The pan-cancer landscape of prognostic germline variants in 10,582 patients

Abstract

Background: While clinical factors such as age, grade, stage, and histological subtype provide physicians with information about patient prognosis, genomic data can further improve these predictions. Previous studies have shown that germline variants in known cancer driver genes are predictive of patient outcome, but no study has systematically analyzed multiple cancers in an unbiased way to identify genetic loci that can improve patient outcome predictions made using clinical factors.

Methods: We analyzed sequencing data from the over 10,000 cancer patients available through The Cancer Genome Atlas to identify germline variants associated with patient outcome using multivariate Cox regression models.

Results: We identified 79 prognostic germline variants in individual cancers and 112 prognostic germline variants in groups of cancers. The germline variants identified in individual cancers provide additional predictive power about patient outcomes beyond clinical information currently in use and may therefore augment clinical decisions based on expected tumor aggressiveness. Molecularly, at least 12 of the germline variants are likely associated with patient outcome through perturbation of protein structure and at least five through association with gene expression differences. Almost half of these germline variants are in previously reported tumor suppressors, oncogenes or cancer driver genes with the other half pointing to genomic loci that should be further investigated for their roles in cancers.

Conclusions: Germline variants are predictive of outcome in cancer patients and specific germline variants can improve patient outcome predictions beyond predictions made using clinical factors alone. The germline variants also implicate new means by which known oncogenes, tumor suppressor genes, and driver genes are perturbed in cancer and suggest roles in cancer for other genes that have not been extensively studied in oncology. Further studies in other cancer cohorts are necessary to confirm that germline variation is associated with outcome in cancer patients as this is a proof-of-principle study.

Keywords: Cancer biology; Driver gene; Germline variants; Non-synonymous mutation; Oncogene; Pan-cancer; Single nucleotide polymorphism; Survival analysis; Tumor suppressor; eQTL.

Fig. 1

Prognostic germline variants identified in analyses 1 through 3. a A description of the three analyses used to identify prognostic germline variants in this figure. b Analysis 1. Germline variants found to be predictive of patient outcome in each cancer. Each dot represents a germline variant that was tested for an association with patient outcome. Variants closer to the outside of the plot are more closely associated with patient outcome. Variants in red are significantly (FDR < 0.10) associated with patient outcome. The alternating black and gray colors reflect alternating chromosomes for the germline variants that were not significant predictors of patient outcome. c Analysis 2. Germline variants found to be recurrently predictive of patient outcome in multiple different cancers. We identified five total germline variants that were recurrently predictive (p < 0.05) of favorable (HR < 1) or poor (HR > 1) patient outcomes in seven or more different cancers. d Analysis 3. A total of 29 groups of cancers created to identify germline variants with weaker effect sizes in larger patient cohorts. Justification for these groups is provided in Additional file 1: Table S3. e Analysis 3. Germline variants found to be predictive of patient outcome in the groups described in Fig. 1d. The format of the figure is the same as in Fig. 1b.

Fig. 2

Selected Kaplan-Meier plots of the prognostic germline variants from analysis 1. The number of patients in each group is indicated next to each line, and the patient outcome measure of each disease is given in Additional file 1: Table S1. The reported p values and hazard ratios were calculated using univariate regression and are different from the p values and hazard ratios reported elsewhere which are based on multivariate regression

Fig. 3

Prognostic germline variants that cause significant amino acid changes (CADD > 25) identified in analyses 4 through 6. a A description of the three analyses used to identify prognostic germline variants in this figure. b Analysis 4. Germline variants causing significant amino acid changes found to be predictive (FDR < 0.10) of patient outcome in each cancer. c Analysis 5. Germline variants causing significant amino acid changes found to be recurrently predictive (p < 0.05) of favorable (HR < 1) or poor (HR > 1) patient outcomes in 5 or more different cancers. d Analysis 6. Germline variants causing significant amino acid changes found to be predictive of patient outcome in patient groups defined in Fig. 1d

Fig. 4

Characteristics of prognostic germline variants and improvement of patient outcome models by the prognostic germline variants. a, b Scatterplots of the prognostic germline variants identified in individual cancers in analysis 1 (a) and in groups of cancers in analysis 3 (b). Each pie chart reflects the distribution of patients that are homozygous for the reference allele, heterozygous, and homozygous for the alternate allele for one prognostic variant. The minor allele was much more likely to be associated with increased risk for poor outcome rather than decreased risk for poor outcome (p = 7.077E−8) in analysis 1 though this trend was not significant in analysis 3 (p = 0.115). c, d Pie charts displaying the genomic locations of the germline variants in analysis 1 (c) and analysis 3 (d). e An example of a receiver operator characteristic (ROC) curve calculated using data from LAML at 366 days of follow-up. The blue line represents the patient outcome predictions made using clinical information alone (C model). The red line represents patient outcome predictions made using clinical information in addition to rs3003628 germline variant status (C + GV model), which we found to be predictive of patient outcomes in LAML. The area under the curve (AUC) was 0.81 for the C model and 0.93 for the C + GV model giving a ΔAUC of 0.12 (12%). f Many of the prognostic germline variants improve clinical outcome model predictions. For each prognostic variant, we created a ROC curve based on the clinical (C) model and the clinical + germline variant (C + GV model), as in Fig. 4e, at each point in time from the 10th-90th percentile of patient progression or death for each cancer. The ΔAUC of the C + GV model versus the C model at each time point was calculated (Additional file 3: Table S4). X-axis: Mean and standard error of ΔAUC. Y-axis: The p values from testing whether or not the AUC of the C + GV model is significantly greater than that of the C model using a Wilcoxon rank sum test. Four examples of prognostic germline variants that significantly increase the AUC are labeled and highlighted in Additional file 3: Table S4

Fig. 5

Literature review of genes associated with the prognostic germline variants and mechanisms by which prognostic germline variants may exert their effects. a The cancer-related functions of genes associated with the prognostic germline variants are quite diverse. b Many of the genes associated with the variants have previously been reported to be tumor suppressor genes or oncogenes. We categorized genes as tumor suppressor genes or oncogenes based on phenotypes reported in the literature, even if the exact mechanism through which the genes act have not yet been determined. c Although many of the variants have been studied in the field, there are many genes that have not yet been studied in the context of human disease and therefore may warrant investigation by the field. d Four of the genes associated with prognostic germline variants are in previously reported cancer driver genes. e Some of the prognostic germline variants cause dramatic amino acid changes and may disrupt well-characterized protein domains. f Some of the prognostic germline variants likely act as expression quantitative trait loci in cis (cis eQTLs) and the expression of these genes are predictive of patient outcome. We found three of these germline variants to also be eQTLs in the genotype tissue expression (GTEx) database in the same tissue that the tumor was derived from. g Some of the prognostic germline variants have been reported to be associated with other diseases related to the tissue from which the tumor was derived

Fig. 6

Examples by which two of the prognostic germline variants may be associated with patient outcome. a–c rs55796947 in MAP2K3/MKK3 is associated with favorable patient outcome in KIRC and results in complete loss of MAP2K3’s protein kinase domain due to a Q73* amino acid change. MAP2K3 inhibition has previously been reported to result in cell cycle arrest and response to chemotherapy drugs. Tumors with the variant show upregulation of genes involved with apoptotic cleavage (a), genes in the apoptotic execution phase (b), and downregulation of E2F targets (c) in a Gene Set Enrichment Analysis (GSEA) of RNAseq data. d–f rs77903511 in the apoptosis inhibitor BIRC5 is predictive of poor patient outcome in UVM (d). This variant is associated with increased BIRC5 expression (e). Elevated BIRC5 expression is associated with poor patient outcome (f)