Tumor-Reactive CD8⁺ T Cells Enter a TCF1⁺PD-1⁻ Dysfunctional State

CD8⁺ T cells can recognize and respond to tumor antigens arising from the aberrant expression of nonmutated genes. Because these antigens may also be expressed on noncancerous/normal cells, they are referred to as self/shared-antigens (SSA; reviewed in refs. 1, 2). In addition to SSA, cancer cells often express tumor-specific antigens (TSA) caused by mutations unique to the tumor and not expressed on normal tissue. Immune checkpoint blockade (ICB) therapies such as antibodies specific for PD-1 or PD-L1 can unleash antitumor T-cell responses, but they can also initiate immune-related adverse effects (irAE), mediated by self-reactive T cells (3, 4). Patients who experience ICB-induced irAE have been shown to have better antitumor responses and survival (5). It has been observed that the target tissue of irAE can correlate with the cancer tissue origin, suggesting that T cells reacting against SSA in tumors also cause irAE. For example, patients with melanoma are more likely to develop vitiligo (6, 7), while patients with lung cancer treated with ICB are more likely to develop pneumonitis (8), likely due to T-cell recognition of SSA expressed in normal and malignant lung tissue (9).

Much of our understanding of the differentiation, fate, and molecular regulation of tumor-reactive T cells has come from studying preclinical cancer mouse models of CD8⁺ T-cell responses against TSA (reviewed in ref. 10). Tumor-specific T cells (TST) differentiate to a dysfunctional (exhausted) state, lacking effector function, expressing multiple inhibitory receptors such as PD-1 and CD39 (11, 12), and harboring characteristic transcriptional and epigenetic programs (13–15). In contrast, the fate of CD8⁺ T cells reactive against self-antigens varies widely (reviewed in refs. 16–18). SSA-reactive T cells (SST) can remain ignorant of cognate antigen, undergo peripheral deletion, or be rendered anergic/self-tolerant, depending on antigen affinity, abundance/persistence, or tissue expression patterns (19–22). Moreover, T-cell responses to SSA may differ in tumors versus normal tissue as SSA might be distinctly regulated or presented in cancer cells. In addition, oncogenic transformation not only transforms cells but impacts inflammatory and/or immune signaling by transformed cells and surrounding stromal cells (reviewed in ref. 23). The unique tumor microenvironment, as compared with normal tissue, can impact CD8⁺ T-cell differentiation (24).

The dysfunctional TST population in tumor-bearing hosts is spatially heterogeneous, with progenitor/stemlike TST expressing the naïve/memory-associated transcription factor TCF1 localizing to secondary lymphoid organs and tertiary lymphoid structures within tumors, and terminally differentiated TCF1^– dysfunctional/exhausted TST found in the tumor parenchyma (reviewed in refs. 25, 26). In contrast, less is known about how SST differentiate in tumor-bearing hosts over time and in different tissues. Therefore, we developed an autochthonous liver cancer model, in which tumors develop slowly over weeks to months, in which we could assess CD8⁺ T-cell responses to SSA and TSA in parallel. We found that both SST and TST lost effector function rapidly upon antigen encounter in hosts with progressing liver lesions. However, while TST persisted within progressing liver lesions, SST did not persist, in large part due to premature cell-cycle exit. We instead found a population of persisting TCF1⁺PD-1^– SST in the spleen, which were nevertheless unable to proliferate or produce effector cytokines even after ICB treatment. T-cell dysfunction has mainly been associated with TCF1^– terminally differentiated T cells in tumors and during chronic infection, but we now identify what we believe to be a novel TCF1⁺PD-1^– dysfunctional CD8⁺ differentiation state in SST in normal tissues and tumor-bearing hosts.

All mice were bred and maintained in a specific pathogen free barrier facility at Vanderbilt University Medical Center (VUMC, Nashville, TN). All studies were approved by the Vanderbilt Institutional Animal Care and Use Committee regulations (IACUC). Mice were age- and sex-matched, between 6 and 12 weeks old when used for experiments, and assigned randomly to experimental groups. Both female and male mice were used. T-cell receptor (TCR)_GAG transgenic mice and Albumin-GAG (Alb-GAG) mice, which were obtained from Philip Greenberg, Fred Hutchinson Cancer Research Center, have been previously described (27, 28). TCR_TAG transgenic mice (stock no. 005236), Cre-ER^T2 (Stock No. 008463), and C57BL/6J Thy1.1 mice (Stock No. 000406) were purchased from The Jackson Laboratory. TCR_TAG and TCR_GAG mice were crossed to Thy1.1 mice to generate TCR_TAG; Thy1.1 and TCR_GAG; Thy1.1 mice, respectively. AST [AlbuminfloxStop-SV40 large T antigen (TAG)] mice, obtained from Natalio Garbi and Günter Hämmerling, German Cancer Research Center and previously described (29), were crossed to Cre-ER^T2 mice to generate AST;Cre-ER^T2. Alb-GAG mice were crossed to Cre-ER^T2 to generate Alb-GAG;Cre-ER^T2 mice, which were then crossed with AST mice to generate Alb-GAG;AST;Cre-ER^T2 (ASTxGAG).

Fluorochrome-conjugated antibodies were purchased from BD Biosciences, eBioscience/Thermo, BioLegend, Tonbo/Cytek Biosciences, and Cell Signaling Technology (for detailed antibody information see Supplementary Table S1). Tamoxifen (Sigma T5648) solution was prepared by warming tamoxifen at 50°C for 1 hour in sterile corn oil with 5% absolute ethanol. Tamoxifen (1 mg) was administered intraperitoneally as a single dose into AST;Cre-ER^T2 or Alb-GAG;AST;Cre-ER^T2 mice either 5 days prior to or 1 day following adoptive transfer, as indicated in experimental schemes.

Spleens from naive TCR_TAG or TCR_GAG, Listeria-inoculated C57BL/6, Alb-GAG, AST;Cre-ER^T2, or ASTxGAG mice were mechanically disrupted to a single-cell suspension with the back of a 3 mL syringe plunger, passed through a 70-μm strainer, and lysed with ammonium chloride potassium (ACK) buffer (150 mmol/L NH₄Cl, 10 mmol/L KHCO₃, 0.1 mmol/L Na₂EDTA). Cells were washed once with cold RPMI1640 (Corning MT10040CV) media supplemented with 10% FCS (RPMI-10; Corning 35010CV). Livers were mechanically disrupted to a single-cell suspension using a glass pestle against a 150-μm metal mesh in cold PBS containing 2% FCS (FCS/PBS) and filtered through a 100-μm strainer. The liver homogenate was spun down at 400 × g for 5 minutes at 4°C, and the pellet was resuspended in 15 mL FCS/PBS, 500 U heparin (NDC 63323–540–05), and 10 mL PBS Buffered Percoll (Cytiva 17089102), mixed by inversion, and spun at 500 × g for 10 minutes at 4°C. Pellets, enriched for liver-infiltrating lymphocytes, were lysed with ACK buffer, and cells were resuspended in RPMI-10 for downstream analysis.

Splenocytes or liver-infiltrating lymphocytes from naïve TCR_GAG or TCR_TAG mice, Listeria-inoculated B6 mice, Alb-GAG mice, AST;Cre-ER^T2 or ASTxGAG mice were stained with Ghost Dye Red 780 Viability Dye (1:2,000 dilution) Tonbo/Cytek 13–0865-T500) and antibodies against surface molecules (CD8, CD90.1, CD90.2, CD38, CD39, CD44, CD62L, CD127, PD-1, TCR Vα7, TCR Vβ12) in FCS/PBS. For detailed antibody information see Supplementary Table S1 and for cell dye information see Supplementary Table S2. Flow cytometry plots shown in figures are gated on live CD8⁺ CD90.1⁺ or CD90.2⁺ transgenic TCR_GAG or TCR_TAG cells. The gating strategy is shown in Supplementary Fig. S1. For intracellular staining, splenocytes or liver-infiltrating lymphocytes were surface stained as above and then fixed and permeabilized using the FoxP3 Transcription Factor Fix/Perm (Tonbo/Cytek TNB-0607) per the manufacturer's instructions before staining for intracellular molecules (TCF1, TOX, Ki67, BCL-2). The samples were then analyzed by flow cytometry (see Flow cytometric analysis). For analysis of effector cytokine production (IFNγ, TNFα), an ex vivo peptide stimulation was performed prior to staining. Splenocytes from naive, Listeria-inoculated, or tumor-bearing AST;Cre-ER^T2 or ASTxGAG mice were mixed with 2 × 10⁶ naive C57BL/6 splenocytes and incubated in RPMI-10 for 4 hours at 37°C in the presence of brefeldin A (Biolegend 420601) and peptide at the following concentrations: GAG peptide (CCLCLTVFL 1.5 μmol/L), TAG peptide (SAINNYAQKL 0.5 μmol/L; Genscript; custom-synthesized). The cells were then surface-stained (CD8, CD90.1) and fixed and permeabilized as described above, stained with antibodies against IFNγ and TNFα and analyzed by flow cytometry.

Flow cytometric analysis was performed using an Attune NxT Acoustic Focusing Cytometer (Thermo Fisher scientific). Flow data were analyzed with FlowJo v10 software (Tree Star Inc.).

Listeria monocytogenes (Lm) ΔactA ΔinlB strain expressing the GAG epitope (CCLCLTVFL, FMuLV GAG_75–83) or Tag-I epitope (SAINNYAQKL, SV40 large T antigen_206–215), which have been previously described (12, 20), were generated by Aduro Biotech and stored at −80°C. Naïve C57BL/6 mice were infected with 3 × 10⁷ c.f.u. LmGAG and 5 × 10⁶ c.f.u. of LmTAG via intravenous injection.

To transfer naive TCR_GAG T cells or TCR_TAG T cells into Alb-GAG, and ASTxGAG, or AST;Cre-ER^T2 mice, splenocytes were sterile isolated and processed from TCR_GAG;Thy1.1 or TCR_TAG;Thy1.1 transgenic mice as described above. Next, an aliquot was taken from the single-cell suspension and counted via microscopy/hemocytometer to obtain the total splenocyte concentration and stained with antibodies for CD8 and CD90.1 to determine the percentage of CD8⁺CD90.1⁺ TCR_GAG or TCR_TAG T cells. This percentage was multiplied by the splenocyte concentration and a cell suspension containing 2 × 10⁶ transgenic T cells/0.2 mL volume in plain RPMI was prepared, and 0.2 mL injected intravenously into each recipient mouse. For the in vivo generation of effector and memory TCR_GAG;Thy1.1 or TCR_TAG;Thy1.1 CD8⁺ T cells, the same procedure was used except that 5 × 10⁵ CD8⁺ splenocytes from TCR_GAG;Thy1.1 or TCR_TAG;Thy1.1 transgenic mice were adoptively transferred into C57BL/6 (B6) mice, which are Thy1.2. Mice were infected the following day with Listeria as described above.

Splenocytes from naive TCR_GAG;Thy1.1 or TCR_TAG;Thy1.1 mice were resuspended after first wash in 2.5 mL of plain, serum-free RPMI1640, rapidly mixed with equal volume of carboxyfluorescein succinimidyl ester (CFSE; Tonbo/Cytek 13–0850-U500; 10 μmol/L) solution, incubated for 5 minutes at 37°C at a final CFSE (5 μmol/L), quenched by mixing CFSE/cell solution with equal volume of pure FCS, washed twice with serum-free RPMI, and resuspended in serum-free RPMI for transfer. CFSE-labeled splenocytes were used for some adoptive transfer studies (indicated in figure legends) and for in vitro culture assays with hepatocytes (described below).

T cells were stained for surface and intracellular markers as described above (see Cell-surface and intracellular cytokine staining). Samples were suspended in FoxP3 Fix/Perm buffer (Tonbo/Cytek TNB-0607), and 4′,6-diamidino-2-phenylindole (DAPI; BioLegend 422801) was added to samples immediately prior to flow cytometric analysis. Samples were analyzed using the Watson Pragmatic cell cycle model in FlowJo v10 to determine cell-cycle phase. For apoptosis analysis, surface staining was performed as described above, and Annexin V PE (BioLegend, 640908) was added to the cells in Annexin V Binding Buffer (BioLegend, 422201) immediately prior to flow cytometric analysis.

PD1-specific (clone RMP1–14) and PDL1-specific (clone 10.F.9G2) antibodies or isotype control (clone LTA-2) were purchased from BioXcell. Antibodies were injected intraperitoneally five times, every other day, at 200 μg per antibody per mouse.

ASTxGAG and C57BL/6 mice were perfused under anesthesia with liver perfusion medium (Gibco 17701038) and liver digestion medium (Gibco 17703034) and then euthanized prior to harvesting the livers. Primary hepatocytes were isolated using Percoll gradient purification (GE) at low-grade centrifugation (50 × g) and incubated on collagen I–precoated plates (Corning 354249) as previously described (30). Hepatocytes were cultured in low-glucose DMEM (Gibco 11885084) supplemented with 5% FCS and 1% penicillin/streptomycin (Gibco 10378016) at 37°C with 5% CO₂. The following day, CFSE-labeled splenocytes from TCR_GAG;Thy1.1 or TCR_TAG;Thy1.1 transgenic mice were prepared as described above, and 100,000 CD8⁺ T cells were added to hepatocyte culture for 3 days. All cells were then removed from the wells and surface stained as described above with antibodies against CD8, CD90.1, and CD69 for flow cytometry analysis.

Previously published and available microarray data (GAG n = 3 biologic replicates (GSE32025; ref. 20) and TAG n = 4 biologic replicates (GSE60501; ref. 12) were analyzed using Broad Institute's Gene Set Enrichment Analysis software (GSEA; http://www.broadinstitute.org/gsea) to determine whether predefined gene sets showed enrichment in T-cell sample groups “GAG” versus “TAG.” Statistical significance was determined by permutation testing and normalized enrichment score (NES).

GraphPad Prism v9 was used for all statistical analyses. Unless otherwise stated, a one-way ANOVA was performed followed by a Tukey test to correct for multiple comparisons. The significance level was set at 0.05. Equal variances were verified by the Brown–Forsythe test.

The data generated in this study are available within the article and its Supplementary Data Files or upon request from the corresponding author.

We previously developed an autochthonous mouse model of liver cancer (AST;Cre-ER^T2), in which we can study TST differentiation and function from the initiation of carcinogenesis onward (12). After a single dose of tamoxifen, AST;Cre-ER^T2 hepatocytes undergo Cre recombinase–mediated induction of the TAG, under the control of the albumin promoter/enhancer. TAG acts as both an oncogene (inhibiting tumor suppressors TP53 and RB (31) to drive liver carcinogenesis) and a tumor-specific antigen (TAG = TSA). By 5 to 7 days post-tamoxifen, most AST;Cre-ER^T2 hepatocytes express TAG, and features of premalignant lesions are evident during the first 30 to 40 days posttamoxifen, with development of full-blown hepatocellular carcinoma by 3 months (Fig. 2 of ref. 12). By adoptively transferring naïve TCR_TAG, which are CD8⁺ T cells that recognize epitope I of TAG (32), into AST;Cre-ER^T2, we can study TST responses throughout carcinogenesis. To compare TSA- to SSA-specific responses, we turned to a model with a defined liver self-antigen developed by the Greenberg lab: the Alb-GAG mouse (Alb-GAG; ref. 28). Alb-GAG mice express the Friend murine leukemia virus gag gene under the control of the albumin promoter/enhancer, leading to hepatocyte-specific expression of the gPr75^gag protein (GAG = SSA) from birth on. Adoptively transferring GAG-specific transgenic CD8⁺ T cells (TCR_GAG; ref. 27) into Alb-GAG allows liver SSA responses to be studied.

We adoptively transferred naive TCR_TAG into AST;Cre-ER^T2 mice 5 days after tamoxifen treatment, just as premalignant lesions were beginning. In parallel, we transferred naive TCR_GAG into Alb-GAG mice and analyzed T-cell numbers, cytokine production, and immunophenotype in livers 5 and 21 days later (Fig. 1A). Both TCR_TAG, which in this context are TST and referred to as TCR_TAG/TST, and TCR_GAG, which in this context are SST and referred to as TCR_GAG/SST, were similarly unable to produce IFNγ and TNFα at 5 and 21 days (assessed by ex vivo re-stimulation with cognate peptide (TAG and GAG; Fig. 1B). While both TCR_TAG/TST and TCR_GAG/SST initially upregulated the activation/inhibitory marker PD1, TST continued to increase PD1 expression over time, while PD1 expression by TCR_GAG/SST dropped at the later time point (Fig. 1C). By day 21 posttransfer, TCR_TAG/TST upregulated CD39, a marker of late/terminal T-cell dysfunction/exhaustion (33), whereas TCR_GAG/SST did not (Fig. 1C). TCR_GAG and TCR_TAG transferred into B6 mice infected with attenuated Listeria monocytogenes expressing either TAG (LM_TAG) or GAG (LM_GAG) epitopes (Supplementary Fig. S1A) differentiated into highly functional effector and memory T cells, producing high levels of both IFNγ and TNFα (Fig. 1B; Supplementary Fig. S1B), demonstrating that TCR_TAG/TST and TCR_GAG/SST functional deficits in AST;Cre-ER^T2 and Alb-GAG mice were context-dependent. Consistent with prior observations (34), the TCR_GAG/SST had markedly impaired persistence, with fewer of these cells found in nonmalignant livers as compared with the TCR_TAG/TST in premalignant livers at early and late timepoints (Fig. 1D).

The improved persistence of TST in liver lesions as compared with SST in normal liver could be because of increased inflammation or other signals present in progressing tumors. To test this idea, we crossed AST;Cre-ER^T2 mice to Alb-GAG mice to create ASTxGAG mice. In ASTxGAG mice, hepatocytes express GAG constitutively from birth and TAG only after induction with tamoxifen. To confirm that both antigens were expressed, we isolated hepatocytes from tamoxifen-treated ASTxGAG and B6 mice, for control, and co-incubated them with naïve TCR_GAG and TCR_TAG (Supplementary Fig. S2A). Both TCR_TAG and TCR_GAG upregulated CD69 in response to ASTxGAG but not B6 hepatocytes (Supplementary Fig. S2B), demonstrating that tamoxifen-treated ASTxGAG hepatocytes similarly activate TCR_TAG and TCR_GAG.

We transferred naïve TCR_TAG and TCR_GAG into ASTxGAG (5 days post-tamoxifen) and B6 (antigen-free) mice (Fig. 2A). Both TCR_TAG and TCR_GAG rapidly lost the ability to produce effector cytokines in ASTxGAG mice, where they are TST and SST, respectively (Supplementary Fig. S3A and S3B), similar to what we had observed in Alb-GAG and AST;Cre-ER^T2 mice (Fig. 1B). Oncogene induction in ASTxGAG livers did not restore the numbers of TCR_GAG/SST to the level of TCR_TAG/TST (Fig. 2B). By day 21 posttransfer, few TCR_GAG/SST remained in ASTxGAG livers, although they did persist in the livers of antigen-free B6 mice (Fig. 2B), demonstrating that presence of the cognate SSA, in this case GAG, led to SST disappearance. TCR_TAG/TST expressed high levels of PD1 and TOX, a DNA-binding protein highly expressed in dysfunctional/exhausted CD8⁺ T cells (35), while TCR_GAG/SST in the same premalignant environment only transiently expressed low levels of PD-1 and TOX (Fig. 2C). In contrast, BCL2, a prosurvival/antiapoptosis factor, was similarly downregulated at day 5 and upregulated at day 21 in both TST and SST (Fig. 2D), which was unexpected to us given the failure of TCR_GAG/SST to persist at day 21. However, fewer TCR_GAG/SST than TCR_TAG/TST were MKI67⁺ at days 5 and 14 (Fig. 2D), suggesting impaired proliferation by the TCR_GAG/SST.

We next sought to determine why TCR_GAG/SST failed to persist in ASTxGAG mice. We labeled naive TCR_TAG and TCR_GAG with CFSE and adoptively transferred CFSE-labeled T cells into the following mice: untreated B6, LM_TAG or LM_GAG-infected B6, or ASTxGAG treated with tamoxifen 5 days prior. TCR_TAG and TCR_GAG were re-isolated from premalignant livers 3d posttransfer for flow cytometric analysis (Fig. 3A). Even at this early time point, fewer TCR_GAG/SST than TCR_TAG/TST were found in ASTxGAG livers (Fig. 3B). To assess cell death, we stained T cells with annexin V and DAPI and found similar proportions of total apoptotic (Annexin V⁺), early apoptotic (Annexin V⁺DAPI^–), and late apoptotic (Annexin V⁺DAPI⁺) TCR_GAG/SST and TCR_TAG/TST in the liver (Fig. 3C; Supplementary Fig. S4A and S4B). Infection-activated TCR_GAG and TCR_TAG underwent apoptosis at a much higher rate than TCR_GAG/SST and TCR_TAG/TST in ASTxGAG mice (Fig. 3C; Supplementary Fig. S4A and S4B), despite the fact that infection-activated TCR_TAG and TCR_GAG accumulated in higher numbers (Fig. 3B). While apoptosis has been thought to be the main driver of deletional tolerance (16), this has mainly been inferred from cell numbers (19) because quantifying apoptosis in vivo is challenging because of rapid clearance of apoptotic cells (36). Given the lack of correlation between apoptosis rates and persistence, we next examined proliferation by assessing CFSE dilution. TCR_GAG/SST were mainly found in earlier divisions (divisions 4–6), in contrast to TCR_TAG/TST, which accumulated in the later divisions (divisions 7–8+; Fig. 3D). Given that TCR_GAG/SST failed to proliferate to the same extent as TCR_TAG/TST, we analyzed cell-cycle distribution on later division TCR_TAG/TST and TCR_GAG/SST using DAPI/DNA content staining. While TCR_TAG/TST in cell divisions 4+ were actively cycling, as shown by the high proportion of these cells in S and G₂–M, TCR_GAG/SST were nearly all in G₀–G₁ (Fig. 3E). Thus, the TCR_GAG/SST proliferated initially but stopped cycling prematurely, leading to decreased numbers of these cells in the liver and tumor lesions.

Given the failure of TCR_GAG/SST to persist in normal and premalignant liver lesions, we next asked whether these cells could home to and/or persist at other sites, including secondary lymphoid organs. We carried out adoptive transfers of naïve TCR_TAG/TST and TCR_GAG/SST into Listeria-infected B6 mice or tamoxifen-treated (5 days prior) ASTxGAG mice (Fig. 4A). Thirty days posttransfer, we recovered similar numbers of TCR_TAG/TST and TCR_GAG/SST from the spleens of ASTxGAG mice (Fig. 4B). TCR_GAG/SST in ASTxGAG spleens were immunophenotypically indistinguishable from memory TCR_GAG generated in LM_GAG-infected B6 mice, having a central memory immunophenotype (CD44⁺CD62L⁺; Supplementary Fig. S5A) with upregulation of IL7R (also known as CD127), TCF1 (Fig. 4C), and BCL2 (Supplementary Fig. S5B), and low PD1, TOX (Fig. 4C), and CD39 (Supplementary Fig. S5B) expression. In contrast, TCR_TAG/TST in ASTxGAG spleens expressed high levels of PD1, TOX (Fig. 4C), and CD39 (Supplementary Fig. S5B) and low levels of CD127 and TCF1 (Fig. 4C). Although TCR_GAG/SST in ASTxGAG spleens had a memory-like immunophenotype, they were incapable of producing effector cytokines (Fig. 4D).

We next wanted to determine whether nontransgenic tumor-reactive CD8⁺ T cells undergo similar differentiation as TCR_TAG/TST or TCR_GAG/SST. Given that endogenous TAG- or GAG-specific CD8⁺ T cells are absent from ASTxGAG mice because of central tolerance (28, 29), we examined endogenous PD-1⁺CD8⁺ T cells, which are enriched for T cells reactive to other tumor antigens (37) arising in TAG-transformed hepatocytes. We found that endogenous PD-1⁺CD8⁺ T cells were heterogeneous, including three subsets: TCF1⁺TOX^– (similar to memory-like TCR_GAG/SST), TCF1⁺TOX1⁺ (similar to early TCR_GAG/SST or TCR_TAG/TST), and TCF1^–TOX⁺ (similar to late TCR_TAG/TST; Supplementary Fig. S5C and S5D). These data suggest that some endogenous nontransgenic tumor-reactive PD-1⁺CD8⁺ T cells also enter the TCF1⁺ memory-like state.

To test whether the induction of the memory-like dysfunctional state was specific to premalignant/malignant hosts, we adoptively transferred TCR_GAG into Alb-GAG hosts and into LM_GAG-infected B6 mice (Supplementary Fig. S6A). As late as 55 days posttransfer, memory-like dysfunctional TCR_GAG/SST were found in the spleens of Alb-GAG mice (Supplementary Fig. S6B) expressing TCF1, CD44, CD62L, and CD127 (Supplementary Fig. S6C), and unable to produce effector cytokines (Supplementary Fig. S6D).

We next asked which transcriptional features were associated with this TCF1⁺ dysfunctional state and how they compared with TCF1^– dysfunctional TST by using our previously reported transcriptional profiling data obtained from late TCR_TAG/TST transferred into AST;Cre-ER^T2 mice (12) and TCR_GAG/SST into Alb-GAG hosts (20). We found that 628 probes mapping to 535 unique genes were differentially expressed in TCR_TAG/TST as compared with TCR_GAG/SST (Supplementary Fig. S7A). Dysfunction/exhaustion-associated genes such as Ctla4, Gzmk, Lag3, Epntpd1 (encoding CD39), and Rgs16, recently shown to promote inhibit ERK1 activation in exhausted T cells (38), were upregulated in TCR_TAG/TST (Supplementary Fig. S7A). In contrast, genes associated with stemlike/progenitor T-cell states and lymphoid homing [Lef1, Tcf7 (encoding TCF1), Sell (encoding CD62L), Ccr7, Klf2, and Bach2] were upregulated in TCR_GAG/SST (Supplementary Fig. S7A). We carried out GSEA (39, 40) to identify pathways associated with TCR_GAG/SST and found enrichment in genes associated with oxidative phosphorylation (Supplementary Fig. S7B), which are also enriched in memory T cells and other quiescent T cells (41). TCR_GAG/SST were also enriched in gene sets associated with memory CD8⁺ T cells (Supplementary Fig. S7C) in contrast to exhausted or effector CD8⁺ T-cell gene sets, which were enriched in TCR_TAG/TST (Supplementary Fig. S7D). Thus, while both TCR_TAG/TST and TCR_GAG/SST lack effector function and are unable to prevent tumor outgrowth, the underlying transcriptional programs controlling their function and phenotype markedly differ.

In hosts with tumors and chronic infection, TCF1⁺PD-1^low stem/progenitor-like exhausted CD8⁺ T cells localized in secondary lymphoid organs and tertiary lymphoid structures preferentially proliferate and respond to ICB targeting PD-1 or PD-L1 (42–48). Thus, we asked whether persistent TCF1⁺ SST would respond better to ICB than TCF1^– TST. We cotransferred TCR_TAG/TST and TCR_GAG/SST into tamoxifen-treated ASTxGAG mice (5 days posttamoxifen, waited 21+ days, by which point TCR_GAG/SST had entered the TCF1⁺PD1^–/low state, and treated ASTxGAG mice with combination anti–PD-1/anti–PD-L1 or isotype control (Fig. 5A). Not only did TCR_GAG/SST fail to regain effector function in response to ICB (Fig. 5B), they did not upregulate MKI67 in response to ICB (Fig. 5C), and we did not observe increased TCR_GAG numbers (Fig. 5D). TCR_TAG/TST also failed to regain effector function in response to ICB (Fig. 5B), consistent with what we previously observed (12), although TCR_TAG/TST did proliferate and show lower PD1 staining in ICB-treated ASTxGAG mice (Fig. 5C). Accordingly, we did not observe any slowing of ASTxGAG tumor progression, as measured by liver weight (Fig. 5E). The failure of TCF1⁺ TCR_GAG/SST to proliferate or regain effector function in response to ICB suggests that TCF1 expression and/or associated memory-like transcriptional and phenotypic programs are not sufficient to enable ICB-mediated T-cell rescue.

To characterize human tumor-reactive T cells from patients, where T-cell specificities are usually unknown, researchers have adopted dysfunctional hallmarks identified from studying tumor-specific or viral-specific T cells, including expression of inhibitory receptors such as PD-1 and CD39, downregulation of TCF1, and upregulation of TOX (26, 49). However, we now find that CD8⁺ T cells specific for SSA enter what we believe to be a previously uncharacterized dysfunctional state immunophenotypically indistinguishable from central memory T cells. Persisting SST were CD44⁺CD62L⁺CD127⁺TCF1⁺, and they did not express PD1, CD39, or TOX. Unlike classical memory T cells generated during acute infection, SST were unable to produce effector cytokines in response to stimulation with cognate antigen, similar to TCF1^–TOX⁺ terminally differentiated TST. We did find that the endogenous tumor-reactive PD1⁺CD8⁺ T-cell population in ASTxGAG mice was heterogeneous, with TCF1⁺TOX1^–, TCF1⁺TOX⁺, and TCF1^–TOX⁺ subsets present, suggesting that the TCF1⁺ memory-like state is not unique to TCR_GAG but present in T cells specific for other SSA/tumor antigens. In line with this, a recent study profiled the immunophenotype and specificity of tumor-infiltrating CD8⁺ T cells (TIL) in patients with melanoma. Classical dysfunctional/exhausted T cells were the dominant subgroup; however, a smaller subset of TIL with memory-like phenotype (PD-1^–TCF1⁺TOX^–) were also present. While many were bystander viral-specific memory T cells, a small percentage (<10%) of the memory phenotype TIL had specificity for patient-derived melanoma cells and also nonmelanoma cells lines (50), suggesting SSA specificity. Thus, our findings on TCR_GAG/SST in the ASTxGAG model may be applicable to other tumor types and to human patients with cancer.

Our findings have important implications for cancer immunology, demonstrating that (i) tumor-antigen type (TSA vs. SSA) can be an important determinant of tumor-reactive T-cell differentiation, (ii) tumor-reactive T cells can lose effector function via classical exhaustion/dysfunction pathways or through differentiation to a memory-like PD-1^–TCF⁺TOX^– nonfunctional state, and (iii) the presence of dysfunctional/exhausted markers (PD-1, CD39) or absence of functional/memory-associated transcription factors (TCF1) does not predict whether tumor-reactive CD8⁺ T cells are functional.

CD8⁺ T-cell differentiation in response to SSA was conserved whether the antigen was expressed in the context of transformed hepatocytes or nontransformed hepatocytes, suggesting that CD8⁺ T-cell differentiation, functional status, and persistence are driven by the nature of the antigen (SSA vs. TSA) and not by the tumor microenvironment. The failure of SSA-reactive T cells to persist has long been attributed to deletional tolerance (16). However, we did not see an increase in the number of apoptotic SST as compared with either TST or to SSA-specific T cells responding to their cognate antigen in the context of acute infection. Rather, SST failed to sustain the initial robust proliferation in response to antigen and exited cell-cycle prematurely, in contrast to TST or infection-activated SSA-specific T cells. Future studies are needed to determine the mechanism by which SST leave cell cycle prematurely while TST do not.

An important question arising is what drives the divergent differentiation paths of TST and SST in our model? Dysfunctional hallmarks such as PD-1, CD39, and TOX, are driven by persistent TCR signaling in the setting of tumors or chronic viral infection and dependent on the transcription factor NFAT (51). TCR_GAG encounter their cognate GAG antigen in both the spleen and liver of Alb-GAG mice, as suggested by the presence of early division TCR_GAG in both organs. However, TCR_GAG persisting in the spleen downregulate PD1, maintain TCF1 expression, reexpress CD127 and BCL2, and harbor gene expression profiles of quiescent T cells, similar to cells undergoing CD8⁺ T-cell differentiation to the memory state during acute infection following pathogen/antigen clearance. Taken together, these findings suggest that SST are not receiving persistent TCR stimulation, in contrast to TST. TCR_GAG may persist in an antigen-free or antigen-low niche within the spleen or other secondary lymphoid organs, though this remains to be investigated. Another determinant of T-cell differentiation in tumors is TCR affinity/avidity (52). Comprehensive profiling of TCR affinity/avidity of TIL in patients with melanoma has revealed that CD8⁺ T cells specific for neoantigens had higher avidity than SSA-specific T cells (50). A recent study using a murine colorectal cancer model found that lower-affinity tumor-infiltrating CD8⁺ T cells have a progenitor phenotype (TCF1⁺TOX^–) while higher affinity T cells maintain a more exhausted phenotype (TCF1^–TOX⁺) with higher proliferation (53). Future studies are needed to determine whether it is the lower affinity/avidity of SSA that skews responding T cells to the TCF1⁺PD-1^– dysfunctional state, or whether other differences between SSA and TSA, such as tissue distribution or expression in secondary lymphoid organs, play a role.

Despite having a memory-like immunophenotype and TCF1 expression, SST failed to regain effector function or even proliferate in response to ICB. Another recent study in a murine lung cancer model found that even though lower-affinity CD8⁺ T cells were more likely to be TCF1⁺ as compared with higher-affinity CD8⁺ T cells, they did not have improved responses to ICB (54). The failure of SST/TCR_GAG to mediate antitumor or autoimmune effects with ICB in ASTxGAG mice demonstrates that TCF1 expression alone does not predict responsiveness to ICB, but it does leave the question open as to why some patients with SST go on to develop irAE after ICB therapy. A recent large retrospective analysis of patients treated with pembrolizumab (anti–PD-1) found that those diagnosed with infection were 80% more likely to develop an irAE, although it was not possible to establish whether the infection preceded the irAE or vice versa (55). Another study in mouse tumor models showed that dendritic cell production of IL12 was required for successful ICB-mediated T-cell responses (56), suggesting that concurrent inflammation or infection may determine whether ICB unleashes SST-mediated autoimmune and/or antitumor responses. Work from our group suggests that there is a “Goldilocks” range for T-cell affinity, in which too high affinity leads to dysfunction while too low affinity induces not dysfunction but functional inertness (52). Lower affinity memory-like SST may not proliferate in the presence of ICB alone but require inflammatory cytokines/dendritic cell priming to push them over the proliferation/effector differentiation threshold.

Future studies using our ASTxGAG model, as well as the development of other similar models with other tissue/tumor-expressed antigens, can unravel how antigen properties such as affinity/avidity and inflammatory context regulate tumor-reactive CD8⁺ T-cell differentiation and whether/how ICB impacts the subsequent balance between antitumor T-cell responses and T cell–mediated irAE.

M.M. Erwin reports grants from NIH during the conduct of the study. N.R. Favret reports grants from NIH during the conduct of the study. C.R. Detrés Román reports grants from NIH during the conduct of the study. M.K.I. Apostolova reports grants from SyBBURE-Searle Undergraduate Research Program and grants from The Barry Goldwater Scholarship and Excellence in Education Foundation during the conduct of the study. M. Philip reports grants from V Foundation, grants from NIH, and grants from Serodino Family Adventure Allee Fund during the conduct of the study. No disclosures were reported by the other authors.

J.J. Roetman: Conceptualization, methodology, writing–original draft, writing–review and editing. M.M. Erwin: Methodology, writing–review and editing. M.W. Rudloff: Methodology, writing–review and editing. N.R. Favret: Methodology, writing–review and editing. C.R. Detrés Román: Methodology, writing–review and editing. M.K.I. Apostolova: Methodology. K.A. Murray: Methodology, writing–review and editing. T.-F. Lee: Methodology, writing–review and editing. Y.A. Lee: Methodology, writing–review and editing. M. Philip: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing.

We thank A. Schietinger and members of the Philip laboratory for helpful discussions. We thank P. Greenberg for providing Alb-GAG and TCR_GAG mice. We thank G. Anzarova and T. Bryson for technical assistance. We thank the Vanderbilt Division of Animal Care, and P. Lauer and Aduro Biotech for providing attenuated Listeria strains. We thank K. Ray for assistance with hepatocyte isolation. This work was supported by the following funding sources: V Foundation Scholar Award (to M. Philip), NIH R37CA263614 (to M. Philip), Serodino Family Adventure Allee Fund (to M. Philip), Vanderbilt-Ingram Cancer Center (VICC) SPORE Career Enhancement Program NIH P50CA098131 (to M. Philip), Vanderbilt Digestive Disease Research Center (VDDRC) Young Investigator and Pilot Award NIH P30DK058404 (to M. Philip), NIH T32GM008554 (to N.R. Favret), NIH T32CA009592 (to C.R. Detrés Román), NIH T32AR059039 (to M.M. Erwin), Barry Goldwater Scholarship (to M.K.I. Apostolova), SyBBURE Searle Program at Vanderbilt (to M.K.I. Apostolova), Young Investigator Award P30DK058404 (to Y.A. Lee), pilot research grant P30DK058404 (to Y.A. Lee), pilot research grant VICC GI SPORE P50CA236733 (to Y.A. Lee), American Cancer Society RSG-22–061–01-MM (to Y.A. Lee), and IRG-19–139–60 (to Y.A. Lee).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.