Identification of genes associated with testicular germ cell tumor susceptibility through a transcriptome-wide association study

Abstract

Transcriptome-wide association studies (TWASs) have the potential to identify susceptibility genes associated with testicular germ cell tumors (TGCTs). We conducted a comprehensive TGCT TWAS by integrating genome-wide association study (GWAS) summary data with predicted expression models from normal testis, TGCT tissues, and a cross-tissue panel that encompasses shared regulatory features across 22 normal tissues, including the testis. Gene associations were evaluated while accounting for variant-level effects from GWASs, followed by fine-mapping analyses in regions exhibiting multiple TWAS signals, and finally supplemented by colocalization analysis. Expression and protein patterns of identified TWAS genes were further examined in relevant tissues. Our analysis tested 19,805 gene-disease links, revealing 165 TGCT-associated genes with a false discovery rate of less than 0.01. We prioritized 46 candidate genes by considering GWAS-inflated signals, correlations between neighboring genes, and evidence of colocalization. Among these, 23 genes overlap with 22 GWAS loci, with 7 being associations not previously implicated in TGCT risk. Additionally, 23 genes located within 21 loci are at least 1 Mb away from published GWAS index variants. The 46 prioritized genes display expression levels consistent with expected expression levels in human gonadal cell types and precursor tumor cells and significant enrichment in TGCTs. Additionally, immunohistochemistry revealed protein-level accumulation of two candidate genes, ARID3B and GINM1, in both precursor and tumor cells. These findings enhance our understanding of the genetic predisposition to TGCTs and underscore the importance of further functional investigations into these candidate genes.

Keywords: colocalization analysis; testicular germ cell tumors; transcriptome-wide association study.

Figure 1

Overview of the study design (A) Prediction models for heritable gene expression included in the analysis. Normal testis from the GTEx panel was selected as the primary target tissue, complemented by additional models from TGCTs from TCGA and cross-tissue features from 22 normal tissues from GTEx (mutually exclusive gene models selected in that order). Prediction models: blup, best linear unbiased predictor; enet, elastic-net regression; lasso, least absolute shrinkage and selection operator regression; susie, sum of single effects regression; top1, single best eQTL. (B) TWAS workflow to identify genes associated with predisposition to TGCTs. Significant genes were defined using a false discovery rate less than 0.01. Robust associations were identified using a gene permutation test p < 0.05. Genes were jointly modeled at each locus to identify conditionally independent genes (Bonferroni-adjusted joint-conditional p < 0.05). Colocalization analysis was subsequently conducted using a posterior probability (PP4) threshold of >0.5 to identify genes likely influenced by a shared causal variant affecting both gene expression and TGCT risk.

Figure 2

Manhattan/Miami plot summarizing the TWAS gene-TGCT associations in relation to GWAS signals Top: summary of TWAS results showing the strength of association of TGCT genetic risk for 19,805 genes. Named TWAS hits indicated by diamonds were independent of GWAS signals (i.e., permutation p < 0.05); color indicates whether the TWAS hit is located at a GWAS locus (purple diamond) or at a non-GWAS region (red diamond). The dotted horizontal line represents the significance threshold at false discovery rate < 0.01. Bottom: strength of SNP-level associations for TGCT risk (−log10 p value) as reported in the TGCT GWAS meta-analysis results. Shown are 1,161,660 SNPs that were included in the FUSION prediction models (i.e., SNPs that are present in the 1000 Genomes reference panel). Dotted horizontal lines indicate the GWAS-suggestive threshold (p < 1 × 10⁻⁵) in light gray and the GWAS-significance threshold (p < 5 × 10⁻⁸) in dark gray.

Figure 3

GWAS signals explained by top leading TWAS genes: UCK2 and GINM1 UCK2 (A) and GINM1 (B) have the lowest permutation p values among the 46 leading genes in Table 1. Each image consists of a top and bottom image. The top image shows all genes located at the risk locus according to the genomic position (hg19). Based on the joint modeling, genes are labeled as leading in dark blue (joint-conditionally independent), conditional on leading genes(s) in light blue, or correlated with leading genes in green (predicted expression R² > 0.9); marginally associated genes (p_TWAS ≥ 0.01 or p_permutation ≥ 0.05) and genes not tested (e.g., due to lack of a significant prediction model) are indicated with empty boxes with dark and light outlines, respectively. The bottom image shows results from the SNP-level GWAS-conditional analysis. Each point illustrates the association between a single SNP at the locus and TGCT status: gray points indicate the marginal association of a SNP with TGCT status (GWAS association) and green points indicate the association of the same SNP with TGCTs after conditioning of predicted expression of the leading gene at each locus (here, UCK2 and GINM1). The horizontal dashed line indicates the genome-wide significance threshold (p = 5 × 10⁻⁸).

Figure 4

Expression patterns of top leading genes at various tissues (A–C) Expression levels of ten most significant genes among genes overlapping a GWAS locus (left) or not (right). Genes are ordered by strength of evidence (p_permutation) with decreasing significance from left to right; expression levels of tissue-specific marker genes are shown for reference on the right. (A) Human embryonal gonads. Single-cell average gene expression (color panel) and proportion of cells with non-zero expression (dot size) at different stages of germ cell differentiation (primordial germ cell [PGC], pre-spermatogonia, and pre-oocyte cell types) as well as in male germ cells and supporting cells (Sertoli and fetal Leydig). (B) Pre-malignant tissue. Microarray expression from microdissected samples characterized as embryonal stem cells (ESCs), spermatogenesis, and germ cell neoplasia in situ (GCNIS). The color panel indicates the log10-average expression levels at each condition. (C) TGCTs. The color panel indicates the log10-average expression levels. (D) Protein localization of ARID3B, in GCNIS adjacent to a non-seminoma (embryonal carcinoma [EC]), and GCNIS adjacent to a seminoma. For each TGCT subtype, both the tumor and GCNIS component are shown in greater magnification at the bottom. Scale bars indicate 100 and 50 μm, respectively.

Figure 5

Network interactions and TGCT-relevant annotations for TWAS leading genes A protein-protein interaction network in STRING (string-db.org) for TWAS leading genes along with genes implicated by GWASs that were not included in the current study due to lack of a prediction model. Disconnected gene nodes without relevant annotations were removed. Line weights indicate the degree of confidence of an interaction between any two proteins based on STRING presets. Pathway annotation was based on manual curation as described in the discussion and Table S9.