Article Figures & Data
Figures
- Figure 1.
A flowchart describing the steps involved in identifying prognostic germline variants.
- Figure 2.
Prognostic germline variants in the TCGA dataset. A, Of the 4.4 million unique variants called in the TCGA dataset, we ran Cox regression on the 196,022 germline variants found in gnomAD with an allele frequency greater than 5% and found in 15 or more of the TCGA lower grade glioma patients. B–E, Similar to the 196,022 germline variants, the 271 prognostic variants are mostly found in protein-coding genes (B), are located in introns (C), and are single nucleotide polymorphisms (SNP; D). E, Most single-nucleotide polymorphisms cause transitions.
- Figure 3.
rs61757955 is a highly prognostic germline variant identified in the TCGA dataset. A, Manhattan plot showing the P values resulting from testing each germline variant by Cox regression, controlling for the 11 variables in bolded in Table 1. Two variants passed the FDR threshold in the TCGA dataset. B, A Kaplan–Meier plot depicting the deleterious outcome associated with rs61757955, adjusting for the eleven covariates. C, ROC curve at 7 years. rs61757955 increases the area under the curve compared with the 11 covariates alone, suggesting that it improves the clinical model. D, Separation of patients on the basis of whether or not they have this germline variant to determine which genes are induced or repressed in patients with rs61757955. Subsequent gene set enrichment analysis reveals that patients with this germline variant exhibit upregulation of the genes involved with KRAS signaling.
- Figure 4.
rs34988193 is a prognostic variant predicted to be highly deleterious. A, A Manhattan plot with the P values resulting from testing each germline variant by Cox regression, controlling for the 11 covariates in bolded in Table 1. rs34988193 is prognostic (FDR < 0.10) in the TCGA when restricting the analysis to the top 0.1% most deleterious variants by CADD. B and C, Kaplan–Meier plots depicting the deleterious outcome associated with rs34988193 in the TCGA (C) and CGGA datasets (D), adjusted for the 11 covariates. D, A schematic showing that this variant is located in the ninth ankyrin repeat of ANKDD1a. The predicted protein structure of ANKDD1a reveals that this variant leads to an amino acid change from lysine to glutamate on the loop of an ankyrin repeat. E, Multiple sequence alignment of ANKDD1a in 22 species showing that lysine is conserved at this position in all of the species with this protein. F, ROC curves comparing the ability of two survival models to label patients as alive or dead after 7 years of follow up. The inclusion of rs61757955 variant status, rs34988193 variant status, GRB2 expression, and ANKDD1a expression significantly improves the survival prediction compared with the 11 covariates bolded in Table 1 alone (P = 0.0279).
Tables
- Table 1.
List of variables that are known to be associated with differences in survival in lower grade glioma patients
Covariate Median (min–max) or number of patients Age 41 (14–87) Gender Female 250 Male 200 Somatic mutation count 50 (0–12255) Percent aneuploidy 11% (5.2E-4%–95%) log(TERT Expression) 1.0 (0.0–9.1) FPKM Principal component 1 (PC1) 0.043 (−0.091–0.064) Principal component 2 (PC2) −0.017 (−0.23–0.17) Principal component 3 (PC3) −0.53 (1.34E-4–0.33) Histologic type Astroctyoma 172 Oligoastrocytoma 113 Oligodendroglioma 165 Tumor location Frontal lobe 265 Temporal lobe 125 Parietal lobe 42 Other 18 Grade G2 212 G3 237 Cannot be assessed 1 Treatment site Henry Fords Hospital 82 Case Western St. Joseph Hospital 90 Other 278 IDH Mutant Wild type 83 Mutant 367 1p/19q Codeletion Absent 303 Present 147 MGMT Promoter methylation Unmethylated 80 Methylated 370 Chr 7 gain/Chr 10 loss Absent 399 Present 51 TP53 Mutant Absent 232 Present 218 NOTE: Eleven variables (bolded) were selected by Lasso for inclusion in the survival model. We used these 11 variables as covariates in our Cox regression model when testing each germline variant.
- Table 2.
Description of the prognostic germline variants identified in this study
Aa Variant Chrom Pos Ref Alt Population frequency Sample size Number of heterozygotes Number of homozygotes Gene name Median expression (FPKM) P HR rs61757955 17 75318086 A G 5.01% 291 21 0 GRB2 42.2 7.08E-10 20.4 rs28672782 11 1446622 C T 16.27% 50 15 5 BRSK2 7.29 <1E-16 1.15E-10 Bb Variant Chrom Pos Ref Alt Population frequency CADD Score PhyloP Score Gene name Dataset Sample size Number of heterozygotes Number of homozygotes P HR rs34988193 15 64943580 A G 30.90% 32 8.42 ANKDD1a TCGA 450 199 52 0.00113 1.73 CGGA 76 18 2 0.0743 1.79 ↵aA description of the two prognostic germline variants (FDR < 0.10) in the TCGA dataset identified when testing all 196,022 germline variants.
↵bA description of the prognostic germline variant (FDR < 0.10) rs34988193 in ANKDD1a identified when the analysis was restricted to only germline variants with a combined annotation dependent depletion (CADD) score greater than 30 in the TCGA and CGGA datasets.
- Table 3.
The association between the germline variant rs61757955 and genomic and histological variables
Variable Mean or Percentage (Wild Type) Mean or Percentage (Mutant) P-value CIC Mutated 15.9% 38.1% 0.017 1p/19q Co-deletion 25.2% 47.6% 0.038 Oligodendroglioma 33.7% 42.9% 0.475 Total Somatic Mutation Count 30.9 30.0 0.766 Percent Aneuploidy 15.1% 11.7% 0.524 Astrocytoma 38.1% 42.9% 0.651 Grade 3 53.0% 42.9% 0.497 IDH Mutated 78.1% 85.7% 0.583 1p/19q Co-deletion 25.2% 47.6% 0.038 MGMT Promoter Methylation 77.8% 81.0% 1.000 Chr 7 Gain/Chr 10 Loss 13.0% 9.5% 1.000 Expression of GRB2 (FPKM) 45.7 44.4 0.636 NOTE: Patients were divided based on whether or not they had the germline variant rs61757955. Patients with the germline variant rs61757955 were more likely (p < 0.05) to have CIC-mutated gliomas and the 1p/19q codeletion (bold).
Supplementary Data
- All Supplementary Figures and Tables - S1. A boxplot representing the percentage of variants called in the whole exome sequenced (WXS) tumor sample that is likely somatic mutations. S2. Correlation between the variant allele frequencies calculated from the four variant sets and the distribution of allele frequencies. S3. Principal components calculated from germline variants from the whole exome sequencing data from the non-tumor samples. S4. Kaplan-Meier plot for the germline variant rs28672782 in BRSK2. Table S1. Quality control checks reveal that somatic mutations and RNA editing did not affect the results of our analysis. Table S2. Correlation between the allele frequencies calculated in our four variant sets and the allele frequencies reported by gnomAD. Table S3. Variants genetically linked to rs61757955 in the European population, the population that is most similar to the TCGA lower grade glioma patient population. Table S4. Results from testing for an association between the germline variant rs34988193 and genomic and histological variables.
Volume 17, Issue 5
