Abstract
Immune therapies have transformed the cancer therapeutic landscape but fail to benefit most patients. To elucidate the underlying mechanisms by which T cells mediate elimination of leukemia, we generated a high-resolution map of longitudinal T cell dynamics within the same tumor microenvironment (TME) during response or resistance to donor lymphocyte infusion (DLI), a widely used immunotherapy for relapsed leukemia. We analyzed 87,939 bone marrow-derived single T cell transcriptomes, along with chromatin accessibility and single T cell receptor clonality profiles, by developing novel machine learning tools for integrating longitudinal and multimodal data. We found that pre-treatment enrichment and post-treatment rapid, durable expansion of ‘terminal’ (TEX) and ‘precursor’ (TPEX) exhausted subsets, respectively, defined DLI response. A contrasting, heterogeneous pattern of T cell dysfunction marked DLI resistance. Unexpectedly, TPEX cells that expanded in responders did not arise from the infusion product but instead from both pre-existing and novel clonotypes recruited to the TME. Our unbiased dissection of the TME using a Bayesian method, Symphony, defined the T cell circuitry underlying effective human anti-leukemic immune responses that may be broadly relevant to other exhaustion antagonists across cancers. Finally, we provide a general analysis paradigm for exploiting temporal single-cell genomic profiling for deep understanding of therapeutic scenarios beyond oncology.
Publisher
Cold Spring Harbor Laboratory