Tumors acquire alterations in oncogenes and tumor suppressor genes in an adaptive walk through the fitness landscape of tumorigenesis. However, the features of this landscape remain poorly understood and cannot be revealed by human cancer genotyping alone. Here, we use a multiplexed, autochthonous mouse platform to model and quantify the initiation and growth of more than one hundred genotypes of lung tumors across four oncogenic contexts: KRAS G12D, KRAS G12C, BRAF V600E, and EGFR L858R. The resulting fitness landscape is rugged (the effect of tumor suppressor inactivation often switches between beneficial and deleterious depending on the oncogenic context), shows no evidence of diminishing-returns epistasis within variants of the same oncogene, and is inconsistent with expectations of a simple linear signaling relationship among these three oncogenes. Our findings suggest that tumor suppressor effects are strongly context-specific, which limits the set of evolutionary paths that can be taken through the fitness landscape.

Cancer progression is a quintessential example of a walk on an adaptive fitness landscape, with tumor growth depending on the cooperation of multiple driver mutations. While sequencing of tens of thousands of clinical samples has revealed a vast set of cancer drivers, far less is known about the interactions of oncogene-tumor suppressor pairs. Due to patient-level selection bias, confounding biological factors and limited sample sizes, a map of these interactions cannot be generated from human data alone and, instead, requires direct perturbational experiments and functional genomics approaches. Here, we model and quantify tumors using an autochthonous mouse platform and Tubaseq, which integrates barcoded lentiviral-sgRNA/Cre vectors and high-throughput barcode sequencing to uncover the number of neoplastic cells in each tumor of each genotype. Across four oncogenic contexts—KRAS G12D, KRAS G12C, BRAF V600E, and EGFR L858R—we analyzed >10,000,000 tumors to estimate the tumorigenesis potential of the oncogene alone as well as the effect of inactivating 28 tumor suppressor genes. We discovered that despite KRAS G12D producing >10x the tumors as G12C, tumor suppressor inactivations provide similar growth effects on both backgrounds. In contrast, although the intrinsic abilities of BRAF V600E and EGFR L858R to drive tumor development fall within the range of the KRAS variants, tumor suppressive effects are categorically different in the context of each oncogene. Many tumor suppressors show clear sign epistasis with the oncogenes, whereby inactivation is advantageous in one context and neutral or deleterious in another. Inactivation of some of the strongest tumor suppressors (e.g., Lkb1, Setd2, and Kmt2d) in KRAS-driven tumors strongly decreases tumor growth in the presence of oncogenic EGFR. While some of these epistatic effects are consistent with a textbook understanding of the RAS pathway, most cannot be predicted based on the linear oncogenic EGFR → KRAS → BRAF pathway model. Analyses of clinical genomics data from AACR Project GENIE confirm that high rates of passenger mutations in KRAS- and BRAF-driven lung tumors, among other factors, prevent the discovery of these interactions from human data alone. However, for EGFR-mutant lung cancers, which are less confounded by high mutational burden, the rates of coincident tumor suppressor mutation are highly correlated with tumor growth effects in our in vivo model. Thus, we find via causal experiments that the landscape of tumor suppression is highly dependent on oncogenic context, with a minority of tumor suppressive effects robust to changes in the oncogene. These findings suggest that the utility of a specific cancer mutation as a prognostic or predictive biomarker of patient outcomes will be dependent on coincident mutations in the tumor and highlight the utility of high-throughput, quantitative autochthonous mouse models in advancing our understanding of cancer biology. Citation Format: Lily M. Blair, Joseph M. Juan, Lafia Sebastian, Vy B. Tran, Wensheng Nie, Gregory D. Wall, Mehmet Gerceker, Ian K. Lai, Edwin A. Apilado, Gabriel Grenot, David Amar, Giorgia Foggetti, Mariana Do Carmo, Zeynep Ugur, Debbie Deng, Alex Chenchik, Maria Paz Zafra, Lukas E. Dow, Katerina Politi, Jonathan J. MacQuitty, Dmitri A. Petrov, Monte M. Winslow, Michael J. Rosen, Ian P. Winters. Oncogenic context shapes the fitness landscape of tumor suppression [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 1172.

Cancer is characterized by diverse genomic alterations that change cell state and can modify therapy responses. In some cases, the relationships between genotype and drug responses are obvious, such as the response of tumors with defined oncogenic mutations to inhibitors of those mutant proteins. However, covalent KRAS G12C inhibitors (G12Ci) induce clinical responses in less than half of patients with oncogenic KRAS G12C-driven lung cancer, illustrating that genomic alterations not obviously related to the drug mechanism can have a dramatic impact. We have developed and optimized an in vivo platform to identify pharmacogenomic interactions that dictate lung cancer responses to therapies. By coupling somatic genome editing, tumor barcoding, ultra-deep barcode sequencing, and robust statistical methods in genetically engineered mouse models of human lung cancer, we can quantify and compare candidate therapies across thousands of clonal tumors of diverse tumor suppressor genotypes. This platform provides a uniquely high-throughput and quantitative assessment of tumors that are initiated de novo within the natural immunocompetent tissue environment. To uncover genotypes that are particularly important in controlling the response of tumors to oncogenic KRAS inhibition, we applied this platform to quantify the impact of 59 tumor suppressor genes on KRASG12C-driven lung tumor responses to G12Ci. Treatment resulted in approximately three-quarters reduction in both average tumor sizes and overall tumor burden, with a clear drug and dose dependence. Approximately one-quarter of inactivated genes significantly and consistently altered tumor responses across numerous G12Ci and replicate studies. Interestingly, most of these genes are thought to be in pathways not directly related to RAS signaling, and they exhibit a striking pattern in which treatment sensitivity is positively correlated with strength of the tumor suppressor effect. This pharmacogenomic profile for G12Ci was categorically different than the profiles we have observed for chemotherapy and other therapies. Overall, these results provide direct causal evidence that certain tumor suppressor genotypes dramatically shift the effectiveness of G12Ci in vivo and generate hypotheses about patients likely to benefit from G12Ci relative to alternatives. Analysis of available human data provides early clinical support for some of these hypotheses. Ongoing efforts to define pharmacogenomic profiles of combination therapies could be of even greater importance. Platforms that can accurately predict how tumor genotype drives responses have the potential to transform precision cancer therapy, enabling more effective patient stratification and therapy combinations. Citation Format: Paul J. McMurdie, Ian P. Winters, Lily M. Blair, Lafia Sebastian, Vy Tran, Gabriel Grenot, Edwin A. Apilado, Edwin A. Apilado, Ian K. Lai, Gregory D. Wall, Dmitri A. Petrov, Monte M. Winslow, Michael J. Rosen, Joseph Juan. Tumor suppressor genotype dramatically impacts lung cancer response to KRAS G12C inhibitors in vivo [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 2789.

Cooperative interactions between genomic alterations in oncogenes and tumor suppressors are a hallmark of cancer. While molecular characterization of tens of thousands of primary tumors has solidified estimates of how frequently specific genomic alterations occur, identifying genetic interactions from co-mutation rates alone remains limited by the vast number of possible combinations as well as inherent biases and confounding effects. Conversely, genetically engineered mouse models can reliably uncover causal effects of specific genetic alterations but are not sufficiently high throughput to generate broad insights into the interplay between genetic alterations in tumors. Therefore, how the effects of one alteration on tumor growth change in the context of other alterations remains largely unknown. We previously addressed these technical limitations by combining CRISPR/Cas9-mediated gene editing, tumor barcoding, and ultra-deep barcode sequencing methods with genetically engineered mouse models of human lung cancer, which together enable multiplexing of tumor genotypes in individual mice and precise, high-throughput quantification of in vivo tumor fitness. Here, we leverage this platform to systematically quantify the growth effects of over 45 known and putative tumor suppressor gene alterations across different common oncogenic point mutations in Kras and Braf. While the set of genes we identified as tumor suppressive in KrasG12D- and KrasG12C-driven lung cancers was largely consistent, inactivation of those genes typically resulted in larger tumor growth advantages in the KrasG12D model. Inactivation of most genes in BrafV600E-driven lung tumors had a strikingly different growth effect than in tumors with either Kras variant, and effects were generally far less pronounced in the BrafV600E context. Interestingly, loss of genes upstream of Braf within the Ras pathway often had strong effects in the KrasG12D or KrasG12C models—loss of Nf1, a negative regulator of Kras, or loss of the wildtype Kras allele enhanced oncogenic Kras-driven tumor growth, while loss of Shp2, a positive regulator of Kras, was mildly detrimental—but had little or no effects in the BrafV600E model. We assessed translatability of our experimental cause-and-effect data through comparisons to correlative data from publicly available human lung cancer genomics databases, and found that genetic interactions predicted by our models and gene co-mutation rates in patient tumors largely align. Together, these findings underscore the need for a contextual understanding of the ways in which tumors are influenced by their genetics and highlight the utility of high-throughput, quantitative autochthonous mouse models in pursuing this endeavor. Citation Format: Ian Winters, Lily Blair, Lafia Sebastian, Gabriel Grenot, Edwin Apilado, Wensheng Nie, Vy Tran, Ian Lai, Gregory Wall, Dmitri Petrov, Monte Winslow, Joseph Juan, Michael Rosen. Oncogenic context drives the landscape of tumor suppression in lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 2196.