Latent Plasticity of Effector-like Exhausted CD8 T cells contributes to memory responses

Abstract

Persistent antigen induces a dysfunctional CD8 T cell state known as T cell "exhaustion" characterized by expression of PD-1 and decreased effector functions 1-3 . Nevertheless, dysfunctional CD8 T cells can mediate control of antigen burden which is long-lasting. While significant heterogeneity of exhausted CD8 T cells has been described, the cells which actively proliferate and exert direct viral control have remained elusive. Here, we define subsets of PD-1 + CD8 T cells marked by expression of CX3CR1 exhibit substantial in situ proliferation and expression of granzyme B during chronic infection for the maintenance of the effector pool through self-renewal independently of previously defined stem-like cells. Unexpectedly, the CX3CR1 + CD8 T cells retain plasticity to be reprogrammed to memory cells through expression of TCF-1 and re-gain polyfunctionality. Thus, we define a subset of effector-like exhausted CD8 T cells with capacity to contribute to the memory pool, offering a prime target for novel immunotherapies.

Fate trajectories of CD8+T cells in chronic LCMV infection

“…All these studies consistently described the memory-like and the terminally exhausted cell state, which we also identified. Some studies [5,18] additionally described an effector-like CX3CR1 + population arising late during the infection, which we also found in our late samples from 14 & 21 dpi.…”

“…Previously published work used scRNAseq to study CD8 T cell differentiation during chronic infection, isolating virus specific T cells at different time-points [4,5,10,18]. They made use of dimensionality reduction and computational tools for trajectory inference.…”

Cytopath: Simulation-based inference of differentiation trajectories from RNA velocity fields

“…We identified correct differentiation branches of CD8 T cells in chronic infection, demonstrated by correct ordering of the experimental time labels and expression levels of branch specific gene expression markers. For this system several phenotypic populations and characteristic markers had been described before, but the connecting differentiation trajectories of those populations are a subject of ongoing research [Chen et al, 2019, Zander et al, 2019, Yao et al, 2019, Raju et al, 2020]. These studies provide evidence for branching in the development process, and only recently in conjunction with simulation-based trajectory inference it was possible to resolve this event in more detail [Cerletti et al, 2020].…”

“…We show that Cytopath successfully recovers known differentiation processes in all datasets with high accuracy. Furthermore, we compare the performance of Cytopath to the best trajectory inference models for each topology [Saelens et al, 2019], which are Slingshot [Street et al, 2018] for tree-like topology, Angle [Saelens et al, 2019] for cell cycle, PAGA (directionality enabled by velocity pseudotime) [Wolf et al, 2019, Bergen et al, 2020 for graph models. We also include comparison to Monocle3 [Cao et al, 2019] as well as two approaches accounting for RNA velocity information, Ve-Tra [Weng et al, 2021] and Cellpath .…”

“…The interpretation of pseudotime ordering for branched processes is particularly challenging and requires an additional trajectory inference step. A comprehensive review by [Saelens et al, 2019] on the performance of various methods on multiple topologies suggests, Slingshot [Street et al, 2018] has the best general performance on datasets with multiple branching lineages, and Monocle3 [Trapnell et al, 2014], as a popular runner-up for pseudotime inference that does not produce explicit trajectories per lineage.…”