ABSTRACT
Lung cancer is the leading cause of cancer death worldwide, with lung adenocarcinoma being the most common subtype. Many oncogenes and tumor suppressor genes are altered in this cancer type and the discovery of oncogene mutations has led to the development of targeted therapies and better clinical outcomes. However, a large fraction of lung adenocarcinomas lacks mutations in known oncogenes, and the genesis and treatment of these oncogene-negative tumors remain enigmatic. Here, we perform iterative in vivo functional screens with quantitative autochthonous mouse model systems to uncover the genetic and biochemical changes that enable efficient lung tumor initiation in the absence of oncogene alterations. Through the generation of hundreds of diverse combinations of tumor suppressor alterations, we demonstrate that inactivation of suppressors of the RAS and PI3K pathways drive the development of oncogene- negative lung adenocarcinoma. Pathway-level human genomic data analysis and histology identified RAS/MAPK and PI3K pathway activation as a common event in oncogene-negative human lung adenocarcinomas. We demonstrate that oncogene-negative tumors and cell lines with activated RAS/MAPK and PI3K pathways are vulnerable to pharmacological inhibition of these signaling axes. Collectively, these results transform our understanding of this prevalent yet understudied subtype of lung adenocarcinoma.
STATEMENT OF SIGNIFICANCE Many lung adenocarcinomas do not have mutations in known proto-oncogenes, and as a result, targeted therapies are unavailable for treating these patients. Here, we uncover driver pathways in a subset of these oncogene-negative lung adenocarcinomas and demonstrate the therapeutic value of inhibiting these pathways.
INTRODUCTION
Tumors typically contain driver gene alterations in both proto-oncogenes and tumor suppressor genes [1]. The classification of cancers based upon their driver oncogenes has enabled a shift from toxic chemotherapies to less toxic and more effective therapies that often target the oncogenes [2]. While oncogenes have clear roles in tumor initiation and maintenance in many cancer types, a significant fraction of tumors lacks identifiable alterations in proto- oncogenes [3, 4]. Consequently, developing targeted therapies for those tumors remains a major clinical challenge.
Lung cancer is the leading cause of cancer death worldwide [5]. Lung adenocarcinoma, the most frequent subtype of lung cancer, has frequent alterations in receptor tyrosine kinase and RAS/RAF pathway oncogenes, including mutations in EGFR and KRAS [6]. However, approximately 30 percent of lung adenocarcinomas are thought to lack a driving oncogene [4, 7, 8]. While oncogene mutations in a very small fraction of these tumors might have been missed due to technical reasons, recent extensive genomic and transcriptomic studies suggest that neither technical reasons nor the presence of novel oncogenes likely explain this large and clinically significant population of lung cancer patients[4–13]. Thus, despite the diagnosis of more than 150,000 patients per year with oncogene-negative lung adenocarcinomas worldwide, the events that drive the initiation and growth of these tumors remain almost entirely unknown. This lack of knowledge has led to the absence of animal models to study this subtype of lung adenocarcinoma and, more importantly, the lack of drugs to target driver pathway dependencies.
Oncogenes and tumor suppressor genes are parts of broader signaling networks that generate and sustain the biochemical changes that drive tumor initiation and growth [1, 14–16]. Combinatorial alterations in tumor suppressor genes could co-operate to activate pathways driving oncogene-negative lung tumors. Genomic analyses of human lung adenocarcinoma have identified complex patterns of mutations in diverse putative tumor suppressor genes [7]. However, the ability to predict which combinations of genomic alterations drive cancer in the absence of oncogene activation based on human genomic data alone remains challenging. While human genomic data can predict combinations of genomic mutations as likely cancer drivers when the mutations co-occur at very high frequencies [17–20], identifying pathogenic combinations of less frequently mutated genes poses a nearly insurmountable challenge. Furthermore, the large numbers of mutations in lung cancers and non-genomic mechanism that often inactivate tumor suppressor genes, further reduce the ability of human cancer genomic studies to identify combinatorial alterations that activate drive pathways in lung cancer [21–24].
Determining the specific pathways involved in tumor initiation can be aided by functional genomic studies in experimental cancer models [25]. Here, we leverage a quantitative mouse model system to assess the ability of hundreds of combinatorial alterations of tumor suppressor genes, acting across many different signaling pathways, to generate oncogene-negative lung tumors in vivo. We uncover genetic drivers and pathway-level changes that drive lung cancer in the absence of oncogene mutations and leverage these findings to identify therapeutic vulnerabilities.
RESULTS
Clinical features of oncogene-negative and oncogene-positive human lung adenocarcinomas are broadly similar
To better understand the genomics of lung adenocarcinomas that lack oncogene mutations, we analyzed data from The Cancer Genome Atlas (TCGA) and AACR Genomics Evidence Neoplasia Information Exchange (GENIE) [26, 27]. We classified tumors as oncogene- positive if they had high-confidence oncogenic alterations in previously described proto- oncogenes, oncogene-indeterminate if they had alterations of unknown significance (variants of unknown significance, VUS) in known proto-oncogenes, and oncogene-negative if they had no alterations in known proto-oncogenes (Methods). We found that a large fraction (17-18%) of lung adenocarcinomas had no activating alterations in known proto-oncogenes, consistent with previous publications (Figure 1a and S1a) [28–30]. Additionally, 15-27% of lung adenocarcinomas were oncogene-indeterminate and thus lung adenocarcinomas without known oncogene mutations accounted for 32-45% of lung adenocarcinoma cases. Patients with oncogene-negative, oncogene-indeterminate, and oncogene-positive lung adenocarcinomas have broadly similar mutational burden and clinical characteristics (Figure S1b-e).
Combinatorial tumor suppressor gene inactivation enables lung tumor development
Given the widespread genetic and epigenetic alterations in human tumors, genomic analyses have limited ability to delineate causal relationships between genomic alterations and tumor development [1]. To determine whether combinatorial tumor suppressor gene inactivation can drive lung tumor initiation in the absence of oncogene activation, we coupled Cre/loxP-based genetically engineered mouse models and somatic CRISPR/Cas9-based genome editing with tumor barcoding and high-throughput barcode sequencing (Tuba-seq) [31–35]. We used Cre/loxP to inactivate each of five “core” tumor suppressor genes (Trp53, Lkb1/Stk11, Keap1, Nf1, and Pten). These genes operate within diverse pathways and are frequently inactivated in human lung cancers, including oncogene-negative lung adenocarcinomas (Figure S2a-b) [35–38]. We used CRISPR/Cas9 to coincidentally inactivate panels of additional tumor suppressor genes in lung epithelial cells in mice with floxed alleles of each of the “core” tumor suppressors, a Cre-reporter allele (R26LSL-Tom (T) [36]), and a Cre-regulated Cas9 allele (H11LSL-Cas9 (C) [37]).
We transduced Nf1f/f;TC, Ptenf/f;TC, Trp53f/f;TC, Lkb1f/f;TC, Keap1f/f;TC, TC, and T mice with two pre-existing pools of barcoded Lenti-sgRNA/Cre vectors that target ∼50 putative tumor suppressor genes that we have previously studied in KRASG12D-driven lung tumors (Lenti- sgTS15/Cre and Lenti-sgTS102/Cre; (Figure 1b, S2c-d, S3a, and Table S1) [31, 32, 35]. The mutation frequency of these genes varied, and mutations in some were enriched in oncogene- negative human lung adenocarcinomas (Table S1) (Figure S2c-d). The combination of Cre/LoxP and CRISPR/Cas9-based genome editing should generate hundreds of combinations of genomic alterations in lung epithelial cells. We previously found that a small percent of lung tumors initiated with Lenti-sgRNA/Cre vectors in other lung cancer models contained multiple sgRNAs, consistent with the transduction of the initial cell with multiple Lenti-sgRNA/Cre vectors [31, 32]. Thus, a high titer of Lenti-sgRNA/Cre pools in this study increases the potential of finding higher-order interactions, in addition to pairwise interactions, that increase the growth advantage of the transduced cells.
One year after transduction with the Lenti-sgTS15/Cre or Lenti-sgTS102/Cre pools, Nf1f/f;TC, Ptenf/f;TC, and Trp53f/f;TC mice developed a modest number of tumors (defined as Tomatopositive expansion >0.5 mm in diameter), while Lkb1f/f;TC and Keap1f/f;TC mice rarely developed any tumors (Figure 1c-d, S3b-c). Interestingly, Nf1f/f;TC, Ptenf/f;TC, and Trp53f/f;TC, and TC mice transduced with the larger Lenti-sgTS102/Cre pool developed many more and larger tumors than the mice transduced with the Lenti-sgTS15/Cre pool. These tumors were positive for TTF1, a marker for lung adenocarcinoma, and negative for P63 and UCHL1, markers for squamous cell and small cell lung cancer, respectively (Figure 1e).
To determine whether some of these tumors contained spontaneous mutations in known proto-oncogenes, we PCR-amplified and sequenced 10 genomic regions that include the hotspot oncogenic mutation sites in Kras, Braf, Nras, and Egfr (Figure S3d, Table S2, and Methods) [33, 38–46]. Across 29 samples, we detected only one oncogene mutation (a KrasG12V mutation in a tumor from a Ptenf/f;TC mouse). Thus, the majority of these tumors arise in the absence of the hotspot mutations in the aforementioned proto-oncogenes. This is consistent with the low mutation rate in mouse models of lung cancer [47] and suggests that the inactivation of combinations of specific tumor suppressor genes in Nf1f/f;TC, Ptenf/f;TC, and Trp53f/f;TC mice drives the development of lung cancer in vivo. Importantly, the overall low number of tumors indicates that inactivation of the “core” tumor suppressor genes alone, and most combinations of tumor suppressor genes tested, are insufficient to generate lung tumors.
Identification of top candidate tumor suppressor genes involved in oncogene-negative lung tumor formation
The Lenti-sgRNA/Cre vectors contain two-component barcodes in which an sgID identifies the sgRNA and a random barcode (BC) uniquely tags each clonal tumor. Thus, high throughput sequencing of the sgID-BC region can identify the sgRNA(s) present in each tumor and quantify the number of cells in each tumor (Figure 1a). To determine which sgRNAs were present in the largest tumors in mice transduced with Lenti-sgTS102/Cre pool, we PCR- amplified the sgID-BC region from genomic DNA from dissected tumors and performed high- throughput sgID-BC sequencing. Most large tumors contained multiple Lenti-sgRNA/Cre vectors targeting different tumor suppressors (Figure S4a-b). Therefore, we calculated the statistical enrichment of each sgRNA based on their relative representation in the dissected tumors (Figure 1f and S4, see Methods).
To further quantify the impact of inactivating each tumor suppressor gene on clonal expansion of lung epithelial cells, we performed tumor barcode sequencing (Tuba-seq) on bulk DNA from one lung lobe from each Nf1f/f;TC, Ptenf/f;TC, Trp53f/f;TC, and TC mouse (Figure 1c). Analysis of the number of cells in clonal expansions further nominated tumor suppressor genes that may contribute to tumor initiation and growth (Figure 1f, and S5). Based on these two analyses, we selected 13 genes for further analysis (Figure 1f). The potential importance of these 13 tumor suppressor genes was often supported by both sgRNAs targeting each gene, suggesting on-target effects. Finally, the presence of sgRNAs targeting our “core” tumor suppressors allowed us to cross-validate our screen. For example, Lenti-sgPten/Cre was enriched in tumors in Nf1f/f;TC mice, and Lenti-sgNf1/Cre was enriched in tumors in Ptenf/f;TC mice (Figure 1f and S4-5).
Inactivation of candidate tumor suppressors efficiently generates lung tumors
To determine the potential of the 13 tumor suppressor genes to initiate oncogene-negative tumors, we generated a pool of Lenti-sgRNA/Cre vectors targeting each of these tumor suppressor genes and one vector with an inert sgRNA (Lenti-sgTS14/Cre pool; Figure 2a). We targeted each gene with the sgRNA that produced the most significant effect on tumor growth and used five times higher titer of each lentiviral vector per mouse than what was used in Lenti- sgTS102/Cre pool, thus increasing the potential for the transduction of the initial cell with multiple Lenti-sgRNA/Cre vectors and detection of higher-order tumor suppressor interactions.
We initiated tumors with Lenti-sgTS14/Cre in Nf1f/f;TC, Ptenf/f;TC, Trp53f/f;TC, TC, and KrasLSL-G12D;T (KT) mice. Less than four months after tumor initiation, several Nf1f/f;TC and Ptenf/f;TC mice showed signs of extensive tumor burden. These mice developed approximately four times more tumors than mice of the same genotypes one year after transduction with the Lenti-sgTS102/Cre (compare Figure 2b-c with Figure 1c-d, and S10c). This suggests that these candidate tumor suppressor genes are enriched for those that generate oncogene-negative lung tumors.
We performed Tuba-seq on DNA from bulk tumor-bearing lungs to determine the number and size of each tumor with each barcoded Lenti-sgRNA/Cre vector. Inactivation of Nf1, Rasa1, and Pten most dramatically increased tumor size and/or tumor number across all mouse genotypes (Figure 2d-e, S6a-b, and Methods). Inactivation of some of the other ten tumor suppressor genes less dramatically but significantly increased tumor size and/or tumor number in a more genotype-specific manner. This suggests that additional molecular pathways altered by these tumor suppressor genes may also lead to early epithelial expansions.
The largest tumors in Nf1f/f;TC, Ptenf/f;TC, Trp53f/f;TC, and TC mice were frequently generated through the inactivation of more than two tumor suppressor genes. Vectors targeting Nf1, Rasa1, and/or Pten were often present in the largest tumors, and the coincident targeting of Nf1, Rasa1, and Pten was the most frequent combination (Figure 2f-g, S6c-h).
To gain greater insight into the contribution of Nf1, Rasa1, and Pten inactivation to the generation of oncogene-negative tumors, we transduced Nf1f/f;TC, Ptenf/f;TC,Trp53f/f;TC, TC, and KT mice with a pool of Lenti-sgRNA/Cre vectors that lacked the vectors targeting Nf1, Rasa1, and Pten (Lenti-sgTS11/Cre) (Figure S7a). Approximately four months after transduction, these mice had many fewer tumors than mice transduced with Lenti-sgTS14/Cre pool (Figure S7b-c and S10c). Tuba-seq analysis confirmed a dramatic decrease in tumor burden relative to mice that received the Lenti-sgTS14/Cre pool (Figure 2h). Thus, the inactivation of Nf1, Rasa1, and Pten emerged as the most important contributors to the generation of oncogene- negative lung tumors.
Extensive experiments generating single and pairwise inactivation of several tumor suppressor genes in individual mice led to the development of very few tumors or hyperplasia only after mutations of Nf1-Rasa1, Nf1-Pten, and Pten-Rasa1, even when initiated with high titers of virus and after long periods of time (Figure S8-S9). Thus, single, and pairwise tumor suppressor gene inactivation is rarely sufficient to generate lung tumors and combinatorial inactivation of three or more tumor suppressor genes likely increases the efficiency of tumor development and/or accelerates the growth of oncogene-negative lung tumors.
Combinatorial inactivation of Nf1, Rasa1, and Pten promotes lung adenocarcinoma development comparably to oncogenic Kras mutation
To dissect the higher-order genetic interactions between Nf1, Rasa1, and Pten, we transduced TC and Trp53f/f;TC mice with a pool of eight lentiviral vectors that would inactivate Nf1, Rasa1, and Pten individually, in pairwise combinations, and all three simultaneously (Lenti- sgTSTriple-pool/Cre, Figure 3a). Three months after tumor initiation, TC mice had hundreds of large tumors that were adenomas and adenocarcinomas (Figure 3b-c and Figure S10a-e). Tuba- seq analysis showed that most of the tumor burden arose as a consequence of concomitant inactivation of all three tumor suppressors, with single and pairwise inactivation of these genes generating very few tumors consistent with our previous observation in individual mice (Figure 3e and S8). To compare the tumor initiation potential of combinatorial Nf1, Rasa1, and Pten mutations with that of a known oncogene, we transduced KrasLSL-G12D;T (KT) mice with Lenti- sgTSTriple-pool/Cre (Figure 3a). Strikingly, coincident inactivation of Nf1, Rasa1, and Pten in TC mice was nearly as potent as oncogenic KRASG12D in driving lung tumor initiation in vivo (Figure 3g and Methods). Additional inactivation of Trp53 in Trp53f/f;TC mice did not increase tumor initiation suggesting that Trp53 is not a major suppressor of oncogene-negative lung adenocarcinoma at these early stages (compare Figure 3b-f and Figure S10a-h).
In molecular evolution studies, generating combinatorial genomic alterations and measuring the fitness of each genotype (growth rate) is often used to infer the possible and the most probable paths from a wild-type state to a complex genotype [48]. Through the generation of all the possible combinatorial alterations of Nf1, Rasa1, and Pten, we quantified the fitness conferred by each mutation and the relative probabilities of different adaptive paths leading from a wild-type state to the triple mutant genotype (see Methods). Our data suggest that inactivation of these three genes can occur in any order, with each additional alteration further increasing the fitness (Figure 3f). The Nf1 Rasa1 Pten mutation sequence emerged as the most probable of all six possible paths.
Finally, To further analyze tumors driven by inactivation of Nf1, Rasa1, and Pten, we initiated tumors in TC and Trp53f/f;TC mice using only the lentiviral vector that targets all three tumor suppressor genes (Lenti-sgNf1-sgRasa1-sgPten/Cre) (Figure S11a). In only three months, these mice developed very large numbers of lung tumors that were almost exclusively lung adenomas/adenocarcinomas (Figure S11b-e). We confirmed the inactivation of Nf1, Rasa1, and Pten in these tumors (Figure S11f). Whole-exome sequencing uncovered no putative oncogene mutations and only a few putative tumor suppressor mutations, none of which occurred in more than one tumor (Table S3). Interestingly, at later timepoints after initiation, tumors in Trp53f/f;TC mice progressed into an invasive Nkx2-1negative Hmga2positive state and metastasized to other organs such as liver similar to what has been reported in KRASG12D;TRP53 mutant lung adenocarcinoma models (Figure S12) [49].
Inactivation of Lkb1 and Keap1 reduce initiation of oncogene-negative tumors
In our initial analyses Lkb1f/f;TC and Keap1f/f;TC mice rarely developed tumors (Figure 1c-d). To directly test the impact of Lkb1 and Keap1 inactivation on the ability of combinatorial inactivation of Nf1, Rasa1, and Pten in generation of oncogene-negative tumors, we initiated tumors with the Lenti-sgTSTriple-pool/Cre pool in Lkb1f/f;TC and Keap1f/f;TC mice (Figure S13a). Consistent with our initial observation, Keap1f/f;TC mice developed many fewer tumors than TC mice (Figure 1c-d and S13b-e). Keap1 inactivation reduced tumor number by ∼10 fold, while having almost no effect on tumor size (Figure S13f). In contrast, Lkb1f/f;TC mice transduced with the Lenti-sgTSTriple-pool/Cre pool developed tumors that appeared larger than those in TC mice (Figure S13b-d). While Lkb1f/f;TC mice had approximately the same number of total neoplastic cells as TC mice, Lkb1f/f;TC mice had greatly reduced tumor number (10-100 fold depending on the engineered genotype) but increased mean tumor size by ∼10 fold (Figure S13e, g). This is consistent with Lkb1 inactivation being detrimental to initiation and/or very early growth of this subtype of oncogene-negative lung tumors while greatly increases growth of a rare subset of these tumors. These findings further underscore the context-specific effects of tumor suppressor genes across lung adenocarcinoma subtypes.
Oncogene-negative murine lung adenocarcinomas have activated RAS and PI3K pathways
NF1 and RASA1 are negative regulators of RAS, while PTEN is a negative regulator of the PI3K-AKT pathway. Therefore, we investigated the impact of inactivating these tumor suppressor genes on RAS and PI3K pathway activation by immunohistochemistry, as well as by RNA-sequencing (RNA-seq) on FACS-isolated Tomatopositive cancer cells. We generated autochthonous tumors in which Nf1, Rasa1, and Pten were inactivated (TC mice with Lenti- sgNf1-sgRasa1-sgPten/Cre; Nf1/Rasa1/Pten tumors), KRASG12D was expressed (KT;H11LSL-Cas9 mice with Lenti-sgInert/Cre; Kras tumors), or KRASG12D was expressed and Pten was inactivated (KT;H11LSL-Cas9 mice with Lenti-sgPten/Cre; Kras/Pten tumors) (Figure S14a, and Methods).
We examined PI3K and RAS pathway activation by immunohistochemistry on Nf1/Rasa1/Pten and Kras/Pten tumors. Nf1/Rasa1/Pten tumors had positive staining for pERK (indicative of RAS pathway activation) and pAKT (indicative of PI3K pathway activation) (Figure 4a). Compared with Kras/Pten tumors, the average pERK staining in Nf1/Rasa1/Pten tumors was less intense and pAKT staining was similar (Figure 4b-c). To compare the expression of genes downstream of the RAS and PI3K-AKT pathways in Nf1/Rasa1/Pten and KRASG12D-driven tumors, we performed single-sample gene set variation analysis (ssGSVA) to derive enrichment scores for individual tumors based on previously reported gene sets representing RAS and PI3K-AKT regulated genes [50, 51]. Compared with Kras/Pten tumors, Nf1/Rasa1/Pten tumors had lower RAS pathway gene signature scores (Figure S14b). PI3K- AKT pathway gene signature scores were similar in Nf1/Rasa1/Pten and Kras tumors (Figure S14c). Interestingly, the rare tumors that eventually developed after pairwise inactivation of Nf1, Rasa1, and Pten also had strong activation of RAS and PI3K pathways (Figure S8 and S14d). This indicates pathway-level biochemical changes, necessary for tumor development, can be acquired either through combinatorial alterations of Nf1, Rasa1, and Pten in a short period of time, or through pairwise alterations of these genes and other unknown factor(s) over longer time periods.
Oncogene-negative human lung adenocarcinomas frequently have activation of RAS and PI3K pathways
To investigate the activation of RAS and PI3K pathways in human oncogene-negative lung adenocarcinomas, we analyzed oncogene-negative (N=35) and oncogene-positive (N=18) lung adenocarcinomas. Immunohistochemistry for pERK and pAKT showed that the average RAS pathway activation was lower, while PI3K pathway activation was similar in oncogene- negative tumors compared with oncogene-positive tumors, which is consistent with our observations from our mouse models (Figure 4d-g and S14e-j). Over 90% of oncogene- negative human tumors had moderate to strong activation of either the RAS or PI3K pathways, while ∼45% of these tumors had moderate to strong activation of both RAS and PI3K pathways (Figure 4h).
These tumors were genomically characterized by Stanford’s Solid Tumor Actionable Mutation Panel (STAMP)[52]. However, activation of the RAS and PI3K pathways were rarely explained by mutations in NF1, PTEN, or other genes profiled by STAMP (Table S5 and S6). This could be due to the noncomprehensive gene panel characterized by STAMP (only 16 out of 49 RAS pathway-related genes and 8 out of 20 PI3K pathway related genes are partially or fully analyzed by STAMP) or the presence of additional unknown genetic and/or epigenetic mechanisms of RAS and PI3K pathway activation.
To assess oncogene-negative lung adenocarcinomas in TCGA and GENIE for genomic alterations that could potentially lead to the activation of RAS and PI3K pathways, we queried a set of well-established negative regulators of the RAS and PI3K pathways for alterations in oncogene-negative tumors in TCGA and GENIE datasets (see Methods and Table S6).
Consistent with activation of RAS and/or PI3K pathways in a large fraction of oncogene- negative lung adenocarcinomas, over 60% of oncogene-negative lung adenocarcinomas in TCGA had alterations in either the RAS or PI3K pathways, and 22% of these tumors had alterations in components of both pathways (Figure 4i). These frequencies were lower in the GENIE dataset, likely because only a fraction of the known genes in these two pathways were analyzed (Figure S15a).
Consistent with previous reports, NF1 and RASA1 alterations were enriched in oncogene- negative tumors; however, coincident alterations in NF1, RASA1, and PTEN were rare, supporting the importance of pathway-level analysis of human genomic data (Figure S15b-c) [53, 54]. Considering previous reports on epigenetic silencing and other non-genomic mechanism of inhibiting tumor suppressor genes such as PTEN [22, 23, 55, 56] , we performed immunohistochemistry for PTEN on 20 oncogene-negative lung adenocarcinomas that did not have genomic PTEN mutations. Consistent with previous reports, we observed low PTEN protein levels in 13 out of 20 of these tumors (Figure S15d-i) [22]. These observations support a model in which activation of the RAS and PI3K pathways can be generated by diverse genomic and/or epigenetic alterations.
Finally, we assessed whether oncogene-negative tumors in our mouse model exhibit transcriptional features that overlap with those of oncogene-negative human lung adenocarcinoma. We generated a gene expression signature of oncogene-negative tumors comprised of genes that are higher in Nf1/Rasa1/Pten tumors relative to Kras tumors in mice. We then calculated gene signature activity scores for each TCGA lung adenocarcinoma tumor on the basis of our mouse-derived oncogene-negative gene expression signature using single-sample GSEA (Table S4). Interestingly, upon stratification of TCGA patients on the basis of whether they harbor oncogene mutations, we found that our oncogene-negative signature was highest in oncogene-negative human lung adenocarcinomas relative to lung adenocarcinomas driven by oncogenic KRAS or other known oncogenes (Figure 4j).
Collectively these data indicate that the molecular and biochemical state of Nf1/Rasa1/Pten mouse tumors recapitulates that of a substantial fraction of oncogene-negative human lung adenocarcinomas. These observations underscore the value of in vivo functional genomics in identifying the biochemical changes that drive human tumor development.
Oncogene-negative lung tumors are vulnerable to inhibition of RAS and PI3K-AKT pathways
Understanding the biochemical changes that drive tumor development can nominate potential therapeutic strategies [38]. To investigate the therapeutic benefit of targeting key nodes in oncogene-negative lung cancer, we initiated tumors in TC mice with a smaller pool of barcoded sgRNA viral vectors targeting Nf1, Rasa1, and Pten. We treated these mice with the SHP2 inhibitor RMC-4550 [57], AKT1/2 inhibitor capivasertib [58, 59], or a combination of the two (Figure 5a and S16a-b). These drugs were chosen based on their ongoing clinical development and ability to reduce activation RAS and PI3K pathways [57, 59].
Direct fluorescence imaging and histology indicated that SHP2 inhibition and combined SHP2 and AKT1/2 inhibition greatly reduced tumor burden (Figure 5b-c and S16c). Tuba-seq analysis provided greater insights into the overall and genotype-specific responses of tumors to the therapeutic interventions. Capivasertib monotherapy was ineffective while RMC-4550 reduced the total tumor burden. While the inefficiency of capivasertib could have been due to incomplete target inhibition the combination of RMC-4550 and capivasertib trended towards being the most efficient therapeutic approach reducing tumor burden by ∼24-34% compared with RMC-4550 alone capivasertib monotherapy was ineffective, while RMC-4550 and the combination of RMC-4550 and capivasertib reduced the total tumor burden significantly, with the combination therapy trending towards being the most efficient therapeutic approach (∼24- 34% reduction in tumor burden in combination therapy compared with RMC-4550 alone) (Figure 5d, S16d-g).
We confirmed the inhibition of RAS and PI3K pathways in lung tumors in mice treated with RMC-4550 and capivasertib by immunohistochemistry (Figure S16h). Furthermore, global gene expression analysis confirmed the downregulation of RAS and PI3K-AKT gene expression signatures after coincident SHP2 and AKT1/2 inhibition (Figure S17a-d). Treated tumors tended to have higher expression of an apoptosis gene expression signature and lower expression of a G2/M gene expression signature, suggesting that this combination treatment induces broad cellular changes in oncogene-negative tumors (Figure S17e-f).
Inhibition of SHP2 and AKT synergizes to reduce the growth of lung adenocarcinoma cell lines
To more extensively characterize the responses to these SHP2 and AKT inhibitors, we generated a panel of Nf1/Rasa1/Pten deficient cancer cell lines from tumors initiated with Lenti- sgNf1-sgRasa1-sgPten/Cre in Trp53flox/flox;TC mice (Figure 6a and S18a-b). As anticipated, RAS and PI3K signaling was reduced in response to treatment with RMC-4550 and capivasertib, respectively (Figure S18c). RMC-4550 and capivasertib each decreased the overall growth of three oncogene-negative cell lines in a dose-dependent manner (Figure 6b and S18d, f). Consistent with our in vivo observations, RMC-4550 and capivasertib synergized to inhibit the growth of oncogene-negative lung adenocarcinoma cell lines (Figure 6c, and S18e, g). RAS and PI3K signaling can promote cell growth and survival [58, 59], and RMC-4550 and capivasertib inhibited proliferation and induced apoptosis to a greater extent than either RMC-4550 or capivasertib alone (Figure 6d-e). RMC-4550 and capivasertib treatment also leads to regression of subcutaneous allografts generated from these mouse oncogene-negative cell lines (Figure 6f-g and S18h-j).
Building on these findings, we assessed activation of RAS and PI3K pathways and driver pathway vulnerabilities in two oncogene-negative human lung adenocarcinoma cell lines, NCI- H1838 (NF1LOF) and NCI-H1623 (RASA1LOF). H1838 and H1623 had activation of RAS and PI3K pathways similarly to oncogene-positive human lung adenocarcinoma cell lines (Figure S18k). Consistent with our findings in mouse oncogene-negative cell lines, RMC-4550 synergizes with capivasertib to inhibit the growth of these oncogene-negative human lung adenocarcinoma cell lines (Figure 6h-i and S18l-m). Collectively, these in vivo and cell culture analyses indicate that oncogene-negative tumors with activated RAS and PI3K pathways are vulnerable to therapeutic inhibition of these pathways.
DISCUSSION
Lung adenocarcinomas that lack defined oncogene alterations afflict as many patients as those driven by either oncogenic KRAS or EGFR. To identify whether combinatorial inactivation of multiple tumor suppressor genes can drive the initiation and growth of lung adenocarcinoma in the absence of oncogene activation, we performed a series of multiplexed in vivo functional genomic screens. By querying an extensive set of combinatorial tumor suppressor gene alterations, we discovered that inactivation of single tumor suppressor genes, as well as pairwise alteration of the majority of tumor suppressors that we assessed, are insufficient to generate lung tumors. Importantly, we uncovered higher-order interactions between tumor suppressor genes as key drivers of oncogene-negative lung adenocarcinomas, with combinatorial inactivation of Nf1, Rasa1, and Pten being as potent as oncogenic KrasG12D in initiating lung tumors in vivo.
Mutations in NF1, RASA1, and PTEN are not strictly mutually exclusive with oncogene alterations in human lung adenocarcinoma, and their inactivation increases initiation and/or growth of KrasG12D-driven lung tumors [35, 60, 61]. Furthermore, while NF1 inactivation is sometimes suggested to be an “oncogenic driver” in lung adenocarcinoma [7, 29, 62], Nf1 inactivation alone is insufficient to initiate lung tumors (Figure S8). Coincident mutations in NF1 and RASA1 are mutually exclusive with other oncogene alterations [53, 54]. However, pairwise alterations of Nf1 and Rasa1 and all other tumor suppressor genes that we tested generated very few tumors even after long time periods. The potent generation of lung adenocarcinomas after combinatorial inactivation of Nf1, Rasa1, and Pten suggests that genomic and/or non-genomic alterations in multiple genes within and across pathways may be required to surpass the thresholds necessary for tumor initiation and growth. We speculate that these thresholds may also be influenced by tumor suppressor genes independent from RAS and PI3K pathways, as well as by environmental factors.
Although cancers harbor diverse genomic and epigenomic alterations, these alterations often converge on key pathways and generate similar biochemical changes [15, 63]. For example, myeloid leukemia can be driven by gain-of-function mutations in KRAS, NRAS, or the receptor tyrosine kinase FLT3, or combined inactivation of multiple negative regulators of RAS pathway such as SPRY4 and NF1 [64, 65]. In contrast to pathway activation generated by oncogene alterations, pathway activation through inactivation of tumor suppressors, can be very diverse, precluding the identification of obvious genomic drivers from gene-centric analysis of human cancer genomic data. Although coincident inactivation of NF1, RASA1, and PTEN is rare in human oncogene-negative lung tumors, mutations in different genes that converge on the RAS and PI3K pathways frequently co-occur in oncogene-negative lung adenocarcinomas (Figure 4i and S15a). Furthermore, activation of the RAS and PI3K pathways in human oncogene-negative lung adenocarcinoma is likely not driven exclusively by genomic alterations in these two pathways, consistent with previous reports and our observations suggesting non-genomic or other mechanisms of downregulation of RAS GAPs and PTEN (Figure 4f-h) [7, 23, 55, 56]. For example, PTEN is mutated in less than 6% of lung adenocarcinomas but is downregulated in up to a quarter of early-stage lung adenocarcinomas through promoter methylation [22, 24]. We also observed low levels of PTEN, in the absence of genetic mutations, in more than half of the oncogene-negative human lung adenocarcinomas (Figure S15i). Thus, despite the ease of DNA sequencing, genomic alterations should serve as a floor, not a ceiling, in estimating the frequency of pathway alteration.
Selection for molecular alterations during cancer evolution is a non-random and tightly constrained process that is strongly influenced by molecular network interactions [66]. The mouse models that we employed are uniquely able to quantify the likely trajectory of selection for alterations that could lead to the driver pathway activation and their possible evolutionary paths. Our data suggest that inactivation of Nf1, Rasa1, or Pten each increases cellular fitness, thus making each next step more likely (Figure 3f and S10g). Interestingly, within other evolutionary systems, this is not always the case, and generation of complex genotypes can be constrained [48, 67–70]. Moreover, given multiple components of RAS and PI3K pathways, single or combinatorial alterations of many of these components in human lung epithelial cells might enable diverse paths toward cellular transformation (Figure 6j).
In this study, we assessed the ability of hundreds of complex tumor suppressor genotypes to generate lung tumors. While activation of RAS and PI3K pathway emerged as the most potent driver of oncogene-negative lung adenocarcinomas, our data also suggest that combinatorial inactivation of tumor suppressor genes outside these two pathways likely has the ability to initiate tumorigenesis (Figure 2 and S6). Given the mutational diversity and complexity of oncogene-negative human lung adenocarcinomas [71], there also remain many other mutational combinations to be investigated. Furthermore, in contrast to previous studies, we did not observe the emergence of lung squamous cell carcinoma as a result of inactivation of Lkb1 and Pten [72], the generation of lung adenocarcinoma after inactivation of Keap1 and Pten (Figure S13)[73], or development of small cell lung cancer as a result of Trp53 and Rb1 inactivation (Figure 1e) [20]. These differences could be driven by the different tropisms of adenoviral vectors (used in all the studies mentioned above) versus lentiviral vectors (used in our study) [74, 75] and/or lower titer of individual lentiviral vectors targeting each gene in our screens. Thus, functional genomic studies across different lung epithelial cell types may be important to comprehensively understand the genesis and progression of different subtypes of lung cancer including squamous and small cell lung cancer.
Knowledge of the genes underlying human cancers is a pillar of cancer diagnostics, personalized medicine, and the selection of rational combination therapies. Our data demonstrate pathway activation in the absence of oncogene mutations in a sizable fraction of human lung adenocarcinoma and document the feasibility of treating oncogene-negative lung adenocarcinomas in mouse and human oncogene-negative lung adenocarcinoma cell lines, and mouse models using inhibitors that are already being tested clinically. Thus, biochemical assessment of oncogenic pathways in tumors is a strong foundation for rational selection of therapies and clinical trial designs. Beyond SHP2 and AKT, extensive efforts have generated inhibitors for many other components of the RAS and PI3K pathways. Thus, further investigation of the therapeutic targeting of key nodes within the RAS pathway (e.g., SOS, MEK, ERK) and PI3K pathway (e.g., PI3K, mTOR), could contribute to the development of the most effective and least toxic therapies.
Our model and findings enable a detailed analysis of tumorigenic mechanisms and clinical manifestations of oncogene-negative lung adenocarcinomas, help to identify biomarkers and new therapeutic targets, and aid in preclinical testing of drugs for the treatment of these tumors. Generating comprehensive molecular and pharmacogenomic maps of oncogene-negative lung adenocarcinomas will transform our basic and translational understanding of this prevalent cancer subtype.
METHODS
Analysis of human lung adenocarcinoma datasets
Somatic mutation data (SNPs and indels, including silent mutations) for 513 TCGA lung adenocarcinoma (LUAD) tumors were downloaded from the UCSC Xena Browser (http://xena.ucsc.edu/) (Link 1 below). TCGA-LUAD clinical and exposure data were downloaded from the GDC Data Portal (https://portal.gdc.cancer.gov/projects/TCGA-LUAD) and the UCSC Xena Browser (Link 2 below). GISTIC2 thresholded copy number variation (CNV) data were downloaded from the UCSC Xena Browser (Link 3 below). Amplifications were defined as “2” and deletions as “-2”. Genes with conflicting CNV values within a single tumor were ignored. Fusion data were obtained from [76]. Fusion and CNV data were filtered to include only data from the 513 samples within the somatic mutation set. Duplicate fusions were collapsed into single fusions. MET-exon skipping data were taken from [77]. Curated survival data from [78] were downloaded from the UCSC Xena Browser (Link 4 below).
Links:
https://tcga.xenahubs.net/download/TCGA.LUAD.sampleMap/LUAD_clinicalMatrix
https://tcga.xenahubs.net/download/survival/LUAD_survival.txt.gz
Data from AACR Project GENIE (hereinafter referred to as GENIE) v8 were downloaded from https://www.synapse.org/#!Synapse:syn22228642 [78], specifically: somatic mutations, copy number alteration (CNA) data, fusion data, panel information (genomic_information.txt), and clinical data (both sample- and patient-level). All data were filtered to only include LUAD tumors. A single tumor was kept for patients with multiple different tumor samples, with priority for earlier sequenced samples and those from primary tumors. If tumor samples appeared identical within the clinical meta-data, the related patient data were excluded.
Determination of oncogenes
To have a conservative estimate of the fraction of lung adenocarcinomas without known oncogenic drivers (oncogene-negative tumors), we generated a list of oncogenes that included any gene that met at least one of these criteria: 1) Genes that have hotspot mutations or specific alterations where cancers or cancer cells with that mutation respond to therapies targeted to the protein product of that mutant gene in patients, 2) The particular alteration in that gene can generate lung adenocarcinoma in genetically-engineered mouse models, 3) The altered gene can generate tumors in other tissues in genetically-engineered mouse models, and 4) Alteration of the indicated gene can lead to the transformation of cells or predicts response to targeted therapies in vitro. Additionally, we excluded genes if their supposed oncogenic alterations co-occur with alterations in other proto-oncogenes (Table S1) in more than 50% of cases.
Classification of mutations and tumors
Mutations (somatic mutations, fusions, CNVs, and MET exon skipping [TCGA only]) were classified as within proto-oncogenes (described above) or not. Mutations within these proto-oncogenes were classified as “accepted oncogenic” mutations if those alterations met at least one of the criteria described above. Any tumor with one accepted oncogenic alteration was classified as “oncogene-positive”. Tumors with accepted oncogenic mutations in more than one gene were classified as “multiple oncogenes mutated”. Any tumor with alterations in a proto- oncogene that was not considered an accepted oncogenic alteration based on the four criteria above was classified as “oncogene-indeterminate”. Thus, these tumors contain variants of unknown significance (VUS) in proto-oncogenes [79]. The remaining tumors, without any mutations in any proto-oncogene, were classified as “oncogene-negative”.
Tumor type counts per database:
TCGA (Total: 513, Oncogene-negative: 91, Oncogene-positive: 283, and Oncogene-
indeterminate: 139)
GENIE (Total: 9,099, Oncogene-negative: 1,645, Oncogene-positive: 6,041, Oncogene-
indeterminate: 1,413)
Oncogene-positive tumors were further classified by the type of oncogenic mutation they had (Figure 1a and S1a).
Clinical characteristics
We divided patients into males or females based on the sex reported by either TCGA or GENIE, if provided. For TCGA, the arithmetic mean for age at diagnosis was computed and reported with a standard error of the mean (SEM). Non-smokers were defined as having tobacco smoking history values of 1 (see public ID 2181650 at https://cdebrowser.nci.nih.gov), while smokers were defined as anything > 1 (current or reformed smokers). The arithmetic mean pack- years smoked for smokers, if reported, was described, along with SEM.
Pan-cancer tumor suppressor genes
We generated a list of tumor suppressor genes based on two previously published reports to compare the number of altered tumor suppressor genes in oncogene-negative tumors with oncogene-positive and oncogene-indeterminate tumors [15, 80]. We manually removed genes with conflicting evidence as a tumor suppressor gene in LUAD. The final list of TSGs is in Table S1.
Calculation of mutation frequencies and absolute number of genes mutated
In general, mutation frequencies for a given gene were calculated as the number of tumors with that gene mutated, divided by the number of tumors screened for mutations in that gene (for TCGA: all tumors were screened for all genes, for GENIE: the panel sequencing information was obtained from genomic_information.txt to determine which tumors were screened for which genes). Mutation frequencies were calculated for point mutations (PM), insertion/deletions (indels), and deletions separately. Additionally, the frequency for a combination of PMs, indels, and deletions was also calculated. The screened set of tumors in GENIE for the latter included only those tumors which were screened for both PMs/indels as well as CNVs for each gene. Reported in Figure S2b are oncogene-negative tumors with either point mutations, indels, or deletions in the indicated gene. In Figure S2c-d, for each gene, a ratio of enrichment of mutations in oncogene-negative over oncogene-positive tumors was calculated as:
The p-values for enrichments were calculated using the two-sided Fisher’s Exact test as implemented by SciPy. For a given set of genes with at least a single tumor screened, the false discovery rate (FDR) was calculated using the Benjamini-Hochberg method on the Fisher’s Exact P-values.
To measure the total number of genes mutated (Figure S1d), a gene was considered mutated if it had at least one point mutation or indel. All these mutations in a tumor were collated, and the number of the unique set of genes was counted as the total number of genes mutated. For counting the number of individual tumor suppressors mutated (Figure S1e), deletions were also included, and the list of pan-cancer tumor suppressors as defined above was used. The Mann-Whitney U test was conducted on the number of respective genes mutated in either oncogene-negative or oncogene-positive tumors.
Survival Analysis
Survival data from [78] were obtained as described above. Kaplan-Meier analysis was performed to estimate probability curves for overall survival (OS) and disease-specific survival (DSS). The logrank test was used to compare oncogene-negative and oncogene-positive tumors.
Gene and pathway alteration co-occurrences
For analysis of simultaneous pairwise alterations of NF1, RASA1, or PTEN within oncogene-negative tumors, we determined the number of tumors with no mutation in NF1, RASA1, or PTEN, with mutation(s) in one gene, or mutations in two genes. Point mutations, indels, and deletions in each gene were included. A tumor needed to have one or more mutations in that gene to be considered mutated. For GENIE, only those tumors screened for both genes for point mutations and indels (according to the panel information file) were investigated. For TCGA, all oncogene-negative tumors were considered. A one-sided Fisher’s exact test was conducted to determine if there were more than the expected number of tumors with both genes mutated.
Gene lists and their acceptable alterations (i.e., not known to be an oncogene alteration) were generated as being in RAS or PI3K pathways [15, 55, 80–98] (Table S6). We determined the list of all tumors screened for each gene in each pathway for the respective type of mutation (point mutation/indel, amplification, deletion, or fusion). For each alteration within each pathway, we determined whether it could activate the corresponding pathway or not according to the above list. A gene was considered mutated if it had at least one accepted mutation within it. A tumor was considered mutated in a given pathway if it had at least one gene mutated in that pathway.
Animal Studies
The use of mice for the current study has been approved by Institutional Animal Care and Use Committee at Stanford University, protocol number 26696. KrasLSL-G12D/+ (Jax # 008179 (K)), R26LSL-tdTomato(ai9) (Jax # 007909 (T)), and H11LSL-Cas9 (Jax # 026816 (C)), Keap1flox, Pten flox (Jax # 006440), Lkb1 flox (Jax # 014143), Nf1 flox (Jax # 017640), and Trp53flox (Jax # 008462) mice have been previously described [36, 37, 41, 99–103]. All mice were on a C57BL/6:129 mixed background except the mice used for derivation of oncogene-negative Nf1, Rasa1, Pten, and Trp53 mutant cell-lines, those used for allograft studies, and some of the Trp53flox/flox;TC mice that were used for metastasis analysis (Figure S12a), which were on a pure C57BL/6 background.
Tumor initiation and selection of Lenti-sgRNA/Cre pools
Tumors were initiated by intratracheal delivery of pooled or individual Lenti-sgRNA/Cre vectors. Barcoded Lenti-sgRNA/Cre vectors within each viral pool are indicated in each figure. Tumors were initiated with the indicated titers and allowed to develop tumors for between 3 and 12 months after viral delivery, as indicated in each figure.
In Figure 1 and Figure S3, we transduced Nf1f/f;TC, Ptenf/f;TC, Trp53f/f;TC, Lkb1f/f;TC, Keap1f/f;TC, TC, and T mice with two pre-existing pools of barcoded Lenti-sgRNA/Cre vectors that target ∼50 putative tumor suppressor genes. These two pools have been previously used to studied the effect of these putative tumor suppressor genes in KRASG12D-driven lung tumors (Lenti-sgTS15/Cre [31, 32] and Lenti-sgTS102/Cre [35]).
Lenti-sgTS15/Cre contained vectors targeting 11 tumor suppressors with one sgRNA per gene in addition to four inert sgRNAs (Lenti-sgTS15/Cre) [31, 32]. Lenti-sgTS102/Cre included vectors targeting 48 tumor suppressors, including all five of the “core” tumor suppressors and most of the tumor suppressors targeted in Lenti-sgTS15/Cre with two or three sgRNAs per gene in addition to five inert sgRNAs (102 sgRNA in total, Lenti-sgTS102/Cre) [35] (See Table S1).
We determined the alteration frequency of many putative tumor suppressor genes, including those targeted using our Lenti-sgTS15/Cre and Lenti-sgTS102/Cre pools, in oncogene- positive and oncogene-negative tumors from TCGA and GENIE [15, 80]. Alterations in only 17 tumor suppressor genes were significantly enriched in oncogene-negative tumors in GENIE and most (12/17) were targeted by the Lenti-sgTS15/Cre and Lenti-sgTS102/Cre pools (Table S1).
We previously found that a small percent of lung tumors initiated with Lenti-sgRNA/Cre vectors in other lung cancer models contained multiple sgRNAs, consistent with the transduction of the initial cell with multiple Lenti-sgRNA/Cre vectors [31, 32]. Thus, from tumor suppressor genes that were found to be mutated in the largest tumors and expansions of experiment in Figure 1, we selected 7 tumor suppressor genes that showed up in Nf1f/f;TC, Ptenf/f;TC, Trp53f/f;TC mice in addition to 6 other tumor suppressor genes that showed significant effect in at least one of these three backgrounds. For Studies in Figure 2 and Figure S7, we used higher titers of Lenti-sgRNA/Cre vectors to increase the potential of finding higher-order interactions that generate lung tumors. We found that simultaneous alterations of Nf1, Rasa1, and Pten was one of the most frequent co-occurring alterations in the largest tumors. Thus, we focused on studying these three tumor suppressor alterations using Lenti-sgTripleTS8/Cre, Lenti- sgTripleTS6/Cre, and Lenti-sgNf1-sgRasa1-sgPten/Cre in the following figures.
The sgRNA sequences used in each experiment are summarized below. For a more detailed description see Table S1:
Lenti-sgTS15/Cre: The exact pool used in [88, 89]
Lenti-sgTS102/Cre: The exact pool used in [90]
Lenti-sgTS14/Cre: Version 1 of sgEP300, sgKmt2c, sgNcoa6, sgRbm10, sgNeo, sgNf1, and sgPten, and version 2 of sgArid1a, sgCdkn2a, sgDnmt3a, sgKdm6a, sgRb1, sgTet2, sgRasa1 from [90]
Lenti-sgTS11/Cre: Lenti-sgTS14/Cre pool excluding sgNf1, sgRasa1, and sgPten
Lenti-sgTripleTS8/Cre: Version 1 of sgNf1, sgNeo, and sgNT, and version 2 of sgRasa1, sgPten, and sgNeo from [90]
Lenti-sgTripleTS6/Cre: Version 1 of sgNf1, sgNeo, and sgNT, and version 2 of sgRasa1, sgPten, and sgNeo from [90]. Vectors targeting only Nf1 and only Rasa1 were removed from this pool.
Lentiviral generation, barcoding, and packaging
The sgRNA sequences, cloning, and barcoding of Lenti-sgRNA/Cre and Lenti- TriplesgRNA/Cre vectors have been previously described [31, 35, 104]. To generate lentivirus, Lenti-sgRNA/Cre vectors were individually co-transfected into 293T cells with pCMV-VSV-G (Addgene #8454) envelope plasmid and pCMV-dR8.2 dvpr (Addgene #8455) packaging plasmid using polyethylenimine. Supernatants were collected 36 and 48 hours after transfection, passed through a 0.45µm syringe filter (Millipore SLHP033RB) to remove cells and cell debris, concentrated by ultracentrifugation (25,000 g for 1.5 hours at 4°C) and resuspended in PBS overnight. Each virus was titered against a standard of known titer using LSL-YFP Mouse Embryonic Fibroblasts (MEFs) (a gift from Dr. Alejandro Sweet-Cordero/UCSF). All lentiviral vector aliquots were stored at -80°C and were thawed and pooled immediately prior to delivery to mice.
Tumor barcode sequencing and analysis
For DNA extraction from single dissected tumors to generate libraries for Tuba-seq, targeted sequencing of selected oncogenes, and whole-exome sequencing, we used Qiagen AllPrep DNA/RNA Micro kit. For Tuba-seq on bulk lungs, genomic DNA was isolated from bulk tumor-bearing lung tissue from each mouse as previously described [31]. Briefly, benchmark control cell lines were generated from LSL-YFP MEFs transduced by a barcoded Lenti-sgNT3/Cre vector (NT3: an inert sgRNA with a unique sgRNA identifying barcode (sgID) and a random barcode (BC)) and purified by sorting YFP+ cells using BD FACS Aria™ II Cell Sorter. Three cell lines (100,000 to 500,000 cells each) were added to each mouse lung sample before lysis to enable the calculation of the absolute number of neoplastic cells in each tumor from the number of sgID-BC reads. Following homogenization and overnight protease K digestion, genomic DNA was extracted from the lung lysates using standard phenol-chloroform and ethanol precipitation methods. Subsequently, Q5 High-Fidelity 2x Master Mix (New England Biolabs, M0494X) was used to amplify the sgID-BC region from 50 ng of DNA from dissected tumors or 32 μg of bulk lung genomic DNA. The unique dual-indexed primers used were Forward: AATGATACGGCGACCACCGAGATCTACAC- 8 nucleotides for i5 index- ACACTCTTTCCCTACACGACGCTCTTCCGATCT-6 to 9 random nucleotides for increasing the diversity-GCGCACGTCTGCCGCGCTG and Reverse: CAAGCAGAAGACGGCATACGAGAT-6 nucleotides for i7 index- GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT-9 to 6 random nucleotides for increasing the diversity-CAGGTTCTTGCGAACCTCAT. The PCR products were purified with Agencourt AMPure XP beads (Beckman Coulter, A63881) using a double size selection protocol. The concentration and quality of the purified libraries were determined using the Agilent High Sensitivity DNA kit (Agilent Technologies, 5067-4626) on the Agilent 2100 Bioanalyzer (Agilent Technologies, G2939BA). The libraries were pooled based on lung weights to ensure even reading depth, cleaned up again using AMPure XP beads, and sequenced (read length 2×150bp) on the Illumina HiSeq 2500 or NextSeq 500 platform (Admera Health Biopharma Services).
Tuba-seq analysis of tumor barcode reads
The FASTQ files were parsed to identify the sgID and barcode (BC) for each read. Each read is expected to contain an 8-nucleotide sgID region followed by a 30-nucleotide barcode (BC) region (GCNNNNNTANNNNNGCNNNNNTANNNNNGC), and each of the 20 Ns represents random nucleotides. The sgID region identifies the putative tumor suppressor gene being targeted, for which we require a perfect match between the sequence in the forward read and one of the forward sgIDs with known sequences. Note that all sgID sequences differ from each other by at least three nucleotides. Therefore, the incorrect assignment of sgID due to PCR or sequencing error is extremely unlikely. All cells generated from the clonal expansion of an original cell transduced with a lentiviral vector carry the same BC sequence. To minimize the effects of sequencing errors on calling the BC, we require the forward and reverse reads to agree completely within the 30-nucleotide sequence to be further processed. In our pipeline, any tumor that is within a Hamming distance of two from a larger tumor is assigned as a “spurious tumor”, which likely results from sequencing or PCR errors and the tumor is removed from subsequent analysis. Reads with the same sgID and barcode are assigned to be the same tumor. The tumor size (number of neoplastic cells) is calculated by normalizing the number of reads to the three benchmarks “spike-in” cell lines added to each sample prior to lysis of the lung and DNA extraction step. The median sequencing depth was ∼ 1 read per 4.8 cells, and the minimum sequencing depth is ∼1 read per 16.5 cells. We have high statistical power in identifying tumors with more than 200 cells, which was used as the minimum cell number cutoff for calling tumors. A minimum cell number of 50 was used for calling expansions in Figures S5 and S6). Minimizing the influence of GC amplification bias on tumor-size calling was done as previously described [31].
Measures of tumor size and growth
We used several metrics of tumor number, burden and size (see Supplemental Figure 4 in [35] for additional details on the calculation of these metric).
Surface tumor size (Methodology: visual inspection), Tomato-positive expansions larger than 0.5 mm in diameter
Relative tumor size / expansion size (Methodology: Tuba-seq): Tumor/expansion size at the indicated percentile was calculated using tumors (clonal cell populations >200 cells) or expansions (clonal cell populations >50 cells) merged from all mice and normalized to the same percentile of sgInert tumors/expansions.
Relative tumor burden (Methodology: Tuba-seq): Tumor burden was calculated as the sum of neoplastic cells per mouse averaged over all mice and normalized to the tumor burden of sgInert tumors.
Relative tumor number (Methodology: Tuba-seq): Tumor numbers above a given size threshold (e.g., 1000 cells) were determined by calculating the number of tumors above the threshold per mouse averaged over all mice and normalized to the tumor number of sgInert tumors.
Relative frequency (Methodology: Tuba-seq): The relative frequency of each sgRNA was calculated in each sample (one sample can contain multiple sgRNAs due to multiple transduction or multiple tumors being present in the sample) and averaged for each sgRNA over all samples for a given mouse genotype.
Frequency in large tumors (Methodology: Tuba-seq): To find synergistic combinations in our data, we ranked all possible combinations of targeted genes by their frequency of co-mutation in the largest tumors. See Method section “Multiple transduction” for how largest tumors and co- mutations of genes were defined.
Tumor burden and tumor number are affected linearly by the titer of each Lenti- sgRNA/Cre vector in the pool. When applicable, we used data on the number of tumors from KT mice (which lack Cas9) to quantify the representation of each Lenti-sgRNA/Cre vectors in the lentiviral pool. Therefore, when calculating tumor burden and tumor number metrics, we normalized the metric to the effective titer based on data from KT mice to account for the viral titer differences among different Lenti-sgRNA/Cre vectors. Tumor/expansion size percentiles, tumor burden, and tumor number were normalized to the values of the same metric for tumors with inert sgRNAs, thus the expression “relative” is used.
For relative tumor/expansion size, relative tumor burden and relative tumor number, confidence intervals and p- values were calculated by a nested bootstrap resampling approach to account for variation in sizes of tumors of a given genotype both across and within mice. First, tumors of each mouse were grouped, and these groups (mice) were resampled. Second, all tumors of a given mouse resampling were bootstrapped on an individual basis (10,000 repetitions). For relative frequency, tumors were bootstrap resampled 10,000 times, and the distribution of inert sgRNA frequencies was used to calculate p-values for enrichment of all other sgRNAs. For “frequency in large tumors”, a permutation test was used to calculate p-values (see section Multiple transduction for details).
Multiple transduction
A fraction of lung tumors initiated with Lenti-sgRNA/Cre vectors contained multiple barcoded Lenti-sgRNA/Cre vectors. If multiple barcodes (sgID-BCs) have unexpectedly similar read counts (Figure S4a,b), we suspect transduction of the initial cell with multiple Lenti-sgRNA/Cre vectors.
To capitalize on these multiple transductions as a way to find higher-order interactions between tumor suppressor genes, we developed a method to identify the combinations of sgRNA that appear to cooperate as potent drivers of tumor growth. Accurate identification of coinfected tumors and grouping of barcodes without over grouping was not a trivial task. We developed methods to identify tumors with likely multiple transductions (i.e., those tumors with complex genotypes with multiple tumor suppressor genes inactivated). For each sgID-BC, we listed all other sgID-BCs from the same sample with read counts within 10% as possible multiple transduction events. A tumor with multiple transductions can be most easily identified among the largest tumors in each mouse as smaller tumors of similar sizes are too abundant. Multiple transductions that lead to synergistic combinatorial tumor suppressor alterations would confer a growth advantage. Thus, synergistic combinatorial alterations of tumor suppressor genes would be expected to be overrepresented among the largest tumors.
To have a dataset with a higher signal-to-noise ratio, we analyzed the largest tumors that were co-infected with up to 6 Lenti-sgRNA/Cre vectors. With this method, for each tumor, we assembled a list of genes that were possibly co-mutated. We then ranked all possible combinations of genes by their frequency in the largest tumors (Figure 2f-g and S6c-h).
An inherent problem with this analysis is that the genotypes that increase tumor growth will be overrepresented amongst the largest tumors even without multiple transductions and specific synergistic interactions. To account for the different number of tumors with different sgIDs, we performed a permutation test, where we control for the number of tumors of each genotype but randomize the sizes of tumors by randomly matching the genotypes with tumor sizes (10,000 repetitions). Synergistic tumor suppressor combinations will occur at significantly higher than expected frequencies based on this permutation test (Figure 2f-g and S6c-h).
Reassuringly, while our analysis resulted in significant enrichment of complex genotypes based on the permutation test, a control analysis performed on smaller tumors within the same mice with high noise to signal ratio resulted in a loss of statistical significance, this shows that our permutation test controls for the bias of different frequency of sgIDs among the tumors.
Fitness landscape and adaptive paths
To investigate the possible adaptive steps that can lead to the complex genotype of coincident inactivation of Nf1, Rasa1, and Pten, we first measured the fitness of all possible combinations of Nf1, Rasa1, and Pten mutations (Figure 3f and S10g). Relative (Malthusian) fitness was calculated based on the number of individuals (cells) at the end (N1) and the beginning of (N0) of a time period [105]. For each genotype, the overall sum of neoplastic cells at the end of the experiment (N1) was calculated as the sum of cells from all tumors in each mouse. As we use KT mice (which lack Cas9 and all sgRNAs have no effect) to approximate the effective titer of our virus pool (see section Measures of tumor size and growth), the initial number of cells transduced (N0) was calculated from the number of tumors generated in control KT mice. Next, the relative fitness for genotype A compared to wild type (wt) was calculated as:
Fitness values relative to wild type are displayed as nodes on the adaptive landscape (Figure 3f and S10g), where genotypes one mutation away from each other are connected by arrows that represent mutations. In the case of the Nf1;Rasa1;Pten triple mutant state, six adaptive paths can lead from wild type to that triple mutant genotype (Figure 3f and S10g). Arrows are shown if the mutation increases the fitness. In figure 3f and S10g all arrows are shown since all mutations increase fitness.
Next, we set out to approximate the relative probabilities of different adaptive paths leading from wild type to the triple mutant genotype with a simple population genetic model. In the model, cell populations start from the wild-type genotype, and they can acquire any of the three mutations present in the triple genotype. In the population of cells, a mutation can arise and then change in frequency until one of two outcomes happens: (i) the frequency of the mutation drops to zero, and the mutation is lost from the population or (ii) the frequency of the mutation reaches 1, when it is present in all the cells and hence is fixed in the population. When a mutation fixes in a population, we consider the genotype of the population to change and that constitutes a “step” on the fitness landscape. We assume a “strong selection weak mutation” regime, where there is no more than one mutation simultaneously present with a frequency less than 1. We also assume that mutations appear randomly and with equal probabilities. Mutations can appear and get lost multiple times in a population, and as long as populations have at least one mutation that increases fitness, one of those mutations will fix in the population eventually.
With the model we are estimating the probability of each adaptive step given that the population starts from the wild-type state. Therefore, the probability of each adaptive step will be influenced by the probabilities of previous step(s) and the sum of probabilities of adaptive steps originating from a given genotype must equal the sum of probabilities of all adaptive steps terminating in the given genotype. If there are multiple adaptive steps originating from the same genotype, they will have probabilities proportional to the fixation probabilities of their respective mutations. The fixation probability of a mutation is proportional to its selective advantage - [106]. As an example, if there are two adaptive steps originating from a genotype with fitness 1.00, one terminating in a genotype with fitness 1.1, the other in a genotype with fitness 1.2, then they have 10% and 20% selective advantage, respectively. Therefore, one adaptive step will happen half as likely as the other, as the selective advantages and therefore the relative fixation probabilities are in a ratio of 1:2.
Targeted sequencing of oncogenic loci for potential spontaneous oncogenic mutation
To determine whether the tumors that develop contained spontaneous oncogenic mutations, we performed Sanger Sequencing and Illumina sequencing (HiSeq 2500 platform; read length 2×150 bp, Admera Health Biopharma Services) on select regions of Kras, Egfr, Braf, and Nras (the 4 frequently mutated oncogenes in lung adenocarcinoma).
PCR products were obtained through amplification with primers listed below on DNA extracted from dissected tumors (Table S2) and cleaned up using ExoSAP (ThermoFisher Scientific, Cat# 78-201) treatment before Sanger.
Kras Codon 12+13 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNTTATTTTTATTGTAAGGCC TGCT
Kras Codon 12+13 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNTTACAAGCGCACGCAG A
Kras Codon 61 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNCCTGTCTCTTGGATATT CTCGAC
Kras Codon 61 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNCAGTTCTCATGTAC TGGTCCCT
Egfr Codon 721 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNCCAGCGGAGAAGCTCC AAAC
Egfr Codon 721 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNATACACTGTGCCAA ATGCTCCC
Egfr Codon 734-756 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNTCTTCTTAATCTCAGGGTC TCTGG
Egfr Codon 734-756 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNCACGTCAAGGATTTCT TTGTTGGC
Egfr Codon 764-793 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNTTACCCAGAAAGGGAT ATGCGTG
Egfr Codon 764-793 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNGGCAACCGTAGGG CATGAG
Egfr Codon 860+863 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNGTGAAGACACCACAGCAT GTCAAG
Egfr Codon 860+863 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNGCTTCCTGATCTACTC CCAGGAC
Braf Codon 503-509 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNGACTGGGAGATTCCTG ATGGAC
Braf Codon 503-509 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNcgtgttatacataccatgtccca c
Braf Codon 637 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNGACCTCACGGTAAAAATA GGTGAC
Braf Codon 637 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNAACTGTTCAAACTGAT GGGACC
Nras Codon 12+13 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNNTTCTACAGGTTTTTGCT GGTGTG
Nras Codon 12+13 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNNNGATTAGCTGGATCG TCAAGGC
Nras Codon 61 Forward Primer:
ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNCGAAAGCAAGTGGTGATT GATGG
Nras Codon 61 Reverse Primer:
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNNNNAAATACACAGAGGAA CCCTTCG
N: random nucleotides added to increase the diversity of PCR products for Illumina Sequencing.
Illumina sequencing was performed on pools of amplicons. The libraries were pooled based on band intensity to ensure even read depth and cleaned up using Sera-Mag Select beads (Thermo Fisher Scientific, Cat# 09-928-107) before undergoing a second round of PCR to attach the sequencing adaptors needed for the HiSeq platform. Second round PCR products were then purified with Sera-Mag Select beads before sequencing.
P5 adapter with i5 Index:
AATGATACGGCGACCACCGAGATCTACACNNNNNNNNacactctttccctacacgac
P7 adapter with i7 Index:
CAAGCAGAAGACGGCATACGAGATNNNNNNgtgactggagttcagacgtg
N’s represent i5 and i7 indices.
Analysis of targeted DNA-sequencing of Kras, Egfr, Braf, and Nras oncogenic loci
Sequenced reads were analyzed using Genome Analysis Toolkit (GATK, Broad Institute [107]). “Somatic short variant discovery” best practices pipeline for tumor samples similarly as for whole exome sequencing (see below). However, for targeted sequencing, identification of duplicate reads (Picard MarkDuplicates algorithm) was omitted as that would result in the loss of reads with matching start and end position, which is normal in targeted sequencing and is not a sign of duplicate artifacts. A mean coverage of 6665-7584 reads was achieved for all samples with 90% of regions having a coverage over 275 reads in all samples. Variant calls made and filtered by GATK Mutect2 function were annotated with Ensembl Variant Effect Predictor [108]. Pick-allele-gene option was used to filter results on the most relevant transcript for each variation. We filtered the results for the known oncogenic codons listed above and variants with a minimum of 5% allele frequency.
Whole exome sequencing
DNA was extracted from 4 individual tumors from TC mice transduced with Lenti-sgNf1- sgRasa1-sgPten, three months after tumor initiation, using Qiagen AllPrep DNA/RNA Micro kit. Whole-exome sequencing library preparation was performed by Admera Health using SureSelect XT Mouse All Exon Kit (Agilent).
Sequenced reads on autosomes were analyzed using Genome Analysis Toolkit (GATK, Broad Institute [107]) “Somatic short variant discovery” best practices pipeline for tumor samples. Mean coverage of 50-72 reads was achieved for all samples, with 90% of regions having coverage over 20 reads in all samples. Variant calls made and filtered by GATK Mutect2 function and were annotated with Ensembl Variant Effect Predictor (VEP [108]). The pick- allele-gene option was used to filter results on the most relevant transcript for each variant. The same exact variants appearing in multiple tumor samples were flagged as germline variant and were removed. We filtered the results for protein-coding variation, variants with a minimum of 5% allele frequency, and removed variations in the olfactory OLFR gene family that are likely germline variations.
Analysis of insertion and deletions
Indel analysis was performed to confirm that insertion and deletions (indels) were generated at the targeted loci as follows: gDNA was isolated from at oncogene-negative mouse cell lines or FACS-sorted Tomatopositive cancer cells using either the AllPrep DNA/RNA(Qiagen) or the DNeasy Blood and Tissue Kit. PCR primers were designed to amplify sgRNA-targeted loci, resulting in 500 to 1000 bp amplicons specific to each locus. Amplicons were purified using PCR purification kit (Qiagen) and sequenced by Sanger sequencing. Cutting efficiency was determined by ICE analysis (https://ice.synthego.com/#/)
Nf1-Amplification Forward primer: GCAATTTTGGGGGAACGCCT
Nf1-Amplification Reverse primer: AAAACCAAGAGAGGTCAGAGCC
Nf1-Sequencing primer: CAGCGATTCTAAAATACCAATGC
Rasa1-Amplification Forward primer: GGAGCACGGTATGTGTCGTT
Rasa1-Amplification Reverse primer: TCCTCTTTAGCGTAGCCAGGAA
Rasa1-Sequencing primer: TTGGTGAAAGCGACGTCTC
Pten-Amplification Forward primer: TGAATACACAGTGGCCTTTGCTT
Pten-Amplification Reverse primer: CAGAGACTGCATCTGGTGGTT
Pten-Sequencing primer: CATTGGGTTAGCTTTCTTAACC
Histology and immunohistochemistry
Lung lobes were inflated with 4% formalin and fixed for 24 hours, stored in 70% ethanol, paraffin-embedded, and sectioned. 4 μm thick sections were used for Hematoxylin and Eosin (H&E) staining and immunohistochemistry (IHC).
Primary antibodies used for IHC were anti-RFP (Rockland, 600-401-379), anti- TTF1(Abcam, ab76013), anti-UCHL1(Sigma, HPA005993), anti-TP63 (Cell Signaling Technology, 13109), anti-phospho-S6 (Cell Signaling Technology, 4858), anti-PTEN (Cell Signaling Technology, 9559), anti-phospho-ERK (Cell Signaling Technology, 4370), anti- phospho-AKT (Thermo Fisher Scientific, 44-621G), and anti-HMHGA2 (Biocheck, 59170AP). IHC was performed using Avidin/Biotin Blocking Kit (Vector Laboratories, SP-2001), Avidin- Biotin Complex kit (Vector Laboratories, PK-4001), and DAB Peroxidase Substrate Kit (Vector Laboratories, SK-4100) following standard protocols.
Images of the H&E-stained slides were analyzed with ImageJ. Tumor areas were converted from pixels to mm2 via a ruler. To quantify the positivity of phospho-ERK and phospho-AKT stained slides, H-scores were calculated using Qupath. The H-score is determined by adding the results of multiplication of the percentage of cells with staining intensity ordinal value (scored from 0 for “no signal” to 3 for “strong signal”) with possible values ranging from 0 to 300 [109]. To normalize potential variations between different rounds of immunohistochemistry, one patient sample was included and stained for both pERK and pAKT in all rounds of staining as a control.
Immunoblotting
3 × 105 cells were seeded into 6-well plates and allowed to adhere overnight in regular growth media and cultured in the presence or absence of 10 µM of Capivasertib, RMC-4550, or a combination of both drugs. After 24 hours, the protein was extracted using RIPA lysis buffer (Thermo Fisher Scientific, 89900) and proteinase/phosphatase inhibitor cocktail (Thermo Fisher Scientific, 78442). Protein concentration was measured using BCA protein assay kit (Thermo Fisher Scientific, 23250). Proteins (30 µg from each sample) were separated by SDS-PAGE and immunoblotted and transferred to polyvinyl difluoride (PVDF) membranes (BioRad, 162-0177) according to standard protocols. Membranes were immunoblotted with antibodies against phosphor-ERK (Cell Signaling Technology, 4370), ERK (Cell Signaling Technology, 9102), phosphor-AKT (Thermo Fisher Scientific, 44-621G), AKT (Cell Signaling Technology, 4691), phospho-S6 (Cell Signaling Technology, 4858), S6 (Cell Signaling Technology, 2217), anti- RASA1 (Abcam, ab2922), anti-PTEN (Cell Signaling Technology, 9559), and HSP90 (BD Bioscience, 610418). Immunoblots were developed using Supersignal® West Dura Extended Duration Chemiluminescent Substrate (Thermo Fisher Scientific, 37071). Initially, the membranes were immunoblotted against non-phosphorylated targets, and after stripping these antibodies using Western Blot Stripping Buffer (Thermo Fisher Scientific, 46430), they were immunoblotted against phosphorylated antibodies. Developing the signal was done using Dura Extended Duration Chemiluminescent Substrate (Thermo Fisher Scientific, 37071). All immunoblots were performed at least three times independently.
Cell Lines and Reagents
Mouse oncogene-negative cell lines were generated from tumors initiated in Trp53flox/flox;TC BL6 mice four months after transduction with Lenti-sgNf1-sgRasa1-sgPten/Cre. After dissociation of tumors (described below), cells were cultured in DMEM supplemented with 10% FBS, 1% penicillin/streptomycin (Gibco), and 0.1% Amphotericin (Life Technologies). HC494 and MMW389T2 (KrasG12D and Trp53 mutant) lung adenocarcinoma cells were previously generated in the Winslow Lab. Human oncogene-negative cell lines (NCI-H1838, NCI-H1623) and oncogene-positive cell lines (A549, H2009, NCI-H2009, SW1573, HOP62, NCI-H358, NCI-H1792) were purchased from ATCC and cultured in RPMI supplemented with 5%FBS, 1% penicillin/streptomycin (Gibco), and 0.1% Amphotericin (Life Technologies). We performed mycoplasma testing using MycoAlert™ Mycoplasma Detection Kit (Lonza). Cell were maintained at 37°C in a humidified incubator at 5% CO2. Mutations in components of RAS and PI3K pathways of NCI-H1838, NCI-H1623 (based on Table S6) are indicated below (extracted from DepMap):
NCI-H1838 (RAS pathway): NF1 (p.N184fs), IQGAP2 (p.P780L)
NCI-H1623 (RAS pathway): RASA1(p.A47fs), FGFR2(p.A355S), ERF(p.G255C)
Clonogenic, apoptosis, and proliferation assays
For clonogenic assays, mouse cells were seeded in triplicate into 24-well plates (4000 cells per well) and allowed to adhere overnight in regular growth media. Cells were then cultured in the absence or presence of the drug as indicated on each figure panel in complete media for 4 days. Growth media with or without drugs was replaced every 2 days. The remaining cells were stained with 0.5% crystal violet in 20% methanol and photographed using a digital scanner. Relative growth was quantified by densitometry after extracting crystal violet from the stained cells using 100% methanol [110].
Clonogenic assay of human oncogene-negative lung adenocarcinoma cell lines were done in spheroids [57]. 400-5000 cells/well were seeded in round bottom ultra-low attachment 96-well plates (Corning) in growth media and incubated for 72 hours at 37°C in 5% CO2. Spheroid formation was confirmed visually, and spheroids were treated in triplicate with dilutions of RMC-4550 and capivasertib in complete growth media. Following drug exposure for five days, cell viability in spheroids was determined using the CellTiter-Glo 3D assay kit (Promega), following the manufacturer’s instructions. Luminescence was read in a Plate Reader. Assay data was normalized to DMSO values.
Drug synergism was analyzed using SynergyFinder (https://synergyfinder.fimm.fi) web application [111]. The degree of combination synergy, or antagonism, was quantified by comparing the observed drug combination response against the expected response, calculated using Loewe’s model that assumes no interaction between drugs [112].
For apoptosis and proliferation assays, 3 × 105 cells were seeded into 6-well plates, and allowed to adhere overnight in regular growth media, and cultured in the presence or absence of 10 µM of Capivasertib, RMC-4550, or a combination of both drugs. After 24 hours, apoptosis and cell proliferation were determined through staining with Fixable Viability Dye eFluor™ 450 (Thermo Fisher Scientific, 65-0863-14), cleaved caspase 3 Antibody (Cell Signaling Technology, 9669), and Click-iT™ EdU Alexa Fluor™ 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific, C-10424) according to the manufacturer’s instructions. Data were acquired using a BD LSR II Flow Cytometer. All experiments were performed independently two times on 3 different cell lines.
In vivo drug response studies
For drug efficacy studies in autochthonous mouse models, TC mice (8-12 weeks old) were divided into 4 groups randomly 3.5 months after tumour initiation. They received the vehicle, capivasertib (100 mg/kg, MedChemExpress), RMC-4550 (30 mg/kg, MedChemExpress), or a combination of both dissolved in 10% DMSO, 40% PEG, 5% Tween 80, and 45% PBS through a gavage needle. Mice were treated daily with drugs for eight days, and the treatment was stopped for two days for recovery, and it continued for two more days before the tissue harvest. The last two doses of combination therapy were half of the initial doses.
Cell line-derived allografts were generated through subcutaneous injection of 300,000 of MY-C3 (Nf1, Rasa1, Pten, and Trp53 mutant) oncogene-negative mouse cell line in 200 μl of PBS in male (6–8 week old) BL6 mice (two tumors per mouse). Once tumors reached an average size of ∼100 mm3 administration of RMC-4550 and capivasertib was started. They received the vehicle, and combination of capivasertib (100 mg/kg, MedChemExpress) and RMC-4550 (30 mg/kg, MedChemExpress) (5 days on, 2 days off) for 17 days.
Tumor dissociation, cell sorting, and RNA-sequencing
Primary tumors were dissociated using collagenase IV, dispase, and trypsin at 37 °C for 30 min. After dissociation, the samples remained continually on ice, were in contact with ice- cold solutions, and were in the presence of 2 mM EDTA and 1 U/ml DNase to prevent aggregation. Cells were stained with antibodies to CD45 (BioLegened, 103112), CD31 (BioLegend, 303116), F4/80 (BioLegend, 123116), and Ter119 (BioLegend, 116212) to exclude hematopoietic and endothelial cells (lineage-positive (Lin+) cells). DAPI was used to exclude dead cells. FACS Aria sorters (BD Biosciences) were used for cell sorting.
RNA was purified using RNA/DNA All Prep kit (Qiagen, 80284). RNA quality of each tumor sample was assessed using the RNA6000 PicoAssay for the Agilent 2100 Bioanalyzer as per the manufacturer’s recommendation. 4.4 ng total RNA per sample was used for cDNA synthesis and library preparation using Trio RNA-Seq, Mouse rRNA kit (Tecan, 0507-32), according to the manufacturer’s instructions. The purified cDNA library products were evaluated using the Agilent bioanalyzer and sequenced on NextSeq High Output 1×75 (Admera Health Biopharma Services).
Analysis of mouse model-derived RNA-seq datasets
Paired-end RNA-seq reads were aligned to the mm10 mouse genome using STAR (v2.6.1d) 2-pass mapping and estimates of transcript abundance were obtained using RSEM (v1.2.30) [113, 114]. The differentially expressed genes between different tumor genotypes and treatment groups were called by DESeq2 using transcript abundance estimates via tximport [115, 116]. The DESeq2-calculated fold changes were used to generate ranked gene lists for input into GSEA [117].
The upregulated genes with absolute log2 fold change greater than 1 and a false discovery rate less than 0.05 in the comparison of Nf1, Rasa1, and Pten mutant oncogene- negative tumors with KrasG12D-driven tumors (KTC+sgInert and KTC+sgPten) were compiled into a signature reflecting the oncogene-negative adenocarcinoma state. This gene signature was utilized in the analysis of human oncogene-positive and oncogene-negative tumors. Scaled estimates of transcript abundance for TCGA LUAD samples were obtained from the GDC data portal (gdc-portal.nci.nih.gov). Each expression profile was then scored on the basis of the mouse-derived gene signature using single-sample GSEA within the Gene Set Variation Analysis (GSVA) package [118].
Data availability
Tuba-seq barcode sequencing and RNA-seq data have been deposited in NCBI’s Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) and are accessible through GEO Series accession number GSE174393. Whole exome sequencing data generated in our study are publicly available in SRA-NCBI (www.ncbi.nlm.nih.gov/sra), under BioProject accession number PRJNA769722.
ACKNOWLEDGEMENTS
We thank the Stanford Shared FACS Facility for flow cytometry and cell sorting services, the Stanford Veterinary Animal Care Staff for expert animal care, Human Pathology/Histology Service Center, Stanford Protein and Nucleic Acid Facility; A. Orantes andS. Mello for administrative support; Stanford’s Molecular Genetic Pathology Laboratory and Henning Stehr for their help in providing genetically profiled tumor tissues. David Feldser, Joseph Lipsick, Eric Collisson, Christopher McFarland, and members of the Winslow and Petrov laboratories for helpful discussions and reviewing the manuscript. We thank Florent Elefteriou and Alejandro Sweet-Cordero for providing mouse strains. M.Y. was supported by a Stanford University School of Medicine Dean’s fellowship, an American Lung Association senior research training grant, and an NIH Ruth L. Kirschstein National Research Service Award (F32- CA236311). G.B., H.C., and J.D.H. were supported by a Tobacco-Related Disease Research Program (TRDRP) Postdoctoral Fellowships (T31FT-1772, 28FT-0019, and T31FT-1619). C.W.M. was supported by the NSF Graduate Research Fellowship Program and an Anne T. and Robert M. Bass Stanford Graduate Fellowship. W-Y.L. was supported by an American Association of Cancer Research Postdoctoral fellowship (17-40-18-LIN). C.L. was the Connie and Bob Lurie Fellow of the Damon Runyon Cancer Research Foundation (DRG-2331). E.L.A and C.I.C were supported by PHS Grant Number CA09302, awarded by the National Cancer Institute, DHHS. E.L.A. was also supported by HHMI Gilliam Fellowship for Advanced Study (GT14928). This work was supported by NIH R01-CA231253 (to M.M.W and D.A.P), NIH R01-CA230919 (to M.M.W.) and NIH R01-CA234349 (to M.M.W and D.A.P.), as well as by the Stanford Cancer Institute, an NCI-designated Comprehensive Cancer Center.
Acknowledgments
The results in Figures 1 and 4 are in part based upon data generated by the TCGA Research Network (https://www.cancer.gov/tcga) and Genomics Evidence Neoplasia Information Exchange (GENIE).
Footnotes
↵8 These authors contributed equally
CONFLICT OF INTERESTS S.K.C. receives grant support from Ono Pharma. C.S. acknowledges grant support from Pfizer, AstraZeneca, Bristol Myers Squibb, Roche-Ventana, Boehringer-Ingelheim, Archer Dx, and Ono Pharmaceuticals. C.S is an AstraZeneca Advisory Board member and Chief Investigator for the MeRmaiD1 clinical trial, has consulted for Pfizer, Novartis, GlaxoSmithKline, MSD, Bristol Myers Squibb, Celgene, AstraZeneca, Illumina, Amgen, Genentech, Roche-Ventana, GRAIL, Medicxi, Bicycle Therapeutics, and the Sarah Cannon Research Institute, has stock options in Apogen Biotechnologies, Epic Bioscience, GRAIL, and has stock options and is co-founder of Achilles Therapeutics. D.A.P. and M.M.W. are founders of, and hold equity in, D2G Oncology Inc.
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.
- 10.
- 11.
- 12.
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.
- 19.
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.
- 40.
- 41.↵
- 42.
- 43.
- 44.
- 45.
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.
- 69.
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.
- 82.
- 83.
- 84.
- 85.
- 86.
- 87.
- 88.↵
- 89.↵
- 90.↵
- 91.
- 92.
- 93.
- 94.
- 95.
- 96.
- 97.
- 98.↵
- 99.↵
- 100.
- 101.
- 102.
- 103.↵
- 104.↵
- 105.↵
- 106.↵
- 107.↵
- 108.↵
- 109.↵
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.↵
- 116.↵
- 117.↵
- 118.↵