T-Cell Exhaustion Signatures Vary with Tumor Type and Are Severe in Glioblastoma

Glioblastoma (GBM) is the most common and feared malignant primary brain tumor, persisting as one of few cancers that remain universally lethal. Despite improvements in standard of care, median survival remains at just 15 to 17 months (1). Immunotherapies continue to yield promise, but the efficacy needed to garner an FDA approval has remained wanting. Success has been limited by marked tumor heterogeneity (2), poor immune infiltration (3), and the tumor's notably potent capacities for subverting local and systemic immune responses. Regarding the latter, it is the T cells required for effective antitumor responses that are particularly victimized (4–6). Their plight is reflected in patient lymphopenia (7), as well as various characterized forms of classical T-cell dysfunction, such as anergy (8, 9) and regulatory T-cell–imposed tolerance (10–13). Countering such dysfunction has not yet reliably yielded sustained and effective T-cell activity within the tumor, however, raising concerns for additional unexplored tumor-imposed states of T-cell dysfunction, such as exhaustion. Despite sporadic studies suggesting T-cell phenotypes that might signal exhaustion, few formal characterizations of T-cell exhaustion in patients or mice with GBM have been undertaken.

T-cell exhaustion is a hyporesponsive state resulting from repeated or prolonged antigenic exposure under suboptimal conditions (14). Initially discovered in the context of chronic lymphocytic choriomeningitis virus (LCMV) infection (14, 15), it is now known to contribute to disrupted T-cell function in cancer as well (16). Exhaustion represents a specific transcriptional program in T cells that is often characterized phenotypically by the increased expression of multiple coinhibitory receptors, many of which constitute classical or alternative immune checkpoints (17). Known receptors include PD-1, CTLA-4, TIM-3, LAG-3, BTLA, 2B4, CD160, CD39, VISTA, and TIGIT. Antagonizing or blocking PD-1 and CTLA-4 (the classical immune checkpoints) are well-recognized FDA-approved anticancer strategies aimed at improving T-cell function in multiple malignancies, including melanoma and non–small cell lung carcinoma. To date, such strategies have shown only limited efficacy in GBM, creating the need to better understand treatment failures. One mechanism conferring resistance to classical checkpoint blockade has been the upregulation of alternative immune checkpoints such as TIM-3 and LAG-3 (18), signifying that T-cell exhaustion might play a substantial role in restraining therapeutic enhancements to T-cell function in GBM. To date, few thorough investigations of classical and alternative immune checkpoint coexpression within GBM have been conducted. Likewise, the prominence of exhaustion among tumor-infiltrating lymphocytes (TILs) and its contribution to local T-cell dysfunction in GBM are poorly described.

We report here that a substantial proportion of TILs derived from human GBM express multiple coinhibitory immune checkpoints. These findings are recapitulated in two different murine models of glioma, where the mounting expression of PD-1, Tim-3, and Lag-3 correspond to diminished TIL responsiveness. TILs from glioma demonstrate significant dysfunction, and their transcriptional signatures and epigenetic profiles match that of traditional bona fide T-cell exhaustion, as described in viral models. Interestingly, in our murine tumor models, patterns of TIL exhaustion prove to be tumor type-specific, with gliomas, lung carcinoma, breast carcinoma, and melanoma all eliciting characteristic, yet distinct exhaustion signatures that do not vary when tumor site is modified, including when each tumor is placed intracranially. Likewise, independent of location, the exhaustion signature and corresponding TIL dysfunction appear to be particularly severe among T cells infiltrating gliomas, highlighting a significant contribution for exhaustion to T-cell dysfunction within these tumors.

All studies were conducted with approval from the Massachusetts General Hospital Cancer Center Institutional Review Board or the Duke Cancer Center Institutional Review Board. All studies were conducted in accordance with recognized ethical guidelines (U.S. Common Rule, 45 CFR 46, 21 CFR 50, 21 CFR 56, 21 CFR 312, 21 CFR 812, and 45 CFR 164.508-514) 21 treatment-naïve GBM patients undergoing primary surgical resection of intracranial (i.c.) GBM were included in the prospective collection of whole blood and tumor tissue. Fifteen healthy age-matched controls were included in the prospective collection of whole blood. Informed consent was obtained from all subjects. Blood specimens were collected into EDTA-containing tubes. All blood and tumor specimens were stored at room temperature and processed within 12 hours. All samples were labeled directly with antibodies for use in flow cytometry, and red blood cells subsequently lysed using eBioscience RBC lysis buffer (eBioscience). Cells were washed, fixed, and analyzed on an LSRII FORTESSA flow cytometer (BD Bioscience).

Female C57BL/6 and VM-Dk mice were used at 6 to 12 weeks of age. C57BL/6 mice were purchased from Charles River Laboratories. VM-Dk mice were bred and maintained as a colony at Duke University. Animals were maintained under specific pathogen-free conditions at Cancer Center Isolation Facility of Duke University Medical Center. The Institutional Animal Care and Use Committee approved all experimental procedures.

Cell lines studied included SMA-560 malignant glioma, CT2A malignant glioma, E0771 breast medullary adenocarcinoma, B16F10 melanoma, and Lewis Lung Carcinoma (LLC). SMA-560 cells are syngeneic on the VM-Dk background, whereas all others are syngeneic in C57BL/6 mice. The SMA-560 cell line is derived from a spontaneous malignant glioma that originally arose on the VM-Dk background. Tumors have low S-100 expression and high glial fibrillary acid protein expression, and are most representative of anaplastic astrocytoma (19). The CT2A cell line is derived from a chemically induced tumor with 20-methylcholanthrene on the C57BL/6 background, and accurately reflects several characteristics of GBM including intratumoral heterogeneity, as well as radio- and chemoresistance (20). SMA-560, CT2A, B16F10, and LLC cells were grown in vitro in DMEM with 2 mmol/L l-glutamine and 4.5 mg/mL glucose (Gibco) containing 10% FBS (Gemini Bio-Products). E0771 cells were grown in vitro in RPMI 1640 (Gibco) containing 10% FBS plus 1% HEPES (Gibco). Cells were harvested in the logarithmic growth phase. For i.c. implantation, tumor cells in PBS were then mixed 1:1 with 3% methylcellulose and loaded into a 250 μL syringe (Hamilton). The needle was positioned 2 mm to the right of the bregma and 4 mm below the surface of the skull at the coronal suture using a stereotactic frame. Note that 1 × 10⁴ SMA-560, CT2A, E0771, and LLC cells or 500 B16F10 cells were delivered in a total volume of 5 μL per mouse. For s.c. implantation, 5 × 10⁵ SMA-560, CT2A, E0771, and LLC cells or 2.5 × 10⁵ B16F10 cells were delivered in a total volume of 200 μL per mouse into the s.c. tissues of the left flank.

Tumors were minced, incubated in 100 units/mL collagenase IV (Sigma-Aldrich) and 0.1 mg/mL DNase I (Roche Diagnostics) in RPMI supplemented with 10% FBS for 20 to 30 minutes in a Stomacher machine set at normal speed, and washed through 70-μm nylon cell strainers (Falcon; BD Biosciences) in PBS with 2% FBS. Cells were immediately stained and subsequently analyzed by flow cytometry.

Human-specific antibodies were purchased from BD Biosciences (CD3: SP34-2; CD4: RPA-T4) or BioLegend (CD8: HIT8a; PD1: EH12.2H7; TIM-3: F38-2E2; LAG3-: 11C3C65; CTLA-4: BNI3; TIGIT: A15153G; CD39: A1; IL2: MQ1-17H12; TNFα: MAb11; IFNγ: 4S.B3. Mouse-specific antibodies were purchased from BD Biosciences (CD3: 145-2C11; CD4: RM4-5; CD8: 53-6.7; PD1: J43; Lag-3: C9B7W; TIGIT: 1G9; CD244.2: 2B4; CD160: CNX46-3; CTLA4: UC10-4F10) or Biolegend (IFNγ: XMG1.2; IL2: JES6-5H4; TNFα: MP6-XT22; Tim-3: RMT3-23; CD39: Duha59; BTLA: 8F4). Appropriate isotype controls were used when applicable. The LIVE/DEAD Fixable Yellow Dead Cell Stain Kit (Thermo Fisher) was used to exclude dead cells. Intracellular staining was performed using the eBioscience Fixation and Permeabilization Buffer kit. For intracellular cytokine staining, cells were stimulated with phorbol 12-myristate 13-acetate (PMA, 50 nmol/L) and ionomycin (500 nmol/L) for 4 to 6 hours at 37°C, 5% CO₂, in the presence of 1 μg/mL brefeldin A (BD biosciences). Surface staining was performed, and cells were fixed and permeabilized with the BD Cytofix/Cytoperm Kit and stained for IFNγ, TNFα, and IL2. Tetramers for mODC1 were generated and used as described previously (21).

Five patient TILs and peripheral blood mononuclear cells (PBMCs) and five control PBMCs were isolated as described above. The repertoire was assessed using the IOTest Beta Mark TCR V beta Repertoire Kit (Cat: IM3497 Beckman Coulter) according to the manufacturer's instructions.

When moribund, tumor-bearing animals were sacrificed and T cells were isolated and sorted from tumors and spleens based on the following markers: Live, CD45⁺, CD3⁺, CD8⁺. CD8⁺ T cells isolated from tumor-draining (TD) cervical lymph nodes (LNs) were additionally stained and sorted for CD44⁺. RNA was extracted using the Qiagen RNAeasy Mini Kit (Cat No: 74104) and DNase treated using Qiagen RNase-Free DNase Set (Cat No: 79254). RNA concentration and purity were determined with NanoDrop and RNA integrity with Agilent Bioanalyzer. cDNA was synthesized and hybridized to the Affymetrix Clariom S Mouse Array.

Genomic DNA was isolated from sorted CD8⁺ T cells (as described previously) from tumors, spleens, or TD cervical LNs using the Quick-DNA Universal Kit (Cat No: D4068; Zymo Research) according to the manufacturer's instructions, and <1 μg of DNA was subjected to bisulfite conversion using the EZ DNA Methylation Kit (Cat No: D5001; Zymo Research). The bisulfite-converted DNA was then amplified for pyrosequencing analysis using the PyroMark PCR Kit (Qiagen; Cat No: 978703). Primers were selected for the CR-B site covering four CpG sites at the PD-1 promoter region as previously published (22). Pyrosequencing was performed on amplified bisulfite-converted DNA using PyroMark Gold Q96 Reagent (Qiagen; Cat No: 972807).

For human studies, the sample size of 21 patients and 10 controls was chosen so that a two-tailed t test comparing groups has 90% power to detect a difference that is 1.1 times the standard deviation of the outcome variable in each group. For animal studies, sample sizes were chosen based on historic experience and were variable based on numbers of surviving mice available at experimental timepoints or technical limitations. For statistical comparisons, unpaired t tests were generally used to compare groups. Analyses were not adjusted for multiple testing.

The checkpoint molecules PD-1, CTLA-4, TIM-3, LAG-3, CD160, 2B4, TIGIT, CD39, and BTLA have been identified as coinhibitory regulators of T-cell effector function during chronic viral infection and within tumors, limiting the scope and duration of T-cell responses (17, 23). Furthermore, their expression on T cells can frequently signal an exhausted state. To begin examining the contribution of exhaustion to T-cell dysfunction in GBM, we used flow cytometry to phenotypically analyze TILs and PBMCs isolated from treatment-naïve patients with GBM enrolled in a specimen collection protocol. PBMCs from healthy controls were isolated for comparison. Twenty-one patients were enrolled in total (patient and control characteristics listed in Supplementary Table S1).

We found that PD-1, LAG-3, TIGIT, and CD39, in particular, were highly expressed on patient CD8⁺ TILs (Fig. 1A and B; gating strategy depicted in Supplementary Fig. S1A). Notably, PD-1 was especially prominent, being found on up to 96% of tumor-infiltrating CD8⁺ T cells, with most of these in turn being PD-1^hi. Although the percentage of TIM-3⁺ CD8⁺ T cells was not statistically significantly greater among TILs than patient or control PBMCs (P = 0.059), TIM-3 was present on >20% of TILs in several patients, and the median fluorescence intensity (MFI) of TIM-3 was significantly greater on TILs than in control blood (Supplementary Fig. S1B). The majority of immune checkpoints studied were present to a greater extent among TILs than among either patient or control PBMCs. An exception was BTLA, which was lower among TILs, consistent with studies showing that BTLA is more highly expressed on naïve T cells and becomes downregulated following antigenic exposure (24). Regarding peripheral blood, patient and control PBMCs did not differ substantially from each other with respect to checkpoint expression.

There was no difference in the percentage of CD8⁺ T cells expressing 2B4, CD160, or CTLA-4 across tissue types or between patients and controls. MFI data for all checkpoint molecules are available in Supplementary Fig. S1B. Upon further characterization of T cells infiltrating human GBM, we found that tumor-infiltrating CD8⁺ T cells demonstrated a predominantly effector memory T-cell phenotype (CD45RA⁻CD62L⁻) as opposed to naïve (CD45RA⁺CD62L⁺; Supplementary Fig. S2), reflecting prior antigenic exposure.

Although high levels of PD-1 expression on effector memory T cells can signal either activation or exhaustion, the mounting expression of additional alternative immune checkpoints frequently reflects the hierarchical loss of effector function, favoring the presence of T-cell exhaustion (17). Therefore, we evaluated the degree of coexpression between PD-1 and the common alternative immune checkpoints TIM-3 and LAG-3 on CD8⁺ T cells from both tumors and blood. Boolean gating was employed to analyze the following T-cell populations: CD8⁺ T cells that were positive for only one marker (e.g., PD-1⁺TIM-3⁻LAG-3⁻), positive for two markers (e.g., PD-1⁺TIM-3⁺LAG-3⁻), or triply positive (PD-1⁺TIM-3⁺LAG-3⁺). Higher percentages of CD8⁺ TILs in human GBM expressed either PD-1 alone, PD-1 in combination with TIM-3 or LAG-3, or all three markers, in comparison with CD8⁺ T cells isolated from either patient or control blood (Fig. 1C). Single expression of TIM-3 or LAG-3 was not common in either TILs or blood. These findings reveal that both TIM-3 and LAG-3 expressions are almost universally accompanied by PD-1 expression (and frequently by one another as well), with such coexpression suggesting a dysfunctional state.

We next assessed cytokine production of TILs and PBMCs to determine whether the high immune checkpoint expression in TILs predicted T-cell hyporesponsiveness. Following stimulation with PMA and ionomycin, intracellular IFNγ, IL2, and TNFα were assessed by flow cytometry. There was no difference in cytokine production between patient and control PBMCs, whereas TILs produced significantly less IFNγ, IL2, and TNFα than control PBMCs (Fig. 1D). Within patient samples, cells that were PD-1 negative or PD-1 single positive produced comparable amounts of IFNγ and TNFα, whereas cells that were PD-1 single positive produced more IL2 than PD-1–negative cells (Fig. 1E), highlighting that high levels of PD-1 alone may represent a state of activation rather than exhaustion. Mounting expression of the alterative immune checkpoints TIM-3 and LAG-3 resulted in loss of function: triply positive PD-1⁺TIM-3⁺LAG-3⁺ T cells were unable to produce IFNγ, IL2, or TNFα (Fig. 1E).

To examine the degree to which these findings might be recapitulated in mice, we employed two common immunologically relevant orthotopic models of murine malignant glioma: SMA-560 and CT2A. Beginning with the CT2A model, we found that PD-1 was expressed by up to 95% of CD8⁺ TILs (Fig. 2A), similar to human GBM TILs. Tim-3, Lag-3, CTLA-4, 2B4, CD160, CD39, and TIGIT expressions were also notably prevalent and appeared to be more common than among GBM patient TILs, although no statistical comparisons across species were made.

CT2A TILs isolated from SMA-560 tumors likewise demonstrated increased expression of multiple immune checkpoints (particularly PD-1, Tim-3, Lag-3, and CD39), but consistently demonstrated less prominent upregulation of CTLA-4, 2B4, CD160, and TIGIT than what was found for CT2A (Fig. 2A and B). Both the CT2A and SMA-560 models demonstrated a substantial proportion of TILs that were triply positive for PD-1, Tim-3, and Lag-3: upon Boolean analysis, a mean of 40% and 46% of infiltrating CD8⁺ T cells in CT2A and SMA-560, respectively, expressed all three immune checkpoints (Supplementary Fig. S3). Similarly, in both murine models, we found that PD-1 was expressed on the majority of CD8⁺ TILs, whereas Tim-3 and Lag-3 were coexpressed with PD-1 rather than present on their own, akin to the observation in human GBM specimens.

To determine whether glioma TILs demonstrate decreased functional capacity compared with lymphocytes isolated from other compartments in tumor-bearing and naïve animals, we harvested lymphocytes from spleens, TD cervical LNs, and tumors at day 18 following i.c. implantation with CT2A. Lymphocytes were also harvested from spleens and LNs of age-matched naïve animals. As might be anticipated, CD8⁺ TILs expressed greater numbers and levels of immune checkpoints than CD8⁺ lymphocytes isolated from spleens or TDLNs (Fig. 3A). Notably, neither spleens nor LNs from tumor-bearing animals demonstrated appreciable differences in the percentages of single-positive PD-1, Tim-3, or Lag-3–expressing T cells when compared with the same lymphoid organs from naïve animals, further revealing such expression patterns as concentrated among TILs. Accompanying the phenotypic variations across TILs, spleen, and LNs, however, were significant differences in T-cell function across the compartments. Following stimulation with PMA and ionomycin, intracellular IFNγ, IL2, and TNFα were assessed by flow cytometry. Significantly fewer TILs were positive for IFNγ, IL2, and TNFα than lymphocytes isolated from either spleens or LNs, signaling a hyporesponsive state among TILs (Fig. 3B). The MFIs of IFNγ, IL2, and TNFα were also significantly lower among TILs than among T cells from other lymphoid compartments (Supplementary Fig. S4A). Representative dot plots are shown in Supplementary Fig. S4B. CD8⁺ TIL staining positively for IFNγ, IL2, or TNFα were not uncommonly found to be PD-1 single positive (consistent with PD-1′s ambiguous identity as an activation marker). TILs triply positive for PD-1, Tim-3, and Lag-3 produced significantly less IL2 than PD-1 singly positive TILs, and triply positive TILs trended toward decreased TNFα production, although a significant difference was not seen. T cells producing IFNγ were not found among triply positive TILs, indicating that mounting expression of alternative immune checkpoints corresponded to worsening function, particularly with regard to IFNγ and IL2 production (Fig. 3C).

To further assess glioma TILs for features of bona fide T-cell exhaustion, as described in models of chronic viral infection, we compared the gene-expression profiles of CD8⁺ CT2A TILs with those of naïve (CD44⁻) CD8⁺ T cells isolated from spleens, or to activated (CD44⁺) CD8⁺ T cells isolated from TDLNs using the murine Clariom S microarray transcriptome-level platform (Affymetrix). The gene-expression profile of TILs differed significantly from the profile of either naïve or effector T cells (Fig. 4A). To determine whether this molecular signature was consistent with T-cell exhaustion, we used gene set enrichment analysis to compare the transcriptional profile of CT2A TILs with previously identified gene sets from virus-specific exhausted CD8⁺ T cells isolated from chronic LCMV clone 13 infections (25). We found enrichment for exhaustion genes and pathways in CT2A TILs that statistically matched patterns observed in virally induced exhaustion (Fig. 4B).

T-cell exhaustion during chronic viral infection is associated with the loss of Tbet^hiPD-1^int T cells, and the accumulation of terminally differentiated Eomes^hiPD-1^hi exhausted T cells (26). Likewise, our data show that GBM TILs demonstrate elevated Eomes and decreased Tbet levels (Fig. 4C). Blimp-1, a transcription factor associated with T-cell exhaustion in chronic viral infections (27), is also elevated among CT2A TILs. As additional confirmation, we examined the methylation status of the PD-1 (pdcd1) gene locus. Chronic viral infection has been shown to enforce demethylation of the PD-1 locus in antigen-specific T cells (28). We found that the PD-1 locus is significantly demethylated in glioma TILs compared with naïve and effector T cells, consistent with expectations for T-cell exhaustion (Fig. 4D).

To determine whether T-cell exhaustion arises moreso amid tumor-specific T cells, we utilized the SMA-560 glioma model. The SMA-560 glioma model expresses the recently characterized ODC1-Q129L mutation, which results in a functionally immunogenic endogenous neoantigen (21). As such, we were able to detect neoantigen-specific T cells within IC SMA-560 gliomas using a newly developed ODC1 tetramer. In so doing, we found that ODC1 tetramer–positive T cells contained higher proportions of triply positive PD-1⁺Tim-3⁺Lag-3⁺ T cells, and were less functional upon stimulation than their tetramer-negative counterparts (Fig. 5A).

We next sought to determine whether T-cell exhaustion likewise preferentially arises in tumor-specific T cells in human GBM. Given the inability to reliably detect and compare neoantigen-specific T cells across heterogeneous patient specimens, we assessed T-cell clonal expansion as a surrogate measure for antigen specificity, with the goal of phenotypically characterizing any expanded clones. We analyzed the T-cell receptor (TCR) Vβ chains of five matched patient TILs and PBMC samples, and five control PBMC samples. Surprisingly, human GBM-infiltrating CD8⁺ T cells exhibited no detectable clonal expansion, precluding phenotypic characterization of expanded antigen-specific clones (Fig. 5B). This lack of clonal expansion suggests either ineffectual tumor antigen presentation or the onset of functional defects amid antigen-specific CD8⁺ T cells immediately upon tumor infiltration. The latter possibility appears more likely, as the limited expansion of CD8⁺ T cells contrasted starkly with ample clonal expansion amid infiltrating CD4⁺ T cells (Fig. 5C). Furthermore, we demonstrated that although CD4⁺ T cells express multiple immune checkpoints (Supplementary Fig. S5A), they remain functional, producing equal amounts of IFNγ as patient and control blood when stimulated (Supplementary Fig. S5B).

In prior experiments (Fig. 2A and B), we revealed that the SMA-560 and CT2A glioma models yield consistent yet characteristically distinct exhaustion signatures among infiltrating T cells. This suggested that varying tumor types might elicit reproducibly distinct patterns of exhaustion. To determine whether the i.c. environment might also contribute to the glioma-induced T-cell exhaustion signatures observed, we implanted CT2A and SMA-560 gliomas both IC and s.c. and analyzed TILs at late stages of tumor growth. The phenotypic T-cell exhaustion signature remained constant for each glioma model, independent of the tumor's location in the brain versus the flank (Fig. 6A).

We then determined whether infiltrating T-cell exhaustion signatures differ substantially across gliomas or other solid tumors commonly metastatic to the brain. To this end, we first validated IC engraftment of melanoma (B16F10), breast (E0771), and lung (LLC) carcinoma cell lines (Supplementary Fig. S6A). Subsequently, we implanted B16F10, E0771, and LLC either IC or SQ into syngeneic C57BL/6 mice, harvested tumors when mice were moribund, and determined the TIL exhaustion patterns. Once again, each tumor yielded a characteristic and distinct TIL exhaustion signature that did not vary with i.c. versus peripheral location (Fig. 6B). Representative staining of PD-1 across tumor models is shown in Supplementary Fig. S6B.

Among tumor types, the SMA-560 and CT2A malignant gliomas each yielded TILs with comparatively severe phenotypic exhaustion patterns (Fig. 6C). This included significantly increased numbers of TILs that were triply PD-1⁺Tim-3⁺Lag-3⁺ (Fig. 6D). Correspondingly, TILs isolated from gliomas exhibited the poorest function, with fewer cells producing IL2 than TILs isolated from nonglioma tumor models (Fig. 6E; MFI data shown in Supplementary Fig. S6C). CT2A gliomas, in particular, harbored the least functional T cells, exhibiting significantly impaired IFNγ and IL2 production compared with TILs from SMA-560 gliomas or other tumor models (Fig. 6E). Likewise, the greater functional impairment corresponded to higher levels of the alternative immune checkpoints TIGIT, CD39, and 2B4 on T cells, compared with those of the SMA-560 glioma model (Fig. 6C). Importantly, the severity of TIL exhaustion patterns did not appear to correspond to levels of T-cell infiltration across tumor types: CT2A and LLC, despite very different exhaustion signatures, both demonstrated low T-cell infiltration (“cold” tumors), whereas E0771, SMA-560, and E0771 all demonstrated greater degrees of infiltration (Supplementary Fig. S7). The degree of T-cell infiltration may not influence T-cell exhaustion profiles, suggesting that the association of T-cell infiltration to treatment response to checkpoint blockade may not be associated with degree of T-cell exhaustion (29). Taken together, these findings suggest that exhaustion patterns among tumor-infiltrating T cells are (1) determined by tumor-intrinsic factors and not environment; (2) strongly influence infiltrating T-cell function; and (3) vary in severity across and reflect tumor type, with gliomas instigating particularly severe exhaustion.

Immune checkpoint blockade has garnered significant interest in recent years following dramatic successes against a variety of solid tumors (30–34) and resultant FDA approval as an immunotherapeutic strategy for multiple malignancies. The efficacy of classical checkpoint blockade (specifically anti–PD-1 and anti–CTLA-4) may be hindered, however, by the emergence of a number of secondary or alternative checkpoints on T cells (18), including TIM-3, LAG-3, BTLA, 2B4, CD160, CD39, and TIGIT (35–37). Mounting expression of these alternative checkpoints on T cells provides additional avenues for T-cell shutdown following activation, but also serves to signal T-cell transition into a dysfunctional, exhausted state (37, 38). Such exhaustion may be reversible in earlier stages, but can rapidly progress to a late phase in which function declines beyond rescue (39, 40).

T-cell dysfunction has long been a hallmark of GBM, as we, and others, have highlighted (4–6). Classically, such dysfunction can be categorized under the labels of senescence, ignorance, anergy, tolerance, and exhaustion (41). Anergy and tolerance are well characterized in GBM, but this study suggests exhaustion as a major contributor to the failure of T cells that otherwise successfully arrive at tumor and initially become activated. Whereas much focus to date has been on activating T cells and ensuring their access to tumors situated within the confines of the brain, we demonstrate that those T cells successfully arriving tumor are still rendered hyporesponsive by the tumor, with exhaustion revealed as a predominant and underappreciated mode of dysfunction. Accordingly, we are among the first to thoroughly characterize immune checkpoint expression on GBM patient CD8⁺ TILs, and to subsequently credential such expression as bona fide exhaustion by performing functional and transcriptome analyses. Our study highlights that phenotypic expression of checkpoints may not accurately reflect an exhausted state: we show that CD4⁺ TILs remain functional despite expression of multiple immune checkpoints, and that CD8⁺ T cells expressing PD-1 alone remain functional. In light of these findings, we demonstrate that T-cell functional status and molecular signatures remain important confirmations of true T-cell exhaustion.

Prevalent among TILs in both patients and mice with GBM are T cells coexpressing PD-1 in conjunction with the alternative inhibitory immune checkpoints, TIM-3 and LAG-3. This coexpression appears to specifically mark TILs as hypofunctional in our study and suggests the importance of TIM-3 and LAG-3 as immune checkpoint targets of import in GBM. Preclinical work has indeed highlighted a possible synergy between PD-1, Tim-3, and Lag-3 blockade in mice with GBM (42). Monoclonal antibodies blocking human TIM-3 and LAG-3 are in early phase clinical trials in GBM and other solid tumors. A phase I clinical trial of anti–LAG-3 alone or in combination with anti–PD-1 is currently recruiting GBM patients (NCT02658981), whereas a phase I clinical trial of anti–TIM-3 alone or in combination with anti–PD-1 is currently recruiting patients with advanced solid tumors (NCT02817633). The results of these trials will help to assess whether combinatorial checkpoint blockade in GBM might increase the likelihood of functional rescue for T cells infiltrating these tumors. Data from these trials will be highly anticipated, given the recent announcements regarding the failure of PD-1 blockade monotherapy in a phase III trial in GBM.

Checkpoint blockade, as with all immunotherapies, encounters multiple challenges in GBM, including considerable intratumoral heterogeneity (2, 43, 44), poor immunogenicity (45, 46), low mutational burden (47), poor T-cell infiltration (48), and substantial tumor-elicited T-cell dysfunction (10). Each of these challenges may need to be addressed before we can anticipate the ingress of prodigious immunotherapeutic success in GBM. GBM has frequently posed more difficulty than other cancers, evident in what remains a 100% mortality rate, despite what is essentially a local disease with only exceedingly rare cases of metastasis. Yet, despite its exclusively local progression, GBM elicits severe immune dysfunction, both locally and systemically. Here, we demonstrate that GBM also elicits a uniquely severe T-cell exhaustion signature compared with other tumor types. This finding is manifested in the increased expression of multiple coinhibitory immune checkpoints and the decreased functional capacity of glioma-infiltrating T cells in comparison with other T cells infiltrating other models of i.c. malignancy. The mounting expression of alternative immune checkpoints may indicate a state of terminal exhaustion that cannot be reversed by traditional checkpoint blockade alone.

Importantly, different tumor histologies (and even different glioma models) in our studies elicited differing exhaustion signatures among infiltrating T cells, regardless of where in the body these tumors were introduced. These results suggest that different tumors might induce reproducibly distinct patterns of exhaustion. They also therefore suggest that the mechanisms whereby tumors induce exhaustion among T cells may not be universal and may represent a “convergent evolution” of sorts. These findings, however, may also begin to provide the foundation for personalized and rationally designed alternative immune checkpoint blockade: checkpoint blockade strategies successful in one cancer type may not be appropriate in others, and may ultimately require thorough characterization of the dysfunction elicited among a given patient's TILs.

Lastly, the lack of influence of tumor location on TIL exhaustion signature suggests a limited role for central nervous system (CNS)-specific factors in the programming of the T-cell exhaustion, a point of optimism for those looking to thwart T-cell exhaustion within the brain. Such strategies, as applied to primary peripheral tumors metastasizing to the brain, may prove equally successful against brain metastases or primary brain tumors, once afforded appropriate CNS access. Future studies will need to examine factors influencing tumor-specific induction of T-cell exhaustion, as well as the capacities for T-cell rescue with rationally designed immune checkpoint blockade or additional strategies.

G. Dranoff is an employee of Novartis. G.P. Dunn holds ownership interest (including patents) in Immunovalent Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Woroniecka, P. Chongsathidkiet, S.H. Farber, C. Jackson, G.P. Dunn, P.E. Fecci

Development of methodology: K. Woroniecka, P. Chongsathidkiet, C. Jackson, L.J. Hansen, Y.-R.A. Yu, D.D. Bigner, K.J. Weinhold, G.P. Dunn, P.E. Fecci

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Woroniecka, P. Chongsathidkiet, K. Rhodin, H. Kemeny, C. Dechant, S.H. Farber, A.A. Elsamadicy, X. Cui, C. Jackson, L.J. Hansen, V. Chandramohan, D.D. Bigner, P.E. Fecci

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Woroniecka, P. Chongsathidkiet, S.H. Farber, L. Sanchez-Perez, A. Giles, P. Healy, G.P. Dunn, P.E. Fecci

Writing, review, and/or revision of the manuscript: K. Woroniecka, P. Chongsathidkiet, S.H. Farber, A.A. Elsamadicy, S. Koyama, C. Jackson, T.M. Johanns, D.D. Bigner, A. Giles, P. Healy, G. Dranoff, K.J. Weinhold, P.E. Fecci

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Woroniecka, X. Cui, T.M. Johanns, D.D. Bigner, P.E. Fecci

Study supervision: L. Sanchez-Perez, P.E. Fecci

The work was supported in part by the National Institutes of Health Duke Brain SPORE Developmental Research Program (P.E. Fecci), the Medical Scientist Training Program at Duke University School of Medicine (K. Woroniecka), the NIH grant K08NS092912 (G.P. Dunn), the American Cancer Society-Institutional Research grant (G.P. Dunn), and the Physician-Scientist Training Program at Washington University School of Medicine (T.M. Johanns).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.