REVIEW ARTICLE Understanding oncogenicity of cancer driver genes and mutations in the cancer genomics era Eduard Porta-Pardo1,2, Alfonso Valencia1,3 and Adam Godzik4 1 Barcelona Supercomputing Center (BSC), Barcelona, Spain 2 Josep Carreras Leukaemia Research Institute (IJC), Badalona, Spain 3 Institucio Catalana de Recerca I Estudis Avancßats (ICREA), Barcelona, Spain 4 Division of Biomedical Sciences, University of California Riverside School of Medicine, Riverside, CA, USA Correspondence One of the key challenges of cancer biology is to catalogue and understand E. Porta-Pardo, Barcelona Supercomputing the somatic genomic alterations leading to cancer. Although alternative defi- Center (BSC), Jordi Girona, 29, 08034 nitions and search methods have been developed to identify cancer driver Barcelona, Spain genes and mutations, analyses of thousands of cancer genomes return a E-mail: eduard.porta@bsc.es or remarkably similar catalogue of around 300 genes that are mutated in at A. Godzik, University of California Riverside least one cancer type. Yet, many features of these genes and their role in can- School of Medicine, 900 University Ave. cer remain unclear, first and foremost when a somatic mutation is truly onco- Riverside, CA 92521, USA genic. In this review, we first summarize some of the recent efforts in E-mail: adam.godzik@medsch.ucr.edu completing the catalogue of cancer driver genes. Then, we give an overview of different aspects that influence the oncogenicity of somatic mutations in the (Received 3 November 2019, revised 23 core cancer driver genes, including their interactions with the germline gen- January 2020, accepted 9 February 2020, available online 28 April 2020) ome, other cancer driver mutations, the immune system, or their potential role in healthy tissues. In the coming years, this research holds promise to doi:10.1002/1873-3468.13781 illuminate how, when, and why cancer driver genes and mutations are really drivers, and thereby move personalized cancer medicine and targeted thera- pies forward. Keywords: cancer drivers; cancer genes; multiscale analysis; personalized medicine; variants of unknown significance The analysis of the first cancer genomes revealed that Genes in this first category are called driver genes, each tumor had acquired hundreds or even thousands those in the latter are named passengers and the same of somatic mutations during its evolution. While at the nomenclature can be used for both individual muta- time there was already a catalogue of genes known to tions and other genetic events. Interestingly, despite be involved in cancer, whole-exome and later whole- the apparent simplicity of this concept, the exact defi- genome sequencing of tumor samples provided the nition of cancer drivers is still debated, as best evi- opportunity to identify cancer genes in an unbiased denced by hundreds of papers offering different and data-driven way. To that end, dozens of computa- practical implementation of algorithms identifying tional biology and bioinformatics groups started devel- them. Most of them are based on the idea that drivers oping tools to analyze these large datasets and should show evidence of positive selection, which can distinguish the genes that contribute to tumor progres- be defined by a statistically significant difference sion from those that are instead neutral. between an observed number of mutations and those Abbreviations CGC, Cancer Gene Census; CRISPR, clustered regularly interspaced short palindromic repeats; GTEx, Genotype–Tissue Expression; ICGC, International Cancer Genome Consortiu; TCGA, The Cancer Genome Atlas; TME, tumor microenvironment; VUS, variant of unknown signifi- cance. FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4233 Federation of European Biochemical Societies This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Understanding cancer driver genes and mutations E. Porta-Pardo et al. The most common cancer driver Box 1. Take-home messages genes have been identified • We are nearing an almost-complete catalogue of Since the creation of the first Cancer Gene Census cancer driver genes (CGC) [1], there have been several major efforts to • The main drivers were discovered decades ago, compile a comprehensive catalogue of cancer driver but we still do not understand many aspects of genes. Most of the recent analyses have exploited data their biology from The Cancer Genome Atlas [2] (TCGA) or the • Cancer driver genes have many variants of International Cancer Genome Consortium [3] (ICGC) unknown significance and the integration of several computational tools to • The germline genome interacts with the somatic identify cancer driver genes [4,5]. Others, like the variants aforementioned CGC, have relied on manual curation • Cancer driver genes interact also with each other of the literature [6]. Over the past 15 years, there have • The oncogenicity of cancer driver genes and been dozens of studies aimed at completing the cata- mutations depends on the tissue and overall con- logue of cancer driver genes [7–10] and, as a result of text (e.g., germline mutations, immune status of these efforts, thousands of genes have been suggested the individual) to drive cancer growth. To evaluate whether there is a consensus on which genes are true drivers and how much we have learned expected by chance. But the background mutation rate during the genomic era of cancer, we have compared and its distribution are not known, and different algo- four of the most cited lists of cancer driver genes that rithms use different assumptions to estimate it. spanned different time points across the last seven years More than a decade and tens of thousands of cancer [4-5,7,10], as well as the first [1] (2004) and the current genomes later, thousands of genes, at some point, have [6] (2019) versions of the CGC (Fig. 1). Of note, the been defined as potential cancer drivers using different genes linked to cancer only by means of germline muta- algorithms. Nevertheless, there is a list of around 300 tions or somatic translocations were excluded from both genes that is consistently identified in almost all analy- CGC lists, as these are not analyzed by most cancer dri- ses: This core list consists of the most important can- ver detection tools. These lists together contain 741 cer driver genes and is unlikely to change in the genes, and there is a set of 280 genes common to two or future. Encouragingly, many of these genes were first more lists. The original CGC contained 94 genes (after identified decades ago by molecular biologists and now the filtering mentioned above). Of these, 26 have been are being ‘rediscovered’ by unsupervised analyses. So, consistently found in all subsequent studies, including while we have not yet identified the precise catalogue ‘classical’ cancer driver genes such as TP53, KRAS, of cancer driver genes or events, nor do we even agree NRAS, HRAS, EGFR, and BRAF. The remaining 68 on their definition, there seems to be a broad consen- are divided between those found at least once in the fol- sus about a ‘core’ group. lowing 15 years (32 genes), and those that were never Besides lacking a ‘final list’ of cancer driver genes, re-identified as somatic drivers (36). Although one might we also do not understand many of the cancer-relevant think that this last group of genes represents false posi- features of these genes. Arguably, one of the most tives, it includes genes with known germline roles in can- important open questions is when a somatic alteration cer such as FANCCA, FANCD2, FANCF, XPC, in a cancer driver gene is truly oncogenic, as personal- ERCC3, and ERCC5. ized cancer care often hinges on its answer. Here, we There are 48 genes that were not part of the original will review the recent efforts that address this question CGC but have been found in all following studies and across multiple biological scales. We will first focus on are now included in the CGC. Among them are some of how different mutations within the same cancer driver the most important discoveries from the first cancer geno- gene might have different effects. Then, we broaden mics era, for instance, B2M, STAG2, IDH1, IDH2, the scope and summarize recent results supporting the ARID1A, SPOP, KDM6A, RHOA, CASP8, or PIK3R1, existence of functional interactions between somatic as well as genes that were initially linked to cancer only mutations in cancer driver genes and other genetic via translocations and are now known to be altered by alterations, either somatic or germline. Finally, we give somatic single nucleotide variants, such as EP300. an overview of the evidence gathered so far about the Notably, the number of unique genes found in each role of the tissue context in determining the oncogenic- list has been shrinking over the years, from 122 unique ity of cancer driver mutations. genes (Tamborero et al. [5], 2013) to 54 and 56 genes 4234 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies E. Porta-Pardo et al. Understanding cancer driver genes and mutations Fig. 1. The quest for new cancer driver genes is approaching its end. (A) Upset plot showing the overlap of six different sets of cancer driver genes published during the last 15 years. (B) Barplot showing the fraction of cancer driver genes that is either unique to each set (orange) or found in at least another study (gray). (C) Barplot showing the number of high-confidence driver genes (i.e., those found at least twice) was found for the first time in the analyzed dataset (Martincorena et al. [7] and Bailey et al. [4], 2018), role [11]. Oncogenes usually have clear hotspots that are suggesting that the number of false positives is strongly enriched in somatic activating missense muta- decreasing over time and that the identification of new tions (e.g., KRAS G12, PIK3CA E545, BRAF V600). cancer driver genes is plateauing (Fig. 1B). In fact, On the other hand, tumor suppressor genes tend to be most of the cancer driver genes found in two studies affected by frameshift or truncating mutations that were discovered in the first TCGA analyses (Tam- completely abrogate the function of the encoded pro- borero et al. [5] and Lawrence et al. [10]; Fig. 1C). tein. Tumor suppressor genes can also have somatic Thus, it seems likely that the most common cancer dri- mutation hotspots that inactivate their function, but ver genes have already been discovered. However, as these are rarer and tend to affect genes that can be both we will see in the following sections, this does not oncogenes and tumor suppressors, depending on the mean that we understand their role in oncogenesis. context. Hence, it is easy to know whether a mutation is oncogenic, as identifying frameshift and truncating mutations is relatively straightforward and there are cat- Variants of unknown significance in alogues of which missense mutations in a given hotspot cancer driver genes have oncogenic effects [12]. The type and distribution of somatic mutations within Nevertheless, there are many cases where a tumor car- cancer driver genes strongly depends on their oncogenic ries a variant of unknown significance (VUS) in a cancer FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4235 Federation of European Biochemical Societies Understanding cancer driver genes and mutations E. Porta-Pardo et al. driver gene. These are often missense mutations located distinction between disease-associated (Group I) and in tumor suppressor genes or outside the known muta- oncogenic mutations (Group II) seems important, as tional hotspots in oncogenes. To put this in perspective, the performance of each group of methods in separat- patients from TCGA have a total of 44 607 somatic ing passenger and driver mutations is different [4]. mutations in cancer driver genes. Only 5435 of these are Group III includes those methods that predict cancer in OncoKB [12], leaving the oncogenicity of the remain- driver mutations using data from three-dimensional ing 39 172 (88%) unknown. Even if we assumed that all protein structures. These methods seem to be more frameshift and truncating mutations in cancer driver accurate than those that use only sequence data [4], genes are oncogenic, there would remain 28 238 mis- but they can only be applied to mutations where the sense mutations of unknown significance (63% of all structure is experimentally determined or can be rea- somatic mutations in these genes; Fig. 2A). sonably modeled. Finally, there is a fourth group of There are two main approaches to analyze the role methods (Group IV) that combine linear and three-di- of these variants of unknown significance: experimen- mensional features using machine-learning approaches. tal and computational. Experimental methods are Whenever possible, it is important to couple compu- more time consuming, but recent advances in satura- tational predictions with experimental data. For exam- tion mutagenesis, CRISPR technology and automation ple, most EGFR mutations in brain tumors of cell culture make the high-throughput analysis of (glioblastoma and lower grade glioma) are located thousands of mutations more accessible to researchers. near its dimerization interface (Fig. 2C). However, we In fact, a subset of cancer driver genes has been ana- only have experimental annotations for a small subset lyzed using deep mutational scans that test virtually all of all of these mutations. Putting side by side the potential mutations in a certain gene. This has been experimental results and the computational predictions done, for example, for TP53 [13,14], BRCA1 [15], (Fig. 2D), a reasonable agreement between the two, HRAS [16], PTEN [17], and MAPK1 [18]. There are albeit with some discrepancies, comes to light. also other analyses that, while not comprehensively studying individual proteins, have reported the onco- Historical contingency and cancer genicity of thousands of somatic mutations in dozens driver genes of different genes [19,20]. Computational methods have also been extensively The paths that life can follow are constrained by previ- explored. Their main advantages are that they are ous events, including seemingly inconsequential genetic orders of magnitude faster and less expensive than variations. This phenomenon, also known as historical experimental methods, allowing researchers to study vir- contingency [21], has implications in tumor evolution, tually any mutation. For example, using 12 different as a mutation might be beneficial in a certain genetic computational tools, we predicted the role of all mis- background and detrimental in another. Similarly, a sense mutations in the cancer driver genes from TCGA tumor might only be able to access certain genotypes, [4]. These predictions had a large agreement with if it has previously acquired other mutations. As we OncoKB [12] annotations (Fig. 2B), with the advantage will see in the following paragraphs, cancer cells are that they gave information on 28 238 missense somatic also subject to historical contingency: The evolutionary mutations not annotated in OncoKB. Importantly, 4864 paths that a tumor can explore depend on the genetic missense mutations in cancer driver genes from TCGA variations it has acquired over time [22]. with no data in OncoKB are predicted to be oncogenic The genetic background of a cancer cell includes (Fig. 2B). both the somatic variants it has acquired over time According to the type of data employed, there are and the germline variants that, by definition, were pre- four different groups of computational methods to sent before any somatic variant ever occurred. For predict the effects of VUS (Table 1). Group I consists example, each individual carries between 20 000 and of methods that use sequence information to distin- 30 000 coding germline variants, some of which even guish between benign and disease-associated muta- completely disrupt entire proteins [23]. Moreover, each tions. These tools have not been designed specifically individual also has hundreds or thousands of germline for cancer but, instead, to separate mutations associ- noncoding variants that influence gene expression, ated with rare diseases, diabetes, asthma, and cancer, including cancer drivers [24]. Finally, once somatic among others, from those that are benign. Methods in evolution begins, it can add hundreds of coding Group II also use sequence information but have been somatic variants and thousands of noncoding ones, trained specifically to distinguish between passenger and the order in which some of them are acquired will and driver mutations using cancer-specific data. The determine the final phenotype of the cancer cell. 4236 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies E. Porta-Pardo et al. Understanding cancer driver genes and mutations Fig. 2. Predicting the oncogenicity of somatic mutations. (A) Number of missense somatic mutations in cancer driver genes in TCGA, according to their oncogenicity annotation in OncoKB. (B) Computational prediction of the oncogenicity of all somatic missense mutations in cancer driver genes found in TCGA. Each column represents an OncoKB category. (C) Subset of somatic missense mutations in the dimerization interface of EGFR found in glioblastoma and lower grade glioma patients from The Cancer Genome Atlas. Mutations are colored according to their OncoKB annotations. (D) A consensus classification of some somatic mutations in EGFR, including all those from panel a. Each tile is colored according to the classification of the corresponding mutation as annotated in OncoKB (bottom), a computational analysis (middle) and a potential consensus between the two (top) FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4237 Federation of European Biochemical Societies Understanding cancer driver genes and mutations E. Porta-Pardo et al. Table 1. List of driver prediction algorithms and their classification rs25673 are five times more likely to have a PTEN (see text for details) somatic mutation in their tumors. The likely reason is Method Group References that these individuals have an intrinsic higher expres- sion level of STK11 and/or GNA11. When adding SIFT I [79] information at the pathway or protein interaction net- PolyPhen-2 I [80] work, the possible connection between these two genes MutAssessor I [81] transFIC I (Ensembl) [82] becomes apparent, as they are both upstream of CADD I (Ensembl) [83] PTEN, so their higher expression could make a MCAP I [84] somatic PTEN mutation more oncogenic than it would REVEL I (Ensembl) [85] be in a different genetic background [28]. These results VEST I [86] highlight the importance of accounting for protein FATHMM II [87] interactions and signaling pathways, already routinely CanDrA II [88] used by many approaches that analyze either germline CHASM II [89] ParsSNP II [90] [29-33] or somatic [34-39] variants alone, when inte- HotMAPS III [91] grating both. HotSpot3D III [92] Interactions between the germline and somatic gen- 3DHotspots.org III [93] omes could also have consequences for genetic risk e-Driver3D III [94] prediction. For example, 25 germline SNPs associated CATH-FunFams III [95] with glioma and glioblastoma have been recently CHASMplus IV [96] tested for their association with the most frequent somatic alterations in these cancer types: IDH1 R132H and 1p/19q deletions [40]. Based on this analy- sis, the authors were capable of building a polygenic The germline genome can interact with cancer driver risk score that predicted not only risk to glioma but, mutations both in cis and in trans (see [25] for an in- specifically, to IDH1-driven glioma. Given the signifi- depth review of the topic). Cis interactions are those cant biological differences between IDH1-mutated and that happen between variants of the same locus, and IDH1 wild-type brain tumors, whether this can be such functional interactions have been described for a extended to other combinations of cancer types and few cancer driver genes. One of the first examples of somatic driver events remains to be seen. Nevertheless, germline–somatic cis interactions was described for the these are significant first steps toward a comprehensive JAK2 somatic mutation V617F. This mutation, which understanding of the interactions between the germline transforms JAK2 into driver of myeloproliferative neo- genome of cancer patients and the somatic mutations plasms, is much more likely to happen in the haplo- acquired by their tumors. type with the minor allele of rs12343867 [26]. Similar Sex of the patient and their ancestry also correlate interactions have been described for somatic EGFR with the type and outcomes of many cancers, high- exon 19 deletion, which is three times more likely in lighting the importance of historical contingency and individuals with the minor rs712829 allele, located in germline–somatic interactions in tumor evolution. The the gene promoter [27]. Finally, it is worth noting that prevalence of many cancer types differs between males deep mutational scans could help discover many cis and females: Thyroid cancer is three times more likely interactions between germline and somatic variants. to occur in women than in men, whereas bladder can- This has been recently shown in TP53, where the effect cer is twice more likely in men than in women, for of dozens of somatic missense mutations depends on example. While this could be attributed to differences the allele of the germline ultra-rare rs35163653 in the environment of each gender, such as prevalence (MAF < 1e-5, p.V217M) [14]. of smoking or differences in hormone levels, multiple Cancer driver mutations can also interact in trans lines of evidence also point to genetics [41]. For exam- with germline variants. This phenomenon has been ple, the frequency of certain somatic driver mutations recently explored using data from TCGA [28], identify- depends on the sex [42]. Also, the sex chromosome X ing 28 germline variants associated with changes in the contains multiple oncogenes and tumor suppressors frequency of 20 somatic variants, suggesting an inter- that can contribute to sex bias and other cancer phe- action between the two. One of the better character- notypes by escaping X-inactivation in females [43,44]. ized interactions in that study is that between the Similarly, the genetic ancestry of an individual also germline variant rs25673 and somatic PTEN muta- correlates with the prevalence of cancer driver muta- tions. Individuals with the minor germline allele at tions. For example, somatic mutations in TP53 and 4238 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies E. Porta-Pardo et al. Understanding cancer driver genes and mutations CCNE1 are more common in cancer from African presence of certain driver mutations correlates with the Americans than in those from Europeans, whereas the quantity and composition of immune cells in the opposite is true for somatic variants in PI3KCA [45]. tumor microenvironment (TME) [51]. Finally, the order in which somatic mutations Regarding the influence of the immune system in the occurred can also influence their phenotype. One of presence of cancer driver mutations, it is mostly medi- the first examples of this phenomenon was described ated by the fact that all somatic mutations can create in a model of colorectal cancer, where tumors only neoantigens: peptides that have not been previously develop when somatic mutations are acquired in a pre- presented to immune cells via HLA and that, there- cise order [46]. Similarly, renal tumors seem to be con- fore, can be identified as foreign by the immune sys- strained to only few evolutionary pathways [47]. tem. If presented in the appropriate context, these Which one of these pathways is taken by the tumor neoantigens can trigger an immune response that ends seems to be determined by the initial somatic driver in the elimination of the cell that carries them, a pro- event. Recently, using TCGA data, this has been sys- cess known as immunoediting. As any other somatic tematically studied in dozens of different cancer types. mutation, those located in cancer driver genes are not The TCGA analysis provided indirect evidence of exempt from immunoediting. In fact, driver somatic somatic historical contingency, as somatic mutations seem to have been selected to be poorly presented in can either be clonal (i.e., they are acquired in the pri- the majority of both, class I [50] and class II HLA mary neoplasm and are thus present in all tumor cells) alleles [52]. At the individual patient level, a common or subclonal (i.e., they are acquired after the tumor immune-evading mechanism of cancer cells is the loss started its expansion and are only present in a subset of expression of HLA alleles that can present their dri- of cells) [22]. An even more dramatic example is seen ver mutations [53]. In fact, the effect of immunoediting in myeloproliferative neoplasms. There are two key is so strong that it can be seen at the population level: driver genes that, when mutated somatically in a mye- The frequency of a cancer driver mutation is nega- loid progenitor, they can potentially become malig- tively correlated with the frequency of the HLA alleles nant: JAK2 and TET2. However, the final phenotype that present the peptides derived from it [52]. of the patient depends on the order in which these However, as explained above, the presence of certain mutations are acquired. If a mutation in JAK2 is cancer driver mutations correlates with differences in acquired before the TET2 mutation, there is an expan- the quantity and composition of the immune infiltrate sion of hematopoietic stem and progenitor cells as well in the tumor microenvironment [51,54]. Whether these as a blockage of the expansion of erythroid progeni- correlations are causal or not remains to be seen in tors. On the other hand, if the order is inverted, there most cases, but some molecular mechanisms have been is an expansion of megakaryocytes and blockage of proposed for a few cases. For example, somatic muta- the hematopoietic cell pool [48]. tions in driver genes with known roles in immune sig- Overall, it seems clear that the evolutionary trajecto- naling, such as CASP8 or HLA, are generally ries of cancer cells are constrained by the genetic vari- associated with higher levels of immune cells in the ants already present in their genomes, regardless of TME, likely because these mutations are, indeed, an their somatic or germline origin. Understanding and immune-evading mechanism. In other cases, however, predicting such constraints could have significant the connection can be more obscure, as in the case of impact in both, the diagnosis (as seen for the polygenic colorectal tumors with KRAS mutations. These risk scores for IDH1 mutations) as well as the treat- tumors are known to have low levels of immune infil- ment of cancer [49]. trate and be resistant to immune-checkpoint blockade. These phenotypes could be due to KRAS repressing the interferon regulatory factor 2 (IRF2), leading to The relationship between the immune high CXCL3 expression and the recruitment of mye- system and cancer driver genes loid-derived suppressor cells to the tumor microenvi- Following the explosion of immune-based therapies to ronment [55]. Another group of cancer driver treat cancer, we are now also improving our under- mutations with a likely mechanism to link them with standing of the complex relationship between the changes to the immune infiltrate of the TME are those immune system and somatic cancer driver mutations. in the Wnt/beta-catenin pathway. Tumors with muta- The relationship between the two seems to be bidirec- tions in this pathway, particularly in CTNNB1, have tional, as the immune system has a strong effect in low levels of immune cells across multiple cancer determining which cancer driver mutations can happen types, likely through the exclusion of BATF3-derived in a cancer patient [50] while, at the same time, the dendritic cells from the TME [56]. Overall, however, FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4239 Federation of European Biochemical Societies Understanding cancer driver genes and mutations E. Porta-Pardo et al. the relationship between somatic driver mutations and also on the mutational signature active in that cancer the immune response against cancer cells will likely be type [13]. Another possibility is that the molecular pro- an important topic in the coming years. cesses altered by different mutations within the same gene can have varying tissue-specific degrees of onco- genicity, as could be the case for PIK3CA mutations Interactions between the tissue of [61,62] (Fig. 4). origin of the tumor and cancer driver All of the above is likely to have a significant impact genes also on personalized cancer care. For example, germ- The cell of origin of the tumor also influences the line mutations in BRCA1 and BRCA2 predispose to oncogenic potential of cancer driver mutations. This is multiple cancer types, specifically to ovarian and evident, for example, in the differences in the preva- breast cancer in women and prostate cancer in men. lence of a given mutation across different cancer types Using a synthetic lethality screen, Jonsson et al. dis- (Fig. 3). Out of the 299 cancer driver genes recently covered that breast cancer cells with mutations in these described in the Pan-Cancer Atlas analysis of TCGA, two genes are sensitive to PARP inhibitors [63]. Since only TP53 has a median somatic mutation frequency BRCA mutations are relatively common in many other over 10% across all cancer types (35%) and only ten cancer types, it was hoped that the synthetic lethality other genes have a median frequency above 1% interaction between PARP and BRCA would also (ARID1A, ATM, BRAF, KMT2C, KRAS, NF1, extend to these other cancer types. Nevertheless, it PIK3CA, PTEN, RB1, and SMARCA4). The remain- seems that the lethal interaction only happens in speci- ing 288 cancer driver genes have a median mutation fic cell lineages, specifically the same ones where germ- frequency below 1%. Moreover, the mutation fre- line BRCA1 and BRCA2 mutations predispose to quency of each cancer driver gene is highly variable. cancer. This highlights the importance of tissue speci- For example, BRAF has a frequency above 50% in ficity, not only to understand oncogenesis [64,65], but melanoma and thyroid adenocarcinoma but below also in determining the success of targeted therapies 10% in all other cancer types (Fig. 3). Something simi- [66]. lar happens with EGFR, with relatively high mutation frequencies in glioblastoma (24%), lung adenocarci- Healthy cells can carry driver noma (7%) and glioma (6%), but below 1% in the mutations remaining 30 cancer types. Overall, there are 43 cancer driver genes that have a mutation frequency above One of the most paradoxical and surprising results 10% in at least one cancer type, but whose median fre- about cancer driver genes is the discovery of healthy quency is below 1%. cells with somatic driver mutations. This was first Moreover, even if somatic driver mutations are shown in skin cells carrying the BRAF V600E muta- shared across cancer types, their role and interactions tion [67], but has been later extended to cells from the can differ depending on the tissue. This is the case of esophagus with NOTCH1 truncating mutations [68], BRAF V600E, which is present in melanoma and col- with more recent studies extending the work to healthy orectal adenocarcinoma patients. Yet, these two tumor colon [69], the colon of patients with inflammatory types differ in their sensitivity to the BRAF inhibitor bowel disease [70], or the endometrium [71]. In fact, vemurafenib. Melanoma patients initially respond very two analyses have studied somatic mutations in the well to the treatment [57], but colorectal cancer entire human body [72,73]. The authors identified patients do not [58]. Similarly, some driver mutations somatic mutations from RNAseq coming from 29 dif- seem to cooperate in some cancer types but are mutu- ferent tissues of over 500 healthy donors that were ally exclusive in others. This is the case, for example, part of the GTEx project (https://www.gtexportal.org/ of KRAS and TP53, which co-occur in pancreatic ade- home/). Virtually all tissues seemed to carry cancer dri- nocarcinoma but are mutually exclusive in lung adeno- ver somatic mutations in some individuals, even if carcinoma [59]. none of them had been diagnosed with cancer. The Cancer driver genes can also show different muta- most extreme example of this phenomenon is probably tional patterns depending on the cancer type [60]. the recently described role of somatic PTEN, KMT2D, These differences could be caused by the distinct muta- and ARID1A mutations in healthy liver [74]. These tional processes active in each cancer type. This has genes are known cancer drivers, but recently Zhu et al. been shown in TP53, where the prevalence of the dif- showed that, under certain circumstances, somatic ferent missense mutations in different cancer types mutations in these genes are actually beneficial to the depends, not only on the effect of the mutation, but homeostasis of the liver [74]. Liver cells that have 4240 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies E. Porta-Pardo et al. Understanding cancer driver genes and mutations Fig. 3. Cancer driver genes are tissue-specific. Each boxplot in the x-axis represents the distribution of mutation frequencies for a cancer driver gene across the 33 cancer types of TCGA. Out of the ten most frequently mutated cancer driver genes (average across all tissues) are highlighted in orange. Only TP53 has an average mutation frequency above 10% Fig. 4. Mutation-hotspot prevalence of PIK3CA depends on cancer type. (A) The mutation frequency of different hotspots (E545, in blue, H1047, in red, and the N-terminal domain, in yellow) differs depending on the cancer type (left). (B) Location of the different hotspots in the PIK3CA–PIK3R1 dimer (in white and green, respectively) structure from PDB file 3HMM FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4241 Federation of European Biochemical Societies Understanding cancer driver genes and mutations E. Porta-Pardo et al. somatic mutations in PTEN, KMT2D or ARID1A Here, we have reviewed the features that seem to have higher fitness. When the liver is under stress and influence the oncogenic role of cancer driver muta- needs to be regenerated, these cells can expand faster tions. First, we have shown that cancer driver genes than their nonmutated counterparts and, thus, regener- have many variants of unknown significance, many of ate the tissue in less time. them potentially benign from the clinical point of view. Overall, it seems that somatic cancer driver muta- However, although new experimental methods, such as tions are pervasive in healthy organs. But, in that case, deep mutational scans, can give us insights into the how is it possible that all of us have thousands of cells oncogenic potential of virtually all mutations in a can- with oncogenic mutations and not develop cancer? The cer driver gene, computational tools are still the only most accepted theory to explain this is that a cell practical alternative in most cases. requires multiple somatic insults before becoming Then, we have reviewed the recent evidence about malignant. This agrees with observations from prema- the role of historical contingency and interactions with lignant stages of certain tumors, where cells already the germline genome in determining the oncogenicity have some driver mutations, but it is not until they of cancer driver mutations. The same somatic muta- reach a minimum threshold, or certain specific driver tion in a cancer driver gene might have different mutations that they actually become malignant [75]. effects depending on which other genetic variants are This is the case, for example, of age-related clonal already present in the cell. This includes both inherited hematopoiesis, which is a natural phenomenon in germline variants, as well as other somatic variants which the pool of hematopoietic stem cells becomes that the (pre)cancerous cell has acquired over time. dominated by a few clones as individuals age. When Moreover, the genetic background of the patient, such clonal expansion is accompanied by somatic namely the pre-existing germline variants, is also likely mutations in driver genes, it can eventually cause acute to affect the oncogenicity of the somatic mutations myeloid leukemia (AML). However, not all driver that happen later in life [78], as evidenced by the dif- mutations carry the same risk to cause AML: While ferences in somatic mutation patterns in individuals TP53 and U2AF1 significantly increase the risk of with different sex or ancestry. While we already have AML, mutations in DNMT3A or TET2 seem to lead numerous examples of such phenomena, we are only to less aggressive cell phenotypes [76]. Moreover, hav- beginning to grasp its importance. ing two or more of these mutations increases the risk We have also discussed the importance of the tissue proportionately [76]. Along the same line, most tumors where driver mutations arise. All cancer driver genes, from adult patients harbor between 5 and 10 cancer with the exception of the omnipresent TP53, are fre- driver mutations irrespectively of their overall muta- quently mutated only in a single or few tissues. More- tion rate [77], suggesting that many tumors need a over, as shown for PIK3CA and EGFR, the mutation minimum number of driver mutations before becoming patterns within a gene can also change depending on the oncogenic. However, another interesting alternative is cancer type. This, together with evidence that the same that the germline genome could modulate the onco- driver mutation in different tissues might lead to very genic potential of somatic mutations. As we have different phenotypes (such as drug sensitivity as in the shown before, there is evidence of interactions between case of BRCA1 and BRCA2), highlights the tissue of somatic and germline variants, so it is possible that origin of somatic mutations must be taken into account driver mutations are only oncogenic when they happen in order to properly assess their oncogenic roles. in the right germline genetic background. Finally, as is Another important question that we will need to oftentimes the case, all of these mechanisms are not address in the coming years is the bidirectional rela- mutually exclusive but, in fact, are likely interacting tionship between the immune system and cancer driver with each other. mutations. Understanding this relationship can be key to find, among others, new drug combinations that extend the scope of immune-based therapies. Conclusions and Perspectives Finally, we have also discussed the growing evidence As we near the end of the beginning of cancer geno- showing that somatic mutations, including those linked mics, new questions emerge around the role of cancer to cancer, seem to be pervasive throughout the body of driver genes and their associated somatic mutations. healthy individuals. This can potentially be explained if One of the most pressing questions that we need to each cancer cell would require a minimum amount of answer is, probably, which mutations are truly onco- driver mutations to become tumorigenic. However, the genic and which are not, as many aspects of personal- sheer number of cells that seem to carry potentially ized cancer care hang from it. oncogenic mutations, together with the surprising 4242 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies E. Porta-Pardo et al. Understanding cancer driver genes and mutations results showing the regenerative role of PTEN, and Lopez-Bigas N (2013) IntOGen-mutations identifies ARID1A, and KMT2D somatic mutations in healthy cancer drivers across tumor types. Nat Methods 10, liver, suggest that other phenomena are likely interven- 1081–1084. ing in the process. 9 McGranahan N, Favero F, de Bruin EC, Birkbak NJ, In conclusion, while we have probably already iden- Szallasi Z and Swanton C (2015) Clonal status of tified the core cancer driver genes, in the coming years actionable driver events and the timing of mutational addressing all of these questions will help understand processes in cancer evolution. Sci Transl Med 7, how, when, and why cancer driver genes and muta- 283ra54. 10 Lawrence MS, Stojanov P, Mermel CH, Robinson JT, tions are really drivers. Garraway LA, Golub TR, Meyerson M, Gabriel SB, Lander ES and Getz G (2014) Discovery and saturation Acknowledgements analysis of cancer genes across 21 tumour types. Nature 505, 495–501. We would like to thank all the patients and scientists 11 Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, involved in The Cancer Genome Atlas. E.P-P received Diaz LA and Kinzler KW (2013) Cancer genome support from a Beatriu de Pinos fellowship (LCF/BQ/ landscapes. Science (80-.) 340, 1546–1558. PI18/11630003) from AGAUR and a La Caixa Junior 12 Chakravarty D, Gao J, Phillips S, Kundra R, Zhang H, Leader Fellowship from Fundacio Bancaria La Caixa. Wang J, Rudolph JE, Yaeger R, Soumerai T, Nissan A.V received support from Instituci o Catalana de MH et al. (2017) OncoKB: a precision oncology Recerca Avancßada (ICREA). AG received support knowledge base. JCO Precis Oncol 2017, 1–16. from NIH R35GM118187. 13 Giacomelli AO, Yang X, Lintner RE, McFarland JM, Duby M, Kim J, Howard TP, Takeda DY, Ly SH, Kim E et al. (2018) Mutational processes shape the References landscape of TP53 mutations in human cancer. Nat 1 Futreal PA, Coin L, Marshall M, Down T, Hubbard T, Genet 50, 1381–1387. Wooster R, Rahman N and Stratton MR (2004) A ensus 14 Kotler E, Shani O, Goldfeld G, Lotan-Pompan M, of human cancer genes. Nat Rev Cancer 4, 177–183. Tarcic O, Gershoni A, Hopf TA, Marks DS, Oren M 2 Weinstein JN, Collisson EA, Mills GB, Shaw KRM, and Segal E (2018) A systematic p53 mutation library Ozenberger BA, Ellrott K, Sander C, Stuart JM, Chang links differential functional impact to cancer mutation K, Creighton CJ et al. (2013) The cancer genome atlas pattern and evolutionary conservation. Mol Cell 71, pan-cancer analysis project. Nat Genet 45, 1113–1120. 178–190.e8. 3 Hudson TJ, Anderson W, Aretz A, Barker AD, Bell C, 15 Starita LM, Young DL, Islam M, Kitzman JO, Bernabe RR, Bhan MK, Calvo F, Eerola I, Gerhard Gullingsrud J, Hause RJ, Fowler DM, Parvin JD, DS et al. (2010) International network of cancer Shendure J and Fields S (2015) Massively parallel genome projects. Nature 464, 993–998. functional analysis of BRCA1 RING domain variants. 4 Bailey MH, Tokheim C, Porta-Pardo E, Sengupta S, Genetics 200, 413–422. Bertrand D, Weerasinghe A, Colaprico A, Wendl MC, 16 Bandaru P, Shah NH, Bhattacharyya M, Barton JP, Kim J, Reardon B et al. (2018) Comprehensive Kondo Y, Cofsky JC, Gee CL, Chakraborty AK, characterization of cancer driver genes and mutations. Kortemme T, Ranganathan R et al. (2017) Cell 173, 371–385. Deconstruction of the ras switching cycle through 5 Tamborero D, Gonzalez-Perez A, Perez-Llamas C, saturation mutagenesis. Elife 6. Deu-Pons J, Kandoth C, Reimand J, Lawrence MS, 17 Mighell TL, Evans-Dutson S and O’Roak BJ (2018) A Getz G, Bader GD, Ding L et al. (2013) Comprehensive saturation mutagenesis approach to understanding identification of mutational cancer driver genes across PTEN lipid phosphatase activity and genotype- 12 tumor types. Sci Rep 3, 2650. phenotype relationships. Am J Hum Genet 102, 943–955. 6 Sondka Z, Bamford S, Cole CG, Ward SA, Dunham I 18 Brenan L, Andreev A, Cohen O, Pantel S, Kamburov and Forbes SA (2018) The COSMIC Cancer Gene A, Cacchiarelli D, Persky NS, Zhu C, Bagul M, Goetz Census: describing genetic dysfunction across all human EM et al. (2016) Phenotypic characterization of a cancers. Nat Rev Cancer 18, 696–705. comprehensive set of MAPK1/ERK2 missense mutants. 7 Martincorena I, Raine KM, Gerstung M, Dawson KJ, Cell Rep 17, 1171–1183. Haase K, Van Loo P, Davies H, Stratton MR and 19 Berger AH, Brooks AN, Wu X, Shrestha Y, Chouinard Campbell PJ (2017) Universal patterns of selection in C, Piccioni F, Bagul M, Kamburov A, Imielinski M, cancer and somatic tissues. Cell 171, 1029–1041.e21. Hogstrom L et al. (2016) High-throughput phenotyping 8 Gonzalez-Perez A, Perez-Llamas C, Deu-Pons J, of lung cancer somatic mutations. Cancer Cell 30, 214– Tamborero D, Schroeder MP, Jene-Sanz A, Santos A 228. FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4243 Federation of European Biochemical Societies Understanding cancer driver genes and mutations E. Porta-Pardo et al. 20 Ng PKS, Li J, Jeong KJ, Shao S, Chen H, Tsang YH, 33 Vanunu O, Magger O, Ruppin E, Shlomi T and Sharan Sengupta S, Wang Z, Bhavana VH, Tran R et al. R (2010) Associating genes and protein complexes with (2018) Systematic functional annotation of somatic disease via network propagation. PLoS Comput Biol 6, mutations in cancer. Cancer Cell 33, 450–462.e10. e1000641. 21 Blount ZD, Borland CZ and Lenski RE (2008) 34 Leiserson MDM, Vandin F, Wu HT, Dobson JR, Historical contingency and the evolution of a key Eldridge JV, Thomas JL, Papoutsaki A, Kim Y, Niu B, innovation in an experimental population of Escherichia McLellan M et al. (2015) Pan-cancer network analysis coli. Proc Natl Acad Sci USA 105, 7899–7906. identifies combinations of rare somatic mutations across 22 McGranahan N and Swanton C (2017) Clonal pathways and protein complexes. Nat Genet 47, 106–114. heterogeneity and tumor evolution: past, present, and 35 Hofree M, Shen JP, Carter H, Gross A and Ideker T the future. Cell 168, 613–628. (2013) Network-based stratification of tumor mutations. 23 Lek M, Karczewski KJ, Minikel EV, Samocha KE, Nat Methods 10, 1108–1118. Banks E, Fennell T, O’Donnell-Luria AH, Ware JS, 36 Bashashati A, Haffari G, Ding J, Ha G, Lui K, Rosner Hill AJ, Cummings BB et al. (2016) Analysis of J, Huntsman DG, Caldas C, Aparicio SA and Shah SP protein-coding genetic variation in 60,706 humans. (2012) DriverNet: uncovering the impact of somatic Nature 536, 285–291. driver mutations on transcriptional networks in cancer. 24 Li Q, Seo JH, Stranger B, McKenna A, Pe’Er I, Genome Biol 13, R124. Laframboise T, Brown M, Tyekucheva S and 37 Hou JP and Ma J (2014) DawnRank: discovering Freedman ML (2013) Integrative eQTL-based analyses personalized driver genes in cancer. Genome Med 6, 56. reveal the biology of breast cancer risk loci. Cell 152, 38 Jia P and Zhao Z (2014) VarWalker: personalized 633–641. mutation network analysis of putative cancer genes 25 Ramroop JR, Gerber MM and Toland AE (2019) from next-generation sequencing data. PLoS Comput Germline variants impact somatic events during Biol 10, e1003460. tumorigenesis. Trends Genet 35, 515–526. 39 Cho A, Shim JE, Kim E, Supek F, Lehner B and Lee I 26 Olcaydu D, Harutyunyan A, J€ager R, Berg T, Gisslinger (2016) MUFFINN: cancer gene discovery via network B, Pabinger I, Gisslinger H and Kralovics R (2009) A analysis of somatic mutation data. Genome Biol 17, 129. common JAK2 haplotype confers susceptibility to 40 Eckel-Passow JE, Decker PA, Kosel ML, Kollmeyer myeloproliferative neoplasms. Nat Genet 41, 450–454. TM, Molinaro AM, Rice T, Caron AA, Drucker KL, 27 Liu W, He L, Ramırez J, Krishnaswamy S, Kanteti R, Praska CE, Pekmezci M et al. (2019) Using germline Wang YC, Salgia R and Ratain MJ (2011) Functional variants to estimate glioma and subtype risks. Neuro EGFR germline polymorphisms may confer risk for Oncol 21, 451–461. EGFR somatic mutations in non-small cell lung cancer, 41 Yuan Y, Liu L, Chen H, Wang Y, Xu Y, Mao H, Li J, with a predominant effect on exon 19 microdeletions. Mills GB, Shu Y, Li L et al. (2016) Comprehensive Cancer Res 71, 2423–2427. characterization of molecular differences in cancer 28 Carter H, Marty R, Hofree M, Gross AM, Jensen J, between male and female patients. Cancer Cell 29, Fisch KM, Wu X, Deboever C, Van Nostrand EL, 711–722. Song Y et al. (2017) Interaction landscape of inherited 42 Li CH, Haider S, Shiah YJ, Thai K and Boutros PC polymorphisms with somatic events in cancer. Cancer (2018) Sex differences in cancer driver genes and Discov 7, 410–423. biomarkers. Cancer Res 78, 5527–5537. 29 Greene CS, Krishnan A, Wong AK, Ricciotti E, Zelaya 43 Dunford A, Weinstock DM, Savova V, Schumacher RA, Himmelstein DS, Zhang R, Hartmann BM, SE, Cleary JP, Yoda A, Sullivan TJ, Hess JM, Zaslavsky E, Sealfon SC et al. (2015) Understanding Gimelbrant AA, Beroukhim R et al. (2017) Tumor- multicellular function and disease with human suppressor genes that escape from X-inactivation tissuespecific networks. Nat Genet 47, 569–576. contribute to cancer sex bias. Nat Genet 49, 10–16. 30 Mostafavi S, Ray D, Warde-Farley D, Grouios C and 44 Vincent-Salomon A, Ganem-Elbaz C, Manie E, Raynal Morris Q (2008) GeneMANIA: A real-time multiple V, Sastre-Garau X, Stoppa-Lyonnet D, Stern MH and association network integration algorithm for predicting Heard E (2007) X inactive-specific transcript RNA gene function. Genome Biol 9, S4. coating and genetic instability of the X chromosome in 31 Jia P, Zheng S, Long J, Zheng W and Zhao Z (2011) BRCA1 breast tumors. Cancer Res 67, 5134–5140. dmGWAS: Dense module searching for genome-wide 45 Yuan J, Hu Z, Mahal BA, Zhao SD, Kensler KH, Pi J, association studies in protein-protein interaction Hu X, Zhang Y, Wang Y, Jiang J et al. (2018) networks. Bioinformatics 27, 95–102. Integrated analysis of genetic ancestry and genomic 32 K€ ohler S, Bauer S, Horn D and Robinson PN (2008) alterations across cancers. Cancer Cell 34, 549–560.e9. Walking the interactome for prioritization of candidate 46 Fearon ER and Vogelstein B (1990) A genetic model disease genes. Am J Hum Genet 82, 949–958. for colorectal tumorigenesis. Cell 61, 759–767. 4244 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies E. Porta-Pardo et al. Understanding cancer driver genes and mutations 47 Turajlic S, Xu H, Litchfield K, Rowan A, Horswell S, Huang KL, Tokheim C et al. (2018) Perspective on Chambers T, O’Brien T, Lopez JI, Watkins TBK, Nicol oncogenic processes at the end of the beginning of D et al. (2018) Deterministic evolutionary trajectories cancer genomics. Cell 173, 305–320.e10. influence primary tumor growth: TRACERx renal. Cell 60 Chang MT, Asthana S, Gao SP, Lee BH, Chapman JS, 173, 595–610.e11. Kandoth C, Gao JJ, Socci ND, Solit DB, Olshen AB 48 Ortmann CA, Kent DG, Nangalia J, Silber Y, Wedge DC, et al. (2016) Identifying recurrent mutations in cancer Grinfeld J, Baxter EJ, Massie CE, Papaemmanuil E, reveals widespread lineage diversity and mutational Menon S et al. (2015) Effect of mutation order on specificity. Nat Biotechnol 34, 155–163. myeloproliferative neoplasms. N Engl J Med 372, 601–612. 61 Burke JE, Perisic O, Masson GR, Vadas O and 49 Amirouchene-Angelozzi N, Swanton C and Bardelli A Williams RL (2012) Oncogenic mutations mimic and (2017) Tumor evolution as a therapeutic target. Cancer enhance dynamic events in the natural activation of Discov 7, 805–817. phosphoinositide 3-kinase p110a (PIK3CA). Proc Natl 50 Marty R, Kaabinejadian S, Rossell D, Slifker MJ, van Acad Sci USA 109, 15259–15264. de Haar J, Engin HB, de Prisco N, Ideker T, 62 Zhang Y, Kwok-Shing Ng P, Kucherlapati M, Chen F, Hildebrand WH, Font-Burgada J et al. (2017) MHC-I Liu Y, Tsang YH, de Velasco G, Jeong KJ, Akbani R, genotype restricts the oncogenic mutational landscape. Hadjipanayis A et al. (2017) A pan-cancer Cell 171, 1272–1283.e15. proteogenomic atlas of PI3K/AKT/mTOR pathway 51 Thorsson V, Gibbs DL, Brown SD, Wolf D, Bortone alterations. Cancer Cell 31, 820–832.e3. DS, Ou Yang TH, Porta-Pardo E, Gao GF, Plaisier 63 Jonsson P, Bandlamudi C, Cheng ML, Srinivasan P, CL, Eddy JA et al. (2018) The immune landscape of Chavan SS, Friedman ND, Rosen EY, Richards AL, cancer. Immunity 48, 812–830.e14. Bouvier N, Selcuklu SD et al. (2019) Tumour lineage 52 Marty R, Thompson WK, Salem RM, Zanetti M and shapes BRCA-mediated phenotypes. Nature 571, Carter H (2018) Evolutionary pressure against MHC 576–579. class II binding cancer mutations. Cell 175, 416– 64 Haigis KM, Cichowski K and Elledge SJ (2019) Tissue- 428.e13. specificity in cancer: the rule, not the exception. Science 53 McGranahan N, Rosenthal R, Hiley CT, Rowan AJ, (80-.) 363, 1150–1151. Watkins TBK, Wilson GA, Birkbak NJ, Veeriah S, 65 Schneider G, Schmidt-Supprian M, Rad R and Saur D Van Loo P, Herrero J et al. (2017) Allele-specific HLA (2017) Tissue-specific tumorigenesis: context matters. loss and immune escape in lung cancer evolution. Cell Nat Rev Cancer 17, 239–253. 171, 1259–1271.e11. 66 Cohen RL and Settleman J (2014) From cancer 54 Rooney MS, Shukla SA, Wu CJ, Getz G and Hacohen genomics to precision oncology - tissue’s still an issue. N (2015) Molecular and genetic properties of tumors Cell 157, 1509–1514. associated with local immune cytolytic activity. Cell 67 Martincorena I, Roshan A, Gerstung M, Ellis P, Van 160, 48–61. Loo P, McLaren S, Wedge DC, Fullam A, Alexandrov 55 Liao W, Overman MJ, Boutin AT, Shang X, Zhao D, LB, Tubio JM et al. (2015) High burden and pervasive Dey P, Li J, Wang G, Lan Z, Li J et al. (2019) positive selection of somatic mutations in normal KRASIRF2 axis drives immune suppression and human skin. Science (80-.) 348, 880–886. immune therapy resistance in colorectal cancer. Cancer 68 Martincorena I, Fowler JC, Wabik A, Lawson ARJ, Cell 35, 559–572.e7. Abascal F, Hall MWJ, Cagan A, Murai K, Mahbubani 56 Spranger S, Dai D, Horton B and Gajewski TF (2017) K, Stratton MR et al. (2018) Somatic mutant clones Tumor-residing Batf3 dendritic cells are required for colonize the human esophagus with age. Science (80-.) effector T cell trafficking and adoptive T cell therapy. 362, 911–917. Cancer Cell 31, 711–723.e4. 69 Lee-Six H, Olafsson S, Ellis P, Osborne RJ, Sanders 57 Flaherty KT, Puzanov I, Kim KB, Ribas A, McArthur MA, Moore L, Georgakopoulos N, Torrente F, GA, Sosman JA, O’Dwyer PJ, Lee RJ, Grippo JF, Noorani A, Goddard M et al. (2019) The landscape of Nolop K et al. (2010) Inhibition of mutated, activated somatic mutation in normal colorectal epithelial cells. BRAF in metastatic melanoma. N Engl J Med 363, Nature 574, 532–537. 809–819. 70 Olafsson S, McIntyre RE, Coorens T, Butler T, 58 Prahallad A, Sun C, Huang S, Di Nicolantonio F, Robinson P, Lee-Six H, Sanders MA, Arestang K, Salazar R, Zecchin D, Beijersbergen RL, Bardelli A Dawson C, Tripathi M et al. (2019) The mutational and Bernards R (2012) Unresponsiveness of colon profile and clonal landscape of the inflammatory bowel cancer to BRAF(V600E) inhibition through feedback disease affected colon. bioRxiv 832014 [PREPRINT]. activation of EGFR. Nature 483, 100–104. 71 Moore L, Leongamornlert D, Coorens T, Sanders M, 59 Ding L, Bailey MH, Porta-Pardo E, Thorsson V, Ellis P, Maura F, Dawson K, Brunner SF, Nangalia J, Colaprico A, Bertrand D, Gibbs DL, Weerasinghe A, Lee-Six Het al. (2019) Abstract 970: the mutational FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of 4245 Federation of European Biochemical Societies Understanding cancer driver genes and mutations E. Porta-Pardo et al. landscape of normal human endometrial epithelium. uncertain significance in clinical exomes at high pp.970–970. sensitivity. Nat Genet 48, 1581–1586. 72 Garcıa-Nieto PE, Morrison AJ and Fraser HB (2019) 85 Ioannidis NM, Rothstein JH, Pejaver V, Middha S, The somatic mutation landscape of the human body. McDonnell SK, Baheti S, Musolf A, Li Q, Holzinger E, Genome Biol 20, 298. Karyadi D et al. (2016) REVEL: an ensemble method 73 Yizhak K, Aguet F, Kim J, Hess JM, K€ ubler K, for predicting the pathogenicity of rare missense Grimsby J, Frazer R, Zhang H, Haradhvala NJ, variants. Am J Hum Genet 99, 877–885. Rosebrock D et al. (2019) RNA sequence analysis 86 Carter H, Douville C, Stenson PD, Cooper DN and reveals macroscopic somatic clonal expansion across Karchin R (2013) Identifying Mendelian disease genes normal tissues. Science (80-.) 364, eaaw0726. with the variant effect scoring tool. BMC Genom 14 74 Zhu M, Lu T, Jia Y, Luo X, Gopal P, Li L, Odewole (Suppl 3), S3. M, Renteria V, Singal AG, Jang Y et al. (2019) 87 Shihab HA, Gough J, Cooper DN, Day INM and Somatic mutations increase hepatic clonal fitness and Gaunt TR (2013) Predicting the functional regeneration in chronic liver disease. Cell 177, 608– consequences of cancer-associated amino acid 621.e12. substitutions. Bioinformatics 29, 1504–1510. 75 Curtius K, Wright NA and Graham TA (2017) 88 Mao Y, Chen H, Liang H, Meric-Bernstam F, Mills Evolution of premalignant disease. Cold Spring Harb GB and Chen K (2013) CanDrA: Cancer-specific driver Perspect Med 7, 19. missense mutation annotation with optimized features. 76 Abelson S, Collord G, Ng SWK, Weissbrod O, PLoS ONE 8, e77945. Mendelson Cohen N, Niemeyer E, Barda N, Zuzarte 89 Wong WC, Kim D, Carter H, Diekhans M, Ryan MC PC, Heisler L, Sundaravadanam Y et al. (2018) and Karchin R (2011) CHASM and SNVBox: toolkit Prediction of acute myeloid leukaemia risk in healthy for detecting biologically important single nucleotide individuals. Nature 559, 400–404. mutations in cancer. Bioinformatics 27, 2147–2148. 77 Sabarinathan R, Pich O, Martincorena I, Rubio-Perez 90 Kumar RD, Swamidass SJ and Bose R (2016) C, Juul M, Wala J, Schumacher S, Shapira O, Unsupervised detection of cancer driver mutations with Sidiropoulos N, Waszak S et al. (2017) The whole- parsimony-guided learning. Nat Genet 48, 1288–1295. genome panorama of cancer drivers. bioRxiv 91 Tokheim C, Bhattacharya R, Niknafs N, Gygax DM, [PREPRINT]. Kim R, Ryan M, Masica DL and Karchin R (2016) 78 Agarwal D, Nowak C, Zhang NR, Pusztai L and Exome-scale discovery of hotspot mutation regions in Hatzis C (2017) Functional germline variants as human cancer using 3D protein structure. Cancer Res potential co-oncogenes. NPJ Breast Cancer 3, 46. 76, 3719–3731. 79 Ng PC and Henikoff S (2003) SIFT: predicting amino 92 Niu B, Scott AD, Sengupta S, Bailey MH, Batra P, acid changes that affect protein function. Nucleic Acids Ning J, Wyczalkowski MA, Liang WW, Zhang Q, Res 31, 3812–3814. McLellan MD et al. (2016) Protein-structure-guided 80 Adzhubei IA, Schmidt S, Peshkin L, Ramensky VE, discovery of functional mutations across 19 cancer Gerasimova A, Bork P, Kondrashov AS and Sunyaev types. Nat Genet 48, 827–837. SR (2010) A method and server for predicting 93 Gao J, Chang MT, Johnsen HC, Gao SP, Sylvester BE, damaging missense mutations. Nat Methods 7, 248–249. Sumer SO, Zhang H, Solit DB, Taylor BS, Schultz N 81 Reva B, Antipin Y and Sander C (2011) Predicting the et al. (2017) 3D clusters of somatic mutations in cancer functional impact of protein mutations: application to reveal numerous rare mutations as functional targets. cancer genomics. Nucleic Acids Res 39, e118. Genome Med 9, 4. 82 Gonzalez-Perez A, Deu-Pons J and Lopez-Bigas N 94 Porta-Pardo E, Garcia-Alonso L, Hrabe T, Dopazo J (2012) Improving the prediction of the functional and Godzik A (2015) A pan-cancer catalogue of cancer impact of cancer mutations by baseline tolerance driver protein interaction interfaces. PLoS Comput Biol transformation. Genome Med 4, 89. 11, 1–18. 83 Kircher M, Witten DM, Jain P, O’roak BJ, Cooper 95 Ashford P, Pang CSM, Moya-Garcıa AA, Adeyelu T GM and Shendure J (2014) A general framework for and Orengo CA (2019) A CATH domain functional estimating the relative pathogenicity of human genetic family based approach to identify putative cancer driver variants. Nat Genet 46, 310–315. genes and driver mutations. Sci Rep 9, 263. 84 Jagadeesh KA, Wenger AM, Berger MJ, Guturu H, 96 Tokheim C and Karchin R (2019) CHASMplus reveals Stenson PD, Cooper DN, Bernstein JA and Bejerano G the scope of somatic missense mutations driving human (2016) M-CAP eliminates a majority of variants of cancers. Cell Syst 9, 9–23.e8. 4246 FEBS Letters 594 (2020) 4233–4246 ª 2020 The Authors. FEBS Letters published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies