DualMET andERBB inhibition overcomes intratumor plasticity in osimertinib-resistant-advanced non-small-cell lung cancer (NSCLC)

Original Article Annals of Oncology

Dual MET and ERBB inhibition overcomes intra-tumor plasticity in osimertinib resistant advanced non-small cell lung cancer (NSCLC)

A. Martinez-Marti1,2,3, E. Felip1,2,3*, J. Matito4, E. Mereu5,6, A. Navarro1,2, S. Cedrés1,2, N. Pardo1,2,3, A. Martinez de Castro1,2, J. Remon1,2, JM. Miquel2, A. Guillaumet-Adkins5,6, E. Nadal7,8, G. RodriguezEsteban5,6, O. Arqués9, R. Fasani10 , P. Nuciforo10, H. Heyn5,6, A. Villanueva 7,11, H. G. Palmer9, A. Vivancos4*. Author Affiliations 1

Department of Medical Oncology, Vall d’Hebron University Hospital, Barcelona, Spain. 2Department of Medical Oncology, Vall d’Hebron Institute of Oncology (VHIO), Barcelona, Spain. 3Autonomous University of Barcelona (UAB), Barcelona, Spain. 4Cancer Genomics Group, Vall d’Hebron Institute of Oncology (VHIO), Barcelona, Spain. 5 CNAG-CRG, Centro Nacional de Análisis Genómico (CNAG) - Center for Genomic Regulation (CRG), Barcelona Institute of Science and Technology (BIST), Barcelona, Spain. 6Pompeu Fabra University (UPF), Barcelona, Spain. 7 Chemoresistance and Predictive Factors Group, Program Against Cancer Therapeutic Resistance (ProCURE), Catalan Institute of Oncology (ICO) Bellvitge Biomedical Research Institute (IDIBELL), L’Hospitalet, Barcelona, Spain. 8Department of Medical Oncology, ICO, IDIBELL, L’Hospitalet, Barcelona, Spain. 9Stem Cells and Cancer Group, Vall d'Hebron Institute of Oncology (VHIO), Barcelona, Spain. 10Molecular Oncology Group, Vall d'Hebron Institute of Oncology (VHIO), Barcelona, Spain. 11Xenopat S.L., Business Bioincubator, Bellvitge Health Science Campus, Barcelona, Spain. *Corresponding authors: Dr. Enriqueta Felip, Vall d'Hebron Institute of Oncology, P. Vall d'Hebron 119-129, 08035 Barcelona, Spain. Phone:+34 932746085; Fax:+34 932746059; E-mail: [email protected]; and Dr. Ana Vivancos, Vall d'Hebron Institute of Oncology, P. Vall d'Hebron 119-129, 08035 Barcelona, Spain. E-mail: [email protected]. The authors declare no potential conflicts of interest.

ABSTRACT Background Third-generation epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) such as osimertinib are the last line of targeted treatment for metastatic non-small cell lung cancer (NSCLC) EGFR-mutant harboring T790M. Different mechanisms of acquired resistance to third-generation EGFR-TKIs have been proposed. It is therefore crucial to identify new and effective strategies to overcome successive acquired mechanisms of resistance. Methods For Amplicon-seq analysis, samples from the index patient (primary and metastasis lesions at different timepoints) as well as the patient derived orthotopic xenograft (PDOX) tumors corresponding to the different treatment arms were used. All samples were formalin-fixed paraffin-embedded (FFPE), selected and evaluated by a pathologist. For ddPCR, twenty patients diagnosed with NSCLC at baseline or progression to different lines of TKI therapies were selected. FFPE blocks corresponding to either primary tumor or metastasis specimens were used for analysis. For single cell analysis orthotopically grown metastases were dissected from the brain of an athymic nu/nu mouse and cryopreserved at -80ºC. © The Author 2017. Published by Oxford University Press on behalf of the European Society for Medical Oncology. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact [email protected]

Results In a brain metastasis lesion from a NSCLC patient presenting an EGFR T790M mutation we detected MET gene amplification after prolonged treatment with osimertinib. Importantly, the combination of capmatinib (c-MET inhibitor) and afatinib (ErbB-1/2/4 inhibitor) completely suppressed tumor growth in mice orthotopically injected with cells derived from this brain metastasis. In those mice treated with capmatinib or afatinib as monotherapy we observed the emergence of KRAS G12C clones. Single cell gene expression analyses also revealed intratumor heterogeneity, indicating the presence of a KRAS-driven subclone. We also detected low frequent KRAS G12C alleles in patients treated with various EGFR-TKIs. Conclusion Acquired resistance to subsequent EGFR TKI treatment lines in EGFR-mutant lung cancer patients may induce genetic plasticity. We assess the biological insights of tumor heterogeneity in an osimertinib-resistant tumor with acquired MET-amplification and propose new treatment strategies in this situation. Keywords: NSCLC, EGFR, T790M, MET, Acquired resistance, Intra-tumor plasticity

KEY MESSAGE: Molecular mechanisms that underline progression to successive EGFR TKIs lines in EGFRmutant lung cancer are poorly understood. Single-cell analysis showed an extreme genetic plasticity. We characterize the role of MET gene amplification and signaling as a mechanism of resistance to osimertinib. The combined blockade of EGFR and cMET overcomes resistance and is proposed as a new treatment strategy.

INTRODUCTION Compared with standard first-line platinum-based chemotherapy, first and second-generation TKIs blocking EGFR signaling have improved outcomes for lung cancer patients with activating mutations in the EGFR gene1-3. However, acquired resistance through a second-site mutation at position 790 (T790M) in the EGFR kinase domain limits the potential of these therapies4. Third-generation T790M inhibitors such as osimertinib5, rociletinib6, olmutinib7, and nazartinib8 are covalent mutant-selective EGFR-TKIs targeting sensitizing mutations in the presence of the T790M. Although these drugs are showing clinical benefit for lung cancer patients9,10, resistance occurs and the lack of further treatment options currently represents a major challenge in the field.

Recent data suggest several tertiary mutations in EGFR, such as C797S, L798I and L718Q as mechanisms of resistance to third-generation TKIs targeting EGFR T790M11,12,13. Finally, osimertinib resistance is being linked to either ERBB2 copy number gain, MET gene amplification, NRAS E63K or KRAS G12S mutations14, 15, 16.

METHODS Here, present the case of a patient with a metastatic lung adenocarcinoma. For the described study we obtained tumor sample from lung tumor and brain metastasis. This metastasis was also used for the PDOX development by injecting cells in mouse brain. All samples from both patient and PDOX, preserved as FFPE, were initially genotyped by Amplicon-seq and the orthotopically-grown metastases from the PDOX were used for the single cell analysis. ddPCR study was performed using all the available samples from patient and PDOX. Also for the ddPCR study, samples from twenty patients diagnosed with NSCLC at different stages of their treatment were selected. Full description in Supplementary methods.

RESULTS To identify new mechanisms of resistance to third-generation EGFR-TKIs and define novel treatment strategies, we analyzed the molecular evolution of tumor samples from an EGFRmutant lung cancer patient treated with consecutive lines of EGFR-TKIs (Figure 1a-c). All available samples were analyzed using targeted re-sequencing detecting mutations in a panel of fifty-seven oncogenes and tumor suppressors11 (Supplementary Table 1) or copy number alterations (CNA) using a nCounter panel. At diagnosis, the patient presented an advanced lung adenocarcinoma with mediastinal lymph nodes, lung and brain metastases initially treated with whole brain radiotherapy (Figure 1c). Since the primary lung adenocarcinoma sample harbored exon 19 deletion in EGFR, the patient was treated with erlotinib (Figure 1d). All lesions initially responded to EGFR blockade until bone metastasis appeared after 9 months of erlotinib treatment (Figure 1c, e). At that time, the patient was included in a phase I clinical trial (AURA trial), receiving treatment with osimertinib. The analysis of cfDNA detected an additional EGFR T790M mutation (Figure 1c, d). Therapy initially reduced brain metastasis and treatment with osimertinib was sustained twenty-one months until the progressive metastatic brain lesion enlarged and required surgical resection (Figure 1c, e). Following brain surgery, osimertinib was continued for and additional three months due to clinical benefit. NGS analyses on this surgical specimen once again showed the deletion of exon 19 in EGFR and the TP53 Q317fs mutation and loss of EGFR T790M mutation (Figure 1d). Additionally, we identified a highlevel amplification of the MET oncogene that was confirmed by fluorescent in situ hybridization17(FISH) (copy number of >40; MET/CEN7 ratio of >5) (Figure 1d, f), and high levels of c-MET protein by immunohistochemistry (Figure 1g). HER2 amplification was

excluded as a resistance mechanism since no amplification was detected by FISH (ERBB2 gene copy number of 6; ERBB2/CEN1718 ratio of 1.1), or by immunohistochemistry (Figure 1f and data not shown). The emergence of this MET amplification in the context of an exon 19 deletion of EGFR and a regression of EGFR T790M mutation led us to combine EGFR and c-MET inhibitors to block the growth of the progressive brain metastasis19. Unfortunately, the patient suffered a rapid relapse and died soon after brain surgery. At the time of surgery of brain metastasis, we obtained surgical tumor tissue to implant orthotopically in immunodeficient nude mice, generating an orthoxenograft or patient-derived orthotopic xenograft (PDOX) model (Figure 2a) 20, 21. PDOXs present high concordance with the original clinical tumors22, 23. In this particular case, PDOX not only faithfully recapitulated the patient´s histology but also preserved MET amplification (Figure 2b, c) and similar EGFR status (total proteins by IHC and CNV using FISH) (Supplementary Figure 3 and Supplementary Table 4). This model allowed us to explore the efficacy of an EGFR inhibitor and c-MET inhibitor combined. Passable biopsies were orthotopically implanted into the brain of thirty-five nude mice that were randomized and treated with vehicle, cisplatin/pemetrexed (standard chemotherapy), osimertinib (EGFR sensitizing and T790M resistance mutation inhibitor), afatinib (ErbB-1/2/4 inhibitor), capmatinib (c-MET inhibitor) and a combination of capmatinib and afatinib (Figure 2a). All treatments were administered during twenty-one days. Capmatinib alone or combined with afatinib showed superior efficacy, significantly increasing the overall survival of mice (Figure 2d). Strikingly, none of the capmatinib/afatinib treated mice displayed weight loss, increased intracranial pressure, presented any tumor evidence, or scaring in the brain or any other analyzed tissues after three hundred days upon tumor implantation. These data demonstrate that capmatinib/afatinib treatment cured all mice. In the case of capmatinib monotherapy two mice died two months after tumor implantation presenting brain tumors upon necropsy. Another two mice died after nine months with no brain tumor, but one presented a lung metastasis and the other a mesenteric lesion. When treated with afatinib alone, all mice progressed with growing brain tumors and had to be sacrificed earlier after treatment initiation. Similarly, PDOX treated with osimertinib did not show any benefit, confirming the resistance observed in the patient. In summary, c-MET, as opposed to EGFR blockade, was effective. The combination of the two however, was the most potent therapy showing curative potential. We then genotyped PDOX samples obtained from mice that progressed to the different treatments (Figure 2g). All xenograft tissues showed the same exon 19 deletion in EGFR, TP53 Q317fs mutation as well as MET amplification detected in the original patient´s brain metastasis (Figure 2c, e, f). In addition, we observed a subclonal TP53 Q165K mutation in some xenografts. Interestingly, we detected the emergence of a subclonal KRAS G12C mutation exclusively in xenograft tumors from mice treated with afatinib or capmatinib as monotherapy.

This data suggested the surfacing of minor preexisting KRAS G12C mutant clones as a mechanism of resistance to effective EGFR or c-MET signaling blockade. In the original patient´s metastatic brain tumor biopsy we actually confirmed the existence of EGFR T790M and KRAS G12C mutations at low allele frequencies using droplet digital PCR24 (ddPCR). To study this phenomenon further, we evaluated clonal distribution within xenograft tumor samples by single cell transcriptome analysis (massive parallel single cell RNA-sequencing, MARS-Seq25, 26). We sequenced 197 randomly selected cells from a tumor xenograft that grew in the brain of a capmatinib treated mouse and presented a KRAS G12C mutation and an exon 19 deletion in EGFR (Figure 2d, e). Using hierarchical clustering, or dimensional reduction representations (tSNE), we grouped single cells based on their differential transcriptional profiles and identified two main subpopulations (Figure 3a, b). We hypothesized that these two subpopulations may represent tumor subclones driven by either KRAS or EGFR activating mutations. To test this hypothesis, we first defined EGFR and KRAS distinctive transcriptional signatures by comparing primary lung adenocarcinoma specimens’ mutant for EGFR or KRAS 27 (Supplementary Table 2-3). Remarkably, KRAS-activated genes were upregulated in the less abundant subclone, while EGFR-related genes were activated in the remaining tumor cells (Figure 3c, d). Indeed, we observed a significantly increased expression of the KRAS- or EGFRsignature genes in the minor and major subpopulation, respectively, supporting their distinct activities in the putative tumor subclones (Student's t-test, Figure 3e, f). The putative EGFRdriven subclone showed a significant association to genes whose expression was altered following targeted EGFR inhibition in vitro (Supplementary Figure 1a-d), further supporting a clonal separation of the oncogenes. Collectively, these results support the existence of two distinct tumor subclones driven by either KRAS or EGFR activating mutations. Surprisingly, we further noticed the increased expression of immune system related genes in the KRAS-driven subclone (Supplementary Figure 1e-f). We analyzed the PD-L1 expression by IHC in patient brain metastasis, PDOX KRAS WT and PDOX KRAS Mut (Supplementary Figure 2). The presence of minor KRAS mutant clones could be a clinically relevant mechanism of resistance to EGFR-TKIs and/or c-MET inhibitors and remain undetectable by standard techniques (NGS, qPCR, Sanger sequencing). Consequently we used the most sensitive genetic assay, ddPCR23 for a retrospectively genetic profiling of EGFR-mutated lung cancer patient samples (Table 1). In the biopsies at the time of progression to EGFR-TKIs from thirteen EGFR-mutated patients, we detected five EGFR T790M and three KRAS G12C mutant tumors. These patients were originally considered wild type for these alterations when evaluated with NGS (Table 1). Furthermore, none of the seven tumor samples evaluated from surgical earlystage NSCLC patients with the presence of mutation in EGFR and naïve to EGFR-TKIs presented KRAS G12C mutations. In one of the samples we detected EGFR T790M.

DISCUSSION In summary, we observed how a lung adenocarcinoma presenting an activating deletion of exon 19 in the EGFR gene acquired a second T790M mutation in the same gene upon treatment with erlotinib, while MET amplification was detected after subsequent osimertinib. In the same line, previous studies showed how MET copy number gain causes gefitinib resistance in CNS lesions utilizing mouse in vivo imaging models28. At this point, we also detected KRAS G12C and EGFR T790M by ddPCR. Importantly, in a PDOX model we demonstrated that this MET amplification is essential for lung cancer cell survival since capmatinib therapy proved very effective. Intriguingly, for the very first time, we show c-MET signaling inhibition with capmatinib to be more potent when combined with afatinib than as a single agent in our mouse model. This afatinib effect contrasted with its complete lack of activity as monotherapy. This benefit of combining afatinib could have been mediated by its previously described capacity to block ERBB3 or ERBB4 activations by heregulin ligand in EGFR mutant lung tumors29. This inhibition of ERBB3/4 or the inhibition of EGFR itself, are both possible mechanism that require further investigation. Our data suggest that this oncogenic ERBB activation would only be relevant for the survival of cancer cells addicted to hyperactive c-MET signaling. In this sense, c-MET and EGFR (ERBB1) form membrane heterodimers in normal and cancer cells leading to their trans-phosphorylation and activation of downstream MAPK pathway. Additionally, c-MET/KRAS/ERK signaling induces the transcription of EGF ligand and EGFR activation as a positive feedback loop. Further analyses will be required to confirm the relevance of such crosstalk between EGFR or ERBB3/4 with c-MET as a molecular determinant of response to combined c-MET and EGFR blockade in advanced lung cancer.

Our results also evidence the extreme plasticity of lung adenocarcinoma genomes that evolve to adapt to as well as survive the pharmacological pressure of third-generation EGFR-TKIs. Could this be a consequence of selecting de novo mutations in lung cancer genomes or is it reflective of the early coexistence of multiple genetic clones with distinctive capacities to resist targetdirected therapies? Our findings support the hypothesis of lung adenocarcinomas consisting of a complex map of genetic clones ready for selection under effective pharmacological pressure. We clearly observed the emergence of KRAS G12C mutant clones upon blocking two upstream activating components of the MAPK pathway such us EGFR or c-MET. Similarly, oncogenic KRAS mutations were described as resistance mechanisms to anti-EGFR antibodies in colorectal cancer 30, 31, a phenomenon that can also involve clonal enrichment upon treatment. Indeed, we observed that drugs blocking EGFR or c-MET signaling preferentially promoted the emergence of genetic alterations in EGFR, MET and KRAS genes; all essential components of the oncogenic TKR/KRAS/MAPK pathway. This particular genetic evolution confirms the strict addiction of lung tumors to TKR/KRAS/MAPK pathway as a driving force of drug-

resistance and disease progression. Consistent with our aforementioned observations, subsequent therapy should be assessed as a combination of the EGFR inhibitor with c-MET inhibitors.

In these highly heterogeneous lung tumor samples, we also noted a subpopulation of cells presenting a distinctive KRAS gene expression signature enriched in immune-related components. Indeed, initial clinical data indicates that KRAS mutant lung adenocarcinomas could be more sensitive to immune checkpoint inhibitors. Thus, we also suggest immunotherapy as a later line of treatment for those patients with EGFR mutant lung tumors that progress to consecutive lines of EGFR-TKIs and present emergence of KRAS mutant as well as potentially immunosensitive clones.

Finally, our data indicated that lung adenocarcinomas might evolve rapidly due to the surfacing of minor pre-existing genetic clones resistant to specific targeted therapies. Therefore, more complex therapies combining EGFR-TKIs with MET inhibitors and/or immunotherapy could be considered for lung cancer patients at earlier stages. This novel approach could prevent drug resistance and disease progression later on. For this reason, the clinical implementation of genetic technologies with higher sensitivity will be crucial in defining the genetic landscape of polyclonal tumors in patients’ candidate to target-directed therapies.

ACKNOWLEDGEMENTS We want to acknowledge the Cellex Foundation for providing facilities and equipment. and Ayudas Merck Serono de Investigación 2016 for its support. We would like to thank to Amanda Wren for excellent technical assistance in writing the manuscript. We thank to Jose Jimenez and Irene Sansano for technical assistance with FFPE and immunohistochemistry.

FUNDING This work was supported by the Spanish Ministries of Health and * Fondo de Investigación Sanitaria-Fondo Europeo de Desarrollo Regional (FEDER)* (PI14/01248, PI13-01339, PIE13/00022, PI16/01898); AECC Scientific Foundation (GCB14-2170); and Fundación Mutua Madrileña (AP150932014). HGP and HH are Miguel Servet researchers funded by the Spanish Institute of Health Carlos III (CPII14/00037, CP14/00229). PN laboratory is funded by the

Tumor Biomarker Research Program of the Banco Bilbao Vizcaya Argentaria Foundation (FBBVA).

DISCLOSURE All authors have declared no conflicts of interest. REFERENCES 1. Maemondo M, Inoue A, Kobayashi K, Sugawara S, Oizumi S, Isobe H, et al. Gefitinib or chemotherapy for non-small-cell lung cancer with mutated EGFR. N. Engl. J. Med. 2010; 362: 2380-2388. 2. Rosell R, Carcereny E, Gervais R, Vergnenegre A, Massuti B, Felip E, et al. Erlotinib versus standard chemotherapy as first-line treatment for European patients with advanced EGFR mutation-positive non-small-cell lung cancer (EURTAC): a multicentre, open-label, randomised phase 3 trial. Lancet Oncol. 2012; 13: 239-246. 3. Sequist LV, Yang JC, Yamamoto N, O'Byrne K, Hirsh V, Mok T, et al. Phase III study of afatinib or cisplatin plus pemetrexed in patients with metastatic lung adenocarcinoma with EGFR mutations. J. Clin. Oncol. 2013; 31:3327-3334. 4. Sequist LV, Waltman BA, Dias-Santagata D, Digumarthy S, Turke AB, Fidias P, et al. Genotypic and histological evolution of lung cancers acquiring resistance to EGFR inhibitors. Sci. Transl. Med. 2011; 3: 75ra26. 5. Cross DA, Ashton SE, Ghiorghiu S, Eberlein C, Nebhan CA, Spitzler PJ, et al. AZD9291, an irreversible EGFR TKI, overcomes T790M-mediated resistance to EGFR inhibitors in lung cancer. Cancer Discov. 2014; 4: 1046-1061. 6. Walter AO, Sjin RT, Haringsma HJ, Ohashi K, Sun J, Lee K, et al. Discovery of a mutant-selective covalent inhibitor of EGFR that overcomes T790M mediated resistance in NSCLC. Cancer Discov. 2013; 3: 1404-1415. 7. Park K, Han JY, Kim DW, Bazhenova LA, Ou SH, Pang YK, et al. 190TiP: ELUXA 1: Phase II study of BI 1482694 (HM61713) in patients (pts) with T790M-positive nonsmall cell lung cancer (NSCLC) after treatment with an epidermal growth factor receptor tyrosine kinase inhibitor (EGFR TKI). J. Thorac. Oncol. 2016; 11 (4 Suppl): S139. 8. Jia Y, Juarez J, Li J, Manuia M, Niederst MJ, Tompkins C, et al. EGF816 exerts anticancer effects in non-small cell lung cancer by irreversibly and selectively targeting primary and acquired activating mutations in the EGF receptor. Cancer Res. 2016; 76(6): 1591-602. 9. Jänne PA, Yang JC, Kim DW, Planchard D, Ohe Y, Ramalingam SS, et al. AZD9291 in EGFR inhibitor-resistant non-small-cell lung cancer. N. Engl. J. Med. 2015; 372: 16891699. 10. Sequist LV, Soria JC, Camidge DR. Update to rociletinib data with the RECIST confirmed response rate. N. Engl. J. Med. 2016; 374:2296-2297. 11. Thress K, Paweletz CP, Felip E, Cho BC, Stetson D, Dougherty B, et al. Acquired EGFR C797S mutation mediates resistance to AZD9291 in non-small cell lung cancer harboring EGFR T790M. Nature Med. 2015; 21:560-562. 12. Chabon JJ, Simmons AD, Lovejoy AF, Esfahani MS, Newman AM, Haringsma HJ, et al. Circulating tumor DNA profiling reveals heterogeneity of EGFR inhibitor resistance mechanisms in lung cancer patients. Nat Commun. 2016;7:11815. 13. Bersanelli M, Minari R, Bordi P, Gnetti L, Bozzetti C, Squadrilli A, et al. L718Q mutation as new mechanism of acquired resistance to AZD9291 in EGFR-mutated NSCLC. J. Thorac. Oncol. 2016; 11(10): e121-123.

14. Planchard D, Loriot Y, André F, Gobert A, Auger N, Lacroix L, et al. EGFRindependent mechanisms of acquired resistance to AZD9291 in EGFR T790M-positive NSCLC patients. Ann. Oncol. 2015; 26: 2073-2078. 15. Eberlein CA, Stetson D, Markovets AA, Al-Kadhimi KJ, Lai Z, Fisher PR, et al. Acquired resistance to the mutant-selective EGFR inhibitor AZD9291 is associated with increased dependence on RAS signaling in preclinical models. Cancer Res. 2015; 12: 2489-2500. 16. Ortiz-Cuaran S, Scheffler M, Plenker D, Dahmen L, Scheel AH, Fernandez-Cuesta L, et al. Heterogeneous mechanisms of primary and acquired resistance to third-generation EGFR inhibitors. Clin. Cancer Res. 2016; 22(19):4837-4847. 17. Schildhaus HU, Schultheis AM, Rüschoff J, Binot E, Merkelbach-Bruse S, Fassunke J, et al. MET amplification status in therapy-naïve adeno- and squamous cell carcinomas of the lung. Clin. Cancer Res. 2015; 21(4):907-915. 18. Rüschoff J, Hanna W, Bilous M, Hofmann M, Osamura RY, Penault-Llorca F, et al. HER2 testing in gastric cancer: a practical approach. Mod. Pathol. 2012; 25: 637-650. 19. Scheffler M, Merkelbach-Bruse S, Bos M, Fassunke J, Gardizi M, et al. Spatial Tumor Heterogeneity in Lung Cancer with Acquired Epidermal Growth Factor ReceptorTyrosine Kinase Inhibitor Resistance: Targeting High-Level MET-Amplification and EGFR T790M Mutation Occurring at Different Sites in the Same Patient. J Thorac Oncol. 2015 Jun;10(6):e40-3. 20. John T, Kohler D, Pintilie M, Yanagawa N, Pham NA, Li M, et al. The ability to form primary tumor xenografts is predictive of increased risk of disease recurrence in earlystage non-small cell lung cancer. Clin. Cancer Res. 2011; 17: 134-141. 21. Fiebig HH, Maier A, Burger AM. Clonogenic assay with established human tumor xenografts: correlation of in vitro to in vivo activity as a basis for anticancer drug discovery. Eur. J. Cancer 2004; 40: 802-820. 22. Hoffman, R.M. Patient-derived orthotopic xenografts: better mimic of metastasis than subcutaneous xenografts. Nat. Rev. Cancer 2015;15:451-452. 23. Fichtner I, Rolff J, Soong R, Hoffmann J, Hammer S, Sommer A, et al. Establishment of patient-derived non-small cell lung cancer xenografts as models for the identification of predictive biomarkers. Clin. Cancer Res. 2008; 14: 6456-6468. 24. Hindson BJ, Ness KD, Masquelier DA, Belgrader P, Heredia NJ, Makarewicz AJ, et al. High-throughput droplet digital PCR system for absolute quantitation of DNA copy number. Anal. Chem. 2011; 83:8604-8610. 25. Jaitin DA, Kenigsberg E, Keren-Shaul H, Elefant N, Paul F, Zaretsky I, et al. Massively parallel single-cell RNA-seq for marker-free decomposition of tissues into cell types. Science 2014; 343:776-779. 26. Paul F, Arkin Y, Giladi A, Jaitin DA, Kenigsberg E, Keren-Shaul H, et al. Transcriptional heterogeneity and lineage commitment in myeloid progenitors. Cell 2015; 163: 1663-1677. 27. The Cancer Genome Atlas Research Network. Comprehensive molecular profiling of lung adenocarcinoma. Nature 2014; 511: 543-550. 28. Nanjo S, Arai S, Wang W, Takeuchi S, Yamada T, Hata A, et al. MET Copy Number Gain Is Associated with Gefitinib Resistance in Leptomeningeal Carcinomatosis of EGFR-mutant Lung Cancer. Mol Cancer Ther. 2017 Mar;16(3):506-515. 29. Yonesaka K, Kudo K, Nishida S, Takahama T, Iwasa T, Yoshida T, et al. The pan-HER family tyrosine kinase inhibitor afatinib overcomes HER3 ligand heregulin-mediated resistance to EGFR inhibitors in non-small cell lung cancer. Oncotarget. 2015; 6: 33602-33611. 30. Misale S, Yaeger R, Hobor S, Scala E, Janakiraman M, Liska D, et al. Emergence of KRAS mutations and acquired resistance to anti-EGFR therapy in colorectal cancer. Nature 2012; 486:532-536. 31. Diaz LA Jr, Williams RT, Kinde I, Hecht JR, Berlin J, Allen B, et al. The molecular evolution of acquired resistance to targeted EGFR blockade in colorectal cancers. Nature 2012; 486: 537-540.

32. Fan J, Salathia N, Liu R, Kaeser GE, Yung YC, Herman JL, Kaper F, et al. Characterizing transcriptional heterogeneity through pathway and gene set overdispersion analysis. Nat Methods 2016; 13:241-244. 33. Marco-Sola S, Sammeth M, Guigó R, Ribeca P. The GEM mapper: fast, accurate and versatile alignment by filtration. Nat Methods 2012; 9:1185–1188. FIGURE LEGENDS File name: Figure 1 format JPEG.jpg Figure 1 Evolution and plasticity of acquired resistance mechanisms to osimertinib in NSCLC harboring EGFR mutation. (a) Study of the molecular profiling of metastatic brain biopsy specimen of female patient with NSCLC exon 19 deletion and T790M mutation treated with osimertinib. (a, c, d) ADC: Adenocarcinoma. (b) Morphological appearance of primary and metastatic lung lesions (haematoxylin and eosin, 20X). (c) Serial of target tumor lesions measures and the lower panel displays anti-EGFR treatment, imaging evaluation and genotyping along the evolution of the metastatic disease. (d) Molecular profiling of paired biopsies: baseline and at the time of progression to erlotinib and osimertinib. n. d.: non-determined. (e) Representative brain MRI and CT scans at the time points indicated are provided; the largest brain target lesion is indicated with an arrow. (f) FISH analyses showing the presence of MET amplification in the brain metastasis after relapse osimertinib (MET gene, green signals; CEN7, red signals; 100X). (g) High expression of cMET and EGFR proteins was observed in brain lesion by immunohistochemistry. No expression for HER2 was found (2.5X). File name: Figure 2 format JPEG.jpg Figure 2 Orthotopic patient-derived xenograft (PDOX) models using the same fresh metastatic brain biopsy of our patient at the time of progression to osimertinib. (a) Different PDOX cohorts that received treatment with vehicle, osimertinib, cisplatin/pemetrexed, afatinib, capmatinib and a combination of capmatinib and afatinib (capmatinib/afatinib). (a, b, e) Cis: cisplatin. Pem: pemetrexed. Cap: capmatinib. Afa: afatinib. (b) Representative images showing high similarity between patient brain metastasis and its PDX (20X) (c) MET gene amplification by FISH in the PDX (MET gene, green signals; CEN7, red signals; 100X). (d) Kaplan-Meier survival analysis for the different PDOX treated cohorts. (e) Genotyping of PDOX samples obtained from mice that progressed to the different treatments. VAF: variant allele frequency. (f) Representation of clonal evolution of the acquired resistance. KRAS G12C and EGFR T790M mutations were only detected by ddPCR in patient lesions. n. d.: non-determined. ADC: Adenocarcinoma. File name: Figure 3 format JPEG.jpg Figure 3 Single cell transcriptome profiles point to the presence of a KRAS-driven subclone. (a) Hierarchical clustering of 197 single cells (columns) derived from a capmatinib-resistant PDOX using the most variable gene sets32. Cells are grouped into two putative subclones (column labels) and correlating gene sets are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). (b) Gene expression variances between cells displayed as t-distributed stochastic neighbor embedding (t-SNE) representation using previous defined distances and cluster identities (as in a). (c) Gene expression signatures derived from KRAS (upper panel) or EGFR (lower panel) mutant primary lung adenocarcinomas27. Gene expression levels of single cells are displayed as relative intensities22. Displayed are the 25 most variant genes and signatures are summarized in the panel above (orange: overrepresented; green: underrepresented). (d) Mutational signature intensities of single cells. Cells are separated by their signature expression levels for EGFR and KRAS mutations. Cells were assigned to clusters as in (a) (e, f) Direct comparison of KRAS (e) or EGFR (f) signature scores between the putative subclones (KRAS:red; EGFR:black). Significant differences between groups (Student’s t-test) are indicated. File name: Table 1 format Word.doc Table 1

Twenty EGFR mutated lung cancer samples were assessed retrospectively by a ddPCR assay. Thirteen tumor samples from EGFR mutated patients at the time of progression to EGFR-TKIs were analyzed. Seven biopsies were evaluated from surgical early-stage NSCLC patients with the presence of EGFR mutation and naïve for EGFR-TKI therapy. Supplementary Information is available in the online version of the letter.

MATERIALS AND METHODS Mutation profiling Patients, samples and DNA extraction For Amplicon-seq analysis, samples from the index patient (primary and metastasis lesions at different timepoints) as well as the derived PDOX tumors corresponding to the different treatment arms were used. All samples were formalin-fixed paraffin-embedded (FFPE), selected and evaluated by a pathologist. For ddPCR, twenty patients diagnosed with NSCLC at baseline or progression to different lines of TKI therapies were selected. FFPE blocks corresponding to either primary tumor or metastasis specimens were used for analysis. A minimum tumor content of 20% was set for both downstream techniques. DNA was extracted from 5x10 µm sliced sections of FFPE material using the Maxwell FFPE Tissue LEV DNA Purification Kit (Promega), according to the manufacturer’s instructions. DNA quality and concentration were measured with Nanodrop 1000 spectrophotometer (Thermo Scientific, Waltham, MA).

Amplicon-seq An initial multiplex-PCR with a proof-reading polymerase was performed. See Supplemental Table 1 for a list of genes included. Indexed libraries were pooled and sequenced in a MiSeq instrument (2X100) at an average coverage of 1000X. Initial alignment was performed with BWA after primer sequence clipping and variant calling was done with the GATK Unified Genotyper and VarScan2 followed by ANNOVAR annotation. Mutations were called at a minimum 3% allele frequency. SNPs were filtered out with dbSNP and 1000 genome datasets. For PDOX samples, murine genome reads were also filtered. All detected variants were manually checked.

Droplet digital PCR (ddPCR) The QX200 Droplet Digital PCR (ddPCR™) System (Bio-Rad Laboratories, Hercules, CA) was used to detect the EGFR p.T790M and KRAS p.G12C variants. Certified probes of the manufacturer were used for the assays (BioRad). The platform allows partitioning of the ddPCR reaction mix into 20000 droplets. Thanks to this process, each droplet harbors an individual PCR reaction. After amplification, droplets are analyzed through a two-color optical detection system and according to the fluorescence signal they are designated as positive or negative for the interrogated variant. Depending of the number of analyzed droplets, sensitivity can reach 0.01% MAF. The 20µL final volume of TaqMan PCR reaction mixture was assembled with 1x ddPCR Supermix for Probes (no dUTP), 900nM of each primer, 250nM of each probe and 100ng of genomic DNA templates (8µL). Each assay (EGFR p.T790M and KRAS p.G12C) was performed in triplicate in separate mixes and loaded in different wells for amplification. The thermal cycling program was performed according to specifications of the manufacturer. After PCR, droplets were read in the Droplet Reader and analyzed with QuantaSoft version 1.7.4. Human reference genomic DNA was included as negative control and used to determine the cutoff for allele calling in each assay. FISH/IHC assays Fluorescent in situ hybridization (FISH) analyses were performed using dual-colors probes for cMET/CEP7 (Zytovision, cat #Z-2087) and HER2/CEP 17 (Dako pharmDx, cat K5731), and EGFR/CEP7 (Agilent Technologies, cat # G100377R-8 and G110902G-8). The following antibodies were used for immunohistochemistry (IHC): anti-cMET clone D1C2 (Cell signaling technologies, cat

#8198), anti-EGFR clone 5B7 (Ventana, cat #790-4347), anti-HER2 (Dako Herceptest, cat #K5207) and anti-PD-L1 clone SP263 (Ventana, cat #790-4905).

Single-cell analysis Primary sample preparation For single cell analysis orthotopically grown metastases were dissected from the brain of an athymic nu/nu mouse and cryopreserved (10% DMSO + 90% non-inactivated FBS) at -80ºC. In order to separate single cells, the samples were rapidly thawed in a water bath in continuous agitation and placed into 25 ml of cold 1x HBSS. Single cells were separated by enzymatic digestion in 5 ml 1x HBSS and 83 ul collagenase IV (10,000 U/ml) for 5 min at 37ºC. Single cells were separated by passing the sample through a 0.9 mm needle and filtration (70um nylon mesh). Cells were washed once in ice cold 1x HBSS and resuspended in DMEM before flow cytometry. In order to enrich human cells during the sorting procedure, tumor cells were stained for 1h at 4ºC with EpCam (CD326, eBioscience, 1:100). Propidium iodide staining identified dead and damaged cells. Library preparation and sequencing To construct single cell libraries from polyA-tailed RNA, we applied massively parallel single-cell RNA sequencing (MARS-Seq)25, 26, a technique enriching for sequencing reads at the 3-end of transcripts. Briefly, EpCam-positive single cells were FACS-sorted into 384-well plates, containing lysis buffer and reverse-transcription (RT) primers. The RT primers contained a single cell barcodes and unique molecular identifiers (UMI) for subsequent de-multiplexing and correction for amplification biases, respectively. Spike-in artificial transcripts (ERCC) were added at a dilution of 1:16x106. PolyAcontaining RNA was converted into cDNA as previously described25, 26 and then pooled using an automated pipeline (liquid handling robotics). Subsequently, samples were linearly amplified by in vitro transcription, fragmented, and 3-ends were converted into sequencing libraries. The libraries consisted of 192 single cell pools. Two multiplexed pools were run in one Illumina HiSeq Rapid two lane flow cell following the manufacturer’s protocol. Primary data analysis was carried out with the standard Illumina pipeline. We produced 83 nt of transcript sequence reads.

Data processing The MARS-Seq technique takes advantage of two-level indexing that allows the multiplexed sequencing of 192 cells per pool and of multiple pools per sequencing lane. Sequencing was carried out as paired-end reads, wherein the first read contains the transcript sequence and the second read the cell barcode and UMIs. Quality check of the generated reads was performed with FastQC quality control suite. Samples that reach the quality standards were then processed to deconvolute the reads to single cell level by demultiplexing according to the cell and pool barcodes. Reads were filtered to remove polyT sequences. Sequencing reads were mapped with the RNA pipeline of the GEMTools 1.7.0 suite33 on the genome references for human (Gencode release 24, assembly GRCh38.p5) and mouse (Gencode release M8, assembly GRCm38.p4). We adapted GEMTools RNASeq pipeline for its usage on single cell reads. The analysis of spike-in control RNA allowed us to discard reads from empty wells or damaged cells. Cells with less than 105 reads or more than 2x106 reads were discarded. Gene quantification was performed using UMI corrected transcript information to correct for amplification biases. Following filtering steps 197 single cells entered subsequent analysis steps.

Data analysis In order to define gene expression signatures associated to mutations in KRAS or EGFR, we accessed RNA sequencing data sets (RNAseq v2; level 3) and mutation status for 230 lung adenocarcinoma samples from The Cancer Genome Atlas (TCGA) data portal27. Samples were grouped in KRAS-mutated (n=31), EGFR-mutated (n=58) and non-mutated (n=141) and significantly differentially expressed genes were assessed using edgeR (FDR<0.001). Gene expression signatures associated to EGFR-specific treatment were obtained from the Library of Integrated Cellular Signatures (LINCS) data portal. Differentially expressed genes following treatment of the lung adenocarcinoma cell line A549 with erlotinib (LSM-1097), gefitinib (LSM-1098) or lapatinib (LSM-1051) were directly assessed from the

data portal. An EGFR inhibitor associated signature (containing 284 genes) was assessed from genes with consistent expression changes. A cell-to-cell transcriptional heterogeneity analysis has been performed for 197 cells and 13,007 genes using Pagoda32. Low-quality cells were removed based on the distribution of the number of non-zero count genes per cell (minimum number of genes detected), removing cells having more than 2 median absolute deviations (MAD) below the median of the minimum number of genes. Genes not expressed in at least 2.5% of the cells and having a minimum level of expression less than the lower quartile of gene expressions were discarded. In addiction of that, we excluded mitochondrial genes from the analysis, which represented a potential confounding effect in this study. Cell models were computed scaling parameters to the matrix size and adapting them to the use of UMI counts. The value of correlation with the expected magnitude was positive for all cells, indicating a good quality fit. The observed expression variance was normalized to the expected for each gene. Further, variability introduced during the experimental phase (e.g. sequencing pools) has been taken into account to control for undesired aspects of heterogeneity. Cell models were then included in the subsequent gene set overdispersion analysis. Subpopulations were identified using informative signatures (described above) and de novo gene sets (based on Pearson’s correlations among cell measurements). The most significant aspects of heterogeneity were represented as heatmap (Fig. 3a). The two cell clusters revealed a subclone (~32% of cells) highly expressing a signature associated to KRAS-mutated tumors (TCGA, KRAS-mutated upregulated, p=5.3e-26; Fig.3e). On the other hand, the major cluster was primarily related to an overdispersed EGFR signature (TCGA, EGFRmutated upregulated, p=1.3e-18, Fig. 3f). Major subpopulations were also evident visualizing cell pairwise distances (obtained from the overall clustering) in a 2-dimension t-distributed stochastic neighbor embedding (t-SNE) plot (Fig. 3b). A more detailed analysis of enriched de novo transcriptional set of genes within the cell clusters highlighted correlating genes involved in immune response, being overrepresented in the putative KRAS-mutated cluster (p=1.6e-26, Supplementary Fig. 1e,f).

Generation and perpetuation of the Orthoxenograft or Patient-Derived Orthotopic Xenograft (PDOX) The human tumor specimen, a small brain metastases biopsy of a lung adenocarcinoma from a patient treated with osimertinib take at Vall d’Hebron University Hospital (VH) -Barcelona, Spain- was aseptically isolated and placed at room temperature in DMEM supplemented with 10% FBS plus 50 U/ml penicillin and 50 mg/ml streptomycin. The patient study was approved by the Institutional Review Board of VH hospital and written informed consent was collected from the patient to implant the tumor in mice. Within two hours from surgical resection biopsy was implanted in nude mice (strain Crl:NUFoxn1nu)(Harlam) at the animal core facility of Bellvitge Biomedical Research Institute (IDIBELL)_ICO. To generate the orthoxenograft or patient derived orthotopic xenograft (PDOX), female mice of 6 to 8 weeks were anesthetized with a continuous flow of 2% to isoflurane/oxygen mixture, and mechanically disaggregated biopsy in saline serum was by intracerebral stereotactic injection into the posterior right part of the cortex four mice with a 25G needle. Mice were inspected twice a week, and monitored for mice weight loss and movement problems as indicators of tumor growth. Animals were housed in individually ventilated cages on a 12-hour light-dark cycle at 21 to 23 ºC and 40% to 60% humidity. Mice were allowed free access to an irradiated diet and sterilized water. At mice sacrifice (50 to 54 days after inoculation), the brain tumor was harvested, cut into small fragments, disaggregate and serially transplanted into 3 to 5 new animals. Engrafted tumors were also cryopreserved in a solution of 90% non-inactivated FBS and 10% dimethyl-sulfoxide and stored in liquid nitrogen for subsequent future implantations and advanced molecular analysis. Representative tumor fragments and the brain, lung and liver tissues from the different mice were fixed and then processed for paraffin embedding. All animal protocols were reviewed and approved according to regional Institutional Animal Care and Use Committees.

Growth curve and drug treatment of engrafted orthoxenograft Twenty mice were implanted with PDOX at passage#2 to generate the kinetic tumor growth curve. Thus, mice were successively sacrificed (n=4/time point) at day 10, 15, 20, 25 and 30, and the brain fixed and processed for paraffin embedding and analyzed by H&E for presence/absence of tumor cells. Thirty-five mice were implanted with a mix of the disaggregate of three independent PDOX at passage#3, and 15 days after implantation the mice were randomly allocated into the different treatment groups (n=6-

8/group) that were included in all cages: I) Placebo; II) Afatinib (20 mg/kg); III) Cisplatin (3.5 mg/kg)+Pemetrexed (100 mg/kg); IV) osimertinib (5 mg/kg); V) Capmatinib (20 mg/kg) and VI) Afatinib (20 mg/kg)+Capmatininb (25 mg/kg). Cisplatin was i.v. administered and Pemetrexed i.p at days 0, 5, and 15. Afatinib, Osimertinib and Capmatinib were orally administered daily during 15 days. For combined Afatinib+Capmatinib, animals were treated first with Afatinib and 2 hours latter with Capmatinib. Afatininb (Selleckchem) was dissolved in 0.5% methylocellulose-0.4% Tween 80; Capmatininb (Selleckchem) in 5% Dimethyl acetamide (DMA)-0.5% methylocellulose-0.4% Tween 80; Osimertinib (Selleckchem) 2,5% Dimethyl sulfoxide (DMSO)-30% Polyethylene glycol 300(PEG-300). Cisplatin (Pharmacia, Pfizer) and Pemetrexed (Alimta, Lilly) were obtained as liquid solution from hospital pharmacy of Catalan Institute of Oncology and diluted in saline. When mice showed symptoms of brain tumor progression (lost of weight and/or activity) mice were sacrificed, its tumor dissected out and representative fragments were either frozen, cryopreserved or fixed and then processed for paraffin embedding. Additionally, the rest of the brain, lung and liver tissues from the different mice were fixed and then processed for paraffin embedding and histologically studies. Kaplan-Meier plots were built for the different treatments and for the best responders treatment groups Capmatinib and combined Afatinib + Capmatinib, the mice were sacrificed (censured) to evaluate by histological analysis the presence of tumor cells (8 months after the end of the treatment)

SUPPLEMENTARY FIGURE LEGENDS

File name: Supplementary Table 1 format JPEG.jpg Supplementary Table 1 Amplicon-Seq VHIO-Card Panel V3 with the list of fifty-seven included genes.

File name: Supplementary Figure 1 format JPEG.jpg Supplementary Figure 1 | Gene expression programs support subclonal structures within the capmatinib-resistant PDOX. Hierarchical clustering of 197 single cancer cells (columns) using previously defined distances (Fig. 3a). Single cells were clustered based on their transcriptional profiles and assigned to two putative subclones (column labels). (a,c) Gene expression signatures derived from EGFR inhibitor treated lung adenocarcinoma cells (A549). Expression levels of genes correlating (a) or anti-correlating (c) with EGFR activity following treatment are displayed as relative intensities42. Displayed are the 25 most variant genes per signature and summarized in the panel above (orange: overrepresented; green: underrepresented). (b,d) Direct comparison of EGFR correlation (b) or anti-correlation (d) signature scores between the putative subclones (KRAS:red; EGFR:black). Significant differences between groups (Student’s t-test) are indicated. (e,f) An immune system related signature was enriched based on a de novo identification of correlating genes42. (e) Displayed are the 25 most variant genes for the signature which is summarized in the panel above (orange: overrepresented; green: underrepresented). (f) Direct comparison of immune response related genes between the putative subclones (KRAS:red; EGFR:black).

File name: Supplementary Figure 2 format.jpg Supplementary Figure 2 | PD-L1 expression by IHC in patient brain metastasis, PDOX KRAS WT (n=3) and PDOX KRAS Mut (n=3). (a) PD-L1 positive staining in patient brain metastasis (PD-L1 stronger staining observed than in PDOX). (b) Similar to patient sample, weak to moderate expression in >50% of PDOX tumor cells.

File name: Supplementary Figure 3 format.jpg

Supplementary Figure 3 | Analysis of the EGFR expression status (total protein by IHC) in (a) patient brain metastasis, (b) PDOX KRAS WT (n=3) and PDOX KRAS Mut (n=3). High EGFR expression levels (no differences between groups).

File name: Supplementary Table 2.1 JPEG.jpg; Supplementary Table 2.2 JPEG.jpg; Supplementary Table 2.3 JPEG.jpg; Supplementary Table 2.4 JPEG.jpg. Supplementary Table 2 | Significantly upregulated genes in KRAS-mutant lung adenocarcinomas (TCGA). logFC: logarithmic fold-change. logCPM: logarithmic counts per million. FDR: False discovery rate corrected p-value

File name: Supplementary Table 3.1 JPEG.jpg; Supplementary Table 3.2 JPEG.jpg; Supplementary Table 3 | Significantly upregulated genes in EGFR-mutant lung adenocarcinomas (TCGA). logFC: logarithmic fold-change. logCPM: logarithmic counts per million. FDR: False discovery rate corrected p-value

File name: Supplementary Table 4.jpg Supplementary Table 4 | Sequential analysis of the EGFR amplification status (CNV using FISH) in relevant samples. Ratio Negative (not amplified due to polisomy Chr7) but high gene copy number. Patient brain met: same results although higher copy number. No differences in EGFR copy number between KRAS WT and Mutant KRAS PDOXs.

Figure 1 Evolution and plasticity of acquired resistance mechanisms to osimertinib in NSCLC harboring EGFR mutation. (a) Study of the molecular profiling of metastatic brain biopsy specimen of female patient with NSCLC exon 19 deletion and T790M mutation treated with osimertinib. (a, c, d) ADC: Adenocarcinoma. (b) Morphological appearance of primary and metastatic lung lesions (haematoxylin and eosin, 20X). (c) Serial of target tumor lesions measures and the lower panel displays anti-EGFR treatment, imaging evaluation and genotyping along the evolution of the metastatic disease. (d) Molecular profiling of paired biopsies: baseline and at the time of progression to erlotinib and osimertinib. n. d.: non-determined. (e) Representative brain MRI and CT scans at the time points indicated are provided; the largest brain target lesion is indicated with an arrow. (f) FISH analyses showing the presence of MET amplification in the brain metastasis after relapse osimertinib (MET gene, green signals; CEN7, red signals; 100X). (g) High expression of cMET and EGFR proteins was observed in brain lesion by immunohistochemistry. No expression for HER2 was found (2.5X).

432x306mm (72 x 72 DPI)

Figure 2 Orthotopic patient-derived xenograft (PDOX) models using the same fresh metastatic brain biopsy of our patient at the time of progression to osimertinib. (a) Different PDOX cohorts that received treatment with vehicle, osimertinib, cisplatin/pemetrexed, afatinib, capmatinib and a combination of capmatinib and afatinib (capmatinib/afatinib). (a, b, e) Cis: cisplatin. Pem: pemetrexed. Cap: capmatinib. Afa: afatinib. (b) Representative images showing high similarity between patient brain metastasis and its PDX (20X) (c) MET gene amplification by FISH in the PDX (MET gene, green signals; CEN7, red signals; 100X). (d) Kaplan-Meier survival analysis for the different PDOX treated cohorts. (e) Genotyping of PDOX samples obtained from mice that progressed to the different treatments. VAF: variant allele frequency. (f) Representation of clonal evolution of the acquired resistance. KRAS G12C and EGFR T790M mutations were only detected by ddPCR in patient lesions. n. d.: non-determined. ADC: Adenocarcinoma.

504x302mm (72 x 72 DPI)

Figure 3 | Single cell transcriptome profiles point to the presence of a KRAS-driven subclone. (a) Hierarchical clustering of 197 single cells (columns) derived from a capmatinib-resistant PDOX using the most variable gene sets42. Cells are grouped into two putative subclones (column labels) and correlating gene sets are summarized in aspects. Displayed are the most variable aspects (rows) and their importance (row colors). (b) Gene expression variances between cells displayed as t-distributed stochastic neighbor embedding (tSNE) representation using previous defined distances and cluster identities (as in a). (c) Gene expression signatures derived from KRAS (upper panel) or EGFR (lower panel) mutant primary lung adenocarcinomas37. Gene expression levels of single cells are displayed as relative intensities42. Displayed are the 25 most variant genes and signatures are summarized in the panel above (orange: overrepresented; green: underrepresented). (d) Mutational signature intensities of single cells. Cells are separated by their signature expression levels for EGFR and KRAS mutations. Cells were assigned to clusters as in (a) (e, f) Direct comparison of KRAS (e) or EGFR (f) signature scores between the putative subclones (KRAS:red; EGFR:black). Significant differences between groups (Student’s t-test) are indicated.

406x304mm (72 x 72 DPI)

Patient sample

Gender

Smoking habit

Previous lines of treatment

Previous lines of TKI

TKI

Activating EGFR mutation

Baseline EGFR T790M (ddPCR)

Baseline KRAS G12C (ddPCR)

Progression to TKI EGFR T790M (ddPCR)

Progression to TKI KRAS G12C (ddPCR)

1

Female

Former

2

2

ex19del

N/A

N/A

13,35%

0,0027%

2

Female

Former

2

1

Gefitinib Nazartinib Erlotinib

ex19del

N/A

N/A

1,60%

0,14%

3

Female

Never

3

2

p.L858R

N/A

N/A

N/A

WT

4

Male

Former

4

1

Erlotinib Osimertinib Erlotinib

ex19del

N/A

N/A

WT

WT

5

Female

Never

4

2

ex19del

N/A

N/A

76,30%

WT

6

Male

Former

4

2

ex19del

N/A

N/A

12,20%

WT

7

Female

Former

3

2

ex19del

N/A

N/A

WT

WT

8

Female

Never

1

1

Erlotinib Nazartinib Afatinib Nazartinib Afatinib Gefitinib Erlotinib

ex19del

N/A

N/A

WT

WT

9

Female

Never

3

2

p.L858R

N/A

N/A

WT

0,75%

10

Female

Never

7

2

p.L858R

N/A

N/A

WT

WT

11

Female

Never

4

3

p.L858R

N/A

N/A

95,75%

WT

12

Female

Never

4

3

ex19del

N/A

N/A

WT

WT

13

Female

Former

7

3

ex19del

N/A

N/A

N/A

WT

14

Female

Never

Naive

0

Erlotinib Gefitinib Erlotinib Gefitinib Dacomitinib Nazartinib Osimertinib Erlotinib Rociletinib Osimertinib Gefitinib Erlotinib Osimertinib Naive

ex19del

WT

WT

N/A

N/A

15

Male

Never

Naive

0

Naive

p.L858R

WT

WT

N/A

N/A

16

Female

Former

Naive

0

Naive

ex19del

WT

WT

N/A

N/A

17

Female

Never

Naive

0

Naive

p.L858R

WT

WT

N/A

N/A

18 19

Male Female

Former Never

Naive Naive

0 0

Naive Naive

Del p.V769 p.L858R

0,33% WT

WT WT

N/A N/A

N/A N/A

20

Female

Never

Naive

0

Naive

ex19del

WT

WT

N/A

N/A

382x262mm (72 x 72 DPI)

406x275mm (72 x 72 DPI)

254x190mm (72 x 72 DPI)

254x190mm (72 x 72 DPI)

254x190mm (72 x 72 DPI)

206x292mm (72 x 72 DPI)

195x289mm (72 x 72 DPI)

204x294mm (72 x 72 DPI)

208x297mm (72 x 72 DPI)

171x280mm (72 x 72 DPI)

163x239mm (72 x 72 DPI)

MATERIALS AND METHODS Mutation profiling Patients, samples and DNA extraction For Amplicon-seq analysis, samples from the index patient (primary and metastasis lesions at different timepoints) as well as the derived PDOX tumors corresponding to the different treatment arms were used. All samples were formalin-fixed paraffin-embedded (FFPE), selected and evaluated by a pathologist. For ddPCR, twenty patients diagnosed with NSCLC at baseline or progression to different lines of TKI therapies were selected. FFPE blocks corresponding to either primary tumor or metastasis specimens were used for analysis. A minimum tumor content of 20% was set for both downstream techniques. DNA was extracted from 5x10 µm sliced sections of FFPE material using the Maxwell FFPE Tissue LEV DNA Purification Kit (Promega), according to the manufacturer’s instructions. DNA quality and concentration were measured with Nanodrop 1000 spectrophotometer (Thermo Scientific, Waltham, MA).

Amplicon-seq An initial multiplex-PCR with a proof-reading polymerase was performed. See Supplemental Table 1 for a list of genes included. Indexed libraries were pooled and sequenced in a MiSeq instrument (2X100) at an average coverage of 1000X. Initial alignment was performed with BWA after primer sequence clipping and variant calling was done with the GATK Unified Genotyper and VarScan2 followed by ANNOVAR annotation. Mutations were called at a minimum 3% allele frequency. SNPs were filtered out with dbSNP and 1000 genome datasets. For PDOX samples, murine genome reads were also filtered. All detected variants were manually checked.

Droplet digital PCR (ddPCR) The QX200 Droplet Digital PCR (ddPCR™) System (Bio-Rad Laboratories, Hercules, CA) was used to detect the EGFR p.T790M and KRAS p.G12C variants. Certified probes of the manufacturer were used for the assays (BioRad). The platform allows partitioning of the ddPCR reaction mix into 20000 droplets. Thanks to this process, each droplet harbors an individual PCR reaction. After amplification, droplets are analyzed through a two-color optical detection system and according to the fluorescence signal they are designated as positive or negative for the interrogated variant. Depending of the number of analyzed droplets, sensitivity can reach 0.01% MAF. The 20µL final volume of TaqMan PCR reaction mixture was assembled with 1x ddPCR Supermix for Probes (no dUTP), 900nM of each primer, 250nM of each probe and 100ng of genomic DNA templates (8µL). Each assay (EGFR p.T790M and KRAS p.G12C) was performed in triplicate in separate mixes and loaded in different wells for amplification. The thermal cycling program was performed according to specifications of the manufacturer. After PCR, droplets were read in the Droplet Reader and analyzed with QuantaSoft version 1.7.4. Human reference genomic DNA was included as negative control and used to determine the cutoff for allele calling in each assay. FISH/IHC assays Fluorescent in situ hybridization (FISH) analyses were performed using dual-colors probes for cMET/CEP7 (Zytovision, cat #Z-2087) and HER2/CEP 17 (Dako pharmDx, cat K5731), and EGFR/CEP7 (Agilent Technologies, cat # G100377R-8 and G110902G-8). The following antibodies were used for immunohistochemistry (IHC): anti-cMET clone D1C2 (Cell signaling technologies, cat #8198), anti-EGFR clone 5B7 (Ventana, cat #790-4347), anti-HER2 (Dako Herceptest, cat #K5207) and anti-PD-L1 clone SP263 (Ventana, cat #790-4905).

Single-cell analysis

Primary sample preparation For single cell analysis orthotopically grown metastases were dissected from the brain of an athymic nu/nu mouse and cryopreserved (10% DMSO + 90% non-inactivated FBS) at -80ºC. In order to separate single cells, the samples were rapidly thawed in a water bath in continuous agitation and placed into 25 ml of cold 1x HBSS. Single cells were separated by enzymatic digestion in 5 ml 1x HBSS and 83 ul collagenase IV (10,000 U/ml) for 5 min at 37ºC. Single cells were separated by passing the sample through a 0.9 mm needle and filtration (70um nylon mesh). Cells were washed once in ice cold 1x HBSS and resuspended in DMEM before flow cytometry. In order to enrich human cells during the sorting procedure, tumor cells were stained for 1h at 4ºC with EpCam (CD326, eBioscience, 1:100). Propidium iodide staining identified dead and damaged cells. Library preparation and sequencing To construct single cell libraries from polyA-tailed RNA, we applied massively parallel single-cell RNA sequencing (MARS-Seq)25, 26, a technique enriching for sequencing reads at the 3-end of transcripts. Briefly, EpCam-positive single cells were FACS-sorted into 384-well plates, containing lysis buffer and reversetranscription (RT) primers. The RT primers contained a single cell barcodes and unique molecular identifiers (UMI) for subsequent de-multiplexing and correction for amplification biases, respectively. Spike-in artificial transcripts (ERCC) were added at a dilution of 1:16x106. PolyA-containing RNA was converted into cDNA as previously described25, 26 and then pooled using an automated pipeline (liquid handling robotics). Subsequently, samples were linearly amplified by in vitro transcription, fragmented, and 3-ends were converted into sequencing libraries. The libraries consisted of 192 single cell pools. Two multiplexed pools were run in one Illumina HiSeq Rapid two lane flow cell following the manufacturer’s protocol. Primary data analysis was carried out with the standard Illumina pipeline. We produced 83 nt of transcript sequence reads.

Data processing The MARS-Seq technique takes advantage of two-level indexing that allows the multiplexed sequencing of 192 cells per pool and of multiple pools per sequencing lane. Sequencing was carried out as paired-end reads, wherein the first read contains the transcript sequence and the second read the cell barcode and UMIs. Quality check of the generated reads was performed with FastQC quality control suite. Samples that reach the quality standards were then processed to deconvolute the reads to single cell level by de-multiplexing according to the cell and pool barcodes. Reads were filtered to remove polyT sequences. Sequencing reads were mapped with the RNA pipeline of the GEMTools 1.7.0 suite33 on the genome references for human (Gencode release 24, assembly GRCh38.p5) and mouse (Gencode release M8, assembly GRCm38.p4). We adapted GEMTools RNASeq pipeline for its usage on single cell reads. The analysis of spike-in control RNA allowed us to discard reads from empty wells or damaged cells. Cells with less than 105 reads or more than 2x106 reads were discarded. Gene quantification was performed using UMI corrected transcript information to correct for amplification biases. Following filtering steps 197 single cells entered subsequent analysis steps.

Data analysis In order to define gene expression signatures associated to mutations in KRAS or EGFR, we accessed RNA sequencing data sets (RNAseq v2; level 3) and mutation status for 230 lung adenocarcinoma samples from The Cancer Genome Atlas (TCGA) data portal27. Samples were grouped in KRAS-mutated (n=31), EGFR-mutated (n=58) and non-mutated (n=141) and significantly differentially expressed genes were assessed using edgeR (FDR<0.001). Gene expression signatures associated to EGFR-specific treatment were obtained from the Library of Integrated Cellular Signatures (LINCS) data portal. Differentially expressed genes following treatment of the lung adenocarcinoma cell line A549 with erlotinib (LSM-1097), gefitinib (LSM-1098) or lapatinib (LSM-1051) were directly assessed from the data portal. An EGFR inhibitor associated signature (containing 284 genes) was assessed from genes with consistent expression changes. A cell-to-cell transcriptional heterogeneity analysis has been performed for 197 cells and 13,007 genes using Pagoda32. Low-quality cells were removed based on the distribution of the number of non-zero count genes per cell (minimum number of genes detected), removing cells having more than 2 median absolute deviations (MAD) below the median of the minimum number of genes. Genes not expressed in at least 2.5% of the cells and having a minimum level of expression less than the lower quartile of gene expressions were discarded. In addiction of that, we excluded mitochondrial genes from the analysis, which represented a potential confounding effect in this study. Cell models were computed scaling parameters to the matrix size and adapting them to the use of UMI counts. The value of correlation with the expected magnitude was positive for all cells, indicating a good quality fit. The observed expression variance was normalized to the expected for each gene.

Further, variability introduced during the experimental phase (e.g. sequencing pools) has been taken into account to control for undesired aspects of heterogeneity. Cell models were then included in the subsequent gene set overdispersion analysis. Subpopulations were identified using informative signatures (described above) and de novo gene sets (based on Pearson’s correlations among cell measurements). The most significant aspects of heterogeneity were represented as heatmap (Fig. 3a). The two cell clusters revealed a subclone (~32% of cells) highly expressing a signature associated to KRASmutated tumors (TCGA, KRAS-mutated upregulated, p=5.3e-26; Fig.3e). On the other hand, the major cluster was primarily related to an overdispersed EGFR signature (TCGA, EGFR-mutated upregulated, p=1.3e-18, Fig. 3f). Major subpopulations were also evident visualizing cell pairwise distances (obtained from the overall clustering) in a 2-dimension t-distributed stochastic neighbor embedding (t-SNE) plot (Fig. 3b). A more detailed analysis of enriched de novo transcriptional set of genes within the cell clusters highlighted correlating genes involved in immune response, being overrepresented in the putative KRAS-mutated cluster (p=1.6e-26, Supplementary Fig. 1e,f).

Generation and perpetuation of the Orthoxenograft or Patient-Derived Orthotopic Xenograft (PDOX) The human tumor specimen, a small brain metastases biopsy of a lung adenocarcinoma from a patient treated with osimertinib take at Vall d’Hebron University Hospital (VH) -Barcelona, Spain- was aseptically isolated and placed at room temperature in DMEM supplemented with 10% FBS plus 50 U/ml penicillin and 50 mg/ml streptomycin. The patient study was approved by the Institutional Review Board of VH hospital and written informed consent was collected from the patient to implant the tumor in mice. Within two hours from surgical resection biopsy was implanted in nude mice (strain Crl:NU-Foxn1nu)(Harlam) at the animal core facility of Bellvitge Biomedical Research Institute (IDIBELL)_ICO. To generate the orthoxenograft or patient derived orthotopic xenograft (PDOX), female mice of 6 to 8 weeks were anesthetized with a continuous flow of 2% to isoflurane/oxygen mixture, and mechanically disaggregated biopsy in saline serum was by intracerebral stereotactic injection into the posterior right part of the cortex four mice with a 25G needle. Mice were inspected twice a week, and monitored for mice weight loss and movement problems as indicators of tumor growth. Animals were housed in individually ventilated cages on a 12-hour light-dark cycle at 21 to 23 ºC and 40% to 60% humidity. Mice were allowed free access to an irradiated diet and sterilized water. At mice sacrifice (50 to 54 days after inoculation), the brain tumor was harvested, cut into small fragments, disaggregate and serially transplanted into 3 to 5 new animals. Engrafted tumors were also cryopreserved in a solution of 90% noninactivated FBS and 10% dimethyl-sulfoxide and stored in liquid nitrogen for subsequent future implantations and advanced molecular analysis. Representative tumor fragments and the brain, lung and liver tissues from the different mice were fixed and then processed for paraffin embedding. All animal protocols were reviewed and approved according to regional Institutional Animal Care and Use Committees.

Growth curve and drug treatment of engrafted orthoxenograft Twenty mice were implanted with PDOX at passage#2 to generate the kinetic tumor growth curve. Thus, mice were successively sacrificed (n=4/time point) at day 10, 15, 20, 25 and 30, and the brain fixed and processed for paraffin embedding and analyzed by H&E for presence/absence of tumor cells. Thirty-five mice were implanted with a mix of the disaggregate of three independent PDOX at passage#3, and 15 days after implantation the mice were randomly allocated into the different treatment groups (n=6-8/group) that were included in all cages: I) Placebo; II) Afatinib (20 mg/kg); III) Cisplatin (3.5 mg/kg)+Pemetrexed (100 mg/kg); IV) osimertinib (5 mg/kg); V) Capmatinib (20 mg/kg) and VI) Afatinib (20 mg/kg)+Capmatininb (25 mg/kg). Cisplatin was i.v. administered and Pemetrexed i.p at days 0, 5, and 15. Afatinib, Osimertinib and Capmatinib were orally administered daily during 15 days. For combined Afatinib+Capmatinib, animals were treated first with Afatinib and 2 hours latter with Capmatinib. Afatininb (Selleckchem) was dissolved in 0.5% methylocellulose-0.4% Tween 80; Capmatininb (Selleckchem) in 5% Dimethyl acetamide (DMA)-0.5% methylocellulose-0.4% Tween 80; Osimertinib (Selleckchem) 2,5% Dimethyl sulfoxide (DMSO)-30% Polyethylene glycol 300(PEG-300). Cisplatin (Pharmacia, Pfizer) and Pemetrexed (Alimta, Lilly) were obtained as liquid solution from hospital pharmacy of Catalan Institute of Oncology and diluted in saline. When mice showed symptoms of brain tumor progression (lost of weight and/or activity) mice were sacrificed, its tumor dissected out and representative fragments were either frozen, cryopreserved or fixed and then processed for paraffin embedding. Additionally, the rest of the brain, lung and liver tissues from the different mice were fixed and then processed for paraffin embedding and histologically studies. Kaplan-Meier plots were built for the different treatments and for the best responders treatment groups Capmatinib and combined

Afatinib+Capmatinib, the mice were sacrificed (censured) to evaluate by histological analysis the presence of tumor cells (8 months after the end of the treatment)