Abstract

   During chronic infection and tumor progression, CD8^+ T cells lose
   their effector functions and become exhausted. These exhausted CD8^+ T
   cells are heterogeneous and comprised of progenitors that give rise to
   effector-like or terminally-exhausted cells. The precise cues and
   mechanisms directing subset formation are incompletely understood.
   Here, we show that growth factor independent-1 (Gfi1) is dynamically
   regulated in exhausted CD8^+ T cells. During chronic LCMV Clone 13
   infection, a previously under-described Ly108^+CX[3]CR1^+ subset
   expresses low levels of Gfi1 while other established subsets have high
   expression. Ly108^+CX[3]CR1^+ cells possess distinct chromatin profiles
   and represent a transitory subset that develops to effector-like and
   terminally-exhausted cells, a process dependent on Gfi1. Similarly,
   Gfi1 in tumor-infiltrating CD8^+ T cells is required for the formation
   of terminally differentiated cells and endogenous as well as
   anti-CTLA-induced anti-tumor responses. Taken together, Gfi1 is a key
   regulator of the subset formation of exhausted CD8^+ T cells.

   Subject terms: Adaptive immunity, Immunotherapy
     __________________________________________________________________

   The mechanisms underlying CD8^+ T cells exhaustion during chronic
   infection and tumor progression formation remain unclear. Here, the
   authors demonstrate that Gfi1 acts as a chromatin remodeler during the
   transition from progenitor to terminally exhausted T cells in chronic
   infection and mediates the formation of terminally differentiated TILs
   and the efficacy of anti-CTLA-4 for tumor treatment.

Introduction

   During chronic viral infections (e.g., lymphocytic choriomeningitis
   virus Clone 13: LCMV Cl-13^[39]1) and tumor progression, persistent
   antigen stimulation causes CD8^+ T cells to progressively lose their
   effector functions, a process known as T cell exhaustion^[40]2–[41]6.
   Exhausted CD8^+ T cells are heterogeneous and maintained by
   self-renewing TCF1^+Ly108^+ progenitor cells^[42]4,[43]7–[44]11. Immune
   checkpoint blockers (ICB) such as anti-CTLA-4 and anti-PD-1/L1 target
   progenitor cells and drive their proliferation and differentiation into
   either Ly108^-CX[3]CR1^+ effector-like or Ly108^-CX[3]CR1^- terminally
   exhausted (Term-Exh) cells^[45]2,[46]4,[47]7,[48]8,[49]11–[50]16.
   Formation and maintenance of exhausted effector-like CD8^+ T cells is
   essential for tumor and viral control^[51]4,[52]7,[53]8,[54]15–[55]17.
   However, the underlying transcriptional and epigenetic mechanisms
   remain incompletely understood, delineation of which would help bolster
   ICB efficacy.

   Accumulating evidence indicates that generation of effector-like and
   Term-Exh subsets involves a substantial chromatin reshaping in
   progenitor cells^[56]10,[57]18–[58]21, which is under the delicate
   control of a network of transcription factors (TFs) and signaling
   pathways. Specifically, TOX is critical for the overall development and
   maintenance of exhausted CD8^+ T cells^[59]22–[60]25, while TCF1
   regulates the formation and maintenance of progenitors that is further
   reinforced by Bach2 and PBAF^[61]9,[62]10,[63]19–[64]21,[65]26. On the
   other hand, Batf and Zeb2 mediate the formation of effector-like and
   Term-Exh cells, but their deletion does not abolish the
   process^[66]9,[67]12,[68]13,[69]18, suggesting that other TFs also play
   a role in their formation.

   Growth factor-independent 1 (Gfi1) is a transcriptional repressor,
   consisting of a SNAG domain^[70]27,[71]28 and six zinc-finger motifs,
   of which the 3–5 motifs are crucial for DNA binding^[72]28. Gfi1
   represses target genes^[73]29–[74]32 through recruitment of histone
   deacetylases (HDACs), demethylases (KDMs)^[75]33,[76]34, and other
   chromatin modifying complexes such as CoREST, NuRD, SWI/SNF, and
   CtBP^[77]35. Studies from our group and others show that Gfi1
   orchestrates the development of thymocytes by repressing Foxo1, Klf2,
   and Irf4, among others^[78]36,[79]37. The absence of Gfi1 favors the
   formation of single positive CD8^+ T thymocytes, but its role in mature
   peripheral CD8^+ T cells, especially exhausted CD8^+ T cells, has not
   been studied^[80]36–[81]41. This can be due in part to the low Gfi1
   expression in resting peripheral CD8^+ T cells, although Gfi1 is
   rapidly upregulated upon antigen stimulation^[82]42,[83]43.

   Given the persistent antigen stimulation in exhausted T
   cells^[84]3,[85]44 and the intimate interactions of Gfi1 with
   chromatin-modifying complexes/epigenetic enzymes, we ask if Gfi1
   regulates CD8^+ T cell exhaustion. Using Gfi1tdTomato reporter mice, we
   show that Gfi1 expression is dynamically regulated in exhausted CD8^+ T
   cell subsets during chronic viral infection and tumor progression. In
   LCMV-specific CD8^+ T cells, Gfi1 expression remains low in a newly
   described Ly108^+CX[3]CR1^+ subset but high in other subsets. Gfi1
   deletion disrupts the repression of epigenetic and transcriptional
   programs in stem-like progenitor cells as well as the formation of the
   effector-like and Term-Exh subsets. In tumor-infiltrating T cells
   (TILs), Gfi1 is highly expressed in a subset of progenitor CD8^+ TILs
   and is required for the accumulation of Tim-3^hi terminally
   differentiated effector TILs. Loss of Gfi1 in T cells dampens effector
   functions of CD8^+ TILs and anti-CTLA-4 efficacy. Overall, we identify
   Gfi1 as a key regulator of exhausted CD8^+ T cell subset
   diversification during chronic viral infection and cancer progression.

Results

Gfi1 expression in CD8^+ T cells is dynamically regulated during chronic
viral infection

   To monitor Gfi1 expression in CD8^+ T cells in response to chronic
   infections, we utilized the Gfi1 reporter mouse model, wherein
   endogenous Gfi1 expression is measured by tdTomato^[86]45 (hereafter,
   Gfi1tdTomato). Gfi1tdTomato mice were infected with LCMV Cl-13, as we
   previously described^[87]2. Mice infected with LCMV Armstrong (Arm)
   were included as controls for acute viral infection. At different times
   post-infection, splenocytes were harvested and stained with
   LCMV-D^bGP33 tetramer to determine Gfi1tdTomato expression in
   LCMV-specific CD8^+ T cells. Gfi1tdTomato expression gradually
   increased in GP33^+ CD8^+ T cells and reached the peak at the later
   phase of both acute and chronic viral infections (Day 30, Fig. [88]1A
   and Supplementary Fig. S[89]1A). Compared to Arm infection, CD8^+ T
   cells from LCMV Cl-13 infected mice expressed lower levels of
   Gfi1tdTomato on Day 30 (Supplementary Fig. S[90]1B). To examine the
   role of Gfi1 in exhausted CD8^+ T cells, we focused on its expression
   in the LCMV Cl-13^+ CD8^+ T cell subsets. At the early stage (Days
   5–8), Gfi1 expression was negligible in all subsets, defined by Ly108
   (a surrogate for the precursor/progenitor marker TCF1^[91]15) and
   CX[3]CR1 (an effector cell marker^[92]46) (Supplementary Fig. S[93]1C).
   On Day 15, when features of exhaustion are being established^[94]9, a
   considerable proportion of Ly108^+CX[3]CR1^- progenitor cells
   (~ 40–50%) expressed Gfi1 (Supplementary Fig. S[95]1C), while no
   significant changes between time points were observed in other subsets,
   indicating an early induction of Gfi1 in exhausted precursor progenitor
   cells.

Fig. 1. Gfi1 is dynamically regulated in exhausted CD8^+ T cell subsets
during chronic viral infection.

   [96]Fig. 1
   [97]Open in a new tab

   A Gfi1tdTomato reporter mice were infected with LCMV Cl-13, followed by
   the assessment of Gfi1tdTomato expression on Days 5 (n = 6), 8 (n = 6),
   15 (n = 6) and 30 (n = 8) post-infection. B–E Identification of Day 30
   exhausted CD8^+ T cell subsets based on CD44, PD-1, TCF1, Ly108 (B),
   Ki67, and CD69 expression (C). D Quantification of Ki67 and CD69
   expressions as in (C) (n = 9). E. Expression and quantification of
   Ly108 and CX[3]CR1 on the 4 exhausted subsets (n = 9). F GMFIs of PD-1,
   2B4, LAG-3, Tim-3, CX[3]CR1, CD44, TCF1, T-bet, Eomes, and TOX in the 4
   exhausted subsets (progenitor, Ly108^+CX[3]CR1^+, effector-like, and
   Term-Exh) (n = 9–10). G Expression and quantification of CXCR6 in the 4
   exhausted subsets (n = 5). H Gfi1tdTomato expression on the 4 exhausted
   CD8^+ T cell subsets (n = 8). Data were pooled from 2 independent
   experiments and depicted with mean ± s.e.m. Data in A were analyzed by
   one-way ANOVA; data in (D–H) were analyzed by repeated one-way ANOVA.
   Holm-Sidak’s post hoc test was used for all analysis. Source data are
   provided as a Source Data file.

   On Day 30 post-infection, T cell exhaustion is well established, and
   the pool of antigen-specific CD8^+ T cells is highly
   heterogeneous^[98]6,[99]9,[100]15. We thus characterized GP33^+ cells
   by assessing their expression of PD-1, CD44, Ly108, TCF1, Ki67, and
   CD69, as previously reported^[101]6 (Fig. [102]1B–D). CD44 expression
   diminishes as CD8^+ T cells become terminally
   exhausted^[103]2,[104]47,[105]48, dividing GP33^+ cells into CD44^hi
   and CD44^int fractions. While CD44^int cells were homogeneously
   Ly108^-TCF1^- Term-Exh cells, CD44^hi cells can be further divided into
   Ly108^+TCF1^+, Ly108^+TCF1^-, and Ly108^-TCF1^- sub-populations
   (Fig. [106]1B). Ki67 expression was highest in Ly108^+TCF1^- cells,
   whereas CD69 expression was highest on CD44^intLy108^-TCF1^- Term-Exh
   cells, consistent with a previous report^[107]6 (Fig. [108]1C, D).
   Ly108 and CX[3]CR1 have also been used to define populations of
   exhausted CD8^+ T cells during chronic LCMV Cl-13 infection^[109]15. We
   therefore examined their expression on the 4 defined subsets
   (Fig. [110]1E). As expected, CD44^intLy108^-TCF1^- Term-Exh cells
   expressed low levels of CX[3]CR1 and Ly108 (Ly108^-CX[3]CR1^-);
   progenitor TCF1^+ cells were Ly108^+CX[3]CR1^-, and
   CD44^hiLy108^−TCF1^− effector-like cells were Ly108^-CX[3]CR1^+
   (Fig. [111]1E and Supplementary Fig. S[112]1D). Intriguingly, the
   fourth population of CD44^hiLy108^+TCF1^- cells expressed intermediate
   CX[3]CR1 and Ly108, that is, they were Ly108^+CX[3]CR1^+ (Fig. [113]1E
   and Supplementary Fig. S[114]1D). Further characterizations of these
   cells revealed that they expressed higher PD-1, 2B4, LAG-3, Tim-3,
   CD44, Eomes, and TOX but lower T-bet than effector-like cells
   (Fig. [115]1F); they also expressed Eomes and Tim-3 at the highest
   level among all the subsets, suggesting they are different from
   effector-like cells.

   Recent studies have described intermediate or precursor effector
   subsets based on CXCR6^[116]12 and KLRG1^[117]13,[118]14 expression. We
   therefore characterized their expression on the 4 exhausted subsets.
   While CXCR6 was not observed on progenitor or Ly108^+CX[3]CR1^+ cells
   (Fig. [119]1G), KLRG1 was expressed on both Ly108^+CX[3]CR1^+ and
   effector-like cells (Supplementary Fig. S[120]1E). Functionally,
   Ly108^+CX[3]CR1^+ cells produced more IFN-γ than other exhausted
   subsets, but this was negligible compared to CD8^+ T cells from LCMV
   Arm infection (Supplementary Fig. S[121]1F, Top). On the other hand,
   effector-like cells produced more perforin than Ly108^+CX[3]CR1^+ cells
   (Supplementary Fig. S[122]1F, Bottom). Contrasting to CD8^+ T cells
   from Arm infection, these LCMV Cl-13 exhausted cells did not produce
   substantial amounts of TNF. We then analyzed Gfi1tdTomato expression in
   these exhausted subsets. Gfi1tdTomato was highly expressed in the
   progenitor, effector-like, and Term-Exh subsets but lowly expressed in
   Ly108^+CX[3]CR1^+ cells (Fig. [123]1H). In sum, using Ly108 and
   CX[3]CR1, we can define 4 exhausted subsets on Day 30 post LCMV Cl-13
   infection: Ly108^+CX[3]CR1^− progenitors, Ly108^+CX[3]CR1^+ cells,
   Ly108^−CX[3]CR1^+ effector-like cells, and Ly108^-CX[3]CR1^- terminally
   exhausted cells (Term-Exh), which differentially express Gfi1.

Ly108^+CX[3]CR1^+ cells possess unique chromatin accessibility and
transcriptional features, some of which are shared with effector-like cells

   Our above characterizations and differential Gfi1 expression of
   Ly108^+CX[3]CR1^+ cells suggest they are a distinct exhausted subset.
   Considering the essential role of epigenetic and transcriptomic
   imprinting of exhausted CD8^+ T
   cells^[124]18,[125]20,[126]21,[127]26,[128]49–[129]51, we hypothesized
   that Ly108^+CX[3]CR1^+ cells would possess distinct chromatin
   accessibility and transcriptomic profiles. To examine this, sorted
   progenitors, Ly108^+CX[3]CR1^+, effector-like, and Term-Exh GP33^+
   CD8^+ cells from WT non-Gfi1tdTomato reporter mice were subject to
   Assay for Transposase-Accessible Chromatin with sequencing (ATAC-seq).
   Since progenitors are the source of the other subsets during chronic
   viral infection^[130]6,[131]9, we focused on statistically significant
   chromatin changes in Ly108^+CX[3]CR1^+ versus progenitor cells, using a
   ≥ 1.2 log2 fold change as the cutoff. Overall, Ly108^+CX[3]CR1^+ cells
   had 3218 differentially accessible chromatin regions (DACR) with
   increased accessibility and 633 DACRs with decreased accessibility
   (Fig. [132]2A). Most of these DACRs were localized at enhancers (distal
   intergenic and intronic regions) in accordance with the concept that
   these regions determine cellular identity^[133]52 (Fig. [134]2B). We
   then visualized these DACRs in effector-like and Term-Exh subsets to
   evaulate their accessibility. As shown in Fig. [135]2C,
   Ly108^+CX[3]CR1^+ and effector-like cells shared similar accessibility
   profiles in some regions (Cluster 1 and Supplementary Data File [136]1;
   e.g., Havcr2 encoding Tim-3). We also identified regions (Clusters 3
   and 4; e.g., Batf3) in which Ly108^+CX[3]CR1^+ cells had the greatest
   increase in accessibility compared to other subsets.

Fig. 2. Ly108^+CX[3]CR1^+ cells possess unique chromatin accessibility and
transcriptional features, some of which are shared with effector-like cells.

   [137]Fig. 2
   [138]Open in a new tab

   Day 30 exhausted GP33^+ CD8^+ T cell subsets were sorted based on Ly108
   and CX[3]CR1 expression for ATAC-Seq (A–G) and RNA-Seq (H). A
   Statistically significant and differentially accessible chromatin
   regions (DACRs) with > 1.2 log2 fold change from pairwise comparisons
   were shown. Prog = Progenitor, Eff-like = Effector-like, and
   Term-Exh = Terminally-exhausted. B Promoter and Enhancer distribution
   of more and less accessible DACRs in the Ly108^+CX[3]CR1^+ cells versus
   progenitors. C Accessibility profiles of DACRS from (B) were visualized
   in effector-like and Term-Exh cells. D Venn diagram depicting the
   intersection of DACRs in Ly108^+CX[3]CR1^+ cells versus progenitors
   with those in effector-like cells versus progenitors (Left), with the
   bar graph showing the unique DACRs in effector-like cells (Middle), and
   heatmap visualization of these regions in progenitors,
   Ly108^+CX[3]CR1^+, and effector-like cells (Right). E Bar graph
   depicting distribution of DACRs in Term-Exh versus Ly108^+CX[3]CR1^+
   cells. F^. Genomic track of accessibility at the Mef2c locus in the 4
   exhausted subsets. G De novo HOMER motif analysis of DACRs from
   Ly108^+CX[3]CR1^+ versus progenitor cells. H Heatmap visualization of
   select transcription factors among the 4 exhausted subsets. ATAC- and
   RNA-seq data were from 2 and 3-4 independent replicates, respectively.
   Each replicate was a pooled sample of 3-5 mice.

   Next, we performed direct pairwise comparisons between
   Ly108^+CX[3]CR1^+ or progenitor cells versus other subsets
   (Fig. [139]2A and Supplementary Data File [140]1). To our surprise, we
   detected only 30 more but no less accessible DACRs in Ly108^+CX[3]CR1^+
   compared to effector-like cells. This limited number of DACRs could be
   due to similar chromatin accessibility profiles between them or a lack
   of statistical power to detect such differences. In contrast, there
   were 9368 and 2682 DACRs between effector-like and progenitor or
   Term-Exh cells, respectively, indicating our data were sufficiently
   powered (Fig. [141]2A). Next, we determined the shared or unique DACRs
   between Ly108^+CX[3]CR1^+ and effector-like subsets when compared to
   progenitor cells (Fig. [142]2D). Strikingly, 77% of identified DACRs in
   Ly108^+CX[3]CR1^+ versus progenitors were also identified in
   effector-like cells (2972 out of 3851 DACRs), but 68% of identified
   DACRs in effector-like versus progenitors were unique to effector-like
   cells and not shared with Ly108^+CX[3]CR1^+ cells (6396 out of 9368
   DACRs) (Fig. [143]2D). Visualization of the inferred 6396 DACRs
   ‘unique’ to effector-like cells revealed that Ly108^+CX[3]CR1^+ cells
   had an intermediate chromatin accessibility profile in these regions
   between progenitor and effector-like cells (Fig. [144]2D and
   Supplementary Data File [145]1, inferred DPvsEff). Specifically, 3465
   DACRs were less accessible (e.g., Brwd1, Hrh4, Wnt3, Smarca2, and
   Setd2) whilst the other 2931 DACRs were more accessible (e.g., Ccl3,
   Zeb2, Klrb1c, Glp1r, and Jun) in effector-like cells than progenitors
   cells. More accessible 2931 DACRs were linked to pathways involving
   GTPase activity and downstream NFAT and AP1 activity important for T
   cell activation^[146]53 (Supplementary Fig. S[147]2B), while less
   accessible 3465 DACRs were associated with Wnt signaling, etc. known to
   inhibit effector differentiation^[148]54. Finally, we identified 6672
   less (e.g., Klrg1, CD86, Kat7) and 1549 more (e.g., Sirt4, Hdac4, and
   Glp1r) accessible DACRs in Term-Exh than Ly108^+CX[3]CR1^+ cells
   (Fig. [149]2E). Several of the identified DACRs between
   Ly108^+CX[3]CR1^+ and progenitors, effector-like or Term-Exh cells
   included loci previously associated with T cell exhaustion and/or
   differentiation such as Zeb2^[150]12, Batf3^[151]55, Hif1a^[152]56,
   Irf4^[153]57, Irf5^[154]58, Batf^[155]9,[156]18,[157]49,
   Glp1r^[158]59,[159]60, and Runx3^[160]49,[161]61–[162]64 as well as
   those with previously unreported roles in T cell exhaustion such as
   Mef2c. Genomic tracks for the Mef2c, Zeb2 and Batf3, and Glp1r loci
   were shown in Fig. [163]2F and Supplementary Fig. S[164]2A. Overall,
   these results suggest that Ly108^+CX[3]CR1^+ and effector-like cells
   share some chromatin accessibility profiles, but there are numerous and
   subtle differences between them. Furthermore, Ly108^+CX[3]CR1^+ cells
   appear to be more distinctively different from progenitors and Term-Exh
   cells.

   We then predicted the transcription factors (TFs) interacting with
   DACRs in Ly108^+CX[3]CR1^+, effector-like, and Term-Exh cells by
   performing de novo motif analysis using HOMER^[165]65. As expected,
   motifs for TFs with previously reported roles in CD8^+ T cell effector
   functions, such as Batf, AP-1/NFAT and IRF family
   proteins^[166]18,[167]49,[168]50,[169]57,[170]66–[171]70 were enriched
   in DACRs identified in Ly108^+CX[3]CR1^+ and effector-like cells
   (Fig. [172]2G and Supplementary Fig. S[173]2C). Interestingly, we
   also observed Gfi1 DNA binding motif
   (5’-AAATC-3’)^[174]28,[175]71,[176]72 in DACRs identified in
   Ly108^+CX[3]CR1^+ cells versus progenitor, effector-like, or Term-Exh
   cells (Supplementary Fig. S[177]2D); regions within the Gfi1 locus were
   differentially accessible among the subsets (Supplementary Fig.
   S[178]2A, Bottom), although the role of Gfi1 in T cell exhaustion has
   been unreported.

   To understand the transcriptomic profiles of the 4 subsets, we
   performed RNA-seq. Pairwise comparisons unveiled 1279 differentially
   expressed genes (DEGs) between Ly108^+CX[3]CR1^+ and progenitor cells,
   499 DEGs between Ly108^+CX[3]CR1^+ and effector-like cells, and 1392
   DEGs between Ly108^+CX[3]CR1^+ and Term-Exh cells (Supplementary Data
   File [179]2), further supporting Ly108^+CX[3]CR1^+ are a distinct
   subset. For illustrative purposes, select transcripts of transcription
   factors (Fig. [180]2H) and effector molecules (Fig. S[181]2E) among
   DEGs were shown. TFs (e.g., Irf8, Irf5, Foxm1, and Hmgb3) and molecules
   associated with cell proliferation, such as Mybl2, were significantly
   increased in Ly108^+CX[3]CR1^+ cells compared to the other subsets.
   Interestingly, Zeb2, Pou2f1, Spi1, E2F2/7/8 proteins, Hmgb2, and Hmgn5
   were significantly elevated in Ly108^+CX[3]CR1^+ and effector-like
   cells compared to progenitors and Term-Exh cells, suggesting these TFs
   may co-regulate effector-like and Ly108^+CX[3]CR1^+ cells. Among DEGs
   between effector-like and Ly108^+CX[3]CR1^+cells, higher expression of
   Prf1 (encodes Perforin), Ccl3, Id2, and Glp1r but lower expression of
   Myb, Mef2c, Hmgn3, Pou2af1, and Spib were found in
   Ly108^+CX[3]CR1^+cells than in effector-like cells. Surprisingly,
   normalized Gfi1 mRNA values from the sequencing experiment were
   comparable between Ly108^+CX[3]CR1^+ and progenitor cells, which were
   lower than Term-Exh and effector-like cells (Supplementary Fig.
   S[182]2F). We speculate the relatively low Gfi1 mRNA values in
   progenitors may be due to a high fraction of Gfi1^lo cells in this
   subset.

CD8^+ T cell intrinsic Gfi1 is required for the formation of effector-like
and Term-Exh cells

   The dynamic regulation of Gfi1 reporter expression, the enrichment of
   Gfi1 motif in DACRs, and the unreported role of Gfi1 in T cell
   exhaustion prompted us to examine whether and how Gfi1 modulates
   exhausted CD8^+ T cells. To this end, we generated mice with specific
   Gfi1 deletion in T cells by crossing floxed Gfi1 mice with CD4-Cre
   transgenic mice (hereafter, Gfi1^cKO)^[183]36,[184]38,[185]41. As we
   previously described^[186]36,[187]41, the zinc finger region 3-5 of
   Gfi1, essential for DNA binding and transcriptional activity^[188]28
   are deleted in Gfi1^cKO T cells. Gfi1^fl/flCD4-Cre^-/- littermates (WT)
   and Gfi1^cKO mice were infected with LCMV Cl-13. On Days 15-16 and
   30-45 post-infection, GP33^+ T cells from the spleen were analyzed.
   Although the frequencies of total GP33^+ CD8^+ T cells were similar on
   Day 15, total cell numbers were lower in Gfi1^cKO mice at this time
   point (p = 0.0549, Supplementary Fig. S[189]3A). On Day 30, when the
   CD8^+ T cell responses contracted, both the frequency and cell number
   of total GP33^+ CD8^+ cells were comparable between Gfi1^cKO and WT
   mice. We then evaluated the impact of Gfi1 deletion on the subset
   diversification of exhausted GP33^+ CD8^+ T cells, categorized by Ly108
   and CX[3]CR1. As shown in Fig. [190]3A, B, Gfi1^cKO cells failed to
   generate effector-like and Term-Exh subsets, but the progenitor and
   Ly108^+CX[3]CR1^+ subsets appeared unaltered. In support of this,
   KLRG1, a killer-type lectin that has been linked to the effector-like
   subset^[191]13,[192]14, was not expressed by Gfi1^cKO cells
   (Supplementary Fig. S[193]3B). Failed formation of effector-like and
   Term-Exh cells was also observed in the GP33^+ CD8^+ T cells from the
   liver (Supplementary Fig. S[194]3C, D), with the cell number of total
   GP33^+ CD8^+ T cells significantly reduced in Gfi1^cKO mice and their
   frequencies being comparable to WT mice (Supplementary Fig. S[195]3D).

Fig. 3. CD8^+ T cell intrinsic Gfi1 is required for the formation of effector
and Term-Exh subsets.

   [196]Fig. 3
   [197]Open in a new tab

   A, B GP33^+ responses were evaluated in WT and Gfi1^cKO mice on Days 15
   and 30 post LCMV Cl-13 infection. Frequencies (A) and cell numbers (B)
   of subsets (in A, n = 11 and 16 for WT mice, 12 and 13 for Gfi1^cKO
   mice on Day 15 and 30, respectively. For (B), n = 11 for WT mice at
   both time points, and n = 12 and 7 for Gfi1^cKO mice on Day 15 and 30,
   respectively). C, D. Frequencies (C) and cell numbers (D) of GP33 +
   subsets in CD4-depleted WT (n = 8 for both time points) and Gfi1^cKO
   mice (n = 7 and 12, respectively, for Day 15 and 30). E CD45.1^+ mice
   reconstituted with WT (n = 7) or Gfi1^cKO (n = 9) CD45.2^+ BM cells
   (chimeric mice) were infected with LCMV Cl-13. GP33^+ CD8^+ T cell
   responses in CD45.1^+ and CD45.2^+ compartments were analyzed on Day 30
   post-infection. F Sorted exhausted subsets (Progenitors: n = 6;
   Ly108^+CX[3]CR1^+: n = 7; effector-like: n = 6; Term-Exh: n = 6) from
   Day 30 LCMV Cl-123-infected Gfi1tdTomato reporter mice were adoptively
   transferred into naïve CD45.1^+ recipients, followed by LCMV Cl-13
   infection on the next day. Progenies of adoptively transferred cells
   were characterized for their expression of Ly108 and CX[3]CR1 on Day 8
   post-infection. Cell counts of subsets were shown by the scatter dot
   plot. Data (mean ± s.e.m) were pooled from 2-4 independent experiments
   and analyzed by two-way ANOVA with Holm-Sidak’s post-hoc test. Source
   data are provided as a Source Data file.

   To further demonstrate the failed formation of effector-like and
   Term-Exh subsets by Gfi1^cKO cells, we assessed the expression of CD44,
   PD-1, Ly108, and TCF1 (Supplementary Fig. S[198]3E). We did not detect
   CD44^int or Ly108^- cells in the absence of Gfi1, further supporting
   the lack of Term-Exh and/or effector-like cells. We also considered
   whether this phenotype was due to increased cell death in the absence
   of Gfi1. However, pan-caspase expression was similar between WT and
   Gfi1^cKO progenitors and Ly108^+CX[3]CR1^+ cells (Supplementary Fig.
   S[199]3F), ruling out that possibility. We did observe a significant
   decrease in Ki67 expression between WT and Gfi1^cKO Ly108^+CX[3]CR1^+
   cells, although no overt difference was observed in progenitor cells
   (Supplementary Fig. S[200]3G).

   We next quantified the expression of inhibitory receptors and TFs
   associated with T cell exhaustion in WT and Gfi1^cKO subsets. TOX,
   LAG-3, and Tim-3 were significantly upregulated in both Gfi1^cKO
   subsets compared to WT cells (Supplementary Fig. S[201]3H).
   Interestingly, we observed increased Eomes and decreased TCF1
   expression only within Gfi1^cKO progenitors, while PD-1 was
   significantly increased in Gfi1^cKO Ly108^+CX[3]CR1^+ cells compared to
   their WT counterparts (Supplementary Fig. S[202]3H). Functionally,
   granzyme B (GzmB) expression was lower in Gfi1^cKO Ly108^+CX[3]CR1^+
   than WT cells (Supplementary Fig. S[203]3I, right), although no overt
   differences in IFN-γ and TNF production were observed between WT and
   Gfi1^cKO cells (Supplementary Fig. S[204]3I, left). In keeping with the
   absence of effector-like cells^[205]15 and reduced cytolytic
   capabilities of Gfi1^cKO Ly108^+CX[3]CR1^+ cells (i.e., reduced GzmB),
   Gfi1^cKO mice had elevated viral titers at later time points (Days 30
   and beyond, Supplementary Fig. S[206]3J).

   CD4^+ T cell deficiency ablates the formation of exhausted
   effector-like CD8^+ T cells^[207]15,[208]73. Since our model deletes
   Gfi1 in both CD8^+ and CD4^+ T cells, we wanted to rule out potential
   confounding effects from Gfi1^cKO CD4^+ T cells by depleting CD4^+ T
   cells in vivo. As previously reported^[209]15,[210]73, CD4 depletion
   promoted exhaustion in WT GP33^+ CD8^+ T cells (Fig. [211]3C versus 3A)
   and resulted in reduced effector-like cells. Like un-manipulated mice,
   no significant differences in total GP33^+ frequencies and numbers were
   observed in CD4-depleted Gfi1^cKO mice on Days 15 and 30 (Supplementary
   Fig. S[212]3K). Interestingly, the numbers of progenitors and
   Ly108^+CX[3]CR1^+ cells were comparable between WT and Gfi1^cKO mice
   with CD4^+ T cell depletion, contrasting their increases in
   un-manipulated Gfi1^cKO mice (Fig. [213]3B, D), suggesting a role of
   CD4^+ T cells in maintaining their overall numbers. Nevertheless, we
   still observed significantly impaired formation of effector-like and
   Term-Exh cells in CD4-depleted Gfi1^cKO mice (Fig. [214]3C, D). To
   firmly establish this is a CD8^+ T cell intrinsic effect, we employed
   the bone marrow chimeric mice (BM) reconstituted with wild-type or
   Gfi1^cKO BM cells. Following reconstitution, these mice were infected
   with LCMV Cl-13. GP33^+ CD8^+ splenocytes were analyzed on Day 30. As
   shown in Fig. [215]3E, WT CD45.1^+ and CD45.2^+ GP33^+ cells
   successfully generated progenitor, Ly108^+CX[3]CR1^+, effector-like,
   and Term-Exh cells, but CD45.2^+ Gfi1^cKO GP33^+ cells were still
   unable to form effector-like and Term-Exh cells even in the presence of
   wild-type CD4^+ T cells. Again, there were no significant differences
   in total GP33^+ frequencies and cell counts in the spleens of BM
   chimeras (Supplementary Fig. S[216]3L). These data highlight a CD8^+ T
   cell intrinsic role for Gfi1 in the formation of effector-like and
   Term-Exh subsets.

Transitional Ly108^+CX[3]CR1^+ cells develop into effector-like and Term-Exh
subsets

   Considering the shared chromatin accessibility and transcriptomic
   landscape between Ly108^+CX[3]CR1^+ and effector-like cells, as well as
   the inability of Gfi1^cKO CD8^+ T cells to form effector-like and
   Term-Exh subsets, we next assessed their developmental potential. To
   this end, we sorted progenitor, Ly108^+CX[3]CR1^+, effector-like, and
   Term-Exh subsets from Day 30 LCMV Cl-13 infected Gfi1tdTomato mice and
   adoptively transferred them into naïve CD45.1^+ recipients, followed by
   LCMV Cl-13 on the next day. The progenies of the transferred cells
   (CD45.2^+) were analyzed on Day 8 post-secondary infection. As
   previously reported, effector-like and Term-Exh cells largely retained
   their phenotype^[217]15, and progenitors were capable of re-generating
   all the subsets (Fig. [218]3F and Supplementary Fig. S[219]4A).
   However, transferred Ly108^+CX[3]CR1^+ cells were only able to form
   effector-like and Term-Exh but not progenitors (Fig. [220]3F).
   Noteworthily, some Ly108^+CX[3]CR1^+ cells retained their identify on
   Day 8 after transfer, indicating some degree of stability. Transferred
   progenitors were recovered at the highest frequencies and numbers,
   although this did not reach statistical significance compared to other
   transferred subsets (Fig. [221]3F and Supplementary Fig. S[222]4A).

   The adoptive transfer studies suggest that differentiation of
   Ly108^+CX[3]CR1^+ cells (Gfi1tdTomato^lo) to effector-like and Term-Exh
   cells (Gfi1tdTomato^hi) is coupled with an increase of Gfi1tdTomato
   expression (Supplementary Fig. S[223]4B). Considering that
   Gfi1tdtomato^lo cells persisted in the progenitor, effector-like, and
   Term-Exh subsets (Fig. [224]1H), we characterized the expression of
   some key TFs, inhibitory receptors, and proteins within Gfi1tdTomato^hi
   and Gfi1tdTomato^lo fractions to shed light on their potential
   interactions with Gfi1 (Supplementary Fig. S[225]4C, D). Eomes, TOX,
   and Tim-3 expression were significantly elevated in Gfi1tdtomato^lo
   compared to Gfi1tdTomato^hi fractions in all 3 subsets (Supplementary
   Fig. S[226]4C, D). The reciprocally lower PD-1 and higher T-bet in
   Gfi1tdTomato^lo than Gfi1tdTomato^lo Term-Exh supported a
   previously-reported strong repression of the Pdcd1 loci by
   T-bet^[227]74, but this was restricted to that subset, as there were no
   significant differences of PD-1 and T-bet expression between
   Gfi1tdTomato^hi and Gfi1tdTomato^lo progenitors and effector-like cells
   (Supplementary Fig. S[228]4C). Finally, CX[3]CR1 expression was higher
   in Gfi1tdTomato^lo progenitor and Term-Exh subsets but lower in
   Gfi1tdTomato^lo effector-like cells compared to their Gfi1tdTomato^hi
   counterparts (Supplementary Fig. S[229]4D). These data were consistent
   with the greater CX[3]CR1 expression in Ly108^+CX[3]CR1^+ and
   effector-like cells (Fig. [230]1E), in line with their developmental
   potentials. Taken together, these data unveil a possible repression of
   Eomes, TOX, and Tim-3 by Gfi1 in exhausted CD8^+ T cell subsets.

   TCF1 maintains progenitors and is downregulated as these cells
   differentiate^[231]4,[232]9,[233]10. Given the downregulated TCF1 and
   Ly108 (surrogate marker for TCF1) in Gfi1tdTomato^lo progenitor cells
   (Supplementary Fig. S[234]4C, D), we asked if the Gfi1tdTomato^lo
   progenitor fraction was more poised to differentiate into other subsets
   than the Gfi1tdTomato^hi progenitor fraction. To address this,
   Gfi1tdTomato^lo and Gfi1tdTomato^hi progenitor cells sorted from Day 30
   LCMV Cl-13 infected Gfi1tdTomato mice were transferred into naïve
   CD45.1^+ mice and subsequently infected with LCMV Cl-13 one day later.
   Eight days post-infection, both progenitor fractions (CD45.2^+)
   regenerated all other subsets, although more cells were recovered from
   transferred Gfi1tdTomato^hi progenitors (Supplementary Fig. S[235]4E).
   Interestingly, Gfi1tdtomato expression was increased in recovered cells
   from transferred Gfi1tdTomato^lo progenitors (Supplementary Fig.
   S[236]4E), indicating a reversible conversion between Gfi1tdTomato^lo
   and Gfi1tdTomato^hi progenitors.

Gfi1 modulates the chromatin accessibility and transcriptomic programs
associated with the formation of exhausted effector-like and Term-Exh CD8^+ T
cells

   Epigenetic changes underpin the formation of T cell subsets during
   chronic viral
   infection^[237]14,[238]18,[239]20–[240]22,[241]26,[242]51,[243]75.
   Given Gfi1’s reported functions in chromatin modifying complexes in
   other cellular contexts^[244]33–[245]35, we posited that Gfi1 would
   impact the chromatin accessibility associated with the differentiation
   of exhausted CD8^+ T cell subsets. To test this idea, sorted splenic
   GP33^+ progenitor and Ly108^+CX[3]CR1^+ cells from Day 30 LCMV Cl-13
   infected Gfi1^cKO mice, together with WT subsets, were subject to
   ATAC-seq and RNA-seq. We identified 693 (647 more and 46 less
   accessible) and 2471 (1109 more and 1362 less accessible) DACRs between
   Gfi1^cKO and WT progenitors and Ly108^+CX[3]CR1^+ cells, respectively
   (Fig. [246]4A, Supplementary Fig. S[247]5A, B and Supplementary Data
   File [248]3).

Fig. 4. Gfi1 controls the chromatin accessibility and transcriptomic programs
required for the formation of exhausted effector-like and Term-Exh CD8^+ T
cells.

   [249]Fig. 4
   [250]Open in a new tab

   Exhausted GP33^+ subsets from WT and Gfi1^cKO mice were sorted on Day
   30 post-LCMV Cl-13 infection for ATAC-seq (A–E) or RNA-seq (F). A
   Differentially accessible chromatin regions (DACRs) between Gfi1^cKO
   versus WT progenitors and Ly108^+CX[3]CR1^+ cells, with a cutoff of 1.2
   log2 fold. B DACRs of progenitor versus Ly108^+CX[3]CR1^+ subsets in WT
   and Gfi1^cKO cells. C Heatmap visualization of DACRs between Gfi1^cKO
   and WT cells were shown. k-means clusters with numbers of DACRs
   indicated in the parentheses for each cluster. D Genomic tracks of
   accessibility at Mef2c and Lef1 loci in WT (progenitor,
   Ly108^+CX[3]CR1^+, effector-like) as well as Gfi1^cKO (progenitors,
   Ly108^+CX[3]CR1^+) subsets. E De novo HOMER motif analysis of DACRs
   from Gfi1^cKO versus WT Ly108^+CX[3]CR1^+ cells. F Heatmap
   visualization of select transcripts for transcription factors in WT and
   Gfi1^cKO GP33^+ subsets. ATAC- and RNA-seq data were from 2 and 3-4
   independent replicates. Each replicate was a pooled sample of 3-5 mice.

   Given that Ly108^+CX[3]CR1^+ cells can be formed with or without Gfi1,
   we asked whether the chromatin accessibility changes observed in WT
   cells during the transition from progenitors to Ly108^+CX[3]CR1^+ cells
   would also occur in Gfi1^cKO cells. Remarkably, only 30 more and 164
   less accessible DACRs were identified in Gfi1^cKO Ly108^+CX[3]CR1^+
   cells versus progenitors, in contrast to 3218 more and 633 less
   accessible DACRs in WT cells (Fig. [251]4B, Supplementary Fig. S[252]5C
   and Supplementary Data File [253]3). For illustrative purposes, we
   visualized all the identified DACRs in both Gfi1^cKO subsets compared
   to their WT counterparts to understand how loss of Gfi1 impacted
   chromatin accessibility and prevented the formation of effector-like
   cells (Fig. [254]4C). Some regions (cluster 4), including Mef2c
   (Fig. [255]4D, denoted with a red star), Foxp1 (Fig. [256]4D, Bottom),
   Irf4, and Nt5e (encodes CD73 and is repressed by Gfi1^[257]76), failed
   to lose accessibility in Gfi1^cKO Ly108^+CX[3]CR1^+ cells, and rather
   increase accessibility compared to WT cells. Of note, these regions
   became less accessible in WT effector-like than in progenitors or
   Ly108^+CX[3]CR1^+ cells (Fig. [258]4D Top and Supplementary Data
   File [259]3), indicating their lost accessibility is needed for the
   formation of effector-like cells. On the contrary, other regions
   (cluster 2), including Dnmt3a, Akt2, Runx1, and Runx3, failed to gain
   accessibility in Gfi1^cKO Ly108^+CX[3]CR1^+ cells versus progenitors,
   which was evident in WT Ly108^+CX[3]CR1^+ cells. Finally, we identified
   regions such as in Havcr2 (encodes Tim-3, Supplementary Fig. S[260]5D,
   Middle), Mef2c, and Ezh1 (Supplementary Fig. S[261]5D, Top) with
   significantly increased accessibility in Gfi1^cKO progenitors and
   Ly108^+CX[3]CR1^+ cells compared to their WT counterparts. Taken
   together, the dearth of DACRs in Gfi1^cKO Ly108^+CX[3]CR1^+ cells
   versus progenitors cells can be explained by the premature increase in
   accessibility in cluster 3 within progenitor cells and the failure of
   Ly108^+CX[3]CR1^+ to lose accessibility in cluster 4 in the absence of
   Gfi1.

   De novo motif analysis on the DACRs between the WT and Gfi1^cKO
   progenitors and Ly108^+CX[3]CR1^+ cells identified enrichment of TFs
   such as Batf^[262]18,[263]49, CTCF^[264]77, and NFAT^[265]78
   (Fig. [266]4E and Supplementary Fig. S[267]5E). Consistent with the
   auto-repression of the Gfi1 locus by Gfi1 itself^[268]32,[269]43, its
   motifs were enriched in accessible regions in both Gfi1^cKO progenitors
   and Ly108^+CX[3]CR1^+ cells compared to WT cells (Fig. S[270]5D, E). We
   detected Gfi1 binding motifs at Havcr2 regions with increased chromatin
   accessibility in Gfi1^cKO compared to WT cells (Supplementary Data
   File [271]3), which was consistent with the increased Tim-3 expression
   in Gfi1tdTomato^lo subsets (Supplementary Fig. S[272]4F) and in
   Gfi1^cKO cells (Supplementary Fig. S[273]3H). Notably, the chromatin
   dysregulation in the absence of Gfi1 was observed only in exhausted
   antigen-specific cells but not naïve CD8^+ T cells, in line with
   negligible Gfi1 protein expression in naive cells (Supplementary Fig.
   S[274]5F).

   RNA-seq analyses of Gfi1^cKO subsets and their WT counterparts also
   revealed dysregulated transcriptomes, which were congruent in part with
   our ATAC-seq analysis. As detailed in Supplementary Data File [275]4,
   we detected 55 DEGs between WT and Gfi1^cKO progenitors, 488 DEGs
   between WT and Gfi1^cKO Ly108^+CX[3]CR1^+ cells, and 515 DEGs between
   Gfi1^cKO Ly108^+CX[3]CR1^+ and progenitor cells. For illustrative
   purposes, select transcripts for TFs and activation markers were
   visualized to highlight the impact of Gfi1 on transcriptomes
   (Fig. [276]4F and Supplementary Fig. S[277]5G). Understandably, some
   transcriptomic programs such as upregulation of Irf8, Spi1, and Zeb2,
   and Hmgn2, and downregulation of TCF7, Id3, and Satb1 were maintained
   in Gfi1^cKO Ly108^+CX[3]CR1^+ cells compared to progenitors
   (Fig. [278]4F and Supplementary Data File [279]4). On the other hand,
   several repression programs in Ly108^+CX[3]CR1^+ cells were
   dysregulated in the absence of Gfi1. Gfi1^cKO Ly108^+CX[3]CR1^+ cells
   aberrantly exhibited upregulated Batf, Lef1, Ikzf2, Hif1a, TOX, Stat4,
   Irf2, Runx2, and Runx3 but downregulated Spib, Bmyc, Myc, Pou2af1,
   Klrg1, and Ifngr2 compared to WT Ly108^+CX[3]CR1^+ cells (Fig. [280]4F
   and Supplementary Data File [281]4). Transcripts for Hrh4, Kit, Glp1r,
   Havcr2, and LAG-3 were significantly increased in both Gfi1^cKO subsets
   compared to WT subsets (Supplementary Fig. S[282]5G). Because CD101
   protein and mRNA is highly expressed in Term-Exh cells^[283]79, we
   analyzed CD101 mRNA transcripts from our sequencing studies and found
   no significant difference between WT and Gfi1^cKO progenitors and
   Ly108^+CX[3]CR1^+ cells (Supplementary Fig. S[284]5H). This further
   supported no formation of Term-Exh cells in the absence of Gfi1. We
   noted that several of the dysregulated transcripts in Gfi1^cKO
   Ly108^+CX[3]CR1^+ (e.g., increased Ikzf2 and Lef1) that were normally
   downregulated in WT effector-like cells compared to progenitors
   (Fig. [285]4F and Supplementary Data File [286]2, [287]4). Overall,
   these data suggest that Gfi1 promotes differentiation into
   effector-like cells by repressing genes associated with stem-like
   progenitor cells.

Gfi1 is necessary for the accumulation of terminally differentiated Tim-3^hi
CD8^+ TILs in late-stage tumors

   Our above results show that Gfi1 plays an important role in exhausted
   CD8^+ T cells during chronic viral infection. Considering that
   tumor-infiltrating CD8^+ T cells (TILs), like LCMV-specific CD8^+ T
   cells, are also known to be exhausted due to constant antigen
   stimulation^[288]3,[289]44,[290]80, we therefore examined whether and
   how Gfi1 modulates CD8^+ TILs. First, to determine if tumor progression
   impacts Gfi1 expression in CD8^+ TILs, we inoculated Gfi1tdTomato mice
   with MB49 tumor cells, a syngeneic model for urothelial adenocarcinoma,
   to establish progressive tumors (Fig. [291]5A). On Days 7, 14, and 22
   post-tumor inoculation, CD8^+ TILs were isolated and analyzed for TCF1
   and Tim-3 expression (Fig. [292]5B). TCF1 mark progenitor TILs
   (TCF1^hi), and Tim-3^hi TILs are commonly regarded as terminally
   differentiated TILs^[293]7,[294]16. While Day 7 CD8^+ TILs cells were
   predominantly TCF1^hi, Tim-3^hi TILs emerged by Day 14 and became the
   dominant subset in late-stage tumors on Day 22. Day 22 Tim-3^hi
   effector cells expressed higher levels of the inhibitory receptors 2B4,
   LAG-3, PD-1, and TF TOX but lower levels of TF Eomes than TCF1^hi
   progenitor TILs (Fig. [295]5C and Supplementary Fig. S[296]6A), while
   T-bet expression was comparable between the two subsets (Supplementary
   Fig. S[297]6A). With respect to their Gfi1tdTomato expression, Day 7
   progenitor TILs contained Gfi1tdTomato^hi, Gfi1tdTomato^med, and
   Gfi1tdTomato^lo subsets (Fig. [298]5D, Top). Although this pattern
   remained at later time points, the frequencies of Gfi1tdTomato^lo cells
   gradually decreased on Days 14-21 (Supplementary Fig. S[299]6B),
   coinciding with the emergence of Tim-3^hi cells that were predominantly
   Gfi1tdTomato^med (Fig. [300]5D, Bottom & S6C). Notably, the frequencies
   of Gfi1tdTomato^hi cells were largely constant over time (Supplementary
   Fig. S[301]6B). The expression of TCF1, Eomes, T-bet, and 2B4 was
   highest in Gfi1tdTomato^hi progenitor cells, whereas the expression of
   TOX, LAG-3, PD-1, and Tim-3 was highest in Gfi1tdTomato^med progenitors
   (Fig. [302]5E and Supplementary Fig. S[303]6D). The low expression of
   TCF1 but high expressions of TOX, PD-1, Tim-3, and LAG-3 by
   Gfi1tdTomato^med progenitor cells mirrored that of Tim-3^hi cells,
   suggesting a close developmental interconnection between them
   (Fig. [304]5C). Altogether, these data indicate that transient
   downregulation of Gfi1 is a shared feature of exhausted progenitor
   CD8^+ T cells during tumor progression and chronic viral infections.

Fig. 5. Gfi1 is necessary for the accumulation of terminally differentiated
Tim-3^hi tumor infiltrating CD8^+ T cells in late-stage tumors.

   [305]Fig. 5
   [306]Open in a new tab

   A Tumor growth dynamics of Gfi1tdTomato reporter mice subcutaneously
   inoculated with 1 × 10^5 MB49 cells. B Identification of TCF1^hi
   progenitor and Tim-3^hi effector CD8^+ TILs with total cell counts and
   ratios of Tim-3^hi to TCF1^hi on Days 7 (n = 11), 14 (n = 13), and 22
   (n = 11) post-inoculation. C Histograms showing the expression of 2B4,
   LAG-3, PD-1, and TOX in TCF1^hi (progenitor) and Tim-3^hi TILs. D Flow
   plots of Gfi1tdTomato expression in progenitor and Tim-3^hi CD8^+ TILs
   at different time points, with the frequencies for subsets on Day 14
   shown in the scatter dot plot (n = 6). E Flow plots showing expression
   of Gfi1tdTomato with TCF1, TOX, Eomes, and T-bet in progenitor cells
   with their GMFIs depicted (n = 5–7). F Tumor growth dynamics and masses
   on Days 7, 12, and 21 in WT (n = 10, 16, & 17) and Gfi1^cKO (n = 11, 16
   & 18) mice post the inoculation of MB49 cells. G Flow plots showing the
   abundance of Tim-3^hi and TCF1^hi progenitor TILs in tumors from WT
   (n = 10, 18, & 17) and Gfi1^cKO (n = 10, 16, & 18) mice as well as
   their ratios on Day 7, 12, and 21 post-tumor inoculation. H
   Distribution of Tim-3^hiT-bet^hi and Tim-3^hiT-bet^lo effector TILs as
   well as Tim-3^loT-bet^int progenitor TILs (Left) and their frequencies
   of TOX^hi and Eomes^hi cells (Right) on Day 21 tumors from WT (n = 8)
   and Gfi1^cKO (n = 10) mice. I Perforin and granzyme (B) production by
   WT (n = 6) and Gfi1^cKO (n = 7) CD8^+ TILs. Data (mean ± s.e.m.) were
   pooled results from 2–4 independent experiments, with each dot
   indicating an individual mouse. Data were analyzed by two-way ANOVA (B
   Left, D, F–H), one-way ANOVA (B, Right), repeated one-way ANOVA (E),
   and unpaired two-tail t test (I). Holm-Sidak’s post hoc test was used
   unless otherwise stated. Source data are provided as a Source Data
   file.

   To investigate the role of Gfi1 in anti-tumor responses, WT and
   Gfi1^cKO mice were inoculated with MB49 tumor cells, and tumor growth
   was monitored over time. While tumor growth was comparable between WT
   and Gfi1^cKO mice at the early stage, Gfi1^cKO mice displayed
   increasingly worse tumor control at later stages, corroborating our
   previous study in B16 melanoma^[307]41 (Fig. [308]5F). Tumor mass was
   also greater in Gfi1^cKO mice on Day 21 but not on Days 7 and 12
   (Fig. [309]5F). Frequencies of total Gfi1^cKO CD8^+ T cells were
   increased in Day 7 tumors but decreased in Day 12 and 21 tumors
   compared to their WT counterparts; Gfi1^cKO CD8^+ cell numbers were
   also significantly reduced in Day 21 tumors (Supplementary Fig.
   S[310]6E–G). In contrast, the frequencies and cell numbers of total
   CD4^+ TILs and FoxP3^+ regulatory T cells (T[reg]) were comparable
   between WT and Gfi1^cKO mice at all time points (Fig. S[311]6E–G). In
   direct correlation with the reduced frequencies and numbers of CD8^+
   TILs in late-stage tumors from Gfi1^cKO mice, frequencies and cell
   numbers of Tim-3^hi CD8^+ TILs were significantly reduced in Gfi1^cKO
   mice on Day 21 (Fig. [312]5G and Supplementary Fig. S[313]6H), while
   the numbers of TCF1^hi progenitor TILs were comparable between WT and
   Gfi1^cKO mice (Supplementary Fig. S[314]6H). Overall, these data
   establish a temporal requirement for Gfi1 in CD8^+ TILs, driving the
   accumulation of Tim-3^hi TILs at later (Day 21) but not earlier stages
   (Day 12) of tumor progression.

   We further characterized Tim-3^hi TILs through examination of T-bet
   expression, which divided them into T-bet^hiTim-3^hi and
   T-bet^loTim-3^hi subsets. T-bet^hiTim-3^hi cells contained higher
   frequencies of TOX^hi cells than T-bet^loTim-3^hi cells, while few
   Tim-3^lo (T-bet^int) progenitor TILs were TOX^hi (Fig. [315]5H, C). Of
   note, T-bet^hiTim-3^hi cells were the dominant subset in Day 21 TILs.
   Gfi1 deletion similarly abrogated the accumulation of both Tim-3^hi
   subsets (Supplementary Fig. S[316]6I). The frequencies of TOX^hi cells
   were significantly reduced in both Gfi1^cKO progenitor and Tim-3^hi
   TILs (Fig. [317]5H), and Eomes^hi cells was only increased in Gfi1^cKO
   progenitor but not Tim-3^hi TILs (Fig. [318]5H, Right). Finally, we
   observed increased LAG-3 but decreased 2B4 expression in Gfi1^cKO
   Tim-3^hi cells, while PD-1 was comparable between WT and Gfi1^cKO cells
   (Supplementary Fig. S[319]6I). Pan-caspase expression was similar
   between WT and Gfi1^cKO cells, excluding increased cell death as the
   underlying mechanism for the failed accumulation of Tim-3^hi cells
   (Supplementary Fig. S[320]6J, Top). We did observe decreased Ki67
   expression within all Gfi1^cKO subsets (Supplementary Fig. S[321]6J,
   Bottom), consistent with a general role for Gfi1 in cell
   proliferation^[322]42,[323]81. IFN-γ production (Supplementary Fig.
   S[324]6K) and the frequency of GzmB^+Prf1^+ (Fig. [325]5I) cells in
   Gfi1^cKO CD8^+ TILs were reduced compared to WT cells. Overall, our
   data suggest an important role of Gfi1 in the transition of TCF1^hi
   progenitor to terminally differentiated Tim-3^hi TILs and their
   effector functions in late-stage tumors.

Gfi1 expression in T cells is required for anti-CTLA-4 efficacy

   ICBs such as anti-CTLA-4/PD-1 enhance anti-tumor responses, at least in
   part, by promoting the proliferation of progenitor TILs and subsequent
   transition into terminally differentiated effector
   TILs^[326]7,[327]8,[328]16. Our above results indicate that transient
   Gfi1 downregulation in TCF1^hi progenitors was necessary for their
   transition into Tim-3^hi effector TILs. We therefore hypothesized that
   anti-CTLA-4 would induce Gfi1 downregulation in CD8^+ TILs. To address
   this, Gfi1tdTomato mice bearing MB49 tumors were treated with
   anti-CTLA-4 on Days 5, 8, and 11, as we previously described^[329]82.
   Anti-CTLA-4 suppressed tumor growth (Supplementary Fig. S[330]7A),
   increased the infiltration of both CD4^+ and CD8^+ T cells
   (Supplementary Fig. S[331]7B), and significantly reduced
   Gfi1tdTomato^hi progenitor cells (Supplementary Fig. S[332]7C),
   supporting a transient Gfi1 downregulation by anti-CTLA-4. To
   investigate if Gfi1 in T cells mediates the therapeutic effects of
   anti-CTLA-4, WT and Gfi1^cKO mice bearing established MB49 tumors were
   similarly treated with anti-CTLA-4. While anti-CTLA-4 significantly
   inhibited tumor growth in WT mice, this did not occur in Gfi1^cKO mice
   (Fig. [333]6A). Similarly, anti-CTLA-4-driven infiltration and/or
   expansion of CD4^+ and CD8^+ TILs was observed in WT but not Gfi1^cKO
   mice (Fig. [334]6B). Interestingly, the increase of IFN-γ production by
   CD4^+ TILs and the reduction of T[reg] in response to anti-CTLA-4 were
   comparable between WT and Gfi1^cKO mice (Supplementary Fig. S[335]7D,
   E). Consistent with the impaired ability of Gfi1^cKO CD8^+ progenitor
   TILs to transition into Tim-3^hi TILs, Tim-3^hi/TCF1^hi ratios of CD8^+
   TILs were significantly reduced in Gfi1^cKO mice compared to WT mice
   (Supplementary Fig. S[336]7F), following anti-CTLA-4 treatments.
   Further characterizations revealed that Gfi1^cKO TCF1^hi CD8^+ TILs
   expressed higher TCF1 but lower TOX than WT cells (Supplementary Fig.
   S[337]7G, H). The expression of TOX and PD-1 was comparable between WT
   and Gfi1^cKO Tim-3^hi subsets, although Tim-3 expression was higher in
   Gfi1^cKO Tim-3^hi cells (Supplementary Fig. S[338]7H). Ki67 expression
   was reduced in all Gfi1^cKO CD8^+ subsets (Supplementary Fig.
   S[339]7H), again supporting a general role of Gfi1 in cell
   proliferation^[340]42,[341]81. Lastly, in contrast to no overt
   differences in Gfi1^cKO CD4^+ TILs, IFN-γ production was significantly
   reduced in Gfi1^cKO CD8^+ TILs than WT cells, upon anti-CTLA-4 therapy
   (Fig. [342]6C).

Fig. 6. Gfi1 expression in T cells is required for anti-CTLA-4 efficacy.

   [343]Fig. 6
   [344]Open in a new tab

   WT and Gfi1^cKO mice inoculated with MB49 cells were treated with
   anti-CTLA-4 on Days 5, 8, and 11 post-inoculation. Tumor growth
   dynamics was depicted in the line graph (A, n = 4). Tumors were
   harvested to evaluate the infiltration of CD4^+ and CD8^+ T cells (B,
   n = 7 for untreated and 8 for treated groups of both WT and Gfi1^cKO
   mice) and IFN-γ production by CD8^+ TILs (C, n = 7 for untreated WT
   mice, = 8 for treated WT mice, = 5 for untreated Gfi1^cKO mice, and = 6
   for treated Gfi1^cKO mice). D WT (n = 4 in the untreated group and = 5
   in the treated group) and Gfi1^cKO (n = 7 in the untreated group and 5
   in the treated group) mice bearing palpable MC38 bladder tumors were
   treated with or without anti-CTLA-4 on Days 5, 8, and 11 post-tumor
   inoculation. Tumor growth dynamics was shown in the line graph. E, F
   Tumors were harvested to evaluate the infiltration of CD4^+ and CD8^+ T
   cells (E n = 8 for untreated WT and 9 for treated WT; n = 11 for
   untreated Gfi1^cKO and 12 for treated Gfi1^cKO) and IFN-γ production by
   CD8^+ TILs (F n = 9 for untreated and treated WT; n = 11 for untreated
   Gfi1^cKO and 12 for treated Gfi1^cKO). Data (mean ± s.e.m.) were pooled
   results from 2-3 independent experiments, with each dot denoting a
   mouse. Two-way ANOVA with Holm-Sidak post hoc test was used. Source
   data are provided as a Source Data file.

   To evaluate this using another tumor model, we inoculated WT and
   Gfi1^cKO mice with MC38 cells, a widely used syngeneic model for
   colorectal adenocarcinoma. As in MB49 bladder tumors, anti-CTLA-4
   efficacy was markedly attenuated in the Gfi1^cKO MC38-bearing mice
   compared to WT mice (Fig. [345]6D). Unlike the MB49 model, anti-CTLA-4
   only significantly increased the infiltration of CD8^+ but CD4^+ T
   cells in MC38 tumors (Fig. [346]6E). While anti-CTLA-4-induced IFN-γ
   production in CD4^+ TIL and T[reg] reduction were comparable between WT
   and Gfi1^cKO mice (Supplementary Fig. S[347]7I, J), anti-CTLA-4-driven
   infiltration of CD8^+ TILs and IFN-γ production were significantly
   lower in Gfi1^cKO mice (Fig. [348]6F). Based on these data, we conclude
   that Gfi1 deletion in T cells primarily impacts CD8^+ TILs and renders
   anti-CTLA-4 therapy less effective.

Discussion

   In this study, we identify 4 exhausted CD8^+ T cell subsets, including
   a previously undescribed Ly108^+CX[3]CR1^+ population, on Day 30
   post-LCMV Cl-13 infection using Ly108 and CX[3]CR1^[349]15 or CD44,
   PD-1, Ly108, TCF1, Ki67, and CD69^[350]6. Of note, Ly108^+CX[3]CR1^+
   cells have been gated together with effector-like cells in previous
   studies^[351]6,[352]12. As we show here, these cells differ in
   phenotype, chromatin accessibility, and transcriptome from true
   effector-like cells. Interestingly, the expression of Gfi1, a
   transcriptional repressor with a previously unreported role in
   exhausted CD8^+ T cells, gradually increases and peaks on Day 30 when
   exhaustion is well established, following LCMV Cl-13 infection. The
   highest level of Gfi1 expression first emerges in precursor progenitor
   cells on Day 15 and remains high in exhausted progenitor,
   effector-like, and Term-Exh subsets but low in Ly108^+CX[3]CR1^+ cells
   on Day 30. Based on these data, we propose that Gfi1 expression
   coincides with the emergence of exhaustion features in progenitor
   cells, that is propagated to effector-like and Term-Exh cells, in line
   with the model proposed by Utzschneider et al.^[353]9. Gfi1 loss in
   CD8^+ T cells blocks the formation of effector-like and Term-Exh
   subsets, whereas the generation of progenitors and Ly108^+CX[3]CR1^+
   cells is unaffected. As a result, viral titers are higher in Gfi1^cKO
   mice at later time points. Considering the phenotype of Gfi1^cKO CD8^+
   T cells on Day 15 post-LCMV Cl-13 infection and the previously proposed
   replacement of short-lived effector cells at this time point^[354]9, we
   hypothesize that Ly108^+CX[3]CR1^+ cells likely emerge ~ 2 weeks
   post-infection, although we were unable to isolate these cells to test
   this hypothesis, due to their continually changing phenotype during the
   first 3-4 weeks after infection^[355]9,[356]15.

   Adoptive transfer of exhausted subsets reveals divergent developmental
   potential of these cells. While effector-like and Term-Exh cells retain
   their phenotype and progenitors expectedly repopulate all the other
   subsets, Ly108^+CX[3]CR1^+ cells only develop into effector-like and
   Term-Exh cells but not progenitors. Interestingly, our data indicate
   that Ly108^+CX[3]CR1^+ cells also retain their identify for some time,
   suggesting some degree of their stability. Previous studies assessing
   the heterogeneity of the effector-like cells have identified
   KLRG-expressing intermediate precursor cells^[357]12–[358]14. Putting
   our findings in that context, we argue that Ly108^+CX[3]CR1^+ cells are
   an even earlier intermediate precursor population to effector-like and
   Term-Exh cells. Collectively, these results support earlier studies
   showing a bifurcation in the development of effector-like and Term-Exh
   cells^[359]12–[360]14, a decision likely made at the Ly108^+CX[3]CR1^+
   and not stem-like progenitor state. While results from the viral
   infection studies with Gfi1^cKO mice indicate Gfi1 expression is needed
   for the differentiation of Ly108^+CX[3]CR1^+ cells into effector-like
   or Term-Exh cells, it nevertheless does not commit a particular fate.
   Future work should focus on factors dictating fate decisions, which
   likely involve CD4^+ T cell help^[361]15.

   Systemic comparisons of Gfi1tdTomato^lo and Gfi1tdTomato^hi fractions
   in the progenitor, effector-like, and Term-Exh subsets suggest
   repression of TOX, Eomes, and Tim-3 by Gfi1. In further support, we
   observe increased expressions of TOX and Tim-3 in both Gfi1^cKO
   progenitors and Ly108^+CX[3]CR1^+ compared to their WT counterparts,
   while an increase in Eomes was only observed in Gfi1^cKO progenitors,
   likely due to its already high expression in WT Ly108^+CX[3]CR1^+
   cells. Complex transcriptomic and epigenetic imprinting govern subset
   diversification. In the absence of Gfi1, progenitor and
   Ly108^+CX[3]CR1^+ cells exhibit increased accessibility at the Mef2c
   locus, which is normally lost in WT effector-like and Term-Exh cells,
   suggesting Gfi1 may promote the formation of effector-like cells by
   suppressing accessibility at this locus. Consistent with this, the
   Mef2c transcript is significantly higher in Ly108^+CX[3]CR1^+ than
   effector-like cells. Our RNA-seq analysis also identify a repressive
   role for Gfi1 in the expression of other genes such as Lef1, Hmgn1,
   Stat4, and Ikzf2. These genes are typically associated with the
   progenitor population but downregulated in effector-like or Term-Exh
   cells. These results, together with the adoptive transfer data and
   functional studies, suggest that Gfi1 represses genes associated with
   the stem-like progenitor cells, thereby facilitating the formation of
   effector-like and Term-Exh cells.

   We notice considerable differences among exhausted CD8^+ T cells from
   chronic viral infections versus tumors, which reflect the distinction
   between systemic versus local antigen distribution, different cellular
   niches, and the rather harsh tumor microenvironment. Despite this, we
   show that Gfi1 is similarly expressed at a high level in a subset of
   TCF1^hi progenitor TILs. Transient downregulation of Gfi1 in the
   progenitor TILs precedes the emergence of terminally differentiated
   effector Tim-3^hi cells, similar to the Gfi1 downregulation in
   Ly108^+CX[3]CR1^+ prior to the formation of effector-like and Term-Exh
   cells during chronic viral infections. Moreover, deletion of Gfi1
   impairs the accumulation of Tim-3^hi TILs in late-stage tumors.
   Interestingly, Gfi1 is seemingly dispensable for the generation of
   Tim-3^hi effector TILs during the early stage of tumor development,
   suggesting a temporal requirement of Gfi1. To this end, a previous
   study report that CD8^+ TILs undergo distinct chromatin states in
   late-stage tumors when they become refractory to ICBs^[362]83. Our
   results suggest the switch of the re-programmability in CD8^+ TILs
   might be partially mediated by Gfi1, warranting future investigations.
   Of note, both Gfi1^cKO progenitor CD8^+ T cells from chronic viral
   infection and tumors exhibit increased Eomes as well as LAG-3
   expression. Despite the reported role of LAG-3 in sustaining TOX
   expression in both tumors and chronic viral infections^[363]84,[364]85,
   we observe increased TOX expression in Gfi1^cKO cells from the chronic
   viral infection but decreased TOX expression in Gfi1^cKO CD8^+ TILs,
   suggesting additional regulatory mechanisms controlling TOX expression
   in chronic infections versus cancer. Although we are unable to directly
   determine Gfi1 binding sites in this study, due to technical challenges
   and limited cell numbers, these results warrant future studies
   examining Gfi1 binding sites and its interactions with TOX, other TFs
   (e.g., Eomes), and inhibitory receptors such as LAG-3.

   T cell exhaustion is a major barrier to immunotherapies (e.g., CAR-T
   and ICBs)^[365]4,[366]8,[367]86,[368]87. Our study identifies an
   important role of Gfi1 in orchestrating CD8^+ T cell response to
   anti-CTLA-4 therapy, the very first FDA-approved ICB to treat patients
   with advanced cancer^[369]88. We reason that fine-tuning Gfi1 activity
   in T cells may prevent or reverse T cell exhaustion to bolster ICB
   efficacy. Considering Gfi1 downregulation is associated with the active
   differentiation of progenitors, we argue that transient and
   intermittent inhibition of Gfi1 with lysine specific histone
   demethylase (LSD1)^[370]33 may facilitate the differentiation of
   progenitors to Ly108^+CX[3]CR1^+ cells and then to effector-like cells,
   thereby improving the control of chronic infections and tumors, as a
   standalone therapy or in combination with ICBs. To this end, a recent
   study reports promising outcomes from combining an LSD1 inhibitor
   (Bomedemstat) with anti-PD-1 in small cell lung cancer^[371]89. Further
   testing of this combination approach should be conducted in melanoma,
   bladder cancer, and colorectal adenocarcinoma, especially those
   resistant to ICBs^[372]90. Elucidating the aforementioned interactions
   of Gfi1 with its binding partners (e.g., CoREST, NuRD, or CtBP
   complexes^[373]35, Eomes, LAG-3, and others yet to be defined) during
   CD8^+ T cell responses will be beneficial in devising rational
   therapeutic strategies to combat cancer and chronic infections.

Methods

Mice

   Gfi1tdTomato mice were generated by Dr. Lacaud^[374]45 and transferred
   to us from Dr. Grimes, with permission from Dr. Lacaud. Gfi1^fl/fl
   mice, as we previously described^[375]36,[376]41, were generated by Dr.
   Hano Hock^[377]38 and provided to us by Dr. Hongbo Chi, in agreement
   with Dr. Hock. They were bred with transgenic CD4-Cre mice as
   heterozygotes at the CD4-Cre locus (Gfi1^cKO) and littermates
   (CD4-Cre^-/-) (WT). Male and female mice were used in all experiments,
   unless specified. B6.SJL-PtprcaPepcb/BoyJ mice (CD45.1^+) were obtained
   from the Jackson Laboratory. CD45.1 mice were crossed with
   Gfi1^fl/flCD4-Cre^-/- (WT) mice to generate CD45.1^+CD45.2^+ mice,
   which were used as recipients in the bone marrow chimera studies.
   C57BL/6J-Ptprcem6Lutzy/J mice (JAXBoy, CD45.1^+) were a gift from Dr.
   Zajac and used in the adoptive transfer experiments. All mice were bred
   and maintained in a specific pathogen-free animal facility at The
   University of Alabama at Birmingham (UAB), under a 12 h/12 h light/dark
   cycle, ambient room temperature (22 °C), and 40– 70% humidity. 7– 12
   weeks old mice were used in the experiments, and protocols were
   approved by the Institutional Animal Care and Use Committee at UAB.

LCMV Infection, bone marrow chimeras, and adoptive transfer of LCMV^+ CD8^+
cells

   For LCMV infections, 4 × 10^6 plaque-forming-units (PFU) of LCMV Cl-13
   (chronic infection) were intravenously injected into Gfi1^cKO and
   littermate control (WT) mice via the tail vein. 2 × 10^5 PFU of LCMV
   Armstrong were injected intraperitoneally. At designated times
   post-infection, sera were collected and frozen until used in the plaque
   assay to determine viral titers using Vero cells. In brief, Vero cells
   were cultured in 6-well plates to confluency. Culture media was
   removed, and 200 μL of diluted sera was added to the plates for 1 h at
   37 °C. Cells were then overlayed with a mixture of 1% agarose and
   medium 199 (Sigma-Aldrich) and cultured for 4 days at 37 °C. Cells were
   overlayed again with a 1:1 mixture of 1% agarose and medium 199
   supplemented with 0.02% of neutral red solution (Sigma-Aldrich). Cells
   were incubated at 37 °C overnight before plaques were enumerated under
   a light source.

   To generate the mixed bone marrow chimeras, recipient CD45.1^+CD45.2^+
   mice were lethally irradiated with 2 doses of 500 Rad with an at least
   4-hour interval in between and reconstituted with 1 × 10^7 bone marrow
   cells depleted of mature T cells from CD45.1^+ mice and CD45.2^+
   Gfi1^fl/flCD4-Cre^-/- (littermate controls, WT) or
   Gfi1^fl/flCD4-Cre^+/- (Gfi1^cKO) mixed at 1:1 ratio. After > 6 weeks of
   reconstitution, mice were similarly infected with LCMV Cl-13. In the
   CD4^+ T cell depletion experiments, in vivo neutralizing antibodies
   against CD4 (Bio X Cell, clone GK1.5, #BE0003-1) (300 μg/mouse) were
   injected intraperitoneally 1 day before and 3 days after LCMV
   infection.

   For adoptive cell transfer, freshly prepared splenocytes from Day 30
   LCMV Cl-13 infected Gfi1tdTomato reporter mice were sorted for GP33^+
   subsets (Ly108^+CX[3]CR1^− progenitors, Ly108^+CX[3]CR1^+ cells,
   Ly108^−CX[3]CR1^+ effector-like cells, and Ly108^-CX[3]CR1^- Term-Exh
   cells,) or Gfi1tdTomato^hi versus Gfi1tdTomato^lo progenitors using BD
   FACSAria SORP-Titan. Sorted cells were adoptively transferred into
   CD45.1 JAXboy mice (~ 8 × 10^4 subset cells or (~ 2 × 10^4
   Gfi1tdTomato^hi/lo progenitors cells per mouse), and mice were infected
   with 4 × 10^6 PFU LCMV Cl-13 on the next day. Spleens from recipient
   mice were harvested on Day 8 post-infection for fate-mapping of
   adoptively transferred cells.

In vivo tumor inoculation and treatment

   The chemically induced murine urothelial adenocarcinoma MB49 cells and
   colorectal adenocarcinoma MC38 cells, kindly provided by Dr. Ashish
   Kamat and Dr. Jim Allison at MD Anderson Cancer Center, was originally
   derived from a male and a female C57BL/6 mouse, respectively. These
   cells were cultured in Dulbecco’s modified Eagle’s medium with 10%
   fetal bovine serum (FBS) at 37 °C, 5% CO[2], with regular testing for
   mycoplasma using the MycoAlert detection kit (Lonza, LT07-118) to
   ensure they were free of mycoplasma contamination. For tumor
   inoculation, exponentially growing MB49 cells (1 × 10^5) and MC38 cells
   (5 × 10^5) with a passage number less than 10 were subcutaneously
   injected into the shaved right flanks of male and female mice,
   respectively. On D5 post-tumor inoculation, mice received anti-CTLA-4
   or isotype antibodies intraperitoneally (i.p.), with the first dose at
   200 μg/mouse and subsequent doses of 100 μg/mouse on D8 and 11. Tumor
   measurements were taken 2–3 times/week, starting from D3 post-tumor
   inoculation. Following our IACUC protocols, mice were euthanized when
   tumors reached 2.0 cm in diameter, developed ulceration, or the mice
   became moribund.

TIL isolation and tissue preparation

   Detailed procedures were described previously^[378]41,[379]91. In
   brief, tumors were collected in ice-cold RPMI 1640 containing 2% FBS
   and minced into fine pieces, followed by digestion with 400 µ/mL
   collagenase D (Worthington Biochemical Corporation, #[380]LS004186) and
   20 µg/mL DNase I (Sigma, #10104159001) at 37 °C for 40 min with
   periodic shaking. EDTA (Sigma, #1233508) was then added at a final
   concentration of 10 mM to stop digestion. Cell suspensions were
   filtered through 70 µM cell strainers, and TILs were obtained by
   collecting the cells in the interphase after Ficoll (MP Biomedicals,
   #091692254) gradient separation. Spleens and livers were collected in
   ice-cold HBSS containing 2% FBS. Single cell suspensions from spleens
   were generated through disruption on a wire screen (infection studies)
   or nylon nano mesh (tumor studies); Red blood cells were lysed with
   0.84% (w/v) NH[4]Cl or ACK lysis buffer (Thermo Scientific). Livers
   were minced and digested with Collagenase IV and DNase I in Hank’s
   balanced salt solution for 30 mins at 37 °C. Digested tissue was
   separated after centrifugation over Ficoll (MP Biomedicals, #091692254)
   and washed with complete media. Cells were resuspended in RPMI 1640
   medium supplemented with 50 μM β-mercaptoethanol, penicillin
   (100 µ/ml), streptomycin (100 mg/ml), and either 1 or 10% fetal bovine
   serum (FBS) or Click’s media (Irvine Scientific, #9195-500 mL)
   containing 10% FBS, 10% Pen/Strep, and 10% Glutamine for flow
   cytometric analyses, as described below.

Flow cytometric analyses

   Flow cytometry data were collected on a BD FACSymphony with BD FACSDiva
   software or an Invitrogen AttuneNXT cytometer with Attune cytometric
   software. All data were analyzed with FloJo. Live-Dead staining was
   done using the aqua fluorescent reactive dye (ThermoFisher Scientific,
   L34966A) prior to surface staining. For surface and tetramer staining,
   cells were stained for 30– 40 mins at 4 °C. For transcription factor
   staining, cells were fixed and permeabilized with the eBioscience
   Foxp3/Transcription Factor staining kit (ThermoFisher Scientific,
   00-5523-00) before staining with antibodies for 30– 45 mins at 4 °C. To
   measure intracellular cytokines, for the LCMV studies, freshly prepared
   splenocytes from Day 30 LCMV Cl-13 infected mice were stimulated with
   GP33 peptides in the presence of monensin (BD Biosciences, #51-2092KZ)
   for 4 h prior to surface staining and intracellular cytokine detection
   using the BD CytoFix/Cytoperm buffer kit. (BD, 554714). For tumor
   studies, freshly isolated TILs were stimulated ex vivo with
   PMA/Ionomycin in the presence of monensin (BD Biosciences, #51-2092KZ)
   for 4 h prior to surface staining and intracellular cytokine detection.
   Antibodies used for flow cytometry included: anti-mouse 2B4 (m2B4 (B6)
   458.1, Biolegend, 1:200), anti-mouse Ly108 (330-AJ, Biolegend, 1:200),
   anti-mouse Tim-3 (RMT3-23, Biolegend, 1:200), anti-mouse/human CD44
   (IM7, Biolegend, 1:200), anti-mouse CD8a (53-6.7. BD, 1:200),
   anti-mouse CD8a (53-6.7, Invitrogen, 1:200), mouse anti-TCF7/TCF1
   (S33-966, BD, 1:100), anti-mouse CX[3]CR1 (SA011F11, Biolegend, 1:200),
   anti-PD-1 (RMP1-30, BD, 1:150), anti-KLRG1 (2F1, Invitrogen, 1:200),
   anti-CXCR6 (SA051D1, Biolegend, 1:200), anti-mouse CD45.1 (A20,
   Biolegend, 1:200), anti-mouse CD45.2 (104, Biolegend, 1:200),
   anti-mouse CD4 (RM4-5, Biolegend, 1:200), anti-mouse/Rat FoxP3
   (FJK-16s, Invitrogen, 1:100), anti-mouse TCRβ (H57-597, Biolegend,
   1:200), anti-IFN-γ (XMG1.2, Biolegend, 1:100), anti-TNF (MP6-XT22,
   Biolegend, 1:100), anti-mouse CD101 (mousehi101, eBioscience, 1:200),
   anti-mouse CD69 (H1.2F3, BD, 1:200), anti-mouse LAG-3 (C9B7W,
   Biolegend, 1:200), anti-mouse Ki67 (11F6, Biolegend, 1:100), anti-mouse
   Eomes (Dan11mag, Invitrogen, 1:100), anti-human/mouse TOX (REA473,
   Miltenyi, 1:100), anti-mouse T-bet (4B10, Biolegend, 1:100), anti-mouse
   Perforin (S16009A, Biolegend, 1:100), anti-human/mouse Granzyme B
   (QA16A02, Biolegend, 1:100). H2-DbGP33 (1:200) tetramers were acquired
   from the National Institutes of Health (NIH) Tetramer Core at Emory
   University.

Bulk RNA-seq

   Freshly prepared splenocytes from Day 30 LCMV CL-13 infected mice were
   sorted for bulk populations of GP33^+ subsets using the BD FACSAria
   SORP Titan. Three to five mice were pooled for each replicate. Cells
   were homogenized in TRIzol LS reagent (Invitrogen, 10296028) and total
   RNA was isolated. Library preparation, quality control, and paired-end
   sequencing was performed at GeneWiz Inc. RNA-seq data analysis was done
   as we previously described^[381]91. In brief, sequences were adapter
   and quality trimmed with Trim Galore (v0.6.10) and aligned to the mm10
   reference genome^[382]92 using Star (v2.7.11a)^[383]93. Counts were
   generated with HTSeq (v2.0.5)^[384]94. Differential analysis using
   glmTreat was performed using edgeR (v3.42.4)^[385]95, and heatmaps were
   generated using limma (v 3.56.2)^[386]96.

Assay for transposase-accessible chromatin with sequencing (ATAC-seq)

   ATAC-seq libraries for bulk-sorted populations, as described above,
   were generated using a previously published protocol^[387]97. Library
   quality control and paired end PE150 sequencing was performed by
   Novogene Co. For data analysis, reads were adapter and quality trimmed
   using Trim Galore (v 0.6.10). Trimmed fastq files were aligned to the
   mm10 reference genome^[388]92 with Bowtie2 (v2.4.3) using the following
   settings bowtie2 -p 10 --very-sensitive -k 10 -X 2000. Genrich (v0.6.1)
   was used to define peaks from biological replicates using the following
   settings –j –r -s 10 -v. Regions aligned to the mitochondrial
   chromosome and the Encode defined blacklist regions^[389]98 were
   excluded during peak calling with Genrich. Bedgraphs from Genrich were
   used to generate bigwig files using the UCSC utility bedGraphToBigWig.
   Narrowpeak files for all samples were merged using BEDTools (v2.28),
   converted to the SAF format, and counts were generated using
   featureCounts (v.2.0.6)^[390]99. Differential analysis using glmTreat
   with lfc ≥ 1.2 or glmQLFTest was performed using edgeR
   (v3.42.4)^[391]95, and heatmaps were generated using limma (v
   3.56.2)^[392]96 and ComplexHeatmap^[393]100. Genomic range objects were
   generated using GenomicRanges (v1.52.1) and annotated with ChipSeeker
   (v1.36.0)^[394]101. Pathway enrichment analysis was performed using
   ClusterProfiler (v4.8.3)^[395]102. Venn diagrams were generated with
   ggVennDiagram (v.15.0)^[396]103. HOMER (v4.11)^[397]65 was used for
   motif analysis. Genomic tracks were viewed and generated with
   Integrative Genomics Viewer (v2.16.1)^[398]104.

Statistical analysis

   All experiments were repeated 2–5 times. Results were depicted as
   mean ± s.e.m. Data were analyzed using a two-sided Student’s t test,
   one-way ANOVA, or two-way ANOVA after confirming their normal
   distribution. All analyses were performed using Prism 10.2.0 (GraphPad
   Software, Inc.), and p < 0.05 was considered statistically significant.

Reporting summary

   Further information on research design is available in the [399]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [400]Supplementary Information^ (1.9MB, pdf)
   [401]41467_2025_59784_MOESM2_ESM.pdf^ (273KB, pdf)

   Description of Additional Supplementary Files
   [402]Supplementary Data 1^ (7.8MB, xlsx)
   [403]Supplementary Data 2^ (418.8KB, xlsx)
   [404]Supplementary Data 3^ (1.7MB, xlsx)
   [405]Supplementary Data 4^ (103.9KB, xlsx)
   [406]Reporting Summary^ (149KB, pdf)
   [407]Transparent Peer Review file^ (3.1MB, pdf)

Source data

   [408]Source Data^ (131.7KB, xlsx)

Acknowledgements