Atlas-Driven Identification of Key Transcription Factors for Programming Protective T Cell States

Cellular immunity, Gene regulatory networks, Genomic engineering

CD8 T cells, crucial for fighting cancer and chronic infections, can adopt diverse functional states. Terminally exhausted (TEX) cells are dysfunctional, while tissue-resident memory (TRM) cells are protective. Despite their opposing roles, these states share remarkably similar gene expression profiles, complicating efforts to target one without affecting the other. To systematically identify the transcription factors (TFs) that selectively govern these cell fates, we built a comprehensive multi-omics atlas integrating transcriptional and epigenetic data across nine distinct CD8 T cell states. Using our analytical platform, we inferred TF activity profiles and catalogued state-specific regulatory fingerprints.

This atlas enabled a comparative analysis of the transcriptionally similar TEX and TRM states. Global TF community analysis revealed distinct biological pathways underlying their functional divergence. We then performed an in vivo CRISPR screen integrated with single-cell RNA sequencing (Perturb-seq) to validate TF roles. This identified several TFs, including ZSCAN20 and JDP2, that selectively govern TEX cell differentiation, while HIC1 and GFI1 were shared regulators of both TEX and TRM states, and KLF6 was unique to TRM cells.

Targeted deletion of the TEX-selective TFs enhanced tumour control in mice and synergized with immune checkpoint blockade, without interfering with the formation of protective TRM cells. Consistently, depleting these TFs in human T cells reduced expression of inhibitory receptors and improved effector function. By decoupling the genetic programmes for exhaustion from those for protective memory, our platform enables more precise engineering of T cell states, accelerating the design of improved cellular immunotherapies.

Within the immune system, CD8 T cells can differentiate into a spectrum of states, acquiring different functions and migration patterns in response to environments like tumours or viral infections. Transcription factors are the master regulators of this differentiation. Understanding how TFs shape these states is therefore essential for programming T cells with therapeutic potential, such as for adoptive cell transfer therapies. A major challenge is identifying the specific TFs that control distinct states, given the substantial heterogeneity and overlapping gene expression profiles even between functionally divergent cell types.

We focused on two critical and transcriptionally similar states: the protective tissue-resident memory (TRM) cell and the dysfunctional terminally exhausted (TEX) cell. In solid tumours, the presence of T cells with TRM characteristics correlates with better patient survival. Conversely, during chronic antigen exposure in cancer or infections like HIV, T cells progressively lose function and enter an exhausted state (TEX). These cells express high levels of inhibitory receptors, lack proliferative and effector capacity, and respond poorly to therapies like anti-PD1 immune checkpoint blockade.

Despite their opposite impacts on disease outcomes, TEX and TRM cells both reside in tissues and share striking transcriptional similarities, including expression of key TFs like BLIMP1, BHLHE40, and NR4A2. They even exhibit highly correlated patterns of open chromatin regions. This overlap complicates efforts to find TFs whose disruption could selectively inhibit harmful TEX development while preserving beneficial TRM formation. We hypothesized that a systematic, atlas-based comparison of TF activity—rather than just expression—across the entire differentiation landscape could reveal these key selective regulators.

Our goal was to create a comprehensive catalogue of TF influence across diverse CD8 T cell states. We developed an analytical pipeline, Taiji, which integrates matched RNA sequencing (RNA-seq) and assay for transposase-accessible chromatin sequencing (ATAC-seq) data to construct a gene regulatory network (GRN). In this network, interactions between TFs and their target genes are weighted based on predicted binding affinity, chromatin accessibility, and expression levels.

To quantify the global influence of each TF within the network, Taiji applies a personalized PageRank algorithm. This assigns an "importance" score to each node (TF) based on the quantity and quality of its connections, providing a measure of TF activity that accounts for upstream regulators, downstream targets, and feedback loops.

We applied Taiji to analyse 121 samples spanning nine well-defined CD8 T cell states from models of acute and chronic viral infection. These included memory precursor, effector, central memory, and TRM states from acute infection, as well as progenitor, effector-like, and terminally exhausted states from chronic infection.

Through comparative statistical analysis, we catalogued TF activity fingerprints. We identified 136 "single-state" TFs, each predominantly active in one specific cell state. This group included novel TFs such as HOXA7 in naive cells and ZSCAN20 and JDP2 in TEX cells. Conversely, 173 "multi-state" TFs, like TCF7 and TBX21, were key regulators in more than one state. As predicted from their transcriptional similarity, TEX and TRM cells shared the highest number of commonly active TF genes.

Our analysis predicted TFs with selective activity in either TEX or TRM cells, which could be leveraged to engineer T cells away from exhaustion and toward functional states without harming TRM formation. We identified 20 single-state TFs for TRM cells (e.g., FOSB, KLF6) and 34 for TEX cells (e.g., ZSCAN20, JDP2, IRF8). We also found 30 multi-state TFs active in both, including well-known factors like PRDM1 and novel ones like HIC1 and GFI1.

We also identified dynamic "TF waves"—coordinated groups of TFs whose activity patterns correlate with specific differentiation trajectories. For example, a TRM-associated wave included AP-1 family members linked to TGFβ response pathways, while a TEX-associated wave involved factors correlating with PD-1 and senescence pathways.

To understand the transcriptional programmes governing each state, we constructed TF-TF association networks based on their shared target genes. This revealed distinct patterns of collaboration. Single-state TFs formed dense, state-specific networks. For instance, TEX-selective TFs like ZSCAN20 and JDP2 showed strong connectivity within TEX regulatory networks.

Multi-state TFs, such as HIC1 and PRDM1, formed different partnership networks in TEX versus TRM cells, reflecting context-specific regulatory architectures. Clustering these associations identified distinct TF communities within each state's network. Pathway analysis of these communities highlighted their functional divergence: TRM communities were enriched for pathways involving cell adhesion and TGFβ response, while TEX communities were linked to apoptosis, catabolism, and proteolysis.

One pathway that emerged as a prominent and previously unrecognized feature of TEX cells was proteasome activity. Proteasome gene signatures were enriched in TEX-like cells from human non-small cell lung cancer patients and mouse tumour models. Functional assays confirmed that proteasome activity was highest in TEX cells from chronic infection and tumour-specific T cells. Importantly, adoptively transferring T cells with high proteasome activity resulted in worse tumour control compared to transferring cells with low activity, validating this pathway as a functional hallmark of TEX dysfunction.

To functionally validate the TFs predicted by our Taiji platform, we performed in vivo Perturb-seq. This combines CRISPR-mediated gene knockout with single-cell RNA sequencing in living animals. We screened a library targeting 19 TF genes, including 12 predicted TEX-selective TFs (like Zscan20 and Jdp2) and 7 multi-state TFs (like Hic1 and Gfi1).

The screen was conducted in two animal models that induce TEX or TRM differentiation. The results confirmed the accuracy of our computational predictions. Knockout of TEX-selective TFs, including the novel factors Zscan20 and Jdp2, specifically reduced the generation of TEX cells without affecting the formation of TRM or other effector/memory populations. Conversely, knockout of the TRM-selective TF Klf6 uniquely impaired TRM development. Knockout of shared multi-state regulators like Hic1 and Gfi1 impacted both cell states.

We next investigated the therapeutic potential of deleting the newly identified TEX-selective TFs, Zscan20 and Jdp2. In mouse tumour models, CD8 T cells lacking these factors showed enhanced control of tumour growth. Furthermore, combining the deletion of these TFs with anti-PD1 immune checkpoint blockade resulted in a synergistic improvement in tumour clearance.

Critically, this enhanced anti-tumour activity did not come at the expense of protective immunity. The formation of TRM cells in tissues and their recall capacity during reinfection were fully preserved in the absence of Zscan20 or Jdp2. This confirms that these TFs selectively enforce the exhaustion programme.

The translational relevance was demonstrated in human T cells. Depletion of ZSCAN20 or JDP2 in human CD8 T cells reduced the expression of inhibitory receptors like PD-1 and TIM-3 and improved effector functions, such as cytokine production, upon stimulation.

By constructing a multi-omics atlas of CD8 T cell states, we have systematically mapped the landscape of transcription factor activity, moving beyond simple gene expression. This approach successfully decoupled the regulatory networks of transcriptionally similar but functionally opposed T cell states. We identified novel TEX-selective TFs whose inhibition can skew T cell differentiation away from dysfunction and toward protection, enhancing anti-tumour immunity without compromising tissue-resident memory.

This platform provides a powerful, rationally guided framework for the precise engineering of T cell states. It accelerates the discovery of therapeutic targets for reprogramming T cells in cancer and chronic infection, paving the way for the design of more effective and durable cellular immunotherapies.