High efficiency CRISPR knock-in demonstrates that TCF1 is insufficient to reverse T cell exhaustion

Abstract

CD8⁺ T cell exhaustion is a regulatory state triggered by chronic antigen stimulation in both cancer and persistent infection. The less differentiated stem-like sub-populations of exhausted T cells have been heavily studied given their importance to the efficacy of current immunotherapies. While the transcription factor TCF1 is both necessary and sufficient for formation and maintenance of these stem-like populations, it remains unclear whether TCF1 can actively de-differentiate more terminally exhausted subsets back into a stem-like state. To address this question, here we utilize and optimize a high efficiency CRISPR knock-in methodology, compatible with mouse in vivo exhaustion models, to engineer T cells that either constitutively over-express TCF1, or conditionally over-express TCF1 following differentiation of the cells into a CX3CR1⁺ intermediate-exhausted state. Strikingly, we find that only constitutive, and not conditional, TCF1 over-expression can increase the size of the stem-like T cell pool. Thus, while TCF1 can slow stem-like T cell differentiation, it is insufficient to revert more differentiated cells back into a stem-like state.

Exhaustion is a functional state that hampers anti-cancer and antiviral CD8 T cell activity, and is preceded by a stem-like state, maintained by the transcription factor TCF1. Here authors develop mouse models that allow a precise understanding of the developmental trajectory between the stem-cell-like and exhausted states of CD8 T cells and find that while constitutive overexpression of TCF1 expands the stem-like T cell pool, TCF1 expression specifically in already exhausted cells is unable to promote dedifferentiation.

Introduction

Chronic antigen stimulation in the context of viral infection and cancer drives CD8⁺ T cells into an exhausted state^1,2. T cell exhaustion is characterized by progressive loss of proliferative capacity, diminished cytokine production and upregulation of inhibitory receptors, such as PD-1^2–7. Exhausted T cells encompass a phenotypically and functionally heterogeneous pool of cells, the least differentiated of which are precursor of exhausted T cells (Tpex)^8,9. Tpex have elevated stem-like properties compared to other exhausted subsets, such as enhanced proliferative and differentiation potential, and thus play important roles in sustaining the exhausted T cell response¹⁰. Tpex constitutively seed and sustain the more differentiated intermediate (Tex-int) sub-population that is important for tumor and virus control, after which they ultimately become terminally exhausted (Tex-term)^9,11–21. Tpex also have clinical relevance as they mediate the proliferative response to immune checkpoint blockade (ICB) therapy in both viral and cancer models^{15,17–19,22}, with elevated intratumoral Tpex predicting better response to ICB therapy in cancer patients^{17,18,23–26}.

Due to the clinical relevance of Tpex, defining the pathways responsible for Tpex formation and maintenance is a major focus of the field. The transcription factor TCF1 is viewed as the “master regulator” of Tpex as it is both necessary and sufficient for Tpex formation and, consequently, for sustaining the exhausted CD8⁺ T cell response to chronic viral infection and within tumors^15,19,27,28. TCF1 directly induces the transcriptional program required for Tpex formation²¹, and its expression is lost once Tpex differentiate^15,16,20. TCF1 gene deletion results in loss of Tpex alongside chromatin accessibility changes associated with differentiation into Tex-int^15,19,27. Conversely, TCF1 over-expression in both tumor and viral models enhances stem-like features, increases Tpex number and improves viral and tumor control^21,27,29,30. Notably, though, these latter studies involved constitutive TCF1 over-expression in all exhausted cell subsets. It thus remains unclear whether TCF1 can actively reverse exhaustion by de-differentiating TCF1 negative subsets (ie. Tex-int and Tex-term) back into Tpex, or whether TCF1 over-expression simply slows Tpex differentiation. This is an important unanswered question with clinical relevance, as reverting differentiated exhausted subsets back into stem-like states could restore ICB responsiveness in treatment refractory patients. However, addressing this question is challenging as it requires conditional TCF1 over-expression within the more differentiated TCF1 negative exhausted cells. Engineering conditional over-expression of a transgene typically requires the time-consuming and costly generation of specialized transgenic or gene knock-in mice, and for this reason, questions such as this one have frequently remained unaddressed.

Recent progress in CRISPR-Cas9 based gene knock-in technologies have provided an alternative and potentially rapid approach to achieving conditional gene over-expression. Cas9-sgRNA ribonucleotide protein (RNP) complexes induce site-specific DNA double-strand breaks, which trigger DNA repair through either non-homologous end joining (NHEJ) or homology-directed repair (HDR). Whereas NHEJ causes indels, HDR can lead to precise integration of donor DNA into the targeted genomic locus when the provided DNA template is flanked by the appropriate homology arms. CRISPR HDR technology has now been employed to generate human CAR T cells³¹, to knock-in transgenes into human T cells^32,33, and to correct point mutations in hematopoietic stem cells as a treatment alternative for sickle cell disease^32,34. HDR knock-in can also be used for conditional gene expression by knocking the transgene into an endogenous gene locus with lineage-specific expression patterns^35–37. Nevertheless, HDR is a much lower efficiency process than NHEJ, and HDR editing within mouse T cells in particular has often failed to achieve the high efficiency levels seen within their human counterparts^38,39. While progress has recently been made to improve knock-in efficacy within mouse T cells^40–42, the bulk of prior studies were conducted in vitro meaning that the applicability of these methods to the study of in vivo T cell biology remains unclear. This has limited the capacity of CRISPR knock-in approaches to address research questions within primary murine CD8⁺ T cells. In contrast, NHEJ-based CRISPR gene knock-out approaches have been successfully and broadly applied to mouse T cells and have typically achieved high knock-out efficacy in a range of contexts, including naïve T cells and in sophisticated inducible systems^39,43–45.

Here, we describe a methodology for high efficiency CRISPR HDR-based gene knock-in within primary murine CD8⁺ T cells that is suitable for both in vitro and in vivo studies. Using this approach, we achieve highly specific, conditional transgene over-expression within Tex-int through transgene knock-in into the Cx3cr1 gene locus, which is selectively expressed within this cell subset. We leverage this approach to conditionally over-express TCF1 selectively within Tex-int cells and compare the resulting phenotype to that induced by constitutive TCF1 over-expression. While conditional TCF1 over-expression causes mild epigenetic, transcriptional and phenotypic changes in edited Tex-int cells, it ultimately fails to fully reprogram these cells back into a Tpex state and, unlike constitutive TCF1 expression, does not increase Tpex numbers. These data indicate that TCF1 is insufficient to reverse exhaustion.

Results

Optimal mouse T cell CRISPR HDR knock-in requires in vitro and in vivo rest before in vivo re-activation

We previously observed low HDR editing efficiency and high DNA template toxicity when using double-stranded DNA templates for HDR editing in primary mouse T cells³⁹. As previous studies have indicated that using AAV DNA repair templates for HDR editing can both reduce DNA toxicity and improve knock-in efficiency^36,40, we decided to trial AAV repair templates instead. Different AAV serotypes can successfully transduce murine cells^42,46–48, and AAV6 has demonstrated some capacity to infect primary murine CD8⁺ T cells^37,40,41,49. To test the efficiency of AAV6 templates for HDR, we employed templates targeting the GFP gene to the start codon of the Thy1 gene (Thy1-GFP). As the Thy1 gene is constitutively expressed in mouse T cells, this editing strategy will lead to constitutive GFP expression alongside loss of Thy1 expression. For these experiments, we used lymphocytic choriomeningitis virus (LCMV)-specific CD8⁺ T cell receptor transgenic T cells (P14 cells) to enable optimization of HDR knock-in for downstream in vivo studies within the well characterized chronic LCMV T cell exhaustion model. Previous work has suggested that HDR editing is most efficient within cycling cells^50,51 so purified P14 cells were first activated using anti-CD3 and anti-CD28 antibodies for 24 hours. Cells were then electroporated with Cas9-sgRNA RNPs targeting the Thy1 repair site prior to immediate incubation with AAV6 containing the Thy1-GFP repair template for 4 hours at a high multiplicity of infection (MOI; 100 K). These MOI and incubation conditions were based on previously published optimized parameters for AAV6 template editing in other cell types^37,52. The excess AAV was then washed off the cells, after which cells were cultured in IL-2 and IL-7 for a further 72 h (Fig. 1a). Using this protocol, we achieved nearly 80% GFP knock-in within P14 cells, with a corresponding loss of Thy1 protein indicating that GFP targeting was occurring specifically at the Thy1 locus (Fig. 1b, c). Of note, we also observed that a proportion of cells were edited by NHEJ, resulting in loss of Thy1 protein (presumably due to indels in the Thy1 gene) with no GFP insertion (Fig. 1b).

Fig. 1. Optimal knock-in efficiency and cell recovery of CRISPR edited T cells requires in vitro and in vivo rest before in vivo activation.

a CD45.1⁺ P14 cells were activated using anti-CD3 and anti-CD28 antibodies in IL-2 and IL-7 supplemented media for 24 hours (h) before electroporation with RNP complex. Immediately after electroporation, cells were incubated for 4 h in the presence of AAV6-Thy1-GFP and then either transferred in vivo or kept in vitro for 48 h before subsequent in vivo transfer (5000 P14 cells per mouse). Mice were infected with 2 × 10⁶ p.f.u. LCMV clone 13 at the point of cell transfer or 6 days later. Created in BioRender. Chen, A. (2026) https://BioRender.com/q96y268. Representative (b) and pooled (c) data showing GFP and Thy1.2 expression in CRISPR edited P14 cells after 72 h of in vitro culture. n = 4 biological replicates. CRISPR edited P14 cells were transferred into mice as in (a). Representative (d) and pooled (e, f) data showing GFP and Thy1.2 expression on splenic P14 cells (d, e), and P14 number (f), at day 8 post-infection (p.i.). Representative and pooled data from 3 independent experiments (n = 3 mice for in vitro condition, n = 9 mice for No rest and In vivo rest groups and n = 10 mice for In vitro rest group). g P14 cells were CRISPR edited as described in (a) except that cells were either transferred in vivo on the same day or left in vitro for 48 or 24 h before transfer in vivo, with mice LCMV infected 4-5 days (d) later. Created in BioRender. Chen, A. (2026) https://BioRender.com/y47c236. Representative (h) and pooled (i, j) data from the experiment in (g) showing splenic P14 phenotype and number at day 8 p.i. Representative data from 4 experiments (n = 12 mice for 6 d and 48 h + 4 d groups and n = 8 mice for 24 h + 5 d group)). Error bars show ± SEM. p values calculated by one-way ANOVA with Tukey’s multiple comparison test (f and j).

We next tested whether this high knock-in efficiency was preserved after in vivo transfer of P14 cells and subsequent activation by chronic LCMV-Clone 13 (Cl.13) infection. To minimize any impact of cell culture on in vivo differentiation, CRISPR-edited P14 cells were initially transferred into mice immediately after the 4 h AAV6 incubation period (“no rest”—Fig. 1a). While we still observed relatively high editing efficiency ( ~ 40% GFP⁺ cells), the results were more variable than that observed in vitro, with a trend towards reduced efficiency (Fig. 1d, e). The reasons for this reduced efficiency of in vivo versus in vitro knock-in are unclear but given that the method was identical prior to transfer into mice, we hypothesized that edited cells had a survival disadvantage when immediately activated in vivo after editing. To test whether a period of rest after HDR editing could potentially overcome this problem, we either cultured CRISPR-edited P14 cells in IL-2 and IL-7 for 48 h (“in vitro rest”) prior to transfer and immediate LCMV infection, or immediately transferred the cells into mice but waited for 6 days before LCMV infection (“in vivo rest”) (Fig. 1a). Both in vitro and in vivo rest led to less variability in GFP⁺ percentage (Fig. 1e) and, consistent with our hypothesis, splenic P14 cell recovery was higher when cells were rested in vivo (Fig. 1f). We finally tested whether combining in vitro and in vivo rest before LCMV infection improved P14 yield and/or editing efficiency (Fig. 1g). While editing efficiency was comparable across all conditions ( ~ 60%), 24 h in vitro rest plus 5 days in vivo rest resulted in the highest P14 yield among the conditions tested (Fig. 1j). Prior in vitro activation is expected to alter in vivo T cell differentiation, however, cell number, cytokine function and subset proportions were largely unaltered (Supplementary Fig. 1a-d), indicating that this protocol minimally impacts P14 cell exhaustion in vivo and can therefore be used to study exhausted T cell differentiation. Thus, we successfully optimized a protocol for high efficiency CRISPR knock-in within mouse T cells compatible with in vivo studies of T cell differentiation.

High efficiency CRISPR HDR editing requires T cell activation and is influenced by insert size

We and others have developed protocols that enable CRISPR gene deletion by NHEJ-induced indels with equally high efficiency in both naïve and activated T cells^39,43. We thus next tested whether the in vitro activation step was necessary for high efficiency CRISPR knock-in. Naïve P14 cells were either immediately edited without prior culture or cultured for 24 h in IL-7 to enable survival without activation prior to editing. Cells were then cultured for 72 h in vitro, with editing efficiency then compared to that seen in cells activated prior to editing. Editing efficiency was equally poor in both naïve conditions compared to pre-activated cells in vitro ( ~ 5% versus ~60% respectively; Supplementary Fig. 2a), and after in vivo transfer and LCMV challenge (Supplementary Fig. 2b). Thus, unlike CRISPR-induced NHEJ, optimal HDR editing requires that cells are actively in cycle, consistent with previous work^39,50.

We next examined how insert size impacts editing efficiency. AAV6 has a maximum DNA packaging capacity of 4.7 kb, which restricts HDR template size to a maximum of ~3 kb when accounting for the additional space required for the homology arms. The Thy1-GFP AAV6 construct from Fig. 1 was modified by adding non-coding regions (ncr) after the polyA tail of the GFP sequence to sequentially increase the insert size close to the maximum possible insert size that could be accommodated within the AAV6 genome (the homology arms were left unaltered) (Fig. 2a). A mock control electroporated with Cas9 protein, but no sgRNA, was also included. Increasing the insert size dramatically diminished editing efficiency (Fig. 2b-d) in line with previous findings⁵³. Thus, editing efficiency is highly dependent on the insert size.

Fig. 2. Insert size is a limiting factor for AAV6 mediated CRISPR HDR knock-in efficiency.

a Schematics showing construct design and length of non-coding regions (ncr) added to Thy1-GFP. Created in BioRender. Chen, A. (2026) https://BioRender.com/v55r550. b-d P14 cells were CRISPR edited as described in Fig. 1. Representative (b) and pooled (c) data showing GFP and Thy1.2 expression in CRISPR edited P14 cells after 72 h of in vitro culture. Mock control cells were electroporated with Cas9 protein alone without sgRNA. n = 3 biological replicates. d Simple linear regression of knock-in efficiency by insert size. n = 3 biological replicates. Error bars show ± SEM. *p < 0.05, ****p < 0.0001 by one-way ANOVA with Holm-Sidak’s multiple comparison test (c).

High efficiency knock-in does not require endogenous gene disruption

While CRISPR HDR knock-in can leverage endogenous regulatory elements to control transgene expression, this is often at the expense of the endogenous gene given that the transgene is typically inserted at the start codon. To trial approaches that preserve endogenous gene expression, we designed a new HDR template for GFP insertion at the Thy1 gene stop codon (Thy1-GFP [end]), with a self-cleaving P2A peptide sequence added before the GFP sequence to promote GFP cleavage from Thy1 protein during translation⁵⁴ (Fig. 3a). Similar editing efficiencies were achieved using both constructs (Fig. 3b, c) and, importantly, the GFP-expressing cells were still largely Thy1.2⁺ in the Thy1-GFP [end] condition (Fig. 3d). However, GFP expression levels were significantly lower when GFP was inserted at the end of the gene (Fig. 3e), possibly due to inefficient P2A protein splicing⁵⁵. Thus, transgenes can be inserted in a manner that preserves endogenous gene expression using this methodology, albeit at the expense of transgene expression level.

Fig. 3. High efficiency constitutive over-expression does not require endogenous gene disruption.

a Schematics of HDR-mediated insertion of GFP DNA templates into the start or the end of the Thy1 locus. Created in BioRender. Chen, A. (2026) https://BioRender.com/6dtwky2. b-h P14 cells were CRISPR edited as described in in Fig. 1. Representative (b, c) and pooled (d, e) data showing GFP and Thy1.2 expression in CRISPR edited P14 cells after 48-72 h of in vitro culture (b, d) or 8 days p.i. (c, d and e). b Representative data from 3 (Thy1-GFP [start]) and 4 (Thy1-GFP [end]) biological replicates. c Representative data from 2 independent experiments (n = 7 mice for Thy1-GFP [start] and n = 6 mice for Thy1-GFP [end]). d Pooled data from 3 biological replicates for Thy1-GFP [start] and 4 biological replicates for Thy1-GFP [end] (in vitro), and 2 independent experiments (in vivo) (n = 7 mice for Thy1-GFP [start] and n = 6 mice for Thy1-GFP [end]). e Pooled data from 2 independent experiments (n = 7 mice per group). f Schematics of HDR-mediated insertion of GFP DNA templates into the endogenous Rosa26 genomic locus. Created in BioRender. Chen, A. (2026) https://BioRender.com/g73u238. Representative (g) and pooled (h) data showing GFP expression in CRISPR edited P14 cells after 72 h of in vitro culture or 8 days p.i. Pooled data from 2 biological replicates (in vitro), or 2 independent experiments (in vivo) (n = 6 mice per group). Error bars show ± SEM. *p < 0.05, ****p < 0.0001 by ordinary two-way ANOVA (d) or unpaired parametric t test (e).

To develop an alternative CRISPR knock-in approach that would enable constitutive transgene expression without disrupting an endogenous coding gene, we designed HDR templates for GFP insertion into the Rosa26 locus (Fig. 3f). CRISPR gene knock-in into the Rosa26 locus has been reported in murine hematopoietic stem and progenitor cells (HSPCs)⁴⁸, but not murine primary CD8⁺ T cells. A ~ 40% knock-in efficiency was achieved in P14 cells both in vitro and in vivo (Fig. 3g, h). This provides an alternative approach for constitutive transgene over-expression by CRISPR knock-in that leaves endogenous gene expression intact.

The Cx3cr1 locus can be leveraged for conditional transgene over-expression within intermediate exhausted T cells

To enable conditional over-expression of transgenes after Tpex differentiation, we decided to target the Cx3cr1 gene locus. CX3CR1 is not expressed within Tpex, but is instead induced after Tpex down-regulate TCF1 and differentiate into a Tex-int state^11,12,20. We initially used a GFP transgene to validate the specificity of transgene expression. To preserve Cx3cr1 gene expression, the GFP transgene was inserted at the end of the Cx3cr1 gene separated by a P2A sequence (as for the Thy1-GFP [end] construct in Fig. 3) (Fig. 4a). While ~40% of CX3CR1⁺ P14 cells were GFP⁺ on day 8 p.i., this doubled by day 20 p.i. (Fig. 4b, c). Of note, while the proportion of CX3CR1⁺ P14 cells was comparable between days 8 and 20 p.i. (Fig. 4d), CX3CR1 MFI within CX3CR1⁺ cells increased 3-fold between these time points (Fig. 4e). In line with this increase in endogenous CX3CR1 expression level, we similarly observed an increase in GFP MFI within the GFP⁺ population at day 20 relative to day 8 (Fig. 4b, f). Importantly, GFP expression was restricted to the CX3CR1⁺ Tex-int population, with negligible expression within the other subsets (Fig. 4g, h). Thus, this targeting strategy enables highly specific Tex-int transgene over-expression while leaving endogenous CX3CR1 expression intact, with high transgene expression observed particularly at later stages of chronic LCMV infection.

Fig. 4. Targeting of the Cx3cr1 locus enables conditional transgene expression within intermediate exhausted T cells.

a Schematics of HDR-mediated insertion of GFP DNA templates into the endogenous Cx3cr1 locus. Created in BioRender. Chen, A. (2026) https://BioRender.com/n79f138. b-h P14 cells were CRISPR edited using the AAV6-CX3CR1-GFP template as described in Fig. 1. Representative (b, h) and pooled (c-g) data showing frequency and mean fluorescence intensity (MFI) of GFP or CX3CR1 expression in CRISPR edited CX3CR1⁺ P14 cells after 8 or 20 days p.i. Data from 2 independent experiments (n = 6 mice per group). i, j P14 cells were CRISPR edited using the AAV6-CX3CR1-GFP and/or AAV6-CX3CR1-mCherry templates. i Representative data showing GFP and mCherry expression in total P14 cells and (j) pooled data showing frequency of single or double expression of GFP/mCherry within edited cells at day 20 p.i. Data from 2 independent experiments (n = 8 mice per group). Error bars show ± SEM. **p < 0.01, ****p < 0.0001 by ordinary two-way ANOVA (c) or two-tailed Mann-Whitney t test (e, f).

Since HDR can result in both mono- and bi-allelic gene targeting³⁵ we investigated the extent of homozygous versus heterozygous knock-in within the Cx3cr1 locus. To assess mono-allelic editing frequencies, AAV HDR templates that targeted GFP or mCherry to the Cx3cr1 locus were used either separately or in combination during the in vitro editing step. Using this approach, we observed that on average ~20% of cells were GFP and mCherry double positive, indicating that a proportion of cells undergo mono-allelic editing (Fig. 4i, j). These data indicate that transgene positive cells are likely a mix of mono- and bi-allelically edited cells.

Conditional TCF1 over-expression is not sufficient to de-differentiate Tex-int back into Tpex

Previous studies have suggested that TCF1 over-expression can increase Tpex numbers^21,27,29, but it is unclear whether this is because TCF1 is slowing Tpex differentiation versus actively de-differentiating TCF1^- cells back into a Tpex state. To test whether TCF1 is sufficient to de-differentiate non-Tpex back into a Tpex state, HDR templates were designed targeting the full-length (p45) TCF1 isoform to the Cx3cr1 locus to enable TCF1 over-expression selectively within Tex-int cells (Fig. 5a). If TCF1 is capable of de-differentiating Tex-int into Tpex, then this editing strategy should increase Tpex numbers. As a positive control to mimic previous constitutive TCF1 over-expression approaches known to boost Tpex numbers, TCF1 was also knocked-in to the Thy1 locus. Given that TCF1 was being over-expressed, Tpex were defined as Slamf6⁺ cells based on prior studies identifying Slamf6 as a specific Tpex marker^{12,16,17,19,20}. There was a substantial ( ~ 5–10 fold) increase in the frequency of TCF1⁺ cells in both CRISPR-edited groups when compared to mock control cells, with significant increases in TCF1⁺ cell frequency on both days 8 and 20 for Thy1-TCF1, but only on day 20 p.i. for CX3CR1-TCF1 (Fig. 5b, c). Despite the lower TCF1⁺ cell frequency in CX3CR1-TCF1 cells at day 20, TCF1 MFI in the TCF1⁺ population within CX3CR1-TCF1 edited cells was similar to the levels found in endogenous Tpex cells, and higher in this group relative to TCF1⁺ cells within the Thy1-TCF1 edited group, indicating that the Cx3cr1 locus can drive higher per cell TCF1 expression within edited cells (Fig. 5d). Importantly, CX3CR1-TCF1 cells only over-expressed TCF1 within the Tex-int populations in contrast to the Thy1-TCF1 cells, which over-expressed TCF1 in all cells (Fig. 5e), including Tpex, which had a higher TCF1 MFI than control in the Thy1-TCF1 but not the CX3CR1-TCF1 group (Supplementary Fig. 3a). While Thy1-TCF1 cells exhibited the expected increase in Slamf6⁺ Tpex numbers at day 8 and 20 p.i., strikingly CX3CR1-TCF1 edited cells failed to demonstrate any Tpex increase at either timepoint (Fig. 5f, g). Similar trends were seen when measuring the number of less differentiated cells using different markers (Granzyme-B^neg and PD-1^lo TOX^lo cells) (Supplementary Fig. 3b-e). Thus, unlike constitutive TCF1 over-expression, conditional TCF1 over-expression was not sufficient to increase Tpex despite higher per cell TCF1 expression.

Fig. 5. Conditional over-expression of TCF1 is not sufficient to reprogram Tex-int into Tpex.

a Schematics of HDR-mediated insertion of TCF1 DNA templates into the endogenous Thy1 or Cx3cr1 genomic loci. Created in BioRender. Chen, A. (2026) https://BioRender.com/t58p429. (b–g) P14 cells were CRISPR edited using the AAV6-Thy1-TCF1 or AAV6-CX3CR1-TCF1 templates and transferred into mice as described in Fig. 1. Representative (b, e) and pooled (c, d) data showing frequency (b) and fold change (FC) of mean fluorescence intensity (MFI) (d) of TCF1 expression in CRISPR edited P14 cells after 8 (b, c) or 20 days p.i. (b, d, e). Representative (f) and pooled (g) data showing frequency of Slamf6⁺ P14 cells after 8 (g) or 20 days p.i. f, g Data from 2 independent experiments for day 8 (n = 8 mice per group for Thy1-TCF1, Thy1 sgRNA and mock groups and n = 7 mice per group for CX3CR1-TCF1 and CX3CR1 sgRNA groups) and data from 3 independent experiments for day 20 (n = 13 mice for Thy1-TCF1 group and n = 12 mice per group for Thy1 sgRNA, CX3CR1-TCF1, CX3CR1 sgRNA and mock groups). h Pooled data showing TCF1 MFI in edited P14 cells at day 20 p.i. that were transduced with a retrovirus (RV) vector containing TCF1, or CRISPR edited using AAV6-Thy1-TCF1, and transferred into mice as described in Fig. 1. EV indicates Empty vector. Data from 3 independent experiments (n = 11 mice per group for RV TCF1 and EV mCherry groups, n = 9 mice for Thy1-TCF1 group and n = 7 mice for mock group). Error bars show ± SEM. p values calculated by ordinary one-way ANOVA with Tukey’s multiple comparison test (c, g, h and d).

Previous work showing that TCF1 can increase Tpex have used retroviral over-expression approaches^21,27,29,30, raising the possibility that failed re-programming was due to weaker TCF1 over-expression by CRISPR knock-in. We thus compared TCF1 over-expression by our CRISPR knock-in approach to traditional retroviral over-expression approaches. As the retroviral vector drives constitutive TCF1 over-expression, cells retrovirally over-expressing TCF1 were compared to cells CRISPR-edited to similarly constitutively over-express TCF1 via the Thy1-TCF1 HDR template. Strikingly, TCF1 MFI within edited cells was higher in cells edited by CRISPR knock-in compared to retroviral transduction (Fig. 5h). Thus, our CRISPR knock-in approach can match, or even out-perform, conventional retroviral over-expression approaches. As CX3CR1-TCF1 knock-in led to even higher TCF1 levels than Thy1-TCF1 knock-in (Fig. 5d), these data suggest that failed reprogramming is not due to insufficient TCF1 over-expression.

To test if using a different TCF1 isoform would change outcomes, we examined the effect of different TCF1 isoforms on Tpex frequencies. However, consistent with prior work²¹, there was no difference in Tpex frequencies when either the p45 or p33 TCF1 isoforms were retrovirally over-expressed, with Tpex frequencies comparable to those seen in Thy1-TCF1 (p45) edited cells (Supplementary Fig. 3f, g). Thus, different TCF1 isoforms do not have different potencies in their capacity to increase Tpex. It was also possible that limiting amounts of the TCF1 co-factor β-catenin may be preventing reprogramming. While retroviral β-catenin over-expression alone boosted Tpex frequencies, combining conditional TCF1 over-expression with β-catenin over-expression failed to increase Tpex beyond that seen with β-catenin over-expression alone (Supplementary Fig. 3h, i). Thus, insufficient β-catenin does not explain failed reprogramming consistent with previous work indicating that combined TCF1 and β-catenin over-expression does not synergistically increase Tpex²⁹.

TCF1 over-expression causes subtle transcriptional and epigenetic changes in Tex-int cells

Given prior findings showing that TCF1 is a master regulator of the Tpex fate, the lack of phenotype within CX3CR1-TCF1 cells was unexpected. To investigate the impact of TCF1 over-expression within CX3CR1⁺ cells at the transcriptional and epigenetic level, we performed combined single-cell RNA-sequencing (scRNA-seq) and assay for transposase-accessible chromatin sequencing (scATAC-seq) on sorted P14 cells from the Thy1-TCF1, CX3CR1-TCF1 and mock groups at day 20 p.i. We identified 10 clusters closely associated with known exhausted T cell differentiation states⁵⁶ (Fig. 6a, b and Supplementary Fig. 4a-d). TCF1 over-expression in the Thy1-TCF1-edited cells caused expansion of Tpex in addition to increased effector-like cells (Clusters 9 and 5, respectively), with the CX3CR1-TCF1 cells demonstrating milder effector-like and/or intermediate cell expansion (Clusters 5 and 0) with little to no increase in Tpex (Cluster 9) (Fig. 6b, c). Thus, constitutive TCF1 expands Tpex and potentially cells in the early stages of differentiation, while conditional TCF1 over-expression has only mild impacts on subset proportions.

Fig. 6. TCF1 over-expression causes mild epigenetic and transcriptional changes in Tex-int cells.

a-g P14 cells were CRISPR edited using the AAV6-Thy1-TCF1 or AAV6-CX3CR1-TCF1 template and transferred into mice as described in Fig. 1. Total P14 cells were sorted 20 days p.i. and analyzed by combined whole cell scRNA-seq and scATAC-seq (multi-ome) analysis. a UMAPs illustrating scRNA-seq, scATAC-seq and Combined data clusters for all cells and for each condition. Proportions of all clusters (b) from (a), or Tpex cluster only (c), per CRISPR genotype in all cells. Heatmaps showing differentially accessible chromatin regions (d) and differentially expressed genes (e) between the Tpex cluster from Mock, and the CX3CR1⁺ clusters from each CRISPR genotype. f Gene Set Enrichment Analysis (GSEA) of Tpex or CX3CR1⁺ gene signatures in CX3CR1⁺ cells from each TCF1 over-expressing group relative to mock or each other. Fast gene set enrichment analysis (FGSEA) was performed using a two-sided rank-based gene set enrichment test, with gene lists ranked by avg_log2FC as input. Enrichment significance indicated by P values was estimated using a gene permutation-based null distribution with adaptive multilevel Monte Carlo sampling implemented via the fgseaMultiLevel function. P values are adjusted using the Benjamini-Hochberg false discovery rate (FDR) across gene sets within each comparison. NES indicates Normalized Enrichment Score, p_adj indicates adjusted p-value (g) Heatmap showing transcription factor motif enrichment within differentially accessible chromatin regions in all clusters from each group. Representative and pooled data showing Granzyme-B (h), CXCR5 (i), PD-1 (j) and TOX (k) expression in CX3CR1⁺ CRISPR edited P14 cells at day 20 p.i. gMFI = Geometric mean MFI. Data from 3 independent experiments (n = 13 mice for Thy1-TCF1 group and n = 12 mice for CX3CR1-TCF1 group). p values calculated by ordinary two-way ANOVA (h-k).

We next analyzed whether TCF1 over-expression within CX3CR1⁺ cells leads to gain or loss of any molecular features of Tpex. Analysis of the differentially accessible chromatin regions (DACR) and differentially expressed genes (DEG) between mock Tpex, and CX3CR1⁺ cells from all 3 groups, revealed an overall subtle, but detectable, gain of Tpex features and loss of CX3CR1⁺ cell features downstream of TCF1 over-expression (Fig. 6d, e, red boxes). Strikingly, CX3CR1+ cells from both TCF1 over-expressing groups had a significant enrichment of a Tpex transcriptional signature, and a significant depletion of a CX3CR1⁺ transcriptional signature, relative to mock (Fig. 6f). In contrast, there was no differential enrichment of these signatures between CX3CR1⁺ cells from the Thy1-TCF1 versus CX3CR1-TCF1 groups, indicating comparable TCF1-associated transcriptional changes in both conditions. Furthermore, a mild enrichment of TCF1 family binding motifs was seen among the DACRs across all non-Tpex cell clusters within the Thy1-TCF1 group, with a milder enrichment seen within the CX3CR1-TCF1 group (Fig. 6g). Thus, TCF1 over-expression within CX3CR1⁺ cells causes a mild transcriptional and epigenetic acquisition of Tpex features, however these changes appear insufficient to de-differentiate these cells back into Tpex.

We next examined if these mild phenotypic changes were evident by flow cytometry through phenotypic analysis of the TCF1⁺ and TCF1^- Tex-int populations (Fig. 5e) in the CX3CR1-TCF1 and Thy1-TCF1 conditions. Consistent with our multi-ome analysis, the TCF1⁺ Tex-int cells exhibited a mild gain of Tpex features relative to their TCF1^- counterparts, such as slightly reduced Granzyme B and PD-1 expression and slightly increased CXCR5 levels (Fig. 6h-j), suggesting partial reprogramming of the TCF1⁺ subset. However, TOX levels were unchanged between subsets, indicating no difference in commitment to exhaustion (Fig. 6k). These phenotypical changes were detected in both Thy1-TCF1 and CX3CR1-TCF1 groups.

Collectively, these results indicate that while TCF1 over-expression can cause mild changes in Tex-int gene expression, epigenetic state and phenotype, TCF1 is insufficient to de-differentiate Tex-int cells back into Tpex.

Discussion

To study TCF1 lineage reprogramming potential, we optimized a methodology involving a high efficiency CRISPR knock-in approach that minimally disrupts in vivo differentiation, and hence could be widely deployed for the in vivo study of T cell biology. Notably, we achieved this high knock-in efficiency using wild-type AAV6 templates, which were previously suggested to inefficiently transduce mouse T cells^40,49. The reasons for this discrepancy are unclear, but cell culture conditions during the AAV6 incubation step, such as culture volume, cell density, FBS concentration and duration of culture can influence AAV6 transduction⁵⁷. Furthermore, AAV6 incubation was conducted within five minutes of nucleofection, which has been shown to improve AAV6 transduction by facilitating virus uptake^52,57. More importantly, we achieved comparable levels of gene over-expression to those seen using conventional over-expression systems, such as retroviral transduction. This approach thus represents a valuable tool in future studies of fundamental mouse T cell biology.

Beyond conditional transgene over-expression, this approach could have broad utility in mouse T cell research. For example, current NHEJ CRISPR-based gene knock-out approaches often yield contaminating wild-type cell populations that do not delete the target gene and can be difficult to identify. Knock-in based gene replacement using GFP circumvents this problem by enabling tagging of bona fide knock-out cells. More broadly, GFP knock-in to specific gene loci could be used to generate gene reporter cells to track expression of genes for which there are no available antibodies. Similarly, target proteins could be tagged with antibody tags (such as FLAG tags) to enable assessment of DNA binding or for co-immunoprecipitation studies where appropriate antibodies are not available. Finally, more sophisticated approaches can also be employed, such as our recently described restriction of payload expression to the tumor microenvironment^37,58, adding degron domains to target proteins to enable acute and inducible protein depletion, knock-in of the Cre gene into lineage specific loci for conditional gene knock-out or lineage tracing within T cells bearing floxed alleles or flox-STOP reporter alleles respectively, or for orthotopic T cell receptor replacement to rapidly generate T cell receptor transgenic cells as has been reported previously^40,41.

Using our CRISPR HDR approach, we show that TCF1 is not sufficient to reverse exhaustion and de-differentiate Tex-int into Tpex, despite mildly altering chromatin state and gene expression. These findings challenge the idea that TCF1 alone can drive the Tpex fate and instead argue that other factors are needed to drive de-differentiation of TCF1^- exhausted cells. This is particularly surprising given the many prior reports demonstrating that TCF1 is a central mediator of stem-like properties within both CD4⁺ and CD8⁺ T cells^{19,21,59–66}. A likely explanation for why constitutive but not conditional TCF1 over-expression promotes Tpex accumulation is that TCF1 is unable to fully remodel the chromatin in terminally exhausted cells into a state permissive for stem-like cell gene expression. In this context, TCF1 will be able to enhance the Tpex gene expression program within pre-existing Tpex, and thereby delay their differentiation, but would have little impact on cell phenotype once cells exit a Tpex state and lose their associated chromatin state. Indeed, our multi-ome data support a model whereby TCF1 slows differentiation, as constitutive TCF1 over-expression led to accumulation of both Tpex and what appeared to be less differentiated non-Tpex effector-like cells. However, there is conflicting data around whether TCF1 can act as a pioneer factor to remodel chromatin during differentiation. During T cell development, TCF1 can directly bind to genomic regions and promote chromatin remodeling necessary for induction of T cell lineage genes and repression of non-T cell lineage programs^67,68. Similarly, in mature CD8⁺ T cells, TCF1 can directly promote chromatin interactions that maintain accessibility at sites associated with CD8⁺ T cell lineage identity⁶⁹. In exhausted T cells during chronic LCMV, constitutive TCF1 over-expression led to increased chromatin accessibility at TCF1 binding sites in parallel with an increase in Tpex numbers, suggesting that TCF1 can directly regulate the chromatin landscape²⁹. However, it is possible that the capacity of TCF1 to act as a pioneer factor is context dependent and may depend on the co-factors present at the time of TCF1 expression. Moreover, prior studies often do not distinguish between whether TCF1-dependent chromatin changes were due to de novo chromatin remodeling by TCF1, versus TCF1-dependent maintenance of accessibility at sites that have been opened by other factors. Indeed, consistent with our findings, prior work has suggested that instead of being a pioneer factor, TCF1 acts as a placeholder to maintain chromatin accessibility at lineage-specific loci within naïve and activated T cells⁷⁰. Similar recent findings in acute infection suggest that de-repression of TCF1 in effector cells is insufficient to redirect these cells to a memory state due to reduced chromatin accessibility of TCF1 binding sites driven by effector-associated transcription factors⁷¹.

These data have clinical implications given the importance of Tpex in a variety of therapeutic contexts, such as cancer immunotherapy. Loss of Tpex at the expense of more terminally exhausted subsets has been linked to therapy failure in both immune checkpoint blockade and adoptive cell therapy^8,72. De-differentiating terminally exhausted subsets into a stem-like state is thus a major focus of the field as it could restore therapy-responsiveness in a subset of refractory patients. Our data argue that exhausted cell de-differentiation is likely complex and may require multiple factors, an important consideration when engineering any therapy aimed at boosting stem-like T cell populations.

Methods

Mice

CD45.2⁺ C57BL/6 (B6) mice (Mus musculus) were purchased from the Walter and Eliza Hall Institute Kew Animal Facility (Kew, VIC, Australia), whereas CD45.1⁺ P14 transgenic⁷³ mice were bred in-house with constant backcrossing to the Kew Animal Facility C57BL/6 background. Both male and female mice were used at 6–14 weeks of age, and all mice were maintained in specific pathogen free conditions. Both male and female mice were used in the study and sex was assigned according to mouse availability and donor mouse sex match. We do not see sex-dependent differences in exhausted T cell differentiation after chronic LCMV infection, although mice within each experiment were still sex matched across groups. Mice were euthanised by cervical dislocation or CO₂ asphyxiation. All animal work was in accordance with protocols approved by the Peter MacCallum Cancer Centre Animal Experimentation Ethics Committee (Protocols E669 and 2024-25) and current guidelines from the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes.

Cell lines

HEK293T cells for retrovirus production were obtained from the American Type Culture Collection (ATCC). Cells were maintained in DMEM medium (Gibco Life Technologies) supplemented with 10% heat-inactivated fetal bovine serum, 2 mM glutamine and 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin. These cells were cultured at 37 °C in a humidified incubator with 5% CO₂.

CD8⁺ T cell isolation and activation

Naïve CD45.1⁺ P14 cells were enriched using the EasySep Mouse CD8⁺ T Cell Isolation Kit (STEMCELL Technologies) from spleens and lymph nodes isolated from the CD45.1⁺ P14 TCR transgenic mouse strain. Spleens and lymph nodes were homogenized through a 70 μm strainer to obtain a single cell suspension prior to red blood cell lysis in 0.83% NH₄Cl. Cells were then resuspended in PBS containing 0.5% Fetal Bovine Serum and 2 mM EDTA for CD8⁺ T cell enrichment. Isolated P14 cells were subsequently activated at a concentration of 2 × 10⁶ cells/mL in a cell culture plate coated with murine anti-CD3 (0.5 μg/mL) and murine anti-CD28 (0.5 μg/mL) antibodies, in the presence of complete RPMI (Gibco) containing 50 μM β-mercaptoethanol (Gibco) supplemented with hIL-2 (100 IU/mL) and mIL-7 (5 ng/mL) for 24 hours. For IL-7 culture, cells were cultured in complete RPMI containing β-mercaptoethanol (50 μM) supplemented with mIL-7 (5 ng/mL) for 24 hours.

CRISPR/Cas9 gene editing and AAV transduction

To perform CRISPR/Cas9 editing, 26.6 pmoles recombinant Cas9 (IDT) and 270 pmoles sgRNA (IDT) were combined and incubated at RT for 10 minutes to form Cas9/sgRNA ribonucleoprotein (RNP) complexes. P14 cells were harvested from culture plates and washed twice with PBS. Up to 20 × 10⁶T cells were resuspended in 20 μL of warm P3 electroporation buffer (containing 82% P3 buffer and 18% Supplement 1; Lonza), combined with RNP and electroporated using a 4D-Nucleofector X Unit (Lonza) using pulse code CM137 for activated murine T cells and DN100 for naïve and IL-7 cultured murine T cells. After electroporation, T cells were resuspended in warm complete RPMI at a concentration of 50 × 10⁶ cells/mL and cells were then immediately added to AAV6 at a multiplicity of infection (MOI) of 100 K for activated cells and 400 K for naïve cells in the presence of 2 uM of M38142 NHEJ inhibitor (MedChemExpress), not exceeding 30 μL of volume per well in a 96-U bottom well plate. Cells were incubated at 37 °C for 4 hours before AAV6 was washed off and cells were either kept in vitro or transferred into mice as specified in the text.

Adoptive transfer and LCMV infection

5 × 10³ CD45.1⁺ P14 cells were adoptively transferred into B6 recipient mice intravenously (i.v.). Mice were then i.v. infected with 2 × 10⁶ p.f.u. Cl.13 on day 5 post transfer of edited P14 cells, unless stated otherwise as in Fig. 1.

HDR templates and AAV vectors

HDR templates and sgRNA were designed using the Benchling online tool (https://benchling.com 2024, USA). The Thy1-GFP HDR template was designed by targeting the start of the second exon of Thy1 genomic locus (sgRNA: GGAGAGCGACGCTGATGGCT) and homology arms were designed to exclude the Thy1 start codon, resulting in deletion of the endogenous Thy1 gene upon HDR knock-in. The GFP sequence was added between the homology arms followed by a polyA tail. Thy1-GFP [end] construct was designed to target Thy1 exon 4 using a different sgRNA (TCTGTGACTGGTTGGGCCCA). Homology arms were designed to exclude the stop codon, a P2A linker was added before the GFP sequence and no polyA tail was added to the construct. Rosa26-GFP HDR template was designed based on previous work⁷⁴, using the same sgRNA, but homology arms were shortened to 600 bp of length. CX3CR1-GFP template was designed targeting the end of exon 2 (sgRNA: GGGTCTCTCCTGCTCTGAAG), homology arms were designed to exclude the stop codon and a P2A linker was added before the GFP sequence. The TCF1 over-expression templates were designed from Thy1-GFP and CX3CR1-GFP by replacing the GFP sequence with the full-length TCF1 cDNA sequence.

Plasmids containing HDR templates were cloned into a pAAV6 backbone and synthesized by Genscript (USA). AAV6 production was performed by Packgene (USA).

Retroviral over-expression

TCF1 p33, TCF1 p45 and β-catenin sequences were cloned into the MSCV-IRES-mCherry plasmid (Addgene #52114) by Genscript. Retroviruses were produced in HEK293T cells with MSCV and pCL-Eco (AddGene #12371) plasmids using FuGENE HD (Promega). P14 cells were activated for 24 hours as per the CRISPR-HDR protocol and then transduced with retroviral supernatant containing Polybrene (2 μg/mL) and hIL-2 (100 IU/mL) by spinfection (1,260 g for 90 minutes at 25 °C), followed by incubation at 37 °C for 4 hours. Retroviral supernatant was then replaced with complete media containing hIL-2 (100 IU/mL) and mIL-7 (5 ng/mL) and cells were kept in vitro for 24 hours before transfer into mice.

Multi-ome analysis

For combined whole cell scRNA-seq and scATAC-seq multi-ome analysis, B6 mice were injected with 5  × 10³ CRISPR-edited P14 CD45.1 cells and infected with Cl.13 after 5 days. 20 days post infection, CD45.1⁺ CD8⁺ cells (ie. P14 cells) were sorted from splenocytes of recipient B6 mice. Sorted cells were incubated with Biolegend TotalSeq A anti-mouse antibody hashtags at a concentration of 0.82 µg/1 × 10⁵ cells/50 µl (#1-3, Cat #155801, #155803 and #155805) to enable sample multiplexing and super-loading on the 10X Genomics single cell sequencing platform. Whole cell scATAC-seq and scRNA-seq multi-ome analysis was performed using a previously reported protocol⁷⁵ (using the OMNI permeabilization method) by the Molecular Genomics Core at the PMCC. Cells were processed using the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kit (10X Genomics) and 1.6 × 10⁴ cells were loaded onto the 10X Genomics Platform for downstream sequencing and analysis. Hash Tag sequences were recovered through sequencing on an Illumina NextSeq machine, while gene expression and ATACseq data were recovered through sequencing on an Illumina NovaSeq machine.

Data matrices from multi-omic data were generated using the 10× Cell Ranger ARC pipeline (version 2.0.0) and mm10 genome. Specifically, cellranger-arc count was used to generate raw feature barcode matrices from FASTQ files. The resulting reads were used to generate antibody capture data matrices with CITE-seq-Count (version 1.4.5). Downstream analysis was performed in R (version 4.4.2) and Seurat (version 5.0.2).

Count matrices were processed using decontX to remove ambient RNA contamination using both raw and filtered feature-barcode matrices as input. The decontaminated counts were then passed on to quality control using the scater and scuttle packages. Per-cell QC metrics were computed using addPerCellQCMetrics. Outlier cells were detected using median absolute deviation (MAD)-based thresholds for total counts (library size), number of detected features, and mitochondrial transcript percentage (nmads = c(low = 2, high = 3)). Cells failing any threshold were discarded as low-quality cells along with mitochondrial genes. Filtered data was then log-normalised using logNormCounts, and doublet detection was performed using scDblFinder (v1.16.0). This filtered data was then converted into a Seurat object, and the doublets were retained until the later HTO demultiplexing step.

Normalised expression matrices were processed and scaled with Seurat’s NormalizeData, FindVariableFeatures and ScaleData workflow, with regression on mitochondrial percentage. Dimensionality reduction with PCA and UMAP was performed using the first 30 PCs. Standard Seurat processing and normalisation steps for a gene expression assay were performed: NormalizeData, FindVariableFeatures, ScaleData, RunPCA, RunUMAP, FindNeighbors and FindClusters.

Antibody capture data matrices from hashtag oligonucleotides (HTO) were demultiplexed using HTODemux (Seurat). Doublets were detected using DoubletFinder (version 2.0.4). Cells that were not HTO singlets and/or droplet singlets were filtered out.

Chromatin accessibility data was analysed using Signac (version 1.12.0) and ArchR (version 1.0.2) using default parameters unless otherwise noted. The union peak list was generated using a hybrid approach utilising MACS3 and ArchR. A first set of peaks was called with ArchR with default parameters based on clusters generated from the latent semantic indexing dimension reduction of the tile matrix, which allowed peaks to be called on unbiased cell clusters. In addition, we called a second set of peaks on this assay using the CallPeaks function which uses the MACS3 algorithm and its default parameters. A union of both two peak lists was used to generate a custom peak list, which was added to the Seurat object using FeatureMatrix and CreateChromatinAssay. Peak annotation was performed using the ClosestFeature function and using EnsDb.Mmusculus version 79 annotations. Standard Signac processing and normalisation steps for a chromatin assay were performed as follows: FindTopFeatures, RunTFIDF, RunSVD and RunUMAP. Signac LinkPeaks was applied to calculate the correlation gene expression and peak accessibility, taking into account its GC-content, accessibility, and length of the peak from the gene transcription start site (TSS).

To perform integrative analysis of both these modalities, the FindMultiModalNeighbors function was applied to calculate the nearest neighbours in the dataset based on a weighted combination of gene expression and chromatin accessibility similarities per cell, based on its respective dimensionalities. Dimensional reduction using RunUMAP was then performed to visualise cell-specific weighted nearest neighbour (WNN) information with graph-based clustering using Seurat DimPlot. Clustering analysis of the weighted shared nearest neighbour (WSNN) information in these assays was performed with FindClusters. Clusters with low-quality metrics were removed, and the final cluster resolution of 1.1 was determined using results from the clustree package (version 0.5.0).

Cell clusters were annotated using a combination of manual annotation of known cell type markers and ProjecTILs (version 3.0.3). A ProjecTILs reference object was generated with the scRNA-seq data from Giles et al.⁵⁶ with the make.reference function. The cell labels were projected onto our query dataset using the Run.ProjecTILs function. Multi-omic data was manually split into CX3CR1⁺ or Tpex subsets based on cell type (CX3CR1⁺ = Cells annotated as ExhInt, TransCTL, Efflike in the multi-ome dataset; Tpex = Cells annotated as Tpex in the multi-ome dataset). Each subset object was subject to re-normalisation per the steps mentioned above. Motif activities for each object were calculated with RunChromVar using BSgenome.Mmusculus.UCSC.mm10 (version 1.4.3) and JASPAR2020 (species 9606, version 0.99.10) as references. The resulting motif information was added to each respective Seurat object as a “chromvar” assay.

Cell cluster proportion bar plots were plotted with dittoSeq (version 1.8.1). Differentially expressed genes (DEGs) and differently accessible chromatin regions (DACRs) were detected with Seurat FindMarkers, supplying gene counts and peak counts as latent variables, respectively. Where applicable, Signac ClosestFeatures was used to define the closest gene to each peak set. DEGs and DACRs were deemed significantly differential if the adjusted p-value was <0.05. TCF1⁺ cluster gene expression heatmap and the DACR heatmap were generated with ComplexHeatmap (version 2.14.0).

Flow cytometric analysis

Spleens were harvested and processed as described above. For cell surface staining, cells were stained on ice in FACS buffer (PBS containing 2.5% FCS and 0.1% sodium azide) using the following anti-mouse Abs and reagents: Zombie Aqua (BioLegend, Cat. no. 423102, 1:1000), anti-CD8-BUV395 (Clone 53-6.7, BD Biosciences, Cat. no. 563786, 1:200), anti-CD45.1-Pacific Blue (Cat. no. 110722, 1:200), -FITC (Cat. no. 110706, 1:200) or -APC (Cat. no. 110714, 1:200) (Clone A20, BioLegend), anti-Thy1.2-PE (Clone 53-2.1, BD Biosciences, Cat. no. 553005, 1:1000), anti-CD44-Pacific Blue (Cat. no. 103020, 1:200) or -APC (Cat. no. 103012, 1:200) (Clone IM7, BioLegend), anti-CXCR5-PECy7 (clone 2G8, BD Biosciences, Cat. no. 560617, 1:25), anti-PD-1-BV785 (clone 29 F.1A12, BioLegend, Cat. no. 135225, 1:200). For intracellular staining, cells were fixed and permeabilized using eBioscience Foxp3/Transcription factor staining buffer set (ThermoFisher Scientific) and stained with rabbit anti-mouse TCF1 mAb (clone C63D9, Cell Signaling Technology, Cat. no. 2203S, 1:400) detected with a polyclonal secondary anti-rabbit IgG AlexaFluor594 Ab (ThermoFisher Scientific, Cat. no. A-11012, 1:1000), anti-Granzyme-B-APC (Clone GB11, Invitrogen, Cat. no. GRB05, 1:200) and anti-TOX-PE (clone TXRX10, Invitrogen, Cat. no. 12-6502-82, 1:100). For intracellular cytokine staining, splenocytes were restimulated with 0.1 μg/mL of the GP33-41 LCMV peptide (Biomolecular Resource Facility, ANU) in media containing 3 μg/mL Brefeldin A (eBioscience). Cell suspensions were incubated at 37 °C, 5% CO₂ for 5 hours, prior to surface staining then fixation with Biolegend Fixation Buffer, and intracellular cytokine staining in eBioscience Permeabilization Buffer containing the following antibodies: anti-IFNγ-PECy7 (clone XMG1.2, eBioscience, Cat. no. 25-7311-82, 1:2000) and anti-TNFα-PE (clone MP6-XT22, BioLegend, Cat. no. 506306, 1:2000). Samples were acquired on a BD LSRFortessa X-20 (BD Biosciences) and analyzed using FlowJo software (BD Biosciences) and GraphPad Prism (GraphPad Software). Gating strategies are illustrated in Supplementary Fig. 5.

Statistics

Microsoft Excel (v16.71) was used for data calculations and statistical analysis was performed using Prism Software (GraphPad, v9.0.2). P values were calculated using an unpaired t-test, one-way ANOVA with a Tukey’s Multiple Comparison test or two-way ANOVA as specified in each figure legend.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.N.M.: conceptualization, investigation, methodology, formal analysis, visualization, writing—original draft, writing—review and editing, A.C.: conceptualization, methodology. N.K.: investigation. SS: investigation. N.Y.L.S.: formal analysis, visualization. K.M.Y.: methodology. I.P.N.: methodology, investigation. S.R.: methodology, investigation, writing—review and editing. C.D.T.D.: investigation. B.H.: investigation. K.M.R.: investigation. I.M.: conceptualization, methodology. P.A.B.: conceptualization, funding acquisition, methodology, project administration, resources, supervision, writing—original draft, writing—review and editing. I.A.P.: conceptualization, funding acquisition, methodology, project administration, resources, supervision, writing—original draft, writing—review and editing.

Peer review

Peer review information

Nature Communications thanks Frank Staal and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data generated in this study are provided in the article itself, its supplementary information and in the Source Data file. The multi-ome (scRNAseq and scATACseq) sequencing data from Fig. 6 and Supplementary Fig. 4 have been deposited in GEO NCBI under the accession code GSE312328. Source data are provided with this paper.

Competing interests

The authors declare the following competing interests. I.A.P. declares research funding from AstraZeneca, Bristol-Myers-Squibb and Roche Genentech. P.A.B. declares research funding from Bristol-Myers-Squibb. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69671-y.