Genome-wide identification and analysis of prognostic features in human cancers

Abstract

Clinical decisions in cancer rely on precisely assessing patient risk. To improve our ability to accurately identify the most aggressive malignancies, we constructed genome-wide survival models using gene expression, copy number, methylation, and mutation data from 10,884 patients with known clinical outcomes. We identified more than 100,000 significant prognostic biomarkers and demonstrate that these genomic features can predict patient outcomes in clinically-ambiguous situations. While adverse biomarkers are commonly believed to represent cancer driver genes and promising therapeutic targets, we show that cancer features associated with shorter survival times are not enriched for either oncogenes or for successful drug targets. Instead, the strongest adverse biomarkers represent widely-expressed housekeeping genes with roles in cell cycle progression, and, correspondingly, nearly all therapies directed against these features have failed in clinical trials. In total, our analysis establishes a rich resource for prognostic biomarker analysis and clarifies the use of patient survival data in preclinical cancer research and therapeutic development.

Distal gene regulation mediated by non-coding RNAs contributes to germline risk for breast and prostate cancer

“…Again, in tumor-specific gene-gene associations, we find that many ncRNAs have associations with multiple pcGenes, leading to vertical bands in the location plots in Figure 2B . For example, in breast tumors, LINC000906 , with predicted activity as an miRNA sponge in breast tumors and correlated with overall survival in breast cancer patients, was associated with 115 pcGenes, the largest association with MS4A5 , a gene whose hypomethylation is shown to be prognostic for multiple cancer types 66, 67 . LINC00115 , a known promoter of breast and ovarian cancer metastasis and progression 68–70 , is another example of an ncRNA associated with multiple different pcGenes, many of which are interferon-related genes ( IFNA17 and IFNW1 ), related to immune system cytotoxicity ( RAC2 and DDB2 ), or coding for secretory proteins ( PRH1 and PRH2 ).…”

Widespread redundancy in -omics profiles of cancer mutation states

“…As a complementary comparison to mutation prediction, we constructed predictors of patient survival using the clinical data available from the GDC, in the TCGA-CDR-SupplementalTableS1.xlsx file. Following the methods described in [44], as the clinical endpoint we used overall survival (OS), except in nine cancer types with few deaths observed where we used progression-free intervals (PFI) as the clinical endpoint (BRCA, DLBC, LGG, PCPG, PRAD, READ, TGCT, THCA and THYM). For prediction, we used Cox regression as implemented in the scikit-survival Python package [45], with patient age at diagnosis and log 10 (sample mutation count) included as covariates, as well as a one-hot encoded variable for cancer type in the pan-cancer case.…”

Identifying tumor cells at the single cell level

“…Interestingly, when stratified by cancer type, tumor signature genes reported to be unfavorable are overrepresented among 5 cancer types: liver, renal, pancreatic, lung and endometrial cancers (Figure 5B). An analogous analysis was done with data taken from [43], where the authors systematically calculated the risk predictive status for all genes in TCGA cancer types. The analysis showed that the cancer specific genes have a significantly higher Stoufer's Z value (a measure of how significantly the gene expression predicts the risk status in any cancer) than the rest of the annotated gene set (Figure 5C).…”

“…-Human protein atlas (https://www.proteinatlas.org/humanproteome/pathology) [42] -Prognostic genes [43] -Gene fusion (ChiTaRs) [44] -SEEK (co-expression database) (https://seek.princeton.edu/seek/) [39] -g:Profiler (https://biit.cs.ut.ee/gprofiler/) [41] -CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA/home.jsp) [38] -MsigDB (GO, Hallmark gene sets) [40] -Atlas of co-essential modules [47] -DepMap Achilles scores (https://depmap.org/portal/download/) [48] -COSMIC (cancer.sanger.ac.uk) [49] Code Availability Ikarus is a python package that can be found on the following link: https://github.com/BIMSBbioinfo/ikarus Code for reproducing the figures can be found on the following link: https://github.com/BIMSBbioinfo/ikarus---auxiliary Author contributions AA conceptualized and planned the project. JD and AB jointly executed all of the computational analyses.…”

“…B) The association of tumor gene signature genes with poor survival outcomes is evident in multiple cancer types. C) Data from [43], show that the tumor gene signature genes have much higher Stoufer's z (association with poor survival outcomes) than rest of protein coding genes. D) Increased association of the tumor signature genes with negative survival is apparent in 22 out of 33 cancer types.…”