Abstract 1686: Characterizing tumor heterogeneity through label-free, morphology-based live cell sorting

Abstract

Abstract The complexity of cancer is compounded by the continuous emergence of malignant cell populations that drive tumor heterogeneity and challenges diagnostic and therapeutic advances. Methods to study cancer at the single cell level are important to better understand tumor heterogeneity. Isolation of tumor cells from tissue typically relies on targeting biomarkers overexpressed in tumor cells, such as EpCAM, resulting in cells that may express common markers with morphologic and molecular heterogeneity. We developed a platform termed Computational Sorting and Mapping of Single cells (COSMOS) that uses high-dimensional morphology analysis of single cells using artificial intelligence, microfluidics and high resolution bright-field imaging to identify and enrich target cells. We used COSMOS to train a deep convolutional neural network (ConvNet) classifier to identify and enrich malignant cells from non-small cell lung cancer (NSCLC) dissociated tumor cell (DTC) samples. The enriched population of NSCLC cells are label-free, unperturbed and viable, making them amenable to many downstream analysis methods but importantly also contain single cell high-dimensional morphology profiles. We verified enrichment of malignant cells by performing RNA and DNA analysis on the enriched sample. Single cell RNA-seq (scRNA-Seq) analysis shows high levels of EpCAM expression in sorted cells. Copy number variation (CNV) analysis demonstrated increased amplitude of deletion and amplification peaks relative to the pre-sorted DTC sample. Further, mutational analysis show increased allele frequency of mutations including P53 and KRAS in post-sorted compared to pre-sorted samples. Morphological heterogeneity within the malignant population is observed by UMAP analysis with the presence of multiple clusters of morphologically unique tumor cell populations detected. We further trained the classifier to detect and enrich for each subpopulation; CNV, mutation, bulk RNASeq, and scRNA-Seq analysis revealed molecular differences between the morphology subgroups. Further work is planned to evaluate the link between the morphological, molecular and functional characteristics of each subpopulation. We demonstrate a platform that can enrich malignant cells based on morphology, yielding a population of cells that are label-free, viable and unperturbed, making them compatible with downstream molecular assays commonly used to study heterogeneity. Additionally the cells can be morphologically profiled at the single cell level yielding a high-dimensional morphological profile that reveals heterogeneous populations of cells that can be further characterized by molecular analysis and could offer a new dimension to understand heterogeneity and biomarker discovery. Citation Format: Andreja Jovic, Kiran Saini, Michael Phelan, Ryan Chow, Simo Zhang, Chassidy Johnson, Nianzhen Li, Thomas J. Musci, Mahyar Salek, Maddison (Mahdokht) Masaeli. Characterizing tumor heterogeneity through label-free, morphology-based live cell sorting [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1686.