Systematic identification of mutations and copy number alterations associated with cancer patient prognosis
- Google, Inc., United States
- Cold Spring Harbor Laboratory, United States
Figures
Figure 1 with 7 supplements
Single base-pair mutations convey limited prognostic information.
(A) Schematic of TP53 mutations and patient survival in the BLCA patient cohort. Red dots indicate missense mutations, blue dots indicate frameshift mutations, and purple dots indicate nonsense …
Figure 1—figure supplement 1
A schematic of the pan-cancer survival analysis pipeline and the datasets used.
(A) An outline of the data processing and analysis performed in this report is presented. 16 tumor types from TCGA were used as a discovery cohort. Mutation data from ICGC and copy number data from …
Figure 1—figure supplement 2
Cox proportional hazards survival analysis and the accuracy of TCGA clinical annotations.
(A) The relation between Z score and P value for −5 ≤ Z ≤ 5 is displayed. (B) Sample data for gene expression values for three hypothetical genes (A, B, and C) illustrating different Z scores. (C) …
Figure 1—figure supplement 3
The mutation status of TP53 is associated with outcome in multiple cancer types.
(A) Kaplan-Meier plots of TP53-mutant and TP53-WT tumors in every cancer type from the TCGA dataset. TP53 is associated with outcome in five of 16 cancer types (BRCA, GBMLGG, HNSC, LUAD, and PRAD). …
Figure 1—figure supplement 4
Hotspot mutations and mutations in multiple cancer driver genes are generally not associated with clinical prognosis.
(A) Lollipop plots of PIK3CA mutations in the BRCA cohort. The left plot displays all non-silent mutations in PIK3CA, while the right plot displays all mutations in ‘hotspot’ codons, for example, …
Figure 1—figure supplement 5
Excluding patients with hypermutated tumors or those who were treated with targeted therapies fails to reveal mutations significantly associated with outcome.
(A) Excluding patients treated with targeted therapies fails to significantly alter mutation Z scores. The most common targeted therapies in the TCGA cohort were BRAF inhibitors (used to treat …
Figure 1—figure supplement 6
Mutations with high variant allele frequencies are no more prognostic than mutations with low variant allele frequencies.
(A) The variant allele frequencies were calculated for all genes in 10 TCGA cohorts. Box plots of the VAFs of four common cancer drivers (TP53, PIK3CA, ARID1A, and NF1) are displayed. Boxes …
Figure 1—figure supplement 7
Prognostic mutations in glioma.
(A) Kaplan-Meier curves of the five genes with the strongest survival associations in GBMLGG (ATRX, EGFR, IDH1, TP53, and PTEN) are displayed. (B) Mutation patterns according to glioma subtype are …
Figure 2 with 4 supplements
Oncogene and tumor suppressor CNAs drive cancer patient mortality.
(A) Examples of driver gene CNAs associated with patient outcome. The copy number of CDKN2A, EGFR, and BRCA2 in the indicated patient cohorts are displayed, as well as Kaplan-Meier curves of patient …
Figure 2—figure supplement 1
Discretized copy number values still hold significant prognostic power.
(A) A heatmap of significant survival associations among the 30 most frequently-mutated cancer driver genes in 16 tumor types from the TCGA are displayed. Z scores were calculated by regressing gene …
Figure 2—figure supplement 2
The prognostic value of cancer CNAs is independent of tumor sample purity.
(A) A bar graph showing Z scores obtained by regressing sample purity, as measured by IHC, against patient survival. Dotted lines indicate Z scores of 1.96 and −1.96, corresponding to a P value < 0.0…
Figure 2—figure supplement 3
CNAs remain prognostic after correcting for tumor stage and grade.
(A) Z scores were calculated for multivariate Cox proportional hazards models including both gene copy number and tumor stage or grade. Density plots display the correlation between Z scores …
Figure 2—figure supplement 4
CNAs remain prognostic after correcting for tumor subtype.
(A) The prognostic value of gene-level CNAs within 18 different cancer subtypes present within the TCGA were analyzed. Density plots display the correlations between Z scores obtained from …
Figure 3 with 3 supplements
Effects of amplicon size and gene mutation status on prognostic CNAs.
(A) 20 prognostic amplifications and 20 prognostic deletions were selected for further analysis (see also Figure 3—figure supplement 2). Of those 40, 14 had at least five patients who had focal CNAs …
Figure 3—figure supplement 1
Gene-level CNAs, TP53 status, total tumor aneuploidy, and total alteration burden.
(A) The total number of arm-length aneuploidies per sample is plotted for TP53-wildtype and TP53-mutant tumors from each TCGA cohort. Boxes represent the second and third quartiles, while error bars …
Figure 3—figure supplement 2
Multivariate analysis of prognostic CNAs.
(A) 20 prognostic deletions and 20 prognostic amplifications were selected for further study. Multivariate Cox proportional hazards models were constructed including both the copy number of the …
Figure 3—figure supplement 3
Prognostic CNAs alter the expression of the gene that they encompass.
(A) The Pearson correlation coefficients between gene copy number and gene expression at 40 prognostic loci are displayed. (B) A scatter plot showing the correspondence between CCND1 copy number and …
Figure 4
Driver gene copy number, but not driver gene mutations, are associated with survival in independent patient cohorts.
(A) Genes mutated in ≥10% of patients in each tumor type in the TCGA were identified, and then compared to the mutation frequency of these genes in the corresponding ICGC cohort or cohorts. The …
Figure 5 with 2 supplements
Robust prognostic biomarkers associated with drug sensitivity in cancer cell lines.
(A) Mutations and CNAs associated with patient outcome in multiple cohorts of glioma/glioblastoma are displayed. Mutations in STAG2 are associated with sensitivity to the PARP inhibitor olaparib, …
Figure 5—figure supplement 1
Multivariate analysis of high-confidence biomarkers with standard clinical criteria.
(A–D) Forest plots of the indicated CNAs in different cohorts in multivariate Cox models that include commonly-measured clinical variables. Circles indicate the hazard ratio, while the bars indicate …
Figure 5—figure supplement 2
Robust prognostic biomarkers associated with drug sensitivity in cancer cell lines.
(A) The number of high-confidence genetic biomarkers identified in each of the indicated tumor types is displayed. The complete list of biomarkers is presented in Supplementary file 8. (B) Mutations …
Additional files
-
Supplementary file 1
Cancer survival cohorts analyzed in this study.
- https://doi.org/10.7554/eLife.39217.024
-
Supplementary file 2
Cox proportional hazards modeling of mutations in the TCGA.
- https://doi.org/10.7554/eLife.39217.025
-
Supplementary file 3
Cox proportional hazards modeling of CNAs in the TCGA.
- https://doi.org/10.7554/eLife.39217.026
-
Supplementary file 4
Cox proportional hazards modeling in the TCGA adjusted for stage, grade, or subtype.
- https://doi.org/10.7554/eLife.39217.027
-
Supplementary file 5
Cox proportional hazards modeling adjusted for TP53 status or total aneuploidy.
- https://doi.org/10.7554/eLife.39217.028
-
Supplementary file 6
Cox proportional hazards modeling of cancer cohorts from the ICGC or curated by cBioportal.
- https://doi.org/10.7554/eLife.39217.029
-
Supplementary file 7
Cox proportional hazards modeling of the MSKCC_2017 cohorts.
- https://doi.org/10.7554/eLife.39217.030
-
Supplementary file 8
High-confidence biomarkers and their associated therapeutic sensitivities.
- https://doi.org/10.7554/eLife.39217.031
-
Transparent reporting form
- https://doi.org/10.7554/eLife.39217.032
