Epigenetic scarring of exhausted T cells hinders memory differentiation upon eliminating chronic antigenic stimulation

Abstract

Exhausted CD8 T cells (T_EX) are a distinct state of T cell differentiation associated with failure to clear chronic viruses and cancer. Immunotherapies like PD-1 blockade can re-invigorate T_EX cells, but re-invigoration is not durable. A major unanswered question is whether T_EX cells differentiate into functional durable memory T cells (T_MEM) upon antigen clearance. Here, using a mouse model, we found that upon eliminating chronic antigenic stimulation, T_EX cells partially (re)acquire phenotypic and transcriptional features of T_MEM cells. These “recovering” T_EX cells originated from the T-cell factor (TCF-1⁺) T_EX progenitor subset. Nevertheless, the recall capacity of these recovering-T_EX cells remained compromised compared to T_MEM cells. Chromatin-accessibility profiling revealed failure to recover core memory epigenetic circuits and maintenance of a largely exhausted open chromatin landscape. Thus, despite some phenotypic and transcriptional recovery upon antigen clearance, exhaustion leaves durable epigenetic scars constraining future immune responses. These results support epigenetic remodeling interventions for T_EX cell targeted immunotherapies.

INTRODUCTION

Memory CD8 T cells (T_MEM) persist long-term, can self-renew in the absence of antigen via homeostatic cytokines, interleukin-7 (IL-7) and IL-15, and generate robust secondary T_EFF responses upon reinfection¹. However, in cancer and chronic viral infections continuous antigen exposure results in “exhaustion”². Exhausted CD8 T cells (T_EX) have altered effector functions, and sustained high expression of inhibitory receptors (IRs), including programmed death-1 (PD-1)^3–5. T_EX lack the key T_MEM property of antigen-independent self-renewal and become addicted to persisting antigen for maintenance^6–8. T_EX have a distinct transcriptional program^{4, 9, 10} and epigenetic landscape^11–16 compared to T_EFF and T_MEM. The epigenetic divergence of T_EX from T_EFF and T_MEM is driven by the transcription factor Tox^17–22 and the T_EX-specific chromatin landscape is stable even following PD-1 pathway blockade¹¹. Thus, T_EX are a distinct type of CD8 T cells with specific functional and phenotypic properties, and a unique underlying transcriptional and epigenetic program. Whether these properties and programs change upon resolution of chronic disease remains poorly understood.

T_EX are heterogeneous, with two main subsets. The PD-1^IntCD44^Hi T_EX progenitor subset (Prog-T_EX) responds to PD-1 pathway blockade, whereas the PD-1^HiCD44^Int terminal subset (Term-T_EX) does not²³. A proliferative hierarchy exists, whereby the PD-1^Int T_EX progenitor subset gives rise to a larger terminal T_EX subset through antigen-driven proliferation^24–28. The transcription factor TCF-1 is critical for the maintenance of this Prog-T_EX subset^25–28. It is now clear that additional heterogeneity exists in T_EX ^29–31. However, all T_EX subsets require Tox²⁹ and are epigenetically distinct from T_EFF and T_MEM³².

Developing T_EX can be rescued from irreversible exhaustion and reprogrammed towards memory if removed from antigen exposure early, but not after prolonged stimulation³³. This is consistent with progressive epigenetic changes¹³, and establishment of NFAT-independent Tox expression later during chronic antigen exposure¹⁷. Nevertheless, a small fraction of T_EX may persist following antigen elimination and some T_EX features may recover^{6, 7, 33, 34}. In patients cured of chronic HCV, previously-exhausted HCV-specific CD8 T cells can persist and display reduced PD-1 expression and improved function, perhaps due to persistence of Prog-T_EX^35–37. A key question is what changes occur phenotypically, functionally, transcriptionally, and epigenetically to T_EX and T_EX subsets upon cure of chronic disease. This question is of increasing relevance, as more patients are being cured of cancer by immunotherapies and of chronic HCV infection by direct-acting antivirals (DAAs)^38–40.

Here, we addressed these questions using the mouse model of chronic lymphocytic choriomeningitis virus (LCMV) infection, the prototypical model of T cell exhaustion. Upon removal from chronic antigen exposure, only a small proportion of T_EX were able to survive and persist, but those persisting T_EX partially (re)acquired some T_MEM phenotypic and transcriptional features while losing some T_EX characteristics. T_EX persisting after antigen withdrawal were mainly derived from TCF-1⁺ T_EX progenitors, indicating preferential survival. However, despite some memory-like phenotypic and transcriptional features, these “recovering” T_EX (REC-T_EX) were highly compromised upon re-challenge infection. ATAC-seq analysis demonstrated that REC-T_EX underwent only minor changes in chromatin accessibility upon withdrawal of antigen and retained a largely T_EX epigenetic identity. This study provides insight into molecular mechanisms of T_EX differentiation upon antigen elimination during chronic disease, and has implications for patients cured of cancer or chronic viral infection.

RESULTS

T_EX downregulate exhaustion markers upon antigen elimination

Although most T_EX are lost upon antigen removal, a fraction persists^{6, 7, 33–37}. The differentiation program of these persisting T_EX remains poorly understood. Here, we used an adoptive transfer approach in a mouse model of chronic viral infection³³ to dissect the mechanisms and differentiation of persisting T_EX. Briefly, CD8 T_EX isolated from spleens of mice chronically infected with LCMV clone 13 (LCMV-Cl13) were isolated 4 weeks post-infection (p.i.) and adoptively transferred into congenically-distinct, infection-free recipient mice. An equal number of antigen-specific T_MEM (generated in LCMV Armstrong (LCMV-Arm) infection at >week 4 p.i) were adoptively transferred into matched recipient mice (Fig. 1a). Recipient mice had previously cleared LCMV-Arm infection, hereafter called LCMV-Arm immune. Immunity in these recipients controlled for any possible low-level virus transfer. One day prior to infection, donor mice were engrafted with a physiological number (10³) of congenically-distinct TCR transgenic P14 cells, specific for the H-2D^b restricted GP_33-41 epitope of LCMV^{41,42, 43}.

Figure 1: Antigen elimination results in acquisition of a partial memory phenotype by T_EX.

(a) Experimental design. CD8+ T cells from LCMV-Cl13 infected mice containing P14 cells (T_EX P14) were adoptively transferred into congenic, infection-free recipient mice. T_MEM P14 cells were transferred into matched recipient mice as a control. (b) Left, representative dot plots for the frequency of donor P14 cells. Right, number of P14 cells per spleen at week 4 post-transfer as a percentage of original number of P14 cells adoptively transferred on d0 (estimating 10% take). N=3 biologically independent experiments. (c) Left, representative dot plots of CD127 and PD-1 expression. Right, percentages of PD-1⁺CD127⁻ and CD127⁺PD-1⁻. (d) Left, representative dot plots of TCF-1 and Tim-3 expression and percentage of TCF-1⁺Tim-3⁻ subset. Right, dot plots and graphs for Tox and PD-1 expression. (e) Representative histograms for expression of PD-1 (left), Tox (middle), or TCF-1 (right). To the right of histograms, corresponding graphs of mean fluorescence intensity (MFI) subtracted from MFI of endogenous naïve cells (ΔMFI). Data (c-e) for PD-1 and CD127 n=12 biologically independent experiments for d0 and n=7 for d28, for Tox and TCF-1 n=4 for d0 and n=2 for d28. (f) Top, representative dot plots of CD160 and PD-1 expression. Bottom, percentages of co-expression of IRs on P14 cells. N=3 biologically independent experiments. (g) Top, representative dot plots of IFNγ production and LAMP-1 (i.e. CD107a) expression in response to peptide stimulation. Bottom, percentages of polyfunctional cells. N=3 and 2 biologically independent experiments for d0 and d28, respectively. (h) Phenotypic profile of T_EX, T_MEM, and REC-T_EX. Level of expression of indicated markers represented as a percentage of the highest expressing cell type. Data presented as mean +/− SD. All analyses for Figure 1 performed by unpaired two-tailed Mann-Whitney test. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Data for d28 from 2-8 mice and 2-5 mice per experiment for the T_EX and T_MEM groups, respectively.

Most donor T_EX were poorly maintained compared to T_MEM upon adoptive transfer to infection-free mice, as expected^{6, 34}, though some T_EX persisted in absence of antigen (Fig. 1b). After 1 month, frequency of CD127⁺PD-1⁻ donor T_EX increased from ~0.37% on adoptive transfer day (d0) to 23.8% (64-fold increase) (Fig. 1c). Additionally, the frequency of TCF-1⁺Tim-3⁻ and PD-1^int Ly108/Slamf6⁺ donor T_EX increased 4-fold and 2.23-fold, respectively (Fig. 1d and Extended Data 1b–1c). Conversely, the frequency of the CD127⁻ PD-1⁺, TCF-1⁻Tim-3⁺, and PD-1^hi Ly108/Slamf6⁻ donor terminal T_EX populations, decreased by 3.7-fold, 2.3-fold, and 2.3-fold, respectively. Moreover, the percentage of donor T_EX co-expressing PD-1 and Tox, decreased by half (77% versus 36.6%) (Fig. 1d). The amount of these proteins expressed/cell also changed with PD-1, Tox and Eomes decreasing, and TCF-1 and CD127 increasing (Fig. 1e and Extended Data 1d–1e). Despite this phenotypic shift away from exhaustion and towards memory, the phenotype of T_EX at week 4 post-transfer did not fully recapitulate that of bona fide T_MEM with residual expression of many exhaustion markers. Examining donor T_EX at later time points (up to d520 post-transfer) showed a trend of continued increasing expression of markers of T_MEM, but even at these late time points Tox expression remained elevated (Extended Data 1g).

We examined co-expression of IRs including PD-1, 2B4 (CD244), Lag-3, CD160, TIGIT and Tim-3 on donor T_EX. The frequency of T_EX co-expressing 4+ IRs significantly decreased upon antigen elimination. However, the frequency of T_EX expressing 1-3 IR(s) remained significantly higher than T_MEM, and the percentage of T_EX cells not expressing any IRs remained lower (Fig. 1f). Donor T_EX showed slightly improved polyfunctionality following 4 weeks in infection-free mice, especially donor T_EX with dual functionality. However, donor T_EX with 4 functions were significantly less frequent than T_MEM, and donor T_EX with zero or one function were increased (Fig. 1g and Extended Data 1f). Similar phenotypic and functional patterns were observed when equal numbers of T_EX and T_MEM P14 cells were co-transferred into the same Arm-immune recipient mice, controlling for any environmental effects (Extended Data 2a–2g) or when examining non-TCR-transgenic H-2D^b tetramer⁺ GP33-specific T_EX (D^b-GP33 T_EX) (Extended Data 2h–2j). Collectively, these data indicate that T_EX transferred into infection free mice displayed mixed T_EX and T_MEM features (Fig. 1h), suggesting partial recovery from exhaustion. These results are consistent with observations post-cure of human HCV infection^35–37. These donor T_EX at 4+ weeks post-transfer will hereafter be termed recovering-T_EX (REC-T_EX).

T_EX undergo transcriptional changes upon antigen elimination

T_EX have a distinct transcriptional program from T_EFF, T_MEM and naïve T cells^{4, 9, 10}; however, it is unclear how the T_EX transcriptional profile changes upon eliminating chronic antigen stimulation. Thus, we examined the transcriptional program of REC-T_EX compared to the originally transferred donor T_EX. Donor T_EX and T_MEM cells were purified on day 0 (d0) and 4 weeks post-transfer to Arm-immune recipient mice and we performed single-cell RNA sequencing (scRNAseq). A control naïve P14 population was examined in parallel to allow benchmarking known differences between naïve and T_MEM or T_EX, but was not examined further below. Donor T_MEM transcriptional profiles remained unchanged upon adoptive transfer. We therefore focused our analysis on T_EX, T_MEM, and REC-T_EX.

The majority of REC-T_EX cells clustered with T_MEM in the scRNAseq tSNE (Fig. 2a). Nine distinct clusters were identified by unsupervised clustering (Extended Data 3b), and the majority of REC-T_EX (68.1%), fell within clusters dominated by T_MEM cells (clusters 0, 2, and 4), whereas only 13.9% of REC-T_EX were within clusters with a majority of T_EX cells (clusters 1, 5, 6, and 7) (Extended Data 3c). Overall, REC-T_EX showed an intermediate transcriptional profile between T_EX and T_MEM, with only three genes unique to REC-T_EX (Extended Data 3d–3e). In REC-T_EX, expression of exhaustion-associated genes, such as Tox, Pdcd1 (PD-1) and Lag3, decreased and memory-associated genes, such as Il7r (CD127) and Tcf7 (TCF-1) increased compared to donor T_EX (Fig. 2b, and Extended Data 3f). However, Tox remained higher and Tcf7 lower in REC-T_EX compared to T_MEM. Unsupervised hierarchical clustering of only REC-T_EX scRNAseq data identified 5 clusters (Fig. 2c). These clusters fell into two main categories; the first contained REC-T_EX-Cluster 2 (REC-T_EX-C2), a cluster that overlapped exclusively with the T_EX signature and had higher expression of Il2rb (encoding CD122) and Jund (Fig. 2d). The second category included all other clusters, that displayed mixed signatures of T_EX and T_MEM (Fig. 2d and Extended Data 4a–4b). These results suggested differing degrees of transcriptional recovery among T_EX following cure of chronic infection.

Figure 2: Recovering-T_EX (REC-T_EX) are transcriptionally distinct from T_EX but intermediate between T_EX and T_MEM.

(a) tSNE plot of scRNAseq transcriptional profile for combined T_EX, T_MEM, and REC-T_EX. (b) Separate tSNE plots of T_EX, T_MEM, and REC-T_EX cells (top row), with overlay of Tox, Tcf7, and Il7r expression patterns (bottom rows). (c) tSNE plot showing unsupervised clustering of REC-T_EX cells. (d) Circos plots for transcriptional profiles of REC-T_EX clusters compared to T_EX and T_MEM signatures, REC-T_EX-Cluster 0 (REC-T_EX-C0) (top), and REC-T_EX-C2 (bottom). (e) Gene ontology (GO) analysis of differentially expressed genes between REC-T_EX (purple) and T_EX (red). −log₁₀ p-value calculated by hypergeometric distribution test. (f) Network analysis by Ingenuity® of the top differentially expressed genes (DEGs) between REC-T_EX and T_EX. (g) Pseudotime trajectory analysis by Monocle 2 for T_EX and REC-T_EX transcriptional profiles generated using top 209 DEGs from scRNAseq data. Arrows indicate suggested direction of differentiation from progenitor T_EX population (State 1), towards more terminally differentiated T_EX populations (States 3, 4 and 5), or towards REC-T_EX following antigen elimination. (h) Pseudotime plot of T_EX and REC-T_EX with overlay of the Tcf7 gene expression (i) Pseudotime plot of T_EX and REC-T_EX with overlay of the signature of genes upregulated in TCF-1+ T_EX progenitor subset from Utzschneider et al. ²⁵. (j) Volcano plot showing selected differentially expressed genes between TCF-1+ T_EX and REC-T_EX. NK, natural killer cells; IFN, interferon.

Pairwise analysis between REC-T_EX and T_EX revealed 239 differentially expressed genes (DEGs); 190 in T_EX and 49 in REC-T_EX. The top gene ontology (GO) terms enriched in REC-T_EX included T cell activation, ferroptosis, homeostasis, and peptide metabolism, whereas T_EX were enriched in signatures of inflammation, such as responses to type I and II interferons (IFNβ and IFNγ) (Fig. 2e). Ingenuity Pathway Analysis (IPA®) of the top differentially expressed genes between T_EX and REC-T_EX also showed enrichment of inflammatory pathways in T_EX. Genes associated with type I interferon responses were the top significant network enriched in T_EX compared to REC-T_EX (Fig. 2f). IPA also revealed enrichment of transcription factors Tcf7 and Zeb2 in REC-T_EX and Tox in T_EX.

To investigate developmental changes, we performed pseudotime analysis of T_EX and REC-T_EX using Monocle 2.0 ⁴⁴ (Fig. 2g). We anchored the trajectory on the transferred donor T_EX cells in state 1 (bottom branch). The predicted trajectory then bifurcated into two branches. One branch bifurcated again, containing cells within states 3 and 4, whereas the other branch contains cells in state 5. The majority of REC-T_EX cells (95.4%) were found in the state 5 branch. State 2 contained only 5 cells at the small interconnecting central branch. T_EX in state 1 had the highest Tcf7 expression likely corresponding to TCF-1⁺ T_EX progenitors^{25, 26}, whereas states 3-5 corresponded to more terminally differentiated TCF-1⁻ T_EX subset (Fig. 2h, and Extended Data 4c). Indeed, the TCF-1⁺ T_EX signature from Utzschneider et al.²⁵ was strongly enriched in state 1 (Fig. 2i), compared to the TCF-1⁻ T_EX signature that enriched in states 3-4 (Extended Data 4d). Together, these data suggested that REC-T_EX likely arise from the TCF-1⁺ progenitor T_EX subset, that persists, and possibly differentiates, following removal from chronic infection.

To investigate how TCF-1⁺ progenitor T_EX may change upon antigen removal, we compared gene expression profiles of TCF-1⁺ T_EX and REC-T_EX. Of 269 DEGs, 227 were enriched in TCF-1⁺ T_EX, including Tox, Tcf7, Slamf6, indicators of recent TCR signaling (Batf and Nr4a2), and type I IFN pathway genes. The 42 genes enriched in REC-T_EX included Il7r, Zeb2, Gzma, and beta chemokines (Fig. 2j). Thus, transcriptional comparison of TCF-1⁺ T_EX and REC-T_EX indicated that TCF-1⁺ T_EX likely undergo further differentiation upon removal from antigen. Additionally, TCF-1⁺ REC-T_EX also displayed 157 DEGs compared to TCF-1⁺ T_EX (Extended Data 4e), further supporting the notion of TCF-1⁺ T_EX differentiation of upon removal from antigen.

TCF-1⁺ T_EX selectively survive and differentiate to REC-T_EX

In the studies above, and in previous work^37,45, the contribution of selective survival versus differentiation of T_EX subsets was unclear. To address this question, we sorted T_EX subsets using Ly108 as a surrogate for TCF-1 expression. Sorting Ly108⁺PD-1^int and Ly108⁻PD-1^hi T_EX allowed high purity separation of TCF-1⁺ T_EX and TCF-1⁻ T_EX, respectively (Fig. 3a). We then adoptively transferred equal numbers of progenitor TCF-1⁺ T_EX (Prog-T_EX) and terminal TCF-1⁻ T_EX (Term-T_EX) into Arm-immune mice and examined persistence and differentiation upon removal from antigen (Fig. 3a, right panel). The number of donor cells recovered from the TCF-1⁺ T_EX was ~10-fold higher than for the TCF-1⁻ T_EX subset after 4 weeks post-transfer (Fig. 3b). This result demonstrates that at least some changes to T_EX upon antigen elimination arise from selective survival of T_EX TCF-1⁺ population.

Figure 3: Differentiation of T_EX into REC-T_EX is associated with selective survival of the progenitor TCF-1⁺ T_EX subset.

(a) Left, dot plots showing sorting purity of Ly108⁺PD-1^int (surrogate for TCF-1⁺) progenitor T_EX (Prog-T_EX) and Ly108⁻PD-1^hi (i.e. TCF-1⁻) terminal T_EX (Term-T_EX). Middle, representative histograms of TCF-1 expression on sorted subsets. Right, experimental design for co-transfer of ~1.4x10⁵ sorted P14 cells of each congenically distinct Prog-T_EX and Term-T_EX subsets into infection-free recipient mice. (b) Left, representative dot plots for the frequency of donor Prog-T_EX and Term-T_EX P14 cells, on the day of adoptive transfer (d0) and ~d30 post-transfer. Middle, percentage of surviving P14 cells for Prog-T_EX and Term-T_EX on d30 post-transfer compared to d0 (estimating 10% take). Right, absolute number of P14 cells per spleen for each subset on d30 post-transfer. (c) Left, representative dot plots of CD127 and PD-1 expression. Right, percentages of CD127⁺ and PD-1⁺CD127⁻ P14 cells for both subsets on d0 and d30 post-transfer, and absolute numbers of CD127⁺ P14 cells per spleen on d30 post-transfer. (d) Left, representative histograms for expression of PD-1 or CD127 on Prog-T_EX and Term-T_EX P14s on d0 (adoptive transfer day), and on d30 post-transfer compared to endogenous CD8+ GP33 tetramer+ T_MEM of recipient mice. Right, plots of PD-1 and CD127 MFI on both subsets. Data (b-d) n=2 biologically independent experiments, 2 donor groups/experiment for d0 and 3-5 mice/experiment for d30. (e) Left, representative histograms for levels of expression of transcription factors Tox, Eomes, or TCF-1 d30 post-transfer, compared to CD8+ naïve T cells. Right, plots of Tox, Eomes, and TCF-1 MFI on both subsets. (f) Left, representative dot plot of CD39 and PD-1 expression. Right, percentages of co-expression of 6 IRs on P14s d30 post-transfer. Data (e-f) from one experiment, 5 mice. All analyses for Figure 3, for co-transferred cells performed by paired two-tailed t-test (Wilcoxon matched-pairs signed rank test). Comparisons of d0 versus d30 analyzed using unpaired two-tailed Student’s t-tests (Mann-Whitney test). *p<0.05, **p<0.01.

Additional changes occurred within donor TCF1⁺ Prog-T_EX, since only ~8% of donor Prog-T_EX population was CD127⁺ at transfer, whereas the REC-T_EX were ~80% CD127+, and the MFI of CD127 was significantly increased (Fig. 3c,d). PD-1 expression also decreased on the donor TCF1⁺ T_EX and the frequency of PD-1⁺CD127⁻ REC-T_EX was reduced ~10-fold (from 84% to 8%) (Fig. 3c,d). Conversely, the progeny of Term-T_EX displayed less differentiation (Fig. 3c,d). Rather, these Term-T_EX sustained higher expression of Tox and Eomes, lower TCF-1 expression (Fig. 3e), and higher co-expression of multiple IRs in the absence of persisting antigen (Fig. 3f). Similar results were obtained when subsets were purified using a Blimp1-YFP reporter to obtain Blimp1-YFP⁻ Prog-T_EX and Blimp1-YFP⁺ Term-T_EX²⁷ (Extended Data 5). Together, these data suggested that REC-T_EX originated from the progenitor TCF-1⁺ T_EX or even a subpopulation within the progenitors, with perhaps additional changes in key pathways, including IL-7R expression, consistent with predictions of the scRNAseq data.

REC-T_EX recall capacity remains compromised compared to T_MEM

The data above indicate that upon eliminating chronic antigenic stimulation T_EX (re)acquired some phenotypic and transcriptional features of T_MEM. We next sought to examine whether REC-T_EX also recover a key functional property of T_MEM, which is the ability to mount potent recall responses upon re-encountering their cognate antigen¹. We sorted REC-T_EX P14 cells and adoptively transferred these cells into new congenic naïve mice together with an equal number of either congenic T_EX or T_MEM. Recipient mice were challenged with LCMV-Arm and the recall response of each donor population was examined 8 days later (Fig. 4a and Extended Data 6a–6b). The REC-T_EX underwent slightly better expansion compared to T_EX which largely failed to expand upon rechallenge (Extended Data 6c). However, these REC-T_EX were >5-fold less efficient in expansion capacity compared to co-transferred T_MEM, indicating that any recovery that occurred in REC-T_EX failed to completely restore memory function (Fig. 4b). Moreover, whereas only ~16.6% (6-33%) of the secondary T_EFF derived from T_MEM were PD-1⁺CD127⁻, secondary T_EFF derived from REC-T_EX were 63% (37-77%) PD-1⁺CD127⁻. Secondary T_EFF derived from REC-T_EX also expressed 3-4 times the amount of PD-1 per cell compared to secondary T_EFF from T_MEM (Fig. 4c,d). Conversely, KLRG1, a marker of robust T_EFF generation, was substantially lower in frequency and MFI on T_EFF from REC-T_EX compared to those derived from T_MEM (Fig. 4c,d). Similar results were obtained in different organs (Extended Data 6d). Consistent with these observations, REC-T_EX-derived T_EFF had a substantially lower percentage of polyfunctional T_EFF than T_MEM derived cells, and inferior cytotoxic capacity (Fig. 4e and Extended Data 6e–6f). After clearance of rechallenge infection, progeny of REC-T_EX were present in similar numbers to T_MEM progeny at d30 p.i. (Fig. 4f). However, REC-T_EX progeny had higher TCF-1, CD127 and CD62L, though these cells also retained more PD-1 and Tox (Fig 4f and Extended Data 6g), suggesting that although some memory-like properties were recovered after rechallenge, there may still be persisting features of the history of exhaustion. Despite the rapid upregulation of PD-1 on REC-T_EX derived secondary T_EFF, blockade of the PD-1 pathway during rechallenge provided only modest benefit and there was no increase in absolute cell number (Fig. 4g–I and Extended Data 6h–k). Thus, despite a slight improvement in secondary expansion of REC-T_EX compared to T_EX upon challenge, REC-T_EX remained significantly inferior to T_MEM even with PD-1 pathway blockade, indicating lack of full recovery of memory properties. Moreover, secondary T_EFF generated from REC-T_EX were phenotypically and functionally compromised compared to secondary T_EFF generated from T_MEM with higher PD-1, lower KLRG1 and suboptimal effector function.

Figure 4: REC-T_EX possess inferior recall capacity compared to T_MEM that is only partially improved with PD-L1 blockade.

(a) Experimental design. Equal number of sorted REC-T_EX and T_MEM was co-transferred into congenic naïve mice, then challenged with LCMV-Arm. Donor cells were examined on d8 post-challenge. (b) Absolute number of P14 cells per spleen on d8 post challenge. (c) Left, representative dot plots of CD127 and PD-1, and CD127 and KLRG1 expression. Right, percentages of different subsets. (d) Left, histograms for expression of PD-1 or KLRG1 for P14 cells. Right, markers’ expression as ΔMFI (compared to endogenous naïve cells). (e) Left, representative dot plots of cytokine production (IFNγ and TNF) and CD107a expression on P14 cells upon GP33-peptide re-stimulation. Right, percentage of polyfunctional REC-T_EX and T_MEM P14 cells. Data (b-e) n=2 biologically independent experiments, 4-5 mice/experiment. (f) Co-transferred REC-T_EX and T_MEM P14s at d30 post-challenge. Representative dot plots of CD127 and CD62L, TCF-1 and Tim-3 expression, and histograms of TCF-1 expression. To the right of dot plots, percentages and absolute numbers of the different P14 subsets, and TCF-1 expression. N=2 biologically independent experiments, 2-5 mice/experiment. (g) Experimental design for LCMV-Arm challenge of co-transferred REC-T_EX and T_MEM, treated with anti-PD-L1 or PBS. (h) Left, percentage of REC-T_EX cells of total P14 cells. Right, fold expansion of REC-T_EX and T_MEM P14 cells. Data presented as mean +/− SD. Unpaired two-tailed Student’s t-tests (Mann-Whitney test). **p<0.01. (i) Left, representative dot plots of CD127 and PD-1, and CD127 and KLRG1 expression in control PBS-treated (top) or anti-PD-L1-treated (bottom) groups. Right, percentages of indicated subsets. (j) Left, histograms for PD-1 or KLRG1 expression on d8 post challenge. Right, expression as MFI. (k) Left, representative dot plots of Tox and PD-1 expression. Right, plots for percentages of Tox⁺PD-1⁺ P14 cells. (l) Left, representative histograms for granzyme B (GzmB) expression. Right, GzmB expression as ΔMFI. Data for (h-l) n=2 independent experiments, 2-4 mice/group/experiment. All analyses for Figure 4, for co-transferred cells by paired two-tailed t-test (Wilcoxon matched-pairs signed rank test). PBS versus αPD-L1 (h-l) analyzed using unpaired two-tailed Student’s t-tests (Mann-Whitney test). *p<0.05, **p<0.01.

Epigenetic scars persist from REC-T_EX exhaustion history

We next investigated the epigenetic state of REC-T_EX by performing transposase accessible chromatin with high-throughput sequencing (ATACseq)⁴⁶ (Extended Data 7a,b). ATACseq analysis of REC-T_EX, T_EX, and T_MEM, using Uniform Manifold Approximation and Projection (UMAP) suggested a chromatin accessibility landscape of REC-T_EX intermediate between T_EX and T_MEM (Fig. 5a). Pearson correlation suggested more similarity in ATACseq signatures between REC-T_EX and T_EX samples than between REC-T_EX and T_MEM samples (mean correlation 116.8 versus 143.7, respectively) (Extended Data 7c). Including naïve P14 cells in Principal Component Analysis (PCA) also suggested that REC-T_EX had a chromatin accessibility profile more similar to T_EX than T_MEM (Extended Data 7d). Pairwise comparisons of open chromatin regions (OCRs) in REC-T_EX, T_EX and T_MEM confirmed that REC-T_EX were more similar to T_EX than to T_MEM (Fig. 5b–d). The number of differential OCRs between REC-T_EX and T_EX was ~1 order magnitude lower than between REC-T_EX versus T_MEM (182 and 1586 OCRs, respectively) (Fig. 5b). Nevertheless, of the 147 OCRs more open in REC-T_EX compared to T_EX (UP_REC-T_EX_vs T_EX) 136 (92.5%) overlap with those more accessible in T_MEM compared to T_EX (Extended Data 7e), suggesting that some changes in REC-T_EX do overlap with features of T_MEM. However, of the 666 OCRs that were significantly more accessible in REC-T_EX versus T_MEM (UP_REC-T_EX_vsT_MEM), 526 OCRs (79%) were also more accessible in T_EX compared to T_MEM, suggesting that the majority of differences between the REC-T_EX and T_MEM OCR landscapes were residual open chromatin signatures from T_EX or “scars” of exhaustion (Fig. 5c).

Figure 5: REC-T_EX remain epigenetically scarred from their exhaustion history.

(a) Uniform Manifold Approximation and Projection (UMAP) of chromatin accessibility profiles for T_EX, T_MEM, and REC-T_EX P14 cells using ATACseq. (b) Number of differential OCRs between REC-T_EX versus T_MEM or T_EX using pairwise analysis by DESeq2. (c) Venn diagram of overlap in OCRs significantly more open in REC-T_EX compared to T_MEM (UP_REC-T_EX_vs T_MEM) and significantly more open in T_EX compared to T_MEM (UP_T_EX_vs T_MEM). (d) Heatmap of differential peaks corresponding to OCRs significantly different between T_EX, T_MEM, and REC-T_EX. Calculated by pairwise analysis using DESeq2. (e) ATACseq tracks at selected genes (Tox, Il7r, Pdcd1, and Klrg1) from naïve, T_MEM, T_EX, and REC-T_EX samples (top to bottom). Shaded boxes indicate peaks with significantly different accessibility between REC-T_EX and T_EX. Open boxes indicate peaks that are not significantly changed between REC-T_EX and T_EX, but are significantly different from T_MEM. The red arrow indicates the Pdcd1 -23kb enhancer.

Further analysis revealed open chromatin changes that were highly biased to REC-T_EX compared to T_EX or T_MEM. However, this cluster (cluster 1, C1) was small, and only 149 OCRs were significantly more accessible in REC-T_EX compared to T_EX and T_MEM. These OCRs included sites near Fasl, Ezh2, Il12r and Id2 (Fig. 5d). Specific clusters of REC-T_EX gene loci that overlapped with T_MEM (REC-T_EX and T_MEM), contained 107 OCRs (cluster C4) with OCRs mapping to genes including Il7r, Il21, Cd226 and Il12rb2. However, the vast majority of T_MEM associated OCRs (n=3242; C5) were not changed in REC-T_EX (Fig. 5d). The OCR pattern in REC-T_EX predominantly resembled that of T_EX, as indicated by the cluster of gene-associated OCRs that remained highly similar between T_EX and REC-T_EX (Fig 5d; cluster 2 of 525 OCRs) including regions near Eomes, Pdcd1, Tox2, Cd101, and Nr4a2.

Some of these changes in individual genes provided insight into mechanisms. For example, an OCR upstream of Il7r became more accessible in REC-T_EX, reflecting a more T_MEM pattern (Fig. 5e). In contrast, in REC-T_EX the accessibility pattern of the Tox locus, a locus that is heavily remodeled in exhaustion¹⁷, remained more similar to T_EX (Fig. 5e). Moreover, accessibility of OCRs associated with Pdcd1 (encoding PD-1) were not significantly changed in REC-T_EX compared to T_EX (Fig 5e), including the well characterized exhaustion-specific enhancer in Pdcd1 at ~23 kb^{11, 12}. A similar OCR stability was observed for other loci such as Klrg1 (Fig. 5e), Batf, Tox2, Eomes, and Stat3 (Extended Data 7f).

To test whether the OCR pattern of REC-T_EX could be explained by persistence of the OCR pattern of TCF-1⁺ Prog-T_EX cells, we compared the chromatin accessibility profile of REC-T_EX to the profiles of TCF-1⁺ and TCF-1⁻ T_EX subsets from Jadhav et al.³². However, REC-T_EX were positioned separately from the TCF-1⁺ T_EX subset and slightly towards T_MEM (Extended Data 7g). We then examined the 182 OCRs that were changed in REC-T_EX compared to T_EX. These OCRs were evenly distributed between the TCF-1⁺ T_EX and TCF-1⁻ T_EX subsets (Extended Data 7h). A similar pattern was observed using a different T_EX subset ATAC-seq dataset ²⁹ (Extended Data 7i). These analyses suggest that, although some chromatin accessibility changes in REC-T_EX reflect features of the Prog-T_EX subset, the global pattern of OCR changes in REC-T_EX are due to contributions from Prog-T_EX biology as well as other changes likely occurring as these cells differentiate in the absence of antigen.

Epigenetic scars in REC-T_EX limit core memory circuits

To identify the transcriptional circuitry associated with this epigenetic “scarring” of T_EX, we performed integrated analysis of scRNAseq and ATACseq data. We first asked what fraction of genes differentially expressed between REC-T_EX and T_EX were associated with changes in OCRs. Of the 190 genes that were significantly upregulated in T_EX compared to REC-T_EX, only 10 (5.2%) were associated with significantly altered chromatin accessibility (Fig. 6a). Similarly, of the 49 genes significantly upregulated in REC-T_EX, only 4 (8.1%) were associated with altered accessibility (Fig. 6a). Thus, the vast majority, 225/239 genes (~94%), of transcriptional changes between T_EX and REC-T_EX were not associated with a change in the open chromatin pattern of that gene. These observations suggest that although the transcriptional and protein expression patterns of REC-T_EX shifted towards a T_MEM-like profile, the underlying cellular identity defined by OCR patterns and, thus, potential for future behavior, remained T_EX-like.

Figure 6: Open chromatin landscape of REC-T_EX hinders access to T_EFF and T_MEM transcriptional circuitry.

(a) Total number of DEGs between REC-T_EX and T_EX from scRNAseq. The fraction of DEGs associated with differential accessibility in OCRs from ATACseq are represented by the white portion of the bar. (b) Heatmap of the 182 differentially accessible OCRs in REC-T_EX versus T_EX. (c) Changes in gene expression associated with the differentially accessible OCRs in REC-T_EX versus T_EX. (d) Transcription factors with significantly enriched motifs in OCRs in REC-T_EX versus T_EX (top half, purple circles) or in T_EX versus REC-T_EX (bottom half, red circles). Pairwise analysis by Homer. (e) Overlap of the transcription factor binding sites (TFBS) in OCRs between REC-T_EX, T_EX, and T_MEM, calculated by Taiji for TCF-4, T-bet and CEBP. (f) Fold difference in the TFBS overlap between REC-T_EX and T_EX divided by overlap between REC-T_EX and T_MEM for top differential TFs identified by HOMER in (d) and Extended Data 8c. (g) Left, heatmap of the PageRank analysis of TFs with the most significant differences in chromatin accessibility of their TFBS among the three groups. Right, fold difference in the TFBS overlap between REC-T_EX and T_EX divided by overlap between REC-T_EX and T_MEM for the identified TFs. (h) Heatmap of the top 100 OCRs showing differential accessibility in REC-T_EX versus T_EX from the T_EFF signature by Pauken et al. ¹¹.

Nevertheless, we analyzed OCRs with differential accessibility between REC-T_EX and T_EX (Fig. 6b) and examined gene expression from the scRNAseq data set associated with these OCRs. The top upregulated genes associated with differentially accessible chromatin regions for REC-T_EX compared to T_EX were Il7r and Emb (Fig. 6c). Some peaks with increased accessibility were associated with genes that had decreased expression in REC-T_EX compared to T_EX (Fig 6c) suggesting potential negative regulatory elements. These included Cd226 and Tox. Of the 140 OCRs that were more accessible in REC-T_EX compared to T_MEM (see Fig. 5c), only a handful (e.g. Zfp362) showed increased expression of the associated gene (Extended Data 8a). Applying the genomic regions enrichment of annotations tool (GREAT), to the comparison between REC-T_EX and T_MEM revealed predicted biological pathways, including signatures associated with T cell activation and differentiation, and regulation of cell death in REC-T_EX (Extended Data 8b). Biological pathways enriched in T_MEM included cell cycle pathways, epigenetics, and regulation of retinoic acid receptor signaling (Extended Data 8b). In contrast, no distinct biological processes were associated with OCR differences between REC-T_EX and T_EX.

HOMER analysis showed the top TFs with enriched motifs in REC-T_EX compared to T_EX were members of the TCF family (TCF-3 and TCF-4), members of the T-box family (T-bet and Tbx5), and ERG (Fig. 6d). Compared to T_MEM, some of the top TFs enriched in REC-T_EX were BATF, AP-1, and Eomes (Extended Data 8c). Compared to REC-T_EX, T_EX were enriched for motifs containing BATF and CEBP binding sites, whereas T_MEM were enriched for Runx1 and 2 binding sites (Fig 6d and Extended Data 8c). Using Taiji, we compared TF binding site (TFBS) overlap between REC-T_EX and either T_EX or T_MEM for the top differential TFs identified above (Fig. 6e). For all of the top 9 TFs with enriched TFBS in REC-T_EX, there was more overlap with T_EX than T_MEM (Fig. 6f).

Next, using PageRank analysis, we identified TFs that had the most significant differences in chromatin accessibility within their TFBS (Fig. 6g). For the top 34 TFs identified, all but two (Tcf7 and Hsf1) showed greater overlap of TFBS between REC-T_EX and T_EX than between REC-T_EX and T_MEM (Fig. 6g). For Tcf7, 77 TFBS characteristic of T_MEM (i.e. not predicted for T_EX) were (re)acquired by REC-T_EX, whereas 47 remained from their exhaustion history. However, for Hic1 and Zeb1, two of the top TFs identified by page rank analysis for REC-T_EX, less TFBS were shared with T_MEM (83 and 71) than with T_EX (152 and 92, respectively) (Extended Data 8d). Together, these analyses suggest that the epigenetic landscape in which these TFs operate remains scarred from the exhaustion history.

Finally, we examined whether the OCR landscape associated with T_EFF might change upon antigen removal. Thus, we identified 545 open chromatin regions enriched for accessibility in T_EFF from published work¹¹. Although there was some change in accessibility at these T_EFF-associated OCRs upon recovery of T_EX (i.e. T_EX versus REC-T_EX; log₂ fold change >0.25) only 7 OCRs were statistically different between T_EX and REC-T_EX (p-value <0.05) (Fig. 6h). These data indicate a similar, and relatively inflexible pattern of effector gene accessibility, perhaps limiting the re-engagement of the effector program upon rechallenge infection (noted in Fig. 4). Mechanistically, Tox is a key TF that drives epigenetic induction of exhaustion and represses development of the OCR landscape of T_EFF¹⁷. Thus, we examined the T_EFF OCRs that failed to open because of Tox activity in T_EX. Of the 2024 “Tox-repressed” OCRs normally found in T_EFF only 58 were differentially accessible in REC-T_EX compared to T_EX (Extended Data 8e). In other words, only ~2.9% of the non-T_EX OCR landscape associated with T_EFF biology was restored in REC-T_EX. Thus, recovery from exhaustion upon removal of antigen stimulation was not associated with a robust or global improvement access to T_EFF genes. This finding is consistent with the poor REC-T_EX recall responses and suboptimal differentiation into secondary T_EFF upon antigen re-exposure compared to T_MEM.

Thus, integrated transcriptional and epigenetic analyses suggested that incomplete functional recovery of T_EX upon eliminating antigen stimulation is largely due to a lack of epigenetic plasticity and a failure to (re)acquire an epigenetic landscape permissive to the core transcriptional T_MEM and T_EFF circuits (Extended Data 9).

DISCUSSION

Reinvigoration of T_EX by checkpoint blockade highlights the reversibility of some aspects of exhaustion^47,2. However, effector properties are only partially and temporarily regained following PD-1:PD-L1 pathway blockade, and the exhaustion OCR landscape remains largely unchanged¹¹. It remains unclear whether non-immunological cure of chronic infections will allow T_EX to (re)-acquire T_MEM differentiation or increase effector potential. To address this question we dissected the program of T_EX following elimination of chronic antigen stimulation. These REC-T_EX generated after experimental “cure” of chronic infection downregulated some features of exhaustion and appeared phenotypically and transcriptionally more memory-like. Despite the appearance of recovery, however, REC-T_EX remained compromised in the ability to respond to reinfection compared to bona fide T_MEM. Profiling chromatin accessibility of REC-T_EX revealed minimal remodeling of the T_EX epigenetic landscape, suggesting that lack of epigenetic plasticity is a major impediment to restoring optimal T_MEM or T_EFF biology.

Approximately 90% of T_EX cells persisting ~1 month after eliminating antigen originated from the TCF-1⁺ T_EX progenitor subset. These observations agree with previous observations in a mouse model of melanoma⁴⁵, and in HCV patients following virological cure³⁷. However, in addition to simply selective survival of this TCF-1⁺ subset of T_EX, phenotypic and transcriptional changes in these REC-T_EX cause them to appear more T_MEM-like. A key change in REC-T_EX upon cure may be the IL-7R-TCF-1 axis. Indeed, upregulation of CD127 protein and Il7r gene expression was associated with increased chromatin accessibility in some regions of the Il7r locus. The increase in CD127 expression by REC-T_EX, especially the TCF-1+ subset, provides a potential mechanism for survival of REC-T_EX and may reveal therapeutic opportunities. Future studies are needed to define whether these cells simply survive or also acquire T_MEM-like homeostatic self-renewal.

Studies in HCV infection have shown that upon virological cure, previously exhausted HCV-specific CD8 T cells can persist and respond to reinfection^{37, 48}. It has been unclear, however, how the recall responses of previously exhausted CD8 T cells compare to bona fide T_MEM. By comparing REC-T_EX to T_MEM directly in a co-transfer setting, we were able to control for key confounding variables since both cell types were present at the same numbers, in the same inflammatory environment and responding to the same level of antigen throughout the reinfection. Under these conditions, re-expansion of REC-T_EX was suboptimal and the responding cells were incapable of generating appropriately differentiated secondary T_EFF. Indeed, REC-T_EX gave rise to cells biased towards T_EX rather than T_EFF upon rechallenge. It should be noted, however, that REC-T_EX were not devoid of any recall capacity and might, in some settings such as chronic infections, provide some degree of protection. Moreover, it is possible that longer periods of antigen-free recovery could improve REC-T_EX differentiation, though even after >500 days, REC-T_EX retain higher Tox. Nevertheless, these observations make several key points. First, when compared on a per cell basis REC-T_EX were inferior to T_MEM for critical properties, such as response to reinfection. Second, the phenotypic and transcriptional similarity of REC-T_EX to T_MEM does not predict future T cell responses. Third, REC-T_EX rapidly upregulate PD-1 upon rechallenge infection, but PD-1 expression alone does not explain inferior recall responses. It is possible that in other settings of more protracted or chronic rechallenge infections or cancer recurrence there may be more benefit to blocking the PD-1 pathway. Nevertheless, these data support the notion of broad, pervasive and durable imprinting of the T_EX program that is at least partially retained in REC-T_EX.

T_EX are epigenetically divergent from T_EFF and T_MEM, suggesting that T_EX are a distinct state or differentiation branch of mature CD8 T cells^{11–13, 16}. This T_EX open chromatin landscape remains largely fixed upon PD-1 pathway blockade¹¹. However, it has been unclear whether T_EX open chromatin landscape remodeling would occur upon “cure” of chronic infections. Our data, rather, find a highly stable open chromatin landscape following removal of chronic antigen stimulus. Indeed, chromatin accessibility in the regulatory regions of many exhaustion genes, like Tox and Pdcd1 remained similar to the original T_EX. The consequence of these patterns was a more rapid and robust upregulation of exhaustion-related gene-products, like PD-1, upon rechallenge. In addition, the accessibility of the overall T_EFF open chromatin landscape remained essentially unchanged following antigen elimination. This effect also extends to poor ability to access T_EFF transcriptional circuits upon reinfection, resulting in limited effector function and reduced expression of canonical T_EFF protein markers, such as KLRG1. One consequence of this epigenetic stability is that REC-T_EX maintain a pattern of transcription factor binding site (TFBS) accessibility that remains largely unchanged from T_EX. Aside from the notable, and likely functionally important, changes in accessibility of Tcf7 TFBS, much of the potential for different TF to function remains hardwired to the T_EX cellular identity. It is possible that coaxing changes in the epigenetic landscape through epigenetic or immunological drugs, might enhance changes in the chromatin landscape upon cure of chronic diseases. However, after a cure that mimics a non-immunological intervention, such as highly effective antiviral treatment, “scars” of the exhaustion epigenetic program remain and strongly influence the future responses of REC-T_EX.

Antiviral cures for chronic LCMV infection do not exist, necessitating these adoptive transfer approaches. However, companion studies from human HCV infection (Tonnerre et al. ⁴⁹, and Sen et al. ⁵⁰) and other recent work ⁵¹, using direct acting antiviral treatment cures chronic viral infection, are consistent with the findings in the LCMV system. It will be interesting in the future to combine immunological and/or epigenetic therapies with cure approaches to potentially enhance the quality and durability of persisting immune memory. Such approaches may also be relevant in cancer where some types of tumors are eradicated, but relapse and recurrence remain an impediment to true long-term cures. However, our studies identify epigenetic “scars” of T cell exhaustion that persist even after cure of chronic infection. These data highlight the need to understand epigenetic plasticity of T_EX more fully and also point to opportunities to improve immunity following cure of chronic disease. Moreover, these findings provide insights into how to assess and potentially manipulate the formation of long-term CD8 T cell immunity following cure of chronic infections and cancer.

METHODS

Mice.

Female C57BL/6 mice (CD45.2+) from Charles River, US National Cancer Institute (NCI) were used for infections at 5-7 weeks of age for all experiments. The numbers of mice used in each experiment are mentioned in the corresponding figure legend. P14 mice transgenic for a TCR recognizing the H-2D^b GP33–41 epitope of LCMV were bred at the University of Pennsylvania (Penn). Blimp-1-yellow fluorescent protein (YFP) reporter mice from Eric Meffre, Yale University, were bred to P14 to generate P14 Blimp-1-YFP mice. All mice were kept on an NCI C57BL/6 background. Mice used in the experiments were age- and sex-matched. Mice were housed in an animal facility at Penn at ~20°C (68°F) with humidity at ~55%, and the dark/light cycle was 12h/12h. All mouse use, experiments, protocols, and breeding conditions were in accordance with Institutional Animal Care and Use Committee (IACUC) guidelines for the University of Pennsylvania, and are in compliance with the ethical guidelines of the University of Pennsylvania that comply with the US national and International guidelines.

LCMV infections.

For acute viral infections, mice were injected intraperitoneally (i.p) with 2x10⁵ plaque-forming units (PFU) of LCMV-Armstrong (LCMV-Arm). For chronic viral infections, mice were injected intravenously (i.v.) with LCMV-Cl13 (4x10⁶ PFU). Mice were infected one day post adoptive transfer of P14 cells. LCMV strains were grown in BHK cells (ATCC, CCL-10) and titrated on VERO cells (ATCC, CCL-81) using plaque assay as previously described ⁴².

Cell preparation.

Single-cell suspensions of splenocytes were generated by mechanically homogenizing spleens using a 70μm cell strainer. Cells were washed and red blood cells lysed using ACK lysis buffer (ThermoFisher). PBMCs were isolated using Histopaque®-1083 gradient (Sigma-Aldrich). When necessary, cells were sorted with a FACS Aria II (BD Biosciences) to >92% purity. Before sorting, CD8 T cells were enriched from total splenocytes using the EasySep^tm CD8 T cell isolation Kit (StemCell) (routinely >85% purity). CD8 T cell-enrichment was also performed before adoptively transferring P14 cells to recipient mice. All in vitro T cell assays were performed using complete RPMI medium (cRPMI): RPMI-1640 (Corning/Mediatech), 10% fetal bovine serum (FBS) (ThermoFisher), 1% HEPES (ThermoFisher), 1% penicillin/streptomycin (ThermoFisher), 1% L-glutamine, and 0.1% β-mercaptoethanol (Sigma-Aldrich).

Adoptive transfers.

For primary infections, recipient mice were engrafted with 1x10³ naïve splenic P14 cells one day before infection with LCMV. For persistence experiments, total CD8+ T cells containing ~1x10⁶ congenic P14 cells (separate transfer experiments, Figure 1 and Extended Data 1), or ~1.6x10⁵ D^b-GP33-specific CD8 T cells (Extended Data 2h–j), or ~2x10⁵ P14 cells of each T_EX and T_MEM P14 cells on distinct congenic backgrounds (co-transfer experiments, Extended Data 2a–g) were adoptively transferred into LCMV-Arm-immune recipient mice (mice that cleared a previous LCMV-Arm infection >3 weeks prior to the adoptive transfer). For subset-transfer experiments, an equal number of Prog-T_EX and Term-T_EX P14 cells on distinct congenic backgrounds were co-adoptively transferred into LCMV-Arm-immune congenic recipient mice, ~1.4x10⁵ cells of each subset sorted on Ly108 and PD-1 expression (Figure 3), or 1.2x10⁵ P14 cells sorted on Blimp1-YFP expression (Extended Data 5). The day of adoptive transfer day is considered d0 for all persistence experiments (Figure 1 and 3, Extended Data 1, 2, and 5). For rechallenge experiments, REC-T_EX P14 cells were co-transferred with an equal number of congenically distinct T_MEM or T_EX, 1-1.5x10³ of each population were transferred into naïve recipient mice.

Antibodies, flow cytometry, and cell sorting.

The following fluorochrome-conjugated antibodies were used (clone mentioned in parentheses): from Biolegend: anti-CD8–BV650, Clone (53-6.7), dilution 1:200; anti-CD44–BV785 (IM7), 1:300; anti-CD44–BV785 (IM7), 1:300; anti-CD45.1–A700 (A20), 1:200; anti-CD45.1–BV605 (A20), 1:200; anti-CD45.1–BV785 (A20), 1:200; anti-CD45.1–FITC (A20), 1:200; anti-CD45.1–PB (A20), 1:200; anti-CD45.2–A700 (104), 1:200; anti-CD45.2–BV421 (104), 1:200; anti-CD45.2–BV605 (104), 1:200; anti-CD45.2–BV785 (104), 1:200; anti-CD62L–BV605 (MEL14), 1:200; anti-CD69–PE-Cy5 (H1.2F3), 1:200; anti-CD107a–A488 (1D4B), 1:500; anti-CD107a–A647 (1D4B), 1:500; anti-CD127–PE-Dazzle594 (A7R34), 1:200; anti-CD160–PE (7H1), 1:100; anti-PD-1–PE-Cy7 (RMP1-30), 1:200; anti-CX3CR1–APC (CX3CR1-173), 1:200; anti-Ly108–PB (330-AJ), 1:200; anti-T-bet–BV421 (4B10), 1:100; anti-TIGIT–PE/Dazzle594 (1G9), 1:200; anti-TIM3–BV605 (RMT3-23), 1:100; and anti-TNF–BV421 (MP6-XT22), 1:100; from eBioscience/Thermo Fisher Scientific: anti-CD4–APC-eFlour780 (RM4-5), 1:200; anti-CD27–APC-eFlour780 (LG.7F9), 1:200; anti-CD39–eFlour660 (24DMS1), 1:200; anti-CD45.2–FITC (104), 1:200; anti-CD45R/B220–APC-eFlour780 (RA3-6B2), 1:200; anti-CD62L–eFlour450 (MEL14), 1:200; anti-2B4–FITC (eBio244F4), 1:100; anti-Eomes–PE (Dan11mag); 1:100; anti-IFN-γ–PE-eFlour610 (XMG1.2), 1:100; anti-IL-2–FITC (JES6-5H4), 1:100; anti-Lag3– (eBioC9B7w), 1:100; and anti-TNF–PE-Cy7 (MP6-XT22), 1:100; from BD Biosciences: anti-CD127–PE-CD594 (SB/199), 1:200; anti-CD244.2/2B4–FITC (2B4), 1:100; anti-IFN-γ–A700 (XMG1.2), 1:100; anti-IL-2–APC (JES6-5H4), 1:100; anti-TCF1/TCF7–PE (S33-966), 1:100; anti-Ki67–A488 (B56), 1:50; and anti-Ki67–BV786 (B56), 1:50; from R&D Systems:anti-CCL3/MIP-1–APC (39624), 1:50; and anti-CCL3/MIP-1–PE (39624), 1:50; from invitrogen: anti-Eomes–PE-eFlour610 (Dan11mag); 1:100; and anti-granzyme B–PE (GB12), 1:50; from Miltenyi: anti-TOX–APC (REA473), 1:100; and from Southern Biotech: anti-KLRG1–FITC (2F1), 1:150. Tetramers of MHC class I H-2D^b GP33 were obtained from the National Institute of Health (NIH) Tetramer Core, dilution 1:300/1:400. Dead cells were excluded using Live/Dead Fixable Aqua Dead Cell Stain Kit, from invitrogen/ThermoFisher, dilution 1:600.

Cells were sorted using a 70 μm nozzle in RPMI 50% FBS maintained at 4°C using a circulating cool-down system. Purity for all samples presented in the figures was >92% (mean 97.9%). The following gating strategies were used for sorting: P14 cells were sorted based on GP33-tetramer+ live CD8+ congenic marker+ (CD45.1, CD45.2). For Blimp1-YFP subsets, progenitor T_EX and terminal T_EX were sorted based on YFP⁺ and YFP⁻, respectively. Flow cytometry data were acquired on a BD LSR II instrument. Cell sorting was performed on a BD FACS-Aria. Data were analyzed using FlowJo software V10.1r7 (FlowJo, LLC).

Staining for flow cytometry.

Surface staining; cell suspensions were surface stained with indicated antibodies for 30 min at 4°C. Dead cells were excluded using Live/Dead (1/600, ThermoFisher) added concurrently with surface antibodies. Intracellular staining was conducted at 4°C for 1 hour. Intracellular cytokine staining; evaluation of in vitro cytokine production was performed by incubating 2x10⁶ cells in 96-well plates for 5h at 37^oC 5% CO₂ in cRPMI supplemented with GolgiStop (1/250; BD bioscience), GolgiPlug (1/500; BD bioscience), and CD107a antibodies (1/500). The cells were re-stimulated with GP33 peptide 0.4 μg/ml (GenScript) to test specific production of cytokines by P14 cells, or re-stimulated with GP276 peptide 0.4 μg/ml (GenScript), or left unstimulated as controls. The cells were stained for surface antigens, then fixed and stained for intracellular cytokines using the BD Fixation/permeabilization kit (BD Bioscience) according to manufacturer’s instructions. Intracellular and nuclear staining of transcription factors; following staining for surface antigens, FoxP3 staining kit (eBioscience/ThermoFisher) was used to stain TFs according to manufacturer’s instructions.

Blockade of PD-1/PD-L1 pathway.

PD-L1 blockade was performed by i.p. injections of 200μl of PBS or PBS containing 200mg rat anti-mouse PD-L1 monoclonal antibody, clone 10F.9G2, from BioXcell, Catalog# BE0101. The injections were performed on days 1, 4, and 7 post-infection with LCMV-Arm.

Cytotoxicity assay.

Splenocytes from naïve CD45.2 C57BL/6 mice were coated with 0.4mM of GP33-41 peptide (specific target cells) or non-coated cells (non-specific cells). Specific target cells and non-specific cells were stained with CTV at 0.5 and 10mM, respectively. Specific target cells and non-specific cells were incubated with sorted REC-T_EX or T_MEM cells (effector to target, E:T = 2:1) or without effector cells as controls for 16 hours in cRPMI. Percentage of specific killing was calculated as: 100 – [(100 x % specific target cells) / % specific target cells without effector cells].

Flow cytometry analysis and calculations.

For analysis of polyfunctionality and inhibitory receptor co-expression; positive gates for each parameter were created in FlowJo (V10.1.r7, Tree Star) and the Boolean tool was used to evaluate the frequency of cells in each possible combination of markers for each sample of interest. Background is not subtracted in the summary of these data. To normalize for MFI at different time points, MFI was calculated in reference to naïve cells. For Spider/Radar plot, MFI of T_EX and REC-T_EX was calculated as a percentage of T_MEM and normalized in reference to T_MEM at both time points for individual experiments, and the mean was presented on the plot.

Single-cell RNA-seq (scRNAseq) library preparation and analysis.

LCMV-specific P14 cells were sorted from mice infected with LCMV-Arm or LCMV-Cl13, as well as mice that received donor P14 T_EX or T_MEM. Splenocytes from mice within the same group were pooled, enriched for LCMV-specific CD8+ T cells, and stained with surface antibodies as described above. Live CD8+ congenic (CD45.1+ or CD45.1/2+) LCMV-GP33-tetramer+ P14 cells were FACS-sorted (1.18–2.4 x10⁴ cells), used to generate single-cell gel-beads in emulsion, and counted using a Countess II Automated Cell Counter (Thermo Fisher Scientific) prior to library preparation. Cells were loaded onto the Chromium Controller (10x Genomics) for a target recovery of 5,000 single cells. The scRNAseq libraries were generated using Chromium Single Cell 3′ Library and Gel Bead Kit v2 (10x Genomics) according to the manufacturer’s protocol. Briefly, emulsions were created and, after reverse transcription, gel-beads in emulsion were disrupted. Barcoded complementary DNA was isolated and amplified by PCR (12 cycles). Following fragmentation, end repair and A-tailing, sample indexes were added during index PCR (8 cycles). Samples were pooled and quantified using a KAPA Library Quantification Kit (Kapa Biosystems) prior to sequencing. Normalized pooled libraries were diluted at 1.8 ρg/ml loaded onto a NextSeq 500/550 High Output Flow Cell (Kit v2, 150 cycles, 400M reads, Illumina) and paired-end sequencing was performed on a NextSeq550 sequencer (Illumina). Sample demultiplexing, barcode processing, alignment, filtering, unique molecular identifier (UMI) counting and aggregating sequencing runs were performed using Cell Ranger software v.2.1.0 (10x Genomics). A total of 14,787 cells were recovered from all libraries, with 29,085 mean reads and 675 median genes detected per cell (3163 naïve cells, 3861 T_MEM, 1964 transferred-T_MEM, 3756 T_EX, and 2043 REC-T_EX). Downstream analysis was performed using Seurat v2.3 ⁵². Briefly, individual library outputs from the Cell Ranger pipeline were loaded, and cells with fewer than 200 detected genes were excluded from downstream analysis. Genes detected in fewer than three cells across the dataset were also excluded. Raw UMI counts were normalized by total expression in the corresponding cell and multiplied by a scaling factor of 10,000 to give UMI count per million total counts and then log-transformed. Variable genes were selected based on average expression and dispersion. Linear dimensional reduction (PCA) was performed using variable genes and the first 7 principal components were determined to be significant using the jackstraw method. tSNE was performed on these significant principle components using default parameters for 1,000 iterations for visualization in two dimensions. Unsupervised clustering was performed using a shared nearest neighbor modularity optimization-based algorithm. Optimal number of clusters was chosen using Identifying K mAjor cell Population groups (IKAP) algorithm ⁵³. Differential expression analysis was performed between each cluster and all other cells using a Wilcoxon rank sum test. Marker genes were identified by the Seurat function FindAllMarkers. Scaled expression data of these marker genes were used for creating the heatmaps. Normalized data are shown in the form of feature plots or violin plots. Trajectory analysis was performed with Monocle v.2 conducted using Monocle 2 v2.10.1 (Qiu et al., 2017b, 2017a). Gene ontology analysis was performed using maker genes in Metascape 3.0 ⁵⁴ (http://metascape.org). Network analysis was performed using Ingenuity Pathway Analysis (IPA) (QIAgen®).

ATACseq library preparation and analysis.

Using the same sorting parameters used for scRNAseq, P14 cells per replicate were sorted from spleens of mice into RPMI medium with 50% FBS (1–1.2x10⁴ cells for REC-T_EX, and 4–5x10⁴ for other cell types). ATACseq sample preparation was performed as described ⁴⁶ with minor modifications. Briefly, sorted were washed twice in cold PBS and resuspended in 50 μl of cold lysis buffer (10nM Tris-HCl, pH 7.4, 10mM NaCl, 3mM MgCl2, 0.1% Tween). Lysates were centrifuged (750xg, 10min, 4°C) and nuclei were resuspended in 50μl of transposition reaction mix (TD buffer [25μl], Tn5 Transposase [2.5μl], nuclease-free water [22.5μl]; (Illumina)) and incubated for 30min at 37°C. Transposed DNA fragments were purified using a Qiagen Reaction MiniElute Kit, barcoded with NEXTERA dual indexes (Illumina) and amplified by PCR for 11 cycles using NEBNext High Fidelity 2x PCR Master Mix (New England Biolabs). PCR products were purified using a PCR Purification Kit (Qiagen) and amplified fragments size was verified on a 2200 TapeStation (Agilent Technologies) using High Sensitivity D1000 ScreenTapes (Agilent Technologies). Libraries were quantified by qPCR using a KAPA Library Quant Kit (KAPA Biosystems). Normalized libraries were pooled, diluted at 1.8μg/ml loaded onto a TG NextSeq 500/550 Mid Output Kit v2.5 (150 cycles, 130 million reads, Illumina) and paired-end sequencing was performed on a NextSeq550 sequencer (Illumina). Raw ATACseq FASTQ files from paired-end sequencing were processed using the script available at the following repository (https://github.com/wherrylab/jogiles_ATAC). Samples were aligned to the GRCm38/mm10 reference genome using Bowtie2. We used samtools to remove unmapped, unpaired, and mitochondrial reads. ENCODE blacklist regions were also removed (https://sites.google.com/site/anshulkundaje/projects/blacklists). PCR duplicates were removed using Picard. Peak calling was performed using MACS v2 (FDR q-value 0.01). For each experiment, we combined peaks of all samples to create a union peak list and merged overlapping peaks with BedTools merge. The number of reads in each peak was determined using BedTools coverage. Differentially accessible peaks were identified following DESeq2 normalization using an FDR cut-off <0.05 unless otherwise indicated. Tracks were visualized using Integrative Genomics Viewer (v.2.3.77, Broad Institute) and GVIZ package on R ⁵⁵. Gene-to-peak associations were determined using the GREAT software package (v.3.0.0, http://bejerano.stanford.edu/great/public/html/) with default settings. Pearson correlations between samples were calculated in R and plotted. PCA was carried out using R (v.3.3.1, R Core Team). Gene ontology term enrichment was performed for each biological condition using GREAT, with default settings and binomial enrichment ⁵⁶. Motif enrichment analysis was performed using HOMER (v.4.6) with default settings ⁵⁷. Transcription Factor Binding Site (TFBS) analysis and PageRank analysis were performed using Taiji ⁵⁸. Statistical analysis of differential chromatin accessibility tests was done using DESeq2 (v.1.18.1), and FDR correction was performed using the Benjamini-Hochberg method in R (v.3.3.1). Significance of gene ontology term enrichments and motif enrichments were calculated with binomial tests and hypergeometric tests. P values and q values <0.05 were considered to indicate a significant difference.

Quantification and statistical analysis.

For each figure the number of replicates for each experiment is indicated in the legend. In the figures, mean and standard deviations (S.D.) are presented, error bars represent the ± S.D. Paired two-tailed t-tests were run on comparisons for co-transfer experiments (Wilcoxon matched-pairs signed rank test), and all other data were analyzed using unpaired two-tailed Student’s t-tests (Mann-Whitney test). Prism 6 (GraphPad Software) was used to calculate statistics. For adoptive transfer time points each individual point represents pooled P14 cells from one infection-matched mouse group, for all other time points individual points represent one mouse.

Data availability.

All data generated during this study are available within the paper. All sequencing data from this study are deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) and are accessible through a GEO Series accession code GSE150370. Any other relevant data are available from the corresponding author on request.

Diagrams.

All diagrams were created using Biorender.com.

Co-submitted articles.

Companion papers from Tonnerre et al.⁴⁹, and Sen et al.⁵⁰.

Extended Data

Extended Data Fig. 1. Upon antigen elimination T_EX downregulate several exhaustion markers and upregulate some T_MEM features.

(a) General gating strategy; dot plots in all figures are for transferred donor P14 cells gated on live CD8+ congenic cells. For surface stains, cells were also gated on D^b-GP33 tetramer+ cells (yellow rectangle). (b) Representative dot plots for the association between TCF-1 and Ly108 (Slamf6) expression on the day of adoptive transfer (d0), and d28 post-transfer. (c) Left, representative dot plots of Ly108 and PD-1 expression. Right, percentages of different T_EX subsets. N=9 biologically independent experiments for d0 and n=4 for d28. (d) Left, representative histograms for CD127. Right, CD127 expression as ΔMFI, corresponding to Figure 1c. (e) Left, representative histograms for Eomes. Right, Eomes expression as ΔMFI, corresponding to Figure 1e. Data presented in (c-e) as mean +/− SD. (f) Representative dot plots of cytokine production (IFNγ) and LAMP-1 (CD107a) expression on unstimulated splenocytes, corresponding to Figure 1g. (g) Left, experimental model, stability of T_EX P14 phenotype post-transfer was examined at day >180 post-transfer (d193-d570) into infection-free mice. Top row, representative dot plots of CD127 and PD-1, CD127 and CD62L, and TCF-1 and Tim-3 expression. To the right of dot plots, percentages of the different T_EX subsets, either compared to compared to T_MEM P14 cells from mice that cleared LCMV-Arm (d67-d238 p.i.). Bottom, representative histograms of Tox expression. To the right of histograms, levels of Tox expression. N=2 biologically independent experiments, 2-5 REC-T_EX mice/experiment. Data presented as mean +/− SD. All analyses for Extended Data 1 performed by unpaired two-tailed Student’s t-tests (Mann-Whitney test). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

Extended Data Fig. 2: REC-T_EX display an intermediate phenotype between T_EX and T_MEM in the P14 co-transfer or D^bGP33 tetramer+ non-P14 transfer approaches.

a) Experimental design for co-transfer of T_EX and T_MEM. (b) Left, dot plots showing percentages of co-transferred T_EX and T_MEM P14 cells on d0 and d28 post-transfer. Right, fold change in number of donor T_EX and T_MEM P14 cells on d28 post-transfer relative to the original number adoptively transferred on d0 (estimating 10% take). N=3 biologically independent experiments, 2-4 mice/experiment. (c) Number of T_EX and T_MEM P14 cells per million CD8 T cells from peripheral blood mononuclear cells (PBMCs) longitudinally as a percentage of the number of cells calculated at baseline (d2.5 post-transfer). N=1 experiment, 4-7 mice/time-point. Data presented as mean +/− S.E.M. Comparisons versus d2.5 analyzed by unpaired two-tailed Student’s t-tests (Mann-Whitney test). (d) Left, representative dot plots of CD127 and PD-1, and TCF-1 and Tim-3 expression. Right, percentages of different subsets. (e) Left panels, representative histograms for expression of indicated markers. Right, plots of ΔMFI. Data for (d-e) for PD-1 and CD127 n=3 independent experiments, and for Tox and TCF-1 n=1. Analyses for (b, d) paired two-tailed t-test (Wilcoxon matched-pairs signed rank test). *p<0.05, **p<0.01. (f) Left, representative dot plots for TIGIT and PD-1 expression. Right, percentages of IRs co-expression on P14 cells. N=1 experiment. (g) Left, representative dot plots of IFNγ production and LAMP-1/CD107a expression on P14 cells. Left, percentages of polyfunctionality on P14 cells. N=2 independent experiments. For (b-g) red and blue percentages are for T_EX and T_MEM P14 cells, respectively. Data for d28 from 2-4 mice per experiment. (h) Experimental design for adoptive transfer of CD8+ T cells containing D^bGP33-specific T_EX or T_MEM. (i) Left, representative dot plots for CD127 and PD-1 expression on donor D^b-GP33 specific T cells. Right, percentages of different subsets. (j) Left, representative histograms for level of expression of PD-1 or CD127. Right, plots for level of expression of PD-1 and CD127. Data (i-j) n=2 independent experiments, 3-8 mice/group/experiment for d28. Data presented as mean +/− SD. Analyses for (i-j) by unpaired two-tailed Student’s t-tests (Mann-Whitney test). ***p<0.001, ****p<0.0001.

Extended Data Fig. 3. REC-T_EX are transcriptionally distinct from both T_EX and T_MEM.

(a) Top, general gating strategy for sorting; sorted cells were gated on live CD8+ D^bGP33-tetramer+ cells. Bottom, dot plots showing purity post-sorting for samples used for scRNAseq. (b) tSNE plot of unsupervised clustering of combined T_EX, T_MEM, and REC-T_EX cells. (b) Contribution of T_EX, T_MEM, and REC-T_EX to each cluster in (b). (c) Heatmap of the top differentially expressed genes (DEGs) between T_EX, T_MEM, and REC-T_EX. (d) Violin plots for genes differentially expressed by REC-T_EX compared to both T_EX and T_MEM. (e) tSNE plots of T_EX, T_MEM, and REC-T_EX cells with overlay of Pdcd1, Lag3, and Gzmb genes.

Extended Data Fig. 4. REC-T_EX are transcriptionally heterogeneous and distinct from the TCF-1⁺ T_EX subset.

a) Circos plots of different REC-T_EX clusters compared to T_EX and T_MEM signatures. (b) Violin plots for individual gene expression by the different REC-T_EX clusters from Figure 2c. (c) Pseudotime plot of T_EX and REC-T_EX with overlay of Tox (d) Pseudotime plot of T_EX and REC-T_EX with overlay of upregulated gene signature in TCF-1⁻ T_EX subset from Utzschneider et al. ²⁵. (e) Volcano plot showing selected differentially expressed genes between TCF-1⁺ T_EX and REC-T_EX.

Extended Data Fig. 5. Partial recovery from exhaustion in REC-T_EX is associated with preferential survival of the progenitor T_EX subset.

(a) Left, representative dot plot of TCF-1 expression in Blimp1-YFP T_EX P14 cells on d20 p.i. with LCMV-Cl13. Middle, dot plots showing sorting purity of Blimp1-YFP⁻ (i.e. TCF-1⁺) progenitor T_EX (Prog-T_EX) and Blimp1-YFP⁺ (i.e. TCF-1⁻) terminal T_EX (Term-T_EX). Right, experimental design for co-transfer of Prog-T_EX and Term-T_EX into infection-free recipient mice. (b) Left, representative dot plots for frequency of donor Blimp1-YFP⁻ Prog-T_EX or Blimp1-YFP⁺ Term-T_EX P14 cells, on the day of adoptive transfer (d0) and d26 post-transfer. Middle, number of P14 cells for Prog-T_EX and Term-T_EX on d0 and d26 post-transfer (estimating 10% take on d0). Right, absolute number of P14 cells per spleen for each subset on d26 post-transfer. (c) Left, representative dot plots of CD127 and PD-1 expression. Right, percentages of CD127⁺ or PD-1⁺CD127⁻ for both subsets on d0 and d26 post-transfer, and absolute numbers per spleen of CD127+ cells for each subset on d26 post-transfer. (d) Left, representative histograms for expression of CD127 or PD-1 on Prog-T_EX and Term-T_EX on d0 and d26 post-transfer, and on endogenous GP33-tetramer+ T_MEM of recipient mice on d26 post-transfer. Right, plots for MFIs. Data from one experiment, 2 mice.

Extended Data Fig. 6. Secondary effectors derived from REC-T_EX and T_MEM are distinct phenotypically and functionally.

(a) Representative dot plots showing sorting purity. (b) Representative dot plot for the frequency of co-transferred donor REC-T_EX and T_MEM P14 cells, on d-1. (c) Absolute number of T_EX and REC-T_EX P14 cells per spleen on d8 post challenge with LCMV-Arm. N=1 experiment, 3 mice. (d) Representative dot plots and plots for PD-1 and KLRG1 expression from different organs. Data representative of two independent experiments, 4-5 mice. (e) Representative dot plots of cytokine production (IFNγ and TNF) and CD107a expression on unstimulated splenocytes. Corresponding to Figure 4e. (f) Cytotoxicity assay of secondary effector cells (d8 post LCMV-Arm challenge) from separately transferred REC-T_EX or T_MEM P14 cells into naïve mice. Left, plots of % specific killing for REC-T_EX compared to T_MEM. Middle and right, number of effector cells and ratio of effector:target cells in REC-T_EX containing wells compared to T_MEM wells. N=1 experiment. Data presented as mean +/− SD. Unpaired two-tailed Student’s t-tests (Mann-Whitney test). (g) Percentages of CD127⁺CD62L⁺ and Tox⁺PD-1⁺, and expression of PD-1, Tox and Eomes on co-transferred REC-T_EX and T_MEM P14s at d30 post-challenge with LCMV-Arm. N=2 biologically independent experiments, 2-5 mice/experiment. (h) Absolute number per spleen for cells derived from T_MEM and REC-T_EX P14 cells (left), endogenous CD8+ T cells (middle), and endogenous GP33-tetramer+ cells (right), on d8 post LCMV-Arm challenge. (i) Left, representative dot plots of T-bet and Eomes expression. Right, percentages of Eomes⁺T-bet⁺. (j) Left, representative dot plots for GzmB expression. Right, percentages of GzmB expressing cells. (k) Left, histograms for expression of Tox and T-bet. Right, plots for levels of expression as ΔMFI (compared to naïve). (l) Representative dot plots and plots for PD-1 and KLRG1 from different organs from anti-PD-L1 treated mice compared to control PBS-treated mice. Data for (h-l) N=2 biologically independent experiments, 2-4 mice/group/experiment. All analyses for Extended Data 6, for co-transferred cells performed by paired two-tailed t-test (Wilcoxon matched-pairs signed rank test). PBS versus αPD-L1 (h-l) analyzed using unpaired two-tailed Student’s t-tests (Mann-Whitney test). *p<0.05, **p<0.01, ***p<0.001.

Extended Data Fig. 7. REC-T_EX remain epigenetically similar to T_EX and bear signatures from both the TCF-1⁺ and TCF-1⁻ T_EX subsets.

(a) Representative dot plots showing sorting purity for ATACseq samples. (b) Correlation of normalized peak enrichment between replicates of for ATACseq samples for each cell type. R² indicates degree of correlation between replicates at p<2.2e-16, calculated by Spearman correlation. (c) Pearson correlation of ATACseq profiles from T_EX, T_MEM, and REC-T_EX samples. (d) Principal component analysis (PCA) of ATACseq profiles for naïve, T_EX, T_MEM, and REC-T_EX P14 cells. (e) Venn diagram showing the overlap between OCRs significantly open in REC-T_EX versus T_EX (REC-T_EXUP_vsT_EX) (from Fig. 5b) and UP_T_MEM _vsT_EX. (f) Representative ATACseq tracks at indicated loci from naïve, T_EX, T_MEM, and REC-T_EX. (g) PCA of ATACseq profiles for T_EX, T_MEM, and REC-T_EX samples from our current dataset compared to T_MEM, CXCR5⁻ Tim-3⁺ T_EX and CXCR5⁻ Tim-3⁺ T_EX from Jadhav et al. ³² (h) Heatmap showing chromatin accessibility of CXCR5⁻ Tim-3⁺ T_EX and CXCR5⁻ Tim-3⁺ T_EX from Jadhav et al. ³² for the 182 differential peaks from REC-T_EX vs. T_EX signature of the current dataset (presented in Figure 6b). (i) Representative ATACseq tracks at indicated loci from T_EX and REC-T_EX samples from the current dataset compared to the four T_EX subsets from Beltra et al. ²⁹

Extended Data Fig. 8. Correlation between transcriptional and epigenetic profiles of REC-T_EX identifies epigenetically fixed networks.

(a) Genes from scRNAseq with significantly different expression between REC-T_EX and T_MEM that also had associated significant differences in the chromatin accessibility. (b) Bubble plot showing significantly different biological processes associated with significantly different OCRs between REC-T_EX and T_MEM identified by GREAT. (c) TFs with significantly enriched motifs in OCRs in REC-T_EX versus T_MEM (top half, purple circles) or in T_MEM versus REC-T_EX (bottom half, blue circles). Pairwise analysis using HOMER. −log₁₀ p-value calculated by hypergeometric distribution test. (d) Venn diagrams showing the overlap in TFBS between T_EX, T_MEM, and REC-T_EX for some transcription factors from PageRank analysis (Figure 6g), Tcf7, Hic1 and Hsf1. (e) Heatmap of the 58 OCRs showing significantly differential accessibility in REC-T_EX compared to T_EX from the ΔTox-T_EX versus wild-type T_EX signature by Khan et al. ¹⁷.

Extended Data Fig. 9. Eliminating chronic antigen stimulation induces varying degrees of transcriptional and epigenetic recovery in T_EX.

Upon eliminating chronic antigen stimulation, exhausted T cells (T_EX) failed to fully differentiate into bona fide memory T cells (T_MEM), but rather gave rise to cells with mixed exhaustion and memory features, recovering-T_EX cells (REC-T_EX). The degree of phenotypic and transcriptional recovery towards memory was not associated with a similar degree of recovery in recall capacity, mainly due to limited changes in the chromatin accessibility landscape. Ag, antigen; IL-7, interleukin-7; MPEC, memory precursor effector cells; SLEC, short-lived effector cells; and T_EFF, effector T cells.Diagram created with Biorender.com.