Pan-cancer analysis of intratumor heterogeneity as a prognostic determinant of survival

Abstract

As tumors accumulate genetic alterations, an evolutionary process occurs in which genetically distinct subclonal populations of cells co-exist, resulting in intratumor genetic heterogeneity (ITH). The clinical implications of ITH remain poorly defined. Data are limited with respect to whether ITH is an independent determinant of patient survival outcomes, across different cancer types. Here, we report the results of a pan-cancer analysis of over 3300 tumors, showing a varied landscape of ITH across 9 cancer types. While some gene mutations are subclonal, the majority of driver gene mutations are clonal events, present in nearly all cancer cells. Strikingly, high levels of ITH are associated with poorer survival across diverse types of cancer. The adverse impact of high ITH is independent of other clinical, pathologic and molecular factors. High ITH tends to be associated with lower levels of tumor-infiltrating immune cells, but this association is not able to explain the observed survival differences. Together, these data show that ITH is a prognostic marker in multiple cancers. These results illuminate the natural history of cancer evolution, indicating that tumor heterogeneity represents a significant obstacle to cancer control.

Keywords: cancer; evolution; heterogeneity; immune surveillance; survival.

Figure 1. Analysis of intratumor heterogeneity, and its prognostic significance, using multifaceted genomic data

SNP6 array data are used to determine allele-specific copy number and tumor purity. These data are integrated with mutation variant allele frequencies in exome sequencing data to infer the cellular prevalence of each mutation. (Sub)clonal populations are identified by clustering mutations by their cellular prevalence. Multivariable models of survival can then be generated, incorporating intratumor heterogeneity and other relevant clinical factors.

Figure 2. The landscape of intratumor heterogeneity across different types of cancer

A. Histograms of the distribution of (sub)clonal populations in 9 cancer types. B. Boxplots of intratumor heterogeneity, expressed as the Shannon Index of diversity, by cancer type. The Shannon Index approaches zero as the number of (sub)clonal populations decreases, or become more equal in size. C. Spectrum of clonality among recurrently mutated (MutSig q<0.10, present in ≥3 cancer types) genes. Each pie is colored by the mean cancer cell prevalence of the gene mutation in that cancer type. The size of the white pie slice indicates the proportion of mutations categorized as subclonal (<70% cellular prevalence). For a more comprehensive survey of genes, see Supplementary Figure 1.

Figure 3. Intratumor heterogeneity is associated with poorer patient survival

A. Kaplan-Meier curve of overall survival by degree of intratumor heterogeneity in the discovery dataset (HNSC). A trend toward poorer overall survival is evident as the number of subclonal populations increases. The log-rank test was used for comparisons. B. Overall survival for low and high ITH HNSC tumors, adjusting for HPV status, stage, and TP53 mutation status. The survival curve is plotted at the mean of other covariates in the multivariable model. C. Survival curves for low and high ITH tumors in 8 additional cancer types. In each case, curves are plotted at the mean of other covariates adjusted for in the regression models. The x-axis ranges from 0-5 years. The y-axis depicts the probability of overall survival, except for PRAD, where relapse-free survival is used. D. Hazard ratios for death (relapse in PRAD), with 95% confidence intervals, for each cancer type, where ITH is adjusted for other covariates as shown in C.

Figure 4. Associations of intratumor heterogeneity with other clinical, molecular, and immune factors

A. Distribution of (sub)clonal populations by HPV status in HNSC, showing a non-significant trend toward higher ITH in HPV+ tumors. B. Distribution of (sub)clonal populations by molecular and histologic subtype in LGG. High ITH was more common in IDH mutant, 1p-19q co-deleted tumors, and in oligodendrogliomas. C. Distribution of (sub)clonal populations by molecular and receptor subtype in BRCA, showing no differences between subgroups. Categorical comparisons were made with the Fisher exact or χ² test. D. Mean levels of RNAseq-derived immune cell infiltration, by cancer type. Error bars represent ± 1 SEM. E. Column graph showing enrichment for immune cell infiltration in tumors with low ITH, by cancer type. The y-axis represents the z-score of absolute increase, or decrease, in immune infiltration in low ITH tumors.