CRISPR/Cas9 screen identifies KRAS-induced COX-2 as a driver of immunotherapy resistance in lung cancer

Oncogenic KRAS impairs anti-tumour immune responses, but effective strategies to combine KRAS inhibitors and immunotherapies have so far proven elusive. In vivo CRISPR-Cas9 screening in an immunogenic murine lung cancer model identifies mechanisms by which oncogenic KRAS promotes immune evasion, most notably expression of immunosuppressive cyclooxygenase-2 (COX-2) in cancer cells. Oncogenic KRAS was a potent inducer of COX-2 in both mouse and human lung cancer which was suppressed using KRAS inhibitors. COX-2 acting via prostaglandin E2 (PGE2) promotes resistance to immune checkpoint blockade (ICB) in both mouse and human lung adenocarcinoma. Targeting COX-2/PGE2 remodelled the tumour microenvironment by inducing pro-inflammatory polarisation of myeloid cells and influx of activated cytotoxic CD8+ T cells, which increased the efficacy of ICB. Restoration of COX-2 expression contributed to tumour relapse after prolonged KRAS inhibition. We propose testing COX-2/PGE2 pathway inhibitors in combination with KRAS G12C inhibition or ICB in patients with KRAS-mutant lung cancer.


ABSTRACT
Oncogenic KRAS impairs anti-tumour immune responses, but effective strategies to combine KRAS inhibitors and immunotherapies have so far proven elusive. In vivo CRISPR-Cas9 screening in an immunogenic murine lung cancer model identifies mechanisms by which oncogenic KRAS promotes immune evasion, most notably expression of immunosuppressive cyclooxygenase-2 (COX-2) in cancer cells.
Oncogenic KRAS was a potent inducer of COX-2 in both mouse and human lung cancer which was suppressed using KRAS inhibitors. COX-2 acting via prostaglandin E2 (PGE2) promotes resistance to immune checkpoint blockade (ICB) in both mouse and human lung adenocarcinoma. Targeting COX-2/PGE2 remodelled the tumour microenvironment by inducing pro-inflammatory polarisation of myeloid cells and influx of activated cytotoxic CD8+ T cells, which increased the efficacy of ICB. Restoration of COX-2 expression contributed to tumour relapse after prolonged KRAS inhibition. We propose testing COX-2/PGE2 pathway inhibitors in combination with KRAS G12C inhibition or ICB in patients with KRAS-mutant lung cancer.

INTRODUCTION
Despite improvements in systemic therapies, lung adenocarcinoma (LUAD) remains the most common cause of cancer-related deaths worldwide (1). Immune checkpoint blockade (ICB), which can reinvigorate anti-tumour immunity, has shown remarkable clinical success in multiple cancer types (2), including LUAD (3), achieving durable responses in a subset of patients. Monoclonal antibodies targeting the immunosuppressive PD-L1/PD-1 axis have become standard of care for LUAD patients, either as a monotherapy (4) or in combination with chemotherapy (5).
However, only a fraction of patients benefits from ICB highlighting the need for combination strategies that will broaden responses to current immunotherapies.
Recent clinical efforts, such as the SKYSCRAPER-01 trial combining PD-L1 blockade with an anti-TIGIT antibody, have failed to lead to improved responses in lung cancer and there is a need to better understand mechanisms of immune evasion in order to design rational combination strategies that are more likely to provide clinical benefit.
Mutations in the oncogene KRAS drive tumourigenesis in 30% of LUAD cases (6). However, the development of inhibitors that directly target KRAS has been notoriously challenging (7). A major breakthrough was achieved with the recent development of mutant-specific KRAS G12C inhibitors (8) which covalently bind to the novel cysteine residue present in nearly half of all KRAS-mutant LUAD patients (9).
This has led to the approval of Amgen's clinical compound sotorasib for the treatment of locally advanced or metastatic KRAS G12C -mutant lung cancer (10).
Whilst these drugs have mild toxicities and achieve clinical responses in a 4 substantial proportion of patients, as with other targeted therapies, responses are often short-lived as resistance inevitably arises (11).
Accumulating evidence suggests that oncogenic signalling extends beyond the tumour cell compartment and engages with the host stromal and immune compartments. Consequently, oncogenic drivers have been shown to play dominant roles in shaping the tumour immune landscape of different cancers and inhibiting anti-tumour immune responses (12). Analysis of clinical samples has demonstrated that KRAS mutations are associated with an immunosuppressive tumour microenvironment (TME) (13) and preclinical studies have identified a number of mechanisms by which oncogenic KRAS can drive immune evasion including promoting the expression of numerous immunosuppressive myeloid chemoattractants (14,15) and immune checkpoint ligands (16). Together these observations provide a rational basis for combining KRAS inhibitors with ICB and numerous preclinical studies have demonstrated that this combination leads to improved therapeutic responses, at least in immunogenic models (15,17,18).
However, recent reports of the CODEBREAK 100/101 clinical trial evaluation of sotorasib in combination with anti-PD-L1/PD-1 antibodies have shown serious toxicities (19,20), casting doubt on the viability of this combination. As an alternative approach, a greater understanding of how oncogenic KRAS drives immune evasion may identify novel immunotherapy combination strategies that could improve outcomes for KRAS-mutant lung cancer patients.
Pooled CRISPR screens have been increasingly used to uncover tumourintrinsic determinants of anti-tumour immunity, identifying numerous genes that either promote sensitivity or resistance to immune control (21-25) . We recently developed a novel immunogenic model of KRAS-mutant lung cancer allowing for the 5 preclinical study of tumour-immune interactions (17). Here we carry out a pooled in vivo CRISPR screen in this novel model to interrogate the role of 240 KRASregulated genes in controlling anti-tumour immunity. This identified several genes that increased sensitivity or resistance to anti-tumour immune responses. Among these, the prostaglandin synthase cyclooxygenase-2 (COX-2), responsible for synthesis of the immunosuppressive molecule prostaglandin E2 (PGE2), was identified as a major driver of immune evasion and resistance to immunotherapy.
Targeting of the COX2/PGE2 axis improved the response of KRAS-mutant lung tumours to anti-PD-1 therapy by inducing pro-inflammatory polarisation of myeloid cells and enhancing T cell infiltration and activation. Importantly, oncogenic KRAS signalling was a strong driver of the COX-2/PGE2 signalling axis in both mouse and human LUAD and COX-2 inhibition delayed tumour relapse after KRAS G12C inhibition.

In vivo CRISPR screen identifies tumour-intrinsic determinants of anti-tumour immunity
It is challenging to carry out large scale genome-wide screens in vivo, whilst also maintaining a sufficiently high representation of the pooled library, as there is a limit on the number of cells that can be orthotopically transplanted. Instead, a rationally selected, smaller, customised library was used. First, we generated a library of lentiviral vectors encoding sgRNAs targeting 240 genes that are regulated by KRAS in human LUAD (Supplementary Table S1). KRAS-regulated genes were identified by differential gene expression analysis of TCGA LUAD samples and LUAD cell lines from the Cancer Cell Line Encyclopedia (CCLE) which were stratified as having high or low RAS pathway activity using a novel RAS transcriptional signature (26). Additional genes were identified using RNA-seq data from KRAS G12Cmutant LUAD cell lines (H358 and H23) treated with a KRAS inhibitor and immortalised type II pneumocytes expressing an ER-KRAS G12V fusion protein which can be readily activated by administration of 4-hydroxytamoxifen (4-OHT) (15). To carry out the screen we utilised the immunogenic KPAR cell line, derived from a genetic KRAS G12D p53 -/lung cancer mouse model as it stimulates endogenous antitumour immune responses and is partially responsive to immunotherapy (17). Next, we engineered the immunogenic KRAS G12D p53 -/-KPAR cell line to express Cas9 under a doxycycline-inducible promoter ( Supplementary Fig. 1A). An inducible system was chosen as it allows temporal control of editing and circumvents any compounding consequences of Cas9 immunogenicity in vivo. Importantly, Cas9 7 expression was abrogated in vitro 48 hours after the removal of doxycycline ( Supplementary Fig. 1B) and was not re-expressed in vivo ( Supplementary Fig. 1C).
Library-transduced KPAR iCas9 cells were treated with doxycycline for four days to allow gene-editing to occur followed by a two-day washout period to abrogate the expression of Cas9 before orthotopic transplantation into C57BL/6 (WT) mice or Rag2 -/-;Il2rg -/mice which lack T cells, B cells and NK cells and therefore do not exert anti-tumour immune responses (Fig. 1A). After three weeks, genomic DNA was isolated from tumour-bearing lungs and subject to deep nextgeneration sequencing (NGS) to compare library representation in tumours growing in immune-competent and immune-deficient mice. In parallel, NGS of genomic DNA from cells passaged in vitro was carried out to identify genes that affect cell viability.
Analysis of genes targeted by sgRNAs that were depleted in vitro identified a number of genes known to affect cell viability in KRAS-mutant tumour cells, including Myc (27) and Fosl1 (28), thereby validating the functionality of the KRAS-target sgRNA library (Fig. 1B). Furthermore, a number of sgRNAs were equally depleted in immune-competent and immune-deficient mice compared to in vitro passaged cells, including those targeting the EMT regulator Zeb1 and the anti-apoptotic caspase inhibitor c-FLIP (encoded by Cflar) (Fig. 1C). These genes therefore supported tumour growth in vivo by mechanisms independent of anti-tumour immunity.
By comparing sgRNAs that were depleted or enriched in immune-competent compared to immune-deficient mice we identified genes that modulate anti-tumour immune responses (Fig. 1D). The most enriched sgRNAs in immune-competent mice were targeting a subunit of the IFNg receptor (Ifngr2) (Fig. 1E). Furthermore, we also saw enrichment of sgRNAs targeting the Ets transcription factor ETV4 ( Supplementary Fig. 1D). Importantly, Etv4 -/-KPAR tumours grew faster than 8 parental tumours in WT mice but not in Rag2 -/-;Il2rg -/mice ( Supplementary Fig. 1E), confirming a role for ETV4 in sensitising tumours to anti-tumour immune responses.

IFNg signalling
Tumour-intrinsic IFNg signalling has previously been shown to be required for anti-tumour immune responses in carcinogen-induced mouse models of cancer (29) and responses to immunotherapy in melanoma (30). To validate the role of tumourintrinsic IFNg signalling in anti-tumour immune responses uncovered in the screen ( Fig. 1E) we generated Ifngr2 -/-KPAR cell lines using CRISPR-Cas9. Flow cytometry confirmed that Ifngr2 -/-KPAR cells had lost the expression of the IFNg-receptor b subunit (Supplementary Fig. 2A). As expected, Ifngr2 -/-KPAR cells were insensitive to IFNg in vitro as we were unable to detect pSTAT1 in response to IFNg ( Supplementary Fig. 2B). Whilst Ifngr2 -/cells grew similarly to parental cells in vitro ( Supplementary Fig. 2C), they grew faster than parental cells when transplanted into immune-competent mice ( Fig. 2A). In contrast Ifngr2 -/and parental tumours grew at 9 similar rates in immune-deficient Rag2 -/-;Il2rg -/mice. These data were validated using a second clone ( Supplementary Fig. 2D) indicating that intact tumour-intrinsic IFNg signalling is required for effective anti-tumour immunity.
Previous studies have demonstrated that oncogenic KRAS regulates tumourintrinsic IFN pathway gene expression (15).

Tumour-intrinsic COX-2 suppresses innate and adaptive anti-tumour immunity
Ptgs2 loss was the strongest sensitiser to anti-tumour immunity in the screen and encodes the enzyme COX-2 which is overexpressed in many cancer types.
COX-2 is responsible for the synthesis of the prostanoid PGE2, which has been shown to suppress anti-tumour immunity in preclinical models of colorectal cancer and melanoma (33). To validate the results obtained in the screen (Fig. 1F), we 11 began by generating Ptgs2 -/-KPAR cell lines using CRISPR-Cas9. Western blotting confirmed the loss of COX-2 expression and PGE2 production in Ptgs2 -/cells ( Supplementary Fig. 3A). We did not observe any difference in the growth of Ptgs2 -/cells and parental cells in vitro ( Supplementary Fig. 3B). However, they grew considerably slower when orthotopically transplanted into immune-competent mice, which as a result had significantly increased survival, with 60% of mice experiencing complete rejection (Fig. 3A). Importantly, COX-2-deficient tumours grew similarly to parental tumours when transplanted into immune-deficient Rag2 -/-;Il2rg -/mice. This was further validated using a second clone ( Supplementary Fig. 3C). Therefore, Ptgs2 -/cell lines showed no cell-autonomous defects in tumour progression but were instead sensitised to anti-tumour immune responses, with immunological rejection occurring in a substantial proportion of mice.
As Rag2 -/-;Il2rg -/mice lack NK cells, T cells and B cells, we wanted to decipher the contribution of innate and adaptive immunity to the reduced growth of COX-2-deficient tumours in immune-competent mice. COX-2-deficient tumours grew faster in mice treated with antibodies depleting either NK cells or CD8 + T cells and grew fastest in mice lacking both subsets (Fig. 3B). Interestingly, tumours grew faster in mice lacking NK cells compared to mice lacking CD8 + T cells, suggesting the innate immune response was largely responsible for the impaired growth of COX-2-deficient tumours, as previously reported (34). However, no mice survived long-term in the absence of CD8 + T cells, demonstrating that the combined action of innate and adaptive immunity was required for tumour rejection. NK cells play a critical role in the control of orthotopic lung tumours during tumour cell seeding in the lung. To ensure the control of COX-2-deficient tumours was not exacerbated by the route of injection we also compared the growth of subcutaneous parental and 12 Ptgs2 -/tumours. Similar to the orthotopic setting, COX-2-deficient subcutaneous tumours grew significantly slower in immune-competent mice ( Supplementary Fig.   3D). Consistent with the role of both innate and adaptive immunity in the rejection of COX-2-deficient tumours, we observed increased frequencies of CD8 + T cells and NK cells as well as CD4 + T cells and Tregs in Ptgs2 -/tumours (Fig. 3C). Increased infiltration of COX-2-deficient tumours by NK cells was confirmed by immunohistochemistry (Fig. 3D).

COX-2/PGE2 signalling drives resistance to ICB in mouse and human LUAD
Given that tumour-intrinsic COX-2 acts as a major driver of immune evasion in KPAR tumours we wanted to assess whether it also contributed to resistance to ICB.
Mice were orthotopically transplanted with parental KPAR or Ptgs2 -/cells and treated with anti-PD-1. Whilst parental KPAR tumours were partially responsive to anti-PD-1, COX-2 deficient tumours were significantly more sensitive to PD-1 blockade with all ICB-treated mice bearing Ptgs2 -/lung tumours surviving long-term ( Fig. 4A). Immunohistochemistry revealed that similar to ICB-treated KPAR tumours, Ptgs2 -/lung tumours were more infiltrated by CD8 + T cells (Fig. 4B), confirming what 14 we observed by flow cytometry (Fig. 3C), however the biggest increase was seen in ICB-treated Ptgs2 -/tumours. Furthermore, flow cytometry analysis demonstrated that PD-1 blockade only led to increased activation of CD8 + T cells in Ptgs2 -/lung tumours (Fig. 4C). In addition, ICB-treated Ptgs2 -/lung tumours showed the greatest expansion of effector memory CD8 + T cells ( Supplementary Fig. 4A) and upregulation of checkpoint molecules (Fig. 4D). Interestingly, anti-PD-1 treatment also led to increased NK cell infiltration in Ptgs2 -/lung tumours, which did not occur in KPAR tumours ( Supplementary Fig. 4B). Furthermore, gene expression analysis revealed that anti-PD-1 induced robust expression of anti-tumour immunity genes only in COX-2-deficient lung tumours (Fig. 4E, Supplementary Fig. 4C). Together these data suggest that tumour-intrinsic COX-2 promotes resistance to ICB by preventing the stimulation of anti-tumour immunity in response to PD-1 blockade.
To determine whether COX-2/PGE2 signalling also affected the clinical response of lung cancer patients to immunotherapy we examined the expression of a previously published COX-2-associated inflammatory gene expression signature (COX-IS) (34) in a cohort of LUAD patients treated with anti-PD-L1/PD-1 for which baseline expression data was available (35). Importantly, expression of the COX-IS was significantly higher in LUAD patients who did not respond to ICB (Fig. 4F).
Furthermore, higher COX-IS expression was associated with significantly worse progression-free survival following ICB (Fig. 4G) and was also predictive of outcome independent of age, gender, smoking status and previous lines of therapy (Fig. 4H).
These results support the notion, as suggested by the mouse model, that the COX-2/PGE2 axis drives immunosuppression and hinders response to ICB in human LUAD.

Inhibition of the COX-2/PGE2 axis delays tumour growth and synergises with ICB
Whilst genetic deletion of tumour-intrinsic COX-2 resulted in a drastic repolarisation of the tumour microenvironment, increased tumour control and sensitisation to ICB, we next sought to assess whether pharmacological blockade of COX-2 could have similar effects. We treated KPAR lung tumour-bearing mice with the COX-2 specific inhibitor celecoxib, which was administered by daily oral gavage. Importantly, celecoxib significantly extended the survival of KPAR-tumour bearing mice to a similar extent seen with PD-1 blockade (Fig. 5E). However, the combination of both celecoxib and anti-PD-1 showed superior efficacy compared to either single-agent alone. Indeed, both celecoxib and anti-PD-1 increased infiltration of tumours with CD8 + T cells which was further increased in the combination treatment arm (Fig. 5F). Furthermore, only the combination treatment led to a significant expansion of effector memory CD8 + T cells (Fig. 5G) and upregulation of checkpoint molecules on both CD8 + and CD4 + T cells ( Supplementary Fig. 5D).

As seen in
Combination treatment also induced the highest levels of PD-L1 on several myeloid cell types ( Supplementary Fig. 5E). This could be due to elevated levels of IFNg ( Fig.   5H) which was significantly upregulated in the combination treatment arm along with other anti-tumour immunity genes.

16
In the clinic, celecoxib has been associated with increased cardiovascular risk (36), which prompted us to explore other therapeutic options to target this immunosuppressive axis. Immune cells express four receptors for PGE2, EP1-4.
As seen with celecoxib, dual EP2-EP4 inhibition led to a significant increase in the  Fig. 5F). In conclusion, these results suggest that pharmacological inhibition of the COX-2/PGE2 axis reverses immunosuppression in the TME, promoting adaptive immunity which enhances the therapeutic efficacy of ICB.

COX-2 expression is driven by oncogenic KRAS and contributes to tumour relapse after KRAS G12C inhibition
Given the known role of oncogenic KRAS in mediating immune evasion we next wanted to understand whether tumour-intrinsic COX-2 expression was 17 regulated by KRAS signalling. To test this, we inhibited KRAS signalling in several mouse and human KRAS-mutant cancer cell lines. Firstly, treatment of KPAR cells in vitro with the MEK inhibitor trametinib led to a drastic reduction in COX-2 protein expression and loss of PGE2 secretion (Fig. 7A). We validated this in the 3LL The advent of KRAS G12C inhibitors has transformed the treatment landscape for KRAS G12C -mutant lung cancer patients. However, responses are often short-lived and combination therapies will be required to overcome the development of adaptive resistance. Whilst oncogenic KRAS pathway reactivation will re-engage proliferative signalling we postulated that restoration of KRAS-mediated immunosuppression may also contribute to tumour relapse. Interestingly, COX-2 expression was restored in 19 KPAR G12C cells in vitro after long-term MRTX849 treatment (Fig. 7I) and in vivo in KPAR G12C tumours that relapsed after MRTX849 treatment (Fig. 7J). Importantly, the combination of MRTX849 and celecoxib delayed tumour relapse and significantly improved the survival of KPAR G12C tumour-bearing mice compared to MRTX849 alone (Fig. 7K).
In summary, these data suggest that oncogenic KRAS is a major driver of COX-2/PGE2 signalling in mouse and human lung cancer and may therefore We demonstrated that tumour-intrinsic IFNg signalling is critical for anti-tumour immunity. Indeed, numerous CRISPR screens have demonstrated that defects in IFNg signalling result in resistance to immunotherapy (21,41) and mutations in JAK1 and JAK2 have been associated with acquired resistance to ICB (42). Given the importance of tumour-intrinsic IFNg signalling in sensitising KPAR tumours to antitumour immune responses, the ability of oncogenic KRAS to suppress IFN pathway signalling may represent a major mechanism of immune evasion in KRAS-mutant LUAD. These results are consistent with our previous findings that KRAS G12C inhibitors can restore tumour-intrinsic IFN signalling in multiple preclinical models of lung cancer (15). Paradoxically, tumour-intrinsic IFNg signalling has also been shown 21 to impede anti-tumour immunity (43), including in a CRISPR screen using an orthotopic KRAS G12D p53 -/lung cancer model similar to the one used in this study (25). Whilst such opposing effects in similar models are surprising, chronic tumourintrinsic IFNg signalling can drive resistance to ICB due to upregulation of immune checkpoint ligands (44). Understanding the contexts in which tumour-intrinsic IFNg signalling promotes or impedes anti-tumour immunity will be important when assessing combination strategies targeting this pathway.
We also identified KRAS-driven expression of COX-2 and secretion of PGE2 as a major mechanism of immune evasion which suppresses both innate and adaptive anti-tumour immune responses. This is consistent with the role of tumourintrinsic COX-2 in driving immune evasion in melanoma and colorectal cancer (33).
The ability of our CRISPR screen to elucidate the function of a secreted molecule can be explained by a recent barcoded CRISPR screen which demonstrated that orthotopic KP tumours grow as clonal lesions (25), suggesting that local secretion of a particular molecule by neighbouring tumour cells that do not contain the same gene deletion is unlikely to be a major problem.
COX-2 derived PGE2 is a pleiotropic molecule that has been shown to act on many cell types including CD8 + T cells, dendritic cells and NK cells (45)  However, these trials do not specifically enrol KRAS-mutant LUAD patients which our data suggests would most benefit from such a combination. The potential benefit of this combination is supported by a recent retrospective analysis showing improved survival of lung cancer patients that were concurrently treated with COX inhibitors whilst receiving immunotherapy (49). Furthermore, we show using a novel drug that dual inhibition of the PGE2 receptors EP2 and EP4 has similar therapeutic benefits to COX-2 inhibition and shows superior synergy when combined with ICB. This supports other work in preclinical models that have demonstrated the benefits of combining EP2 and EP4 antagonists with ICB (37,38). EP2-4 inhibition therefore has the potential to enhance the efficacy of immunotherapy in the clinic, whilst possibly avoiding the toxicities associated with celecoxib treatment.

24
The ability of KRAS G12C inhibition to suppress the COX-2/PGE2 signalling axis in vivo may in part explain the synergy observed in combination with ICB in preclinical models (15,17,18,40). However, given the apparent poor tolerability of the combination of KRAS G12C inhibition by sotorasib and PD(L)-1-targeted ICB observed in the clinic (50) it may be more feasible to target KRAS-driven immune evasion mechanisms such as COX-2. The benefits of KRAS G12C inhibition in the clinic are also seriously confounded by the rapid emergence of acquired resistance which can be driven by many different oncogenic mutations within the RAS signalling pathway (11). Clinical trials such as CODEBREAK 101 are attempting to overcome this by combining KRAS G12C inhibitors with other targeted therapies such as RTK inhibitors; however, given the genetic complexity underlying resistance, the feasibility of this approach remains unclear. Instead, targeting KRAS-driven immune suppression may prove more successful. Our work shows that long-term KRAS G12C inhibition results in restoration of the COX-2/PGE2 axis which may contribute to tumour relapse. Furthermore, our data suggest that combination of KRAS G12C inhibitory drugs and COX-2 or EP2-4 prostaglandin receptor inhibition may be successful in the treatment of immune hot lung cancer. One might speculate that it could possibly avoid the toxicities reported for sotorasib and PD(L)-1 blockade.

In vivo tumour studies
All animal studies were approved by the ethics committee of the Francis Crick cells were plated in 96-well plates and growth was monitored for 5d using an IncuCyte Zoom system (Essen Biosciences).

In vitro treatments
27 Drugs or cytokines were added in fresh media 24h after seeding cells at stated concentrations and samples were obtained at indicated time points.

Flow cytometry
Mouse tumours were cut into small pieces and incubated with collagenase (1 mg/ml; ThermoFisher) and DNase I (50 U/ml; Life Technologies) in HBSS for 45 min at 37°C.
Cells were filtered through 70 μm strainers (Falcon) and red blood cells were lysed using ACK buffer (Life Technologies). Samples were stained with fixable viability dye

Immunohistochemistry
Tumour-bearing lungs were fixed in 10% NBF for 24h and transferred to 70% ethanol. Fixed lungs were processed into paraffin-embedded blocks. Tissue sections were stained with haematoxylin and eosin, using the automated TissueTek Prisma 28 slide stainer. For immunohistochemistry staining, sections were boiled in sodium citrate buffer (pH 6.0) for 15 min and incubated with the following antibodies for 1h: anti-CD8 (4SM15, Thermo Scientific), anti-NCR1 (EPR23097-35, Abcam) and anti-Arg1 (D4E3M, Cell Signalling). Primary antibodies were detected using biotinylated secondary antibodies and HRP/DAB detection. Slides were imaged using a Leica Zeiss AxioScan.Z1 slide scanner. Tumour-infiltrating immune cells were quantified using QuPath.

RT-qPCR
RNA was extracted from cell lines or frozen lung tumours using RNeasy kit (Qiagen).
Single tumour nodules were plucked from tumour-bearing lungs. Tumours from multiple mice were included in each analysis. Frozen tumour samples were homogenised prior to RNA extraction using QIAshredder columns (Qiagen). cDNA was generated using the Maxima First Strand cDNA Synthesis Kit (Thermo Fisher Scientific) and qPCR performed using Applied Biosystems Fast SYBR Green reagents (Thermo Fisher Scientific). mRNA relative quantity was calculated using the DDCT method and normalised to Sdha, Hsp90 and Tbp. For heatmap visualisation, relative mRNA expression for each gene was log-transformed and mediannormalised.

29
Cells were treated as indicated for 48h. Conditioned media was collected and PGE2 concentration determined using the Prostaglandin E2 parameter assay kit (R&D), as per manufacturer's instructions.

siRNA experiments
Cells were seeded in a 6-well plate and reverse-transfected with 50nM siGENOME siRNA pools targeting mouse Myc (Dharmacon) using DharmaFECT 4 transfection reagent (Dharmacon) according to the manufacturer's instructions. 24h after transfection, cells were treated for 24h with trametinib (10nM). Control cells were mock-transfected (no siRNA) or transfected with siGENOME RISC-free control siRNA (Dharmacon).

MicroCT imaging
Mice were anesthetised by inhalation of isoflurane and scanned using the Quantum GX2 micro-CT imaging system (Perkin Elmer). Serial lung images were reconstructed and tumour volumes subsequently analysed using Analyse (AnalyzeDirect) as previously described (51)

Western blotting
Cells were lysed using 10X Cell Lysis Buffer (Cell Signalling) supplemented with protease and phosphatase inhibitors (Roche). Protein concentration was determined using a BCA protein assay kit (Pierce) and 15-20 μg of protein was separated on a 4-  sgRNA sequences along with on-target and off-target scores are stated in Appendix 1. 50 non-targeting sgRNAs were also used from the mouse GeCKOv2 library.

Stable cell lines, plasmids and lentivirus infection
Annealed oligonucleotides corresponding to each sgRNA were combined into 10 different pools, each containing 125 sgRNAs and 5 non-targeting controls, and cloned into a lentiviral vector (pLenti_BSD_sgRNA). Lentiviral libraries were produced using each of the 10 pools and stored at -80°C. Reads were normalised to total read counts per sample using MAGeCK and log2-fold change between groups calculated.

Bioinformatic analysis
The CCLE RNA-seq data were obtained from the CCLE repository hosted at The Broad (https://data.broadinstitute.org/ccle_legacy_data). We used the classification of RAS-low and RAS-high as previously described (26

ACKNOWLEDGEMENTS
We thank the science technology platforms at the Francis Crick Institute including Biological Resources, Advanced Sequencing, Scientific Computing, Bioinformatics and Biostatistics, Flow Cytometry, Experimental Histopathology, and Cell Services.