Abstract Advances in single cell technologies have ushered in a new multi-omics era that has provided more granular information on cell types and functions at unprecedented resolution. Morpholomics (the information encoded by cell morphology) has traditionally been used as a readout of cell identity, state, and function; but scalability, interpretation, and sorting based on morphology has remained challenging. The Deepcell® platform, REM-I, performs high-dimensional morphology analysis of unlabeled, single cells using deep-learning on high-resolution bright-field images captured in microfluidic flow to profile and sort target cells in real-time and at scale. Morphology analysis of single cells is achieved by a self-supervised deep learning foundation model, ‘Human Foundation Model’ (HFM), which combines deep learning features with morphometrics. HFM determines cell identities using morphological features associated with distinct processes characterized by granules, vesicles, cell size, pigmentation. and others. To demonstrate the ability of HFM modeling to distinguish and characterize a diversity of cell types representative of tumor microenvironments, we processed various sample types on the REM-I platform. Human melanoma, immune cells, stromal cells, and dissociated biopsies were combined to recapitulate heterogeneous tumor samples. HFM analysis enabled morphology-based exploration of the resulting bright-field images, revealing the presence of various morphologically distinct populations within each sample. Tumor cells were morphologically distinct and clustered separately from non-tumor cells. Immune cells exhibited subtle but separable morphologies, including T cells of distinct activation states. To demonstrate the morphology-based sorting capabilities of REM-I, we applied it to a ‘reference sample’ consisting of several fixed cell lines, at proportions ranging from 1% to >90%. Using only brightfield images of single cells analyzed through HFM, we used the platform to enrich each cell line from the ‘reference sample’ with high yield and purity, and identified cell types constituting a small percentage (~1%) of the total population. We next showed that unlabeled cells enriched on REM-I that are characterized as unperturbed and viable are amenable to downstream molecular analyses. These cell populations were sorted and the retrieved live cells were analyzed by scRNA-Seq (10x Chromium), demonstrating enrichment of various cell types of the tumor microenvironment in each cell population relative to the pre-sorted sample. These results illustrate: 1) the ability to perform downstream molecular analyses on sorted cells from REM-I, 2) the platform can be used to parse the tumor microenvironment, and 3) the link between morphology, cell identity, and molecular signatures. Citation Format: Andreja Jovic, Christian Corona, Kiran Saini, Tiffine Pham, Ryan Carelli, Vivian Lu, Senzeyu Zhang, Stephane C. Boutet, Maddison Masaeli. Integrating deep learning morphology profiling, cell sorting, and scRNASeq reveals sample diversity and enriched sub-populations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6934.
Read full abstract