Detection of driver mutations and genomic signatures in endometrial cancers using artificial intelligence algorithms

Abstract

Analyzed endometrial cancer (EC) genomes have allowed for the identification of molecular signatures, which enable the classification, and sometimes prognostication, of these cancers. Artificial intelligence algorithms have facilitated the partitioning of mutations into driver and passenger based on a variety of parameters, including gene function and frequency of mutation. Here, we undertook an evaluation of EC cancer genomes deposited on the Catalogue of Somatic Mutations in Cancers (COSMIC), with the goal to classify all mutations as either driver or passenger. Our analysis showed that approximately 2.5% of all mutations are driver and cause cellular transformation and immortalization. We also characterized nucleotide level mutation signatures, gross chromosomal re-arrangements, and gene expression profiles. We observed that endometrial cancers show distinct nucleotide substitution and chromosomal re-arrangement signatures compared to other cancers. We also identified high expression levels of the CLDN18 claudin gene, which is involved in growth, survival, metastasis and proliferation. We then used in silico protein structure analysis to examine the effect of certain previously uncharacterized driver mutations on protein structure. We found that certain mutations in CTNNB1 and TP53 increase protein stability, which may contribute to cellular transformation. While our analysis retrieved previously classified mutations and genomic alterations, which is to be expected, this study also identified new signatures. Additionally, we show that artificial intelligence algorithms can be effectively leveraged to accurately predict key drivers of cancer. This analysis will expand our understanding of ECs and improve the molecular toolbox for classification, diagnosis, or potential treatment of these cancers.

Fig 1. Mutation distribution in endometrial cancers.

A. Percent coding and non-coding mutations in endometrial cancers. B. Distribution of coding mutations by endometrial cancer histology. C. Percent coding driver (CHASM) and pathogenic (VEST4) mutations out of total mutations in A in endometrial cancers. D. Distribution of nucleotide substitutions in coding endometrial mutations (blue) compared with coding mutations in other cancers (red). Significant p values between the two samples are shown (t-test: independent samples, unequal variance, two tails). E. Chromosomal re-arrangements in endometrial cancers. Data extracted using the CONNAN function on COSMIC.

Fig 2. Genes and pathways affected by endometrial cancer mutations.

A. Highly mutated genes detected in endometrial cancer patients. Shown are only those genes that are mutated in at least 100 patients out of 3213 (66 genes). For each of the 66 genes, we computed their occurrence in endometrial cancers compared with occurrence in all other cancers (expressed as percent). Red stars represent those genes with highest frequency of driver mutations. Complete data in S2 and S4 Tables. B. Biological processes affected by most frequent mutations in endometrial cancers. Only those processes with a string score above 0.1 are shown. C. Percent driver and passenger mutations for genes in A as determined by the CHASM algorithm.

Fig 3. Driver mutations and copy number variations of most significantly altered genes.

A. Domain diagrams of genes with location of driver mutations shown. Only driver mutations identified from CHASM are mapped onto these diagrams and the high frequency ones are indicated in red. The number of incidences of a certain mutation is shown in parentheses. All gene maps were made based on previous reports: PPP2R1A [, –88], FGFR2 [73, 89], PTEN [–94], PIK3CA [95, 96], CTNNB1 [68, 70, 97, 98], TP53 [99, 100], and BCOR [101]. B. Copy number variations of most significantly altered genes. Each dot represents one sample where high copy number was detected.

Fig 4. Top five driver mutations that alter protein electrostatic surface potential.

Surface rendering of the protein structure is shown with basic or positive surface potential colored blue, acidic or negative colored red, and neutral colored white. The WT or mutant residue location is identified by a black or white circle. A. PPP2R1A WT R183 residue compared to B. PPP2R1A mutant Q183 residue. C. PTEN WT R130 residue with a tartrate molecule shown in yellow sticks compared to D. PTEN mutant P130 residue. E. PIK3CA WT E545 residue compared to F. PIK3CA mutant A545 residue. G. CTNNB1 WT D32 residue compared to H. CTNNB1 mutant H32 residue. I. CTNNB1 WT G34 residue compared to J. CTNNB1 mutant E34 residue.

Fig 5. Major molecular pathways affected by driver mutations in endometrial cancers.

Please see text for details and discussion.