An immunogenic model of KRAS-mutant lung cancer for study of targeted therapy and immunotherapy combinations

Mutations in oncogenes such as KRAS and EGFR cause a high proportion of lung cancers. Drugs targeting these proteins cause tumour regression but ultimately fail to cure these cancers, leading to intense interest in how best to combine them with other treatments, such as immunotherapies. However, preclinical systems for studying the interaction of lung tumours with the host immune system are inadequate, in part due to the low tumour mutational burden in genetically engineered mouse models. Here we set out to develop mouse models of mutant KRAS-driven lung cancer with an elevated tumour mutational burden by expressing the human DNA cytosine deaminase, APOBEC3B, to mimic the mutational signature seen in human lung cancer. This failed to substantially increase clonal tumour mutational burden and autochthonous tumours remained refractory to immunotherapy. However, by establishing clonal cell lines from these tumours we generated an immunogenic syngeneic transplantation model of KRAS mutant lung adenocarcinoma that was sensitive to immunotherapy. Unexpectedly, we found that anti-tumour immune responses were not directed against neoantigens but instead targeted derepressed endogenous retroviral antigens. The ability of KRASG12C inhibitors to cause regression of KRASG12C-expressing versions of these tumours was markedly potentiated by the adaptive immune system, providing a unique opportunity for the study of combinations of targeted and immunotherapies in immune-hot lung cancer.


ABSTRACT
Mutations in oncogenes such as KRAS and EGFR cause a high proportion of lung cancers. Drugs targeting these proteins cause tumour regression but ultimately fail to cure these cancers, leading to intense interest in how best to combine them with other treatments, such as immunotherapies. However, preclinical systems for studying the interaction of lung tumours with the host immune system are inadequate, in part due to the low tumour mutational burden in genetically engineered mouse models. Here we set out to develop mouse models of mutant KRAS-driven lung cancer with an elevated tumour mutational burden by expressing the human DNA cytosine deaminase, APOBEC3B, to mimic the mutational signature seen in human lung cancer. This failed to substantially increase clonal tumour mutational burden and autochthonous tumours remained refractory to immunotherapy. However, by establishing clonal cell lines from these tumours we generated an immunogenic syngeneic transplantation model of KRAS mutant lung adenocarcinoma that was sensitive to immunotherapy. Unexpectedly, we found that anti-tumour immune responses were not directed against neoantigens but instead targeted derepressed endogenous retroviral antigens. The ability of KRAS G12C inhibitors to cause regression of KRAS G12C -expressing versions of these tumours was markedly potentiated by the adaptive immune system, providing a unique opportunity for the study of combinations of targeted and immunotherapies in immune-hot lung cancer. It has therefore become critical to further elucidate the molecular determinants that underpin the interaction between the tumour and the immune system. Increasing evidence suggests that tumour-cell-intrinsic oncogenic signalling, including KRAS Identifying rational therapeutic approaches to extend the clinical benefits of current ICBs in NSCLC requires the use of preclinical models that recapitulate the interactions between tumour cells and the immune system, which is not possible in conventional xenograft models lacking a functional immune system. Genetically engineered mouse models (GEMMs) have been extensively used to gain mechanistic insights into the biology of KRAS-mutant lung cancer and to assess the efficacy of novel therapeutics.
Such models recapitulate key aspects of the human disease in an immune-competent setting, however, they fail to elicit strong anti-tumour immune responses 8,9 and therefore have limited use for studying tumour-immune interactions. Genetically engineered mouse cancer models usually feature a small number of introduced strong driver mutations, sufficient for tumourigenesis, and acquire few additional mutations. 4 Tumours arising from these models therefore have a low tumour mutational burden (TMB) compared to their human counterparts 10 , limiting the presentation of neoantigens to the adaptive immune system. This problem has been overcome by the forced expression of highly immunogenic antigens, such as ovalbumin 9,11 , but it is unclear whether the strong anti-tumour immune responses elicited by such foreign antigens reflect those in human cancers which occur towards less potent neoantigens and tumour-associated antigens.
To address this issue, we set out to generate a novel mouse model of KRAS driven lung adenocarcinoma with increased tumour mutation burden and potentially increased immunogenicity. Approaches used included the use of carcinogens and also the over-expression of a member of the APOBEC family of single-stranded DNA deaminases, which are responsible for inducing mutations in a range of cancers 12 . We were ultimately successful in generating a transplantable KRAS mutant lung cancer model that is partially sensitive to immunotherapy and shows a response to KRAS targeted agents that is clearly boosted by the adaptive immune system. This transplantable lung cancer model will be a valuable tool for studying strategies for combining targeted agents against the RAS pathway with immunotherapies. 5

Autochthonous KP lung tumours do not engage with the adaptive immune system
The introduction of adenovirus expressing Cre recombinase (AdCre) into the lungs of Kras LSL-G12D/+ ;p53 fl/fl (KP) mice leads to expression of oncogenic Kras G12D and deletion of p53 in lung epithelial cells, resulting in the induction of lung adenocarcinoma 13 . This system represents one of the most widely used mouse models of lung cancer. To assess the immunogenicity of lung tumours arising in KP mice, we crossed them onto a Rag2 -/background, which lacks mature T and B cells, and monitored tumour growth by micro-computed tomography (micro-CT) imaging.
Adaptive immunity was unable to constrain the growth of KP tumours as they grew at similar rates in immune-competent (Rag2 +/-) and immune-deficient (Rag2 -/-) mice ( Fig.   1A). Furthermore, immunohistochemistry staining revealed that tumours arising in immune-competent hosts lacked T cell infiltration (Fig. 1B). To assess whether an adaptive immune response could be generated against KP tumours, we treated tumour-bearing mice with a combination of anti-PD-L1 and anti-CTLA-4 (Fig. 1C). This combination therapy failed to delay tumour growth (Fig. 1D, E) and did not lead to an increase in the survival of tumour-bearing mice (Fig. 1F). It has previously been shown that MEK inhibition enhances anti-tumour immunity and synergises with anti-PD-L1 in KRAS-mutant CT26 colorectal tumours 14 . However, we found that the combination of anti-PD-L1 and trametinib failed to control KP tumour growth compared to trametinib alone ( Supplementary Fig. 1A, B). These data suggest that the adaptive immune system is unable to recognise autochthonous KP tumours.

Human APOBEC3B does not induce immunogenicity in the KP model
In comparison with human lung cancers seen in the clinic, KP mouse lung tumours exhibit very few mutations, which are necessary to generate neoantigens that can make tumour cells visible to the immune system 10 . APOBEC3B is a single-stranded DNA cytosine deaminase that induces C>T/G substitutions in several solid cancers 12,15,16 and has been associated with intra-tumoural heterogeneity in lung adenocarcinoma 17,18 . Furthermore, analysis of lung adenocarcinoma (LUAD) samples from The Cancer Genome Atlas (TCGA) revealed that the mutational rate of nonsynonymous APOBEC mutations was lower in comparison with other types of 6 mutation, suggesting that non-synonymous mutations generated by APOBEC could be immunogenic and preferentially eliminated by the immune system ( Fig. 2A).
We therefore decided to express human APOBEC3B in the KP model to increase the frequency of mutations in these tumours to promote the generation of neoantigens that could stimulate adaptive anti-tumour immune responses. We inserted a human However, KPA tumours grew at similar rates to KP tumours ( Fig 2F). Interestingly, immunohistochemical analysis revealed that A3Bi expression was associated with moderate CD8 + T cell infiltration ( Fig. 2G and Supplementary Fig 3C) which was confined to the periphery (Supplementary Fig 3D). In contrast, CD8 + T cells were entirely absent from KP tumours. To assess whether this increased CD8 + T cell recruitment promoted immune-control of KPA tumours we crossed KPA mice onto a Rag1 -/background to yield KPAR mice and evaluated tumour growth by micro-CT imaging in immune-competent KPA and immune-deficient KPAR animals. We observed no differences in tumour number or tumour growth in KPAR mice compared with KPA mice (Supplementary Fig. 3E, F). We also treated KP and KPA tumourbearing mice with anti-PD-1 and anti-CTLA-4 and observed no differences between the two groups (Fig. 2H). We then performed whole-exome sequencing (WES) of KP and KPA tumours to assess whether A3Bi expression resulted in increased tumour mutational burden. We found that the total number of subclonal exonic SNVs was moderately increased in A3Bi-expressing tumours, however the majority of these were not typical A3Bi T(C>T/G) mutations (Fig. 2I). Consistent with this finding, KPA 7 tumours possessed more predicted subclonal neoantigens, however few of these were likely to be directly due to APOBEC3B activity (Fig. 2J), perhaps indicating the possibility of indirect mechanisms linking APOBEC3B with the formation of new mutations.
Altogether, these findings suggest that A3Bi expression in the KP model of lung adenocarcinoma did not produce sufficient immunogenic mutations to elicit an adaptive immune response. This may be in part because of the heterogeneity of A3Bi expression in the tumours, the subclonal nature of any potential neoantigens, or alternatively due to insufficient numbers of mutations induced in this system.

APOBEC3B increases subclonal mutations in urethane-induced tumours but does not promote immunogenicity
Murine KP tumours develop extremely rapidly, inducing a life-threatening tumour burden in about 14 to 18 weeks. We reasoned that the aggressive nature of the KP model did not allow sufficient time for APOBEC3B to induce mutations during tumour development, leading to only a few detectable SNVs with low allelic frequency.
Carcinogen-induced tumours tend to be less aggressive than GEMMs and develop more slowly. To extend the length of tumour development, we exposed the mice to urethane before A3Bi expression. Urethane is a carcinogen which induces A>T/G substitutions and initiates lung tumours by inducing an activating mutation at codon Q61 in Kras 19,20 . To model A3Bi expression in carcinogen-induced tumours we initiated tumours with urethane in Rosa26 A3Bi/CreER(t2) mice (UrA3Bi) which when treated with tamoxifen express A3Bi in all tissues. Since APOBEC mutations are often late events in tumour evolution 17 , we delayed the induction of A3B by three weeks after the first injection of urethane ( Supplementary Fig. 4A).
We confirmed the expression of A3Bi in the lungs after delivery of tamoxifen in UrA3Bi mice (Fig. 3A). A3Bi expression was downregulated in tumours compared with adjacent lung, consistent with what we observed in KPA tumours. Despite this, UrA3Bi tumours had more advanced histological grades compared with tumours induced by urethane alone (Fig. 3B). However, tumour growth and the number of tumours per animal were similar between urethane and UrA3Bi tumours ( . We performed WES to assess whether A3Bi generated mutations in these tumours. As expected, urethane exposure generated a substantial number of clonal exonic SNVs (Fig. 3E). Similar to the KPA model, UrA3Bi tumours contained higher numbers of subclonal exonic SNVs (Fig. 3E) and predicted neoantigens (Fig. 3F) compared to urethane-induced tumours, however A3Bi expression failed to induce typical APOBEC T(C>T/G) mutations.
To summarise, as with the KP model, APOBEC3B expression failed to induce immunogenicity in carcinogen-induced models of lung cancer.

Establishment of immunogenic clonal cell lines from KPAR tumours
We hypothesised that the lack of immunogenicity in APOBEC3B-expressing autochthonous tumours might be due to the subclonality of mutations, which have been shown to be less effective at generating effective adaptive immune responses 21,22 . We therefore established cell lines from these models which were  9 We carried out WES to assess the mutational burden of the KPAR clonal cell lines, the parental polyclonal cell line (KPAR1) and another autochthonous KPAR tumour taken from the same mouse. Single-cell cloning moderately increased the frequency of detectable mutations (Fig. 4C). All single-cell clones contained more mutations compared with the parental cell line or KP tumours (Fig. 2I). However, the number of mutations in all clonal cell lines was still very low compared to other transplantable syngeneic cancer cell lines, and immunogenic KPAR1.3 cells did not possess more predicted neoantigens than non-immunogenic KPAR1.1 cells (Fig. 4D, Supplementary   Table S1). Notably, very few mutations were typical APOBEC T(C>T/G) mutations ( Supplementary Fig. 5A). Furthermore, we were unable to detect antigen-specific CD8 + T cells against any of the predicted neoantigens when pulsing tumour-infiltrating lymphocytes (TILs) isolated from KPAR1.3 tumours in an IFNg enzyme-linked immune absorbent spot (ELISpot) assay (Supplementary Table S1).
Given that A3Bi failed to induce any immunogenic mutations in the KPAR1. Together these results indicate that the immunogenicity of the KPAR1.3 cell line was not due to tumour mutational burden, but elevated expression of endogenous retroviral antigens that stimulate endogenous CD8 + T cell responses.

KPAR tumours generate an adaptive immune response
Although immunogenicity of the KPAR1.3 cell line was not due to neoantigens generated by non-synonymous mutations as we initially hypothesised, the novelty of an immunogenic transplantable murine lung cancer cell line warranted further characterisation. We therefore used flow cytometry to characterise the tumour Taken together, these data demonstrate that orthotopic KPAR tumours generated an adaptive anti-tumour immune response which was absent in orthotopic KPB6 tumours.

KPAR tumours are responsive to ICB
Given that the growth of KPAR tumours were partially restrained by adaptive immunity and orthotopic tumours were highly infiltrated with activated immune cells, Furthermore, the majority of CD4 + T cells in subcutaneous tumours were Tregs whilst in orthotopic tumours CD4 + effector T cells were more abundant (Supplementary Fig.   7G). 12 To summarise, KPAR1.3 tumours were sensitive to anti-PD-1 or anti-CTLA-4 immune checkpoint blockade therapy, the response to which was dependent on the site of tumour growth.

Generation of KPAR G12C cells to assess the immunomodulatory properties of KRAS G12C inhibitors
The recently developed class of KRAS G12C inhibitors has been shown to promote anti-tumour immune responses in the immunogenic CT26 G12C model of colorectal To test the effect of KRAS G12C inhibitors in the KPAR lung cancer model we used prime-editing technology to generate the KPAR G12C cell line. WES revealed that KPAR cells were homozygous for KRAS G12D so we edited both alleles to KRAS G12C ( Supplementary Fig. 8A). Cell-viability assays demonstrated that KPAR G12C cells showed impaired viability in response to treatment with AZ-8037, a recently described Supplementary Fig. 8B). Furthermore, immunoblotting revealed that AZ-8037 inhibited pERK in KPAR G12C cells ( Supplementary Fig. 8C). To assess whether KRAS G12C inhibition could stimulate anti-tumour immunity in vivo we tested the response of KPAR G12C subcutaneous tumours to AZ-8037 in both immunecompetent and immune-deficient (Rag1 -/-) mice. Vehicle-treated KPAR G12C tumours grew slower in immune-competent mice compared to Rag1 -/mice, similarly to what we observed with the parental KPAR tumours (Fig. 7A). AZ-8037 treatment caused marked tumour regression in both immune-competent and Rag1 -/mice, however the response was much more durable in immune-competent mice as all tumours remained responsive during the duration of treatment whilst tumours in Rag1 -/mice began to grow back before termination of treatment (Fig. 7A). Furthermore, after the treatment was terminated one of the six treated mice showed a durable cure (Fig. 7B). We also used CRISPR technology to edit the KPB6 cell line, which harbours a wildtype KRAS and KRAS G12D allele, to generate the KPB6 G12C cell line which lost the wildtype allele by indel generation and contained a KRAS G12C allele ( Supplementary Fig. 8D). Cellviability assays and immunoblotting demonstrated that KPB6 G12C cells were sensitive to AZ-8037 ( Supplementary Fig. 8E, F). In contrast to KPAR G12C tumours, the response of KPB6 G12C tumours to AZ-8037 was comparable in immune-competent and Rag1 -/mice (Fig. 7C), with tumours beginning to lose responsiveness before treatment ended, and then growing back rapidly after the cessation of treatment with 13 no long-term responses achieved (Fig. 7D). Given that adaptive immunity contributes to the efficacy of KRAS G12C inhibition in KPAR tumours, we next wanted to assess the effects of KRAS G12C inhibition on the tumour microenvironment. qPCR analysis of orthotopic KPAR G12C tumours revealed that KRAS G12C inhibition induced a proinflammatory microenvironment with increased antigen presentation, cytokine production, interferon signalling, immune cell infiltration and T cell activation (Fig 7E).
These results suggest the efficacy of KRAS G12C inhibition in the immunogenic KPAR model was partially due to the generation of an adaptive anti-tumour immune response which resulted in durable regressions in immune-competent hosts.

DISCUSSION
There is a need for improved models of lung cancer that are immunogenic to enable us to better understand the interplay between the tumour and the immune system and assess the efficacy of novel therapeutic interventions. We and others have tried several approaches to make lung cancer GEMMs more immunogenic. We treated KP mice with carcinogens, expressed APOBEC3B in p53-deleted urethane-induced tumours, and -based on the assumption that chromosome rearrangement could lead to neoantigens -also expressed Mad2, a spindle checkpoint protein associated with aneuploidy, in KP tumours 28 . However, none of these strategies generated immunogenic tumours that grew differentially in immune-competent and immunedeficient backgrounds, induced T cell responses or responded to immune checkpoint blockade (data presented here and unpublished data). In this study, APOBEC3B expression only moderately increased the tumour mutational burden in KP and urethane-induced lung tumours and was not sufficient to make these tumours immunogenic. The lack of substantial numbers of APOBEC3B induced mutations in these models was potentially a consequence of a detrimental impact of APOBEC3B expression during early stages of tumour development, which is reflected by the downregulation of A3Bi expression in both KP and urethane-induced lung tumours.
The development of a KP model with a temporal regulation of APOBEC3B could help addressing this limitation. Furthermore, APOBEC3B-induced mutations were mainly subclonal in the autochthonous tumours, as described in patients 18 . The subclonality of many mutations in these models and the heterogenous A3Bi expression we observed in tumours could explain the absence of effective anti-tumour immune responses, since sensitivity to ICB correlates with clonal neoantigen burden in NSCLC 29 . However, despite containing many more clonal exonic mutations than A3Biexpressing KP tumours, urethane-induced lung tumours were also refractory to ICB. inhibitors is partially due to the engagement of the adaptive immune system in immune-hot tumours. The KPAR model therefore offers the possibility to explore combinations of immunotherapy with KRAS G12C inhibition to overcome the acquired resistance anticipated following this novel targeted therapy 42 .
In conclusion, we have created a novel model of immunogenic KRAS-driven lung adenocarcinoma, which we anticipate will contribute to the development of new combinations of therapies, including those involving immune checkpoint blockade and KRAS G12C inhibition. 18

Flow cytometry
Mouse tumours were cut into small pieces, incubated with collagenase (1 mg/ml; ThermoFisher) and DNase I (50 U/ml; Life Technologies) for 45 min at 37°C and filtered through 70 µm strainers (Falcon). Red blood cells were lysed for 5 min using ACK buffer (Life Technologies). Cells were stained with fixable viability dye eFluor870 (BD Horizon) for 30 min and blocked with CD16/32 antibody (Biolegend) for 10 min.
Cells were then stained with one of three antibody cocktails for 30 min (see Supplementary Table 1). Intracellular staining was performed using the Fixation/Permeabilization kit (eBioscience) according to the manufacturer's instructions. Samples were resuspended in FACS buffer and analysed using a BD Symphony flow cytometer. Data was analysed using FlowJo (Tree Star).
For FACS analysis in vitro, cells were trypsinised, washed with FACS buffer and stained for eMLV envelope glycoprotein using the 83A25 monoclonal antibody followed by a secondary staining with anti-rat IgG2a (PE). Samples were run on LSRFortessa (BD).

Histopathology and Immunohistochemistry
Tumour-bearing lungs were fixed in 10% NBF for 24 h followed by 70% ethanol.

Micro-CT imaging
Mice were anesthetised by inhalation of isoflurane and scanned using the Quantum GX2 micro-CT imaging system (Perkin Elmer) at a 50μm isotropic pixel size. Serial lung images were reconstructed and tumour volumes subsequently analysed using Analyse (AnalyzeDirect).
We corrected variant allele frequencies for tumour purity using two different methods depending on the model. In the case of the KP induced tumours we calculated the ratio of reads per base across the floxed exons of Trp53 (1-9) and the undeleted exon 11. In the cases of the Urethane induced models we used the VAF of the Kras Q61 mutation. We assumed this mutation to be the primary driver of tumour activation since it is found in all the urethane induced tumours. All Kras Q61 loci were found to be copy number neutral. We called absolute copy number counts from the relative ratios using cnvkit call. The ratios were scaled using the purity estimates calculated above and integer copy numbers assigned using the follow log2 ratio thresholds, with used these tumour purity estimates and the relative copy number ratios to estimate the clonality of SNVs. In the case of KPA(R) cell lines, we assumed clonality of most CNAs and scale the CN to the X chromosome (haploid in male cell lines and diploid in female cell lines).
We distinguished clonal from sub-clonal mutations by estimating the fraction of tumour cells carrying the observed mutation using the method detailed in Turajlic et al. 49 . and normalised to at least three housekeeping genes.

Forward primer Reverse primer
A3Bi

Tumour Cell Viability
For short-term viability assays, 1.5 x 10 3 KPAR1.3 G12C or 2 x 10 3 KPB6 G12C cells were seeded in 96-well plates and grown in the presence of different inhibitors for 72 h. Cell viability was assessed using CellTiter-Blue (Promega).

Western blotting
Cells were lysed using protein lysis buffer (Sigma) with protease and phosphatase inhibitor cocktails (Sigma