Abstract

   Epigenetic scarring of terminally dysfunctional (T[Dysf]) CD8^+ T cells
   hinders long-term protection and response to immune checkpoint blockade
   during chronic infections and cancer. We developed a faithful in vitro
   model for CD8^+ T cell terminal dysfunction as a platform to advance T
   cell immunotherapy. Using TCR-transgenic CD8^+ T cells, we found that
   1-week peptide stimulation, mimicking conditions in previous models,
   failed to induce a stable exhaustion program. In contrast, prolonged
   stimulation for 2–3 weeks induced T cell dysfunction but triggered
   activation-induced cell death, precluding long-term investigation of
   exhaustion programs. To better mimic in vivo exhaustion, we provided
   post-effector, chronic TGF-β1 signals, enabling survival of chronically
   stimulated CD8^+ T cells for over 3 weeks. These conditions induced a
   state of terminal dysfunction, marked by a stable loss of effector,
   cytotoxicity, and memory programs, along with mitochondrial stress and
   impaired protein translation. Importantly, transcriptomic and
   epigenetic analyses verified the development of terminal
   exhaustion-specific signatures in T[Dysf] cells. Adoptive transfer of
   T[Dysf] cells revealed their inability to recall effector functions or
   proliferate after acute lymphocytic choriomeningitis virus rechallenge.
   This tractable model system enables investigation of molecular pathways
   driving T cell terminal dysfunction and discovery of therapeutic
   targets for cancer or chronic infections.

   Keywords: Immunology, Oncology

   Keywords: Adaptive immunity, Cancer immunotherapy, Epigenetics
     __________________________________________________________________

   [41]graphic file with name jciinsight-10-191220-g048.jpg
     __________________________________________________________________

   Here, we developed the first faithful in vitro model of terminal CD8 T
   cell dysfunction that enables investigation of epigenetic scarring of
   exhausted T cells

Introduction

   CD8^^+ T cell effector function is critical for clearing viral
   infections or tumor cells. When antigen-primed CD8^^+ T cells
   experience prolonged antigenic stimulation and inflammatory
   microenvironments, they progressively lose their cytotoxic capacity and
   become dysfunctional. This state of T cell dysfunction, also called
   exhaustion, is characterized by high expression of surface inhibitory
   receptors, as well as a gradual loss of effector functions,
   proliferation capacity, and memory potential, along with impaired
   mitochondrial function and protein translation ([42]1, [43]2). The
   gradual progression of dysfunction, coupled with variations in spatial
   cues, produces heterogeneous exhausted T cell (TEX) populations
   ([44]2–[45]5). These include a progenitor subset with self-renewal and
   effector capacity; a cytolytic subset with strong cytotoxic function;
   and a terminally exhausted subset with poor function, survival, and
   proliferative capacities ([46]1–[47]3, [48]6). Immune checkpoint
   blockade (ICB) is a revolutionary cancer immunotherapy that
   reinvigorates both the progenitor and cytolytic TEX subsets by blocking
   signals from inhibitory receptors, such as cytotoxic T lymphocyte
   antigen 4 and programmed cell death 1 (PD-1), but has limited impact on
   terminally exhausted T cells — a major obstacle for ICB therapy in many
   patients with cancer.

   We and others have shown that the epigenome plays a central role in
   locking TEX fate commitment ([49]3, [50]7–[51]14). Importantly, de novo
   DNA methylation enforces the silencing of effector and memory programs
   in TEX cells, restraining ICB response ([52]8). Recent studies have
   revealed key roles for microenvironmental signals in regulating T cell
   dysfunction, such as strength and duration of TCR stimulation, chronic
   TGF-β1 exposure, type I IFN signaling, and nutrient limitation ([53]1,
   [54]2). Additionally, hypoxia — adapted for in vitro modeling ([55]15)
   — showed a key role in driving and potentially accelerating T cell
   dysfunction ([56]16, [57]17). Yet the specific signals critical for
   initiating and imprinting exhaustion-associated epigenetic programs
   remain unclear. This major gap in the field is due to the diverse range
   of cues that CD8^^+ T cells encounter in vivo within tumor or infected
   tissue microenvironments. This complexity makes it challenging to
   isolate and understand the individual and combined effects of
   exhaustion-inducing signals on epigenetic regulation of TEX cells.

   As such, reductionist approaches will be critical for dissecting how
   signals in each microenvironment drive or counteract epigenetic
   programming of T cell exhaustion. Recent efforts to develop in vitro
   models of T cell exhaustion have aimed to address these challenges
   ([58]18–[59]21). In previous models, CD8^^+ T cells acquire
   exhaustion-like features, including inhibitory receptor upregulation,
   reduced proliferation, and mitochondrial stress. However, several
   models rely primarily on persistent TCR stimulation as the sole driver
   of T cell exhaustion, with repeated antigen-dependent or independent
   (anti-CD3) stimulation under a short timeline of 5–10 days
   ([60]18–[61]21). The fundamental obstacle we have found in TCR-based
   approaches is that repeated TCR stimulations of CD8^^+ T cells leads to
   activation-induced cell death ([62]9, [63]20). Thus, persistent TCR
   signaling alone fails to replicate the extended timeline of T cell
   exhaustion observed during chronic infection or cancer, in which at
   least 2–3 weeks of chronic stimulation in the presence of other
   microenvironmental signals is necessary to establish a terminally
   dysfunctional (T[[Dysf]]) state ([64]11, [65]22–[66]24). Furthermore,
   it remains unclear whether in vitro–induced dysfunction is stable after
   antigen withdrawal — a core defining feature of the epigenetically
   fixed state of terminal exhaustion observed in vivo ([67]11,
   [68]22–[69]24). Therefore, it is essential to develop a robust model
   system that faithfully recapitulates the progressive nature of T cell
   exhaustion and enables tracking of chronically stimulated T cells over
   extended periods. Such a system is vital for developing reliable
   discovery platforms to identify novel therapeutic targets aimed at
   preventing or reversing T cell dysfunction and enhancing the
   effectiveness of T cell immunotherapies.

   Here, we developed an antigen-dependent model system of T cell
   exhaustion using TCR-transgenic, lymphocytic choriomeningitis
   virus–glycoprotein 33-41–specific (LCMV-GP33–specific) CD8^^+ T cells.
   We found that prolonged (2–3 weeks) rather than short-term antigen
   stimulation (1 week) was required to induce stable T cell exhaustion.
   To overcome activation-induced cell death and better mimic in vivo
   conditions, we introduced chronic TGF-β1 signals at the post-effector
   phase, which supported CD8^^+ T cell survival. The combined signals and
   long-term viability faithfully induced terminal dysfunction in CD8^^+ T
   cells (T[[Dysf]]), which displayed a remarkable loss of effector and
   memory programs, mitochondrial stress, and impaired protein
   translation. Transcriptome and epigenome analyses of these
   dysfunctional lymphocytes verified the acquisition of
   exhaustion-specific signatures. Furthermore, adoptive T cell transfer
   studies showed that T[[Dysf]] cells had impaired effector recall
   responses and proliferation capacity. Our model enables investigation
   of the long-term developmental process that leads to terminal
   dysfunction in antigen-specific CD8^^+ T cells, providing a faithful
   platform for identifying new therapeutic targets to block or reverse
   epigenetic processes stabilizing terminal dysfunction.

Results

Acute antigenic stimulation fails to induce a dysfunctional program in CD8^^+
T cells.

   To determine whether 1 week of repeated TCR stimulation induces a
   stable dysfunctional program in CD8^^+ T cells, we utilized naive
   TCR-transgenic murine CD8^^+ T cells, “P14 cells,” which respond to
   LCMV-derived GP33 peptide. We compared P14 cells initially activated
   using anti-CD3 plus anti-CD28 stimulation on day 0–2 (Acute-2d) versus
   those that received repeated GP33 peptide stimulations following this
   activation phase until day 7 (Acute-7d) ([70]Figure 1A). After initial
   activation (days 0–2), over 98% of P14 cells upregulated CD44 and PD-1,
   with most coexpressing PD-1 and Tim3 — indicating robust effector
   differentiation ([71]Figure 1, B–D, and [72]Supplemental Figure 1, A
   and B; supplemental material available online with this article;
   [73]https://doi.org/10.1172/jci.insight.191220DS1). By day 7, repeated
   stimulation maintained high frequencies of CD44^^+PD-1^^+ and
   PD-1^^+Tim3^^+ populations in Acute-7d cells ([74]Figure 1, C and D,
   and [75]Supplemental Figure 1B), mimicking inhibitory receptor
   upregulation on effector CD8^^+ T cells in vivo ([76]25). To assess
   stability of these features, we rested antigen-experienced P14 cells
   for 12 days without GP33 stimulation and rechallenged them with GP33
   peptide on days 12 and 19 ([77]Figure 1A). The frequencies of
   CD44^^+PD-1^^+ and PD-1^^+Tim3^^+ populations gradually declined over
   time but remained elevated in Acute-7d P14 cells versus Acute-2d cells
   by day 19 ([78]Figure 1, C and D, and [79]Supplemental Figure 1, B–D).
   These findings suggest that Acute-7d P14 cells do not have stable
   maintenance of inhibitory receptors following rest, unlike TEX cells
   retaining high levels of inhibitory receptors in vivo ([80]26, [81]27).

Figure 1. Acute antigenic stimulation for 1 week fails to induce a
dysfunctional program in CD8^+ T cells.

   [82]Figure 1
   [83]Open in a new tab

   (A) Schematic for in vitro stimulation of GP33-specific P14 CD8^+ T
   cells isolated from spleens of naive mice. P14 cells were activated
   with anti-CD3 and anti-CD28 day 0–2, followed by resting (“Acute-2d”),
   or by 2 rounds of GP33 peptide stimulation day 2–7 followed by resting
   until day 19 (“Acute-7d”). (B) Representative FACS plots showing
   expression of PD-1 and CD44 or Tim3 on P14 cells from day 0 to day 2
   and summary bar graphs showing frequency of PD-1^+CD44^+ or
   PD-1^+Tim3^+ P14 cells on days 0 and 2. (C) Longitudinal tracking of %
   PD-1^+CD44^+ P14 cells for either Acute-2d (bottom line, green) or
   Acute-7d (top line, blue) or (D) % PD-1^+Tim3^+ P14 cells from day
   0–19. (E) Expression levels (shown by geometric mean fluorescence
   intensity, gMFI) of CD62L from Acute-2d or Acute-7d P14 cells on days 7
   and 19. (F) Bar graphs showing % of CD62L^+Ccr7^+, total CD62L^+, and
   total Ccr7^+ P14 cells on day 19. (G) Schematic for adoptive transfer
   of 100K Acute-2d or Acute-7d in vitro–stimulated P14 cells in
   congenically distinct C57BL/6 mice on day 19, followed by acute LCMV
   infection and analysis day 6–7 postinfection. (H) Fold expansion of P14
   cells relative to endogenous GP33-specific CD8^+ T cells on day 6–7
   postinfection. (I) Representative FACS plots and % Ifnγ^+CD107a^+ P14
   cells after ex vivo GP33 peptide restimulation of splenocytes. For B–F
   n = 4 biological replicates, representative of 2 to 3 independent
   experiments. For H and I, data were pooled from 2 independent
   experiments, each with n = 3–4 biological replicates per group. *P <
   0.05, ***P < 0.001, ****P < 0.0001. Comparisons were determined by
   2-way ANOVA (C and D) or Mann–Whitney U test (unpaired, 2-sided) (B, E,
   F, H, and I). Error bars indicate mean ± SEM.

   A hallmark of T cell exhaustion is the failure to restore memory
   programs after antigen removal ([84]11–[85]14, [86]24). To determine
   whether repeated antigen stimulation for 1 week affects the recovery of
   memory programs, we tracked expression levels of memory- and
   stemness-associated programs before and after resting (Ly108, L
   selectin [CD62L], Ccr7, and IL-7 receptor [Il7r; CD127]). On day 7,
   Acute-7d P14 cells initially downregulated Ly108 and CD62L, but Ly108
   was recovered after resting while CD62L remained higher in Acute-2d
   cells ([87]Figure 1E and [88]Supplemental Figure 1I). The frequencies
   of CD62L^^+ Ccr7^^+ cells were higher in rested Acute-2d cells versus
   Acute-7d cells, while Il7r remained high in both groups by day 19
   ([89]Figure 1F and [90]Supplemental Figure 1J). Notably, the
   proliferative activity (measured by Ki67 expression) was comparable
   between the two groups on day 19, with a modest but nonsignificant
   increase in Acute-2d cells ([91]Supplemental Figure 1H). These findings
   suggest that 1 week of repeated stimulation does not lead to a loss of
   memory features, mirroring the retained memory potential in effector
   CD8^^+ T cells isolated 1 week following chronic LCMV infection
   ([92]24).

   We next evaluated effector function after GP33 rechallenge at different
   time points. Both groups exhibited high polyfunctionality (Ifnγ and Tnf
   coexpression, CD107a: degranulation activity), which increased with
   resting ([93]Supplemental Figure 1, G and K–N). Acute-7d P14 cells had
   significantly higher degranulation activity and increased expression of
   cytotoxic molecules granzyme B (Gzmb) and Perforin compared with
   Acute-2d cells ([94]Supplemental Figure 1, E–G), indicating that
   repeated acute antigen stimulation may enhance cytolytic effector
   differentiation in CD8^^+ T cells. We also observed lower apoptosis
   (annexin V) in Acute-2d, while both groups retained viable populations
   after resting ([95]Supplemental Figure 1O). Additionally, mitochondrial
   mass (MitoTracker-Green) and oxidative phosphorylation activity
   (MitoSox) were similar ([96]Supplemental Figure 1, P and Q). Overall,
   our findings demonstrate that 1 week of antigen stimulation alone is
   insufficient to induce stable dysfunction, suggesting that longer
   antigenic stimulation or additional microenvironmental cues are
   required for terminal exhaustion.

   Finally, we tested recall responses by adoptively transferring Acute-2d
   or Acute-7d CD8^^+ T cells (day 19) into mice, followed by acute LCMV
   infection ([97]Figure 1G), as terminally TEX cells have limited
   capacity to mount a response to antigen rechallenge ([98]11, [99]24).
   Both groups underwent extensive antigen-dependent proliferation, though
   Acute-2d cells exhibited greater expansion ([100]Figure 1H), possibly
   due to enhanced survival or CD62L-mediated lymphoid homing ([101]Figure
   1, E and F). Yet, after expansion, CD62L expression was comparable
   between the groups ([102]Supplemental Figure 2B). Upon ex vivo GP33
   peptide stimulation, both groups showed strong polyfunctionality,
   expressing high levels of Ifnγ, CD107a, Perforin, Gzmb, and Tnf
   ([103]Figure 1I and [104]Supplemental Figure 2, D, E, and H). Markers
   of proliferation, activation, and memory (CD62L, Ki67, CD44, Ly108,
   Tcf1, PD-1, Tim3) remained comparably elevated ([105]Supplemental
   Figure 2, A–C, F, G, I, and J), with similar expression patterns in
   liver and lungs ([106]Supplemental Figure 2, K–T). Overall, these
   findings confirm that repeated antigen stimulation for 1 week fails to
   induce terminal dysfunction, as these cells retain robust recall
   capacity.

Chronic antigenic stimulation faithfully induces terminal dysfunction.

   To develop a model system for terminal dysfunction in CD8^^+ T cells,
   we extended the duration of repeated GP33 peptide stimulation for an
   additional 2 weeks (Chronic GP33 stim) ([107]Figure 2A). Compared with
   acutely stimulated Acute-7d condition, 2–3 weeks of prolonged antigen
   stimulation produced a higher frequency of PD-1^^+Tim3^^+ P14 cells
   ([108]Figure 2B). However, the chronically stimulated P14 cells
   exhibited substantial loss of viability, with a progressive decline in
   total cell numbers and increased cell death, likely due to
   activation-induced cell death — a major limitation for studying CD8^^+
   T cells under prolonged periods of TCR stimulation in vitro ([109]20)
   ([110]Figure 2C and [111]Supplemental Figure 3F).

Figure 2. Chronic antigenic stimulation faithfully induces terminal
dysfunction.

   [112]Figure 2
   [113]Open in a new tab

   (A) Schematic for in vitro stimulation of P14 cells: Activated P14
   cells were acutely stimulated by GP33 peptide from day 2–7 and rested
   until day 19 (Acute-7d condition, blue) or repeatedly GP33-stimulated
   from day 2–19 with chronic TGF-β1 exposure (day 7–19, red T[Dysf]) or
   without TGF-β1 signals (gray). Longitudinal tracking of (B) frequencies
   of PD-1^+Tim3^+ P14 cells, (C) P14 cell numbers/200 μL, (D) %Ifnγ^+ and
   (E) %Tnf^+ from day 7–19 for Acute-7d (blue), Chronic GP33 stim (gray),
   or Chronic GP33+TGF-β1 (red). (F) Representative FACS plots and bar
   graphs of Ifnγ and Tnf expression, (G) %Ifnγ^+CD107a^+, or (H)
   %Tnf^+CD107a^+ P14 cells on day 19 after GP33 peptide rechallenge. (I)
   Representative FACS plots and bar graphs of Tim3 and Ly108 expression.
   (J) Summary bar graph of %Il7r^+CD62L^+ P14 cells on day 19. Expression
   level (geometric MFI; gMFI) of (K) MitoTracker Green dye (for
   mitochondrial mass) and (L) MitoSox Red dye (for mitochondrial ROS)
   within P14 cells on day 19. (M) Schematic for tumor-killing assay of
   naive (yellow), Acute-7d, or T[Dysf] P14 cells cocultured with
   GP33-expressing CT2A glioma tumor cells. (N) Bar graph showing numbers
   of dead CT2A-GP33 cells per 10 viable P14 cells. All n = 4 biological
   replicates, representative of 2 to 3 independent experiments. Adjusted
   P value *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.
   Comparisons were determined by 2-way ANOVA (B–E) or 1-way ANOVA with
   Tukey’s multiple comparisons (F–L and N). Error bars indicate mean ±
   SEM.

   Previous studies to model T cell exhaustion in vitro have used
   plate-bound anti-CD3/CD28 stimulation (day 0–2) followed by repeated
   plate-bound anti-CD3 stimulations for 5–10 days ([114]20, [115]21).
   While extended strong TCR stimulation induced high levels of PD-1 and
   Tim3 ([116]Supplemental Figure 3, A–C), this method also induced
   extensive cell death that limited reliable functional assessment and
   tracking of the epigenetic remodeling toward terminal T cell exhaustion
   ([117]Supplemental Figure 3, A, D, and E). To better mimic in vivo
   conditions where CD8^^+ T cells can survive and develop exhaustion
   following prolonged antigen exposure, we provided continuous TGF-β1
   signals between days 7 and 19 (Chronic GP33 stim + TGF-β1) to
   ameliorate activation-induced cell death ([118]Figure 2A) ([119]9). Our
   previous work demonstrated that prolonged, post-effector TGF-β1
   exposure drives terminal dysfunction in human CD8^^+ T cells that are
   chronically stimulated in an antigen-independent manner ([120]9).
   Indeed, we found that chronic GP33 plus TGF-β1–stimulated P14 cells not
   only showed improved cell viability compared with chronic GP33 alone or
   plate-bound stimulation approaches ([121]Figure 2C and
   [122]Supplemental Figure 3, D–F) but also further increased PD-1 and
   Tim3 coexpression ([123]Figure 2B). To evaluate the impact of TGF-β1
   exposure on CD8^^+ T cell survival, we performed annexin V and
   live/dead staining on chronically stimulated CD8^^+ T cells. TGF-β1
   supplementation significantly improved the viability of P14 T cells
   under chronic stimulation, with reduced apoptosis and increased
   proportions of live cells compared with chronic GP33 stimulation alone
   on both days 14 and 20 ([124]Supplemental Figure 3, F–H). These
   findings suggest that TGF-β1 attenuates activation-induced cell death.
   The improved viability of P14 cells under chronic GP33 plus TGF-β1
   stimulation (hereafter referred to as T[[Dysf]]) enabled robust
   analysis of exhaustion programming during prolonged antigenic
   stimulation.

   We next examined whether chronic GP33 plus TGF-β1 stimulation induces
   dysfunction characteristic of terminally exhausted CD8^^+ T cells. By
   tracking effector cytokine production in chronically versus acutely
   stimulated P14 cells, we observed that chronic GP33 stimulation — with
   or without TGF-β1 — led to a progressive decline in effector cytokine
   output ([125]Figure 2, D and E), indicative of a transition toward
   dysfunction. Importantly, T[[Dysf]] exhibited a significant loss of
   polyfunctionality compared with acutely stimulated cells, with reduced
   frequencies of Ifnγ^^+Tnf^^+, Ifnγ^^+CD107a^^+, and Tnf^^+CD107a^^+
   cells, reflecting severe impairment in both cytokine production and
   degranulation capacity ([126]Figure 2, F–H). Furthermore, T[[Dysf]]
   cells showed substantial loss of memory-associated markers, including
   reduced expression and frequencies of Il7r and CD62L, compared with
   both acutely and chronically stimulated controls ([127]Figure 2J and
   [128]Supplemental Figure 4E), further indicating impaired memory
   differentiation.

   We also examined exhaustion-associated surface markers and found that
   T[[Dysf]] cells exhibited a shift toward a terminally exhausted
   phenotype. Alongside sustained high coexpression of PD-1 and Tim3,
   T[[Dysf]] cells showed decreased frequencies and expression of Ly108 —
   a marker of the progenitor subset of TEX cells — leading to an
   increased proportion of Tim3^^+Ly108^^– cells, a phenotype of terminal
   exhaustion ([129]2, [130]28, [131]29) ([132]Figure 2I and
   [133]Supplemental Figure 4F). Furthermore, T[[Dysf]] cells demonstrated
   increased frequencies of Tim3^^+CD62L^^– cells, further supporting
   their progression toward a terminally exhausted state
   ([134]Supplemental Figure 4A). Notably, T[[Dysf]] cells showed elevated
   expression of CD103, a well-established downstream target gene of
   TGF-β1 signaling that marks a terminally exhausted subset of human
   tumor-reactive CD8^^+ T cells ([135]30–[136]33) ([137]Supplemental
   Figure 4D). In addition, T[[Dysf]] cells showed increased expression of
   Gzmb — commonly associated with terminal dysfunction ([138]28, [139]34)
   — as well as higher levels of perforin compared with Acute-7d cells,
   likely reflecting the effects of sustained TCR stimulation
   ([140]Supplemental Figure 4, B and C). To assess mitochondrial stress,
   a key feature linked to terminal dysfunction in CD8^^+ T cell
   exhaustion ([141]15, [142]21, [143]35), we tracked changes in
   mitochondrial functions within T[[Dysf]] P14 cells and found a
   significant reduction in mitochondrial mass accompanied by increased
   mitochondrial ROS levels — a marker of progressive T cell exhaustion in
   vivo ([144]Figure 2, K and L). These results indicate that chronic
   antigen plus TGF-β1 stimulation over 3 weeks establishes a
   dysfunctional program in CD8^^+ T cells.

   To assess the cytotoxic function of in vitro–differentiated P14 cells,
   we utilized an antigen-dependent tumor-killing assay, in which we
   cocultured naive or chronically or acutely stimulated P14 cells with
   GP33-expressing CT2A glioma cells for 18–20 hours ([145]Figure 2M).
   Consistent with their impaired effector functions, T[[Dysf]] cells
   demonstrated significantly reduced tumor-killing activity compared with
   acutely stimulated P14 cells ([146]Figure 2N), further demonstrating
   their exhausted state. Overall, these findings reveal a fundamental
   role for chronic TGF-β1 signals in driving both human and mouse T cell
   dysfunction ([147]9). Our model system should now enable in vitro
   studies of how chronically stimulated CD8^^+ T cells progress toward
   dysfunction under extended antigenic stimulation and assessment of
   microenvironmental cues beyond TCR signaling alone in this process.

Chronic TCR plus TGF-β1 signaling establishes the transcriptomic signature of
terminal exhaustion and impairs protein translation.

   Next, we examined whether in vitro chronic antigen plus TGF-β1
   stimulation establishes the transcriptional circuits of T cell
   exhaustion observed in vivo. We performed RNA-Seq on
   antigen-experienced P14 cells, FACS-purified on day 7 (Effector stage)
   or day 19 (Acute-2d, Acute-7d, and T[[Dysf]]) ([148]Supplemental Figure
   5A). Principal component analysis (PCA) revealed that T[[Dysf]] cells
   underwent distinct transcriptional programming and clustered closely
   with terminally TEX cells from chronic LCMV infection ([149]28)
   ([150]Figure 3A and [151]Supplemental Figure 5B). In contrast, the
   transcriptomes of acutely stimulated P14 cells (Acute-2d and Acute-7d)
   closely resembled those of memory CD8^^+ T cells, including TCM and TEM
   cell subsets from acute LCMV infection and progenitor TEX cells from
   chronic LCMV infection ([152]28, [153]36). These results indicate that
   prolonged TCR and TGF-β1 stimulation accurately models the
   transcriptional landscape of terminal exhaustion.

Figure 3. Chronic TCR plus TGF-β1 signals establish the terminal exhaustion
transcriptome and impair protein translation.

   [154]Figure 3
   [155]Open in a new tab

   (A) Principal component analysis (PCA) plot comparing transcriptional
   (RNA) signatures for in vitro antigen-stimulated P14 cells (isolated on
   day 7, “Effector”; or day 19, “T[Dysf],” “Acute-2d”, “Acute-7d”) to
   published signatures for PD-1^+CD8^+ TEX subsets isolated on day >45 of
   chronic LCMV infection (progenitor “Tprog” CD101^–Tim3^–; cytolytic
   “T[cyto]” CD101^–Tim3^+; terminally exhausted “T[term]” CD101^+Tim3^+)
   ([156]28) or memory CD8^+ T cell subsets on day 48 postinfection with
   acute LCMV (central memory “TCM,” effector memory “TEM”) ([157]36). (B)
   Heatmap showing differentially expressed genes (DEGs) between T[Dysf]
   and Acute-7d P14 cells. DEGs were plotted based on relative Z-score of
   normalized RNA counts and grouped based on shared expression patterns.
   (C) Gene set enrichment analysis (GSEA) plots for DEGs upregulated in
   T[Dysf] versus Acute-7d in comparison to published signatures for TEX
   subsets on day >45 of chronic LCMV infection ([158]28). (D) GSEA score
   plot compared with Hallmark gene signature for DEGs upregulated in
   Acute-7d versus T[Dysf]. (E) Summary bar graph for intracellular
   puromycin levels (gMFI) in TEX subsets isolated from LCMV clone
   13–infected mice on day 21 postinfection or (F) in vitro–stimulated P14
   cells on day 19 after GP33 peptide rechallenge. N = 2 biological
   replicates for RNA-Seq, or n = 4–6 biological replicates for E and F,
   representative of 2 to 3 independent experiments. Statistical
   significance was determined by (A) PCA, (B) DESeq2 analysis, or (D)
   GSEA using Partek software or (C) using UCSD and Broad Institute GSEA
   software. Comparisons (E) were determined by 1-way ANOVA with Tukey’s
   multiple comparisons or (F) Mann-Whitney U test (unpaired, 2-sided).
   Adjusted P value *P < 0.05, **P < 0.01, ****P < 0.0001. Error bars
   indicate mean ± SEM.

   T[[Dysf]] cells showed significant upregulation of
   exhaustion-associated genes, including inhibitory receptors shared with
   Effector cells (Pdcd1, Havcr2, Tigit, Lag3), and surface markers and
   transcription factors (TFs) linked to terminal exhaustion (Cd101, Cd38,
   Tox, Nr4a1/2) ([159]Figure 3B and [160]Supplemental Figure 5C). They
   also upregulated Itgae (encoding CD103, known to be induced by TGF-β1
   signals) ([161]37) and negative regulators of TGF-β1 signaling (Smurf1,
   Ski, Skil, Pmepa1) ([162]38). Notably, T[[Dysf]] cells retained RNA
   expression of some effector molecules (Ifng, Gzma, Gzmb, Gzmk), though
   discordance between Ifng RNA and protein levels may reflect
   posttranscriptional regulation, impaired protein synthesis, or the
   chronic stimulation conditions under which they were isolated, in
   contrast with acutely stimulated cells, which were resting before
   transcriptional analysis. Conversely, acutely stimulated cells retained
   higher expression of stemness- and memory-associated genes (e.g., Sell,
   Ccr7, Tcf7, Lef1, Cd27), along with T cell migration– and
   chemotaxis-linked genes (e.g., S1pr1, Klf2, Cxcr3, Cxcr4, S1pr4)
   ([163]Figure 3B and [164]Supplemental Figure 5C).

   To further determine whether the upregulated DEGs within T[[Dysf]]
   cells recapitulate the transcriptional signature of TEX cells in vivo,
   we performed gene set enrichment analysis (GSEA) using progenitor,
   cytolytic, and terminally TEX subsets from chronic LCMV infection
   ([165]28). Importantly, T[[Dysf]] cells were enriched for the
   terminally exhausted signature, while Acute-2d and Acute-7d cells
   closely resembled progenitor TEX cells, and day 7 Effector P14 cells
   aligned with the cytolytic subset signature ([166]Figure 3C and
   [167]Supplemental Figure 5, D and E). Comparing single-cell
   transcriptional signatures of human tumor-infiltrating lymphocytes
   (TILs) from 21 types of cancer ([168]33), we found that T[[Dysf]] cells
   resembled TEX subsets, including KIR^^+TXK^^+ NK-like and terminal
   exhausted TILs ([169]Supplemental Figure 6A), while Acute-7d P14 cells
   aligned with functional subsets of TILs, closely resembling terminally
   differentiated effector memory T cell signature ([170]Supplemental
   Figure 6B). To gain insights into the biological functions of
   upregulated genes in Acute-7d versus T[[Dysf]] cells, we performed Gene
   Ontology and Hallmark GSEA, which revealed significant enrichment of
   pathways associated with T cell activation, proliferation, adhesion,
   IL-2/TNF/IFNG cytokine signaling, and glycolysis in acutely stimulated
   T cells ([171]Figure 3D and [172]Supplemental Figure 6C). However,
   upregulated genes in T[[Dysf]] cells were enriched for cell
   cycle–related processes, E2F targets, and G2M checkpoint hallmarks
   ([173]Supplemental Figure 6, D–G), previously linked with dysfunctional
   CD8^^+ T cells ([174]9, [175]39). Thus, T[[Dysf]] cells
   transcriptionally resemble terminally TEX cells in chronic infection
   and cancer.

   TEX cells also exhibit impaired protein translation during metabolic
   stress ([176]40). To evaluate protein translational activity in TEX
   cells from chronic LCMV infection, we performed a puromycin
   incorporation assay, in which puromycin is incorporated into nascent
   polypeptide chains during active protein synthesis ([177]41).
   Terminally exhausted virus-specific CD8^^+ T cells had significantly
   reduced puromycin incorporation compared with progenitor or cytolytic
   TEX subsets ([178]Figure 3E). Similarly, T[[Dysf]] cells restimulated
   in vitro on day 19 showed reduced puromycin incorporation compared with
   Acute-7d cells ([179]Figure 3F and [180]Supplemental Figure 6H).
   Importantly, puromycin levels positively correlated with Ifnγ and PD-1
   protein expression in both groups ([181]Supplemental Figure 6, I and
   J). Together, these results demonstrate that T[[Dysf]] cells acquire
   the transcriptional features of terminally TEX cells in mice or humans
   and also impaired protein synthesis, which may explain the discordant
   RNA-protein expression patterns observed in TEX cells ([182]42).

Stable CD8^^+ T cell dysfunction requires prolonged stimulation exceeding 2
weeks.

   The hallmark of terminal CD8^^+ T cell exhaustion is the stability of
   the dysfunctional state ([183]3, [184]10). To determine whether
   T[[Dysf]]cells adopt this irreversible phenotype, we performed recovery
   experiments in which chronically stimulated P14 cells were withdrawn
   from antigenic stimulation and TGF-β1 signaling at different time
   points and allowed to rest under homeostatic conditions ([185]Figure
   4A). We first tested whether 12 days of chronic stimulation was
   sufficient to establish stable dysfunction by terminating GP33 and
   TGF-β1 exposure on day 12, followed by a 7-day rest until day 19
   (T[[Dysf]]-12d condition). T[[Dysf]]-12d cells demonstrated substantial
   functional recovery, with restored effector cytokine production and
   degranulation activity, reflected by increased frequencies of
   Ifnγ^^+Tnf^^+ and Tnf^^+CD107a^^+ cells, comparable to acutely
   stimulated cells ([186]Figure 4, C and D). This was accompanied by
   improved cell viability ([187]Figure 4B). In contrast, T[[Dysf]] cells
   rested from day 19 to day 26 failed to regain effector function and
   maintained reduced cytokine and degranulation capacity ([188]Figure 4,
   C and D). The viability of T[[Dysf]] cells also declined throughout the
   resting period ([189]Figure 4B).

Figure 4. Stable CD8^+ T cell dysfunction requires prolonged stimulation
exceeding 2 weeks.

   [190]Figure 4
   [191]Open in a new tab

   (A) Schematic showing resting phase of T[Dysf] P14 cells under
   homeostatic conditions. Naive P14 CD8^+ T cells were stimulated with
   GP33 peptide and TGF-β1 for either 12 days (T[Dysf]-12d) or 19 days
   (T[Dysf]), then rested in GP33 and TGF-β1–free medium for 7 days. (B)
   Bar graph depicting frequency of viable P14 cells comparing T[Dysf]-12d
   pre-rest (day 12) and post-rest (day 19) versus T[Dysf] pre- and
   post-rest (day 19 and 26, respectively). (C and D) Functional
   assessment of polyfunctional cytokine production. Representative FACS
   plots and bar graphs showing frequencies of (C) Ifnγ^+Tnf^+ and (D)
   Tnf^+CD107a^+ P14 cells following GP33 peptide rechallenge for P14
   CD8^+ T cells on day 12, 19, and 26 following resting for 7 days. (E)
   Bar graph showing frequency of Total Il7r^+ P14 cells. (F) Expression
   level (normalized gMFI) of Ly108, (G) Gzmb, (H) PD-1, (I) Tim3, (J)
   CD101, and (K) CD103 within P14 cells normalized to Acute-7d on day 19.
   All n = 4 biological replicates, representative of 2 to 3 independent
   experiments. Adjusted P value *P < 0.05, **P < 0.01, ***P < 0.001,
   ****P < 0.0001. Comparisons were determined by 1-way ANOVA with Tukey’s
   multiple comparisons (B–K). Error bars indicate mean ± SEM.

   Phenotypic analysis further distinguished reversible from stable
   dysfunction. T[[Dysf]]-12d cells downregulated exhaustion markers
   (PD-1, Tim3, CD101, CD103, and Gzmb) and reexpressed the
   memory-associated markers Ly108 and Il7r, though CD62L recovery
   remained modest ([192]Figure 4, E–K, and [193]Supplemental Figure 7, A
   and B), suggesting a flexible dysfunctional state. In contrast,
   T[[Dysf]]cells rested from day 19 retained high expression of
   exhaustion markers and failed to reexpress Il7r and Ly108 ([194]Figure
   4, E and F, and [195]Supplemental Figure 7A), indicating phenotypic
   rigidity and stable commitment to terminal exhaustion. To further
   validate the stability of this state, we extended the recovery period
   to 11 days of resting (day 19 to day 30). Even after prolonged rest,
   T[[Dysf]] cells failed to restore cytokine production, degranulation,
   or expression of memory markers, while exhaustion markers remained
   elevated ([196]Supplemental Figure 7, C–K). These findings reveal a
   threshold in stimulation duration for stable exhaustion programming.
   While less than 2 weeks of chronic stimulation results in reversible
   functional impairment, extending stimulation to about 3 weeks
   establishes irreversible reprogramming characteristic of terminal
   dysfunction. The inability of T[[Dysf]] cells to recover function or
   memory traits, even after prolonged rest, validates our in vitro model
   as a faithful representation of the irreversible T cell exhaustion
   observed in vivo.

Dysfunctional P14 cells recapitulate the heterogeneity of exhausted CD8^^+ T
cells generated in vivo.

   T cell dysfunction is a gradual process resulting in heterogeneous
   populations. On day 19, we identified distinct subsets: a dominant
   PD-1^^hiTim3^^+ population and a minor PD-1^^intTim3^^– subset,
   suggesting that heterogeneity arises during chronic stimulation
   ([197]Figure 5, A and B, and [198]Supplemental Figure 8A). Resting
   Acute-7d P14 cells downregulated PD-1 and Tim3, leading to the
   emergence of a PD-1^^loTim3^^– subset ([199]Figure 5A). Across all
   these subsets, acutely stimulated P14 cells maintained significantly
   higher expression levels of effector cytokines and memory markers
   (Ifnγ, Tnf, CD107a, Ly108, Il7r, and CD62L) compared with T[[Dysf]]
   cells ([200]Supplemental Figure 8, B–D and H). Remarkably, the
   PD-1^^hiTim3^^+ T[[Dysf]] subset exhibited reduced Perforin and Ki67
   ([201]Supplemental Figure 8, F and G), while maintaining heightened
   expression of Tox ([202]43) ([203]Supplemental Figure 8E), consistent
   with terminal dysfunction.

Figure 5. Dysfunctional CD8^+ T cells recapitulate the heterogeneity of
exhausted CD8^+ T cells generated in vivo.

   [204]Figure 5
   [205]Open in a new tab

   (A) Representative FACS plots of PD-1 and Tim3 expression for P14 CD8^+
   T cells on both day 7 (Effector 7d-stim) and day 19 (Acute-7d or
   Chronic GP33^+TGF-β1: T[Dysf]) after rechallenging with GP33 peptide.
   (B) Summary bar graph showing % of PD-1– and Tim3-expressing subsets on
   P14 CD8^+ T cells on day 19. (C) Schematic for the resting phase of
   T[Dysf] P14 cells under homeostatic conditions. (D) Representative FACS
   plot of PD-1 and Tim3 expression for P14 CD8^+ T cells on day 23
   following resting for 4 days. (E) Longitudinal tracking on day 19 and
   23 of the frequency of PD-1^intTim3^– (purple) and PD-1^hiTim3^+ (red)
   P14 cells, (F) frequency of CD107a^+Ifnγ^+ P14 cells, and (G) frequency
   of Tnf^+Ifnγ^+ P14 cells within PD-1^intTim3^– and PD-1^hiTim3^+
   subsets. (H) Expression levels (gMFI) of TOX, (I) PD-1, (J) Tim3, (K)
   CD103, (L) Cx3cr1, and (M) Ki67 within PD-1^intTim3^– and PD-1^hiTim3^+
   subsets on day 23. (N) Bar graph showing frequencies of divided P14
   cells with ≥3 proliferation cycles by CFSE staining. All n = 4
   biological replicates, representative of 2 to 3 independent
   experiments. Adjusted P value ^#P < 0.05 for PD-1^hiTim3^+ cells (F),
   **P < 0.01, ***P < 0.001, ****P < 0.0001. Comparisons were determined
   by 1-way ANOVA (B), 2-way ANOVA (E–G), or Mann-Whitney U test
   (unpaired, 2-sided) (H–N). Error bars indicate mean ± SEM.

   To test stability of T[[Dysf]] P14 subsets, we CFSE-labeled and rested
   T[[Dysf]] cells for 4 days (day 19–23, [206]Figure 5C). The
   PD-1^^hiTim3^^+ subset remained severely dysfunctional with high
   expression of inhibitory receptors and Tox ([207]Figure 5, D–J, and
   [208]Supplemental Figure 8, I–L). In contrast, PD-1^^intTim3^^– cells
   recovered polyfunctionality and expressed higher levels of Cx3cr1, a
   marker of the cytolytic TEX subset, while PD-1^^hiTim3^^+ cells
   retained CD103 expression ([209]Figure 5, K and L). PD-1^^intTim3^^–
   cells also showed greater homeostatic proliferation, with higher
   proportions undergoing at least 3 division cycles during rest
   ([210]Figure 5, M and N). Together, these findings show that 3 weeks of
   chronic stimulation drives stable CD8^^+ T cell dysfunction, with the
   dominant PD-1^^hiTim3^^+ subset exhibiting features of terminal
   exhaustion.

Terminal dysfunction of in vitro chronically stimulated CD8^^+ T cells is
stabilized by exhaustion-specific epigenetic programs.

   T cell exhaustion is marked by distinct epigenetic changes that lock
   terminally exhausted CD8^^+ T cells into a stable dysfunctional state
   ([211]3, [212]7–[213]14). This remodeling involves loss of chromatin
   accessibility at effector function–, stemness-, and memory-associated
   genes while maintaining accessibility at exhaustion-linked loci
   ([214]11, [215]23, [216]44). To define these changes within our in
   vitro–generated P14 cells, we performed ATAC-Seq on Acute-7d and
   T[[Dysf]] (PD-1^^hiTim3^^+) P14 cells isolated on day 19 ([217]Figure
   6A). T[[Dysf]] cells acquired a distinct open chromatin landscape, with
   over 24,600 differentially open chromatin regions (OCRs; P < 0.01, fold
   change ≥ 2), including ~13,900 regions with reduced accessibility and
   ~10,700 regions with increased accessibility relative to Acute-7d cells
   ([218]Figure 6B). These OCRs were mostly intronic and intergenic
   ([219]Supplemental Figure 9, A and B), suggesting regulatory roles.
   Regions with increased accessibility in T[[Dysf]] cells included TGF-β1
   signaling genes (e.g., Tgfb1, Tgfbr2, Pmepa1, Skil, Smad9), inhibitory
   receptors (e.g., Pdcd1, Cd200r2), MAPK pathway genes (e.g., Dusp16,
   Map2k2, Map2k4), and exhaustion-associated TFs (Nr4a3, Stat3, Irf8)
   ([220]45, [221]46) ([222]Figure 6B). In contrast,
   decreased-accessibility regions included stemness/memory genes (e.g.,
   Ccr7, Sell, Slamf6, Cxcr5, and Il15ra) and effector/TCR signaling genes
   (e.g., Cd28, Cd44, Il2ra, Icos, Icosl, Itgal, Ifngr1, and Ifngr2)
   ([223]2, [224]29, [225]47) ([226]Figure 6B). Pathway enrichment
   analysis showed that closed OCRs in T[[Dysf]] cells were enriched for
   IL-2/STAT5, TCR signaling, Ifnγ, and inflammatory response pathways,
   while open OCRs were enriched for PI3K/AKT/mTOR, Hedgehog and Notch
   signaling, and GTPase activity pathways ([227]Supplemental Figure 9,
   C–F). These results demonstrate extensive epigenetic remodeling in
   T[[Dysf]] cells, consistent with a stable, epigenetically reprogrammed
   dysfunctional state.

Figure 6. Terminal dysfunction of in vitro–generated CD8^+ T cells is
stabilized by exhaustion-associated epigenetic programs.

   [228]Figure 6
   [229]Open in a new tab

   (A) Schematic for ATAC-sequencing of P14 cells from the models
   described in Figure 2A and Figure 5C. T[Dysf] cells were FACS-sorted on
   day 19 of chronic stimulation as the PD-1^+Tim3^+ (double-positive,
   “T[Dysf]-DP”) and rested for 4 days. Rested T[Dysf]-DP cells were
   sorted on day 23 as PD-1^+Tim3^– (single-positive, “Rest-SP”) or
   PD-1^+Tim3^+ (double-positive, “Rest-DP”). (B) Heatmap showing
   differentially open chromatin regions (OCRs) between Acute-7d and
   T[Dysf]-DP cells (day 19) plotted based on relative Z-score with
   example genes listed. Red indicates “open” and blue indicates “closed”
   regions. (C) PCA comparing chromatin accessibility profiles for in
   vitro Acute-7d or T[Dysf] subsets to published signatures for in vivo
   memory or exhausted CD8^+ T cells (23). (D) Representative snapshots of
   accessible chromatin peaks (mapped in Integrative Genomics Viewer, IGV)
   within in vitro P14 or in vivo naive, memory, progenitor, or exhausted
   CD8^+ T cells (44) at gene loci for Ccr7, (E) Ifng, and (F) Pdcd1. (G)
   Alluvial plot tracking the number of chromatin regions from Acute-7d
   cells that became open/closed in T[Dysf]-DP (day 19) compared with
   Acute-7d and regions from T[Dysf]-DP that remained unchanged or became
   open/closed in rested T[Dysf] subsets (day 23). (H) Venn diagram
   comparing transcription factor (TF) motifs enriched within closed OCRs
   in T[Dysf]-DP cells and regions that became open in Rest-SP cells or
   (I) open OCRs in T[Dysf]-DP cells and regions that became closed in
   Rest-SP cells. (J) Average % CpG methylation at Tcf7 and (K) Pdcd1
   differentially methylated regions (DMRs) within naive or in
   vitro–generated P14 cells (day 19). N = 2 biological replicates,
   representative of 2 to 3 independent experiments. Statistical
   significance was determined by DESeq2 (B and G) or PCA (C) using Partek
   software. Comparisons in J and K were determined by repeated measures
   1-way ANOVA with Tukey’s multiple comparisons. Adjusted P value **P <
   0.01, ***P < 0.001, ****P < 0.0001.

   To determine whether this epigenetic landscape mirrors in vivo
   exhaustion-specific programming, we compared ATAC-Seq profiles of
   T[[Dysf]] and Acute-7d P14 cells to virus-specific CD8^^+ T cells from
   acute and chronic LCMV infections ([230]23). PCA of the top 5,000 most
   variable peaks showed that T[[Dysf]] cells clustered with terminally
   exhausted cells, while Acute-7d cells aligned more with memory CD8^^+ T
   cells ([231]Figure 6C). Importantly, OCRs between T[[Dysf]] and
   Acute-7d cells reflected patterns observed in terminally exhausted
   (PD-1^^+Tim3^^+CXCR5^^–) versus progenitor (PD-1^^+Tim3^^–CXCR5^^+) TEX
   cells or memory CD8^^+ T cells ([232]44). For instance, T[[Dysf]] cells
   exhibited similarly reduced accessibility at memory/progenitor
   TEX-associated genes (e.g., Ccr7, Ifng, Tnf, Cxcr5, Sell, Slamf6, Il7r,
   Tcf7) and increased accessibility at terminal exhaustion genes (e.g.,
   Pdcd1, Havcr2, Cd244a) ([233]Figure 6, D–F, and [234]Supplemental
   Figure 9, I–Q). Thus, T[[Dysf]] cells undergo exhaustion-specific
   chromatin remodeling consistent with terminally TEX cells in vivo,
   while Acute-7d cells retain a memory-like open chromatin landscape.

   To assess stability after antigen stimulation withdrawal, we performed
   ATAC-Seq on PD-1^^hiTim3^^+ cells (“double-positive”; Rest-DP) and
   PD-1^^intTim3^^– (“single-positive”; Rest-SP) P14 cells isolated on day
   23 following withdrawal from chronic stimulation ([235]Figure 5, C and
   D, and [236]Figure 6A). Both populations retained the chromatin
   accessibility profiles of T[[Dysf]] and terminally TEX cells
   ([237]Figure 6, C–F, and [238]Supplemental Figure 9, I–Q), with ~92.8%
   of closed and ~81.9% of open T[[Dysf]]-specific OCRs unchanged
   ([239]Figure 6G and [240]Supplemental Figure 10, A and B), indicating
   strong epigenetic stability. However, Rest-SP cells showed partial
   reversal in a subset of regions: ~5.9% (820 OCRs) of closed regions
   regained accessibility, including Cd44, Sell, Btla, and Icos; TFs
   (Izkf1, Ikzf3, Bach2, Stat4, Zeb2, Klf12, and Foxo3); and cytokine
   receptors (Il12rb2, Il18r1) ([241]Figure 6G and [242]Supplemental
   Figure 10, A–C). Conversely, ~16.7% (~1,790 OCRs) of open regions
   became less accessible, including peaks at TFs (e.g., Bcl2, Tox, Irf4,
   Prdm2, and Foxp1), histone-modifying enzymes (e.g., Ezh2 and Hdac5),
   and signaling molecules (e.g., Mapk10, Wnt3, Wnt11, Notch4, and Tbk1)
   ([243]Figure 6G and [244]Supplemental Figure 10, A–C). Reopened regions
   in Rest-SP cells were enriched for IL-2 and TCR signaling pathways,
   while regions that lost accessibility were enriched for GTPase activity
   and MAPK signaling pathways ([245]Supplemental Figure 9, G and H).
   These results demonstrate that T[[Dysf]] cells retain a largely stable,
   epigenetically “scarred” chromatin state after removal of chronic
   signals — particularly in the PD-1^^hiTim3^^+ cell population — closely
   resembling the fixed epigenetic profile of terminally TEX cells in vivo
   ([246]11, [247]23, [248]44).

   To gain insights into mechanisms underlying this stability, we
   performed TF motif enrichment analysis on T[[Dysf]]-specific OCRs that
   remained unchanged following resting. Stable closed OCRs were enriched
   for motifs of TFs regulating effector and memory programs (e.g., BATF,
   Bach2, AP1, NF-κB, c-Myc, KLF, and STATs), while reopened peaks in
   Rest-SP cells were enriched for Tcf7, LEF1, KLF1/3, and Foxo1/3 motifs
   ([249]Figure 6H), suggesting limited reengagement of memory-regulatory
   TFs. Conversely, stable open OCRs in Rest-SP and Rest-DP cells were
   enriched for exhaustion-linked motifs (e.g., Elk1/4, Smad, IRF2/3, and
   Tbx21), while Rest-SP cells exhibited reduced accessibility at OCRs
   containing NFAT, Nur77, IRF4, RUNX1/2, and STAT1/5 motifs, suggesting
   partial rewiring of TCR and cytokine-dependent signaling ([250]Figure
   6I). Overall, these findings reveal that specific TF networks preserve
   the dysfunctional epigenetic state by maintaining accessibility at
   exhaustion-related motifs and closing memory- and effector-regulatory
   regions, limiting recovery even after stimulation withdrawal.

   Our previous work demonstrated that de novo DNA methylation drives
   epigenetic silencing of memory and effector programs within TEX cells
   ([251]8, [252]9). Additionally, specific DNA demethylation changes at
   the Pdcd1 locus promote PD-1 upregulation in TEX cells, beginning in
   the effector phase of CD8^^+ T cell differentiation ([253]26, [254]48).
   To determine whether T[[Dysf]] cells acquire distinct DNA methylation
   programs associated with stable exhaustion, we performed targeted DNA
   methylation analysis at key differentially methylated genomic regions
   (DMRs) ([255]8) ([256]Supplemental Figure 11, A–C). Our analysis
   revealed that T[[Dysf]] cells acquired substantial de novo DNA
   methylation at both Tcf7 and Ccr7 loci, consistent with downregulation
   of Tcf1 and Ccr7 proteins ([257]Figure 6J and [258]Supplemental Figure
   11D), while acutely stimulated cells recovered both molecules by day 19
   ([259]Supplemental Figure 11, E–I). In contrast, the Pdcd1 DMR was
   demethylated in both groups ([260]Figure 6K), consistent with DNA
   demethylation at Pdcd1 during early effector differentiation ([261]26).
   However, PD-1 remained higher on T[[Dysf]] cells on day 19, likely due
   to sustained antigen exposure ([262]26, [263]27, [264]49)
   ([265]Supplemental Figure 11, E–G). Thus, acquisition of de novo DNA
   methylation at stemness and memory-associated loci limits recovery,
   paralleling epigenetic constraints in these programs within terminally
   exhausted CD8^^+ T cells in vivo.

Dysfunction of in vitro–generated T[[Dysf]] cells is irreversible during in
vivo rechallenge.

   Finally, we assessed the in vivo recall responses of our in
   vitro–generated T[[Dysf]] cells, as a key hallmark of T cell exhaustion
   is the compromised ability to respond after antigen rechallenge
   ([266]11, [267]24). We sorted congenically distinct Thy1.1^^+ P14 cells
   from T[[Dysf]] or Acute-7d groups on day 19 and adoptively transferred
   them into naive C57BL/6 animals. One day later, mice were challenged
   with acute LCMV infection, and P14 recall responses were assessed after
   6–7 days ([268]Figure 7A). While Acute-7d cells underwent extensive
   proliferation, T[[Dysf]] cells showed impaired expansion ([269]Figure
   7B), as well as sustained Ly108 downregulation ([270]Figure 7C). Upon
   ex vivo GP33 peptide restimulation, T[[Dysf]] cells exhibited reduced
   effector functions (Ifnγ and Perforin), while Acute-7d cells maintained
   heightened polyfunctionality ([271]Figure 7, D and E). T[[Dysf]] cells
   also expressed higher levels of exhaustion-associated molecules, such
   as Tox and Gzmb, in spleen, liver, and lungs ([272]Figure 7, F and G,
   and [273]Supplemental Figure 12, A and B). To test whether resting
   T[[Dysf]] cells improves their recall capacity, we adoptively
   transferred T[[Dysf]] P14 cells after a 7-day rest period, followed by
   acute LCMV infection ([274]Figure 7H). Despite rest, T[[Dysf]] cells
   remained impaired in proliferation and expression of Ly108 and Il7r
   ([275]Figure 7, I and J, and [276]Supplemental Figure 12C), showed
   diminished effector function (e.g., Ifnγ, Tnf, and CD107a) ([277]Figure
   7K), and retained high levels of exhaustion markers (PD-1, Tim3, Gzmb,
   and Tox) ([278]Figure 7, L–N, [279]Supplemental Figure 12, D–F).
   Together, these findings demonstrate that ~3 weeks of in vitro chronic
   stimulation drives CD8^^+ T cells into a stable, terminally
   dysfunctional program that mirrors in vivo terminal exhaustion,
   including severely impaired recall capacity.

Figure 7. Dysfunction of in vitro–generated T[Dysf] cells is irreversible
during in vivo rechallenge.

   [280]Figure 7
   [281]Open in a new tab

   (A) Schematic for adoptive transfer of “Acute-7d” (blue) or “T[Dysf]”
   (red) P14 cells into congenically distinct C57BL/6 mice on day 19,
   followed by acute LCMV infection and analysis on day 6–7 postinfection.
   (B) Total P14 cell numbers from spleens, livers, and lungs. Panels C–G
   and I–N are for spleens. (C) Bar graph showing expression level (gMFI)
   of Ly108 or (D) % Ifnγ^+ P14 cells after ex vivo GP33 peptide
   rechallenge. (E) Representative FACS plots showing Ifnγ and Perforin
   expression and summary bar graph showing % Ifnγ^+Perforin^+ P14 cells.
   (F) Bar graph showing expression level (gMFI) of Gzmb or (G) TOX. (H)
   Schematic for adoptive transfer of “Rested Acute-7d” (dark blue) or
   “Rested T[Dysf]” (dark red) P14 cells on day 26 into congenically
   distinct C57BL/6 mice, followed by acute LCMV infection and analysis on
   day 6–7 postinfection. (I) Bar graph showing total number of P14 cells
   isolated from spleens or (J) expression level (gMFI) of Ly108. (K)
   Representative FACS plots showing Ifnγ and Tnf or CD107a expression and
   summary bar graph showing % of Ifnγ^+Tnf^+ or Ifnγ^+CD107a^+ P14 cells.
   (L) Representative FACS plots showing PD-1 and Tim3 expression and (M)
   summary bar graph showing % of PD-1^+Tim3^+ P14 cells. (N) Bar graph
   showing the expression level (gMFI) of Gzmb. For all FACS plots (E, K,
   and L), Acute-7d plots are representative of 1 recipient mouse, while
   the T[Dysf] plots show data combined from all T[Dysf] recipient mice.
   For B, data were pooled from 2 independent experiments with n = 3–5
   biological replicates per group for each experiment. For all other
   panels, n = 4–6 biological replicates from 2 to 3 independent
   experiments. Adjusted P value *P < 0.05, **P < 0.01, ***P < 0.001,
   ****P < 0.0001. Comparisons were determined by Mann-Whitney U test
   (unpaired, 2-sided). Error bars indicate mean ± SEM.

Discussion

   Epigenetic scarring of T cell exhaustion programs is a major
   cell-intrinsic barrier to T cell–based immunotherapies. Faithful model
   systems for T cell exhaustion are essential for understanding the
   individual and cooperative contributions of factors driving the
   epigenetic progression toward terminal exhaustion. While common animal
   models of T cell exhaustion, such as chronic LCMV infection or tumor
   models, are among the most reliable systems for these investigations,
   they present inherent complexities that make it challenging to dissect
   the precise contributions of individual factors to the epigenetic
   programming of TEX cells. Therefore, there is a major need to develop
   reliable reductionist approaches that enable precise control and
   isolation of individual signals, facilitating a deeper understanding of
   how specific factors influence the programming events underlying T cell
   exhaustion.

   Recent studies have introduced new in vitro models of T cell
   dysfunction through repeated polyclonal anti-CD3 or antigen-specific
   stimulation of murine CD8^^+ T cells ([282]18–[283]21). However, a key
   limitation has been the poor survival and widespread cell death of
   repeatedly TCR-stimulated CD8^^+ T cells, preventing chronic
   stimulation from extending beyond 5–10 days ([284]20). This restricts
   the ability to replicate in vivo settings, such as chronic viral
   infections, where CD8^^+ T cells experience prolonged stimulation over
   ~3 weeks before developing a terminally dysfunctional state ([285]2,
   [286]24). Indeed, we found that ~1–2 weeks of repeated stimulation
   induces some exhaustion-like features, marked by increased coexpression
   of inhibitory receptors; however, these conditions fail to establish
   epigenetically stable terminal dysfunction. Instead, after acute
   antigen stimulation is removed, CD8^^+ T cells rapidly recover effector
   functions and memory programs. These findings align with the
   demonstrated plasticity of day 8–15 effector CD8^^+ T cells, which can
   regain function and memory potential when removed from chronic
   LCMV-infected environments and transferred to antigen-free conditions
   ([287]24). Thus, short-term antigen stimulation alone is insufficient
   for inducing terminal dysfunction; instead, it drives effector
   differentiation with preserved memory potential, mimicking CD8^^+ T
   cell responses during acute viral infections, during which viruses are
   typically cleared within 1–2 weeks ([288]50).

   Here, we have developed a faithful animal cell model for terminal
   exhaustion by delivering chronic TGF-β1 signals between days 7 and 19
   to chronically antigen-stimulated P14 CD8^^+ T cells. This in vitro
   model builds on our previous work showing that chronic, post-effector
   TGF-β1 signaling is necessary for driving terminal dysfunction in
   chronically TCR-stimulated human CD8^^+ T cells ([289]9). Here, we
   demonstrate that prolonged TGF-β1 signals are critical for maintaining
   the survival of chronically antigen-stimulated P14 (T[[Dysf]]) cells
   while also promoting an epigenetically stabilized, terminally
   dysfunctional state. Under these conditions, T[[Dysf]] cells exhibited
   the full range of phenotypic, functional, transcriptional, epigenetic,
   and metabolic characteristics of terminal exhaustion. The stability of
   this terminal exhaustion program was validated by resting T[[Dysf]]
   cells following antigen withdrawal, which revealed persistent
   epigenetic “scarring” of the dysfunctional state. A limited degree of
   epigenetic flexibility was observed only within the minor
   PD-1^^intTim3^^– subset that emerged after resting T[[Dysf]] cells.
   Furthermore, adoptive transfer studies of T[[Dysf]] cells — both before
   and after resting — demonstrated that T[[Dysf]] cells remained
   terminally dysfunctional, as they failed to regain effector functions
   or proliferate in response to acute LCMV infection.

   Our tractable model of T cell exhaustion described here offers several
   advantages for studying molecular mechanisms that underline terminal
   dysfunction. Specifically, the ex vivo culture model (a) enables
   tracking of the developmental trajectory of T cell exhaustion over
   extended periods; (b) facilitates mechanistic studies of various
   signals from infected tissue or tumor microenvironments; (c) supports
   high-throughput molecular investigations of terminal T cell exhaustion,
   improving both the technical and economic efficiency, compared with
   genetic or therapeutic screening studies performed in vivo; (d)
   provides a more reliable tool for identifying key factors that
   stabilize the dysfunctional state in CD8^^+ T cells; and (e) offers
   physiologically relevant conditions for studying antigen-specific T
   cell responses by using antigen-dependent stimulation, rather than
   polyclonal, CD3-dependent stimulation. The T[[Dysf]] model can be
   leveraged to design a discovery platform for the development of more
   effective immunotherapies aimed at restoring the function of severely
   dysfunctional T cells, ultimately enhancing treatment outcomes for
   patients with cancer or chronic infections.

Methods

Sex as a biological variable.

   Our study was designed by accounting for sex as a biological variable.
   We examined both male and female animals, and similar findings are
   reported for both sexes.

Animals.

   All mice used in the LCMV infection experiments from this study were
   housed at The Ohio State University animal facilities under the
   IACUC-approved guidelines. This facility included an ambient
   temperature kept in a range of 68°C–76°C, relative humidity 30%–70%,
   and a 12-hour dark/12-hour light cycle (light 6 am to 6 pm). Mice were
   fed standard Teklad 7912 chow (Envigo). Transgenic mice expressing a
   TCR specific for the LCMV peptide GP33 (P14 donor mice) were bred
   in-house at The Ohio State University on a C57BL/6 background.
   Recipient C57BL/6 mice were purchased from The Jackson Laboratory and
   used at approximately 8 weeks of age. Recipient mice were sex-matched
   with donor mice.

In vitro chronic stimulation of mouse P14 CD8^^+ T cells.

   On day 0 of the in vitro chronic stimulation model, CD8^^+ T cells were
   isolated from spleens of healthy, uninfected male or female P14
   transgenic mice (≥8 weeks old). Spleens were mashed through a 100 μm
   filter and subjected to RBC lysis. Splenocytes were enriched for CD8^^+
   T cells using the EasySep Mouse CD8^^+ T Cell Isolation Kit (StemCell
   Technologies), then stained for spectral cytometry to confirm purity,
   viability, and GP33 tetramer specificity. CD8^^+ T cells were seeded
   into 6-well, flat-bottom, tissue culture–treated plates coated with
   mouse purified anti-CD3 (5 μg/mL; clone: 17A2; catalog 100102) and
   anti-CD28 (10 μg/mL; clone: 37.51; catalog 02102) (both from
   BioLegend), at a density of 1 × 10^^6 to 3 × 10^^6 CD8^^+ T cells/well
   in 3 mL of complete RPMI medium (10% fetal bovine serum and 1×
   Penicillin/Streptomycin from Gibco, 200 IU/mL of recombinant human IL-2
   from PeproTech, and 50 μM of 2-Mercaptoethanol from MilliporeSigma) to
   stabilize IL-2 activity and CD8^^+ T cell priming ([290]51). Plates
   were incubated at 37°C and 5% CO[[2]] for 2 days to allow T cell
   activation.

   On day 2, activated P14 CD8^^+ T cells were pooled, washed, and
   resuspended in T cell media (complete RPMI medium containing 20 IU/mL
   of recombinant human IL-2 and 10 ng/mL of recombinant human IL-15 from
   PeproTech). Cells were seeded into 96-well, U-bottom plates at 1 ×
   10^^5 to 2 × 10^^5 viable cells/well. For the 2d-stimulated condition,
   cells were cultured without GP33 peptide. For Acute-7d and Chronic
   groups, GP33 peptide (0.3 μg/mL) was added. On day 7, Acute-7d cells
   were maintained in peptide-free T cell media. For the Chronic + TGF-β1
   T[[Dysf]] condition, recombinant human TGF-β1 (5 ng/mL; PeproTech) was
   added to the GP33 peptide–containing medium beginning on day 7.
   Anti-CD28 was omitted during the chronic stimulation phase to simulate
   the declining costimulatory environment during chronic viral infections
   and tumors, where CD28 expression is progressively lost as TEX cells
   become terminally exhausted. Media were replenished every 2–3 days with
   fresh GP33 peptide and TGF-β1 until day 19.

   On day 19 of chronic stimulation, P14 cells were sorted for viable
   CD8^^+ T cells (Sony MA900 Cell Sorter) and used for adoptive transfer,
   in vitro recovery experiments, or ATAC-Seq or snap-frozen at –80°C to
   use for RNA and DNA methylation sequencing analyses.

   For recovery experiments, sorted P14 CD8^^+ T cells from day 12 or 19
   of chronic stimulation were seeded at 5 × 10^^4 cells/well in complete
   RPMI medium containing IL-15 (10 ng/mL). Cells were cultured for 4–12
   days and analyzed on days 23, 26, or 30 of in vitro culture.

Plate-bound in vitro stimulation of mouse P14 CD8^^+ T cells.

   Following a 2-day activation phase with mouse purified anti-CD3 (5
   μg/mL) and anti-CD28 (10 μg/mL) (BioLegend), activated P14 CD8^^+ T
   cells were resuspended in complete RPMI medium containing 20 IU/mL of
   recombinant human IL-2 and 10 ng/mL of recombinant human IL-15
   (PeproTech). P14 cells were then seeded into tissue culture–treated,
   96-well, flat-bottom plates coated with anti-CD3 (5 μg/mL) at a density
   of 1 × 10^^5 to 2 × 10^^5 viable cells per well. Cells were transferred
   to freshly coated anti-CD3 plates every 2 days and analyzed on day 5
   and 9.

Functional analysis of in vitro–stimulated P14 CD8^^+ T cells.

   The effector function and cytokine production of in vitro–stimulated
   P14 cells were analyzed on days 0, 2, 5, 7, 12, 19, 23, 26, and 30
   after initial activation. Cells were collected at these time points and
   stimulated for either 3 hours with PMA/ionomycin plus protein transport
   inhibitors (eBioscience Cell Stimulation Cocktail, Thermo Fisher
   Scientific) or for 5 hours with GP33 peptide (0.3 μg/mL) plus protein
   transport inhibitors (eBioscience Brefeldin A and Monesin, Thermo
   Fisher Scientific) at 37°C. A spectral cytometry antibody targeting
   mouse CD107a (Brilliant Violet 421, BioLegend) was added at the
   beginning of each stimulation to track CD8^^+ T cell degranulation
   activity. Stimulation was immediately followed by antibody staining for
   spectral cytometry analysis.

Puromycin incorporation assay.

   For puromycin incorporation analysis (to measure protein translation
   capacity), P14 cells were stimulated for 5 hours with GP33 peptide (0.3
   μg/mL) plus protein transport inhibitors as above. During the final 30
   minutes of incubation, puromycin was added to the media at a
   concentration of 10 μg/mL (MilliporeSigma). During intracellular
   staining following stimulation, anti-puromycin antibody (Alexa Fluor
   488, BioLegend) was used to label and measure puromycin uptake by
   spectral cytometry.

Adoptive transfer and in vivo rechallenge of in vitro–stimulated P14 CD8^^+ T
cells.

   For adoptive transfer experiments, antigen-experienced P14 cells were
   sorted on day 19 or on day 26 (following 7 days of rest) of the in
   vitro stimulation protocol to isolate the viable CD8^^+ T cells using
   the Sony MA900 Cell Sorter. Immediately following sorting, P14 cells
   from the 3 described conditions (Acute-2d, Acute-7d, and T[[Dysf]])
   were adoptively transferred into wild-type, congenically distinct
   (Thy1.1^^–Thy1.2^^+) C57BL/6 mice (male or female depending on P14 cell
   identity, age-matched at ≥8 weeks old) at 100,000–150,000 P14
   cells/mouse from 1 of the 3 conditions via i.v. retro-orbital
   injection. One day following adoptive transfer, recipient mice were
   infected with LCMV Armstrong virus (2 × 10^^5 PFU/mouse) via i.p.
   injection. Mice were euthanized on day 6–7 postinfection, and lymphoid
   (spleens) and peripheral tissues (livers and lungs) were harvested and
   processed into single-cell suspensions as previously described
   ([291]9). Tissue samples were then used for ex vivo GP33 peptide
   stimulation for 5 hours followed by spectral cytometry staining to
   assess the function and phenotype of the transferred P14 cells.

Spectral cytometry antibody staining.

   Single-cell suspensions were stained with a comprehensive panel, and
   data were collected using a Cytek Aurora 4-laser spectral cytometer and
   SpectroFlow software (version 3.3). Dead cells were stained with Ghost
   Dye Violet 510 (catalog 50-105-2992, Tonbo Biosciences, 1:600
   dilution). Mouse surface antibodies used for staining throughout this
   study include Brilliant Violet 421 anti-CD107a (LAMP-1; clone 1D4B,
   1:150 dilution; catalog 121617), Brilliant Violet 605 anti-CD62L
   (MEL-14, 1:400 dilution; catalog 104437), Brilliant Violet 605
   anti-CD103 (2E7, 1:400 dilution; catalog 121433), Brilliant Violet 650
   anti-Ccr7 (4B12, dilution 1:200; catalog 120137), Brilliant Violet 711
   anti-Cx3cr1 (SA011F11, 1:300 dilution; catalog 149031), Pacific Blue
   anti-Ly108 (SLAMF6; 330-AJ, 1:150 dilution; catalog 134608),
   PerCP/Cyanine5.5 anti-CD8a (53–6.7, 1:300 dilution; catalog 100734),
   Alexa Fluor 700 anti-CD44 (IM7, 1:400 dilution; catalog 103026),
   APC/Cyanine7 anti-CD279 (PD-1; 29F.1A12, 1:200 dilution; catalog
   135224), PE/Cyanine7 anti-Tim3 (RMT3-23, 1:200 dilution; catalog
   119716), PE anti-CD127 (Il7r; A7R34, 1:300 dilution; catalog 135009)
   (all from BioLegend), and Brilliant Violet 711 Annexin V (catalog
   563972, BD Horizon, 1:300 dilution). LCMV GP33-specific CD8^^+ T cells
   were detected by tetramerization of GP33 monomers (supplied by the NIH
   Tetramer Core Facility) conjugated to streptavidin-APC (eBioscience).

   Cells were then fixed and permeabilized for intracellular staining
   using the 10× Permeabilization Buffer (Invitrogen). Mouse intracellular
   antibodies used for staining throughout this study include APC
   anti–IFN-γ (XMG1.2, 1:200 dilution; catalog 505810), Brilliant Violet
   650 anti-TNF (MP6-XT22, 1:400 dilution; catalog 506333), Brilliant
   Violet 785 anti-Tbet (4B10, 1:200 dilution; catalog 644835) PE/Dazzle
   594 anti–Granzyme B (QA16A02, 1:200 dilution; catalog 372216), PE
   anti-Perforin (S16009A, 1:200 dilution; catalog 154306), Alexa Fluor
   488 anti-Puromycin (2A4, 1:200; catalog 381506) (all from BioLegend),
   Alexa Fluor 488 anti–TCF-7/TCF-1 (S33-966, 1:200 dilution; catalog
   567018, BD Biosciences), eFluor 660 anti-TOX (TXRX10, 1:200 dilution;
   catalog 50-6502-82, eBioscience), and Brilliant Ultra Violet 737
   anti–Ki-67 (SolA15, 1:150 dilution; catalog 367-5698-82, eBioscience).

Cytotoxicity assay of P14 CD8^^+ T cells.

   For in vitro cytotoxicity assay, GP33-expressing CT2A glioma tumor
   cells (a gift from Hiroshi Nakashima, Harvard University, Boston,
   Massachusetts, USA) ([292]52) were labeled with CFSE dye (1 μM, Sigma),
   then seeded into a 96-well, flat-bottom plate at a concentration of
   60,000 cells per well in complete RPMI medium. The tumor cells were
   incubated at 37°C and 5% CO[[2]] overnight to allow monolayer
   formation. The following day, each well was seeded with 60,000 P14
   cells collected from either the acutely or chronically stimulated
   conditions at the indicated time point. The cocultured tumor cells and
   P14 cells were then incubated overnight (18–20 hours), followed by
   staining for spectral cytometry. Tumor cell killing activity was
   determined by measuring the number of dead tumor cells per 10 viable
   CD8^^+ T cells by spectral cytometry.

RNA-Seq.

   RNA was isolated from sorted P14 cell pellets using the Arcturus
   PicoPure RNA Isolation Kit (catalog KIT0204, Applied Biosystems). RNA
   samples were submitted for library prep and sequencing by Azenta Life
   Sciences. RNA-Seq data were analyzed by Partek Flow software. The
   generated FASTQ files were used for prealignment quality
   assurance/quality control, then aligned to the mouse mm10 genome by
   STAR. The resultant aligned reads were quantified to the mm10 RefSeq
   Transcripts using Quantify to annotation model, followed by filtering
   to exclude features where maximum < 50.0. DESeq2 analysis was performed
   first by normalization of the filtered gene counts for median ratio to
   compare among the P14 conditions. DEGs were determined as genes with
   fold change ≥ 2 and P < 0.05 between the compared groups. The DEGs were
   plotted into heatmaps and used for PCA compared with the
   transcriptional signatures of TEX subsets from acute or chronic LCMV
   infection in mice ([293]28, [294]36) using Partek Flow software. GSEA
   for Hallmark and Gene Ontology pathways and the pan-cancer single-cell
   RNA-Seq gene signatures for human TILs ([295]33) was performed using
   Partek Flow software. The lists of significant biomarker genes for TEX
   subsets were calculated using the normalized RNA counts of TEX subsets
   (generated in vivo) ([296]28) on Partek (P < 0.05, fold change ≥ 2).
   Enrichment analyses using these transcriptional signatures of TEX
   subsets were performed on GSEA software (UCSD and Broad Institute,
   v4.3.3).

ATAC-Seq.

   P14 cells were isolated for ATAC-Seq using the Sony MA900 Cell Sorter,
   including Acute-7d or T[[Dysf]] PD-1^^+Tim3^^+ on day 19 and resting
   T[[Dysf]] cells isolated as PD-1^^+Tim3^^– (Rest-SP) or PD-1^^+Tim3^^+
   (Rest-DP) on day 23. ATAC was performed using Omni-ATAC, as previously
   described ([297]53), with minor modifications. Briefly, nuclei were
   isolated from cells using ATAC lysis buffer (10 mM Tris-HCl pH 7.5, 10
   mM NaCl, 3 mM MgCl[[2]], 0.1% NP-40, 0.1% Tween 20, and 0.01%
   Digitonin) and then transposed for 30 minutes at 37°C using Tagment DNA
   TDE1 Enzyme (Illumina, 20034197). Transposed DNA was purified, and
   libraries were generated using Nextera XT DNA Library Preparation Kit
   (Illumina, FC-131-2001). The concentration and size distribution of
   libraries were determined by Qubit (Thermo Fisher Scientific) and
   Agilent Tapestation, respectively. The ATAC libraries were then
   sequenced by Azenta Life Sciences. The data were analyzed using
   cutadapt 1.18, Bowtie2, Picard 2.18.17, samtools, and deepTools 3.3.0.
   First, the FASTQ files were trimmed using cutadapt, then aligned to
   mouse mm10 genome assembly using Bowtie2. Next, duplicate reads were
   removed using Picard 2.18.17, and bigWig files were generated with
   deepTools. Differentially OCRs on day 19 were defined as regions with
   fold change ≥ 2 (for open in T[[Dysf]]) or ≤ –2 (for closed in
   T[[Dysf]]) and P < 0.01 for T[[Dysf]] versus Acute-7d cells and plotted
   as heatmaps using Partek Flow software. PCA was performed using Partek
   to compare OCRs for Acute-7d, T[[Dysf]], and Rest-SP or Rest-DP subsets
   to published chromatin landscapes for memory or exhausted CD8^^+ T
   cells ([298]23) using the top 5,000 variable peaks. Accessible
   chromatin peaks were visualized using IGV software at the indicated
   genomic sites. To visualize chromatin accessibility transitions
   specific to T[[Dysf]] PD-1^^+Tim3^^+ cells, we generated an alluvial
   plot. Peaks were classified as “open” or “closed” (stage 2) based on
   comparisons to the Acute-7d condition, using thresholds of adjusted P <
   0.01 and fold change ≥ 2 or ≤ –2. These peaks were then evaluated in
   resting states (Rest-DP and Rest-SP) to assign stage 3 labels — open,
   closed, or unchanged — using the same statistical cutoffs. The
   Sankey-style plot depicts the number of peaks transitioning from the
   Acute-7d baseline (stage 1) through their T[[Dysf]]-specific state
   (stage 2) to their resting fate (stage 3), highlighting dynamic
   regulatory patterns across conditions. TF enrichment analysis was
   performed using HOMER (findMotifsGenome.pl) with the mm10 genome and
   the -size given option to preserve original region lengths.

Targeted DNA methylation sequencing.

   DNA was isolated from FACS-purified P14 cell pellets (sorted on day 19
   of in vitro stimulation) and used for bisulfite conversion using the EZ
   DNA Methylation-Direct Kit (Zymo). PCR was then performed on the
   bisulfite-converted DNA at DMRs for key genes related to CD8^^+ T cell
   exhaustion states (Tcf7, Ccr7, Pdcd1). The following primers were used
   for each region: mTcf7 (forward primer:
   5′-GGTTAGTTTGAGTTTGGTTTAGAGTAGTGAG-3′, reverse primer:
   5′-CCTCTTACCTAAATTTCCCTACAAAATACC-3′), mCcr7 (forward primer
   5′-GGAGTTTGGGATAAAAGTTTTTAATGG-3′, reverse primer:
   5′-CCAAACCCACTCTAAACCCTATATTAC-3′), and mPdcd1 (forward primer:
   5′-GGTTGAGAGAGATTGAAATTAGGGTTAG-3′, reverse primer:
   5′-CAACAAAACTAACAAACCTAAAACAACT-3′). Following PCR amplification, the
   amplicon size was verified by gel electrophoresis. Gel bands were
   excised, and amplicon DNA was purified from each region using the
   Zymoclean Gel DNA Recovery Kit (Zymo). The purified amplicon DNA was
   then used for library preparation using the Native Barcoding Kit 24 V14
   (Oxford Nanopore Technologies) ([299]9). The barcoded DNA library was
   loaded onto an R10 flow cell and run on the MinION Mk1B sequencer from
   Oxford Nanopore. FASTQ files generated from the sequencing run were
   used for downstream analysis for genome alignment and determining
   percentage of CpG methylation at each amplified region using a
   customized NanoEM pipeline ([300]54).

Statistics.

   Statistical analyses were performed using GraphPad Prism v10. For flow
   cytometry and DNA methylation data analysis, comparisons between 2
   groups were performed using the Mann-Whitney U test (unpaired,
   2-sided). For comparison of 3 or more groups, 1-way ANOVA tests (for 1
   factor) or 2-way ANOVA tests (for 2 factors) were performed with a
   Tukey’s multiple-comparison correction. Biological and technical
   replicates (n) are stated in the figures. All data were expressed as
   mean ± SEM, and a P value of less than 0.05 was considered
   statistically significant. Statistical significance for RNA-Seq data
   was determined by PCA, DESeq2 analysis, or GSEA using Partek Flow
   software (v12) or UCSD and Broad Institute GSEA software. Statistical
   significance for ATAC-Seq data was determined by PCA and DESeq2
   analysis using Partek Flow software.

Study approval.

   All research conducted in this study complies with the ethical
   regulations under the approved protocols by The Ohio State University
   IACUC (IACUC #2019A00000055-R1) and the Institutional Biosafety
   Committee (IBC #2020R00000129).

Data availability.

   Values for all data points in graphs are reported in the
   [301]Supporting Data Values XLS file. ATAC-Seq and RNA-Seq datasets
   have been deposited in the NCBI BioProject PRJNA1299473 and Gene
   Expression Omnibus under the accession codes [302]GSE304670 and
   [303]GSE304671, respectively.

Author contributions

   Conceptualization: HEG. Methodology: HEG, AY, AAS, and AL.
   Investigation: HEG, AY, AAS, AL, AMY, KP, AS, MRA, and EMO.
   Visualization: HEG, AY, AAS, and AL. Supervision: HEG. Writing —
   original draft: HEG, AY, AAS, and AL. Writing — review and editing:
   HEG, EMO, AY, AAS, and AL.

Supplementary Material

   Supplemental data
   [304]jciinsight-10-191220-s045.pdf^ (9.6MB, pdf)
   Supporting data values
   [305]jciinsight-10-191220-s046.xlsx^ (121.1KB, xlsx)

Acknowledgments