One cell, many fates: transcriptional networks governing CD8 + T cell differentiation

A recent study in Nature shows that CD8+ T cell differentiation trajectories are governed by transcription factors, with distinct single-state and multi-state regulators directing cell fate decisions. Activation of naïve CD8+ T cells by cognate antigen initiates their differentiation into a spectrum of functionally distinct and specialized states, ranging from memory cells with self-renewal capacity to cytotoxic effector cells that mediate immediate clearance of pathogens and tumors. Acute virus infection models showed that the large pool of cytotoxic effector cells generated during the primary response contracts into a small memory population with enhanced self-renewal capacity, forming the basis of the classical linear model of CD8+ T cell differentiation which proposed that naïve T cells first develop into effector subsets that later give rise to long-lived memory cells.1, 2 However, later lineage-tracing and single-cell profiling studies3, 4 have challenged this unidirectional view, suggesting that memory and effector cells can emerge in parallel. Despite these advances, the regulatory mechanisms that govern these differentiation states remain incompletely understood. Now, Chung et al. compiled a comprehensive catalogue of transcription factor (TF) activity across diverse cytotoxic T cell states to define the regulatory circuits underlying these states.5 A PageRank algorithm was used to assess the TF activity across cell states, identifying both single-state TFs that selectively influence a specific state and multi-state TFs that are active in more than one differentiation state. This provides a framework for understanding how CD8+ T cell differentiation states are regulated by specific transcription factors and their potential interactions with target genes. Conventional CD8+ T cell programs that generate memory and effector T cell states can be altered by chronic antigenic stimulation such as those encountered during chronic virus infections or within tumor microenvironments. Sustained T cell receptor signaling in these settings induces prolonged expression of inhibitory receptors such as programmed cell death-1 (PD-1) that activate regulatory programs to restrict T cell activation and limit immunopathology.6 This CD8+ T cell differentiation state, commonly referred to as CD8+ T cell exhaustion, is characterized by poor effector function, impaired cytokine production and can be viewed as an adaptive strategy to protect host tissues from excessive immunopathology rather than a simple loss of immune competence. In contrast to earlier studies that mainly focused on either conventional effector memory differentiation or exhausted T cell states, Chung et al. integrated transcriptional datasets spanning a broad spectrum of CD8+ T cell states. These included central memory (TCM), tissue-resident memory (TRM), memory precursor (MP), terminal effector (TE) and effector memory (TEM) CD8+ T cells that arise during acute LCMV infection, as well as progenitors of exhaustion (TEXprog), effector-like exhaustion (TEXeff) and terminally exhausted (TEXterm) CD8+ T cell states found in chronically LCMV infected mice, enabling a unified analysis of transcriptional regulation across divergent differentiation trajectories.5 Transcription factor activity analysis identified single-state-specific TFs, active in a single differentiation state and multi-state TFs, which are active across two or more states. This included Zeb1, Bach1 and Irf7, which were active in naïve, TCM and TEXprog cells, respectively, while several TFs such as TCF1 and T-BET exhibited activity across multiple cell states. A group of TFs, including BATF, IRF8 and NFATc1, were enriched in exhaustion-related cell states as compared to conventional states, consistent with their role in promoting CD8+ T cell exhaustion.7, 8 However, it is important to note that in the absence of chronic antigenic stimulation, these TFs can also promote T cell memory and effector differentiation in a context-dependent manner.8, 9 Comparative TF activity analysis revealed that TRM and TEXterm cell states exhibited a relatively similar transcriptional profile and shared multiple TFs, suggesting overlapping regulatory circuits in tissue residency and terminal exhaustion programs (Figure 1). These multi-state TFs included previously characterized factors such as Nr4a2 and Bhlhe40, as well as novel transcription factors identified in this study, including Hic1 and Gfi1. Hic1 has been established as a critical regulator of TRM CD8+ T cell development in the small intestine,10 however, it has not been linked to T cell exhaustion. In the current study, Perturb-seq that couples in vivo CRISPR-mediated perturbation with single-cell RNA-seq was used to functionally validate novel transcription factors. Genetic ablation of Hic1 during chronic virus infection led to a reduction in TEXterm CD8+ T cell frequency while the TEXprog population increased. A similar phenotype was observed for GFI1, where the TEXprog population expanded and terminally exhausted cells showed reduced frequency, identifying it as a positive regulator of CD8+ T cell exhaustion. In contrast, recently we demonstrated that GFI1 can promote CD8+ T cell stemness by enhancing their proliferative capacity, thereby sustaining the long-lived persistence of memory CD8+ T cell populations.11 GFI1 was selectively expressed in memory populations, including TCM and TEXprog, consistent with its role in promoting memory T cell longevity. These divergent outcomes underscore a potential limitation of the in silico approach which indicated that GFI1 was not active in memory populations, such as TCM or MP. Thus, computational TF activity predictions may not fully capture context-specific regulatory roles. Beyond shared transcriptional programs and multi-state TFs, the study by Chung et al. also uncovered novel transcription factors such as Jdp2, Zfp324 and Zscan20 that exhibited specific activity in TEXterm cells but not TRM.5 These findings were functionally validated using Perturb-seq in two independent infection models, one analyzing exhausted CD8+ T cell populations in the spleen of chronically lymphocytic choriomeningitis virus (LCMV) infected mice and the other examining resident memory populations isolated from the small intestine following acute LCMV infection. This revealed that Hic1 and Gfi1 regulated both the development of TEXterm and TRM CD8+ T cells, confirming them as multi-state TFs. In contrast, loss of Jdp2, Zfp324 and Zscan20 selectively impaired TEXterm formation without affecting TRM development, identifying these TFs as both newly identified and TEXterm-specific that could be targeted to modulate terminal exhaustion. These TEXterm-specific TFs were validated as key targets whose disruption improved anti-tumor responses. Specifically, genetic ablation of Zfp324 and Zscan20 in tumor-specific CD8+ T cells improved tumor control, while loss of Hic1, a multi-state TF, did not impair tumor control. Moreover, depletion of Zfp324 and Zscan20 synergized with anti-PD-1 therapy, further reducing tumor size and enhancing survival, highlighting their potential to improve the efficacy of immune checkpoint blockade. These findings underscore a critical translational opportunity, as selective targeting of exhaustion-specific TFs can prevent CD8+ T cell dysfunction while preserving effector and memory differentiation, offering a rational framework for engineering more effective T cells for immunotherapy. Collectively, this study provides a valuable framework for understanding the transcriptional regulation of CD8+ T cell differentiation and identifies shared and state-specific transcription factors as promising targets to enhance adoptive T cell therapy. At the same time, it highlights a key limitation of this approach, as computational TF activity predictions may not fully capture context-specific roles, particularly in memory populations or tissue-residenT cells. Future work will need to dissect the complexity of TF networks and understand context-dependent mechanistic regulation to fully harness these insights for therapeutic applications. Conceptualization: M.Z.C. and G.T.B. Writing – original draft: M.Z.C. and G.T.B. The authors declare no conflicts of interest. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.