Finding Driver Pathways in Cancer: Models and Algorithms

ABSTRACT: BACKGROUND: Cancer sequencing projects are now measuring somatic mutations in large numbers of cancer genomes. A key challenge in interpreting these data is to distinguish driver mutations, mutations important for cancer development, from passenger mutations that have accumulated in somatic cells but without functional consequences. A common approach to identify genes harboring driver mutations is a single gene test that identifies individual genes that are recurrently mutated in a significant number of cancer genomes. However, the power of this test is reduced by: (1) the necessity of estimating the background mutation rate (BMR) for each gene; (2) the mutational heterogeneity in most cancers meaning that groups of genes (e.g. pathways), rather than single genes, are the primary target of mutations. RESULTS: We investigate the problem of discovering driver pathways, groups of genes containing driver mutations, directly from cancer mutation data and without prior knowledge of pathways or other interactions between genes. We introduce two generative models of somatic mutations in cancer and study the algorithmic complexity of discovering driver pathways in both models. We show that a single gene test for driver genes is highly sensitive to the estimate of the BMR. In contrast, we show that an algorithmic approach that maximizes a straightforward measure of the mutational properties of a driver pathway successfully discovers these groups of genes without an estimate of the BMR. Moreover, this approach is also successful in the case when the observed frequencies of passenger and driver mutations are indistinguishable, a situation where single gene tests fail. CONCLUSIONS: Accurate estimation of the BMR is a challenging task. Thus, methods that do not require an estimate of the BMR, such as the ones we provide here, can give increased power for the discovery of driver genes.     

Distinguishing between driver and passenger mutations in individual cancer genomes by network enrichment analysis

Background

In somatic cancer genomes, delineating genuine driver mutations against a background of multiple passenger events is a challenging task. The difficulty of determining function from sequence data and the low frequency of mutations are increasingly hindering the search for novel, less common cancer drivers. The accumulation of extensive amounts of data on somatic point and copy number alterations necessitates the development of systematic methods for driver mutation analysis.

Results

We introduce a framework for detecting driver mutations via functional network analysis, which is applied to individual genomes and does not require pooling multiple samples. It probabilistically evaluates 1) functional network links between different mutations in the same genome and 2) links between individual mutations and known cancer pathways. In addition, it can employ correlations of mutation patterns in pairs of genes. The method was used to analyze genomic alterations in two TCGA datasets, one for glioblastoma multiforme and another for ovarian carcinoma, which were generated using different approaches to mutation profiling. The proportions of drivers among the reported de novo point mutations in these cancers were estimated to be 57.8% and 16.8%, respectively. The both sets also included extended chromosomal regions with synchronous duplications or losses of multiple genes. We identified putative copy number driver events within many such segments. Finally, we summarized seemingly disparate mutations and discovered a functional network of collagen modifications in the glioblastoma. In order to select the most efficient network for use with this method, we used a novel, ROC curve-based procedure for benchmarking different network versions by their ability to recover pathway membership.

Conclusions

The results of our network-based procedure were in good agreement with published gold standard sets of cancer genes and were shown to complement and expand frequency-based driver analyses. On the other hand, three sequence-based methods applied to the same data yielded poor agreement with each other and with our results. We review the difference in driver proportions discovered by different sequencing approaches and discuss the functional roles of novel driver mutations. The software used in this work and the global network of functional couplings are publicly available at http://research.scilifelab.se/andrej_alexeyenko/downloads.html.

Electronic supplementary material

The online version of this article (doi:10.1186/1471-2105-15-308) contains supplementary material, which is available to authorized users.     

Statistical analysis of pathogenicity of somatic mutations in cancer

Recent large-scale sequencing studies have revealed that cancer genomes contain variable numbers of somatic point mutations distributed across many genes. These somatic mutations most likely include passenger mutations that are not cancer causing and pathogenic driver mutations in cancer genes. Establishing a significant presence of driver mutations in such data sets is of biological interest. Whereas current techniques from phylogeny are applicable to large data sets composed of singly mutated samples, recently exemplified with a p53 mutation database, methods for smaller data sets containing individual samples with multiple mutations need to be developed. By constructing distinct models of both the mutation process and selection pressure upon the cancer samples, exact statistical tests to examine this problem are devised. Tests to examine the significance of selection toward missense, nonsense, and splice site mutations are derived, along with tests assessing variation in selection between functional domains. Maximum-likelihood methods facilitate parameter estimation, including levels of selection pressure and minimum numbers of pathogenic mutations. These methods are illustrated with 25 breast cancers screened across the coding sequences of 518 kinase genes, revealing 90 base substitutions in 71 genes. Significant selection pressure upon truncating mutations was established. Furthermore, an estimated minimum of 29.8 mutations were pathogenic.     

A new deep learning technique reveals the exclusive functional contributions of individual cancer mutations

Cancers are caused by genomic alterations that may be inherited, induced by environmental carcinogens, or caused due to random replication errors. Postinduction of carcinogenicity, mutations further propagate and drastically alter the cancer genomes. Although a subset of driver mutations has been identified and characterized to date, most cancer-related somatic mutations are indistinguishable from germline variants or other noncancerous somatic mutations. Thus, such overlap impedes appreciation of many deleterious but previously uncharacterized somatic mutations. The major bottleneck arises due to patient-to-patient variability in mutational profiles, making it difficult to associate specific mutations with a given disease outcome. Here, we describe a newly developed technique Continuous Representation of Codon Switches (CRCS), a deep learning-based method that allows us to generate numerical vector representations of mutations, thereby enabling numerous machine learning-based tasks. We demonstrate three major applications of CRCS; first, we show how CRCS can help detect cancer-related somatic mutations in the absence of matched normal samples, which has applications in cell-free DNA–based assessment of tumor mutation burden. Second, the proposed approach also enables identification and exploration of driver genes; our analyses implicate DMD, RSK4, OFD1, WDR44, and AFF2 as potential cancer drivers. Finally, we used CRCS to score individual mutations in a tumor sample, which was found to be predictive of patient survival in bladder urothelial carcinoma, hepatocellular carcinoma, and lung adenocarcinoma. Taken together, we propose CRCS as a valuable computational tool for analysis of the functional significance of individual cancer mutations.     

Integrative Analysis of Somatic Mutations Altering MicroRNA Targeting in Cancer Genomes

Determining the functional impact of somatic mutations is crucial to understanding tumorigenesis and metastasis. Recent sequences of several cancers have provided comprehensive lists of somatic mutations across entire genomes, enabling investigation of the functional impact of somatic mutations in non-coding regions. Here, we study somatic mutations in 3′UTRs of genes that have been identified in four cancers and computationally predict how they may alter miRNA targeting, potentially resulting in dysregulation of the expression of the genes harboring these mutations. We find that somatic mutations create or disrupt putative miRNA target sites in the 3′UTRs of many genes, including several genes, such as MITF, EPHA3, TAL1, SCG3, and GSDMA, which have been previously associated with cancer. We also integrate the somatic mutations with germline mutations and results of association studies. Specifically, we identify putative miRNA target sites in the 3′UTRs of BMPR1B, KLK3, and SPRY4 that are disrupted by both somatic and germline mutations and, also, are in linkage disequilibrium blocks with high scoring markers from cancer association studies. The somatic mutation in BMPR1B is located in a target site of miR-125b; germline mutations in this target site have previously been both shown to disrupt regulation of BMPR1B by miR-125b and linked with cancer.     

Germline fitness-based scoring of cancer mutations

A key goal in cancer research is to find the genomic alterations that underlie malignant cells. Genomics has proved successful in identifying somatic variants at a large scale. However, it has become evident that a typical cancer exhibits a heterogenous mutation pattern across samples. Cases where the same alteration is observed repeatedly seem to be the exception rather than the norm. Thus, pinpointing the key alterations (driver mutations) from a background of variations with no direct causal link to cancer (passenger mutations) is difficult. Here we analyze somatic missense mutations from cancer samples and their healthy tissue counterparts (germline mutations) from the viewpoint of germline fitness. We calibrate a scoring system from protein domain alignments to score mutations and their target loci. We show first that this score predicts to a good degree the rate of polymorphism of the observed germline variation. The scoring is then applied to somatic mutations. We show that candidate cancer genes prone to copy number loss harbor mutations with germline fitness effects that are significantly more deleterious than expected by chance. This suggests that missense mutations play a driving role in tumor suppressor genes. Furthermore, these mutations fall preferably onto loci in sequence neighborhoods that are high scoring in terms of germline fitness. In contrast, for somatic mutations in candidate onco genes we do not observe a statistically significant effect. These results help to inform how to exploit germline fitness predictions in discovering new genes and mutations responsible for cancer.     

Chromatin accessibility of primary human cancers ties regional mutational processes and signatures with tissues of origin

Somatic mutations in cancer genomes are associated with DNA replication timing (RT) and chromatin accessibility (CA), however these observations are based on normal tissues and cell lines while primary cancer epigenomes remain uncharacterised. Here we use machine learning to model megabase-scale mutation burden in 2,500 whole cancer genomes and 17 cancer types via a compendium of 900 CA and RT profiles covering primary cancers, normal tissues, and cell lines. CA profiles of primary cancers, rather than those of normal tissues, are most predictive of regional mutagenesis in most cancer types. Feature prioritisation shows that the epigenomes of matching cancer types and organ systems are often the strongest predictors of regional mutation burden, highlighting disease-specific associations of mutational processes. The genomic distributions of mutational signatures are also shaped by the epigenomes of matched cancer and tissue types, with SBS5/40, carcinogenic and unknown signatures most accurately predicted by our models. In contrast, fewer associations of RT and regional mutagenesis are found. Lastly, the models highlight genomic regions with overrepresented mutations that dramatically exceed epigenome-derived expectations and show a pan-cancer convergence to genes and pathways involved in development and oncogenesis, indicating the potential of this approach for coding and non-coding driver discovery. The association of regional mutational processes with the epigenomes of primary cancers suggests that the landscape of passenger mutations is predominantly shaped by the epigenomes of cancer cells after oncogenic transformation.     

Guanine Holes Are Prominent Targets for Mutation in Cancer and Inherited Disease

Single base substitutions constitute the most frequent type of human gene mutation and are a leading cause of cancer and inherited disease. These alterations occur non-randomly in DNA, being strongly influenced by the local nucleotide sequence context. However, the molecular mechanisms underlying such sequence context-dependent mutagenesis are not fully understood. Using bioinformatics, computational and molecular modeling analyses, we have determined the frequencies of mutation at G•C bp in the context of all 64 5′-NGNN-3′ motifs that contain the mutation at the second position. Twenty-four datasets were employed, comprising >530,000 somatic single base substitutions from 21 cancer genomes, >77,000 germline single-base substitutions causing or associated with human inherited disease and 16.7 million benign germline single-nucleotide variants. In several cancer types, the number of mutated motifs correlated both with the free energies of base stacking and the energies required for abstracting an electron from the target guanines (ionization potentials). Similar correlations were also evident for the pathological missense and nonsense germline mutations, but only when the target guanines were located on the non-transcribed DNA strand. Likewise, pathogenic splicing mutations predominantly affected positions in which a purine was located on the non-transcribed DNA strand. Novel candidate driver mutations and tissue-specific mutational patterns were also identified in the cancer datasets. We conclude that electron transfer reactions within the DNA molecule contribute to sequence context-dependent mutagenesis, involving both somatic driver and passenger mutations in cancer, as well as germline alterations causing or associated with inherited disease.     

Mutational signatures of de-differentiation in functional non-coding regions of melanoma genomes

Much emphasis has been placed on the identification, functional characterization, and therapeutic potential of somatic variants in tumor genomes. However, the majority of somatic variants lie outside coding regions and their role in cancer progression remains to be determined. In order to establish a system to test the functional importance of non-coding somatic variants in cancer, we created a low-passage cell culture of a metastatic melanoma tumor sample. As a foundation for interpreting functional assays, we performed whole-genome sequencing and analysis of this cell culture, the metastatic tumor from which it was derived, and the patient-matched normal genomes. When comparing somatic mutations identified in the cell culture and tissue genomes, we observe concordance at the majority of single nucleotide variants, whereas copy number changes are more variable. To understand the functional impact of non-coding somatic variation, we leveraged functional data generated by the ENCODE Project Consortium. We analyzed regulatory regions derived from multiple different cell types and found that melanocyte-specific regions are among the most depleted for somatic mutation accumulation. Significant depletion in other cell types suggests the metastatic melanoma cells de-differentiated to a more basal regulatory state. Experimental identification of genome-wide regulatory sites in two different melanoma samples supports this observation. Together, these results show that mutation accumulation in metastatic melanoma is nonrandom across the genome and that a de-differentiated regulatory architecture is common among different samples. Our findings enable identification of the underlying genetic components of melanoma and define the differences between a tissue-derived tumor sample and the cell culture created from it. Such information helps establish a broader mechanistic understanding of the linkage between non-coding genomic variations and the cellular evolution of cancer.     

Signatures of accelerated somatic evolution in gene promoters in multiple cancer types

Cancer-associated somatic mutations outside protein-coding regions remain largely unexplored. Analyses of the TERT locus have indicated that non-coding regulatory mutations can be more frequent than previously suspected and play important roles in oncogenesis. Using a computational method called SASE-hunter, developed here, we identified a novel signature of accelerated somatic evolution (SASE) marked by a significant excess of somatic mutations localized in a genomic locus, and prioritized those loci that carried the signature in multiple cancer patients. Interestingly, even when an affected locus carried the signature in multiple individuals, the mutations contributing to SASE themselves were rarely recurrent at the base-pair resolution. In a pan-cancer analysis of 906 samples from 12 tumor types, we detected SASE in the promoters of several genes, including known cancer genes such as MYC, BCL2, RBM5 and WWOX. Nucleotide substitution patterns consistent with oxidative DNA damage and local somatic hypermutation appeared to contribute to this signature in selected gene promoters (e.g. MYC). SASEs in selected cancer gene promoters were associated with over-expression, and also correlated with the age of onset of cancer, aggressiveness of the disease and survival. Taken together, our work detects a hitherto under-appreciated and clinically important class of regulatory changes in cancer genomes.     

Mutations of C-Reactive Protein (CRP) -286 SNP,APC and p53 in Colorectal Cancer: Implication for a CRP-Wnt Crosstalk

C-reactive protein (CRP) is an established marker of inflammation with pattern-recognition receptor-like activities. Despite the close association of the serum level of CRP with the risk and prognosis of several types of cancer, it remains elusive whether CRP contributes directly to tumorigenesis or just represents a bystander marker. We have recently identified recurrent mutations at the SNP position -286 (rs3091244) in the promoter of CRP gene in several tumor types, instead suggesting that locally produced CRP is a potential driver of tumorigenesis. However, it is unknown whether the -286 site is the sole SNP position of CRP gene targeted for mutation and whether there is any association between CRP SNP mutations and other frequently mutated genes in tumors. Herein, we have examined the genotypes of three common CRP non-coding SNPs (rs7553007, rs1205, rs3093077) in tumor/normal sample pairs of 5 cancer types (n = 141). No recurrent somatic mutations are found at these SNP positions, indicating that the -286 SNP mutations are preferentially selected during the development of cancer. Further analysis reveals that the -286 SNP mutations of CRP tend to co-occur with mutated APC particularly in rectal cancer (p = 0.04; n = 67). By contrast, mutations of CRP and p53 or K-ras appear to be unrelated. There results thus underscore the functional importance of the -286 mutation of CRP in tumorigenesis and imply an interaction between CRP and Wnt signaling pathway.     

Universal Global Imprints of Genome Growth and Evolution – Equivalent Length and Cumulative Mutation Density

Background

Segmental duplication is widely held to be an important mode of genome growth and evolution. Yet how this would affect the global structure of genomes has been little discussed.

Methods/Principal Findings

Here, we show that equivalent length, or , a quantity determined by the variance of fluctuating part of the distribution of the -mer frequencies in a genome, characterizes the latter''s global structure. We computed the s of 865 complete chromosomes and found that they have nearly universal but (-dependent) values. The differences among the of a chromosome and those of its coding and non-coding parts were found to be slight.

Conclusions

We verified that these non-trivial results are natural consequences of a genome growth model characterized by random segmental duplication and random point mutation, but not of any model whose dominant growth mechanism is not segmental duplication. Our study also indicates that genomes have a nearly universal cumulative “point” mutation density of about 0.73 mutations per site that is compatible with the relatively low mutation rates of (15)10/site/Mya previously determined by sequence comparison for the human and E. coli genomes.     

VarWalker: Personalized Mutation Network Analysis of Putative Cancer Genes from Next-Generation Sequencing Data

A major challenge in interpreting the large volume of mutation data identified by next-generation sequencing (NGS) is to distinguish driver mutations from neutral passenger mutations to facilitate the identification of targetable genes and new drugs. Current approaches are primarily based on mutation frequencies of single-genes, which lack the power to detect infrequently mutated driver genes and ignore functional interconnection and regulation among cancer genes. We propose a novel mutation network method, VarWalker, to prioritize driver genes in large scale cancer mutation data. VarWalker fits generalized additive models for each sample based on sample-specific mutation profiles and builds on the joint frequency of both mutation genes and their close interactors. These interactors are selected and optimized using the Random Walk with Restart algorithm in a protein-protein interaction network. We applied the method in >300 tumor genomes in two large-scale NGS benchmark datasets: 183 lung adenocarcinoma samples and 121 melanoma samples. In each cancer, we derived a consensus mutation subnetwork containing significantly enriched consensus cancer genes and cancer-related functional pathways. These cancer-specific mutation networks were then validated using independent datasets for each cancer. Importantly, VarWalker prioritizes well-known, infrequently mutated genes, which are shown to interact with highly recurrently mutated genes yet have been ignored by conventional single-gene-based approaches. Utilizing VarWalker, we demonstrated that network-assisted approaches can be effectively adapted to facilitate the detection of cancer driver genes in NGS data.     

Identification of candidate genes for lung cancer somatic mutation test kits

Over the past three decades, mortality from lung cancer has sharply and continuously increased in China, ascending to the first cause of death among all types of cancer. The ability to identify the actual sequence of gene mutations may help doctors determine which mutations lead to precancerous lesions and which produce invasive carcinomas, especially using next-generation sequencing (NGS) technology. In this study, we analyzed the latest lung cancer data in the COSMIC database, in order to find genomic “hotspots” that are frequently mutated in human lung cancer genomes. The results revealed that the most frequently mutated lung cancer genes are EGFR, KRAS and TP53. In recent years, EGFR and KRAS lung cancer test kits have been utilized for detecting lung cancer patients, but they presented many disadvantages, as they proved to be of low sensitivity, labor-intensive and time-consuming. In this study, we constructed a more complete catalogue of lung cancer mutation events including 145 mutated genes. With the genes of this list it may be feasible to develop a NGS kit for lung cancer mutation detection.     

Evaluating the Effect of 3′-UTR Variants in DICER1 and DROSHA on Their Tissue-Specific Expression by miRNA Target Prediction

Untranslated gene regions (UTRs) play an important role in controlling gene expression. 3′-UTRs are primarily targeted by microRNA (miRNA) molecules that form complex gene regulatory networks. Cancer genomes are replete with non-coding mutations, many of which are connected to changes in tumor gene expression that accompany the development of cancer and are associated with resistance to therapy. Therefore, variants that occurred in 3′-UTR under cancer progression should be analysed to predict their phenotypic effect on gene expression, e.g., by evaluating their impact on miRNA target sites. Here, we analyze 3′-UTR variants in DICER1 and DROSHA genes in the context of myelodysplastic syndrome (MDS) development. The key features of this analysis include an assessment of both “canonical” and “non-canonical” types of mRNA-miRNA binding and tissue-specific profiling of miRNA interactions with wild-type and mutated genes. As a result, we obtained a list of DICER1 and DROSHA variants likely altering the miRNA sites and, therefore, potentially leading to the observed tissue-specific gene downregulation. All identified variants have low population frequency consistent with their potential association with pathology progression.     

Identification of Druggable Cancer Driver Genes Amplified across TCGA Datasets

The Cancer Genome Atlas (TCGA) projects have advanced our understanding of the driver mutations, genetic backgrounds, and key pathways activated across cancer types. Analysis of TCGA datasets have mostly focused on somatic mutations and translocations, with less emphasis placed on gene amplifications. Here we describe a bioinformatics screening strategy to identify putative cancer driver genes amplified across TCGA datasets. We carried out GISTIC2 analysis of TCGA datasets spanning 14 cancer subtypes and identified 461 genes that were amplified in two or more datasets. The list was narrowed to 73 cancer-associated genes with potential “druggable” properties. The majority of the genes were localized to 14 amplicons spread across the genome. To identify potential cancer driver genes, we analyzed gene copy number and mRNA expression data from individual patient samples and identified 40 putative cancer driver genes linked to diverse oncogenic processes. Oncogenic activity was further validated by siRNA/shRNA knockdown and by referencing the Project Achilles datasets. The amplified genes represented a number of gene families, including epigenetic regulators, cell cycle-associated genes, DNA damage response/repair genes, metabolic regulators, and genes linked to the Wnt, Notch, Hedgehog, JAK/STAT, NF-KB and MAPK signaling pathways. Among the 40 putative driver genes were known driver genes, such as EGFR, ERBB2 and PIK3CA. Wild-type KRAS was amplified in several cancer types, and KRAS-amplified cancer cell lines were most sensitive to KRAS shRNA, suggesting that KRAS amplification was an independent oncogenic event. A number of MAP kinase adapters were co-amplified with their receptor tyrosine kinases, such as the FGFR adapter FRS2 and the EGFR family adapter GRB7. The ubiquitin-like ligase DCUN1D1 and the histone methyltransferase NSD3 were also identified as novel putative cancer driver genes. We discuss the patient tailoring implications for existing cancer drug targets and we further discuss potential novel opportunities for drug discovery efforts.