Abstract 3772: Inter- and intra-tumoral variations in ASCL1, NEUROD1, and POU2F3 transcriptional programs underlie three distinct molecular subtypes of small cell lung cancers

Abstract

Abstract Accounting for 15% of all lung cancer diagnoses, small cell lung cancer (SCLC) is an aggressive malignancy with dismal clinical outcomes, due in part to failure to define SCLC molecular subsets and identify the unique, targetable vulnerabilities therein. Recent data has begun to delineate these subsets by uncovering inter-tumoral heterogeneity in features such as DNA damage response, EMT, and neuroendocrine (NE) status. An integrated analysis of clinical samples to determine the implications of this heterogeneity broadly on SCLC classification has not yet been performed. Using RNAseq data from 81 SCLC tumor samples, we applied non-negative matrix factorization (NMF) which identified three clusters, each enriched for unique transcriptional programs driven by ASCL1 (30/81), NEUROD1 (24/81), or POU2F3 (27/81). These three genes encode transcription factors which define mutually exclusive NE-high, NE-low, and novel tuft cell variants of SCLC. Bulk RNAseq analyses of SCLC CTC-derived xenograft (CDX) models validated the clustering analysis in vivo. However, single-cell RNAseq from these same models reveals subtle evidence of intra-tumoral and intra-cellular heterogeneity in transcriptional programs that appear mutually exclusive in bulk analyses, suggesting plasticity among these variants. Guided by our NMF results, we performed RNAseq-based supervised clustering to classify each of 60 SCLC cell lines into ASCL1-driven, NEUROD1-driven, and POU2F3-driven clusters. Then, using reverse phase protein array data for these lines, we derived proteomic signatures for each cluster as follows: ASCL1-driven: E-cadherinhigh/TTF-1high/cMYClow, NEUROD1-driven: E-cadherinlow/TTF-1low/cMYChigh, and POU2F3-driven: E-cadherinhigh/TTF-1low/cMYChigh (ANOVA p<0.001 for each). Other targetable proteins vary significantly among these clusters including BCL-2 (lower in NEUROD1-driven; p<0.001) and PD-L1 (higher in ASCL1-driven; p=0.04). Correlating these clusters with IC50 values for over 500 drugs we find unique vulnerabilities. For example, POU2F3-driven lines were more sensitive to PARP inhibitors (PARPi), despite lacking biomarkers of PARPi sensitivity in SCLC (e.g. high SLFN11, low ATM). ASCL1-driven lines were more resistant to aurora kinase inhibitors (AURKi), as predicted given their lower expression of cMYC, a predictor of AURKi sensitivity in SCLC. Our data suggest that SCLC is subdivided on the basis of three unique transcriptional programs and that each subtype is characterized by diverse protein expression and drug responses. Single-cell analyses suggest, however, that multiple transcriptional programs may coexist within a single tumor or, even, a single cell, thus providing a potential novel mechanism for lineage switching and therapeutic resistance. Citation Format: Carl M. Gay, Lixia Diao, C. Allison Stewart, Yuanxin Xi, Robert J. Cardnell, Stephen G. Swisher, Jack A. Roth, Bonnie S. Glisson, Jing Wang, John V. Heymach, Lauren A. Byers. Inter- and intra-tumoral variations in ASCL1, NEUROD1, and POU2F3 transcriptional programs underlie three distinct molecular subtypes of small cell lung cancers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 3772.