Low TCR Binding Strength Results in Increased Progenitor-like CD8⁺ Tumor-Infiltrating Lymphocytes

The strength and impact of the T-cell receptor (TCR)–peptide-MHC interaction on antitumor immunity is the topic of many years of research [reviewed in (1–5)]. Naïve T cells that receive strong signals through their TCR are more rapidly activated than those that receive weak signals (6, 7). Adoptive cell therapy experiments using T cells with a range of avidities and affinities for a self/melanoma antigen demonstrate that optimal antitumor immune responses occur when the monomeric interaction is about 10 μm (8) and that stronger binding does not improve responses. To determine the effect of TCR binding strength on gene expression and epigenetic programs, Shakiba and colleagues examined a panel of altered peptide ligands for a TCR against SV40 large T antigen epitope 1 (9). These experiments suggest that there is a “Goldilocks zone” in which an intermediate strength of engagement provides reduced T-cell exhaustion and improves T-cell effector function. These results suggest a hypothesis that two TCR-dependent pathways act on the T-cell repertoire within the tumor microenvironment (TME): proliferation and differentiation, both of which are increase in higher affinity T cells (9–12). The summation of these influences and how they impact T cells are key in the quality of antitumor T-cell responses.

T-cell exhaustion is a spectrum of hypofunctional conditions induced by chronic antigen exposure, which commonly occurs in response to chronic viral infections and antigens in the TME [reviewed in (13–16)]. Differentiation to terminal exhaustion is irreversible; T cells lose their capacity to de-differentiate into less exhausted phenotypes or into memory cells and become more susceptible to cell death (17–20). As T cells progress from naïve cells towards exhaustion, they have increased effector functions, including IFNγ and granzyme B production (17, 18, 21). However, exhausted-effector cells ultimately lose the capacity to proliferate and effectively respond to antigen. A subset of T cells expressing the progenitor transcription factor Tcf7 remains proliferative and replenishes the terminally exhausted subset (17, 18, 21, 22).

In this study, we used the immunogenic CT26 tumor model in BALB/c mice to examine the impact TCR affinity for antigen has on the endogenous, polyclonal CD8⁺ tumor-infiltrating lymphocyte (TIL) response against a single tumor-associated antigen (TAA). This antigen, gp70_423–431, is derived from an endogenous retrovirus, restricted by the H-2L^d MHC class I molecule, and is recognized by CD8⁺ T cells (23, 24). T cells with a broad array of affinities for this antigen can be distinguished and isolated with a peptide-loaded H-2L^d MHC tetramer. Using a technique established by others and us, we used fluorescent-tetramer binding to suggest the relative affinity of TILs to antigen (9, 24–28). We also developed a novel version of this technology by conjugating oligo-barcodes to MHC-tetramers to distinguish relative affinity by cellular indexing of transcriptome and epitopes by sequencing (CITE-seq). Using this model system, we showed that T-cell dysfunction and differentiation in the TME was dependent on differences in TCR binding strength.

We additionally identified sub-populations of CD8⁺ TILs that were significantly enriched for low tetramer binding (Tet-Low). Tet-Low TILs were slower to progress towards terminal phases of T-cell exhaustion and retained a progenitor-like phenotype for longer in the TME than TILs with high tetramer binding (Tet-High). We found increased expression of quiescence genes and decreased expression of T-cell exhaustion and effector genes amongst Tet-Low TILs. Together, these results suggested that Tet-Low TILs persist longer, proliferate less, and have less tumoricidal potential. Using a transgenic mouse characterized by a low-affinity TCR (29), we confirmed that low-affinity TILs were less effective at controlling tumor growth, but sustained a progenitor-exhausted state and failed to differentiate into terminal phases of T-cell exhaustion. We propose further differentiation is necessary for effector functions that directly control tumor growth. These experiments set up a dynamic model in which Tet-Low TAA-specific TILs progress slowly towards exhaustion, but do not receive the stimulation necessary to mount an effective antitumor response. Tet-High T cells progress rapidly to phases of terminal exhaustion, expressing effector molecules (e.g., Ifng and Gzmb), and thus, are more tumoricidal.

Six- to 8-week-old female BALB/c (BALB/cAnNCr) mice were purchased from Charles River Laboratories. The 1D4-TCR transgenic mice were made by the Mouse Genetics Core Facility at National Jewish Health. 1D4-TCR encodes a Vβ8.3/Vα6 (TRBV13-1/TRAV21) TCR (30) with low affinity (100 μm) for the self gp70_423–431 peptide complexed with H-2L^d (31). The alpha and beta TCR chains were inserted into shuttle vectors (32), which were subsequently injected into embryos of gp70-deficient mice [generated in-house as described previously (28)] then backcrossed onto the Rag2-deficient gp70-deficient mice [{(C.12956(B6)-Rag2^tm1fwaN12) and BALB.B6 env-/-, respectively}; all mice were on a gp70-deficient background to avoid negative selection of the TCR]. Germline integration and subsequent expression of the TCR genes were established by PCR and tetramer staining. All animal protocols were reviewed and approved by the Institutional Animal Care and Use Committee at the University of Colorado School of Medicine.

CT26 cells were purchased from ATCC in 2014 and tested for Mycoplasma contamination by PCR at the Barbara Davis Center Bioresource Core prior to freezing. Aliquots were used for no more than 1 month after thawing CT26 cells were cultured in “complete medium”: RPMI with l-glutamine (Life Technologies), 10% FBS (Hyclone), 100 U/mL each penicillin and streptomycin (Life Technologies), 1 mmol sodium pyruvate (Life Technologies), 10 mmol HEPES (Life Technologies), and 1× minimum essential medium nonessential amino acids (Life Technologies).

BALB/c, gp70-deficient and 1D4-TCR mice were injected subcutaneously with 1 × 10⁵ CT26 tumor cells in 100 μL 1X PBS (Life Technologies) in both hind flanks (33). Starting 7 days after injection, mice were palpated every other day to detect tumor growth. Tumors and spleens were harvested 7, 10, or 14 days posttumor challenge as indicated. Unless otherwise indicated, tumors were harvested 14 days posttumor challenge.

TILs were isolated as described (33). Briefly, CT26 tumors were harvested from BALB/c mice, weighed, minced using a razor blade, and treated for 25 minutes at 37°C with 0.1 mg/mL Liberase (Research Grade, Dispase Low) in 5 mL serum-free RPMI1640 medium according to the manufacturer's instructions (Roche Life Science). Tumors were filtered through a 100-μm cell strainer and washed in complete medium supplemented with 0.1 mmol/L beta-mercaptoethanol (Sigma). Spleens were mechanically dissociated and filtered through a 100-μm strainer prior to RBC lysis using the manufacturer's instructions (ThermoFisher).

For in vitro culture of TILs and single-cell RNA sequencing (scRNA-seq) experiments, tumors from 15 mice were pooled together to make a single sample. Before sorting via FACS, CD8⁺ T cells were enriched using a Mouse CD8 Negative Selection Kit (STEMCELL Technologies) according to the manufacturer's instructions. CD8-enriched TILs were prepared for sorting as described below, sorted into Tet-High, Tet-Low, and Tet-negative groups, and immediately processed. For in vitro culture, sorted CD8⁺ T cells were plated at 1×10⁶ cells/mL with 20 U/mL IL2 and incubated at 37°C and 5% CO₂ for 1 week. For other experiments, cells from a single mouse directly following harvest were used for downstream applications without CD8 purification.

The gp70_423–431 peptide (also known as AH1, SPSYVYHQF)-loaded H-2L^d MHC tetramers were obtained from the NIH Tetramer Core Facility. Tetramers were titrated to ensure an excess of tetramer was added to each sample when staining. Biotinylated H-2L^d monomers loaded with the gp70_423–431 peptide were also obtained from the NIH Tetramer Core Facility and tetramerized with a PE and oligonucleotide-barcode-conjugated streptavidin molecule (BioLegend, TotalSeq-A0951 PE Streptavidin product # 405251). Barcoded MHC tetramers were purified after tetramerization using fast protein liquid chromatography on an AKTA Pure Purification System equipped with a Superdex 200 column to isolate tetramerized MHC molecules from monomers and other smaller products. Using Unicorn version 6.3 software, the samples were run over the column at 0.75 mL/minute in 1X PBS (Life Technologies) with 0.1% sodium azide (Sigma) and tetramer-containing fractions were collected and pooled. Fractions were concentrated with Centricon concentrators (Millipore) to approximately 1 mg/mL.

For flow cytometry analyses, 2 × 10⁶ live cells were washed in PBS and stained with a viability dye (Supplementary Tables S1 and S2) for 30 minutes in the dark, per the manufacturer's instructions. Samples were then washed with flow buffer [1× PBS, 2% FBS, and 0.1% (w/v) sodium azide, as described (33)] and stained for surface markers. For all experiments with tetramer staining, Fc Block (TruStain FcX, anti-mouse CD16/32, BioLegend, product #101320) was added to each sample directly before tetramer was added, which was incubated for 1 hour. 50 μL of an antibody cocktail was added to each sample and incubated for 30 minutes. If staining with tetramer, surface marker antibodies were added during the final 30 minutes of tetramer staining. Staining with CITE-seq antibodies (Supplementary Table S3) was performed after tetramer staining and directly before flow cytometry antibodies were added. Before cell sorting, cells were diluted to ∼1 × 10⁷ cells/mL sorting buffer (flow buffer with 1 mmol EDTA) and sorted into complete medium on a BD Aria Fusion instrument. Other flow cytometry experiments used the Cytek Aurora instrument (with panels in Supplementary Tables S1 and S2). Data were analyzed using FlowJo software (Tree Star). The CD8⁺ T cell gating strategy is shown in Supplementary Fig. S1. CD8⁺ subpopulations were further gated and analyzed for programmed cell death protein 1 (PD-1), TIM3 or tetramer binding. Tetramer-negative gates were set based on CD8⁺ T cells from a spleen from a “naive" mouse, which do not bind tetramer. The tetramer-positive population was divided roughly in half to set the gates for Tet-High and Tet-Low, based on day 14 TILs when performing analysis over multiple days (Supplementary Fig. S1). When sorting based on tetramer staining, gates did not overlap, so some cells between Tet-High and -Low were precluded from the gating, impacting the Tet-High populations more than the Tet-Low populations.

CITE-seq was performed on the CT26 TILs using BioLegend TotalSeq-A antibodies (Supplementary Table S3) and the Chromium Single Cell 3′ Reagent Kits v3, following the manufacturer's protocols. Viability was assessed using TC20 (Bio-Rad), and cells were diluted to target 5,000 cells. The samples went onto the 10x Chromium machine for gel bead emulsions (GEM) generation and barcoding. Briefly, 10x Barcoded Gel Beads were mixed with the surface protein labeled cells in a master mix and partitioning oil to obtain single-cell GEMs. Gel beads were dissolved allowing for incubation of both the poly-A RNA and cell surface proteins within each GEM resulting in full-length cDNA and DNA from the cell surface protein. GEMs were cleaned following the Chromium Single Cell 3′ Reagent Kits v3 protocol. The TotalSeq-A Antibodies with 10x Single Cell 3′ Reagent Kit v3 3.1 protocol was followed, and a cDNA amplification mix was prepared to generate Antibody-Derived Tag (ADT) libraries. Sequencing was performed on the Illumina NovaSeq 6000.

Using the BioLegend TotalSeq-A antibodies and protocol along with the Chromium Single Cell 3′ Reagent Kits v3, scRNA-seq and CITE-seq were performed on two separate occasions. Tumor samples from 15 mice were pooled together for each experiment. Cells were received in the National Jewish Health Genomics Core, viability was assessed using TC20 (Bio-Rad), and cells were diluted to an appropriate volume to target 5,000 cells. The samples then went onto the 10x Chromium machine for GEMs (gel beads in emulsion) generation and barcoding. Briefly, 10x Barcoded Gel Beads were mixed with the surface protein-labeled cells in a master mix and partitioning oil to obtain single cell GEMs. Gel beads were dissolved allowing for incubation of both the poly A RNA and cell surface proteins within each GEM resulting in full-length cDNA and DNA tagged to the cell surface protein. GEMs were then cleaned following step 2.1 in the Chromium Single Cell 3′ Reagent Kits v3 protocol. TotalSeq-A Antibodies with 10x Single Cell 3′ Reagent Kit v3 3.1 was followed at Step II. cDNA amplification mix was prepared according to this protocol to allow for ADT libraries. The libraries were finished with the Protocol – TotalSeq-A Antibodies with 10x Single Cell 3′ Reagent Kit v3 3.1. 150-bp paired-end reads were sequenced at the University of Colorado Genomics Shared Resource lab on the Illumina NovaSeq 6000.

scRNA fastq files were mapped to the mouse genome version GRCm38 and counted using 10X Cell Ranger 3.0.2 software generating cell/gene counts (34). CITE-seq fastq files were counted using CITE-seq-count 1.4.2 (35). Tables were imported into R version 4.0.3 "Bunny-Wunnies Freak Out" (http://www.r-project.org/index.html) with Seurat v3.1 for analysis (36–39). Cells were excluded from analysis if they had less than 200 features, greater than 30,000 RNA molecules, greater than 7 counts/feature ratio, greater than 7% mitochondrial RNA, and tetramer counts over the 95 percentile. In addition, due to a bimodal distribution of tetramer counts in the high and low libraries indicating a large number of “tetramer-negative cells”, the cells with tetramer counts less than the minimum tetramer count value between bimodal peaks on a histogram were removed. Histogram breaks were set at equal to one third the maximum of the 95th percentile of tetramer counts value. Data were normalized, and the top 5,000 variable genes, plus Cd8a and Cd38, were used for integration using the "LogNormalize" method. The top 26 principal components from the integrated data with a resolution of 0.5 was chosen for clustering the cells. Tetramer barcode counts were not used as a variable in the clustering of RNA data and generation of the Uniform Manifold Approximation and Projection (UMAP). ADT CITE-seq data were analyzed and clustered as according to Seurat v3.1 multimodal vignette with a resolution of 0.4. All additional plots and analysis were performed on the normalized RNA/ADT counts in the RNA/ADT assay data slots with Seurat tools. Venny (https://bioinfogp.cnb.csic.es/tools/venny/index.html) was used to generate Venn diagrams and AUcell (40) was used to apply gene sets data to the UMAP.

We examined scRNA-seq gene lists generated from CD8⁺ TILs from the B16 tumor model (41) to further validate the labels we assigned to the CD8⁺ TIL clusters in the CT26 tumor model Venny (https://bioinfogp.cnb.csic.es/tools/venny/index.html) was used to generate Venn diagrams and AUcell (40) was used to apply gene sets data to our UMAP.

ScRNA-seq and CITE-seq data are publicly available at Gene Expression Omnibus accession number GSE212980.

Using the CT26 colon carcinoma model of BALB/c, which expresses the immunodominant MHC class 1 antigen gp70_423–431, we tracked antigen-specific T cells using an H-2L^d MHC-tetramer (23, 24). We determined how the binding strength between T cells and TAA influenced critical pathways in tumor immunity, by analyzing the tetramer staining intensity of polyclonal, antigen-specific TILs. Using FACS, we stratified CD8⁺ TILs by tetramer binding intensity (Supplementary Fig. S1): high tetramer binders (Tet-High), low tetramer binders (Tet-Low), and cells that failed to bind tetramer (Tet-negative) as described in the materials and methods. After 7 days of culture with IL2, the TIL retained their relative tetramer binding strength (Fig. 1A and B). The consistent fluorescence intensity showed that tetramer binding was not a transient characteristic of CD8⁺ TILs but was durable overtime and reflected the relative binding strength of TCRs for tumor antigen. Therefore, we used the fluorescence intensity of tetramer staining to ascertain the relative binding strength of T cells from a polyclonal T-cell repertoire. To investigate if the differences in tetramer binding were due to differences in the number of TCR complexes on the cell surface, we normalized tetramer binding to CD3 expression (Fig. 1C and D). T cells remained significantly different based on tetramer binding: high, low, and negative (Fig. 1D and E). These data provide evidence that tetramer staining represents differences in TCR binding strength, rather than simply differences in TCR surface expression.

We next examined Tet-High and Tet-Low CD8⁺ TILs at 7, 10, and 14 days post CT26 tumor injection. Although day 7 tumors were smaller, they were identifiable, and harbored gp70_423–431-specific T cells. The highest frequency of Tet-Low TILs was observed on day 7, greater than Tet-High TILs. The frequency of Tet-High TILs increased between day 7 and 10 and remained high through day 14 (Fig. 2A), demonstrating that Tet-High TILs accumulate in the TME over time. The rate at which Tet-High TILs differentiated toward terminal exhaustion, identified by the surface expression of inhibitory receptors PD-1 and TIM3 (42, 43), was increased relative to the rate of exhaustion in Tet-Low cells (Fig. 2B and C). By days 10 and 14, we observed that Tet-High TILs were significantly more exhausted than Tet-Low TILs, although neither population prevented tumor growth. Also, at days 10 and 14, significantly fewer Tet-Low cells expressed both PD-1 and TIM3 compared with Tet-High cells (Fig. 2D); however, we observed no differences between the groups with PD-1⁺TIM3⁻ expression (Fig. 2E). Finally, there were more Tet-Low TILs without PD-1 and TIM3 expression than Tet-High TILs (Fig. 2F). The rate of activation and exhaustion, as determined by PD-1 and TIM3 expression, demonstrates the influence TCR binding strength has on polyclonal CD8⁺ TILs, and supports the notion that high-affinity interactions increase the rate of T-cell fate and differentiation.

Given the differential rate that Tet-Low TILs reached terminal exhaustion compared with Tet-High TILs, we sought to determine what transcriptional differences persisted between the two groups at day 14. This timepoint is more physiologically relevant than the earlier time points because the levels of Tet-High and Tet-Low T cells have equilibrated, and this later time allows for the migration and development of dysfunctional CD8⁺ TILs (Fig. 2D). We pooled multiple tumors for FACS to separate Tet-High, Tet-Low, and Tet-negative CD8⁺ TILs for scRNA-seq (Fig. 3A). We detected TCR binding strength by employing MHC-tetramers tagged with both fluorescent molecules for FACS and an oligo-barcode for sequencing. Like the fluorescent tetramer, this reagent enabled determination of relative TCR binding strength (Fig. 3B). We detected selected surface protein expression with the use of oligo-barcode–conjugated antibodies. We integrated the CD8⁺ TIL scRNA-seq transcriptome data using 5002 of the most variant genes within our samples. Further analysis identified 15 clusters of transcriptionally distinct cells and projected the cells onto a UMAP (Fig. 3C). The tetramer barcode was not used in generating the UMAP. The immunophenotype of each cluster, as determined by comparing gene expression to current literature, is listed (Supplementary Table S4). Tetramer-barcode counts were visualized onto the UMAP to indicate the amount of tetramer bound to each cell (Fig. 3D). We also visualized the location of the Tet-High, Tet-Low, and Tet-negative scRNA-seq libraries constructed from the FACS-isolated TILs to identify the relative tetramer binding in the UMAP clusters (Fig. 3E). The results clearly showed a bias towards antigen-specific cells (enriched on the right-hand side of the UMAP). The segregation of tetramer-positive cells from Tet-negative cells indicated that antigen specificity for the TAA has an influence on the transcriptional profiles of CD8⁺ TILs, even at this late time point. In this dataset, most tumor-reactive cells recognized gp70_423–431, given the segregation of tetramer-binding clusters (clusters 0, 2, 5, 6, 8, 9, 12) from Tet-negative clusters (clusters 1, 3, 4, 7, 10, 11, 13), and the limited presence of Tet-negative cells within these tetramer-binding clusters. However, a few CD8⁺ TILs did not recognize the tetramer but were within specific tetramer-positive clusters (clusters 2 and 6), suggesting they may recognize another antigen within the TME (44). The antigen-specific clusters had a mix of Tet-High and Tet-Low cells, both by cell counts (Fig. 3F) and by the frequency of Tet-High or Tet-Low cells that fell within each cluster (Fig. 3G). However, there was significant enrichment of Tet-Low cells within clusters 6 and 12, which occupied the UMAP space closest to the Tet-negative clusters. The enrichment of Tet-Low CD8⁺ TILs within clusters 6 and 12 and enrichment of Tet-High cells within dividing cluster 2, 5, and 8 demonstrate that TCR binding strength impacts the transcriptional profiles of CD8⁺ TILs.

We summarized the UMAP clusters in Supplementary Table S4 using the complete list of differentially expressed genes between each cluster (Supplementary Table S5) and between only tetramer-positive or tetramer-negative clusters (Supplementary Table S6), the transcriptional profiles from the literature, and Fig. 4. In addition, we examined scRNA-seq data from CD8⁺ TILs from the B16 tumor model (41) to further validate the labels we assigned to the CD8⁺ TIL clusters in the CT26 tumor model (Supplementary Fig. S2). We used blended UMAPs to examine the broad framework of differential transcriptional phenotypes of T cells within the clusters (Fig. 4A–D). CD44 and CD62 L protein levels (Fig. 4A) and the corresponding RNA expression (Cd44 and Sell, Fig. 4B) showed most gp70_423–431-specific TILs were antigen-experienced (CD44⁺), and the nonspecific TILs had a naïve-like phenotype (CD62L⁺). The naïve like T cells and the antigen-specific T cells in clusters 6 and 12 expressed Tcf7 (Fig. 4C), an essential transcription factor for maintaining a progenitor T-cell phenotype. The remaining Tet⁺ T cells were high in Tox expression, a transcription factor that regulates terminal exhaustion in T cells. The comparison of Mki67, a marker of proliferation, and Btg2, an anti-proliferation factor, showed that cluster 5 and the cells located at the bottoms of clusters 8 and 12 were dividing cells (Fig. 4D). To further identify the phenotypes of the clusters, we assessed the most differentially expressed genes within each cluster (Fig. 4E) and additionally examined expression of other genes in T cell–related gene signatures, such as other transcription factors, effector molecules, and exhaustion markers (Fig. 4F–H). These expression profiles showed a spectrum of T-cell exhaustion that was highest in cells in the upper right corner of the UMAP space and an increase in progenitor characteristics (left side of the UMAP; Fig. 4F–H). Cells occupying the center of the UMAP, in between the antigen-specific and tetramer-negative clusters, were activated, but more progenitor-like relative to the cells on the far right of the UMAP.

Cells within clusters 6 and 12, which were enriched for Tet-Low TILs, were largely Tcf7⁺, Tox⁻, and Havcr2⁻ (which encodes the protein TIM3), indicating more progenitor-like phenotypes that retained the capacity to further differentiate into terminally exhausted effector cells or persist as progenitors (Fig. 4B and D). The other tetramer-positive clusters highly expressed a variety of inhibitory immune receptors, the transcription factor Tox, and lacked Tcf7 expression (Fig. 4C).

In addition to scRNA-seq, we also labeled samples with a panel of CITE-seq antibodies (Supplementary Table S3). The oligo-barcode tags from these antibodies were used to determine the expression of surface proteins in combination with the RNA transcriptome profile within each individual cell. Clustering was performed based on the detection of surface proteins in our samples using CITE-seq ADT sequencing counts and projected onto a UMAP (Fig. 5A). The geographical relationship of the protein-derived clusters was very similar to the clusters determined by the RNA transcriptome. RNA transcriptome clusters 0, 2, 5, and 8 were largely compressed into CITE-seq ADT cluster A based on surface protein expression that included high expression of inhibitory receptors (Fig. 5B and C). Progenitor-like RNA transcriptome clusters 6 and 12 also clustered together into CITE-seq ADT cluster C based on protein expression that included low expression of inhibitory receptors (Fig. 5B and C). The progenitor exhausted RNA transcriptome cluster 9 was uniquely clustered in CITE-seq ADT cluster H, which had a unique mix of inhibitory receptors (Fig. 5B and C). We visualized the CITE-seq ADT-generated clusters onto the scRNA-seq UMAP (Fig. 5B) and the RNA clusters onto the CITE-seq ADT UMAP (Fig. 5C) to further demonstrate the relationship between the transcriptome-based clusters and the protein expression. We reached the same conclusions regarding the cluster profile of cells within our samples based on the unbiased RNA transcriptome and protein expression as projected on the UMAPs, and these independent and unbiased UMAP analyses validated the clustering of one another.

We performed a Pearson correlation between each of the CITE-seq markers in tetramer-positive cells to investigate relationships between proteins and displayed these r values as a heat map (Fig. 5D). This analysis showed a clear relationship between expression of the inhibitory receptors PD-1, TIM3, and LAG3. Critically, it also showed a positive relationship between TIM3 expression and tetramer-barcode counts, further supporting the relationship between TCR affinity for an antigen and the rate of T-cell exhaustion within the TME.

We next identified proteins that significantly differed in the Tet-High and Tet-Low scRNA-seq clusters and visualized the proteins PD-1, TIM3, LAG-3, and CD73 as violin plots for each cluster (Fig. 5E–H). We found a significant increase in expression of TIM3 on Tet-High cells within the more progenitor-like clusters 6 and 12, implicating a divergence in these clusters towards T-cell exhaustion in the Tet-High cells. In addition, progenitor-exhausted cluster 9 had significantly reduced expression of TIM3 on Tet-Low cells, further implicating the divergence of Tet-High cells towards exhaustion, even in these transitory clusters. Although not significant, we observed a trend of increased CD73 expression in Tet-Low cells within the more progenitor-like tetramer-positive clusters relative to CD73 expression in the Tet-High cells. CD73, an ectonucleotidase shown to reduce antitumoral activity in CD8 T cells (45), may represent a target that could be leveraged to improve Tet-Low cells that have retained more progenitor-like characteristics.

Given the enrichment of Tet-Low TILs within transcriptome clusters 6 and 12–ADT cluster C, we performed differential gene expression analysis between the Tet-High and Tet-Low cells. A volcano plot highlighting the most differentially regulated genes within clusters 6 and 12 demonstrates a divergence between the Tet-High and Tet-Low cells in progenitor-like cells from the same cluster (Fig. 6A). We determined that many of the most differentially expressed genes within clusters 6 and 12 associated with exhaustion marker expression, such as Tox, Pdcd1, Lag3, and Havcr2. Differential expression of exhaustion genes implicates TCR binding affinity in this process. Transient pre-exhaustion clusters 6 and 12 retained fewer Tet-High cells, supporting the hypothesis that Tet-High cells proceeded through this phenotypic stage more rapidly. We further compared the differences between Tet-High and Tet-Low cells in terminally exhausted cluster 0 and plotted the results on a volcano plot (Fig. 6B). Tet-High cells were significantly increased in expression of effector function molecules, such as Ifng, Gzmb, Prf1, and Ccl4, with no differences in common exhaustion genes. Thus, CD8⁺ TILs that have reached terminal exhaustion retained some effector functions, and Tet-High TILs retained more of these functions. However, there were fewer differences between Tet-High and Tet-Low cells within cluster 0, further suggesting that cluster 0 was a terminal phase of T-cell differentiation and that both Tet-High and Tet-Low cells reached this stage. Tet-Low cells in cluster 0 had higher expression of quiescent markers, Btg1 and Btg2, which may indicate that they were more recently exhausted, or a mechanism by which they were slower to become exhausted. Finally, we investigated gene expression differences between Tet-High and Tet-Low cells within tetramer-positive clusters pooled together (Fig. 6C). Except for Havcr2 expression, there were no significant differences in expression of common exhaustion markers in Tet-High and Tet-Low cells amongst all antigen-specific clusters. However, we did observe significant enrichment of multiple effector molecules in Tet-High cells and enrichment of quiescent markers in Tet-Low cells. Full gene lists for all intra-cluster differences within each antigen-specific cluster and clusters 6 and 12 pooled together is in Supplementary Table S7.

Tet-High and Tet-Low cells from scRNA-seq clusters were individually examined for differential gene expression. Results were plotted as violin plots for quiescent genes (Fig. 6D) and effector molecules (Fig. 6E). Exhaustion marker expression was lower in clusters 6, 9, and 12 (Fig. 4H), and within cluster 6 Tet-Low cells had less exhaustion marker expression than Tet-High cells (Fig. 6F). The loss of quiescence genes as Tet-High CD8⁺ TILs progressed towards exhaustion (cluster 0) demonstrates another critical difference between Tet-High and Tet-Low cells, which retained more genes that regulate quiescence.

We observed some overarching differences between transcripts in Tet-High and Tet-Low cells (Fig. 6G). Specifically, enrichment of quiescent markers in the Tet-Low cells, such as Btg1, Btg2, Slfn2, Zfp36I2, and the proliferation inhibitor Pdcd4, which suggests a mechanism by which Tet-Low CD8⁺ TILs were maintaining a more progenitor-like phenotype. The Tet-Low cells had significantly higher expression of the early activation marker Cd69. The Tet-Low cells had significantly reduced expression of effector molecules such as Ifng, Gzmb, Prf1, Ccl4, and Ccl3. Critically, the total Tet-High and Tet-Low cells were not a significantly different in Pdcd1, Tox, Tcf7, or Lag3 expression, although there was still a significant difference in Entpd1 and Havcr2 expression. These results showed that both Tet-High and Tet-Low cells proceeded to terminal differentiation, although the number of Tet-Low cells that occupied progenitor-like phenotypes was increased.

To determine if TILs with low TCR binding strength to tumor antigen retain a more progenitor-like phenotype, we developed a transgenic mouse with the previously well-characterized TCR, “1D4” (29–31). The genetic background of this mouse was Rag- and gp70-deficient (28). We compared TILs from the 1D4 mice to two immunocompetent strains of mice: gp70-deficient and wild-type (WT) mice (BALB/c, gp70-sufficient). As determined by tumor weight and total viable cell counts after 14 days of tumor growth, the 1D4 transgenics had significantly larger tumors than WT (BALB/c) mice, and the WT mice had significantly larger tumors than gp70-deficient mice (ref. 28; Fig. 7A and B). Gp70-deficient mice had a T-cell repertoire with higher affinity to gp70_423–431 than WT mice, as determined by tetramer binding (28). As a percentage of total viable cells within the tumor, gp70-deficient and the 1D4 mice had similar numbers of CD8⁺ TILs, and gp70-deficient mice had significantly more than the WT (Fig. 7C). CD8⁺ TILs were further characterized in these mice by their expression of PD-1 and TIM3. In the WT and gp70-deficient mice, PD-1⁺ TIM3⁻ cells made up a small population of total CD8⁺ TILs; however, TILs from the 1D4 mice were predominantly PD-1⁺ TIM3⁻ (Fig. 7D). Furthermore, TILs from the 1D4 mice were predominantly progenitor exhausted cells (PD-1⁺ TIM3⁻ SLAMF6⁺), whereas in the WT and gp70-deficient mice this population was significantly smaller (Fig. 7E). Progenitor exhausted cells are a distinct transitory state, capable of differentiating into terminally exhausted cells (17). Enrichment of 1D4 T cells in this transitory state suggested that low-affinity T cells may require further stimulation to fight the tumor and to proceed to terminal exhaustion. Comparing clusters 6 and 12 enriched for TILs with low tetramer binding suggested that low-affinity TILs may represent a targetable population of progenitor cells for induction of tumor killing activity.

In this study, we determined how the polyclonal, endogenous CD8⁺ TIL response specific for a single TAA was affected by TCR binding strength. Using a barcoded MHC-tetramer, we successfully identified antigen-specific cells by CITE-seq and further distinguished Tet-High from Tet-Low TIL with scRNA-seq. The likelihood that the binding of MHC tetramers differentially influenced transcription in Tet-High and Tet-Low cells was minimal. In the sequencing experiments, Tet-High and Tet-Low TILs were harvested from the same mice and were treated identically, including exposure to the MHC-tetramer molecules. Each sample was harvested from the H-2L^d gp70_423–431 antigen-rich environment and stained with tetramer for one hour before being placed on ice, reducing the risk of activation from tetramer staining. On average and within cluster 6, a Tet-Low enriched cluster, Cd69 expression, a marker of early TCR activation, was more highly expressed in Tet-Low TILs than Tet-High TILs, further suggesting that tetramer staining was not disproportionately affecting Tet-High TIL activation. Finally, tetramer staining was unlikely to promote T-cell exhaustion in the timeframe of the staining process, which was the predominant phenotype observed in TILs from this model, and one of the most notable differences between Tet-High and Tet-Low samples.

Using tetramer staining to examine CD8⁺ TILs, we determined that the rate of T-cell exhaustion occurred more rapidly in Tet-High TILs, and by late timepoints, Tet-High and Tet-Low TIL occupied similar transcriptional clusters. We observed an enrichment of progenitor-like phenotypes in Tet-Low TILs as the tumor became established, supporting the hypothesis that Tet-Low T cells persisted in less differentiated transcriptional profiles. Progenitor-like clusters 6 and 12 from the scRNA-seq data were significantly enriched for Tet-Low T cells. Together they represented a functionally similar subset of CD8⁺ TILs, as they clustered together based on CITE-seq ADT markers. Critically, these clusters expressed less Tox, Pdcd1, Havcr2 (TIM3), Lag3, and Ifng but had increased expression of Tcf7, Il2, IL7r, and Klf2, consistent with a progenitor-like phenotype. The monoclonal T cells analyzed by the Schietinger group show a similar transcriptional profile as the polyclonal TIL response shown here; specifically, the low-affinity TCR interactions express more Tcf7 and Il7r, whereas high-affinity clones express more Entpd1 and Tox (9). However, we also detected some differences between Tet-High and Tet-Low populations, which were not associated with TCR binding strength by the Schietinger group, such as Havcr2 expression. These discrepancies are likely due to the differences in scRNA-seq vs. bulk sequencing analysis that may mask differences in Tet-High and Tet-Low cells.

TCR engagement is the most significant contributor to the transcriptional profile of TILs as was established by the differences between tetramer-negative and -positive cells, which divided the UMAP space. A sub-dominant TAA from the CT26 tumor, gp70_37–44, has been identified from the same endogenous retrovirus as gp70_423–431 (44), and CT26 cells have numerous point mutations in expressed genes (46). Thus, there is a possibility for T-cell recognition of these peptides, explaining the presence of some tetramer-negative cells within predominantly tetramer-positive clusters. The distribution of Tet-High and Tet-Low cells within the UMAP may suggest that, although TCR affinity is critical at earlier stages of T-cell activation (3, 47), by later time points transcriptional differences between Tet-High and Tet-Low populations are less significant. Specifically, T cells reach similar transcriptional profiles, even if at different rates. Furthermore, Tet-High cells were enriched in the dividing clusters and found to have increased expression of multiple effector molecules such as Ifng, Gzmb, Prf1, and Ccl4, whereas Tet-Low cells had increased expression of multiple quiescent genes such as Btg1, Btg2, and Slfn2, even in the more exhausted cluster 0, suggesting that in terminally exhausted clusters, Tet-Low cells retained a more quiescent transcriptional profile. One explanation for the enrichment of Tet-Low cells in progenitor-like clusters is that new infiltrates into the TME may be predominantly Tet-Low cells replenishing the progenitor-like phenotypes. In contrast, Tet-High cells may proliferate more rapidly once they encounter antigen and progress out of progenitor states, while increasing their contribution to the total percentage of TILs. This logic also predicts why we observed more Tet-Low cells relative to Tet-High cells at early timepoints in TME development. Another explanation for the difference in the rate of T-cell exhaustion between Tet-High and Tet-Low cells is that Tet-High cells may be starting these transcriptional programs more rapidly than Tet-Low TILs. This possibility was supported by an increase in the transcriptional differences between Tet-High and Tet-Low cells within progenitor-like clusters relative to exhausted cells in cluster 0 and all other tetramer-positive clusters. The transcriptional differences within the progenitor clusters were largely T-cell exhaustion—related genes and suggest that progenitor clusters may be the clusters in which Tet-High and Tet-Low cells deviated in their progression towards T-cell exhaustion. The transcriptional differences found between Tet-High and Tet-Low cells in the exhausted cluster 0 included quiescent genes not found to be different in progenitor clusters 6 and 12. In terminal phases of T-cell exhaustion, Tet-Low cells retained higher expression of key quiescent markers relative to their Tet-High counterparts, suggesting Tet-Low cells retained their progenitor characteristics through higher expression of quiescent genes.

Using a transgenic mouse with a low-affinity TCR, 1D4, low-affinity TILs were predominantly in a progenitor-exhausted state (PD-1⁺TIM3⁻SLAMF6⁺), which in WT mice, only made up a small proportion of total endogenous CD8⁺ TILs. The increase in progenitor-like cells in the low-affinity 1D4 TILs confirms the results others have identified in clonal low-affinity TCR interactions (9) and the results from our polyclonal CD8⁺ TIL populations (clusters 6 and 12). WT mice were better than 1D4 transgenic mice at controlling tumor growth, demonstrating the impaired efficacy of 1D4 low-affinity progenitor CD8⁺ TILs. Gp70-deficient mice were the most efficient at controlling tumor growth, although TILs from these mice had significantly reduced levels of progenitor exhausted cells relative to 1D4 mice. TILs from the gp70-deficient mice have higher affinity for gp70_423–431 (28). The increase in progenitor exhausted cells from 1D4 TILs relative to WT or gp70-deficient TILs further supported the hypothesis that Tet-Low cells are slower or less frequently differentiate into terminal phases of T-cell exhaustion and that their weak TCR interaction also results in insufficient activation to induce potent tumoricidal responses. Conversion of Tet-Low T cells into antitumor TILs may require additional stimulation not received from the native antigen. Improved stimulation of low-affinity interactions has been shown in the CT26 model using peptide-mimotope vaccines (24, 29, 48). Progenitor-like Tet-Low cells, enriched in clusters 6 and 12, have not yet reached terminal differentiation as have cells in cluster 0, which may make them more responsive to therapeutic interventions; however, isolation of any single cluster of TILs is likely insufficient to control tumor growth directly. Thus, the persistence of Tet-Low cells in progenitor-like phenotypes may be targetable to improve TIL cytotoxicity. However, Tet-High populations, which have increased effector functions but decreased progenitor characteristics, may require alternate interventions.

B.P. O'Connor reports grants from NCI during the conduct of the study; other support from Seagen outside the submitted work. J.E. Slansky reports grants from Cancer League of Colorado, NIH/NCI; and grants from NIH/NAIAD during the conduct of the study. No disclosures were reported by the other authors.

Z.L.Z. Hay: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, project administration, writing–review and editing. J.R. Knapp: Data curation, formal analysis, investigation, visualization, writing–review and editing. R.E. Magallon: Investigation. B.P. O'Connor: Conceptualization, data curation, formal analysis, project administration, writing–review and editing. J.E. Slansky: Conceptualization, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

R01 CA226879 (to J.E. Slansky), T32 AI007405 (to Z.L.Z. Hay), and University of Colorado Cancer Center shared resources (P30CA046934).

We thank R01 CA226879 (J.E. Slansky), and T32 AI007405 (Z.L.Z. Hay) for funding support and the Cancer League of Colorado for funding the 1D4 mouse. We thank the NIH Tetramer Core Facility (contract number 75N93020D00005) for providing the H-2L^d tetramers and monomers, University of Colorado Cancer Center Genomics Shared Resource (P30CA046934, RRID:SCR_021984), and CU|AMC ImmunoMicro Flow Cytometry Shared Resource (RRID:SCR_021321). At National Jewish Health, we thank the Genomics Facility, the Mouse Genetics Core Facility, and the Flow Cytometry Facility for their contributions. We stand together (49).

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.