Co-occurring genomic alterations in non-small cell lung cancer biology and therapy

Abstract

The impressive clinical activity of small molecule receptor tyrosine kinase inhibitors (TKIs) for oncogene-addicted subgroups of non-small cell lung cancer (NSCLC) [for example those driven by activating mutations in the gene encoding epidermal growth factor receptor (EGFR) or rearrangements in the genes encoding the receptor tyrosine kinases anaplastic lymphoma kinase (ALK), ROS proto-oncogene 1 (ROS1), and rearranged during transfection (RET)] has established an oncogene-centric molecular classification paradigm in this disease. However, recent studies have revealed considerable phenotypic diversity downstream of tumor-initiating oncogenes. Co-occurring genomic alterations, particularly in tumor suppressor genes such as TP53 and LKB1 (also known as STK11), have emerged as core determinants of the molecular and clinical heterogeneity of oncogene-driven lung cancer subgroups through their effects on both tumor cell-intrinsic and non-cell-autonomous cancer hallmarks. In this review, we discuss the impact of co-mutations on the pathogenesis, biology, micro-environmental interactions, and therapeutic vulnerabilities of NSCLC and assess the challenges and opportunities that co-mutations present for personalized anti-cancer therapy, as well as the expanding field of precision immunotherapy.

Table of Contents Summary

Co-occurring genomic alterations contribute to the heterogeneity of driver oncogene-defined non-small cell lung cancer (NSCLC) subgroups. This Review discusses the effects of co-mutations on the pathogenesis, biology, microenvironmental interactions, and therapeutic vulnerabilities of NSCLC.

Introduction

The identification in 2004 of activating oncogenic mutations in the tyrosine kinase domain of epidermal growth factor receptor (EGFR) in a subset of patients with non-small cell lung cancer (NSCLC) that exhibited dramatic clinical responses to the first-generation EGFR tyrosine kinase inhibitor (TKI) gefitinib launched the field of targeted therapy in NSCLC and reinforced the concept of oncogene addiction as a pillar of modern cancer therapeutics^{1, 2}. Subsequent discovery of ALK re-arrangements in 2007³ in 3–7% of NSCLC expanded the spectrum of targetable genomic alterations in this disease⁴. Since then, several additional driver events with robust transforming potential have been reported, including oncogenic ROS1⁵, RET⁶, NTRK1⁷ and NRG1⁸ fusions, oncogenic somatic mutations in BRAF (V600E and non-V600E)^9–11, intragenic insertions in ERBB2 (also known as HER2)¹² and exon 14 skipping mutations in the MET proto-oncogene^13–15. Pivotal clinical studies established the superiority of molecularly targeted therapy compared with platinum-doublet chemotherapy [G] for EGFR-mutant, ALK-rearranged and ROS1-rearranged NSCLC and led to the FDA approval of several first, second and third-generation small molecule inhibitors of mutant oncoproteins^16–32. The robust clinical activity of these targeted agents, coupled with the apparent mutual exclusivity of strong oncogenic drivers in NSCLC, cemented a driver oncogene-centric paradigm in NSCLC oncogenesis and molecular classification. This prevailing model, commonly represented graphically as an “oncogenic pie chart”, constitutes the bedrock of NSCLC clinical practice and the framework that underpins the design and implementation of a generation of precision oncology clinical trials aimed at matching patients with available targeted therapies based on identification of a single genomic driver event (Figure 1). However, accumulating evidence points towards the existence of substantial clinical heterogeneity within oncogenic-driver defined NSCLC subgroups that is currently incompletely accounted for by the single oncogenic driver model. In this review, we discuss the emerging role of co-occurring genomic alterations as major determinants of both tumor cell-intrinsic as well as non-cell-autonomous cancer hallmark traits, including their impact on the composition of the tumor microenvironment and response to systemic anti-cancer therapies.

Figure 1. Single oncogenic driver paradigm of lung adenocarcinoma molecular classification.

The dominant contemporary model of non-small cell lung cancer pathogenesis and molecular classification is based on identification of single and largely non-overlapping oncogenic driver events. Oncogenic pie charts are presented for early-stage (a) and metastatic (b) lung adenocarcinomas (LUADs). The prevalence of individual genomic alterations in early-stage disease is based on combined analysis of whole exome sequencing data from the PanCancer Atlas cohort of The Cancer Genome Atlas (TCGA) (n=785)^177–179, as well as the cohorts reported by Imielinski et al (n=148)⁶² and Kadara et al (n=108)¹³², following exclusion of patients with stage 4 disease (n=741 patients in total). The prevalence of MET splice site alterations, MET amplification, ERBB2 amplification, HRAS and NRAS mutations as well as ALK, ROS1 and RET fusions was based on data from the TCGA and Imielinski cohorts only. Oncogenic driver alterations in advanced or metastatic LUAD (encompassing both treatment-naïve patients as well as patients that received prior anti-cancer therapies) are based on next-generation sequencing of pre-defined panels of cancer-relevant genes from patients treated at Memorial Sloan Kettering Cancer Center (N=860,MSK-IMPACT panel⁵⁹) and samples referred to Foundation Medicine (n=4402,FoundationOne panel¹⁵) (n=5262 patients with advanced/metastatic LUAD in total). The prevalence of alterations in NF1, NRAS, HRAS, MAP2K1, FGFR1/2 and RIT1 is based on data from MSK-IMPACT only. It is notable that although the prevalence of oncogenic KRAS mutations is similar in both early and advanced stage LUADs the frequency of other driver alterations (for example truncating NF1 mutations) differs substantially depending on the disease stage. The increased prevalence of EGFR mutations in the metastatic dataset may partially reflect referral bias. Data were visualized and downloaded from the open source web program cBioPortal^{180, 181} or curated from the scientific literature.

Diversity in driver-defined subgroups.

There is mounting evidence that substantial molecular and clinical heterogeneity exists within oncogenic driver-defined subgroups of NSCLC (“intra-driver heterogeneity”). Despite known associations between certain NSCLC oncogenic subtypes and distinct tumor histopathologic features or growth patterns, NSCLCs driven by the same dominant oncogenic alteration can vary considerably in their histological appearance and immunohistochemical profile. For example, KRAS-mutant lung adenocarcinomas (LUADs) demonstrate dual propensity towards either solid growth pattern with positivity for the NKX2–1 homeobox transcription factor (also known as TTF1) or, alternatively, invasive mucinous adenocarcinoma histology and corresponding lack of NKX2–1 expression³³. At the molecular level, considerable efforts have focused on segregating LUAD into molecular subtypes on the basis of multi-dimensional molecular profiling, coupled with unsupervised clustering computational approaches³⁴. Enrichment for specific oncogenic drivers has been observed within distinct subtypes, however cluster membership typically transcends initiating oncogenes, thus providing further evidence for intra-driver molecular diversity³⁴.

Most importantly, overwhelming evidence indicates that intra-driver molecular diversity translates into heterogeneous clinical behavior and variable sensitivity to anticancer therapies. Across clinical trials of first-line targeted therapy for oncogene-addicted subgroups of NSCLC, rates of objective response [G] typically range between 50% and 83% and complete responses [G] are rare; in addition, some patients exhibit de novo resistance^{16–26, 28–32, 35}. Even more variable are duration of response to targeted therapy, progression-free survival and overall survival^{16–26, 28–32, 35}. Phenotypic variability and therapeutic response heterogeneity are particularly evident within KRAS-mutant LUAD. The pervasive diversity of this oncogenotype was aptly demonstrated in a study that applied affinity propagation clustering analysis³⁶ to mRNA expression data from 106 genomically-annotated NSCLC cell lines; strikingly, variation in mRNA expression within KRAS-mutant NSCLC cell lines was equivalent to that observed across the entire cell line panel³⁷. Inter- and intra-driver heterogeneity are also evident following treatment with inhibitors of the immune checkpoint molecules PD-1 or PD-L1, with only ~20% of unselected NSCLC patients deriving durable clinical benefit^38–43.

What are the molecular underpinnings of this remarkable intra-driver heterogeneity in NSCLC? In many cases, divergent clinical behavior can be directly attributed to the distinct effects of individual oncogenic alleles. Multiple studies have affirmed the favorable prognostic impact of exon 19 EGFR deletions compared with exon 21 L858R amino acid substitution, although the molecular basis for this association has not been conclusively determined^{44, 45}. Furthermore, EGFR exon 20 in-frame insertion mutants are recalcitrant to all currently FDA-approved EGFR TKIs due to insertion-imposed steric hindrance of the drug binding pocket, but exhibit sensitivity to poziotinib - a smaller and more flexible inhibitor - in vitro and in vivo^{46, 47}. Among ALK-rearranged NSCLC, both the fusion partner as well as EML4-ALK fusion variantshave been considered candidate modifiers of transforming potential and response to ALK TKIs^{48, 49}. For example, the PRKAR1A-ALK fusion was consistently demonstrated to be less sensitive to first, second and third generation ALK TKIs⁴⁸. In addition, EML4-ALK variant 3 was associated with more frequent secondary resistance mutations (including the G1202R solvent front mutation) compared to EML4-ALK variant 1 and, consequently, longer progression-free survival with the 3^rd generation ALK inhibitor lorlatinib, that is active against the EML4-ALK^G1202R mutation⁴⁹. Similarly, in RET-rearranged NSCLC, non-KIF5B-RET fusions have been associated with significantly higher response rates to RXDX-105 (a RET and BRAF inhibitor) but not to the potent and selective RET inhibitor LOXO-292⁵⁰. Finally, distinct KRAS mutant alleles differentially engage downstream effectors with KRAS^G12C or KRAS^G12V preferentially activating RALA or RALB signaling and KRAS^G12D triggering increased PI3K–AKT and MAPK/ERKpathway activation⁵¹. Currently, the prognostic and predictive utility of KRAS alleles in NSCLC remains unclear but is likely to increase in view of the ongoing clinical development of covalent, direct KRAS^G12C inhibitors^52–56. Nonetheless, distinct types of somatic mutations or gene rearrangements appear to only partially account for intra-driver heterogeneity because marked differences in biological behavior can also be observed between NSCLC that bear identical oncogenic alterations in driver genes. Taken together, these studies challenge the single-oncogene paradigm in NSCLC by unveiling multiple layers of heterogeneity within oncogenic subgroups that can only partially be attributed to the driver oncoprotein itself.

Co-occurring genomic alterations

The compendium of co-occurring genomic alterations in NSCLC are potentially more impactful than distinct mutations in oncogenic drivers with regard to determining tumor heterogeneity. LUADs and lung squamous cell carcinomas (LUSCs) are characterized by a high average number of somatic mutations per Mb in comparison to many other tumor types⁵⁷. Although passenger mutations account for the largest fraction of this mutational burden, combinations of somatic mutations in bona fide cancer driver genes are identified in the majority of LUAD and a substantial fraction of LUSC, even when next generation sequencing platforms are limited to the evaluation of pre-defined sets of cancer-relevant genes. Importantly, large-scale profiling studies utilizing either whole exome sequencing or broad targeted sequencing panels in NSCLC tumors have revealed multiple non-random patterns of co-occurring or mutually exclusive mutations, which typically vary depending on the particular oncogenic driver mutation^{15, 34, 58–62}. From an evolutionary standpoint, co-selection of oncogenic alterations implies functional co-operation that converges on improved fitness, whereas mutual exclusivity indicates redundancy (potentially manifesting as soft exclusivity) or antagonism (resulting in more strict patterns of mutual exclusivity due to deleterious effects of the combined alterations)^{63, 64}. Thus, from its inception, NSCLC develops through a network of evolving genetic interactions that collectively determine cancer hallmark traits^{65, 66}. This further suggests that early oncogenic events may channel tumor evolution towards distinct trajectories and influence the likelihood of positive or negative selection of subsequent genomic alterations. It is important to note, however, that even genomic alterations that do not show statistically significant patterns of co-occurrence may still have important interactions biologically. For example, mutations in TP53 (which encodes p53) are under-represented among KRAS-mutant LUAD compared to other oncogene-driven subgroups, yet p53 inactivation is common and impactful in KRAS-mutant LUAD^67–71. The importance of co-mutations as mediators of diverse NSCLC phenotypes has only recently attracted focus and their functional impact remains largely uncaptured within current molecular stratification frameworks.

Knowledge of the clinical context is paramount when evaluating the functional importance as well as prevalence of co-occurring genomic alterations. In particular, it is critical to distinguish between early-stage, surgically resected tumors and locally advanced or metastatic disease because several distinct patterns of co-mutations are enriched in metastatic disease, likely reflecting acquisition of traits that promote tumor progression and metastatic dissemination^{72, 73}. In addition, selective pressure imposed by previous anticancer therapy can substantially influence patterns of co-mutations; therefore, detailed knowledge of prior therapeutic exposures is critical for accurate interpretation and understanding of the functional effect of co-mutation patterns.

Determination of the clonal or sub-clonal nature as well as timing of individual co-alterations may also provide important information regarding their contributions to different stages of carcinogenesis and impact on therapeutic response. Early clonal events are more likely to impact core cancer hallmarks that are critical for tumor initiation. In addition, targeting clonal events is more likely to yield sustained responses; although both clonal and sub-clonal events can contribute to clinical resistance, clonal events are more likely to result in primary resistance. In the landmark TRACERx study, multiregion sequencing of 100 early-stage NSCLCs provided a measure of the extent of clonal driver events (1–18 in LUAD and 1–14 in LUSC) and sub-clonal driver events (0–10 in LUAD and 0–12 in LUSC) and established a catalog of clonal alterations⁷⁴. Furthermore, evidence from other tumor types supports the notion that within a network of epistatic oncogenic interactions the chronology – or order – of individual genomic alterations can impart distinct phenotypic outcomes. For example in myeloproliferative disorders, the order in which mutations in JAK2 and TET2 arise in hematopoietic stem and progenitor cells can affect age of disease onset, influence the likelihood of the disease manifesting as polycythemia vera versus essential thrombocythemia and result in different propensities for development of thrombosis^{75, 76}.

Finally, when assessing patterns of co-occurring events in human NSCLC it is important to also consider the impact of mutational processes and immune selection. Several distinct mutational signatures sculpt the genome of NSCLC – including signatures of tobacco exposure and APOBEC-mediated cytidine deamination⁵⁷. Certain recurrent oncogenic mutations, for example classical mutations in PIK3CA (which encodes a catalytic subunit of PI3K), occur within APOBEC deaminase trinucleotide motifs and are enriched in tumors with a high APOBEC mutational footprint⁷⁷. While the full extent to which mutational processes account for unique combinations of somatic genomic alterations in NSCLC is currently unknown, there is evidence that APOBEC-mediated mutagenesis fuels sub-clonal diversification and branched evolution^{78, 79}. Furthermore, tumor genomes can be shaped by immunosurveillance through early elimination of clones that present strong antigenic neo-peptides. For example, the major histocompatibility complex (MHC) class I genotype of individual patients was demonstrated to impose restrictions on the tumor mutational landscape and predict for selection of distinct driver mutations⁸⁰. Such immunoediting likely influences patterns of co-mutations in NSCLC and these associations warrant further study. On the other hand, imposition of a cold tumor immune microenvironment [G] as a result of tumor cell-intrinsic processes may relax immune selection and result in a more diverse spectrum of co-mutations.

Co-mutations within LUAD subgroups

KRAS-mutant LUAD

Activating mutations in KRAS are the most prevalent oncogenic driver event in both early-stage and metastatic LUAD, occurring in 25–32% of tumors^{34, 59, 60, 62, 67}. As noted previously, KRAS-mutant LUADs are intrinsically heterogeneous in their biology and clinical behavior. We previously identified three robust and reproducible transcriptomic sub-groups of KRAS-mutant LUAD by applying non-negative matrix factorization consensus clustering^81–83 to RNASeq data from 68 tumors from The Cancer Genome Atlas (TCGA) dataset⁶⁷. Remarkably, superimposition of somatic genetic alterations of key tumor suppressor genes revealed non-overlapping patterns of co-occurring genomic alterations in the three subgroups: one subgroup was dominated by co-occurring TP53 alterations (thereafter referred to as KP), whereas co-mutations or genomic loss in LKB1 (also known as STK11) were a hallmark of the second cluster (referred to as KL), that was further enriched in somatic mutations in KEAP1 and ATM. Bi-allelic inactivation of the CDKN2A/CDKN2B locus was significantly enriched in the third cluster (referred to as KC), that was defined by lack of NKX2–1 expression. Notably, distinct KRAS alleles were not differentially distributed between the three clusters – with the exception of enrichment for KRAS^G12D in the KC subgroup in some cohorts. These findings established co-mutations as major determinants of the molecular diversity of KRAS-mutant LUAD.

Landmark large scale sequencing studies have established a census of major KRAS co-mutations in both early-stage and advanced LUAD^{15, 34, 59, 62}. The significance of co-occurrence for individual pairs of genetic alterations varies depending on the size of the clinical cohort, the number of possible interactions that are surveyed and the sequencing platform. However, co-mutations in a set of core genes including LKB1, KEAP1, ATM and RBM10 are consistently enriched in KRAS-mutant LUAD (Figure 2). Additional significantly co-altered genes reported in some studies include PTPRD, U2AF1, POLE, NTRK3 and LRP1B. Mutations in TP53 and inactivation of CDKN2A, CDKN2B or combined CDKN2A/CDKN2B loss due to bi-allelic deletion, are common and functionally relevant co-alterations, although they are not enriched in KRAS-mutant compared to other oncogene-driven subgroups. Mutations in other established drivers within the receptor tyrosine kinase-RAS-RAF network including EGFR, ERBB2, BRAF, NF1, as well as ALK, ROS1 and RET rearrangementsare largely non-overlapping with KRAS, although the strength of their negative association varies depending on the individual gene.

Figure 2. Spectrum of major co-occurring genomic alterations in KRAS- and EGFR-mutant lung adenocarcinoma.

Volcano plots (left graphs) summarizing enrichment of individual co-alterations in: KRAS-mutant compared with KRAS-wild-type LUADs (a) and EGFR-mutant compared with EGFR-wild-type LUADs (b). The magnitude of co-mutation enrichment is indicated on the x-axis and is expressed as log₂ (% in KRAS-mutant / % in KRAS-wild-type) or log₂ (% in EGFR-mutant / % in EGFR-wild-type) respectively, whereas the statistical significance of the association is plotted on the y-axis and is expressed as –log₁₀P value (derived from a Fisher’s exact test). Significantly enriched co-mutations based on a q value <0.05 (derived from Benjamini-Hochberg procedure¹⁸²) are highlighted in red, whereas under-represented genomic events are highlighted in blue. The prevalence of each co-alteration in KRAS-mutant and KRAS-wild-type groups (or EGFR-mutant and EGFR-wild-type groups) is shown in the adjacent frequency plots (right graphs of parts a and b). Targeted next generation sequencing-based molecular profiling (MSK-IMPACT platform) from 860 patients with metastatic LUAD treated at Memorial Sloan Kettering Cancer Center were included in this enrichment analysis that was performed using the cBioPortal web program^{180, 181}. Oncogene-driver specific, non-random patterns of co-occurring alterations in key tumor suppressor genes are evident for both KRAS-mutant and EGFR-mutant tumors.

Insights from genetically engineered mouse models and cell line studies have been pivotal towards elucidating the phenotypic sequelae of the most prominent KRAS co-mutations in NSCLC. Somatic deletion of Lkb1 is insufficient for initiation of lung carcinogenesis in mice as a singular event but dramatically accelerates Kras^G12D-driven carcinogenesis and fosters early metastatic dissemination⁸⁴. In addition, loss of Lkb1 results in epigenetic reprogramming and an expanded tumor histological repertoire, with high incidence of squamous or adenosquamous carcinomas, in agreement with data supporting enrichment of LKB1 mutations in human adenosquamous NSCLC^84–86. Mechanistically, trans-differentiation [G] is mediated by LKB1 loss-triggered down-regulation of the Polycomb Repressive Complex 2 (PRC2) subunit EED and relief of PRC2-mediated repression of squamous differentiation genes⁸⁶. Epigenetic reprogramming upon LKB1 loss in KRAS-mutant cells is further fueled by a metabolic network, that promotes increased flux of glucose-derived carbon towards serine biosynthesis and the methionine salvage pathway, bolstering synthesis of S-adenosyl methionine (SAM), a critical substrate for DNA methylation⁸⁷. Generation of SAM via this pathway, coupled with up-regulation of DNA methyl-transferases, underpins an increase in global levels of CpG methylation in KL cells that is critical for tumor maintenance. Furthermore, KL NSCLC cells exhibit increased dependence on dTTP synthesis and rely on an unorthodox pathway of pyrimidine biosynthesis that utilizes mitochondrially generated carbamoyl phosphate^{88, 89}. The unique metabolic phenotypes associated with combined expression of oncogenic KRAS and LKB1 inactivation – but not with mutations in either gene alone - may at least partially explain their preferential co-occurrence in human NSCLC.

Intricately linked with LKB1 inactivation is loss of KEAP1, an adaptor protein that mediates ubiquitination and proteasomal degradation of NRF2, a key transcription factor in cellular antioxidant, metabolic, cyto-protective, and anti-inflammatory pathways⁹⁰. Somatic mutations in LKB1 and KEAP1 significantly co-occur with mutant KRAS and with each other in NSCLC^{15, 59, 63, 91}. In the conditional Kras^G12D/+;Trp53^Fl/FL mouse model, loss of Keap1 increases both tumor burden and the percent of high-grade lesions, pointing towards roles in both tumor initiation and progression⁹². At the cellular level KEAP1 loss results in increased cellular proliferation in vivo and an altered metabolic profile characterized by increased glucose-derived carbon flux towards the pentose phosphate and serine–glycine biosynthetic pathways^93–95 with enhanced dependence on glutaminolysis for tricarboxylic acid (TCA) cycle anaplerosis⁹², a dependence that is further enhanced by co-occurring LKB1 alterations⁹⁶. Thus, NRF2- mediated metabolic reprogramming and regulation of redox homeostasis likely underpin the strong co-selection of KEAP1 with LKB1 and KRAS mutations in NSCLC at least partially because oncogenic KRAS itself promotes oxidative stress and anabolic metabolism and because KL NSCLC cells depend on the pentose phosphate pathway for NAPDH generation and for detoxification of reactive oxygen species because these cells have defective fatty acid oxidation^{94, 97, 98}. This suggests that up-regulation of a NRF2-driven transcriptional program may represent a bottleneck in the evolution of LKB1-deficient NSCLC. Potentiation of cellular anabolic, antioxidant and detoxification pathways collectively support the aggressive clinical phenotype of KEAP1-mutant NSCLC that is concordant with its role as an independent negative prognostic indicator^{92, 99}.

Loss of p53 or expression of either dominant negative or dominant gain-of-function Trp53 mutants also co-operate with oncogenic Kras to induce LUADs with shortened latency and increased metastatic proclivity, although these tumors are less aggressive than those with Lkb1 loss^{69–71, 84}. Notably, the selective pressure for p53 inactivation is most critical in aggressive, high-grade lesions that exhibit high levels of ERK signaling, whereas engagement of p53-mediated signaling is minimal in low-grade adenomas, thus highlighting stage and signal intensity - dependent patterns of co-operativity^100–102. This notion is further supported by identification of TP53 mutational inactivation as a clonal and predominantly early event in established NSCLC that precedes genome doubling and subsequent branched evolution^{74, 103}.

Mutations in ATM, encoding an apical kinase in the DNA damage response pathway, also significantly co-occur with mutant KRAS. In murine models the impact of Atm inactivation on Kras-driven lung carcinogenesis is context-dependent and varies according to the functional status of p53¹⁰⁴. In a p53-proficient setting, bi-allelic loss of Atm is tolerated but does not promote Kras^G12D-initiated neoplasia. In contrast, complete Atm inactivation is incompatible with cellular viability in the context of Kras^G12D expression and bi-allelic Trp53 inactivation, suggesting that excessive DNA damage in this context removes incipient cancer cells from the proliferative pool. Interestingly, Kras-driven lung carcinogenesis is accelerated by incomplete Atm loss in a p53 deficient settingThus, data from genetically engineered mouse models (GEMMs) point towards a context-dependent, conditional haplo-insufficient role for Atm loss in Kras^G12D-driven lung tumorigenesis. Selection against complete ATM inactivation may explain the mutual exclusivity of ATM and TP53 mutations in human LUAD as well as the enrichment of ATM mutations in the KL subgroup⁶⁷. However, ATM haplo-insufficiency has not been convincingly established yet in human LUAD, where there is evidence for complete lack of ATM expression by immunohistochemistry in a significant proportion of LUAD¹⁰⁵.

Intriguingly, analysis of patterns of co-occurrence and mutual exclusivity between a set of 505 pre-selected candidate functional genomic events in 6456 tumors from the Pan-Cancer TCGA Dataset using a novel algorithmic approach (SELECT algorithm) identified somatic mutations in RBM10 as the top-scoring KRAS co-occurrence motif in both NSCLC and colorectal adenocarcinoma⁶³. RBM10 encodes a splicing regulator that is involved in cellular growth control via regulation of NOTCH signaling^{106, 107}. In vivo depletion of Rbm10 in mice using CRISPR/Cas9- mediated gene editing concurrently with activation of endogenous oncogenic Kras^G12D confers a modest fitness advantage that is lost when Trp53 or Lkb1 are also inactivated¹⁰⁸. The precise phenotypic consequences of RBM10 inactivation in NSCLC and the mechanisms that underpin its oncogenic cooperation with KRAS remain incompletely understood.

Somatic genomic alterations in CDKN2A [encoding the p16 and p14^ARF (p19^ARF in the mouse) tumor suppressors] and CDKN2B (encoding p15) are observed in ~20% and ~12% of metastatic KRAS-mutant NSCLC respectively and bi-allelic loss of the CDKN2A/CDKN2B locus is a hallmark of the KC subgroup^{15, 67}. KC tumors are characterized by lack of NKX2–1 expression and frequent activation of a gastrointestinal transcriptional program (manifesting histologically as invasive mucinous carcinoma in some cases), enrichment for the Kras^G12D mutation and poor prognosis. Several of these features are recapitulated in mice where endogenous expression of oncogenic Kras^G12D is coupled with bi-allelic deletion of Cdkn2a/Cdkn2b, resulting in concurrent abrogation of p16, p19^ARF and p15. Both isolated Cdkn2a loss (leading to inactivation of p16 and p19^ARF) as well as combined Cdkn2a/Cdkn2b inactivation accelerate Kras^G12D-driven lung carcinogenesis and promote loco-regional metastatic spread but combined loss of p16, p19^ARF and p15 elicits a more marked phenotype than Cdkn2a loss alone with enhanced cellular proliferation, frequent loss of NKX2–1 expression and up-regulation of the embryonal protein HMGA2, increased burden of poorly differentiated, high-grade tumors, enhanced metastatic proclivity and curtailed survival¹⁰⁹. Mechanistically, loss of NKX2–1 unleashes a hepatocyte nuclear factor 4-alpha (HNF4A)-driven gastric differentiation program whereas concomitant loss of HNF4A promotes de-repression of HMGA2¹¹⁰.

EGFR-mutant LUAD

Although EGFR –mutant tumors represent the prototypical oncogene-addicted LUAD subgroup that spearheaded adoption of the single-driver model, the overwhelming majority of EGFR-mutant lung tumors harbor one or more co-mutations, even when the analysis is limited to pre-defined sets of cancer-relevant genes within established panels (FoundationOne, Guardant360, MSK-IMPACT)^{15, 59, 72, 73}. The spectrum of enriched genomic co-alterations in advanced EGFR-mutant LUAD is dominated by recurrent mutations in a core set of genes including TP53 (54.6% % - 64.6%), RB1 (9.6%−10.33%), CTNNB1 (which encodes β-catenin; 5.3%−9.6%), and PIK3CA (9%−12.4%) as well as amplifications involving EGFR itself (22% - 25.5%), NKX2–1 (12.2% - 16.7%), CDK4 (7%−10%), CDK6 and CCNE1^{15, 72, 73} (Figure 2). The spectrum and prevalence of co-mutations does not appear to vary depending on the specific initiating EGFR mutation and is similar across the three most common subtypes (EGFR exon 19 deletion, EGFR^L858R and EGFR exon 20 insertions)⁴⁷. Prior therapy is associated with increased average number of co-alterations. Mutations in PIK3CA and CTNNB1 are more frequent in advanced stage tumors compared with early stage LUADs, pointing towards functional roles in malignant progression and metastasis, whereas alterations in TP53 (62.5%), RB1 (9.5 −12.5%) and NKX2–1 (12.5%) appear to occur with comparable frequencies in early- and advanced-stage tumors^{15, 34, 72, 73}.

Somatic mutations in TP53 represent by far the most prevalent co-alteration in EGFR-mutant LUAD (54.6%−64.5%) and their clinical significance has been evaluated in several studies. TP53 mutations are mostly truncal events (present in all geographically distinct segments of the tumor) that occur early during tumor evolution and prior to whole genome doubling, and are frequently accompanied by truncal loss of heterozygosity at the TP53 locus, indicating strong selective pressure for complete TP53 inactivation in early stage LUADs¹⁰³. Furthermore, tumors bearing co-mutations in TP53 exhibit higher degrees of copy number genomic instability (aneuploidy), and a higher somatic mutation burden, both on the trunk and in the branches of the tumor phylogenetic tree¹⁰³. Therefore, TP53 co-mutations impact the natural history of EGFR-mutant NSCLC at least partially by allowing tolerance of a greater degree of genomic instability that results in both larger numbers of co-occurring truncal drivers as well as late sub-clonal diversification with focal emergence of high amplitude amplifications and deletions in mediators of therapeutic resistance¹⁰³. In keeping with a more complex genomic landscape and a larger burden of clonal or sub-clonal co-drivers, multiple clinical studies have identified TP53 co-alterations as a negative prognostic marker in EGFR-mutant LUAD and a consistent predictor of worse clinical outcomes following EGFR TKI therapy^{72, 73, 111–116}.

Mutational inactivation of RB1 is a clonal and early genetic event in 9.5%−12.5% of EGFR-mutant LUAD^{72, 73, 103, 117}. The majority of RB1-mutant tumors also harbor TP53 co-alterations, underscoring the critical contributions of these archetypal tumor suppressor genes to cell cycle control. TP53 and RB1 co-mutations mark the earliest ancestors of EGFR-mutant LUAD that transform to small cell carcinoma following exposure to EGFR TKIs and dramatically increase the risk of small cell transformation, although loss of RB1 is insufficient to directly induce neuroendocrine trans-differentiation.^117–119. Alterations in other regulators of G1/S cell cycle transition including amplification of CDK4, CDK6 and CCNE1 are prevalent and appear to be enriched in tumors that express the EGFR^T790M gatekeeper mutation that confers TKI resistance, although data regarding their preferential occurrence in EGFR-mutant compared to EGFR wild-type LUAD are less consistent between studies⁷³. Similarly, genomic alterations –most commonly deletion events - in the CDKN2A and CDKN2B genes are observed in ~24.6% and 20.2% of EGFR-mutant tumors and these alterations are typically truncal, further underscoring the significance of G1/S checkpoint dysregulation in the early stages of lung carcinogenesis driven by mutant EGFR^{72, 73, 103}.

Activating mutations in CTNNB1 represent one of the most consistently co-selected alterations in EGFR-mutant LUAD across different studies. CTNNB1 mutations are rare in early-stage EGFR-mutant LUAD (1.8% in the TCGA cohort) but their prevalence increases in late-stage tumors (5.3% - 9.6%), in agreement with earlier studies that identified a central role for WNT signaling in LUAD metastasis and experimental data demonstrating increased invasive potential of EGFR and CTNNB1 co-mutated NSCLC cells in vitro^{15, 34, 72, 73, 120, 121}. However, in a mouse LUAD model driven by compound Egfr^L858R/T790M mutations genetic deletion of Ctnnb1 reduced tumor burden indicating non-redundant functions in tumor initiation¹²². Interestingly, mutations in CTNNB1 have been reported to occur more frequently in LUAD with the EGFR^T790M mutation following exposure to first or second generation EGFR TKIs suggesting enhanced genetic interaction in this setting⁷³. Mutant EGFR has further been shown to directly tyrosine phosphorylate β-catenin resulting in its stabilization and nuclear accumulation¹²⁰.

PIK3CA mutations, including classical kinase (H1074R and H1074L) and helical (E545K and E542K) domain mutations are observed in 9%−12.4% of advanced stage EGFR-mutant LUAD and, like CTNNB1 mutations, are encountered preferentially in advanced-stage tumors^{15, 34, 72, 73}. In vitro, co-occurring PIK3CA mutations promote cellular invasion and migration whereas in vivo they are associated with worse overall survival in some studies but do not appear to impact response rates and progression-free survival with first or second line EGFR TKI therapy^{73, 115, 123}.

NKX2–1 amplification is significantly enriched in EGFR-mutant LUAD and constitutes a classical example of a context-dependent genetic interaction. In mouse models of Kras-mutant LUAD, Nkx2–1 loss fosters metastasis and Nkx2–1 haplo-insufficiency promotes both initiation and progression of invasive mucinous adenocarcinomas¹²⁴; therefore, in this genomic background Nkx2–1 functions as a tumor suppressor gene. In contrast, hemizygous Nkx2–1 loss suppresses Egfr^L858R-driven lung carcinogenesis, indicating that sustained NKX2–1 expression is essential for tumor initiation downstream of mutant Egfr¹²⁵. Mechanistically, NKX2–1 transactivates the receptor ROR1, which directly binds to EGFR and sustains EGFR–ERBB3 heterodimerization, ERBB3 phosphorylation and pro-survival PI3K–AKT signaling; in addition, ROR1 can interact with and phosphorylate SRC, providing a parallel pathway to AKT activation¹²⁶. Thus, the function of NKX2–1 as a lineage survival oncogene in EGFR-mutant NSCLC provides a plausible explanation for its preferential amplification in this oncogenic subgroup.

ALK, ROS1, RET and other oncogenic fusion-driven molecular subgroups

Recent studies have also begun to shed light on the co-mutation landscape and genomic architecture of LUAD driven by oncogenic fusions^{15, 127}, although the clinical significance of co-alterations in this setting is less well characterized. Interestingly, advanced-stage ALK-rearrangement -positive LUAD are enriched in somatic alterations in CDKN2A (32.5%) and CDKN2B (26.5%), but are less likely to harbor TP53 alterations (23.8%−26.5%) compared with other driver subgroups^{15, 128}. TP53 co-mutations promote genomic instability and are an independent negative prognostic factor in ALK-re-arrangement-positive LUAD, regardless of the type of systemic therapy used^128–130. The prevalence of additional co-mutations in this group is low,¹²⁸ and both the rarity of co-drivers and strong addiction to the ALK fusion oncoprotein may account for the long progression-free survival observed in patients with ALK-rearrangements with the potent and selective second and third generation ALK inhibitors alectinib, brigatinib and lorlatinib^{23, 26, 27}. Similarly to LUAD with ALK fusions, both RET and ROS1 fusion-positive LUAD are characterized by high rates of concurrent CDKN2A loss (29.8% and 30.4% for RET and ROS1-rearranged tumors respectively) and CDKN2B loss (25% and 17.7% respectively) and relative paucity of TP53 mutations, although the frequency of TP53 mutations appears to be somewhat higher compared to ALK-rearranged tumors (34.6%−45.5% for RET-and 45.6% for ROS1- rearranged tumors)^{15, 131}. The key finding that TP53 somatic mutations are underrepresented across LUAD driven by different oncogenic fusions was validated in a subsequent study that further identified frequent bi-allelic SETD2 deletions in this group¹²⁷. The functional consequence of these associations is currently incompletely understood. The co-alteration spectrum of LUAD driven by NRG1 or NTRK1 fusion events has not been elucidated to date.

Other oncogenic subgroups.

A distinct pattern of co-occurring alterations is observed in LUAD driven by MET exon 14 skipping mutations. Specifically, these tumors are characterized by highly significant enrichment of MDM2 and CDK4 amplification (41.6%) compared with other driver oncogenes, as well as amplification of MET itself¹⁵. In contrast, mutations in TP53 (33.57%) are under-represented, whereas loss of CDKN2A (24.1%) and CDKN2B (17.5%) occur with similar frequencies to that in the overall population of patients with NSCLC¹⁵. The spectrum of co-occurring alterations in BRAF-mutant NSCLC mirrors the background frequency of alterations in TP53 (53.3%), LKB1 (16.2%), ATM (5.8%), NF1 (6.9%), PIK3CA (6.6%), KEAP1 (6.6%), MYC (10.8%), NKX2–1 (7.3%), although alterations in RB1, MDM2, CDKN2A (16.6%) and CDKN2B (11.2%) are less frequent within this molecular subgroup¹⁵. Finally, patients with ERBB2-mutant NSCLC exhibit preferential amplification of NKX2–1 (19.4%) and ERBB2 itself (14.4%) as well as frequent mutations in RB1 (8.9%), but the frequencies of co-mutations in TP53 (51.7%), CDKN2A (27.2%), CDKN2B (17.2%), PIK3CA (5%), CTNNB1 (4.4%) and MDM2 amplification (7.2%) are similar to that observed in the overall population of patients with NSCLC¹⁵.

Effects on the immune microenvironment

In addition to their impact on cell-autonomous cancer hallmarks, co-mutations can also shape the NSCLC microenvironment and determine its immune contexture (Figure 3). Inactivating LKB1 genomic alterations, present in ~25% of KRAS-mutant LUAD, have emerged as a major driver of the cold, non-T cell-inflamed microenvironment in NSCLC, characterized by paucity of infiltrating CD3+, CD4+ and CD8+ T-cells and low tumor cell expression of PD-L1, despite intermediate to high tumor mutational burden (TMB)^{67, 68, 132–134}. These findings are recapitulated in the Kras^LSL-G12D/+; Lkb1^Fl/Fl GEMM, where Cre-mediated Lkb1 ablation triggers marked influx of tumor-associated neutrophils with T cell suppressive properties including increased expression of Arginase 1 (ARG1) and Interleukin 10 (IL-10)¹³⁵. Mechanistically, Lkb1 loss in this model results in altered tumor cytokine milieu with increased expression of interleukin 1β (IL-1β), IL-6, CXCL7 and G-CSF that foster myeloid cell recruitment¹³⁵. In addition, LKB1 inactivation induces epigenetic repression of STING (also known as TMEM173), thus promoting insensitivity to cytosolic dsDNA accumulation¹³⁶. Silencing of STING in this context is triggered by enhanced activity of the EZH2 and DNMT1 methyltransferases due – at least in part- to increased production of SAM through diversion of glucose towards the serine biosynthetic pathway in LKB1-deficient cells^{87, 136}. Increased expression of MYC has also been observed following LKB1 loss¹³⁷ and may provide an additional mechanistic clue to the immune inert phenotype of LKB10-deficient NSCLC because IL-23 and CCL9-mediated inflammation and exclusion of B cells, T cells and NK cells have been reported to underpin the strong oncogenic cooperation between KRAS and MYC in lung cancer pathogenesis¹³⁸. Finally, LKB1 inactivation has also been reported to impinge on non-immune components of the microenvironment of Kras^G12D-mutant mouse tumors, including increased collagen deposition as a result of elevated lysyl oxidase (LOX) expression and effects on angiogenesis^{139, 140}.

Figure 3. Impact of co-mutations on the microenvironment of KRAS-mutant lung adenocarcinoma.

Schematic representation of co-mutation-associated changes in the immune and non-immune microenvironment of KRAS-mutant lung adenocarcinoma (LUAD). (a) LKB1 inactivation promotes epigenetic suppression of STING and insensitivity to cytosolic DNA that accumulates in the cytoplasm of KRAS- and LKB1-mutant (KL) cells due to dysfunctional mitochondria¹³⁶. KL tumors are further characterized by a pro-inflammatory cytokine milieu with accumulation of immunosuppressive neutrophils, marked paucity of CD4+ and CD8+ T-cells and evidence of T-cell exhaustion^{68, 135}. The potential contributions of immune cell metabolic restriction, altered angiogenesis and acidification of the tumor microenvironment (highlighted in blue) to the immune-inert phenotype of KL tumors remain as yet unexplored, but represent plausible directions for future study. (b) MYC fosters immune evasion of murine Kras^G12D- driven LUADs through IL-23- mediated expulsion of T, B and NK cells and CCL9-mediated macrophage recruitment and secretion of immunosuppressive VEGF¹³⁸. (c) KEAP1 mutations, which frequently co-occur with mutations in LKB1, particularly in the context of KRAS-mutant LUAD, have also been associated with low intra-tumoral density of infiltrating T- and B- lymphocytes, although the possible role of KEAP1 loss on NK cell infiltration remains unclear¹⁴¹. Stabilization of NRF2 as a result of KEAP1 inactivation may further promote reduced expression of STING through post-transcriptional regulation¹⁴². (d) Finally, somatic TP53 mutations have been shown to mediate NF-κB pathway activation in Kras-mutant murine models of LUAD¹⁴⁶. Although TP53 mutations have been associated with reduced production of chemokines required for the recruitment of NK and T cells in some models and human tumors, in the context of KRAS-mutant LUAD TP53 co-alterations promote an inflamed tumor immune microenvironment with increased production of interferon γ (IFNγ) and increased expression of PD-L1 on the surface of tumor cells^{67, 68, 186}.

Inactivating mutations in KEAP1 have also been associated with an altered NSCLC immune microenvironment¹³⁴. In a conditional GEMM of LUAD (Keap1^Fl/Fl;Pten^Fl/Fl), co-deletion of Keap1 and Pten resulted in immunologically cold tumors, akin to Lkb1-mutant NSCLC¹⁴¹. Interestingly, NRF2 was recently identified as a negative regulator of STING expression via effects on STING mRNA stability¹⁴² thus suggesting a tantalizing mechanistic connection between the effects of KEAP1 and LKB1 inactivation that warrants further study. Additional immune phenotypes may be uniquely associated with KEAP1 loss; for example, increased peri-tumoral accumulation of natural killer (NK) cells in KEAP1-mutant tumors was reported in a cohort of surgically resected early-stage LUAD¹³².

Finally, TP53 co-mutations are associated with an inflamed tumor immune microenvironment and increased tumor cell PD-L1 expression in KRAS-mutant NSCLC and GEMMs. This is at least in part due to activation of the nuclear factor κB (NF-κB) pathway driven by p53 loss, as well as increased tolerance of a higher mutational burden that may ostensibly result in enhanced immunogenicity due to increased neoantigen load^{143, 144, 145, 146}.

The impact of co-mutations on other oncogene-driven subgroups of NSCLC, including those driven by EGFR mutations, ALK, ROS1 and RET translocations, as well as ERBB2 and MET exon 14 skipping mutations has not hitherto been determined and represents an area of active investigation. This will be particularly pertinent for BRAF-mutant NSCLCs, which are characterized by high tumor cell PD-L1 expression and more favorable clinical response to PD-1 and PD-L1 inhibitors^{147, 148}.

Effects on drug sensitivity

Large-scale efforts aimed at linking tumor genomic alterations with sensitivity to cytotoxic and targeted therapies have uncovered a wealth of pharmacogenomic interactions in NSCLC and other cancer types^149–152. These seminal high-throughput studies yielded multiple novel associations but also highlighted challenges in therapeutic response modeling that underscore the genomic complexity and biological heterogeneity of cancer. Interestingly, logic models – generated using the LOBICO (“Logic Optimization for Binary Input to Continuous Output”computational approach- that combine multiple input features such as mutations in cancer genes, gene fusions, recurrent copy number aberrations and binarized pathway activity scores (derived from gene expression profiling outperform single-gene models for prediction of drug sensitivity^{153, 154}. Thus, co-occurring alterations can function as robust, and in many settings more precise, biomarkers of therapeutic response than single-gene predictors.

Chemical and genetic screens in panels of molecularly annotated NSCLC cell lines as well as candidate target approaches have uncovered several KRAS co-mutation-driven molecular dependencies and collateral vulnerabilities. KL NSCLC cell lines are characterized by unique sensitivity to depletion of multiple components of the coatomer 1 (COPI) complex and pharmacological inhibition of lysosomal acidification (for example by exposure to bafilomycin A) as a result of critical dependence on lysosomal macromolecule degradation for supply of TCA cycle substrates¹⁵⁵. Other studies have linked LKB1 loss with enhanced sensitivity to energetic stress triggered by the biguanides metformin and phenformin¹⁵⁶ or the combination of phenformin with the mTOR inhibitor MLN0128¹⁵⁷, as well as to endoplasmic reticulum stress induced by 2-deoxy-D-glucose¹⁵⁸. Enhanced dependence on nucleotide (and especially dTTP) synthesis further underpins the selective sensitivity of KL cells to deoxythymidylate kinase (DTYMK) depletion and to combined treatment with gemcitabine - a deoxycytidine analog that inhibits DNA synthesis and further depletes dNTP pools by targeting ribonucleotide reductase¹⁵⁹- and CHK1 inhibitors, that abrogate the CHK1-mediated checkpoint response to replicative stress^{160, 161}. Furthermore, LKB1 deficient cells are selectively vulnerable to inhibition of the ATP1A1 Na+/K+-ATPase by cardiac glycosides¹⁶² and to several structurally distinct inhibitors of the HSP90 family of molecular chaperones⁶⁷. While some of these vulnerabilities are associated with LKB1 inactivation irrespective of concurrent KRAS mutations, others, such as addiction to lysosomal enzymatic degradation, appear to be specific to the KL oncogenotype and thus represent de facto co-mutation-dependent vulnerabilities. In contrast, KL lung tumors exhibit resistance to MEK inhibitors in mouse models and LKB1 deficiency by immunohistochemistry is associated with lack of benefit from the addition of the vascular endothelial growth factor A (VEGFA) inhibitor bevacizumab to platinum doublet chemotherapy^{163, 164}. Beyond the KL genotype, KRAS and KEAP1 co-altered NSCLC cells display chemically tractable selective dependence on GLUT8-mediated uptake for effective diversion of glucose towards the serine biosynthetic pathway¹⁶⁵ and rely on glutaminolysis for TCA cycle anaplerosis; thus, they are selectively sensitive to glutaminase inhibition in both cell line and mouse models⁹². In agreement with the role of NRF2 as a transactivator of antioxidant as well as phase II detoxifying and cytoprotective enzymes that mediate resistance to cytotoxic chemotherapy, KEAP1 co-mutations are associated with resistance to multiple inhibitors of oncogenic kinases within the receptor tyrosine kinase–MAPK pathway in vitro¹⁶⁶ and significantly worse clinical outcomes with platinum-doublet chemotherapy in KRAS-mutant LUAD⁹⁹ or, as shown in preliminary data, with chemo-immunotherapy using pemetrexed-carboplatin (or cisplatin) plus the PD-1 antibody pembrolizumab in non-squamous NSCLC¹⁶⁷. Finally, ATM co-alterations increase the sensitivity of KRAS and BRAF-mutant NSCLC cell lines to MEK inhibitor-induced apoptosis and genetic deletion of Atm is associated with increased sensitivity to poly(ADP-ribose) polymerase (PARP) and ATR inhibitors as well as to radiation therapy in mouse models of Kras-mutant LUAD^{168, 169}. Notably, despite apparent mutual exclusivity between classical activating mutations in RAS pathway genes, oncogenic co-operativity has been observed between atypical, weakly activating mutations¹⁷⁰. In this setting signaling inputs from multiple co-altered RAS pathway genes coordinately contribute towards thresholds of oncogenic activity that are critical for transformation and tumor maintenance but also bestow therapeutic vulnerabilities; for example, co-mutations in NF1 and RASA1, encoding two critical RAS pathway GTPase-activating proteins (GAPs), drive addiction to the MEK–ERK signaling axis and confer enhanced sensitivity to MEK inhibitors in a subset of both LUAD and LUSC^{171, 172}.

Of particular relevance is the impact of co-occurring alterations on clinical outcomes with EGFR and ALK TKIs as well as other targeted therapies. TP53 co-mutations have consistently been associated with shorter progression-free survival following upfront treatment with 1^st or 2^nd generation EGFR TKIs and there is further evidence that they adversely impact clinical outcomes with the third generation, mutant-selective EGFR TKIs, in patients whose tumors have acquired the EGFR^T790M gatekeeper mutation^{72, 115}. In a study of 200 EGFR-mutant patients with extensive molecular profiling at baseline, pre-existing MET (present in 2% of cases) or ERBB2 (4% prevalence) amplification were also associated with significantly shorter progression-free survival with first-line 1^st or 2^nd generation EGFR TKI therapy, whereas among patients with acquired EGFR^T790M mutation from a distinct cohort, co-mutations in RB1 and PTEN and amplification of MDM2 were independently associated with worse progression-free survival following treatment with 3^rd generation EGFR TKIs^{72, 173}. Co-alterations in BRAF, CDKN2A, CDKN2B, fibroblast growth factor receptor 3 (FGFR3) and amplification of MET and EGFR itself were all enriched in patients with acquired resistance to EGFR TKIs compared to pre-treatment tumors, indicating roles in mediating the drug resistant phenotype. Interestingly, co-mutations in PIK3CA don’t impact response to first, second or third generation EGFR TKIs^{113, 123, 174}. Importantly, co-occurring clonal alterations in both TP53 and RB1, present in ~9 % of EGFR-mutant LUAD at baseline, substantially increase the risk of transformation to small cell carcinoma upon treatment with EGFR TKI; therefore co-occurring alterations can affect not only the likelihood and duration of response to targeted therapy but also impact mechanisms of acquired resistance¹¹⁷. It is currently unknown whether the likelihood of acquisition of an EGFR^T790M secondary resistance mutation can also be influenced by the co-mutation status of the tumor.

In keeping with their prominent role in shaping tumor immunobiology and immune contexture, co-occurring genomic alterations can further impact clinical response to immune checkpoint inhibitors. This is particularly evident in KRAS-mutant LUAD. Inactivating somatic mutations in LKB1, present in ~25% of KRAS-mutant LUAD, have emerged as a major genomic driver of primary resistance to PD-1 and PD-L1 inhibition, despite KL LUAD harboring intermediate to high TMB⁶⁸. Importantly, the negative impact of LKB1 genomic alterations on clinical outcomes with anti-PD1 or anti-PD-L1 therapy extends to PD-L1 positive tumors⁶⁸. Therefore, somatic genomic alterations may represent independent predictors of clinical outcomes with immune checkpoint inhibitors, in addition to previously established markers such as PD-L1 expression and TMB. De novo resistance to immune checkpoint blockade following LKB1 loss is further associated with primary resistance to combined anti-PD-1 and anti-CTLA4 therapy with nivolumab and ipilimumab¹⁷⁵. In contrast, KRAS-mutant tumors bearing co-mutations in TP53 exhibit high rates of clinical response to PD-1 axis immunotherapy and markedly improved progression-free and overall survival compared to KL⁶⁸. In addition to LKB1, co-mutations in KEAP1 have also been implicated in de novo resistance to PD-1 blockade⁹⁹ and both LKB1 and KEAP1 are associated with inferior clinical outcomes with chemo-immunotherapy with pemetrexed-carboplatin (or cisplatin)-pembrolizumab, particularly among PD-L1-positive and TMB-high tumors¹⁶⁷. In this context, double LKB1;KEAP1 mutant tumors exhibita particularly recalcitrant clinical response phenotype¹⁶⁷. Finally, mutations in PTEN have also been nominated as a candidate driver of primary resistance to immune checkpoint inhibition in NSCLC, in agreement with similar reports in melanoma^{175, 176}.

Conclusions and perspectives.

As our understanding of the genomic landscape of NSCLC deepens, broad tumor genomic profiling becomes increasingly accessible and our therapeutic armamentarium continues to evolve, there is growing appreciation that the current single oncogenic driver model fails to adequately capture the clinical complexity of NSCLC and warrants revision. Co-occurring genomic alterations in oncogenic drivers and tumor suppressor genes have emerged as major tenets of the molecular diversity of NSCLC. Antecedent knowledge of a key set of major co-mutations may therefore allow more granular insights into NSCLC biology; facilitate development of improved clinical response prediction algorithms; anticipate and forestall the emergence of acquired resistance; and enable development of novel, highly personalized therapeutic approaches in the next wave of precision oncology clinical trials. Based on accumulated and emerging evidence we propose a next-generation, dynamic model for the molecular classification of NSCLC that encompasses the molecular and clinical diversity affected by co-mutations (Figure 4). Immediate priorities and challenges for the future are to catalog, functionalize and systematically evaluate the therapeutic utility of the full spectrum of co-occurring alterations in NSCLC and, simultaneously, to expeditiously translate the most robust and critical insights into more precise therapeutic strategies that yield improved clinical outcomes for NSCLC patients. These tasks will require novel computational tools and high throughput in vivo platforms as well as large, prospectively assembled collaborative clinical datasets and efficient and flexible umbrella clinical trial [G] designs.

Figure 4. Next-generation model for the molecular stratification of lung adenocarcinoma.

Oncogenic subgroups of lung adenocarcinoma (LUAD) are divided into smaller subsets on the basis of key co-occurring genomic alterations. Co-mutations constitute major determinants of tumor molecular diversity and can impact both tumor cell-autonomous and non-cell-autonomous cancer hallmarks; determine prognosis; predict response to systemic therapies and influence mechanisms of innate and acquired resistance. For simplicity, only KRAS and EGFR co-alterations are depicted graphically. For KRAS-mutant LUADs the previously identified KL, KP, and KC transcriptome-based subgroups are also indicated⁶⁷; co-mutations in LKB1, KEAP1 and ATM are significantly enriched in the KL subgroup, whereas co-occurring alterations in TP53 and bi-allelic inactivation of CDKN2A/CDKN2B are hallmarks of the KP and KC subgroups respectively. Co-mutations in RBM10 don’t appear to exhibit predilection for any of the three KRAS transcriptomic subgroups. It should therefore be noted that several of the reported co-alterations within oncogene-defined groups are not mutually exclusive. Although co-mutation-defined cohorts are represented as slices of equal size, both the spectrum and prevalence of individual co-mutations evolve according to disease stage, prior treatment exposures, immune editing and the mutational processes that are operational at each stage of carcinogenesis.

Glossary

Platinum-doublet chemotherapy

Cisplatin or carboplatin-based combinations with a second chemotherapeutic agent, most commonly pemetrexed (LUAD), taxanes (LUAD or LUSC) and gemcitabine (LUSC)

Objective response

Measurable decrease in tumor burden of a predefined amount in response to therapy

Complete responses

The disappearance of all signs of cancer in response to treatment, including both target and non-target lesions (with reduction of all lymph nodes to <10mm in short axis), without emergence of any new lesions

A less strict pattern of mutual exclusivity in which combinations of somatic mutations in different genes occur less frequently than expected but do not result in toxic functional effects and can therefore rarely co-occur in the same cell

Cold tumor immune microenvironment

Tumor microenvironment characterized by lack or paucity of infiltrating T cells

Trans-differentiation

Conversion of one differentiated somatic cell type to another without passage through an intermediate pluripotent or progenitor cell state

Umbrella clinical trial

A clinical trial that assesses multiple targeted therapeutic strategies in a single cancer type