleles of all three represent 27% 358 \nof alleles, which that containing protective alleles for each is 64% (Table S22). Comparing the 359 \nall-risk (A/C/ATTCTGTC for rs3769823, rs3769821, and rs59308963) to all-protective (G/T/-) 360 \nhaplotypes in melanocytes, we observed significantly lower reporter activity for the risk 361 \nhaplotype in the forward direction (cloned relative to the direction of CASP8 transcription 362 \n(Figure S17A). This direction matches the eQTL direction of effect for rs3769823 and CASP8 in 363 \nmelanocytes, melanomas, skin, and mammary tissues (Figure S18A-F). In contrast, reporter 364 \nassays in melanoma cells and breast cancer cells showed the opposite direction of effect, with 365 \nhigher reporter expression driven from the all-risk haplotype (Figures S17B, C). These data 366 \nsuggest a complex cis-regulatory architecture overlapping annotated melanocyte promoter for 367 \nmultiple CASP8 isoforms, that allelic patterns of regulation may be highly dependent on both 368 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nadjacent functional variants as well as cellular context, and where the observed direction of 369 \neffect in melanocytes in vitro is consistent with CASP8 as plausible causal gene. 370 \nThe genomic region containing rs3769823, rs3769821 and rs59308963 physically 371 \ninteracts with FLACC1 372 \nGiven that rs3769823, rs3769821, rs59308963 are all correlated with not only CASP8 373 \nexpression but also FLACC1 in melanocytes (rs3769823, P = 7.14 x 10-4, slope = -0.23, Figure 374 \nS18A; rs3769821, P = 2.42 x 10-5, slope = -0.28; rs59308963, P = 0.01, slope = -0.19; Table 375 \nS10), we investigated whether these or other credible causal variants physically interact with 376 \nFLACC1. We first assessed interactions using a region-specific Capture-C assay performed in 377 \nfive independent human primary melanocyte cultures, where capture baits were tiled across the 378 \nentire region of association 48,49 at 2q33.1. Here, we observed a direct physical contact between 379 \nthe restriction fragments containing these three variants and the FLACC1 promoter (Figure 380 \nS19A, Table S23 and S24). We did not observe physical interaction between these variants and 381 \nany other gene within the TAD containing the association signal. We further assessed this 382 \ninteraction using chromatin conformation capture (3C) in two independent melanocyte cultures 383 \nas well as the UACC903 melanoma cell line, using the region harboring these three variants as 384 \nbait and using primers spanning CASP8 and FLACC1. Here, we confirmed a physical 385 \ninteraction between these variants and the FLACC1 promoter in both melanocyte cultures, with 386 \nlittle evidence of a strong association in melanoma cells (Figure S19B). These data suggest 387 \nthat the transcriptional regulatory region harboring rs3769823, rs3769821, and rs59308963 388 \ncould potentially play a role in regulation not only of CASP8, but also FLACC1. 389 \nMultiple transcription factors bind to rs3769823 in an allele-preferential manner 390 \nGiven potential cis-regulatory roles for rs3769823, rs3769821, rs59308963, we further assessed 391 \nwhether these variants influence the binding of nuclear proteins to the sequences 392 \nencompassing them. We first performed electromobility shift assays (EMSAs) using nuclear 393 \nextracts from melanoma cell lines, primary melanocytes, or breast cancer cell lines. We 394 \nobserved preferential binding of nuclear protein to the A-risk allele compared to G-protective 395 \nallele of rs3769823 across all cell lines tested, where competition with unlabeled probe 396 \ncontaining the A-risk allele is stronger than that containing the G allele, demonstrating specificity 397 \n(Figures 4B and S20). For rs3769821, we observed suggestive evidence of preferential binding 398 \nof nuclear protein to the T-protective allele compared to C-risk allele in melanocytes, melanoma, 399 \nand breast cancer cell lines, albeit with weak specificity (Figure S21A). Finally, for rs59308963, 400 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nEMSA results show preferential binding to the insertion-risk allele compared to deletion allele 401 \nwith weak specificity in melanocytes, melanoma, and breast cancer cells (Figure S22A). These 402 \ndata suggest a strong and consistent pattern of allelic protein binding to rs3769823, consistent 403 \nwith the large effect observed in reporter assays, and weaker evidence for allelic binding for the 404 \nother two variants. 405 \nWe also performed quantitative mass spectrometry with oligonucleotides corresponding to the 406 \nrisk or the protective allele of each of these variants incubated with nuclear extracts from 407 \nmelanoma or breast cancer cell lines to directly identify allele preferential binding proteins. For 408 \nrs3769823, we noted shared allele-specific binding proteins between both cell lines; specifically, 409 \nIRF2 and E4F1 were observed to preferentially bind the A-risk allele, while REST and POU2F1 410 \nbound the G-protective allele (Figures 4C and S23A). The data for A-specific binding proteins 411 \nare consistent with the patterns observed from the EMSA data, suggesting E4F1 or IRF2 as 412 \npotential risk-allele binding transcription factors. Further, sequence-based motif prediction was 413 \nalso consistent with these mass spectrometry data, indicating that the sequence around 414 \nrs3769823 forms a consensus binding site for both E4F1 and IRF2, favoring binding to the A-415 \nrisk allele (Figure S24). For rs3769821, most of the potential preferential binding was found to 416 \nbe the T-protective allele via EMSA, however mass spectrometry found few T-specific binders 417 \nand no overlapping T-binding proteins between cells; PRDM14 and ZBTB9 were both found to 418 \npreferentially bind the C-allele (Figure S21B). We also identified BNC2 as a strong risk-419 \ninsertion binding protein to the risk-associated allele of rs59308963 in melanoma cells (Figure 420 \nS22B).  421 \nBased on the clear results and strong allelic pattern of cis-regulatory activity for rs3769823, we 422 \nsought to verify whether E4F1 or IRF2 proteins bound the A-allele by using antibodies against 423 \nthese proteins in conjunction with EMSAs. Antibodies against either E4F1 or IRF2 consistently 424 \nresulted in loss of A allele-specific protein binding in both melanoma and breast cancer cell lines 425 \n(Figures 4D and S23B). To further establish the binding of E4F1 or IRF2 to rs3769823, we 426 \nperformed chromatin immunoprecipitation (ChIP) for E4F1 or IRF2 followed by quantitative PCR, 427 \nnoting an enrichment of binding at rs3769823 in melanoma cell lines with homozygous for 428 \nrs3769823-A as well as primary melanocytes heterozygous for this variant (Figures 4E and 429 \nS25B). We assessed enrichment at other locations across the CASP8 gene that had previously 430 \nbeen shown via ChIP-seq to be bound by IRF2 or E4F1, and DHS data from ENCODE (Figure 431 \nS25A), however ChIP only showed enrichment of both in the region over rs3769823 432 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n(UACC903(A/A), PE4F1 = 1.46 x 10-3, PIRF2 = 0.016; C87(A/G), PE4F1 = 3.25 x 10-3, PIRF2 = 9.47 x 433 \n10-3; Figure S25B). Using the heterozygous melanocyte culture, we also assessed allelic 434 \nenrichment in the immunoprecipitants using quantitative real-time PCR genotyping assay and 435 \nnoted a significant enrichment of the A allele as compared to the G allele at rs3769823 in 436 \nprimary melanocytes for both transcription factors (Figure S25C). Together, these data clearly 437 \nsuggest that E4F1 and IRF2 may act through rs3769823 at this locus. 438 \nE4F1 and IRF2 as potential repressor and activator of CASP8 transcription, respectively 439 \nGiven the allelic binding observed for E4F1 and IRF2 to the A-risk allele of cis-regulatory SNP 440 \nrs3769823, we next explored relationships between levels of these transcription factors and 441 \nexpression of CASP8 and FLACC1 in various cell and tissue types. In primary melanocytes, we 442 \nobserve clear negative correlation between both E4F1 mRNA levels with CASP8 (E4F1, P = 443 \n3.80 x 10-6, Pearson r = -0.43; Figure S26, Table S25) while IRF2 is not significant (IRF2, P = 444 \n0.39, r = 0.08; Figure S26, Table S25); multiple linear regression shows both rs3769823 and 445 \nE4F1 to be significant (E4F1 P = 1.70 x 10-15; rs3769823 P = 5.39 x 10-7; IRF2 P = 0.43; Table 446 \nS26). These data are consistent with a known role for E4F1 as a transcriptional repressor50 and 447 \nsuggests a role in regulation of CASP8. This direction of effect is consistent with transcriptional 448 \nactivity observed in reporter assays in melanocytes potentially being attributable in part of a 449 \nrepressive effect for E4F1 when considering the longer three-variant risk haplotype, but not the 450 \nshort (138 bp) fragment assays considering rs3769823 alone. In contrast, in TCGA melanoma 451 \ntumors, while significant negative correlation with E4F1 was observed (E4F1, P = 2.26 x 10-4, r 452 \n= -0.20; Figure S26, Table S25), a much stronger positive with CASP8 was observed for IRF2 453 \n(IRF2, P = 1.69 x 10-27, r = 0.54; Figure S26, Table S25); both transcription factors and 454 \nrs3769823 are significant via multiple linear regression (E4F1 P = 0.0018; IRF2 P = 2 x 10-16; 455 \nrs3769823 P = 0.035; Table S27), again with the strongest effect for IRF2. We did not find 456 \nsignificant evidence for an interaction between rs3769823 and either of the transcription factors 457 \n(Table S27). We were able to replicate this positive correlation specifically between IRF2 and 458 \nCASP8 in independent panels of melanoma cell lines (P = 4.23 x 10-4, r = 0.38; Table S25) and 459 \nthe Leeds Melanoma Cohort (P = 8.74 x 10-4, r = 0.30; Table S25). This direction of effect is 460 \nconsistent with a potential role for IRF2 as a potential activator in both individual and haplotype-461 \nbased reporter assays, but inconsistent with the direction of the relatively weak eQTL for 462 \nCASP8 in melanoma tumors. We formally tested for potential interactions between E4F1 and 463 \nrs3769823 in melanocyte expression data as well as between E4F1 and/or IRF2 and rs3769823 464 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nin TCGA melanoma tumors but did not find significant supporting evidence (Tables S26 and 465 \nS27).  466 \nWe also assessed transcripts encoding two T-protective allele binding proteins for correlations 467 \nwith potential target genes, observing a positive correlation between the transcriptional 468 \nrepressor REST, as well as POU2F1, and CASP8 in melanocytes (REST, P = 7.21 x 10-11; r = 469 \n0.58; POU2F1, P = 8.40 x 10-3, r = 0.25), a weak negative correlation for POU2F1 in TCGA 470 \nmelanomas (P = 0.04, r = -0.11), and a positive correlation for REST in the Leeds Melanoma 471 \nCohort (P = 1.75 x 10-14, r = 0.30) (Table S25). This positive correlation with CASP8 in 472 \nmelanocytes and some melanoma tumors is inconsistent with the established role for REST as 473 \na transcriptional repressor. Still, this direction of effect is consistent with a potential role for 474 \nREST repressing reporter activity in both individual (short fragment) reporter assays for 475 \nrs3769823 as well as both previously-reported MPRA assays. Multiple linear regression 476 \nconsidering all three individually correlated risk-A and protective-G binding alleles and genotype 477 \nfor rs3769823 in melanocytes suggests all four are significant (REST P = 6.71 x 10-7; rs3769823 478 \nP = 4.29 x 10-6; POU2F1 P = 0.0019; E4F1 P = 0.010; Table S28). We note similarly complex 479 \npatterns of correlation between both risk- and protective- allele rs3769823 binding proteins in 480 \nnon-melanocytic tissues (GTEx skin, GTEx breast mammary, and TCGA breast invasive 481 \ncarcinoma; Table S25) and expression of CASP8 as well as FLACC1 (Table S25). These data 482 \nare consistent with the activity of the regulatory region encompassing rs3769823 being 483 \nmodulated by a complex interplay of multiple allele-specific transcriptional regulators interacting 484 \nwith risk and protective alleles in a tissue-specific manner. 485 \nFinally, we also investigated correlations between CASP8 and allele-specific binding proteins for 486 \nrs3769821 (Table S29) and rs59308963 (Table S30), observing negative correlations with the 487 \ntranscripts for rs3769821 risk-allele binding protein ZBTB9 and the rs59308963 risk-allele 488 \nbinding protein BNC2 (ZBTB9, TCGA P = 2.15 x 10-11, TCGA r = -0.35, Leeds P = 3.40 x 10-3, 489 \nLeeds r = -0.12; BNC2, TCGA P = 2.70 x 10-4, TCGA r = -0.19, Leeds P = 1.43 x 10-13, Leeds r 490 \n= -0.29; Tables S29 and S30). For FLACC1, we note similar correlations in melanocytes to 491 \nthose observed for CASP8, with negative correlation observed for E4F1 (P = 3.40 x 10-6, r = -492 \n0.43), and positive correlations observed for REST and POU2F1 (REST, P = 1.72 x 10-10, r = 493 \n0.57; POU2F1, P = 4.71 x 10-12, r = 0.61), but do not observe consistent correlations for IRF2 or 494 \nREST across melanoma datasets (Table S25). Similar to what is observed for rs3769823, these 495 \ncorrelations vary considerably between melanocytic and non-melanocytic tissues (Tables S29 496 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nand S30) observing similarly complex patterns of correlation. Collectively, these data suggest a 497 \npotentially complex gene regulatory interplay, prominently of multiple allele-specific rs3769823 498 \nbinding proteins, as well as other transcriptional regulators, some of which may bind this region 499 \ndifferentially in risk and protective haplotypes.  500 \nTo determine the effects of depleting the two rs3769823 A-risk allele-specific binding proteins 501 \non CASP8 transcription in melanoma, primary melanocyte cultures, and breast cancer cell lines, 502 \nwe knocked these genes down using siRNAs. Knockdown of E4F1 resulted in an increase in 503 \nCASP8 levels while IRF2 knockdown results in reduction of CASP8 RNA levels in melanoma 504 \n(Figure 5A and 5B, S27A and S27B), breast cancer (Figure S28A and S28B), and 505 \nmelanocytes (Figure S29A), consistent with respective roles for E4F1 and IRF2 as a potential 506 \nrepressor and activator in cells of melanocytic lineage. Short fragment (138 bp) reporter assays 507 \nfor rs3769823 in conjunction with knockdown of A-risk allele binding protein IRF2 resulted in a 508 \nsignificant decrease of expression driven primarily from the A-risk allele compared to the 509 \nprotective G-allele in both forward and reverse directions in melanoma (UACC903 cells, Pforward 510 \n= 1.22 x 10-7, Preverse = 2.84 x 10-8, Figure 5C; UACC1113, Pforward = 1.65 x 10-5, Preverse = 2.20 x 511 \n10-3, Figure S27C) and breast cancer cell lines (T47D, Pforward = 2.92 x 10-7, Preverse =3.18 x 10-7, 512 \nFigure S28C), consistent with a potentially prominent role for IRF2 in CASP8 regulation in 513 \nmelanoma cells. Consistent with the weaker correlations between E4F1 and CASP8 in 514 \nmelanoma cells, E4F1 knockdown only modestly increased reporter activity for the A-risk allele, 515 \nand only in the forward direction, suggesting a potentially weak repressor role for E4F1 in the 516 \ncontext of melanoma cells (UACC903, Pforward = 8.05 x 10-4, Preverse = 0.74, Figure 5C; 517 \nUACC1113, Pforward = 3.47 x 10-3, Preverse = 0.02, Figure S27C). In primary melanocytes, reporter 518 \nassays for rs3769823 with knockdown of A-risk-allele binding E4F1 resulted in a significant 519 \nincrease of promoter activity in the A-risk allele compared to the protective G-allele in both 520 \ndirection (C23 melanocytes, Pforward = 1.27 x 10-9, Preverse = 0.013, Figure S29B), suggesting a 521 \npotentially repressor role for E4F1 in regulation of CASP8 in melanocytes.  522 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n 523 \n 524 \nFigure 5. E4F1 or IRF2 and rs3769823 regulate CASP8 expression in UACC903 melanoma 525 \ncells. (A) E4F1 or IRF2 were knocked down using respective pools of four different siRNAs in 526 \nUACC903 melanoma cells, and E4F1, IRF2 and CASP8 levels were measured. GAPDH-527 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nnormalized E4F1, IRF2 or CASP8 mRNA levels are shown as fold change over those from non-528 \ntargeting siRNA. A representative experiment from three biological replicates is shown 529 \n(individual datapoints, mean, and SEM are plotted). P-values are shown from one 530 \nrepresentative set. (B) Protein levels were examined using anti-E4F1, anti-IRF2, or anti-GAPDH 531 \nantibody with UACC903 cell lysates of siRNAs transfected with either non-targeting control, 532 \nE4F1, or IRF2 siRNAs. GAPDH was used as a loading control. One representative set of three 533 \nreplicate experiments is shown. (C) Individual luciferase assays were performed by co-534 \ntransfecting rs3769823 luciferase constructs with E4F1, IRF2, or non-targeting control siRNAs 535 \ninto UACC903 melanoma cells. Renilla-normalized relative luciferase activities are shown 536 \nrelative to an empty construct containing only a minimal promoter (TATA). One representative 537 \nset is shown from three biological replicates. Mean with SEM is plotted, n = 6 technical 538 \nreplicates. A two-tailed t-test assuming unequal variances was used to calculate all P values 539 \nshown against control siRNA. 540 \n 541 \n 542 \nDiscussion  543 \nIn this study, we focused on a melanoma susceptibility locus over the CASP8 gene on 2q33.1 544 \nthat was originally identified by Barrett and colleagues5 and has since been replicated in a large 545 \nmelanoma GWAS meta-analysis (rs10931936-T, Pfixed = 2.12 x 10-12, Prandom = 2.17 x 10-8, OR = 546 \n1.08)10. In order to comprehensively identify additional independent signals as well as nominate 547 \none or more sets of credible causal variants associated with such signals, we used multiple 548 \ncomplementary approaches. Conditional and joint analyses previously found no evidence for a 549 \nsecond genome-wide significant signal at 2q33.110; here, we observe no evidence for a 550 \nprominent independent melanoma risk signal of marginal significance. Fine-mapping using 551 \nBayesian approaches similarly identified one prominent signal at this locus with little evidence 552 \nfor additional signals. 553 \nGermline variants is this region have been identified as associated with multiple cancers, 554 \nincluding breast cancer (rs3769821) 11, keratinocyte cancers (cutaneous BCC and SCC 555 \ncombined, rs6743068) 13, cutaneous BCC (rs6714430) 13, cutaneous SCC (rs10931936) 13, and 556 \nnon-small cell lung cancer (rs3769821) 51 where the direction of effect is the same and the risk 557 \nalleles share a common haplotype with the melanoma risk allele of rs10931936. In addition, a 558 \nhighly correlated risk signal was previously identified for prostate cancer (rs59308963) 52 where 559 \nthe direction of effect is opposite to that of melanoma and the other cancers. This locus was 560 \nalso identified via GWAS for chronic lymphocytic leukemia (CLL; rs3769825) 53, however this 561 \nsignal appears to be considerably less correlated with the melanoma risk signal (r2 with 562 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nrs10931936 = 0.26). For several of these cancers with summary data readily available, we 563 \nevaluated evidence for colocalization, with the results suggesting sharing of one or more causal 564 \nvariants between melanoma and signals for breast cancer, cutaneous BCC, and cutaneous 565 \nSCC and further suggesting that our work to fine-map identify functional, potentially causal 566 \nvariants for melanoma risk may be relevant for these other cancers.  567 \nUsing multiple complementary fine-mapping approaches, we identified a set of 27 potential 568 \ncausal variants. Collectively, these variants present multiple potential alternative or 569 \ncomplementary functional mechanisms for influencing gene function and cancer risk. Notably, 570 \nrs3769823 is a missense protein-coding variant in codon 14 of an alternative first exon of 571 \nCASP8 utilized by multiple isoforms and expressed in human primary melanocytes (isoforms G 572 \nand H; Figure S6). However, a majority of algorithms designed to predict variant function 573 \nsuggest this variant is likely benign and K14R is not located within conserved domains or known 574 \nor predicted sites of post-translational modification. Despite this, we did not directly test for an 575 \neffect of the two alleles of this variant on CASP8 protein function and cannot rule out a role for 576 \naltered protein sequence as a mechanism underlying risk at this locus. Zhang and colleagues 577 \nstudied the functionality of rs3769823 by expressing the two alleles in lung cancer cell lines and 578 \nnoted differences in proliferation and cell migration in vitro and in vivo 54. Notably, expression 579 \nfrom the K14R (protective) allele generated more CASP8 protein, which the authors posited 580 \nmay be due to altered protein stability. Downstream phenotypic assays did not distinguish 581 \nbetween the effects of protein abundance and protein sequence-specific effects, and potential 582 \ntranscriptional differences for the two constructs were not controlled for, so a protein-coding 583 \nfunction for this variant on CASP8 remains possible but still unclear. Further work is needed to 584 \nassess potential effects of this protein-coding change on CASP8 protein function.  585 \nWe also observed a significant CASP8 sQTL in melanocyte and other GTEx eQTL datasets 586 \nrelevant to cancers with associations for this locus (marked by rs10804111; r2 to rs10931936 = 587 \n0.62; D’ = 1) reflecting alternative splicing events at the exon 8 to exon 9 junction. The 588 \nalternative splicing event is more strongly associated with the protective allele generates an 589 \nalternative CASP8 transcript, isoform H. This transcript retains only part of canonical intron 8 590 \nand introduces a frameshift in the transcript, leading to premature termination. The resultant 591 \nprotein retains the CASP8 death effector domain but lacks the catalytic domain, and notably, 592 \nHimeji and colleagues demonstrated that this protein could inhibit apoptosis mediated by wild-593 \ntype CASP8 in a dominant-negative manner39. Thus, the transcript preferentially driven by the 594 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nprotective allele could potentially lead to loss of CASP8 function; this stands in functional 595 \ncontrast to the melanocyte eQTL observed where the protective allele of rs10931936 is 596 \nassociated with higher rather than lower levels of CASP8 transcript. Nonetheless, while LD of 597 \nrs10804111 with rs10931936 is comparable to that of other fine-mapped credible causal 598 \nvariants for melanoma at this locus, none of the fine-mapping approaches we applied nominates 599 \nrs10804111 as a credible causal variant itself. Still, much as for the protein-coding change 600 \ncaused by rs3769823, we cannot rule out a role for altered CASP8 splicing contributing to 601 \nmelanoma risk.  602 \nIn contrast, eQTL and TWAS results in human primary melanocytes10,36 provide strong evidence 603 \nsuggesting cis-gene regulation as a likely mechanism underlying melanoma risk attributable to 604 \nthis locus, where melanoma risk-associated alleles were significantly correlated with lower 605 \nlevels of CASP8. Both TWAS and colocalization approaches suggest that melanoma risk and 606 \nmelanocyte eQTL for CASP8 likely share one or more common causal variant, with 607 \ncolocalization strongly nominating rs3769823. Both also provide some support for FLACC1, 608 \nwhere in contrast to the CASP8 eQTL, higher FLACC1 levels are associated with the melanoma 609 \nrisk allele. Several fine-mapped melanoma risk variants at this locus are clearly located within 610 \nmelanocyte gene-regulatory regions, including regions annotated as active promoters, 611 \nconsistent with a potential role for this signal regulating CASP8 levels. Consistent with observed 612 \nFLACC1 QTLs, melanocyte-specific capture-C and 3C assays showed a physical interaction 613 \nbetween fine-mapped melanoma risk variants and the FLACC1 promoter region (Figure S19) 614 \nsuggesting risk variants indeed may play a role in regulating both genes. Notably, none of these 615 \nanalyses suggest a prominent role for other nearby genes, including the apoptosis regulatory 616 \ngenes CASP10 and CFLAR. 617 \nBased on these data, we comprehensively assessed fine-mapped risk variants for potential cis-618 \nregulatory function, first identifying eight variants overlapping with cis-regulatory histone marks 619 \nin disease relevant tissue (melanocytes, keratinocytes, human mammary epithelial cells, and 620 \nbreast myoepithelial primary cells). We assessed each for allelic cis-regulatory potential both in 621 \ntwo previously published massively parallel reporter assays performed in melanocytes and/or 622 \nmelanoma cells, identifying rs3769823 as the only significant variant, with allelic effects in both 623 \nmelanoma cells and to a lesser degree melanocytes. We likewise tested each in individual 624 \nluciferase reporter assays in two melanoma cell lines, again identifying rs3769823 as the only 625 \nvariant with consistently significant allelic cis-regulatory activities in both cell lines and in both 626 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nforward and reverse orientations where the direction of reporter effect was consistent with the 627 \nmelanocyte FLACC1 eQTL but opposite that of CASP8 (e.g., the risk allele is associated with 628 \nhigher reporter expression). Nonetheless, in individual assays, several variants displayed allelic 629 \nactivity in one orientation, including two additional variants rs3769821 and rs59308963, located 630 \nnearby rs3769823 in the same annotated gene promoter region, suggesting potential functional 631 \ncontributions from more than one variant. Testing commonly occurring 632 \nrs3769823/rs3769821/rs59308963 haplotypes in a reporter assay revealed cell-type dependent 633 \neffects; the risk-associated A/C/ATTCTGTC haplotype was associated with lower expression 634 \nrelative to the protective G/T/- haplotype in melanocytes but higher expression in melanoma 635 \ncells in the forward orientation relative to CASP8 transcription (Figure S17A, B). This stands in 636 \ncontrast to the results from Camp and colleagues, who tested risk and protective haplotypes of 637 \nrs3769823/rs3769821 and observed lower reporter gene expression driven from the risk 638 \nhaplotype in both normal breast and breast cancer cell lines 55. The differing direction of effect 639 \nobserved in melanocytes for rs3769823 in short-fragment versus longer-fragment haplotype 640 \nanalysis suggests that while rs3769823 is strongly functional, the allelic effects are influenced 641 \nby other nearby and potentially functional variants. 642 \nAll three variants showed allelic patterns of protein binding via EMSA. Previously, in the context 643 \nof lung cancer susceptibility, Long and colleagues observed a cis-regulatory element shared 644 \nacross many lung cell types overlapping rs3769823 and noted that this variant is predicted to 645 \nalter IRF8 DNA binding 56. Here, we applied a quantitative mass spectrometry workflow and 646 \nidentified multiple allele-specific nuclear binding proteins from both melanoma and breast 647 \ncancer cells that bound these variants in an allele specific manner. In particular, we identified 648 \nproteins that typically act either as a transcriptional repressor (E4F1) or activator (IRF2) as both 649 \npreferentially binding the A-risk allele of rs3769823, and verified binding of both factors via ChIP 650 \nin melanoma cells and primary melanocytes. While we did not observe significantly different 651 \nIRF8 binding in this context, the DNA-binding motifs of IRF2 and IRF8 are quite similar. We 652 \nobserved a similar situation for G-protective allele specific binding proteins, where both an 653 \nactivator (POU2F1) and repressor (REST) were found to bind to this allele in an allele 654 \npreferential manner. Thus, allelic transcriptional activity may be highly dependent on relative 655 \navailability of these factors as well as competition for binding under different cellular contexts. 656 \nIndeed, gene expression correlation analyses suggest, for example, that E4F1 may play a 657 \nprominent role for CASP8 regulation in melanocytes, while the activator IRF2 may play a larger 658 \nrole in regulation of CASP8 in melanomas. Indeed, reporter assay data for rs3769823 in 659 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nconjunction with E4F1 or IRF2 knockdown are consistent with this, where in melanoma cells 660 \nknockdown of IRF2 resulted in a significant decrease of expression driven from the A-risk allele, 661 \nwhile knockdown of E4F1 resulted in a weaker increase of reporter activity driven from the A-662 \nallele. Given the similarity of DNA-binding motifs for IRF proteins, it may be possible that in 663 \naddition, tissue-specific availability of other IRF family members could also drive or repress 664 \ngene expression via rs3769823 and possibly explain the pleiotropy at this signal for prostate 665 \ncancer where the direction of effect is opposite that of melanoma and the other cancers 15. In 666 \ncontrast knockdown of E4F1 in melanocytes resulted in a significant increase of activity driven 667 \nfrom the A-risk allele. We further observed additional allele specific binding proteins for 668 \npotentially functional adjacent variants rs3769821 and rs59308963, which could well further 669 \nmodulate transcriptional output themselves as well as potentially contribute to preference of 670 \ntranscription factor binding to rs3769823. 671 \nOur work provides strong evidence for two potential causal genes at this locus, CASP8 and 672 \nFLACC1. Of these, CASP8 represents a very strong a priori candidate cancer risk gene, given 673 \nits long-established role as an apoptotic initiator caspase and the fact that lower CASP8 levels 674 \nare associated with cancer risk. Little is known about the function of FLACC1 (formerly 675 \nALS2CR12). It has been found to be component of sperm tail flagellum 57, and functional 676 \nstudies of ALS2CR12-like genes in C. elegans suggest that similar genes play a role in gene 677 \nand or repetitive DNA silencing, a function appearing to be dependent on the biogenesis of 678 \nsmall RNAs 58,59. Notably this region harbors multiple genes involved in apoptosis, including 679 \nCASP10, encoding another apoptotic initiator caspase 60-62 sharing similar substrate specificity 680 \nto CASP8 63 but possibly acting as a negative regulator of CASP8-mediated cell death 64. Also 681 \nlocated within this region is CFLAR (also known as c-FLIP), encoding a protein structurally-682 \nsimilar to CASP8 but lacking caspase activity and acting to prevent CASP8 dimerization and 683 \ninhibiting apoptosis 19,65-69. Despite the interacting roles of these genes, we observe little 684 \nevidence that the risk signal drives differences in expression of these antiapoptotic genes. 685 \nNonetheless, we cannot rule out potential regulation of these genes from this signal under 686 \nspecific contexts. We do note, however, that at least in human melanocytes we observe no 687 \nevidence of physical interaction between fine-mapped cancer risk variants (Figure S19) and the 688 \npromoter regions of either gene. 689 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nIn summary, our study functionally interrogated the multi-cancer risk locus on chromosome 690 \nband 2q33.1, providing strong evidence for cis-regulation of CASP8 via multiple functional 691 \nvariants as a primary causal mechanism influencing cancer risk. 692 \n 693 \nMaterial and methods 694 \nMelanoma GWAS summary data 695 \nFor all analyses using melanoma GWAS data, we used summary statistics from both 696 \nhistologically confirmed, as well as self-reported melanoma cases (from 23andMe and UK 697 \nBiobank), and controls as previously described10. In total, summary data are derived from 698 \n36,760 melanoma cases and 375,188 controls; all participants provided informed consent and 699 \nparticipation was IRB approved. Participants from 23andMe provided informed consent and 700 \nparticipated in the research online under a protocol approved by the external AAHRPP-701 \naccredited IRB, Ethical and Independent Review Services (E and I Review). 702 \nConditional and joint analysis of melanoma GWAS 703 \nConditional and joint analysis was performed with summary melanoma GWAS meta-analysis 704 \ndata by using Genome-wide Complex Trait Analysis (GCTA, v.1.26.0)28 to identify independent 705 \nassociated variants. We set the collinearity threshold (-cojo-collinear) to R2 = 0.05 to detect only 706 \ncompletely independent SNPs. Linkage disequilibrium (LD) between SNPs was calculated using 707 \n5,000 randomly selected individuals of European ancestry from UK Biobank. 708 \nColocalization of the 2q33.1 melanoma association signal with other GWAS and QTL data  709 \nFor colocalization analyses, we used GWAS summary data from breast cancer (122,977 cases 710 \nand 105,974 controls; dbGaP accession number phs001265.v1.p1; downloaded from 711 \nhttps://bcac.ccge.medschl.cam.ac.uk/bcacdata/oncoarray/oncoarray-and-combined-summary-712 \nresult/gwas-summary-results-breast-cancer-risk-2017/)11 and keratinocyte cancers (47,742 713 \ncases and 634,414 controls; dbGaP accession number phs000360.v3.p1)13, including 714 \ncutaneous BCC (31,787 cases and 619,351 controls)13 and cutaneous squamous cell 715 \ncarcinoma (SCC; 9,674 cases and 625,657 controls)13. Colocalization analyses for the 2q33.1 716 \nlocus were performed using both Hypothesis Prioritization in multi-trait Colocalization 717 \n(HyPrColoc)29 and eQTL and GWAS Causal Variant Identification in Associated Regions 718 \n(eCAVIAR)30 as implemented in the ezQTL web-based tool 719 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n(https://analysistools.cancer.gov/ezqtl/)70. For both, we used the set of SNPs in +/- 100 kb 720 \nwindow size from the melanoma GWAS lead SNP rs1093193636,38. For HyPrColoc, a posterior 721 \nprobability (PP) above 50% (0.5) was considered to display a positive colocalization and SNP 722 \nscore above 5% (0.05) was considered to share a common causal variant. For eCAVIAR, the 723 \ncolocalization posterior probability (CLPP) score was calculated with a maximum number of two 724 \ncausal SNPs in the locus; a CLPP score above 1% (0.01) was considered to display a positive 725 \ncolocalization.   726 \nGWAS fine-mapping 727 \nFine-mapping was performed using several approaches. Firstly, in order to inclusively nominate 728 \ncandidate causal variants for assessment, we used the log-likelihood ratio (LLR) of each SNP in 729 \nthe region relative to lead variant rs10931936, and considered all variants within a ratio of 730 \n>1:1000. Variants not successfully imputed from the GWAS that thus could not be fine-mapped 731 \nin this manner nor via other methods requiring summary data, hence we used an LD-based 732 \nthreshold. Specifically, we chose this threshold to be consistent with the observed LD (r2) 733 \nbetween LLR-mapped credible set variants with, where the minimum LD from this set with 734 \nrs10931936 was r2 = 0.625 (1000 Genomes Projects Phase 3 EUR data; 1000G EUR). We thus 735 \nconsidered variants not present in the summary dataset where 1000G EUR r2 ≥ 0.625. We also 736 \nperformed Bayesian fine-mapping of melanoma GWAS summary data10 for the 2q33.1 risk 737 \nregion using both Deterministic Approximation of Posteriors (DAP-G)32,33 and Probabilistic 738 \nAnnotation INtegraTOR (PAINTOR)34,35. DAP-G (version 1.0) analysis used a 500 Kb window 739 \nsize centered over the association signal (median position of chr2:202,123,966) and allowing for 740 \na maximum number of five causals. A UK Biobank (UKBB) LD reference panel (from 337K 741 \nUKBB participants, downloaded from https://alkesgroup.broadinstitute.org/UKBB_LD/)71 was 742 \naligned with the summary statistics from the melanoma GWAS meta-analysis10 using R (version 743 \n4.1.0) for input into DAP-G. PAINTOR analyses were performed as described previously46 using 744 \nPAINTOR 3.0 with a window size of 100 Kb, maximum causals set to 4, 4 melanocyte-specific 745 \nepigenomic annotations and pairwise LD derived from 1000 Genomes phase 3 EUR (1000G 746 \nEUR) data computed using PLINK version 1.9 and R version 4.1.0. Functional annotations 747 \ninclude a set of 2,000 melanocyte-specific expressed genes from our melanocyte dataset36, 748 \nmelanocyte enhancers, transcribed regions, and a histone mark (H3T11ph) from ENCODE and 749 \nRoadmap.   750 \nPrediction of variant effect on protein function and structure 751 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nWe predicted the functional impact of missense variant rs3769823 on function of CASP8 protein 752 \nthrough the PredictSNP2 37 web-based tool (https://loschmidt.chemi.muni.cz/predictsnp2/), 753 \nwhich integrates five functional prediction methods including CADD (Combined Annotation 754 \nDependent Depletion) 72, DANN (Deleterious Annotation of genetic variants using Neural 755 \nNetwork) 73, FATHMM (Functional Analysis Through Hidden Markov Model) 74, FunSeq2 75, and 756 \nGWAVA (Genome-Wide Annotation of VAriants)76. We also predicted potential effects of 757 \nCASP8K14R on secondary structure of the protein through JPred4 77 web server 758 \n(https://www.compbio.dundee.ac.uk/jpred4/), which also predicts solvent accessibility and 759 \ncoiled-coil regions. Predictions from AlphaMissense78 for rs3769823 are not available. 760 \nQuantitative trait locus (QTL) data and other expression data 761 \nQTL data from 106 primary human melanocyte cultures primarily of European ancestry were 762 \ngenerated as described previously for both gene expression (eQTL)36 as well as for splicing 763 \n(sQTL) and CpG methylation (meQTL)38 (dbGaP accession: phs001500.v1.p1). Conditional 764 \nanalysis of the CASP8 melanocyte eQTL was performed using QTLTools 765 \n(https://qtltools.github.io/qtltools/)79, combining genotype and expression level data, along with a 766 \nmatrix of covariates including the top three genotype principal components and the top 15 767 \nPEER factors. For other tissues, we used pre-analyzed cis-eQTL data from the Genotype-768 \nTissue Expression (GTEx) project (v.7), which were downloaded from the GTEx portal 769 \n(https://gtexportal.org/home/index.html)40. eQTL analysis was performed previously36 using 770 \nrelative linear copy-number values as a covariate for 349 melanoma tumors from The Cancer 771 \nGenome Atlas Project (TCGA) skin cutaneous melanoma (SKCM) dataset41,42. eQTL data for 59 772 \nearly passage melanoma cell lines was performed as described previously43 using expression 773 \nand genotyping data from NCBI GEO (accessions GSE78995 for expression data; GSE99193 774 \nfor genotyping data). We specifically assessed genes within the TAD harboring the lead SNP at 775 \n2q33.1 (rs10931936; chr2:201,760,000-203,080,000; hg19) based on Hi-C results from 776 \nSKMEL5 melanoma cells as visualized by the 3D Genome Browser 777 \n(http://3dgenome.fsm.northwestern.edu/view.php). 778 \nCASP8 isoform analysis 779 \nTo assess splicing QTLs (sQTLs), we used the same genotype data, population structure 780 \ncovariates, and statistical approaches as used for melanocyte eQTL analysis36, replacing 781 \nnormalized gene expression levels with normalized splice junction events. STAR80 was used to 782 \nmap the RNA-Seq reads onto the genome (hg19) and then LeafCutter81 was applied to quantify 783 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nthe splice junctions following the procedures described by the authors 784 \n(http://davidaknowles.github.io/leafcutter/articles/sQTL.html). Taqman real-time PCR assays 785 \ntargeting unique junctions of CASP8 transcript isoforms were obtained from Thermo Fisher 786 \nScientific (Waltham, MA; all isoform transcripts: Hs01018151_m1; exon 8 to 9 junction: 787 \nHs01018160_m1; exon 8L (alternative spliced) to 9 junction: Hs04405665_m1). RNA was 788 \nisolated from primary melanocyte cultures derived from 106 individuals mainly of European 789 \ndecent36, and cDNA was synthesized using iScriptTM Advanced cDNA Synthesis Kit (Bio-Rad, 790 \nHercules, CA). Taqman assays were performed in triplicates (technical replicates) and 791 \nnormalized to TBP or PPIA levels.  792 \nExon-trap analysis of alternatively spliced exons of CASP8 genes 793 \nrs10804111 is an intronic variant; no variants strongly linked to the CASP8 melanocyte sQTL 794 \nwere located in consensus splice or branch point sequences, and none could create a cryptic 795 \nsplice site responsible for the observed allelic splicing pattern. We therefore assessed potential 796 \neffects of the lead sQTL SNP rs10804111 in a minigene assay. The intron from the exon 8L to 9 797 \njunction is >5,500 bp and not amenable to cloning in a simple minigene vector. We therefore 798 \nconstructed minigene vectors to contain CASP8 exons 8/8L and surrounding intron (100 bp 799 \nupstream of exon 8 and 100 bp downstream intronic sequence from the exon 8L junction) and 800 \nCASP8 exon 9 and surrounding intronic sequence (100 bp upstream and 100 bp downstream 801 \nintronic sequence), linked together with 201 bp of intronic sequence surrounding each allele of 802 \nrs10804111 (Figure S8). CASP8 exons 8 to 9 (including intron 8; 1,181 bp), along with the 5’ 803 \nflanking of exon 8 and 3’ flanking of 9 sequences (100 bp for each) were custom-synthesized 804 \nfrom Thermo Fisher Scientific and cloned in sense orientation using XhoI and BamHI restriction 805 \nsites of the Exontrap vector pET01 (MoBiTec, Germany) to generate the pET01-806 \nrs10804111(C/T) mini-gene constructs containing each allele of rs10804111. Sequence-verified 807 \npET01 constructs were transfected into UACC903 melanoma cells using LipofectamineTM 2000 808 \n(Invitrogen, Waltham, MA) in 6-well format. Cells were harvested 48 h after transfection, and 809 \ntotal RNA was extracted with the QIAGEN QIAcube using RNeasy kit with on-column DNase I 810 \ntreatment (Qiagen, Germany). 1 μg of total RNA was converted into cDNA with SuperScriptTM III 811 \nreverse transcriptase (Invitrogen) and a vector-specific cDNA primer listed in Table S31. 812 \nTaqman assays targeting unique junctions of pET01-rs10804111 were obtained from Thermo 813 \nFisher Scientific (vector exon 2 (VE2) as a control: APH6CGV; VE1 to VE2 junction: APPRMTJ; 814 \nCASP8 exon 8 to exon 9 junction: APKA42T; CASP8 exon 8L to exon 9 junction: APNKT7M; 815 \nlisted in Table S1). Taqman assays were performed in triplicate and normalized to VE2 levels. 816 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nCapture-C data 817 \nCapture-C data were generated and analyzed as previously described 48,49. Briefly, regions for 818 \nbait design were determined for each locus via identification of the set of SNPs with a LLR < 819 \n100 relative to the leading SNP of a given GWAS signal. For the 2q33.1 locus, capture baits 820 \nwere tiled across the entire region of association (chr2:202,114,359 to 202,205,025); baited 821 \nfragments are listed in Table S32. Capture-C libraries were made using the Arima HiC Kit 822 \n(Arima Genomics, Carlsbad, CA) and the KAPA HyperPrep Kit (KAPA Biosystems, Wilmington, 823 \nMA) following the manufacturer’s protocol. Data are presented for region-specific Capture-C 824 \nruns with five independent human primary melanocyte cultures (C56, C140, C205, C24, and 825 \nC27) with three biological replicates for each. Paired-end sequencing reads from biological 826 \nreplicates were pre-processed with the HiCUP pipeline82 and aligned to human genome version 827 \n19 via bowtie283,84. Chromatin interaction loops were detected at one- and four-fragment 828 \nresolutions via Capture Hi-C Analysis of Genomic Organization (CHiCAGO) pipeline v.1.16.085. 829 \nBecause we observed no distant interactions between fine-mapped variants and gene promoter 830 \nat one-fragment resolution, data presented are from four-fragment analysis. Visualization was 831 \nperformed using the WashU Epigenome Browser86. 832 \nChromatin conformation capture (3C) 833 \nChromatin conformation capture (3C) assays were done based on the protocol from the Dekker 834 \nLab87. RP11-1078D1 bacterial artificial chromosome (BAC) clones were purchased from 835 \nBACPAC Resources Center and purified with the QIAGEN large-construct Maxi Kit to cover the 836 \n2q33.1 genomic region (chr2:202,045,328 to 202,245,875). BAC libraries were made by HindIII 837 \ndigestion of BAC plasmids, followed by ligation and DNA purification. To generate melanoma 838 \ncell and melanocyte 3C libraries, we fixed and lysed 1 x 108 cells, followed by HindIII digestion 839 \nand ligation. Both the BAC libraries, and melanoma and melanocytes 3C libraries were 840 \namplified with a Taqman assay containing a primer localized to various regions of 841 \nchr2:202,045,328 to 202,245,875 and a fixed primer harboring rs3769823, rs3769821, and 842 \nrs59308963 plus a FAM-labeled probe annealing to 3’ of the fixed primer. Primers are listed in 843 \nTable S31. PCR cycle conditions were as follows: 95°C, 5min; [95°C, 30s; 62°C, 30s; 72°C, 844 \n30s]x34; 72°C, 10min. Amplification for different primer pairs from the melanoma and melanocyte 845 \n3C libraries was normalized to that of BAC libraries. 846 \nEpigenomic annotations 847 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nChromatin Primary Core Marks Segmentation by HMM (ChromHMM), DNase Hypersensitivity 848 \n(DHS), and histone modification data for primary human melanocyte cultures, primary human 849 \nkeratinocyte culture, and human breast mammary cultures “breast variant human mammary 850 \nepithelial cells” and “breast myoepithelial primary cells” (shown as vHMEC and BMYO, 851 \nrespectively, in Figure 1) were obtained from the Roadmap Epigenomics Project (downloaded 852 \nthrough the UCSC Genome Browser; http://genome.ucsc.edu/). Peaks from FAIRE-seq and 853 \nH3K27ac ChIP for 11 melanoma cultures samples were obtained from the Gene Expression 854 \nOmnibus (GEO; accession GSE60666)45. 855 \nCell culture 856 \nWe originally obtained melanoma cell lines (UACC903 and UACC1113) from the University of 857 \nArizona Cancer Center (UACC), which were maintained in RPMI-1640 medium with L-glutamine 858 \n(Quality Biological, 112-025-101) supplemented with 10% FBS (GenClone, 25-514H), 20 mM 859 \nHEPES (pH 7.9; Gibco, 15630), and 100 U/ml penicillin-streptomycin (Gibco, 15140). Primary 860 \nmelanocytes from foreskin healthy newborn males, mainly of European descent, were grown in 861 \nM254 media (Invitrogen, M254500) with HMGS-2 supplement (Invitrogen, S0165). MCF7 and 862 \nT47D breast cancer cell lines were purchased from ATCC. MCF7 cells were grown in DMEM 863 \nwith L-glutamine (Quality Biological, 112-014-101) supplemented with 10% FBS, and 100 U/ml 864 \npenicillin-streptomycin. T47D cells were maintained in RPMI 1640 supplemented with 10% FBS, 865 \nand 100 U/ml penicillin-streptomycin. All cell lines were grown in a 5% CO2 humidified incubator 866 \nat 37 °C and were tested for mycoplasma contamination every 3-6 months.  867 \nLuciferase reporter assays 868 \nSequences encompassing each variant were PCR-amplified (primers listed in Table S31) from 869 \ngenomic DNA of HapMap CEU panel samples with the appropriate genotypes to obtain clones 870 \nwith each genotype, and cloned into the HindIII and XhoI sites of the pGL4.23[luc2/minP] 871 \n(Promega, Madison, MI) luciferase vector in the 5’-to-3’ or 3’-to’5’ orientation using In-Fusion HD 872 \nCloning Kit (Takara Bio). Sequence-verified pGL4.23 constructs were co-transfected with the 873 \npGL4.73[hRluc/SV40] Renilla luciferase control vector (Promega) into melanoma cell lines 874 \n(UACC903 and UACC1113) using LipofectamineTM 2000 (Invitrogen) in 24-well format. 875 \nLuciferase activity was measured 24 h after transfection with the Dual Luciferase Reporter 876 \nAssay System (Promega). Luciferase activity was normalized to the Renilla luciferase activity. 877 \nAll the experiments were performed in six replicate wells and repeated for at least three 878 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nbiological replicates. Significance between alleles was assessed using a two-tailed, unpaired t-879 \ntest assuming unequal variance.  880 \nMassively parallel reporter assay data 881 \nWe used data from two massively parallel reporter assay (MPRA) datasets assessing allelic cis-882 \nregulatory potential for candidate causal variants from this locus in melanoma cells and 883 \nmelanocytes46,47. Firstly, in a prior MPRA analysis (MPRA v.1)46, 823 variants from 20 genome-884 \nwide significant loci from a prior melanoma meta-analysis8 were tested in a melanoma cell line, 885 \nUACC903, as well as HEK293FT cells. For this study, an LD-based selection criteria was used 886 \n(r2 > 0.4 with the sentinel SNP using 1,000 Genomes phase 3 EUR data), further filtering 887 \ncandidate causal variants to test only those located within annotated open chromatin regions 888 \nand promoter/enhancer histone marks found in primary melanocytes and/or short-term cultures. 889 \nA second MPRA study (MPRA v.2)47 investigated 54 loci from a more recent melanoma GWAS 890 \nmeta-analysis10 in both melanoma cells and melanocytes. Here, 1,992 candidate causal 891 \nvariants were tested from 54 genome-wide significant loci. Candidate causal variants were fine-892 \nmapped using log-likelihood ratios (LLR < 1:1,000 with the lead SNP) or where variants were 893 \nnot successfully imputed or tested in the GWAS using an LD-based criteria (r2 > 0.8 with the 894 \nlead SNP, 1,000 Genomes EUR).  895 \nEMSAs and supershift assays 896 \nNuclear extracts from melanoma cell lines (UACC903 and UACC1113) or human melanocytes 897 \nwere prepared using the NE-PER Nuclear and Cytoplasmic Extraction Kit (Thermo Fisher 898 \nScientific). Nuclear extract lysates for MCF7 and T47D breast cancer cell lines were purchased 899 \nfrom Abcam (ab14860 and ab14896, respectively). Oligonucleotides for each SNP were 900 \nsynthesized with or without biotin labeling at the 5’ end and HPLC purified (21-28nt, Life 901 \nTechnologies, listed in Table S31). Forward and reverse strands were then annealed to 902 \ngenerate double stranded 5’-end labelled probes or unlabeled competitors. 50 nM probes were 903 \nadded to 2 µg of nuclear extract preincubated with 1 µg of poly d(I-C) (Roche) in LightShiftTM 1 × 904 \nbinding buffer (Thermo Fisher Scientific) for 30 min on ice. Competition experiments were 905 \nperformed by adding 1-100 fold more unlabeled competitor oligonucleotides to the reaction 906 \nmixture 5 min before the addition of labelled probes. Anti-E4F1 (Abcam, ab70615) or anti-IRF2 907 \n(Abcam, ab245658) antibodies for supershift assays were incubated with nuclear extract for 1 h 908 \nat 4 °C prior to adding poly d(I-C). The reactions were run on 5 – 10% TBE gels (Bio-Rad 909 \nCriterion) on ice at 120 V, transferred to Biodyne B membranes (VWR), transferred blots were 910 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\ncrosslinked (Stratagene UV Stratalinker 1800) and detected using the LightShift 911 \nChemiluminescent EMSA Kit (Thermo Fisher Scientific) and imaged on Chemidoc Touch (Bio-912 \nRad). One representative data is shown from three biological replicates.  913 \nQuantitative mass spectrometry  914 \nMCF7 breast cancer cells were grown in RPMI 1640 (Gibco) supplemented with 10% FBS, 100 915 \nU/ml penicillin and 100 μg/ml streptomycin (Gibco). The same culture conditions were used for 916 \nUACC903 melanoma cells with additional 20 mM HEPES (pH 7.9). Nuclear lysates for DNA 917 \naffinity purifications were collected as described previously88. Oligonucleotide probes with 5’-918 \nbiotinylation of the forward strand were ordered from Integrated DNA Technologies (Table S31). 919 \nDNA pulldowns and on-bead trypsin digestion were performed in Eppendorf tubes as described 920 \npreviously89. Briefly, forward and reverse oligos were annealed with a 1.5X molar excess of the 921 \nreverse strand. For each pulldown, annealed DNA oligos (500 pmol) were immobilized on 10 μl 922 \n(20 μl slurry) Streptavidin-Sepharose beads (GE Healthcare, Chicago, IL). Immobilized DNA 923 \noligos were incubated with nuclear extracts (500 μg) from UACC903 and MCF7, and 10 μg of 924 \nnon-specific competitor DNA (5 μg polydIdC, 5 μg polydAdT). After washing away unbound 925 \nproteins, beads were resuspended in elution buffer (2 M Urea, 100 mM TRIS (pH 8), 10 mM 926 \nDTT), peptides were alkylated with 50 mM iodoacetamide, and on-bead digested with 0.25 μg 927 \ntrypsin. Tryptic peptides were labeled by stable isotope dimethyl labeling on StageTips, as 928 \ndescribed previously89. Samples were eluted from the StageTips and matching light and 929 \nmedium labeled samples were combined. Samples were loaded onto a 30 cm column (heated 930 \nat 40 °C) packed in-house with 1.8 μm Reprosil-Pur C18-AQ (Dr Maisch) and eluted using a 931 \ngradient from 9 to 32% Buffer B (80% acetonitrile, 0.1% formic acid) over 114 min at a flow rate 932 \nof 250 nL/min using an Easy-nLC 1000 (Thermo Fisher Scientific). Peptides were sprayed 933 \ndirectly into either a Orbitrap Fusion Tribrid mass spectrometer (Thermo Fisher Scientific) or a Q 934 \nExactive mass spectrometer (Thermo Fisher Scientific). The mass spectrometers were operated 935 \nas described previously46,89. Thermo RAW files were analyzed with MaxQuant 1.6.0.1 by 936 \nsearching against the UniProt curated human proteome (released June 2017) with standard 937 \nsettings90. Protein ratios were normalized by median ratio shifting and used for outlier calling. An 938 \noutlier cutoff of 1.5 inter-quartile ranges in two out of two biological replicates was used. 939 \nrs3769823 and rs3769821 were assayed in both UACC903 and MCF7 cells, while rs59308963 940 \nwas assayed only in UACC903. 941 \nChromatin immunoprecipitation and genotyping 942 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nFollowing the manufacturer’s instructions of the Active Motif ChIP-IT High-Sensitivity kit, actively 943 \ngrowing melanoma cells (UACC903 and UACC1649) or C87 primary melanocytes were fixed 944 \nwith 1% formaldehyde when 80-90% confluent. Nuclei of 1.0 – 1.5 × 107 cells were prepared 945 \nand sheared with ME220 Focused-ultrasonicator (Covaris, Woburn, MA) for 20 min following the 946 \ninstructions. 10 – 30 µg sheared chromatin from 1.5 – 4.5 × 106 cells were used for each 947 \nimmunoprecipitation reaction with anti-E4F1, anti-IRF2 or anti-IgG (Abcam, ab37415) following 948 \nthe instructions. Purified immunoprecipitated DNA or input DNA was analyzed by SYBR Green 949 \nqPCR for enrichment of target sites using the primers listed in Table S31. ChIP DNA or input 950 \nDNA from UACC1649 or C87 melanocyte cell lines (heterozygous for rs3769823) was assayed 951 \nfor genotyping rs3769823 by Taqman genotyping assay (Assay ID: C__25808407_20). All 952 \nexperiments were performed in triplicate (technical replicates) and repeated for at least three 953 \nbiological replicates. 954 \nsiRNA-mediated knockdown of E4F1 and IRF2  955 \nSMARTpool ON-TARGETplus Human siRNAs to E4F1 (L-011847-00-0005), IRF2 (L-011705-956 \n02-0005) and ON-TARGETplus Non-targeting controls (D-001810-01-05) were purchased from 957 \nDharmacon (Lafayette, CO). Transfections were carried out using LipofectamineTM RNAiMAX 958 \n(Invitrogen) according to the manufacturer’s instructions. Cells were collected after 48h for RNA 959 \nand protein extractions.  960 \nqPCR 961 \nRNA was isolated using RNeasy Mini Column (Qiagen), which was always complemented by 962 \nDNase treatment. cDNA was synthesized from the total RNA using iScriptTM Advanced cDNA 963 \nSynthesis Kit (Bio-Rad). Gene expression levels were quantified by qPCR using Taqman 964 \nassays for E4F1 (Hs00231773_m1), IRF2 (Hs01082884_m1), CASP8 (Hs01018151_m1, 965 \nHs01018149, Hs01018160_m1, Hs04405665_m1 and Hs01022432_m1), and GAPDH 966 \n(Hs99999905_m1) from Thermo Fisher Scientific. Gene expression levels of E4F1, IRF2 and 967 \nCASP8 were normalized to GAPDH. Each experiment was performed in triplicate and replicated 968 \nthree times. Significance was assessed using a Student’s two-tailed T-test assuming unequal 969 \nvariances. 970 \nWestern blotting 971 \nWhole-cell extracts prepared in RIPA buffer with protease inhibitors (Complete Protease 972 \nInhibitor, Roche) or phosphatase inhibitors (PhosSTOP, Roche). Samples were quantified using 973 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nDC Protein Assay (Bio-Rad) and electrophoresed on 4–12% Bis Tris Plus Bolt gels in MES 974 \nbuffer (Invitrogen). Proteins were transferred to nitrocellulose using an iBlot2 (Invitrogen). Blots 975 \nwere blocked with 5% non-fat dry milk in Tris-buffered saline with 0.1% Tween-20 (TBST). 976 \nPrimary and secondary antibodies were diluted in 1-2% milk in TBST, and all washes were 977 \nperformed with TBST. Blots were rinsed briefly with PBS before the addition of ECL Prime 978 \nWestern Blotting Detection Reagent (Amersham, UK). Images were captured on ChemiDoc™ 979 \nGel Imaging System (Bio-Rad). The antibodies of anti-E4F1 (Santa Cruz, sc-514718), anti-IRF2 980 \n(Santa Cruz, sc-101069,) and anti-β-actin (Santa Cruz, sc-47778) were used as primary 981 \nantibodies, and goat anti-mouse HRP (Santa Cruz, sc-2005) for secondary antibodies. 982 \nAnalysis of melanocyte and melanoma expression data 983 \nTranscriptomic data were generated as described previously from human primary 984 \nmelanocytes36, TCGA-SKCM36,41,42, early passage melanoma cultures43, the Leeds Melanoma 985 \nCohort91, and GTEx tissues (v.7; https://gtexportal.org/home/index.html)40. In brief, RNA 986 \nSequencing by Expectation maximization (RSEM, version 1.2.31, 987 \nhttp://deweylab.github.io/RSEM/) was used to quantify the gene expression followed by the 988 \nquantile normalization to obtain the final RSEM for primary melanocytes, TCGA-SKCM, and 989 \nGTEx tissues as previously described36. For background correction and quantile normalization, 990 \nRobust Multi-array Average (RMA) algorithm with default settings (Affymetrix) was performed for 991 \nmelanoma cell lines43. A primary melanoma transcriptomic dataset from the Leeds Melanoma 992 \nCohort study on 703 patients was generated using the Illumina DASL Human HT12 v4 array 993 \nplatform, as previously described91 (European Genome-phenome Archive accession 994 \nEGAS00001002922; https://ega-archive/org/studies/EGAS00001002922). 16 samples were 995 \nremoved in quality control (expression data from only one extraction per patient) leaving a 996 \ncohort of 687 patients91. LUMI92 was used for background correction and quantile normalization 997 \nas previously reported92. For GTEx (v.7) transcriptomic data, genes were selected based on 998 \nexpression thresholds of > 0.5 RSEM and ≥ 6 reads in at least 10 samples, and expression 999 \nvalues for each gene were inverse quantile normalized to a standard normal distribution across 1000 \nsamples and described previously36. We used R (version 4.1.0) for Pearson correlation to 1001 \nanalyze the correlations between candidate target genes and allelic-specific binding 1002 \ntranscription factors. The significance level was set at P < 0.05 and [slope] > 0.4 for all these 1003 \ntests. Using transcriptomic and genotype datasets from multiple tissues, we also ran multiple 1004 \nlinear regression on R (version 4.1.0) to evaluate the influence of allelic-specific binding 1005 \ntranscription factor or genetic variable of rs3769823 on the CASP8.  1006 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n 1007 \nAcknowledgements 1008 \nThis research was supported in part by the Intramural Research Program of the National 1009 \nInstitutes of Health (NIH). The contributions of the NIH author(s) are considered Works of the 1010 \nUnited States Government. The findings and conclusions presented in this paper are those of 1011 \nthe author(s) and do not necessarily reflect the views of the NIH or the U.S. Department of 1012 \nHealth and Human Services. We would like to thank members at the National Cancer Institute’s 1013 \nCancer Genomics Research Laboratory (CGR) for help with sequencing efforts and Rouf 1014 \nBanday from the National Cancer Institute for assistance with minigene assay design and data 1015 \ninterpretation. This work has been supported by the Intramural Research Program (IRP) of the 1016 \nNational Cancer Institute, US National Institutes of Health. We thank Jeremy Bravo Narula for 1017 \nassistance with the manuscript. We thank all the cohorts, funders, and investigators who 1018 \ncontributed to the melanoma GWAS, as originally acknowledged by Landi, 23andMe, and 1019 \ncolleagues10; data from this GWAS were used toward fine-mapping. JN receives funding from 1020 \nHorizon Europe 101136622. The Leeds Melanoma Cohort recruitment, follow up and data 1021 \ncollection was supported by CRUK C588/A19167, C8216/A6129, C588/A10721 and NIH 1022 \nCA83115. Mark Iles is supported in part by the National Institute for Health and Care Research 1023 \n(NIHR) Leeds Biomedical Reearch Centre (BRC) (NIHR203331). The views expressed are 1024 \nthose of the author(s) and not necessarily those of the NHS, the NIHR or the department of 1025 \nHealth and Social Care. 1026 \n 1027 \n 1028 \nDeclaration of interest 1029 \nThe authors declare no competing interests. 1030 \n 1031 \nData and code availability 1032 \nConditional fine-mapping data are available in Tables S1 and S11; Bayesian fine-mapping 1033 \nresults are available in Tables S5-S7. Colocalization analyses are available are in Tables S2, 1034 \nS3, S12, S13, S19, and S20. Data from the 2020 melanoma GWAS meta-analysis performed by 1035 \nLandi and colleagues was obtained from dbGap (dbGap: phs001868.v1.p1), with the exclusion 1036 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\nof self-reported data from 23andMe and UK Biobank. The full GWAS summary statistics for the 1037 \n23andMe discovery dataset will be made available through 23andMe to qualified researchers 1038 \nunder an agreement with 23andMe that protects the privacy of the 23andMe participants. 1039 \nPlease visit the 23andMe website for more information and to apply to access the data. 1040 \nSummary data from the remaining self-reported cases are available from the corresponding 1041 \nauthors of that manuscript (Matthew H Law, matthew.law@qimrberghofer.edu.au; Mark M Iles, 1042 \nm.m.iles@leeds.ac.uk; and Maria Teresa Landi, landim@mail.nih.gov). Melanocyte eQTL data 1043 \nand RNA-seq expression data from 106 individuals are available from dbGap (dbGap: 1044 \nphs001500.v1.p1), and pre-analyzed cis-eQTL data from GTEx (v.7) were downloaded from the 1045 \nGTEx portal. eQTL data from 59 early melanoma cell lines was available from NCBI Gene 1046 \nExpression Omnibus (accessions GSE78995 for expression data; GSE99193 for genotyping 1047 \ndata). Leeds Melanoma Cohort data from 703 patients was available in the European Genome-1048 \nphenome Archive accession EGAS00001002922. MPRA data are available in the NCBI Gene 1049 \nExpression Omnibus as a SuperSeries under the accession number GEO: GSE129250 and 1050 \nGSE210356. Capture-C data have been previously published48,49; interaction data including 1051 \nboth called interactions as well as raw sequencing data are available through ArrayExpress 1052 \n(accession: E-MTAB-15079). Specifically, for the 2q33.1 locus restriction fragments baited for 1053 \nregion-specific Capture-C assays are provided in Table S32, and loops called by ChICAGO are 1054 \nin Tables S23 and S24. Pearson correlations are available in Table S25, S29 and S30. Multiple 1055 \nlinear regression results are available in Tables S26-S28. Luciferase assay fragments, other 1056 \nprimers and Taqman assays are listed in Table S31. 1057 \n 1058 \nWeb resources 1059 \n23andMe, https://research.23andme.com/collaborate/#dataset-access/ 1060 \nGTEx portal, https://www.gtexportal.org/home/ 1061 \nhttps://ldlink.nci.nih.gov/?tab=home 1062 \nNCBI GEO, https://www.ncbi.nlm.nih.gov/geo/ 1063 \nRegulomeDB, https://regulomedb.org/regulome-search/ 1064 \nThe Cancer Genome Atlas (TCGA) Research Network,  1065 \nhttps://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga 1066 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n 1067 \nReferences 1068 \n1. Brown, K.M., Macgregor, S., Montgomery, G.W., Craig, D.W., Zhao, Z.Z., Iyadurai, K., Henders, 1069 \nA.K., Homer, N., Campbell, M.J., Stark, M., et al. (2008). Common sequence variants on 1070 \n20q11.22 confer melanoma susceptibility. Nat Genet 40, 838-840. 10.1038/ng.163. 1071 \n2. Gudbjartsson, D.F., Sulem, P., Stacey, S.N., Goldstein, A.M., Rafnar, T., Sigurgeirsson, B., 1072 \nBenediktsdottir, K.R., Thorisdottir, K., Ragnarsson, R., Sveinsdottir, S.G., et al. (2008). ASIP and 1073 \nTYR pigmentation variants associate with cutaneous melanoma and basal cell carcinoma. Nat 1074 \nGenet 40, 886-891. 10.1038/ng.161. 1075 \n3. Falchi, M., Bataille, V., Hayward, N.K., Duffy, D.L., Bishop, J.A., Pastinen, T., Cervino, A., Zhao, Z.Z., 1076 \nDeloukas, P., Soranzo, N., et al. (2009). Genome-wide association study identifies variants at 1077 \n9p21 and 22q13 associated with development of cutaneous nevi. Nat Genet 41, 915-919. 1078 \n10.1038/ng.410. 1079 \n4. Bishop, D.T., Demenais, F., Iles, M.M., Harland, M., Taylor, J.C., Corda, E., Randerson-Moor, J., 1080 \nAitken, J.F., Avril, M.F., Azizi, E., et al. (2009). Genome-wide association study identifies three 1081 \nloci associated with melanoma risk. Nat Genet 41, 920-925. 10.1038/ng.411. 1082 \n5. Barrett, J.H., Iles, M.M., Harland, M., Taylor, J.C., Aitken, J.F., Andresen, P.A., Akslen, L.A., 1083 \nArmstrong, B.K., Avril, M.F., Azizi, E., et al. (2011). Genome-wide association study identifies 1084 \nthree new melanoma susceptibility loci. Nat Genet 43, 1108-1113. 10.1038/ng.959. 1085 \n6. Macgregor, S., Montgomery, G.W., Liu, J.Z., Zhao, Z.Z., Henders, A.K., Stark, M., Schmid, H., 1086 \nHolland, E.A., Duffy, D.L., Zhang, M., et al. (2011). Genome-wide association study identifies a 1087 \nnew melanoma susceptibility locus at 1q21.3. Nat Genet 43, 1114-1118. 10.1038/ng.958. 1088 \n7. Iles, M.M., Law, M.H., Stacey, S.N., Han, J., Fang, S., Pfeiffer, R., Harland, M., Macgregor, S., 1089 \nTaylor, J.C., Aben, K.K., et al. (2013). A variant in FTO shows association with melanoma risk not 1090 \ndue to BMI. Nat Genet 45, 428-432, 432e421. 10.1038/ng.2571. 1091 \n8. Law, M.H., Bishop, D.T., Lee, J.E., Brossard, M., Martin, N.G., Moses, E.K., Song, F., Barrett, J.H., 1092 \nKumar, R., Easton, D.F., et al. (2015). Genome-wide meta-analysis identifies five new 1093 \nsusceptibility loci for cutaneous malignant melanoma. Nat Genet 47, 987-995. 10.1038/ng.3373. 1094 \n9. Ransohoff, K.J., Wu, W., Cho, H.G., Chahal, H.C., Lin, Y., Dai, H.J., Amos, C.I., Lee, J.E., Tang, J.Y., 1095 \nHinds, D.A., et al. (2017). Two-stage genome-wide association study identifies a novel 1096 \nsusceptibility locus associated with melanoma. Oncotarget 8, 17586-17592. 1097 \n10.18632/oncotarget.15230. 1098 \n10. Landi, M.T., Bishop, D.T., MacGregor, S., Machiela, M.J., Stratigos, A.J., Ghiorzo, P., Brossard, M., 1099 \nCalista, D., Choi, J., Fargnoli, M.C., et al. (2020). Genome-wide association meta-analyses 1100 \ncombining multiple risk phenotypes provide insights into the genetic architecture of cutaneous 1101 \nmelanoma susceptibility. Nat Genet 52, 494-504. 10.1038/s41588-020-0611-8. 1102 \n11. Michailidou, K., Lindstrom, S., Dennis, J., Beesley, J., Hui, S., Kar, S., Lemacon, A., Soucy, P., Glubb, 1103 \nD., Rostamianfar, A., et al. (2017). Association analysis identifies 65 new breast cancer risk loci. 1104 \nNature 551, 92-94. 10.1038/nature24284. 1105 \n12. Stacey, S.N., Helgason, H., Gudjonsson, S.A., Thorleifsson, G., Zink, F., Sigurdsson, A., Kehr, B., 1106 \nGudmundsson, J., Sulem, P., Sigurgeirsson, B., et al. (2015). New basal cell carcinoma 1107 \nsusceptibility loci. Nat Commun 6, 6825. 10.1038/ncomms7825. 1108 \n13. Liyanage, U.E., Law, M.H., Han, X., An, J., Ong, J.S., Gharahkhani, P., Gordon, S., Neale, R.E., 1109 \nOlsen, C.M., andMe Research, T., et al. (2019). Combined analysis of keratinocyte cancers 1110 \nidentifies novel genome-wide loci. Hum Mol Genet 28, 3148-3160. 10.1093/hmg/ddz121. 1111 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n14. Hart, K., Landvik, N.E., Lind, H., Skaug, V., Haugen, A., and Zienolddiny, S. (2011). A combination 1112 \nof functional polymorphisms in the CASP8, MMP1, IL10 and SEPS1 genes affects risk of non-1113 \nsmall cell lung cancer. Lung Cancer 71, 123-129. 10.1016/j.lungcan.2010.04.016. 1114 \n15. Lubahn, J., Berndt, S.I., Jin, C.H., Klim, A., Luly, J., Wu, W.S., Isaacs, S., Wiley, K., Isaacs, W.B., 1115 \nSuarez, B.K., et al. (2010). Association of CASP8 D302H polymorphism with reduced risk of 1116 \naggressive prostate carcinoma. Prostate 70, 646-653. 10.1002/pros.21098. 1117 \n16. Shi, Y. (2002). Mechanisms of caspase activation and inhibition during apoptosis. Mol Cell 9, 1118 \n459-470. 10.1016/s1097-2765(02)00482-3. 1119 \n17. Boldin, M.P., Goncharov, T.M., Goltsev, Y.V., and Wallach, D. (1996). Involvement of MACH, a 1120 \nnovel MORT1/FADD-interacting protease, in Fas/APO-1- and TNF receptor-induced cell death. 1121 \nCell 85, 803-815. 10.1016/s0092-8674(00)81265-9. 1122 \n18. Muzio, M., Chinnaiyan, A.M., Kischkel, F.C., O'Rourke, K., Shevchenko, A., Ni, J., Scaffidi, C., Bretz, 1123 \nJ.D., Zhang, M., Gentz, R., et al. (1996). FLICE, a novel FADD-homologous ICE/CED-3-like protease, 1124 \nis recruited to the CD95 (Fas/APO-1) death--inducing signaling complex. Cell 85, 817-827. 1125 \n10.1016/s0092-8674(00)81266-0. 1126 \n19. Kischkel, F.C., Hellbardt, S., Behrmann, I., Germer, M., Pawlita, M., Krammer, P.H., and Peter, 1127 \nM.E. (1995). Cytotoxicity-dependent APO-1 (Fas/CD95)-associated proteins form a death-1128 \ninducing signaling complex (DISC) with the receptor. EMBO J 14, 5579-5588. 1129 \n20. Boatright, K.M., Renatus, M., Scott, F.L., Sperandio, S., Shin, H., Pedersen, I.M., Ricci, J.E., Edris, 1130 \nW.A., Sutherlin, D.P., Green, D.R., and Salvesen, G.S. (2003). A unified model for apical caspase 1131 \nactivation. Mol Cell 11, 529-541. 10.1016/s1097-2765(03)00051-0. 1132 \n21. Pop, C., Fitzgerald, P., Green, D.R., and Salvesen, G.S. (2007). Role of proteolysis in caspase-8 1133 \nactivation and stabilization. Biochemistry 46, 4398-4407. 10.1021/bi602623b. 1134 \n22. Oberst, A., Pop, C., Tremblay, A.G., Blais, V., Denault, J.B., Salvesen, G.S., and Green, D.R. (2010). 1135 \nInducible dimerization and inducible cleavage reveal a requirement for both processes in 1136 \ncaspase-8 activation. J Biol Chem 285, 16632-16642. 10.1074/jbc.M109.095083. 1137 \n23. Catts, V.S., and Weickert, C.S. (2012). Gene expression analysis implicates a death receptor 1138 \npathway in schizophrenia pathology. PLoS One 7, e35511. 10.1371/journal.pone.0035511. 1139 \n24. Thome, M., Schneider, P., Hofmann, K., Fickenscher, H., Meinl, E., Neipel, F., Mattmann, C., 1140 \nBurns, K., Bodmer, J.L., Schroter, M., et al. (1997). Viral FLICE-inhibitory proteins (FLIPs) prevent 1141 \napoptosis induced by death receptors. Nature 386, 517-521. 10.1038/386517a0. 1142 \n25. Budihardjo, I., Oliver, H., Lutter, M., Luo, X., and Wang, X. (1999). Biochemical pathways of 1143 \ncaspase activation during apoptosis. Annu Rev Cell Dev Biol 15, 269-290. 1144 \n10.1146/annurev.cellbio.15.1.269. 1145 \n26. Krueger, A., Schmitz, I., Baumann, S., Krammer, P.H., and Kirchhoff, S. (2001). Cellular FLICE-1146 \ninhibitory protein splice variants inhibit different steps of caspase-8 activation at the CD95 1147 \ndeath-inducing signaling complex. J Biol Chem 276, 20633-20640. 10.1074/jbc.M101780200. 1148 \n27. Yang, J., Lee, S.H., Goddard, M.E., and Visscher, P.M. (2011). GCTA: a tool for genome-wide 1149 \ncomplex trait analysis. Am J Hum Genet 88, 76-82. 10.1016/j.ajhg.2010.11.011. 1150 \n28. Yang, J., Ferreira, T., Morris, A.P., Medland, S.E., Genetic Investigation of, A.T.C., Replication, 1151 \nD.I.G., Meta-analysis, C., Madden, P.A., Heath, A.C., Martin, N.G., et al. (2012). Conditional and 1152 \njoint multiple-SNP analysis of GWAS summary statistics identifies additional variants influencing 1153 \ncomplex traits. Nat Genet 44, 369-375, S361-363. 10.1038/ng.2213. 1154 \n29. Foley, C.N., Staley, J.R., Breen, P.G., Sun, B.B., Kirk, P.D.W., Burgess, S., and Howson, J.M.M. 1155 \n(2021). A fast and efficient colocalization algorithm for identifying shared genetic risk factors 1156 \nacross multiple traits. Nat Commun 12, 764. 10.1038/s41467-020-20885-8. 1157 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n30. Hormozdiari, F., van de Bunt, M., Segre, A.V., Li, X., Joo, J.W.J., Bilow, M., Sul, J.H., Sankararaman, 1158 \nS., Pasaniuc, B., and Eskin, E. (2016). Colocalization of GWAS and eQTL Signals Detects Target 1159 \nGenes. Am J Hum Genet 99, 1245-1260. 10.1016/j.ajhg.2016.10.003. 1160 \n31. Shabalin, A.A. (2012). Matrix eQTL: ultra fast eQTL analysis via large matrix operations. 1161 \nBioinformatics 28, 1353-1358. 10.1093/bioinformatics/bts163. 1162 \n32. Wen, X., Lee, Y., Luca, F., and Pique-Regi, R. (2016). Efficient Integrative Multi-SNP Association 1163 \nAnalysis via Deterministic Approximation of Posteriors. Am J Hum Genet 98, 1114-1129. 1164 \n10.1016/j.ajhg.2016.03.029. 1165 \n33. Lee, Y., Luca, F., Pique-Regi, R., and Wen, X. (2018). Bayesian Multi-SNP Genetic Association 1166 \nAnalysis: Control of FDR and Use of Summary Statistics. bioRxiv. 1167 \n34. Kichaev, G., Yang, W.Y., Lindstrom, S., Hormozdiari, F., Eskin, E., Price, A.L., Kraft, P., and 1168 \nPasaniuc, B. (2014). Integrating functional data to prioritize causal variants in statistical fine-1169 \nmapping studies. PLoS Genet 10, e1004722. 10.1371/journal.pgen.1004722. 1170 \n35. Kichaev, G., and Pasaniuc, B. (2015). Leveraging Functional-Annotation Data in Trans-ethnic 1171 \nFine-Mapping Studies. Am J Hum Genet 97, 260-271. 10.1016/j.ajhg.2015.06.007. 1172 \n36. Zhang, T., Choi, J., Kovacs, M.A., Shi, J., Xu, M., Program, N.C.S., Melanoma Meta-Analysis, C., 1173 \nGoldstein, A.M., Trower, A.J., Bishop, D.T., et al. (2018). Cell-type-specific eQTL of primary 1174 \nmelanocytes facilitates identification of melanoma susceptibility genes. Genome Res 28, 1621-1175 \n1635. 10.1101/gr.233304.117. 1176 \n37. Bendl, J., Musil, M., Stourac, J., Zendulka, J., Damborsky, J., and Brezovsky, J. (2016). 1177 \nPredictSNP2: A Unified Platform for Accurately Evaluating SNP Effects by Exploiting the Different 1178 \nCharacteristics of Variants in Distinct Genomic Regions. PLoS Comput Biol 12, e1004962. 1179 \n10.1371/journal.pcbi.1004962. 1180 \n38. Zhang, T., Choi, J., Dilshat, R., Einarsdottir, B.O., Kovacs, M.A., Xu, M., Malasky, M., Chowdhury, 1181 \nS., Jones, K., Bishop, D.T., et al. (2021). Cell-type-specific meQTLs extend melanoma GWAS 1182 \nannotation beyond eQTLs and inform melanocyte gene-regulatory mechanisms. Am J Hum 1183 \nGenet 108, 1631-1646. 10.1016/j.ajhg.2021.06.018. 1184 \n39. Himeji, D., Horiuchi, T., Tsukamoto, H., Hayashi, K., Watanabe, T., and Harada, M. (2002). 1185 \nCharacterization of caspase-8L: a novel isoform of caspase-8 that behaves as an inhibitor of the 1186 \ncaspase cascade. Blood 99, 4070-4078. 10.1182/blood.v99.11.4070. 1187 \n40. Consortium, G.T., Laboratory, D.A., Coordinating Center -Analysis Working, G., Statistical 1188 \nMethods groups-Analysis Working, G., Enhancing, G.g., Fund, N.I.H.C., Nih/Nci, Nih/Nhgri, 1189 \nNih/Nimh, Nih/Nida, et al. (2017). Genetic effects on gene expression across human tissues. 1190 \nNature 550, 204-213. 10.1038/nature24277. 1191 \n41. Cancer Genome Atlas, N. (2015). Genomic Classification of Cutaneous Melanoma. Cell 161, 1192 \n1681-1696. 10.1016/j.cell.2015.05.044. 1193 \n42. Hoadley, K.A., Yau, C., Hinoue, T., Wolf, D.M., Lazar, A.J., Drill, E., Shen, R., Taylor, A.M., 1194 \nCherniack, A.D., Thorsson, V., et al. (2018). Cell-of-Origin Patterns Dominate the Molecular 1195 \nClassification of 10,000 Tumors from 33 Types of Cancer. Cell 173, 291-304 e296. 1196 \n10.1016/j.cell.2018.03.022. 1197 \n43. Choi, J., Xu, M., Makowski, M.M., Zhang, T., Law, M.H., Kovacs, M.A., Granzhan, A., Kim, W.J., 1198 \nParikh, H., Gartside, M., et al. (2017). A common intronic variant of PARP1 confers melanoma 1199 \nrisk and mediates melanocyte growth via regulation of MITF. Nat Genet 49, 1326-1335. 1200 \n10.1038/ng.3927. 1201 \n44. Roadmap Epigenomics, C., Kundaje, A., Meuleman, W., Ernst, J., Bilenky, M., Yen, A., Heravi-1202 \nMoussavi, A., Kheradpour, P., Zhang, Z., Wang, J., et al. (2015). Integrative analysis of 111 1203 \nreference human epigenomes. Nature 518, 317-330. 10.1038/nature14248. 1204 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n45. Verfaillie, A., Imrichova, H., Atak, Z.K., Dewaele, M., Rambow, F., Hulselmans, G., Christiaens, V., 1205 \nSvetlichnyy, D., Luciani, F., Van den Mooter, L., et al. (2015). Decoding the regulatory landscape 1206 \nof melanoma reveals TEADS as regulators of the invasive cell state. Nat Commun 6, 6683. 1207 \n10.1038/ncomms7683. 1208 \n46. Choi, J., Zhang, T., Vu, A., Ablain, J., Makowski, M.M., Colli, L.M., Xu, M., Hennessey, R.C., Yin, J., 1209 \nRothschild, H., et al. (2020). Massively parallel reporter assays of melanoma risk variants identify 1210 \nMX2 as a gene promoting melanoma. Nat Commun 11, 2718. 10.1038/s41467-020-16590-1. 1211 \n47. Long, E., Yin, J., Funderburk, K.M., Xu, M., Feng, J., Kane, A., Zhang, T., Myers, T., Golden, A., 1212 \nThakur, R., et al. (2022). Massively parallel reporter assays and variant scoring identified 1213 \nfunctional variants and target genes for melanoma loci and highlighted cell-type specificity. Am J 1214 \nHum Genet 109, 2210-2229. 10.1016/j.ajhg.2022.11.006. 1215 \n48. Xu, M., Mehl, L., Zhang, T., Thakur, R., Sowards, H., Myers, T., Jessop, L., Chesi, A., Johnson, M.E., 1216 \nWells, A.D., et al. (2021). A UVB-responsive common variant at chromosome band 7p21.1 1217 \nconfers tanning response and melanoma risk via regulation of the aryl hydrocarbon receptor, 1218 \nAHR. Am J Hum Genet 108, 1611-1630. 10.1016/j.ajhg.2021.07.002. 1219 \n49. Thakur, R., Xu, M., Sowards, H., Yon, J., Jessop, L., Myers, T., Zhang, T., Chari, R., Long, E., Rehling, 1220 \nT., et al. (2025). Mapping chromatin interactions at melanoma susceptibility loci uncovers 1221 \ndistant cis-regulatory gene targets. Am J Hum Genet 112, 1625-1648. 1222 \n10.1016/j.ajhg.2025.04.015. 1223 \n50. Rizos, H., Diefenbach, E., Badhwar, P., Woodruff, S., Becker, T.M., Rooney, R.J., and Kefford, R.F. 1224 \n(2003). Association of p14ARF with the p120E4F transcriptional repressor enhances cell cycle 1225 \ninhibition. J Biol Chem 278, 4981-4989. 10.1074/jbc.M210978200. 1226 \n51. Dai, J., Lv, J., Zhu, M., Wang, Y., Qin, N., Ma, H., He, Y.Q., Zhang, R., Tan, W., Fan, J., et al. (2019). 1227 \nIdentification of risk loci and a polygenic risk score for lung cancer: a large-scale prospective 1228 \ncohort study in Chinese populations. Lancet Respir Med 7, 881-891. 10.1016/S2213-1229 \n2600(19)30144-4. 1230 \n52. Schumacher, F.R., Al Olama, A.A., Berndt, S.I., Benlloch, S., Ahmed, M., Saunders, E.J., Dadaev, T., 1231 \nLeongamornlert, D., Anokian, E., Cieza-Borrella, C., et al. (2018). Association analyses of more 1232 \nthan 140,000 men identify 63 new prostate cancer susceptibility loci. Nat Genet 50, 928-936. 1233 \n10.1038/s41588-018-0142-8. 1234 \n53. Berndt, S.I., Skibola, C.F., Joseph, V., Camp, N.J., Nieters, A., Wang, Z., Cozen, W., Monnereau, A., 1235 \nWang, S.S., Kelly, R.S., et al. (2013). Genome-wide association study identifies multiple risk loci 1236 \nfor chronic lymphocytic leukemia. Nat Genet 45, 868-876. 10.1038/ng.2652. 1237 \n54. Zhang, X., Qin, N., Fan, J., Zhang, C., Sun, Q., Gu, Y., Zhu, M., Zhang, E., Dai, J., Jin, G., et al. (2022). 1238 \nA causal variant rs3769823 in 2q33.1 involved in apoptosis pathway leading to a decreased risk 1239 \nof non-small cell lung cancer. Cancer Biol Med 19, 1385-1396. 10.20892/j.issn.2095-1240 \n3941.2022.0068. 1241 \n55. Camp, N.J., Lin, W.Y., Bigelow, A., Burghel, G.J., Mosbruger, T.L., Parry, M.A., Waller, R.G., Rigas, 1242 \nS.H., Tai, P.Y., Berrett, K., et al. (2016). Discordant Haplotype Sequencing Identifies Functional 1243 \nVariants at the 2q33 Breast Cancer Risk Locus. Cancer Res 76, 1916-1925. 10.1158/0008-1244 \n5472.CAN-15-1629. 1245 \n56. Long, E., Yin, J., Shin, J.H., Li, Y., Li, B., Kane, A., Patel, H., Sun, X., Wang, C., Luong, T., et al. 1246 \n(2024). Context-aware single-cell multiomics approach identifies cell-type-specific lung cancer 1247 \nsusceptibility genes. Nat Commun 15, 7995. 10.1038/s41467-024-52356-9. 1248 \n57. Choi, E., and Cho, C. (2011). Expression of a sperm flagellum component encoded by the 1249 \nAls2cr12 gene. Gene Expr Patterns 11, 327-333. 10.1016/j.gep.2011.03.003. 1250 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n58. Leyva-Diaz, E., Stefanakis, N., Carrera, I., Glenwinkel, L., Wang, G., Driscoll, M., and Hobert, O. 1251 \n(2017). Silencing of Repetitive DNA Is Controlled by a Member of an Unusual Caenorhabditis 1252 \nelegans Gene Family. Genetics 207, 529-545. 10.1534/genetics.117.300134. 1253 \n59. Reddy, K.C., Dror, T., Sowa, J.N., Panek, J., Chen, K., Lim, E.S., Wang, D., and Troemel, E.R. (2017). 1254 \nAn Intracellular Pathogen Response Pathway Promotes Proteostasis in C. elegans. Curr Biol 27, 1255 \n3544-3553 e3545. 10.1016/j.cub.2017.10.009. 1256 \n60. Kischkel, F.C., Lawrence, D.A., Tinel, A., LeBlanc, H., Virmani, A., Schow, P., Gazdar, A., Blenis, J., 1257 \nArnott, D., and Ashkenazi, A. (2001). Death receptor recruitment of endogenous caspase-10 and 1258 \napoptosis initiation in the absence of caspase-8. J Biol Chem 276, 46639-46646. 1259 \n10.1074/jbc.M105102200. 1260 \n61. Sprick, M.R., Rieser, E., Stahl, H., Grosse-Wilde, A., Weigand, M.A., and Walczak, H. (2002). 1261 \nCaspase-10 is recruited to and activated at the native TRAIL and CD95 death-inducing signalling 1262 \ncomplexes in a FADD-dependent manner but can not functionally substitute caspase-8. EMBO J 1263 \n21, 4520-4530. 10.1093/emboj/cdf441. 1264 \n62. Wang, J., Chun, H.J., Wong, W., Spencer, D.M., and Lenardo, M.J. (2001). Caspase-10 is an 1265 \ninitiator caspase in death receptor signaling. Proc Natl Acad Sci U S A 98, 13884-13888. 1266 \n10.1073/pnas.241358198. 1267 \n63. Fischer, U., Stroh, C., and Schulze-Osthoff, K. (2006). Unique and overlapping substrate 1268 \nspecificities of caspase-8 and caspase-10. Oncogene 25, 152-159. 10.1038/sj.onc.1209015. 1269 \n64. Horn, S., Hughes, M.A., Schilling, R., Sticht, C., Tenev, T., Ploesser, M., Meier, P., Sprick, M.R., 1270 \nMacFarlane, M., and Leverkus, M. (2017). Caspase-10 Negatively Regulates Caspase-8-Mediated 1271 \nCell Death, Switching the Response to CD95L in Favor of NF-kappaB Activation and Cell Survival. 1272 \nCell Rep 19, 785-797. 10.1016/j.celrep.2017.04.010. 1273 \n65. Irmler, M., Thome, M., Hahne, M., Schneider, P., Hofmann, K., Steiner, V., Bodmer, J.L., Schroter, 1274 \nM., Burns, K., Mattmann, C., et al. (1997). Inhibition of death receptor signals by cellular FLIP. 1275 \nNature 388, 190-195. 10.1038/40657. 1276 \n66. Humphreys, L., Espona-Fiedler, M., and Longley, D.B. (2018). FLIP as a therapeutic target in 1277 \ncancer. FEBS J 285, 4104-4123. 10.1111/febs.14523. 1278 \n67. Leverkus, M., Neumann, M., Mengling, T., Rauch, C.T., Brocker, E.B., Krammer, P.H., and 1279 \nWalczak, H. (2000). Regulation of tumor necrosis factor-related apoptosis-inducing ligand 1280 \nsensitivity in primary and transformed human keratinocytes. Cancer Res 60, 553-559. 1281 \n68. Wachter, T., Sprick, M., Hausmann, D., Kerstan, A., McPherson, K., Stassi, G., Brocker, E.B., 1282 \nWalczak, H., and Leverkus, M. (2004). cFLIPL inhibits tumor necrosis factor-related apoptosis-1283 \ninducing ligand-mediated NF-kappaB activation at the death-inducing signaling complex in 1284 \nhuman keratinocytes. J Biol Chem 279, 52824-52834. 10.1074/jbc.M409554200. 1285 \n69. Siegmund, D., Hadwiger, P., Pfizenmaier, K., Vornlocher, H.P., and Wajant, H. (2002). Selective 1286 \ninhibition of FLICE-like inhibitory protein expression with small interfering RNA oligonucleotides 1287 \nis sufficient to sensitize tumor cells for TRAIL-induced apoptosis. Mol Med 8, 725-732. 1288 \n70. Zhang, T., Klein, A., Sang, J., Choi, J., and Brown, K.M. (2022). ezQTL: A Web Platform for 1289 \nInteractive Visualization and Colocalization of Quantitative Trait Loci and GWAS. bioRxiv. 1290 \n71. Weissbrod, O., Hormozdiari, F., Benner, C., Cui, R., Ulirsch, J., Gazal, S., Schoech, A.P., van de 1291 \nGeijn, B., Reshef, Y., Marquez-Luna, C., et al. (2020). Functionally informed fine-mapping and 1292 \npolygenic localization of complex trait heritability. Nat Genet 52, 1355-1363. 10.1038/s41588-1293 \n020-00735-5. 1294 \n72. Kircher, M., Witten, D.M., Jain, P., O'Roak, B.J., Cooper, G.M., and Shendure, J. (2014). A general 1295 \nframework for estimating the relative pathogenicity of human genetic variants. Nat Genet 46, 1296 \n310-315. 10.1038/ng.2892. 1297 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n73. Quang, D., Chen, Y., and Xie, X. (2015). DANN: a deep learning approach for annotating the 1298 \npathogenicity of genetic variants. Bioinformatics 31, 761-763. 10.1093/bioinformatics/btu703. 1299 \n74. Shihab, H.A., Rogers, M.F., Gough, J., Mort, M., Cooper, D.N., Day, I.N., Gaunt, T.R., and 1300 \nCampbell, C. (2015). An integrative approach to predicting the functional effects of non-coding 1301 \nand coding sequence variation. Bioinformatics 31, 1536-1543. 10.1093/bioinformatics/btv009. 1302 \n75. Fu, Y., Liu, Z., Lou, S., Bedford, J., Mu, X.J., Yip, K.Y., Khurana, E., and Gerstein, M. (2014). 1303 \nFunSeq2: a framework for prioritizing noncoding regulatory variants in cancer. Genome Biol 15, 1304 \n480. 10.1186/s13059-014-0480-5. 1305 \n76. Ritchie, G.R., Dunham, I., Zeggini, E., and Flicek, P. (2014). Functional annotation of noncoding 1306 \nsequence variants. Nat Methods 11, 294-296. 10.1038/nmeth.2832. 1307 \n77. Drozdetskiy, A., Cole, C., Procter, J., and Barton, G.J. (2015). JPred4: a protein secondary 1308 \nstructure prediction server. Nucleic Acids Res 43, W389-394. 10.1093/nar/gkv332. 1309 \n78. Cheng, J., Novati, G., Pan, J., Bycroft, C., Zemgulyte, A., Applebaum, T., Pritzel, A., Wong, L.H., 1310 \nZielinski, M., Sargeant, T., et al. (2023). Accurate proteome-wide missense variant effect 1311 \nprediction with AlphaMissense. Science 381, eadg7492. 10.1126/science.adg7492. 1312 \n79. Delaneau, O., Ongen, H., Brown, A.A., Fort, A., Panousis, N.I., and Dermitzakis, E.T. (2017). A 1313 \ncomplete tool set for molecular QTL discovery and analysis. Nat Commun 8, 15452. 1314 \n10.1038/ncomms15452. 1315 \n80. Dobin, A., Davis, C.A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S., Batut, P., Chaisson, M., and 1316 \nGingeras, T.R. (2013). STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15-21. 1317 \n10.1093/bioinformatics/bts635. 1318 \n81. Li, Y.I., Knowles, D.A., Humphrey, J., Barbeira, A.N., Dickinson, S.P., Im, H.K., and Pritchard, J.K. 1319 \n(2018). Annotation-free quantification of RNA splicing using LeafCutter. Nat Genet 50, 151-158. 1320 \n10.1038/s41588-017-0004-9. 1321 \n82. Wingett, S., Ewels, P., Furlan-Magaril, M., Nagano, T., Schoenfelder, S., Fraser, P., and Andrews, 1322 \nS. (2015). HiCUP: pipeline for mapping and processing Hi-C data. F1000Res 4, 1310. 1323 \n10.12688/f1000research.7334.1. 1324 \n83. Langmead, B., and Salzberg, S.L. (2012). Fast gapped-read alignment with Bowtie 2. Nat 1325 \nMethods 9, 357-359. 10.1038/nmeth.1923. 1326 \n84. Langmead, B., Wilks, C., Antonescu, V., and Charles, R. (2019). Scaling read aligners to hundreds 1327 \nof threads on general-purpose processors. Bioinformatics 35, 421-432. 1328 \n10.1093/bioinformatics/bty648. 1329 \n85. Cairns, J., Freire-Pritchett, P., Wingett, S.W., Varnai, C., Dimond, A., Plagnol, V., Zerbino, D., 1330 \nSchoenfelder, S., Javierre, B.M., Osborne, C., et al. (2016). CHiCAGO: robust detection of DNA 1331 \nlooping interactions in Capture Hi-C data. Genome Biol 17, 127. 10.1186/s13059-016-0992-2. 1332 \n86. Li, D., Hsu, S., Purushotham, D., Sears, R.L., and Wang, T. (2019). WashU Epigenome Browser 1333 \nupdate 2019. Nucleic Acids Res 47, W158-W165. 10.1093/nar/gkz348. 1334 \n87. Naumova, N., Smith, E.M., Zhan, Y., and Dekker, J. (2012). Analysis of long-range chromatin 1335 \ninteractions using Chromosome Conformation Capture. Methods 58, 192-203. 1336 \n10.1016/j.ymeth.2012.07.022. 1337 \n88. Spruijt, C.G., Baymaz, H.I., and Vermeulen, M. (2013). Identifying specific protein-DNA 1338 \ninteractions using SILAC-based quantitative proteomics. Methods Mol Biol 977, 137-157. 1339 \n10.1007/978-1-62703-284-1_11. 1340 \n89. Makowski, M.M., Willems, E., Fang, J., Choi, J., Zhang, T., Jansen, P.W., Brown, K.M., and 1341 \nVermeulen, M. (2016). An interaction proteomics survey of transcription factor binding at 1342 \nrecurrent TERT promoter mutations. Proteomics 16, 417-426. 10.1002/pmic.201500327. 1343 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint \n\n90. Cox, J., and Mann, M. (2008). MaxQuant enables high peptide identification rates, individualized 1344 \np.p.b.-range mass accuracies and proteome-wide protein quantification. Nat Biotechnol 26, 1345 \n1367-1372. 10.1038/nbt.1511. 1346 \n91. Nsengimana, J., Laye, J., Filia, A., O'Shea, S., Muralidhar, S., Pozniak, J., Droop, A., Chan, M., 1347 \nWalker, C., Parkinson, L., et al. (2018). beta-Catenin-mediated immune evasion pathway 1348 \nfrequently operates in primary cutaneous melanomas. J Clin Invest 128, 2048-2063. 1349 \n10.1172/JCI95351. 1350 \n92. Du, P., Kibbe, W.A., and Lin, S.M. (2008). lumi: a pipeline for processing Illumina microarray. 1351 \nBioinformatics 24, 1547-1548. 10.1093/bioinformatics/btn224. 1352 \n 1353 \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 6, 2026. ; https://doi.org/10.64898/2026.01.06.697866doi: bioRxiv preprint","source_license":"Public-Domain","license_restricted":false}