Leukocyte cones from unidentified healthy donors were purchased from National Health Service Blood and Transplant (NHSBT). Human peripheral blood mononuclear cells (PBMCs) were isolated using SepMate™ Isolation Tubes (StemCell Technologies, #85450). The blood from leukocyte cones was transferred to Sepmate™ isolation tubes containing Histopaque 1077 (Sigma #10771) before being centrifuged at 1200 x g for 10 minutes at room temperature with the brake on. The top layer of cells enriched for PBMCs was transferred into a separate Falcon tube and washed twice with 1X Phosphate Buffered Solution (PBS; Sigma #D8537). The PBMCs were then resuspended in SepMate™ Cell Separation Buffer (StemCell Technologies #20144) for subsequent T cell isolation.

Primary human CD4⁺ and CD8⁺ T cells were isolated using EasySep™ Human T cell isolation kits (StemCell Technologies CD4; #17952 or CD8; #17953) according to the manufacturer’s instructions using EasySep™ magnets (Stem Cell Technologies, #18001).

Primary samples were purchased, tracked, and stored according to Human Tissue Act regulations (IRAS; 206950 for storage < 1 month, or 246390 for > 1 month).

T cells were cryopreserved in heat-inactivated fetal bovine serum (FBS; Thermo Fisher, #10270106) supplemented with 10% Dimethyl sulfoxide (DMSO; Sigma, #D2650-5X5ML) at 20x10⁶ cells/vial and stored in liquid nitrogen.

When required, cells were thawed rapidly at 37°C before resuspension and continued culture in complete T cell media; Roswell Park Memorial Institute (RPMI)-1640 (Gibco Thermo Fischer Scientific, 21875091) media supplemented with 10% Heat Inactivated (HI) FBS and 50 IU/mL rhIL-2 (R&D Systems, #202-IL-050). Throughout all experimental conditions, 50IU/mL rhIL-2 was used and primary cells were maintained at 1x10⁶ cells/mL.

A375 cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM; Sigma Aldrich, #D6429) supplemented in 10% FBS. Cells were routinely monitored for mycoplasma.

All cells were grown in 37 °C humidified incubators with 5% CO₂.

T cells were rested overnight in complete media after thawing. For stimulation, cells were resuspended in activation media; complete T cell media supplemented with 15 μL/mL ImmunoCult™ Human aCD3/aCD28 T Cell Activator (Stem Cell Technologies, #10971) for 48 hours. After this initial stimulation, single stimulated controls (Ts) were maintained in complete media throughout the protocol. On the other hand, exhausted T cells (Tex) were generated by repetitive stimulation (in the same method as described above) every 48–72 hours for a total of six stimulations. At each stimulation, cell density adjustments and media replenishment were performed in parallel with complete medium for the Ts cells.

1x10⁶ cells per condition were electroporated with ribonucleoprotein (RNP)-Cas9 complexes using a 4D Nucleofector™ 96 well unit (Lonza) in P3 Primary Cell Buffer (Lonza, #V4SP-3096) and the EH-115 pre-set pulse program. Guide RNA sequences are provided in Supplementary Table S1.

For blocking experiments, nucleofection was performed on non-activated T cells while for reversal experiments nucleofection was performed in exhausted T cells.

For endpoint Immunocult™ stimulation, 1x10⁵ cells from Ts, Tex controls and KO cultures were seeded in a round bottom 96-well plate in 100 μL/well activation medium and incubated for 48 hours. Supernatants were then harvested for cytokine analysis and stored at - 80°C until required.

In the blocking CRISPR format, endpoint stimulation was performed 48 hours after the last (sixth) stimulation. In the reversal CRISPR format, endpoint stimulation was set-up 8 days post-nucleofection during which the cells were rested in complete media.

A375 cells were seeded in 96-well flat bottom plates at 5x10³ cells/well and placed in an IncuCyte S3 or SX5 Live cell analysis incubator (Essen Bioscience) overnight with scans performed every 3–4 hours. The following day, the A375 media was replaced with complete T cell media supplemented with 2X Immunocult™ activator (CD8⁺ T cells: 7.5 µL/mL, CD4⁺ T cells: 30 µL/mL). Tex and Ts cells resuspended in complete T cell media were then added at a target:effector (T:E) ratio of 1:5. Plates were returned to the IncuCyte imager for 48 hours. At the end of the co-culture period, T cells were removed and replaced with 100 μL/well 1X PBS, before cancer cell viability was determined using CellTiterGlo^® according to the manufacturer’s instructions.

Pelleted cells were stored at -80°C until required. Protein lysates were extracted using lysis buffer supplemented with phosphatase and protease inhibitors for 15 minutes on ice. Cells were then centrifuged at 14000 rpm at 4°C for 20 minutes. Protein concentrations were quantified using Pierce BCA Protein Assay (ThermoFisher, #23227). Western blot was performed as previously described (64). Blots were imaged using an Odyssey^® M Imaging system (LI-COR Biosciences) and Empiria Studio^® Software was used for densitometry analysis. Combined Linear Range testing was performed to enable quantitative densitometry analysis. Unless stated otherwise, relative protein expression was quantified relative to total protein staining.

All antibodies used for western blot are provided in Supplementary Table S2.

A minimum of 5x10⁴ cells per condition were seeded in a v-bottom polypropylene 96-well plate (Greiner Bio-One, #651201), washed with 1X PBS, and stained with Zombie NIR Fixable Viability Kit (BioLegend, #423106), at room temperature for 20 minutes protected from light. Subsequently, the cells were resuspended in human Fc block (BD; #564220) for 10 minutes at 4°C. Antibodies (Supplementary Table S3) were prepared at a 2X concentration in eBioscience™ Flow Cytometry Staining Buffer (FACS buffer; Thermo Fisher, #00-4222-26) and added to the cell directly after blocking, for 20 minutes at room temperature protected from light. After staining, the cells were washed in FACS buffer and fixed with Cytofix™ Fixation Buffer (BD; #554655) for 15 minutes at 4 °C. Finally, the cells were re-suspended in 1X PBS for acquisition using a BD FACSLyric™ Clinical Cell Analyzer or Attune NxT flow cytometer and analyzed using FlowJo v10.8.1 software.

1-2x10⁵ cells per condition were seeded and washed as above and stained with Zombie Green Fixable Viability Kit (BioLegend, #423111) at room temperature for 20 minutes protected from light. Cells were washed once in FACS buffer and then in 1X PBS. MitoTracker™ Deep Red dye (Invitrogen, #M46753) was diluted 1:1000 in PBS and added directly to the cells for 20 minutes at 37 °C protected from light. The cells were washed twice in PBS. Fixation, acquisition and analysis were performed as described above.

2x10⁵ cells per sample were seeded, washed and stained with viability as above. Surface staining was carried out with antibodies (Supplementary Table 4) prepared in BD Horizon™ Brilliant Stain Buffer Plus (BD, #566385) and FACS buffer, and cells were incubated in antibody mix for 30 minutes at 4 °C. Following this, samples were washed in FACS buffer and resuspended in Fixation/Permeabilization buffer from eBioscience™ Foxp3/Transcription Factor Staining Buffer Set (Invitrogen, #00-5523) for 1 hour at 4 °C. Cells were washed 1x in Permeabilization buffer and stored in PBS at 4 °C. On the day of acquisition, antibodies for intracellular stain were prepared in Brilliant Stain Buffer Plus and Permeabilization buffer, and cells were incubated with antibody mix for 1 hour at 4 °C. Samples were then resuspended in 150µl PBS. Samples were acquired using the Sony ID7000^™ and unmixing was performed using the instrument-associated software. Subsequent analysis was carried out using FlowJo v10.10.0 software. Data was quality controlled and cleaned using built-in FlowClean function. Samples were gated for live, single cells, FOXP3^- CD56^- population (gating strategy shown in Supplementary Figure S3A), and downsampled to 10,000 cells per sample. This population was used for manual gating and Uniform manifold approximation and projection (UMAP) analysis. Dimensional reduction was performed using UMAP FlowJo plugin v4.1.1 (65) and clustering performed using FlowSOM v4.1.0 with default settings and number of meta-clusters set to 10 (66). Built-in plugin Cluster Explorer was used to visualize marker expression and cluster characterization.

Data was generated using the GEM-X Flex Gene Expression workflow. Samples were fixed at 4 °C for 16 hours according to 10x Genomics guidelines (CG000782, Rev C). Gene expression was measured using barcoded probe pairs designed to hybridize to mRNA specifically. Samples were hybridized to barcoded human transcriptome probes and pooled. Using a microfluidic chip, single cell suspension pool of ~232,000 cells was partitioned into nanoliter-scale Gel Beads-in-emulsion (GEMs). A pool of ~737,000 10x GEM Barcodes was sampled separately to index the contents of each partition. Inside the GEMs, probes were ligated and the 10x GEM Barcode was added, and all ligated probes within a GEM share a common 10x GEM Barcode. Barcoded and ligated probes were then pre-amplified in bulk, after which gene expression libraries were generated (User Guide: CG000787 GEM-X Flex Gene Expression Reagent Kits - Multiplex) and sequenced (NovaSeq X. Sequencing read configuration: 28-10-10-90).

Sequencing data was processed using Cell Ranger software (cellranger-9.0.0; cellranger-multi pipeline) to align reads to the Chromium Human Transcriptome Probe Set v1.0 and demultiplex samples. Seurat package functions were used to convert Cell Ranger output count matrices into individual Seurat objects. Quality control metrics were calculated to generate filtered Seurat objects of cells with between 500 and 5000 detected features (nFeature_RNA) and with less than 15% mitochondrial gene expression. Following this, the filtered Seurat objects were merged into a single combined object and normalized using default methods. Subsequently the top 3,000 highly variable genes were identified using the variance stabilizing transformation (VST) method (Seurat:: FindVariableFeatures) and the expression values for each gene scaled. Principal component analysis (PCA) was performed using the Seurat::RunPCA() function, restricted to the previously identified variable features and computing the top 30 principal components (67). To account for batch effects between donors, harmonization was applied using the harmony package (68). Uniform Manifold Approximation and Projection (UMAP) was computed using the first 10 Harmony dimensions.

The filtered and harmonized dataset was aligned to a reference atlas of human CD8⁺ TILs (version 1) using ProjecTILs and scGATE packages (69, 70). During this process of projection, log-normalization and batch correction were performed, and the query was mapped into the reference PCA and UMAP space to annotate CD8⁺ subpopulations. All cells not confidently annotated by the reference were excluded from the resulting annotated dataset. Differential gene expression analysis was performed on subsets of the integrated dataset to identify marker genes distinguishing two experimental conditions. The Seurat::FindMarkers()` function was used with minimum expression threshold of 0.5 and a log-fold change threshold of 0.25. The results were visualized with volcano plots highlighting significant genes (adjusted p-value < 0.05 and fold change > 2) using the EnhancedVolcano package (71).

A supervised pseudotime approach was employed using the psupertime package (72) to incorporate a time series based on the progression of samples collected over the course of the exhaustion protocol. For input, the top 2000 variable features in the whole Seurat dataset were calculated using VST. A further custom list of 31 genes identified from the literature as relevant to T cell states was added to this to generate psupertime models for Tex and Ts groups of samples using the following label ordering: S4, S6, S6R, S6 stim. A subset of the dataset, restricted to relevant features for supervised trajectory inference, was created and converted into a SingleCellExperiment object, which was then split into sample groups. The Tex model was trained using a dataset comprising 37,250 cells. Following feature selection,1,406 genes were taken forward for training this model. Of these, 990 genes (70%) were identified as relevant to the trajectory. The model achieved a mean accuracy of 85%. The Ts model was trained using a dataset comprising 34,645 cells. Following feature selection, 1,327 genes were taken forward for training this model. Of these, 970 genes (73%) were identified as relevant to the trajectory. The model achieved a mean accuracy of 83%. Gene level plots were generated to visualize the expression dynamics over time for both trajectories. Dotplots capturing gene expression differences between sample conditions were generated using SCpubr::do_DotPlot() (73).

Data files were submitted to BioStudies (EMPL-EBI) with link: https://www.ebi.ac.uk/biostudies/ArrayExpress/studies/E-MTAB-15435?key=7414ac65-199d-4d09-89d2-b659bb32606e and the accession number of E-MTAB-15435. Any additional information required to reanalyze the data reported in this paper is available upon request.

Cytokine concentrations in supernatants following endpoint stimulation or cancer cell line co-culture were determined using Meso Scale Discovery (MSD^®) multiplex human U-plex CAR-T Combo immunoassay kit (Meso Scale Discovery, #K15338K-4), or U-plex TGF-β Combo Human (#K15241K-4) according to the manufacturer’s instructions. Plates were read on Meso QuickPlex SQ 120MM instrument. Data analysis was conducted on MSD Discovery Workbench software.

Statistical significance was determined by the tests outlined in figure legends, where n describes number of biological replicates, unless otherwise stated, and were performed using GraphPad Prism v 10.1.2.

In this study, we developed an in vitro T cell exhaustion protocol, utilizing chronic stimulation with soluble anti-CD3/anti-CD28 Immunocult™ to generate human exhausted CD8⁺ and CD4⁺ T cells (Tex) (Figure 1A). To provide a functional single stimulated T cell control (Ts), following an initial stimulation, cells were rested in complete T cell media during the chronic stimulation period (Figure 1A). Over the course of the chronic stimulations, both CD8⁺ (n=5) and CD4⁺ (n=6) T cells expanded well, showing a median cumulative fold expansion of 33.6-fold (95%CI 23.8-135.1) and 31.9-fold (95%CI 9.8-134.1) respectively (Supplementary Figure S1A). Tex cells for both compartments also maintained a high level of viability (81% CD8⁺, 95% CD4⁺Supplementary Figure S1B).

To characterize our in vitro generated exhausted T cells, Tex and Ts populations were assessed for expression of their inhibitory receptors PD-1 and Tim-3. At the end of the chronic stimulation protocol, we observed significant increases in the PD-1⁺Tim-3⁺ subpopulation in exhausted CD8⁺ and CD4⁺ T cells compared to matched functional cells (Figure 1B). The increase in %PD-1⁺Tim-3⁺ subpopulation was robust across donors increased by 3.36-fold in CD8⁺ T cells (n=13) and 2.75-fold in CD4⁺ T cells across donors (n=8; Figure 1B). Moreover, extended characterization of CD8⁺ (n=2) and CD4⁺ (n=2) T cells throughout our in vitro exhaustion protocol showed reducing expression of the activation markers 4-1BB, CD25, CD69, HLA-DR, and ICOS in Tex cells over the course of the repeated stimulation protocol (Supplementary Figure S3B). In addition to immunophenotypic analysis, both CD8⁺ and CD4⁺ Tex cells showed signs of reduced mitochondrial mass indicative of impaired mitochondrial metabolism, as demonstrated by reduced MitoTracker™ staining, compared to Ts controls despite more recent stimulation (Supplementary Figure S1C).

Immunophenotypic and functional responses to re-stimulation with Immunocult™ was also assessed. This re-stimulation was performed on Tex and Ts cultures in parallel (referred to as endpoint stimulation in Figure 1A). In response to endpoint stimulation, the proportion of cells co-expressing PD-1 and Tim-3 in Tex cells was maintained for both CD8⁺ and CD4⁺ T cell compartments, reaching the same levels as Ts, which expectedly upregulate PD-1 and Tim-3 expression upon activation (Supplementary Figures S1E, F). Beyond these markers, we found the co-stimulatory molecule and surrogate marker of T cell activation, 4-1BB, was significantly reduced in Tex cells (Figure 1C). Across multiple donors, the proportion of 4-1BB⁺ cells were 3.74-fold higher in CD8⁺ Ts and 2.47-fold in CD4⁺ Ts cells compared to matched Tex cultures. Furthermore, both CD8⁺ and CD4⁺ Tex cells showed significantly reduced IFN–γ, TNF-α, and IL-2 secretion compared to Ts cells (Figure 1C). Together these phenotypic and functional markers confirm impaired responses of our chronically stimulated T cells that correspond to an exhausted phenotype.

To expand on the characterization of the dysfunctional T cell states emerging from our in vitro T cell exhaustion, samples were collected at different stages along the chronic and endpoint stimulation protocol for Tex and Ts CD8⁺ cultures, and transcriptomic profiles were compared by scRNA-seq (Figure 2A). Overlays of the data obtained from 2 donors highlighted very good consistency, indicating reproducibility of the in vitro exhaustion (Supplementary Figures S2A, B). Expectedly, Tex and Ts samples transcriptionally show some overlaps, albeit distinct subsets observed (Supplementary Figure S2C). We used ProjecTILs to project our merged scRNA-seq dataset onto a reference CD8⁺ human TILs atlas (69, 70, 74) that was assembled from 20 different samples spanning 7 cancer type studies and has been previously described to capture CD8⁺ exhausted (“TEX”) and precursor exhausted (“TPEX”) subtypes. This allowed us to compare our in vitro cultured T cells to TILs and to annotate the exhausted T cell subpopulations within our samples (Supplementary Figure S2D). Over 75% of the single cell dataset could be effectively mapped onto T cell transcriptional profiles within the CD8⁺ human TILs atlas, indicating the physiological relevance of our in vitro model of T cell exhaustion (Supplementary Table S5). Based on the atlas annotations, we observed that a larger proportion of our Tex samples were classified as exhausted compared to the Ts samples (64% vs 49% TEX) at the end of the in vitro exhaustion protocol. This higher frequency of exhausted T cells in the Tex samples further increased after endpoint stimulation (86% vs 40% TEX) (Figure 2B). Notably, central memory and particularly effector memory T cell populations were consistently more prevalent in the Ts samples compared to Tex samples, which further demonstrates the loss of functional T cell states that result from our in vitro chronic stimulation. Surprisingly, the Ts samples showed a moderate level of cells that were also annotated as TEX. We chose to query differentially expressed genes between the TEX-labelled subsets in Ts and Tex samples following endpoint stimulation. This analysis revealed that despite transcriptional overlaps with the TEX classification, IL2RA and numerous cytokines like IFNG, CCL3, CCL4 and CSF2 were amongst the most highly upregulated genes in TEX-annotated Ts samples suggesting that this subset within the Ts retain polyfunctionality in comparison to the Tex-TEX subset (Supplementary Figure S2E).

When examining the changes in key exhaustion and activation markers across the protocol samples and along pseudotime models, we observe higher gene expression and a greater proportion of cells expressing ENTPD1, TOX, HAVCR2 and SLAMF6 in Tex samples compared to Ts samples, with further induction following endpoint stimulation. On the contrary, expression of activation and co-stimulatory markers IL2RA (CD25), TNFRSF9 (4-1BB), TNFRSF4 (OX-40), and cytokines IFNG, TNF and GZMB were more highly expressed and further induced in Ts compared to Tex samples (Figure 2C, Supplementary Figure S2F). This is consistent with our earlier immunophenotyping and functional assessment and overall support the acquisition of a dysfunctional state through chronic stimulation.

To complement the scRNA-seq analysis and expand the characterization of CD4⁺ cells, high dimensional spectral flow cytometry was performed using equivalent samples (CD8⁺ n=2, CD4⁺ n=2). The proportion of Naïve, Central Memory (CM) Effector Memory (EM) and Terminally differentiated Effector Memory (TEMRA) T cell states within the Tex and Ts cultures, based on CCR7 and CD45RA cultures were assessed (Supplementary Figure S3C). Although CM cells made up the majority of Tex and Ts cultures for both CD8⁺ and CD4⁺ T cells, within the CD8⁺ compartment Tex cells had a smaller population of naïve and TEMRA cells compared to matched Ts samples at an intermediate point (stim 4), maintaining that difference in naïve population and at the end of exhaustion (stim 6). In response to endpoint stimulation, CD8⁺ Ts cells enlist an EM population not seen in Tex (15.60% in Ts compared to 1.15% in Tex) (Supplementary Figure S3D top panel). For CD4⁺ T cells, the cell state proportions between Tex and Ts cells appear similar at stim 4 but differences are clear by the end of exhaustion. Specifically, the proportion of EM reduces to from 3.46% at stim 4 to 0.4% in Tex cells, whilst this population increases in matched Ts (2.44% at stim 4 to 19.50% at stim 6; Supplementary Figure S3D bottom panel). This EM population is retained in CD4⁺ Ts cells upon endpoint stimulation (and rested conditions) but does not reappear in Tex. Whilst this dual-marker gating does not seek to provide the granularity of the transcriptional data, it does recapitulate the enrichment of an effector population in Ts, suggesting maintenance of dysfunction across the exhaustion protocol on the Tex populations, even upon resting.

Dimensional reduction and clustering of the spectral cytometry data were performed to further explore populations present in the spectral cytometry dataset. For CD8⁺ T cells, visualization of the samples on the UMAP demonstrates the change in distribution of the cells across the exhaustion protocol and most notably with endpoint stimulation (Supplementary Figure S4A). The largest cluster (cluster 4) represented 32% of cells and was shared across Tex and Ts samples (Supplementary Figure S4B). This could represent a TEMRA-like population with intermediate CD45RA expression alongside CD45RO⁺ CCR7⁺ expression but lacking expression of co-inhibitory markers (Supplementary Figure S4D). This may relate to the TPEX population shown in scRNA-seq analysis to be shared across samples. Tex evolution from stim 4 to stim 6 of the exhaustion process shows an increase in cluster 3 (3.71% in Tex stim 4, 36.6% in Tex stim 6) (Supplementary Figure S4C). Cluster 3 is characterized by CD39, PD-1 and intermediate Tim-3 expression with reduced activation marker expression (fewer markers and lower intensity) relative to other populations. This population also expresses CD28, HLA-DR and EOMES, a transcription factor which in addition to its role in T cell differentiation is also a contributor to terminal exhaustion (Supplementary Figure S4D). In parallel, there is a decrease from Tex stim 4 to stim 6 in cluster 7 (Supplementary Figure S4C), which co-expresses several co-inhibitory markers (PD-1, Tim-3, LAG3) and has a stronger expression of activation markers CD69, ICOS and CD25 (Supplementary Figure S4D).

Comparing CD8⁺ Tex and Ts samples at the end of exhaustion, Ts has a lower proportion of cluster 3 described above to have a dysfunctional phenotype and notably has a higher proportion of naïve-like cells (cluster 10; CD45RO^- CCR7⁺ CD45RA⁺ CD28^int) and a non-proliferative central memory population (cluster 5; CD45RO⁺ CCR7⁺ CD45RA^int) (Supplementary Figures S4C, D). Of interest, this naïve-like cluster additionally shows EOMES expression which could indicate transition toward central memory phenotype. On endpoint stimulation, Ts enlist an effector-like population (cluster 9) which shows EOMES expression and strong upregulation of both costimulatory and coinhibitory markers, reflective of activation in response to stimulation. Of note, CCR7 expression is still present here and therefore we cannot directly map this cluster to the manually gated EM population. In comparison, cluster 9 is less prominent in Tex following endpoint stimulation, where we instead identify clusters 1 and 6 which both have high CD39 expression and lower costimulatory marker expression than cluster 9. Cluster 1 also shows co-expression of PD-1, Tim3 and LAG3 (Supplementary Figures S4C, D).

For CD4⁺, distribution of cells on the UMAP demonstrates a clear shift in Tex across stages of the exhaustion protocol which is distinct from the pattern observed in Ts samples (Supplementary Figure S5A). The largest cluster represents 50% of cells included and is shared across Ts stim 4 and Tex/Ts stim 6 samples (Supplementary Figures S5B, C). This CD45RO⁺ CCR7⁺ CM-like cluster retains expression of CD28, with limited expression of costimulatory or coinhibitory receptors. As Tex cells progress from stim 4 to stim 6 of the exhaustion protocol, a CD39⁺ population expressing PD-1, LAG3 and intermediate levels of Tim3 is preserved (cluster 1). This population shows expression of fewer costimulatory markers than clusters that appear activated (9 and 10). Cluster 5, which is specific to Tex stim 4, shows a similar profile with lower expression of transcription factors TCF1 and EOMES (Supplementary Figure S5D). Compared to Tex stim 4, Tex stim 6 samples also show loss of cluster 8 and 9 which both show signs of activation with both increased costimulatory and coinhibitory markers. This indicates a reduction in capacity to enlist activation and a shift towards dysfunction across the exhaustion protocol in CD4⁺ cells.

In CD4⁺ Tex and Ts at the end of exhaustion (stim 6), the dysfunctional population described above (cluster 1) more strongly represented in the Tex sample. In parallel, Ts is characterized by naïve-like cluster 7 (CD45RO^-, CCR7⁺ CD45RA⁺ CD28⁺), and cluster 4 with similar CCR7/CD45RA expression, but loss of CD28 and EOMES upregulation indicative of transitioning to differentiation. Also present in Ts at end of exhaustion are clusters 3 (CM-like with low expression of costimulatory and inhibitory receptors) and 6, expressing PD-1 and a subset of costimulatory markers. On endpoint stimulation, Tex shows an increased proportion of the dysfunctional-phenotype cluster 1 seen at both stim 4 and stim 6. In comparison, Ts is made up of a higher proportion of clusters 3 and 4 (present at the end of exhaustion), as well as cluster 10 demonstrating an activated effector phenotype by upregulation of EOMES, strong costimulatory receptor expression and intermediate expression of inhibitory receptors. This cluster does however retain CCR7 and TCF1 expression. Cluster 9 is also present in this CD4+ post endpoint stimulation sample, with a similar phenotype to activated population in cluster 10 but with reduced proliferation. Overall, CD4⁺ Tex is characterized by a high CD39⁺ PD-1⁺ LAG3⁺ across stimulation and following endpoint stimulation, whilst Ts samples dynamically demonstrate the presence of a naive-like population and activation in response to endpoint stimulation (Supplementary Figures S5C, D).

To assess the feasibility of downstream assays and manipulation of these cells, we also used these methods to characterize Tex and Ts cells which were rested in IL-2 media for 8 days from the end of exhaustion (Figure 2A). It was observed in the scRNA-seq analysis that CD8⁺ Tex cells showed a mild reduction in exhaustion populations during the resting period (TEX 64% to 46%, TPEX 3% to <1%, Figure 2B), with relative expansion of the naïve-like population. Notably, after resting Tex samples still lack the EM population seen in Ts at the end of exhaustion and retained in Ts rested samples. Similar observations noted relative to EM population in rested Tex condition which remains smaller compared to matched rested Ts condition, via spectral flow cytometry (Supplementary Figure S3D). Pseudotime analysis of genes of interest also shows strong similarity in expression profile at end of exhaustion and in rested samples for both Tex and Ts (Figure 2C). For CD4⁺ cells, this comparison relied on flow cytometry data, where characterization based on CCR7/CD45RA expression showed a similar pattern: retention of an EM population in Ts during the resting period which represents only 3% of the parallel Tex sample (Supplementary Figure S3D). Further to this, distribution of the clusters defined in CD4⁺ samples by high dimensional flow cytometry analysis demonstrate that rested samples closely mirror the composition of the relevant Tex or Ts sample at end of exhaustion (Supplementary Figure S3D).

Finally, we assessed whether the reduced functionality of our in vitro generated Tex cells translated into impaired control of cancer cell line growth using antigen-agnostic co-cultures with the melanoma cell line A375 in the presence or absence of Immunocult™ activator (Supplementary Figure S8A). In the absence of Immunocult™ activator neither Tex nor Ts cells showed cancer cell line control (Supplementary Figure S8B). However, in the presence of Immunocult™, Tex cells showed ~20% less cancer cell line control than matched Ts cells (CD8+ n=3 donors; CD4+ n=2 donors) when assessed using A375 confluency or orthogonal CellTiterGlo^® readouts across multiple Immunocult™ concentrations (Supplementary Figure S8C).

Altogether these data confirm the successful induction of an exhaustion phenotype following our in vitro chronic stimulation protocol for both CD8⁺ and CD4⁺ T cells, demonstrated by perturbed immunophenotype, cytokine secretion, transcriptomics, mitochondrial dysfunction, and control of cancer cell line growth.

Having established a robust in vitro protocol for generating dysfunctional, exhausted T cells, we wanted to develop an RNP CRISPR pipeline to interrogate prospective therapeutic targets that may block T cell exhaustion. Genetic ablation of RASA2 has previously been demonstrated to ameliorate the effects of chronic NY-ESO-1 antigen induced CD8⁺ T cell exhaustion in the context of repetitive co-culture with tumor cells. We therefore sought to examine the impact of RASA2 KO in our T cell exhaustion assay. In contrast to current protocols and literature, we first performed a nucleofection in non-activated T cells using Cas9-RNP complex to target RASA2 (Figure 3A). This enabled us to confirm no impact of nucleofection on T cell viability or activation (Supplementary Figure S9B) before the cells were exhausted using our protocol as described above.

At the end of the exhaustion protocol, RASA2 sgRNA had >70% KO efficiency in CD8⁺ T cells, compared to the non-targeted controls (Unnucleofected, NTC sgRNA and LUC sgRNA; Figure 3B). Interestingly, no significant difference was seen in the proportion of PD-1⁺Tim-3⁺ CD8⁺ T cells between the exhausted controls and RASA2 KO following chronic stimulation. (Supplementary Figure S10A). After endpoint stimulation, no significant differences were observed in PD-1 or Tim-3 expression between controls and RASA2 KO cells. However, across donors, RASA2 KO cells did show a significant increase in 41BB expression (n=8; 1.29-fold; Supplementary Figure S10B), suggesting a degree of increased activation in response to re-stimulation. In support of this, cytokine secretion following endpoint stimulation showed RASA2 KO cells to have significant increases in TNF-α (3.21-fold), and IFN-γ (1.56-fold) secretion (Figure 3C), and although statistical significance was not achieved, numerical increases in Granzyme B (1.63-fold) and Granzyme A (1.44-fold) secretion were also observed (Figure 3C). In contrast, no overall difference in IL-2 secretion levels were observed as the impact of RASA2 KO varied between donors (Supplementary Figure S10B). Altogether, our data suggests that RASA2 depletion in the CD8⁺ compartment can impair the development of T cell dysfunction. The increased effector protein secretion arising from RASA2 KO recapitulates the findings of Carnevale et al. with our in vitro T cell exhaustion protocol.

Following validation of RASA2 KO in CD8⁺ T cells, we wanted to further characterize the role of RASA2 in blocking T cell exhaustion by extending analysis to the CD4⁺ T cell compartment. Similar to the experimental setup in CD8⁺ cells, RASA2 sgRNA was nucleofected with Cas9 into non-activated CD4⁺ cells prior to the exhaustion protocol (Figure 3A). In the CD4⁺ compartment, RASA2 protein expression was reduced with >80% KO efficiency (Figure 3D). In line with our CD8⁺ results, RASA2 KO increased cytokine secretion of CD4⁺ T cells, compared to exhausted controls. Specifically, significant increases of TNF-α (2.04-fold), IFN-γ (1.41-fold), Granzyme B (1.68-fold) and Granzyme A (1.55-fold) secretion were observed (n=8 donors; Figure 3E). As before, immunophenotyping showed PD-1 and Tim-3 co-expression remained unchanged in RASA2 depleted cells at the end of exhaustion (Supplementary Figure S10A). PD-1, Tim-3 and 4-1BB expression levels in response to endpoint stimulation also remained unaffected by RASA2 ablation in CD4⁺ cells, despite the improvement in effector cytokine response (Supplementary Figure S10B). In summary, this data indicates, for the first time, that ablation of RASA2 can also enhance effector function and block the development of dysfunction in CD4⁺ T cells.

Using RASA2 depletion as an exemplar, we show that our RNP CRISPR pipeline can validate T cell modulators which could be targeted to block the development of exhaustion. However, we wanted to utilize our platform to evaluate targets that could be key for reversing this dysfunctional state. To do this we first had to determine whether the dysfunctional phenotype established from our exhaustion protocol was stable over a resting period. As described earlier, scRNA-seq analysis of the CD8⁺ Tex S6R sample showed mild reduction in TEX-annotated populations following the resting period while Pseudotime analysis of genes of interest shows strong similarity in expression profile at end of exhaustion (S6) and in rested samples (S6R) for both Tex and Ts (Figures 2B, C). At a protein level, CD8⁺ Tex cells maintained an increased PD-1⁺ Tim-3⁺ population compared to the Ts controls after the 8-day resting period and retained the inability to upregulate 4-1BB in response to subsequent stimulation (Supplementary Figure S6B upper panels). After resting, both Granzyme B and IL-2 secretion of Tex cells remained statistically different to Ts controls (n=5 donors; Supplementary Figure S7B). Furthermore, despite not achieving statistical significance, reduced IFN-γ and TNF-α secretion (0.58-fold and 0.62-fold reduction, respectively) was also observed (n=5 donors; Supplementary Figure S7B). Overall, these data suggest maintenance of the dysfunctional phenotype after resting.

For CD4⁺ T cells, the impact of the resting period relied on flow cytometry and cytokine analysis. In terms of the proportions of the PD-1⁺TIM-3⁺ population after resting and the propensity of the cells to upregulated 4-1BB in response to stimulation, the impact of resting was more variable across donors (Supplementary Figure S6B bottom panels). However, characterization of T cell states based on CCR7/CD45RA expression and high dimensional clustering showed a similar pattern for CD4⁺ T cells at the end of exhaustion and following resting (Supplementary Figures S3C right panel, Supplementary Figure S5A) as described above. Analysis of the secretory profile showed that the significantly increased Granzyme A secretion of Tex cells compared to Ts was maintained after resting, while numerical reductions in IL-2 secretion (0.45-fold) of Tex cells was also observed albeit this did not reach statistical significance (Supplementary Figure S7). We also observed increased secretion of TGFβ on CD4⁺ Tex compared to equivalent Ts across S4 and S6 points of the exhaustion protocol (Supplementary Figure S12A), suggesting an immunosuppressive TGFβ feedback loop. Production of active form (cleaved) was observed by Western blot substantiating the cytokine readout (Supplementary Figure S12B). We focused on CD4⁺ TGF-β function given their dominant role in secreting this immunosuppressive cytokine compared to CD8⁺ T cells. Nevertheless, from our CD8 scRNA-seq pseudotime analysis, we did observe TGFβ expression in exhausted CD8⁺ maintained from stim 4 onwards, while a declining trend was noted for the Ts equivalent (Supplementary Figure S12C).

These data indicate dysfunction in CD4⁺ T cells is maintained after resting but this phenotype appears to be less stable than the CD8⁺ compartment.

Consequently, we investigated whether RASA2 depletion could reverse T cell exhaustion and restore effector function. For that purpose, we performed CRISPR-Cas9 KO in exhausted T cells. Following nucleofection, the cells were maintained in complete media for 8 days and then functionally assessed in their response to endpoint stimulation (Figure 4A).

Importantly exhausted CD8⁺ cells had high levels of viability (80-90%) prior to electroporation (Day 0) and 8 days post nucleofection (Figure 4B). After 8 days, RASA2 KO in exhausted CD8⁺ T cells was confirmed by western blot and showed ~65% KO efficiency (Figure 4C). In this reversal set up, a significant increase in PD-1 and Tim-3 co-expression (1.68-fold) was observed with RASA2 KO at the end of the resting period (Figure 4D). At post endpoint stimulation, 41BB was upregulated at a small extent, while similarly small increase was observed for PD1. (Figure 4D). As well as different immunophenotypic impacts of RASA2 KO in exhausted CD8⁺ T cells, we also observed an enhanced cytokine response. In this assay format, RASA2 KO led to significant increases in secretion of IL-2 (1.81-fold), TNF-α (1.63-fold), IFN-γ (1.25-fold) and Granzyme B (1.46-fold) secretion while the impact of RASA2 KO on Granzyme A secretion was found to vary between donors (n=6; Figure 4E). Overall, our novel approach of performing KO in exhausted cells verified that RASA2 depletion restores effector capacity of dysfunctional CD8⁺ T cells and highlights the importance of examining functional readouts when characterizing T cell exhaustion states.

Having seen the restorative impact of RASA2 KO in exhausted CD8⁺ cells, we extended our approach to test the reversibility of CD4⁺ T cell exhaustion. High levels of viability (80-90%) were observed with exhausted CD4⁺ T cells throughout the reversal nucleofection workflow (Figure 5A). In the CD4⁺ compartment, we successfully depleted RASA2 in exhausted cells with ~65% reduction in protein expression compared to non-targeting controls (Figure 5B). Similar to the results observed in the CD8⁺ compartment, RASA2 depleted CD4⁺ cells showed a significant increase (1.50-fold) of PD-1⁺ Tim-3⁺ cells at the end of the resting period (Figure 5C left panels), however, in contrast to the phenotype in CD8⁺ cells, RASA2 KO did not impact PD-1, Tim-3 or 4-1BB expression in response to re-stimulation in exhausted CD4⁺ T cells (Figure 5C middle and right panels). Nonetheless, cytokine analysis showed a significant increase in IL-2 (2.39-fold), Granzyme B (2.28-fold) and Granzyme A (1.52-fold) secretion upon RASA2 KO in exhausted CD4⁺ T cells. A moderate increase in TNF-α (1.42-fold) and IFN-γ (1.26-fold) secretion upon RASA2 KO was also observed, however variable magnitudes of effect between donors led to significance not being reached (Figure 5D).

Together, our results conclude that RASA2 depletion can re-invigorate exhausted CD4⁺ T cells, as demonstrated by increased effector protein secretion, and that this is potentially through a different mechanism than in exhausted CD8⁺ T cells.