1840 Whole-exome somatic analysis of mouse cancer models and implications for preclinical immunomodulatory drug development

Bruno Zeitouni1, Jason Davis2, Cordula Tschuch1, Anne-Lise Peille1, Yana Raeva1, Manuel Landesfeind1, Sheri Barnes2, Julia Schueler1 1Charles River, Discovery Research Services Germany GmbH, Am Flughafen 12 - 14, D-79108 Freiburg, Germany 2Charles River, Discovery, 3300 Gateway Centre Blvd, Morrisville, NC 27560, USA

1 INTRODUCTION 2 MOUSE CANCER MODELS 3 WHOLE-EXOME DATA Figure 1. The Charles River analysis pipeline for somatic mutation detection from mouse WES data. Raw sequence data were analysed by a mouse-specific bioinformatics pipeline from read Tumor Sequenced Model Passage Sequencing raw data Cancer immunotherapy is evolving as the new mainstream approach to fight tumor progression. Index Model Name Cancer Type Original Strain WES Availability mapping onto the mouse to the variant calling and filtering, including the removal of the germline The use of immunotherapeutic agents mediating immune checkpoint receptor blockade has Designation Type Number found in the sequenced matching normal DNA of the corresponding mouse strain. Gene coverage demonstrated a better anti-tumor efficacy compared to conventional therapies. Because of their 1 ATNC 4T1 Breast BALB/c Yes Cell line 8N4 was determined by applying the RPKM (Reads Per Kilobase per Million mapped reads) normalisation. 2 ATNC A20 Lymphoma BALB/c Yes Cell line 11N8 Read Mapping on Mouse growth in immunocompetent mice, syngeneic mouse cancer models are particularly relevant for 3 ATNC B16F10 Melanoma C57BL/6 Yes Cell line 6N3 mm10 genome (“C57BL/6”) with BWA Figure 2. WES mapping metrics results. Each sample was sequenced with an preclinical drug testing. The characterization of their genetic alterations is essential to select the 4 ATFR Breast 11509-F Breast FVB/N Yes Tumor 3 5 ATNC Cloudman S91 Melanoma DBA2 Yes Cell line 21N4 expected coverage of 120X. WES achieved a mean base coverage on targets of models that best fit the molecular requirements of a successful immunotherapy. Mouse mapping data 165X and an average of 68% of reads mapped on Agilent targets. 6 ATNC Colon26 Colon BALB/c Yes Cell line 10N3 7 ATNC CT26 Colon BALB/c Yes Cell line 11N3 Variant calling with 3 different tools : In this study, we analysed somatic mutation profiles from whole- (WES) in a 8 ATNC EMT6/P Breast BALB/c Yes Cell line N5 Mapped reads on targets (%) Mean coverage on targets (number of reads) panel of 14 different mouse models including genetically-engineered mice (GEM) and syngeneic 9 ATFR KP NSCLC C57BL/6 Yes Tumor 1 GATK-Lite, Samtools and Freebayes 100% 350 10 ATFR KP1 NSCLC C57BL/6 Yes Cell line 11N3 cell lines, covering 6 major cancer types with well-characterized responses to known immune 300 11 ATFR KP4 NSCLC C57BL/6 Yes Cell line 5N1 Variant annotation with genome databases: 80% checkpoint inhibitors. 12 ATFR Lewis Lung NSCLC C57BL/6 No NA NA 250 dbSNP (strain specificity) Gene 13 ATNC MC38 Colon C57BL/6 Yes Cell line 15N6 60% 200 14 ATFR Renca Renal BALB/c Yes CDX 2 coverage 40% 150 Variant filtering based on: with RPKM Table 1. The panel of mouse cancer models used in the study. The 4 models KP, KP1, KP4 and Breast 11509-F are GEM models, the others are values 100  Mapping/calling statistics and functional annotations 20% syngeneics developed by injection of established cell lines onto the mouse strain (C57BL/6, BALB/c, FVB/N and DBA2). All models were sequenced 50 by WES either directly from the cell line, from the tumor of the xenograft or from the cell line-derived xenograft (CDX), along with DNA from original  Germline mutations found in the normal strain DNA

0% 0 F mouse strains. The enrichment of exonic regions was done using the Agilent SureSelect Mouse kit V1 prior to Illumina HiSeq 2500 paired-end - 4 EXPLORATORY ANALYSIS OF KP 4T1 A20 KP1 KP4 CT26 MC38 sequencing. Renca B16F10 EMT6/P MUTATION PROFILES Somatic mutations : SNPs and INDELS Gene coverage Colon26 Cloudman S91 Cloudman Breast 11509 Breast (a) (b) (a) (b) KP KP1 KP KP4 KP1 KP4 A B A B A B 5 CORRELATION WITH IMMUNO-ONCOLOGY SENSITIVITY DATA KP C57BL/6 [CELLRANGE] KP4 # Mutations in common 221 230 315 [CELLRANGE] 484 300 Responders Figure 5. Correlation analysis of mutation BALB/c 147 22 57 - Same predicted genotype 101 104 310 Non-responders load and in vivo sensitivity data using Mouse [CELLRANGE] - Het in A and Hom in B 120 123 5 Anti-mPD1 Anti-mCTLA4 Anti-mPD1+Anti-CTLA4 strains DBA2 checkpoint inhibitors in mouse models. [CELLRANGE] # Mutations specific to A 79 70 588 Spearman correlation = 0.24 Spearman correlation = 0.48 Spearman correlation = 0.64 208 p-value = 0.4886 p-value = 0.005824 Correlation scatterplots are shown between FVB/N - No mutation signal in B 58 40 494 p-value = 0.5161 107 13 the mutation rate and the Tumor/Control KP [CELLRANGE] - Filtered by pipeline in B 21 30 94 (T/C) values from anti-PD1, anti-CTLA4 or

GEM KP4 [CELLRANGE] # Mutations specific to B 682 254 169 combination of both. Using a cut-off of 50% 575 models Breast 11509-F [CELLRANGE] KP1 - No mutation signal in A 530 141 125 of responders, non-responders models are - Filtered by pipeline in A 152 113 44 color-coded in red and responders in light- KP1 [CELLRANGE] 903 /C values (%)

/C values (%) blue. The regression line is shown in blue. T 4T1 [CELLRANGE] T Figure 4. Mutation overlap analysis between the KP GEM models. The Spearman correlation and the t-test p- A20 [CELLRANGE] (a) Venn diagram of mutations between the KP tumor and KP1, KP4 cell lines. The mutation count increased 1.3- fold for KP4 and 2.9-fold for KP1 during value are indicated. A significant correlation was observed in combination data showing a B16F10 [CELLRANGE] cell line establishment. The KP1 model derived more in terms of mutations compared to the KP4 model. Mouse higher sensitivity in models with high syngeneic EMT6/P [CELLRANGE] (b) Table of pairwise mutation overlap from (a). The number of mutations is shown for each comparison of samples named A or B. 50% of common

Median of optimal mutation rates. Colon26 was sensitive to Median of optimal cell lines mutations were found to be homozygous in KP and heterozygous in KP1/KP4. Specific mutations were found in each model either because no signal was Median of optimal T/C values (%) Cloudman S91 [CELLRANGE] detected in the other model (meaning no mutation detected) or because of the pipeline filtering parameters applied. anti-CTLA4 whereas CT26 was only Colon26 [CELLRANGE] sensitive to the combination with anti-PD1. Although the 2 melanoma models show MC38 [CELLRANGE] similar mutations rates, only Cloudman S91 Model / Gene Alk Apc Atm Braf Brca1 Brca2 Cdk4 Cdkn2a Erbb3 Fgfr1 Fgfr2 Flt1 Jak2 Jak3 Kit Kras Pdgfra Pten Ret Sos1 Stat3 Stk11 Trp53 CT26 [CELLRANGE] 4T1 DEL A942S P31X responded to checkpoint inhibitors while B16- [CELLRANGE] Renca A20 R307I N134S E93D R752S S237G Mutation rate (m/MB) Mutation rate (m/MB) Mutation rate (m/MB) F10 was resistant to all 3 treatments, B16F10 G386R Q1039* K92T C264R K1997N DEL L731H S145F T131P K748N R152Q N128D Table 2. Mutational landscape of suggesting other causes of response. Breast11509-F somatic variants found in selected CloudmanS91 D2086G G506V G1679E A121V R914Q G12D targetable cancer genes. Amino- Figure 3. Heterogeneity in the number of somatic mutations between the mouse cancer models. Colon26 R2066K DEL S107F V88I V8M acid changes and homozygous whole G12D (a) Barplot of mutation rates in mouse strains (in green), GEM models (in blue) and syngeneic cell lines (in orange). CT26 R2066K V92M DEL S107F V88I V8M gene deletions are shown in green 6 CONCLUSION The total number of detected mutations is indicated in parenthesis. A average mutation rate in of 37 somatic G1469X EMT6/P D546A G209* N233T and red respectively. Consistently, the mutations per megabase (m/MB) was detected, ranging from 5 m/MB in the model KP to 140 m/MB in the syngeneic S264I renal line Renca. Mutation rates were markedly lower in GEM-derived models than in syngeneic lines (average of 9 KP G12D KP GEM models harboured the Kras Mouse models of cancer are a relevant tool for preclinical studies specifically for immuno-oncology. The molecular KP1 G12D DEL G12D mutation and a Trp53 gene vs 50 m/MB). KP4 G12D DEL characterization of these models, as presented here with gene mutation data from WES, allow the identification of the G242V deletion except for KP, probably due (b) Dendrogram of models based on mutated genes. A binary distance was used before a complete-link hierarchical MC38 E1171D W487C DEL E92K A340G M28I markers of their heterogeneity (mutation load, mutations in cancer genes) helping to optimize their use in drug discovery. S258I to the normal DNA of the mouse clustering of models. Consistently, the KP GEM model and KP-derived models clustered together, as did the colon T638M Renca D2086G R1785Q *** E653D K575E R287* R210C They will support the development of innovative immunotherapeutic agents and the discovery of biomarkers to classify the Colon26 and CT26 too. T351N stroma in the sequenced tumor. *** E1604D, E1510K, G1023S, V1010I, C933S, S765P patient cohort profiting the most from these new compounds.