Leveraging cross-cancer shared heritability to better understand the genetic architecture of cancer

摘要 全基因组关联研究（GWAS）和全转录组关联研究（TWAS）已经确定了 数百种常见的、中等效应的等位基因和基因与癌症风险相关，但大多数癌症 遗传性仍然无法解释。迄今为止，大多数癌症流行病学研究都集中在个体癌症上 类型我们建议利用癌症之间的共享遗传力来进行最大的跨癌症研究。 GWAS和TWAS迄今为止。为了实现我们的目标，我们将使用来自12个实体的单个和汇总GWAS数据， 癌症（乳腺癌、结肠直肠癌、子宫内膜癌、食道癌、神经胶质瘤、头颈癌、肺癌、黑素瘤、卵巢癌， 胰腺，前列腺和肾脏）基于超过400，000例病例和900，000例对照，扩展了我们先前的研究。 与六个新的癌症点和超过10万个新的癌症病例合作。 我们将进行总体和基于子集的跨癌症GWAS荟萃分析，以确定新的癌症 危险等位基因（Aim 1a）。我们还将开发明确测试多效性效应的统计方法， 仅进行汇总统计，并将其应用于已知和新的癌症SNP（目标1b）。创新和 应用跨癌TWAS的方法，利用两种肿瘤中基因表达的遗传调控， （TCGA）和正常（GTEx）组织（Aim 2）。最后，我们将使用既利用GWAS 来自dbGaP和UK Biobank的汇总统计数据和个体水平数据，以及功能注释数据 来自ENCODE和RoadMap表观基因组学项目，对癌症进行深入的遗传性分析。 具体来说，我们将建立风险等位基因的相对效应大小模型，作为等位基因频率和基因组频率的函数。 注释（目标3a），并首次评估多种癌症中显性效应的存在 (Aim 3 b）。建议的目标建立在我们以前成功使用大型GWAS汇总统计数据的基础上， 建立和量化多种癌症的共同遗传贡献。他们还建立在我们成熟的轨道上 开发和应用统计方法进行多表型关联研究的记录， 遗传力估计 我们的申请是为了响应PA-17-239：“现有数据的二次分析和集成， 阐明癌症风险和相关结果的遗传结构”。我们汇集了 来自12个不同癌症GWAS联盟的研究人员，创造了一个前所未有的机会， 癌症易感基因座作为拟议研究的一部分，我们将开发一系列新的统计方法 可以广泛应用于具有共同遗传基础的其他疾病组。实现我们的目标将 导致发现新癌症风险等位基因并鉴定参与肿瘤发展的共有途径 跨癌症。它还将为未来测序研究的设计和分析提供信息，以确定低- 与癌症风险相关的频率和罕见变异，通过提供合理效应量的指导， 所需的样本量和基因组特征最有可能窝藏大效应的低频变异。

ABSTRACT Genome-wide association studies (GWAS) and transcriptome-wide association studies (TWAS) have identified hundreds of common, modest-effect alleles and genes associated with cancer risk, but much of cancer heritability remains unexplained. To date, most epidemiological studies of cancer focus on individual cancer types. We propose to leverage the shared heritability across cancers to conduct the largest cross-cancer GWAS and TWAS to date. To achieve our goal, we will use individual and summary GWAS data from 12 solid cancers (breast, colorectal, endometrial, esophageal, glioma, head and neck, lung, melanoma, ovarian, pancreatic, prostate and renal) based on more than 400,000 cases and 900,000 controls expanding our prior work with six new cancer sites and more than 100,000 new cancer cases. We will conduct overall and subset-based cross-cancer GWAS meta-analysis to identify novel cancer risk alleles (Aim 1a). We will also develop statistical methods that explicitly test for pleiotropic effects using summary statistics only and apply these to both known and novel cancer SNPs (Aim 1b). We will develop and apply methods for cross-cancer TWAS, leveraging the genetic regulation of gene expression in both tumor (TCGA) and normal (GTEx) tissue (Aim 2). Finally, we will use novel methods that leverage both GWAS summary statistics and individual-level data from dbGaP and UK Biobank, as well as functional annotation data from the ENCODE and the RoadMap Epigenomics projects to conduct in-depth heritability analysis of cancer. Specifically, we will model the relative effect sizes of risk alleles as a function of allele frequency and genomic annotation (Aim 3a), and for the first time assess the presence of dominance effects across multiple cancers (Aim 3b). The proposed Aims build on our previous success in using large GWAS summary statistics to establish and quantify the shared genetic contribution to multiple cancers. They also build on our proven track record for developing and applying statistical methods to conduct multi-phenotype association studies and heritability estimation. Our application is in response to PA-17-239: “Secondary Analysis and Integration of Existing Data to Elucidate the Genetic Architecture of Cancer Risk and Related Outcomes”. We have brought together investigators from 12 different cancer GWAS consortia, creating an unprecedented opportunity to identify novel cancer susceptibility loci. As part of the proposed research, we will develop a series of new statistical methods that can be broadly applied to other disease groups with a shared genetic basis. Completion of our Aims will lead to discovery of novel cancer risk alleles and identify shared pathways involved in tumor development across cancers. It will also inform the design and analysis of future sequencing studies to identify low- frequency and rare variants associated with cancer risk, by providing guidance on plausible effect sizes, required sample sizes and the genomic features most likely to harbor large-effect low-frequency variants.