Perfecting CAR Engraftment to a Tee (Cell) through Characterization of Single Cell Transcriptome Product and Understanding Neurotoxicity

Background: Autologous chimeric antigen receptor T (CAR-T) cell therapies show significant clinical activity against hematologic malignancies including multiple myeloma (MM). Numerous factors influence the effectiveness of this therapeutic approach. The fitness of a CAR-T cell product is an important factor affecting T cell engraftment and persistence, a pre-requisite for effective CAR-T cell therapy. In an effort to deconstruct the cellular heterogeneity of CAR-T cell products and explore relationships between T cell states within the cellular product, gene expression and CAR-T cell engraftment following adoptive transfer, we performed single cell RNA-sequencing (scRNA-seq) on CAR-T cell products from a phase I trial of BCMA-specific CAR-T cells in relapsed/refractory MM patients that included cohorts receiving a similar CAR-T cell dose with and without preconditioning with cyclophosphamide (NCT02546167, Cohen et al, J Clin Invest 2019). Method: scRNA-seq was performed on 25 unique products with an average of 8823 cells captured per product at an average depth of 40779 reads per cell. Batch effects were controlled by integration using established metrics for quality control. The linked inference of genomic experimental relationships (LIGER) method exhibited the most effective integration of 228,752 single cells across the 25 products. Annotation was completed by comparing transcriptomes of cells to the Blueprint/ENCODE references of pure immune cells. Differential gene expression (DGE) analysis was conducted by both Wilcoxon rank sum test testing and negative binomial distribution. Gene enrichment analysis was performed by comparing DGE with the Molecular Signatures Database v7.4. Cell to cell communication utilized the KEGG signaling pathway maps and curated lists of interactions from the literature. The intercellular communication probability was estimated on the DE genes before statistically significant intercellular communications were calculated by a permutation test, with dominant senders, receivers, mediators, and influencers identified using graph theory. Patients were grouped based upon “good” or “poor” engraftment using a peak blood vector copies per cell cutoff of 10,000. Association of transcriptional variation with neurotoxicity was explored for all patients. Results: The CD4:CD8 ratio of 2.51 single cells in the data was comparable to the observed average CD4:CD8 ratio of total T cells in the product based upon flow cytometric analysis of th