Oncogenic Signaling Pathways in The Cancer Genome Atlas

Highlights

Alteration map of 10 signaling pathways across 9,125 samples from 33 cancer types

Reusable, curated pathway templates that include a catalogue of driver genes

57% of tumors have at least one potentially actionable alteration in these pathways

Co-occurrence of actionable alterations suggests combination therapy opportunities

Summary

Genetic alterations in signaling pathways that control cell-cycle progression, apoptosis, and cell growth are common hallmarks of cancer, but the extent, mechanisms, and co-occurrence of alterations in these pathways differ between individual tumors and tumor types. Using mutations, copy-number changes, mRNA expression, gene fusions and DNA methylation in 9,125 tumors profiled by The Cancer Genome Atlas (TCGA), we analyzed the mechanisms and patterns of somatic alterations in ten canonical pathways: cell cycle, Hippo, Myc, Notch, Nrf2, PI-3-Kinase/Akt, RTK-RAS, TGFβ signaling, p53 and β-catenin/Wnt. We charted the detailed landscape of pathway alterations in 33 cancer types, stratified into 64 subtypes, and identified patterns of co-occurrence and mutual exclusivity. Eighty-nine percent of tumors had at least one driver alteration in these pathways, and 57% percent of tumors had at least one alteration potentially targetable by currently available drugs. Thirty percent of tumors had multiple targetable alterations, indicating opportunities for combination therapy.

Introduction

Cancer is a disease in which cells have acquired the ability to divide and grow uncontrollably (Hanahan and Weinberg, 2000, Hanahan and Weinberg, 2011), usually through genetic alterations in specific genes. Advances in DNA sequencing over the past decade have made it possible to systematically study these genetic changes, and we now have a better understanding of the commonly involved processes and signaling pathways (Garraway and Lander, 2013, Vogelstein et al., 2013). As more genetic alterations become targetable by specific drugs, DNA sequencing is becoming part of routine clinical care (Hartmaier et al., 2017, Schram et al., 2017, Sholl et al., 2016, Zehir et al., 2017). However, there is considerable variation in the genes and pathways altered across different tumor types and individual tumor samples, and a complete understanding of the genes and pathways altered in all cancer types is essential to identify potential therapeutic options and vulnerabilities.

Several important signaling pathways have been identified as frequently genetically altered in cancer, including the RTK/RAS/MAP-Kinase (hereafter also called RTK-RAS for brevity) pathway, PI3K/Akt signaling, and others (Vogelstein and Kinzler, 2004). Members of these pathways and their interactions have been captured in a number of pathway databases, such as Pathway Commons (Cerami et al., 2011), which aggregates a number of databases, including REACTOME (Joshi-Tope et al., 2005) and KEGG (Kanehisa and Goto, 2000). Genes in key pathways are not altered at equal frequencies, with certain genes recurrently altered and well-known in cancer, while others are only rarely or never altered.

The detection of recurrence of rare alterations often requires large numbers of samples (Lawrence et al., 2014). This is confounded by the challenge to distinguish between functionally relevant (or “driver” alterations) and non-oncogenic “passenger” events (Gao et al., 2014), especially in tumor types with a high background mutation burden (Alexandrov et al., 2013, Lawrence et al., 2013). In these cases, many mutations, even when they occur in cancer genes, may have no functional effect. This topic is further addressed in Bailey et al. (Bailey et al., 2018).

Previous studies by The Cancer Genome Atlas (TCGA) have incrementally mapped out the alteration landscape in signaling pathways. Certain pathways, such as RTK-RAS signaling or the cell-cycle pathway, are altered at high frequencies across many different tumor types, whereas other pathways are altered in more specific subsets of malignancies (e.g., alterations in the oxidative stress response pathway are strongly associated with squamous histologies). With >10,000 samples characterized by TCGA, there is an opportunity to systematically characterize and define the alterations within well-known cancer pathways across all tumor types and map out commonalities and differences across pathways. The existence of shared genomic features across histologies has been highlighted by several studies (Ciriello et al., 2013, Hoadley et al., 2014, Hoadley et al., 2018), but these studies traditionally used a gene-centric, as opposed to pathway-centric, approach. Identifying relationships of inter- and intra-pathway recurrence, co-occurrence or mutual exclusivity across different types of cancers can help elucidate functionally relevant mechanisms of oncogenic pathway alterations that might inform treatment options.

Here, we worked within the framework of the TCGA PanCancer Atlas initiative (Cancer Genome Atlas Research Network et al., 2013c) to build a uniformly processed dataset and a unified data analysis pipeline aimed at exploring similarities and differences in canonical cancer pathway alterations across 33 cancer types. The focus of this effort is on mitogenic signaling pathways with vidence for functional alterations; other oncogenic processes, such as alterations in DNA repair (Knijnenburg et al., 2018), the spliceosome (Seiler et al., 2018), ubiquitination (Ge et al., 2018), or metabolic pathways (Peng et al., 2018), as well as the effects of splicing mutations (Jayasinghe et al., 2018), are covered by other efforts of the TCGA PanCancer Atlas project.

Results

Dataset

We evaluated all samples in the TCGA PanCancer Atlas collection for which the following data types were available: somatic mutations (whole-exome sequencing), gene expression levels (RNA-Seq), DNA copy-number alterations (Affymetrix SNP6 arrays), and DNA methylation (Infinium arrays). This resulted in a final set of 9,125 samples from 33 different cancer types (Figure 1A, Table S1). In order to account for molecular or histological subtypes, these cancer types were further stratified into a total of 64 genomically distinct tumor subtypes, as previously defined by the individual TCGA analysis working groups (Figure 1A, Table S1). All genomic data and clinical attributes per sample can be visualized through the cBioPortal for Cancer Genomics at (Cerami et al., 2012).

Definition of Pathways and Alterations

We evaluated 10 canonical signaling pathways with frequent genetic alterations, starting with key cancer genes explored in these pathways in previous TCGA publications, and focused on pathway members likely to be cancer drivers (functional contributors) or therapeutic targets. The pathways analyzed are: (1) cell cycle, (2) Hippo signaling, (3) Myc signaling, (4) Notch signaling, (5) oxidative stress response/Nrf2, (6) PI-3-Kinase signaling, (7) receptor-tyrosine kinase (RTK)/RAS/MAP-Kinase signaling, (8) TGFβ signaling, (9) p53 and (10) β-catenin/Wnt signaling (Figures 2 and S1, Table S2). Alterations in DNA repair pathways, epigenetic modifiers, splicing, and other cellular processes frequently altered in cancer were not included, as these primarily provide a background of genomic instability, rather than specifically proliferative potential.

We began by compiling and reviewing the full set of cancer-type specific pathway diagrams from the compendium of TCGA manuscripts published between 2008 and 2017 (Brennan et al., 2013, Cancer Genome Atlas Network, 2012a, Cancer Genome Atlas Network, 2012b, Cancer Genome Atlas Network, 2015a, Cancer Genome Atlas Network, 2015b, Cancer Genome Atlas Research Network, 2008, Cancer Genome Atlas Research Network, 2011, Cancer Genome Atlas Research Network, 2013a, Cancer Genome Atlas Research Network et al., 2013b, Cancer Genome Atlas Research Network, 2014a, Cancer Genome Atlas Research Network, 2014b, Cancer Genome Atlas Research Network, 2014c, Cancer Genome Atlas Research Network, 2014d, Cancer Genome Atlas Research Network, 2017a, Cancer Genome Atlas Research Network et al., 2017b, Davis et al., 2014), each of which included the pathway genes found to be genetically altered in the individual tumor types. These pathway diagrams are publicly available as pre-defined network templates within the visualization tool (Bahceci et al., 2017). By taking the union of pathway members across multiple TCGA studies, we produced a consolidated list of candidate member genes for each of the ten pathways. These were then further curated based on updated literature (including but not limited to the references in Table S2), public pathway databases, and expert opinion (Figure 1B).

The selected genes in the ten pathways were then assessed for recurrent alterations within and across different tumor types as follows (Figure 1B): Alterations of pathway members were classified as activating events (usually specific recurrent missense mutations, i.e., hotspot mutations, amplifications, or fusions involving oncogenes) or inactivating events (truncating mutations, specific recurrent missense or inframe mutations, deletions, as well as fusions and promoter hypermethylation of tumor suppressor genes). Individual alterations were also scrutinized for two features: statistical recurrence across sets of tumor samples and presumed functional impact. We first assessed statistical recurrence using MutSigCV (Lawrence et al., 2014) for mutations and GISTIC 2.0 (Mermel et al., 2011) for copy-number alterations. In order to identify likely functional variants, we then used recurrence across tumor samples at the residue level (linear and 3D mutational hotspots; Chang et al., 2016, Chang et al., 2018, Gao et al., 2017; see STAR Methods) and prior knowledge about specific variants via the OncoKB knowledge base, which contains information about the oncogenic effects and treatment implications of variants in > 400 cancer genes (Chakravarty et al., 2017a). Epigenetic silencing through promoter DNA hypermethylation of tumor suppressor genes was evaluated using the RESET algorithm (see STAR Methods). Gene fusions and structural rearrangements were called from RNA-Seq data using a combination of the STAR-Fusion, EricScript and BreakFast algorithms (Gao et al., 2018, see STAR Methods), and likely passenger events were filtered out based on OncoKB annotation. Through this process, genes without evidence for recurrent or previously known oncogenic alterations were removed from the preliminary pathway templates. The resulting curated pathway templates and the identified genetic alterations were vetted for functional importance by individual pathway experts or the corresponding TCGA PanCancer Atlas pathway-specific analysis working groups, when applicable (Figure 1B). The pathway member genes and the genetic alterations considered as oncogenic are listed in Table S3, and binary genomic alteration matrices are provided as Table S4 (see STAR Methods).

The resulting comprehensive dataset of different types of alterations across many tumor types form the basis of all subsequent analyses regarding pathways, patterns of co-occurrence and mutual exclusivity, as well as potential therapeutic implications. The simplified pathway diagrams in Figure 2 show the most frequently altered genes in the ten pathways, including alteration frequencies as well as the types of oncogenic alterations identified in each of the genes.

Pathway Alteration Frequencies per Tumor Type

For each tumor type and subtype, we computed the fraction of samples with at least one alteration in each of the 10 signaling pathways (Figure 3). A tumor sample was considered as altered in a given pathway if one or more genes in the pathway contained a recurrent or known driver alteration (as described above). Despite the fact that non-recurrent and not previously known alterations were filtered out as likely passenger events and were not included in the alteration frequencies, the microsatellite instability (MSI) and polymerase ε (POLE) mutant subtypes of gastrointestinal and uterine tumors, which had the highest mutation burden, also had the highest overall frequencies of pathway alterations. This is possibly due to the frequent inactivating mutations introduced by the predominant mutation mechanisms in these tumor types (Boland and Goel, 2010, Rayner et al., 2016).

The RTK-RAS pathway was the signaling pathway with the highest median frequency of alterations (46% of samples) across all cancer types. The tumor subtypes with the highest fraction of alterations in this pathway were (in descending order): melanoma (SKCM, 94% altered), the genomically-stable subtype of colorectal cancer (CRC GS, 88%), Her2-enriched breast cancer (BRCA Her2-enriched, 82%), pancreatic cancer (PAAD, 78%), IDH1-wild-type glioma (LGG IDHwt, 82%), lung adenocarcinoma (LUAD, 74%), and thyroid carcinoma (THCA, 84%). Some tumor types, such as lung squamous cell carcinoma (LUSC), EBV-positive esophagogastric cancer (STES EBV), and non-hypermutated uterine cancer (UCEC CN high and CN low), had high rates of alterations in the PI3K pathway, altered in 68%, 80%, 86%, and 95% of samples, respectively. While cell-cycle alterations were common in many tumor types, the pathway was only rarely altered in uveal melanoma (UVM), thymoma (THYM), testicular cancer (TGCT), and acute myeloid leukemia (AML). Alterations in the Wnt pathway were the most variable across cancer types. Colorectal cancer had near universal activation of this pathway, while others, such as renal cell carcinomas and breast cancer, had very low frequencies of alteration in genes in this pathway. The oxidative stress response/Nrf2 pathway had the lowest overall frequency of alteration (1% of samples), and it was altered most frequently in lung squamous (25% altered) and esophagogastric squamous cell carcinoma (STES ESCC, 23%).

Particularly interesting gene alterations across tumor types were observed in the RTK-RAS pathway. KRAS was the most frequently altered gene (9% across all samples), followed by BRAF (7%) and EGFR (4%) (Figure 4A). KRAS alterations were most common in pancreatic carcinoma (PAAD, 72%), genomically stable colorectal cancer (69%), and lung adenocarcinoma (33%) (Figure 4B). BRAF alterations (predominantly known hotspot mutations) were found in melanoma and thyroid carcinoma, altered in 51% and 62% of samples, respectively. EGFR alterations were predominantly found in glioblastoma (GBM, 50%), low grade glioma IDHwt (52%), HPV-negative head and neck cancer (HNSC HPV-, 13%), lung adenocarcinoma (13%), and esophagogastric squamous carcinoma (14%), while ERBB2 alterations were found most commonly in breast cancer and chromosomally unstable esophagogastric carcinoma (STES CIN 26% altered), as well as cervical carcinoma (CESC 23% altered). While most of the alterations described here were previously reported as functional contributors, we identified relatively rare potentially oncogenic alterations in SOS1 (<1%). SOS1 encodes a guanine-nucleotide-exchange factor (GEF) involved in the activation of Ras proteins. Specific germline mutations in this gene are involved in Noonan syndrome (Lepri et al., 2011), and recurrent somatic mutations in SOS1 were recently identified in otherwise RAS-pathway driver-negative lung adenocarcinoma samples (Campbell et al., 2016). We identified recurrent (hotspot) mutations (A90V/T, N233Y/S) and other known activating mutations (M269I/V, G434R, R552S/K/G/M, E846K) in SOS1 in a total of 1% of lung adenocarcinoma samples, 1% of uterine carcinomas, independent of subtype, and at lower frequencies in several other cancer types (Figure 4C). This finding suggests that rarely altered novel cancer genes can be identified as more tumor samples are profiled. A more detailed analysis of RAS pathway alterations is published separately, including a description of downstream transcriptional changes due to malfunctioning Ras signaling and results suggesting that multiple hits in the Ras pathway are capable of increasing overall Ras activity in RAS wild-type tumors (Way et al., 2018).

The alteration frequencies of the most commonly altered genes in the other nine pathways are in Figure 5 (full heatmaps providing frequencies of alteration for every gene in each pathway are in Figures S2 and S3) . In some pathways, the alterations are distributed over many genes (e.g., cell cycle, PI3K), while in others the alterations mainly affect only a few genes (Wnt, Myc, Nrf2). Several pathways are featured in more detail as separate publications, including: (1) PI3K pathway (Zhang et al., 2017): aberrations in the PI3K pathway were found predominantly as activating events in PIK3CA (less commonly in PIK3CB) and inactivating events in PTEN or PIK3R1 with PIK3CA and PTEN alterations being most commonly found in head and neck cancer, breast cancer, gastrointestinal and gynecological tumors; (2) TGFβ pathway (personal communication): The TGFβ pathway had the highest alteration rate in pancreatic and gastrointestinal cancers, while renal and brain cancers, among others, had almost no alterations in this pathway; and (3) Myc pathway (Schaub et al., 2018): Myc pathway alterations were most common in tumor types with amplification of chromosome 8, which contains MYC, such as breast cancer, ovarian cancer (OV), and others.

Mutual Exclusivity and Co-occurrence among Pathway Alterations

Individual tumors typically have multiple functional alterations affecting more than a single pathway. Some pathways may be the target of more than one alteration per patient or distinct pathways may be co-altered in one tumor. Patterns of mutual exclusivity between alterations across large patient cohorts have been associated either with functional redundancy, indicating that once one occurred and is selected the second will not provide a further selective advantage, or with synthetic lethality, indicating that cells cannot survive with both alterations (Etemadmoghadam et al., 2013, Mina et al., 2017). Patterns of co-occurrence of alterations in many tumor samples, on the other hand, indicate functional synergies and, importantly, may reflect resistance to therapy targeting one of the alterations (Nissan et al., 2014)

To explore significantly co-occurring and mutually exclusive alterations by pathway or by gene, we used the SELECT method (Mina et al., 2017). Among the 410 alterations characterized, we identified 156 pairs of mutually exclusive alterations and 117 pairs of co-occurring alterations (Table S5).

Upon mapping these significant pairs to the affected pathways, we found numerous mutually exclusive pairs within the p53, cell-cycle, RAS, and PI3K pathways, suggesting that one alteration is sufficient to functionally alter each of these pathways or that more than one might be disadvantageous (Figure 6A). On the other hand, the Hippo, RTK, and, to a lesser extent, Wnt pathways often had multiple alterations per tumor sample, suggesting co-occurring events that mediate synergistic activation of each pathway (Figure 6A). The SELECT method also identified several significant interdependencies between distinct pathways (Figure S4). For example, activation of RTKs was significantly mutually exclusive with alterations that promote either RAS or PI3K signaling, consistent with RTKs being able to activate either pathway without the need for additional alterations. Notable exceptions were significant co-occurrent alterations in the FGF receptors FGFR2 and FGFR3 and alterations in the PI3K pathway (Figure S4A). The p53 and cell-cycle pathways were frequently co-altered. Indeed, numerous alterations affecting Rb-mediated cell-cycle control were found co-occurring with TP53 mutations. These included amplification of CCNE1, mutation of CDKN2A, RB1 loss, and amplification of CDK6 and E2F3 (Figure S4B). Interestingly, TP53 mutations were found mutually exclusive with CDKN2A deletion, consistent with the latter invariably affecting both p16, regulating the cell cycle, and ARF, promoting p53-dependent apoptosis. Similarly, MDM2 amplification was significantly mutually exclusive with RB1 and CDKN2A loss. However, MDM2 is proximal to CDK4 in the genome and the two genes were almost always co-amplified. Overall, these results indicate that p53 signaling and cell-cycle control are frequently co-altered across multiple tumor types, either through two independent events (e.g., mutations of TP53 and RB1), or through a single alteration that is able to affect both pathways (e.g., CDKN2A deletion).

The strongest co-occurrence among pathways was found between alterations of the PI3K and Nrf2 pathways. Here, gain of function mutations and amplifications of the NFE2L2 gene (encoding for NRF2) significantly co-occurred with PIK3CA amplification and tended to co-occur with PIK3CA mutations and PIK3CB amplification (Figure 6C). Interestingly, even though NFE2L2 amplification was largely mutually exclusive with loss of STK11 (a.k.a. LKB1), the latter significantly co-occurred with loss of function mutations of KEAP1, a negative regulator of NFE2L2. Co-occurring Nrf2-PI3K pathway alterations were most frequent in lung tumors (both squamous cell and adenocarcinoma), esophageal carcinomas, head and neck squamous cell carcinoma and uterine carcinoma, independent of subtype (Figure 6D). In these tumor types, alterations in NFE2L2 and KEAP1 were recurrent and almost perfectly mutually exclusive, and they frequently co-occurred with PIK3CA activation or STK11 loss (Figure 6E). PI3K pathway activation promotes NRF2 accumulation, which, in turn, mediates metabolic pathways required to sustain cell proliferation and protection from reactive oxidative species (Mitsuishi et al., 2012). NRF2 is however kept in check by inhibitory molecules such as KEAP1 and CUL3 (Figure 6F). The observed co-occurrence between alterations of the PI3K and Nrf2 pathways suggests that bypassing these inhibitory mechanisms (either by loss of KEAP1 or CUL3, or by direct over-activation of NFE2L2) is synergistic with active PI3K signaling. Importantly, tumors with these events might depend on NRF2 activity to tolerate PI3K pathway over-activation.

The RTK-RAS pathway contained numerous and some of the most significantly mutually exclusively altered genes, as well as several gene pairs that were significantly concurrently altered (Figure 6G). In particular, alterations promoting EGFR activation (gain of function mutations, fusion, and amplification) were involved in the highest number of significant pairs. EGFR amplification was significantly mutually exclusive with activation of its paralog growth factor receptor Her2 (ERBB2,Figure 6H, Box 1) and with key drivers of the RAS pathway, including oncogenic mutations in BRAF and KRAS as well as loss of NF1 and RASA1 (Figure 6G). Since oncogenic EGFR can be synthetically lethal with mutated KRAS and can mediate resistance to BRAF inhibition in colon cancer and melanoma (Sun et al., 2014, Unni et al., 2015), these results suggest a similar antagonistic interaction with loss of NF1 or RASA1. Overall, alterations of either of these genes were recurrent across multiple tumor types, although almost never in the same patient (Figure 6H, Box 2). On the other hand, in glioblastoma and IDH wild-type low grade glioma, EGFR amplifications were highly co-occurrent with either EGFR mutations or gene fusions (Figure 6H, Box 3) or with focal amplifications of chromosome 4q12, where both KIT and PDGFRA are located (Figure 6H, Box 4). It should be noted that the majority of EGFR and PDGFRA fusions were found coincident with amplifications in these genes, indicating that, potentially, in these cases the same structural variant was detected as both a copy number gain and a fusion (Alaei-Mahabadi et al., 2016). Interestingly, co-amplification of EGFR and PDGFRA has been proposed to be an early event in glioblastoma development, where the two receptors heterodimerize under EGF stimulation and respond to EGFR-inhibitors (Chakravarty et al., 2017b).

Overall, these results provide a map of the cross-talk between pathways and pathway components, reflecting functional interactions and dependencies that could be therapeutically exploited.

Therapeutic Actionability

DNA sequencing has been used routinely to inform the choice of targeted therapy in specific cancer types for several years, and some institutions now apply it more broadly to guide clinical trial enrollment for many additional cancer types. A relatively small number of alterations in a subset of tumor types are currently biomarkers for standard care targeted therapies, and a larger number are potential biomarkers for investigational therapies, some with promising clinical results. Using the OncoKB knowledge base of clinically actionable alterations (Chakravarty et al., 2017a), we systematically assessed all alterations in each sample of each cancer type, distinguishing between standard care actionability (Levels 1 or 2) and investigational therapies (Levels 3 and 4). Overall, 51% of tumors had at least one potentially actionable alteration in the ten signaling pathways, and 57% had at least one actionable alteration when including genes outside of these pathways, most notably BRCA1/2 and IDH1/2 (all numbers referenced below include these additional genes). Apart from the Her2-enriched breast cancer samples, most of which have a standard care targeted therapy, melanoma was the tumor type with the highest fraction of tumors with a Level 1 or 2A alteration (46%) (Figure 7A), mainly due to frequent BRAF mutations (Figure 7B), followed by esophagogastric cancers (ERBB2 amplifications). Luminal A breast cancer was the tumor type with the highest frequency of biomarkers with promising investigational data (Level 3A), driven by the high prevalence of PIK3CA, AKT1 and ERBB2 mutations. Several tumor types had frequent mutations that are biomarkers for drug sensitivity in other cancer types (Level 3B), including endometrial cancer, where PIK3CA mutations are common. Uveal melanoma and testicular non-seminoma had the lowest percentage of potentially targetable samples (2.5% and 8.5%, respectively); thymoma, mesothelioma (MESO), and renal clear cell carcinoma (KIRC) also had low frequencies of potentially actionable alterations.

Thirty percent of tumor samples had two or more potentially targetable alterations (Figure 7C). Among these, the MSI-H and POLE-mutated tumor subtypes had the highest proportion of samples with multiple potentially actionable alterations (not considering the fact that patients with MSI-H tumors are now eligible for immunotherapy). Other tumor types with a high frequency of samples with multiple targetable alterations included non-hypermutated endometrial cancer (64%), colorectal cancer (37%), and breast cancer (28%).

Finally, we searched for candidate drug combinations that could prove effective across different tumor types based on the occurrence of actionable alterations. Hypermutant MSI and POLE subtypes had a high fraction of samples of actionable alterations corresponding to various drug combinations. In other tumor subtypes, a combination of CDK4 and MDM2 inhibitors was the most commonly indicated combination (1% total), in particular in dedifferentiated liposarcomas (SARC DDLPS), in which 78% of the cases had co-amplification of the two targets (Figure 7D). By a similar consideration linking actionable alterations of targets to their inhibitors, a combination of HER2 and PI3K inhibitors might be beneficial across multiple tumor types, in particular Her2-enriched breast cancer (17%), uterine carcinosarcoma (UCS, 7%), chromosomally unstable endometrioid carcinoma (UCEC CN high, 7%), and cervical adenocarcinoma (7%) (Figure 7D). Additional candidate combination therapies include PI3K and MEK inhibitors in EBV+ stomach tumors (10%), CDK4 and PI3K inhibitors in glioblastoma multiforme (7%), HER2 and MEK inhibitors in pancreatic cancer (7%), PI3K and RAF inhibitors in melanoma (SKCM, 12%), and IDH and PI3K inhibitors in IDH-mutant low grade glioma (14%) (Figure 7D). While there are many steps from the observation of combinations of genetic alterations to valid combination therapies, this survey indicates the wide landscape of potential tumor-type specific novel therapeutic combinations that can be explored in experimental and clinical contexts.

Discussion

Signaling pathways are somatically altered in cancer at varying frequencies and in varying combinations across different organs and tissues, indicative of complex interplay and pathway cross-talk. Understanding the extent, detailed mechanisms, and co-occurrence of the oncogenic alterations in these pathways is critical for the development of new therapeutic approaches that can improve patient care.

Here we performed a comprehensive characterization of 10 selected signaling pathways across the 33 cancer types analyzed by TCGA. This report constitutes the first pan-cancer exploration that uses a uniformly processed dataset and a standardized set of pathway templates, curated through a combination of computational methods and expert review (Figures 1 and 2). The results highlight similarities and differences in frequencies of alteration of individual pathways in different cancer types and specific molecular subtypes (Figure 3). They also underscore the potential for discovering previously uncharacterized alterations in pathway genes that occur at low frequencies and might otherwise remain statistically unnoticeable (see SOS1, Figure 4). More generally, even though a small set of critical genes contains a very large fraction of alterations in these pathways (Figures 4 and 5), there is a complex interplay of co-occurring and mutually exclusive alterations within and across pathways (Figure 6). In spite of the accumulating wealth of biological knowledge and the accepted oncogenic relevance of these pathways, the number of currently approved biomarkers linked to standard of care therapies remains sparse (Figure 7), but additional drug targets in these pathways will hopefully emerge, and candidates for combination therapy will be explored.

This analysis of targetable alterations only included currently approved therapies or investigational therapies with reported promising results. These predominantly target the RTK-RAS, PI3K, cell-cycle, and p53 pathways. While some of these therapies are standard care, many are still investigational, and further testing is required to assess how effective different targeted therapies will be across tumor types and in tumors with different co-mutation spectra. Efforts are underway to develop therapies that target additional pathways, some of which are in clinical trials (Table S6) (Park and Guan, 2013), (Whitfield et al., 2017), (Whitfield et al., 2017), (Aster and Blacklow, 2012), (Takebe et al., 2014), (Buijs et al., 2012), (Sheen et al., 2013), (Pai et al., 2017). In the Wnt signaling pathway, for example, two approaches involve drugs targeting PORCN, which is involved in the processing of wingless proteins, and monoclonal antibodies directed at proteins in the Frizzled gene family. While the Nrf2 pathway does not have therapies directly targeting any of the pathway members included in this study, alterations in Nrf2 pathway members (NFE2L2 and KEAP1) are used as part of the inclusion criteria in the Phase 2 trial of a TORC1/2 inhibitor. Clinical trials involving these pathways exemplify opportunities in precision medicine to associate additional functional alterations as part of inclusion criteria (Table S6). Not all apparently functional mutations, however, represent therapeutic targets, as illustrated, e.g., by the unusually large number of mutations in the MSI-H and POLE-mutated tumor subtypes, of which only a small fraction plausibly dominate oncogenesis. The observed co-occurrence patterns indicate a potential for combination therapies in some tumor types. The development of targeted combination therapies has been challenging for several reasons, including lack of safety data for combinations, the relatively slow pace of adoption of clinically approved multi-panel gene tests and of clinical trials testing combinations of multiple targeted therapies. However, there is a growing corpus of promising preclinical data indicating such combinations can be effective, such as the combination of MDM2 and CDK4 inhibitors (Laroche-Clary et al., 2017), and the combination of PI3K inhibitors and HER2 inhibitors in HER2-positive/PIK3CA mutant breast cancer patients, even when single gene-therapy approaches (e.g., PI3K monotherapy for PIK3CA mutant tumors) have thus far not had definitive clinical impact.

Although we cover a diverse range of oncogenic processes that spans most tissues and organ systems (Figures 1 and 3), we did not include some tumor types in the scope of this TCGA project, including most hematologic cancers. Furthermore, in spite of the relatively large set of samples, this effort is still underpowered to reliably discover tumor-type specific alterations that occur at very low frequencies; these will require further exploration using larger tumor-type specific sample sets.

The original aim and scope of TCGA was to genomically characterize primary, untreated tumors with a basic set of genetic alterations and transcript profiles. As the program is now completed, a future challenge is to expand these analyses to larger sample sets, additional data types, such as metabolite levels, a wider range of epigenetic states, post-translational modifications of proteins, and to investigate metastatic disease and genomic alterations that arise in post-treatment samples, as well as analyzing the role of a wider range of germline alterations and their interplay with somatic events. These new avenues of research will benefit from pathway-level analysis for which the templates and template curation pipelines presented here constitute a promising starting point. Similarly, as the catalog of clinically actionable alterations continues to grow, understanding intra- and inter-pathway dependencies, such as the ones considered here, will be crucial for the development of effective combination therapies that address or prevent resistance to initially successful single agent therapies.

The curated pathway templates and the uniformly processed dataset of alteration calls in 9,125 tumor samples are publicly available (Tables S3 and S4) and can be easily accessed through the PathwayMapper tool (), which allows alteration frequencies to be visually overlaid on the pathway templates; and, via the cBioPortal for Cancer Genomics (). This pathway landscape in The Cancer Genome Atlas is meant to provide a valuable resource for clinical oncologists, for cancer researchers and for a broad scientific community interested in cancer precision medicine.

Acknowledgments

This work was supported by NIH Grants U54 HG003273, U54 HG003067, U54 HG003079, U24 CA143799, U24 CA143835, U24 CA143840, U24 CA143843, U24 CA143845, U24 CA143848, U24 CA143858, U24 CA143866, U24 CA143867, U24 CA143882, U24 CA143883, U24 CA144025, P30 CA016672.

Authors Contributions

F.S.-V., M.M., J.A., W.K.C., A.L., E.M.V.A., A.D. C., G.C., C. S. and N. S. designed the study; F.S.-V., M.M., J.A., W.K.C., A.L., J.G., K.L., S.D., H.S.K., Z.H., A.O., B.G., J.G., H.Z., R.K., I.B., L.D., U.D., C.K., Q.G., M.H.B,W.-W.L., S.M.F., I.S., H.L., L.D., A.D.C., G.C., and N.S. collected and annotated the data; F.S.-V., M.M., J.A., W.K.C., A.L., D.C., W.Z., H.S., P.W.L., F.D., D.L.L., S.S., G.P.W., C.S.G., Y.X., C.W., A.I., A.H.B., T.G.B., A.J.L., G.D.H., T.G., L.N.K., G.M., C.H., A.D.T., M.J.F., F.M., M.M., E.V.A, A.D.C., G.C., C.S. and N.S. analyzed and interpreted the data. The Cancer Genome Atlas Research Network was responsible for the project administration. F.S.-V., M.M., J.A., W.K.C., A.L., E.M.V.A., A.D.C., G.C., C.S. and N.S. drafted the manuscript. All authors participated in editing or reviewing of the manuscript, and all authors approved the submitted manuscript.

Declaration of Interests

Michael Seiler, Peter G. Smith, Ping Zhu, Silvia Buonamici, and Lihua Yu are employees of H3 Biomedicine, Inc. Parts of this work are the subject of a patent application: WO2017040526 titled “Splice variants associated with neomorphic sf3b1 mutants.” Shouyoung Peng, Anant A. Agrawal, James Palacino, and Teng Teng are employees of H3 Biomedicine, Inc. Andrew D. Cherniack, Ashton C. Berger, and Galen F. Gao receive research support from Bayer Pharmaceuticals. Gordon B. Mills serves on the External Scientific Review Board of Astrazeneca. Anil Sood is on the Scientific Advisory Board for Kiyatec and is a shareholder in BioPath. Jonathan S. Serody receives funding from Merck, Inc. Kyle R. Covington is an employee of Castle Biosciences, Inc. Preethi H. Gunaratne is founder, CSO, and shareholder of NextmiRNA Therapeutics. Christina Yau is a part-time employee/consultant at NantOmics. Franz X. Schaub is an employee and shareholder of SEngine Precision Medicine, Inc. Carla Grandori is an employee, founder, and shareholder of SEngine Precision Medicine, Inc. Robert N. Eisenman is a member of the Scientific Advisory Boards and shareholder of Shenogen Pharma and Kronos Bio. Daniel J. Weisenberger is a consultant for Zymo Research Corporation. Joshua M. Stuart is the founder of Five3 Genomics and shareholder of NantOmics. Marc T. Goodman receives research support from Merck, Inc. Andrew J. Gentles is a consultant for Cibermed. Charles M. Perou is an equity stock holder, consultant, and Board of Directors member of BioClassifier and GeneCentric Diagnostics and is also listed as an inventor on patent applications on the Breast PAM50 and Lung Cancer Subtyping assays. Matthew Meyerson receives research support from Bayer Pharmaceuticals; is an equity holder in, consultant for, and Scientific Advisory Board chair for OrigiMed; and is an inventor of a patent for EGFR mutation diagnosis in lung cancer, licensed to LabCorp. Eduard Porta-Pardo is an inventor of a patent for domainXplorer. Han Liang is a shareholder and scientific advisor of Precision Scientific and Eagle Nebula. Da Yang is an inventor on a pending patent application describing the use of antisense oligonucleotides against specific lncRNA sequence as diagnostic and therapeutic tools. Yonghong Xiao was an employee and shareholder of TESARO, Inc. Bin Feng is an employee and shareholder of TESARO, Inc. Carter Van Waes received research funding for the study of IAP inhibitor ASTX660 through a Cooperative Agreement between NIDCD, NIH, and Astex Pharmaceuticals. Raunaq Malhotra is an employee and shareholder of Seven Bridges, Inc. Peter W. Laird serves on the Scientific Advisory Board for AnchorDx. Joel Tepper is a consultant at EMD Serono. Kenneth Wang serves on the Advisory Board for Boston Scientific, Microtech, and Olympus. Andrea Califano is a founder, shareholder, and advisory board member of DarwinHealth, Inc. and a shareholder and advisory board member of Tempus, Inc. Toni K. Choueiri serves as needed on advisory boards for Bristol-Myers Squibb, Merck, and Roche. Lawrence Kwong receives research support from Array BioPharma. Sharon E. Plon is a member of the Scientific Advisory Board for Baylor Genetics Laboratory. Beth Y. Karlan serves on the Advisory Board of Invitae.