Immunophenotyping of Newly Diagnosed and Recurrent Glioblastoma Defines Distinct Immune Exhaustion Profiles in Peripheral and Tumor-infiltrating Lymphocytes

Glioblastoma (GBM) is the most prevalent and aggressive primary tumor of the central nervous system and as tumor recurrence is almost a certainty, new treatment modalities are urgently needed to specifically target and kill invading tumor cells (1). Unleashing the abilities of tumor-specific T cells could meet these needs (2, 3); however, extensive immunosuppression and immune evasion hinders the success of immunotherapeutic strategies (4).

A variety of mechanisms to escape a tumor-specific T-cell–mediated immune response have been identified in glioma and other cancer entities: (i) expression of immune checkpoint molecules inducing T-cell exhaustion, for example, through programmed death-1 (PD-1) receptor or CTLA-4 signaling (5, 6); (ii) modulation of the local tumor microenvironment by attracting regulatory T cells (Tregs) secreting immunosuppressive cytokines (7); (iii) escape from T-cell recognition caused by mutation or downregulation of HLA molecules (8); and (iv) a high degree of heterogeneity, denying a common target for the immune system (9). A major goal of current immunotherapeutic strategies is to overcome these mechanisms of T-cell suppression.

Two opposing subpopulations play a central role in tumor-specific immunity: CD8⁺ cytotoxic T cells (CTLs) and Tregs (3, 10, 11). CTLs represent the effector arm of the tumor-specific immune response and they are more prevalent in GBM than CD4⁺ T helper cells (12). Increased frequencies of CD3⁺ T-cell infiltration in general or CD8⁺ CTLs have been associated with a better outcome in various studies (12–15). On the other hand, Tregs are actively attracted to the tumor environment (16, 17) and support the immune escape by interacting with resident macrophages, suppressing T_Effector-cell activation and impairing tumor-specific immune responses (16, 18–20). Even though controversial, higher frequencies of Tregs in the tumor are reportedly associated with a worse prognosis and faster recurrence of GBM (15, 21, 22). Local immunosuppression has become a hallmark of GBM, and recent studies have demonstrated multiple additional pathways of immune escape beyond Treg-mediated suppression (2, 11, 20). Effector T cells are silenced by a combination of soluble factors, such as IL10 and TGFβ (18, 23, 24), as well as direct inhibition by myeloid-derived suppressor cells (MDSCs, CD33⁺ HLA-DR⁻; ref. 25), tumor-associated macrophages, and Tregs (20, 26). Together with the expression of immune checkpoint molecules, such as PD-L1 by tumor cells (27, 28), this combination presumably leads to the exhaustion of tumor-specific effector T cells (4).

GBM-mediated immunosuppression is not limited to the local tumor microenvironment as peripheral CD4⁺ T cells display an exhausted phenotype (29). Moreover, peripheral Treg counts, IL10 production, and the frequency of MDSCs are increased in patients with GBM compared with healthy donors (14), with circulating T cells expressing high levels of CTLA4 and monocytes elevating PD-L1 expression (26). It is however still uncertain which other pathways and surface molecules beyond the PD-1/PD-L1 axis might mediate and define the exhaustion of T cells.

Beyond their functional phenotype, tumor-infiltrating T cells are also characterized by their antigen specificity. T cells recognize their cognate antigen upon engagement of their unique T-cell antigen receptor (TCR) with peptide presented by HLA molecules. Profiling the immune repertoire by quantifying the TCR composition in blood and tumor tissue using targeted next-generation sequencing of the TCRVβ locus represents a novel technique that enables the assessment of clonal T-cell diversity. Hsu and colleagues and Sims and colleagues have provided first insight into TCR clonality in patients with GBM, and described a distinct TCR signature in the peripheral blood (30). However, the composition of the TCR repertoire in recurrent GBM and the question whether a recall antigen response takes place remain unresolved.

A detailed analysis of the T-cell memory, activation, and exhaustion phenotype as well as phenotypic changes during disease progression has not yet been performed in GBM. Here, we compared the T-cell phenotype and clonality of primary and recurrent tumors. As new immunotherapeutic approaches almost exclusively enter as clinical trials in the recurrent disease stage, our study delivers important insights into how the glioma-modulated immune system and tumor-specific T-cell response is configured in recurrent GBM. This information can help to determine how upcoming immunotherapeutic approaches might activate the tumor-specific immune response most efficiently.

Patients with newly diagnosed primary (pGBM) and recurrent (rGBM) isocitrate dehydrogenase-1 (IDH1) wild-type glioblastoma WHO IV were included. In the long-term survivor cohort (LTS), we only included patients with IDH wild-type (IHC and sequencing) glioma and survival >5 years (Supplementary Tables S1 and S2). Control samples were obtained from anonymized, age-matched healthy donors from the Department of Transfusion Medicine, University Hospital Hamburg-Eppendorf (Hamburg, Germany). All patients signed informed consent. This study was approved by the local ethics council of the Hamburg chamber for physicians and was performed in accordance with the Helsinki declaration of 1975.

Peripheral blood was collected before surgical tumor removal into EDTA-containing tubes. Ficoll gradient (PromoCell) centrifugation was performed for the isolation of peripheral blood lymphocytes (PBLs). PBLs were either directly processed or used after being frozen in RPMI/10% DMSO. Fresh tumor samples were immediately processed. Superficial blood was washed off before samples were minced and digested for 30 minutes at 37°C in Hank's medium with 1.5 mg/mL collagenase (Biochrom) and 10 μg/mL DNase (Roche). Digested tumor was strained, before tumor-infiltrating lymphocytes (TILs) were isolated using a Ficoll gradient. Peripheral blood mononuclear cells (PBMCs), and TILs were resuspended in flow cytometry staining buffer (eBioscience).

Flow cytometric analysis was performed on PBLs and TILs using two multicolor antibody staining panels (Supplementary Table S3). After F_C blocking, samples were stained in flow cytometry staining buffer (eBioscience) for 45 minutes at room temperature with the antibody cocktails, washed, and resuspended in buffer prior to analysis. Analysis was performed on a BD LSRFortessa flow cytometer and using the FACSDiva software (Becton Dickinson).

PBMCs were freshly isolated from EDTA-blood of patients with GBM and healthy controls using Ficoll gradient centrifugation. A total of 1 × 10⁶ PBMCs were incubated with PHA-L (Sigma Aldrich) at 1 μg/mL in 500 μL RPMI with 10% FCS for 24 hours at 37°C. Supernatants were frozen before being analyzed (Fig. 5A and B). For stimulation of purified CD3 T cells (Fig. 5C), we used PBMCs and fresh GBM tissue resections and sorted CD3⁺ cells using positive separation with magnetic beads (Miltenyi Biotec). Briefly, the columns were loaded with 1 × 10⁷ of either PBMCs or freshly isolated single-cell tumor suspension, washed, and eluted with 500 μL or 1 mL, respectively. CD3 purity was confirmed by flow cytometry including a live/dead staining. CD3⁺ TILs sorted out of fresh tissue were compared with IL2/OKT3 expanded TILs from unmatched patients with GBM. Next, 50,000 CD3⁺ T cells were activated with a combination of anti-CD3, -CD28, and -CD2 beads (T-cell activation/expansion kit from Miltenyi Biotec) for 72 hours. Supernatants were then collected and frozen for cytokine analysis. All samples were analyzed using the THelper-cell Cytokine Legendplex Kit (Biolegend) on a BD LSRFortessa flow cytometer (Becton Dickinson).

Genomic DNA from peripheral immune cells was isolated from >1 × 10⁶ frozen PBMCs or 2–3 mL of frozen citrate-blood using the genomic DNA Miniprep Kit (Macherey-Nagel). For TIL TCRVß repertoire sequencing, vital fresh tumor samples (∼1 cm³) were snap frozen in liquid nitrogen before genomic DNA was extracted as described above. The samples for TCRVß sequencing were prepared as described before (31). In brief, the TCRß genes containing the entire Vß, Dß, and Jß segments were amplified in a multiplex PCR, using primers further optimized from the previously published BIOMED2-TCRß-A and –B primer pools (31, 32). Using genomic DNA as template, the amplicons were tagged with Illumina adapters and indices in two consecutive PCR reactions, as described before (31). Phusion HS II (Thermo Fisher Scientific) was used for all PCRs. Both PCR reactions were size separated by agarose gel electrophoresis and amplicons with the expected size were purified using the NucleoSpinVR Gel and PCR Clean-up kit (Macherey-Nagel). Amplicon concentrations and purity were determined on Qubit (Qiagen) and on an Agilent 2100 Bioanalyzer (Agilent Technologies), respectively.

Next-generation sequencing (NGS) was performed as described before (31), using an Illumina MiSeq sequencer (500–600 cycle single indexed, paired-end runs (V2 or V3 chemistry). The adapter-ligated PhiX library was used as an internal control for Illumina sequencing runs. Analysis of demultiplexed and FastQ-formatted TCRß data was performed using the MiXCR analysis tool (33). A clonotype was defined by the CDR3 amino acid sequence for further analysis and only sequences with a read count ≥2 were included in the analysis.

Statistical analysis was performed using GraphPad Prism and IBM SPSS v22. t Test and Mann–Whitney U test were used for baseline data, as well as one-way ANOVA or Fisher exact as indicated in the figures. Plots were graphed using GraphPad Prism, R Foundation's R v2.12, Adobe Illustrator CC 2017. T-SNE analysis of flow cytometry data was performed in R using the package “Rtsne” (v0.13) parameterized with perplexity 23 and 20,000 iterations (34). Results were plotted using the package “ggplot2” (v2.2.1). Heatmaps were created using the function heatmap.2 from the R package “gplots” (v3.0.1).

We analyzed PBLs and TILs from patients with newly diagnosed primary GBM (pGBM, n = 19 PBL and 11 TIL samples) and recurrent GBM (rGBM, n = 16 PBL and 11 TIL samples) and compared them with PBLs from long-term IDH wild-type (wt) glioma survivors (LTS, n = 5) as well as age-matched healthy donors (HD, n = 15). Cohort composition and patient characteristics are summarized in Supplementary Table S1). We analyzed the expression of CD45RA, CCR7, and CD28 on CD8⁺ and conventional CD4⁺ T cells, which from here on will be only named CD4⁺ T cells. The marker combination enabled identification of the different stages of T-cell differentiation, namely T_NV = naïve (CD45RA⁺, CCR7⁺, CD28⁺), T_CM = central memory (CD45RA⁻, CCR7⁺, CD28⁺), T_TM = transitional memory (CD45RA⁻, CCR7⁻, CD28⁺), T_EM = effector memory (CD45RA⁻, CCR7⁻, CD28⁻), and T_TE = terminal effector (CD45RA⁺, CCR7⁻, CD28⁻) cells (Supplementary Fig. S1A and S1B). According to these markers, TILs exhibited a profoundly different phenotype compared with PBLs, as demonstrated by an exclusive infiltration of differentiated CD4⁺ and CD8⁺ T cells, a mixture of T_TM and T_EM in the CD8⁺ compartment and a majority of T_TM in the CD4⁺ compartment (Fig. 1A). Comparisons of the PBL profiles of pGBM and rGBM patients with HDs only showed minor differences in CD8⁺ and CD4⁺ T cells. Interestingly, compared with other PBL groups, LTS patients were unique in that they contained a significantly higher proportion of naïve (T_NV) CD8⁺ cells than all other groups (T_NV: P = 0.021–0.0023, mean 35.91% ± 8.3 SEM; Fig. 1B). In the TIL compartment, patients with rGBM displayed a significantly enlarged proportion of CD8⁺ T_EM cells (P = 0.015, mean 53.63% ± 6.1 vs. 31.96% ± 5.4 SEM) as well as CD4⁺ T_EM cells (P = 0.015, mean 32.44% ± 6.6 vs. 12.88% ± 3.3 SEM) compared with patients with pGBM. In contrast, CD8⁺ TILs with a T_TM phenotype were less common in patients with rGBM compared with patients with pGBM (P = 0.022, mean 20.48% ± 4.8 vs. 38.90% ± 5.7 SEM).

CD4⁺ cells constituted the major T-cell population in all groups (Fig. 1C). However, increased frequencies of CD8⁺ cells were observed in TILs, especially in the patients with recurrent tumors (pGBM: P = 0.045, mean 36.43% ± 2.3 vs. 28.39% ± 2.6 SEM; rGBM: P = 0.032, mean 48.48% ± 5.6 vs. 33.37% ± 3.9 SEM), suggesting an ongoing cytotoxic response to the tumor. In the periphery, both pGBM and rGBM, but not LTS patients showed a significant shift toward the CD8⁺ population compared with healthy donors. Interestingly, Tregs, defined as CD3⁺ CD4⁺ CD25^high CD127^low cells (Supplementary Fig. S2A), were more prevalent in the blood of patients with primary tumors, compared with HDs (P = 0.044, mean 6.81% ± 0.4 vs. 5.50% ± 0.5 SEM; Fig. 1C, right). No differences in Treg populations were detected between PBLs and TILs.

To assess whether patients with GBM had major differences in the lymphocyte population, we assessed routinely obtained differential blood counts (Supplementary Fig. S1D). In our cohort, patients with rGBM (n = 32) exhibited significantly lower absolute leukocyte (P = 0.005) and lymphocyte (P = 0.003) counts compared with pGBM (n = 46) patients, while the percentage of eosinophils was elevated in rGBM (P = 0.041).

To further characterize the immune signature of the main T-cell subsets, we developed a panel of 12 function-related markers and their combination resulted in 39 subpopulations for the three main T-cell subsets (CD8⁺, CD4⁺, and Treg cells). This panel included the activation markers HLA-DR and CD25, the homeostatic proliferation marker CD127, exhaustion-/senescence markers (PD-1, KLRG1, CD57, and Tim-3) and ATP-metabolizing enzymes CD39 and CD73, in addition to the already described differentiation markers CD45RO, CCR7, and CD28 (Supplementary Fig. S2A and S2B). The frequency of cells expressing a certain marker or marker combination for each sample cohort was assessed, and hierarchical clustering was applied to group samples according to their phenotypic similarities (Fig. 2A). The heatmap shows that TILs clearly clustered apart from PBLs, and highlights their activated/memory phenotype, as defined by the surface marker signature of the T_TM/T_EM differentiation (CD45RA⁻ CCR7⁻ CD28^+/−). Moreover, the distinct clustering was driven by the expression of HLA-DR, PD-1, Tim-3, and CD39, and absence of KLRG1 in both CD4⁺ and CD8⁺ cells, as well as expression of CD73 in CD8⁺ T cells (Fig. 2A).

To obtain a global impression of similarities and differences in cell surface marker expression in all sample groups, we used the dimensionality reduction algorithm t-distributed stochastic neighbor embedding (t-SNE, Fig. 2B; ref. 35). t-SNE analysis confirmed the two separated clusters for TILs and PBLs, as apparent in Fig. 2A. Comparing the TIL groups, primary and recurrent GBM samples could be easily distinguished. In peripheral blood, no conspicuous clusters could be identified at first view, except for the LTS samples (magenta), which defined a subpopulation distinct from the TIL samples and apart from the majority of blood samples. To further elucidate which markers contributed to the phenotypic differences between the groups of patients, we manually assigned the samples to groups based on their hierarchical clustering (Fig. 2A) and distribution in the t-SNE plot. Next, the phenotypic traits corresponding to each of the clusters were represented in a heatmap, showing the medians of the population frequencies within each group (Fig. 2C). This figure reveals that the most striking differential traits between the two TIL groups are the high expression of PD-1, and CD39, and HLA-DR in rGBM. In peripheral blood, the trait that most distinctively discriminated rGBM from the other samples was the high proportion of cells expressing PD-1, while the increased presence of naïve T cells (T_NV) and a low percentage of PD-1⁺ cells characterized LTS samples. Moreover, the most divergent phenotypes among GBM patient groups are rGBM and LTS, probably reflecting their differing clinical outcomes.

To assess the relevance of PD-1 and other exhaustion markers in the different groups of samples, we compared the expression of PD-1, KLRG1, CD57, and Tim-3, as well as the marker combination CD45RO and HLA-DR (for activated memory T cells) in CD4⁺, CD8⁺, and Treg cells (Fig. 3; Supplementary Fig. S2). In addition, we used spider plots to visualize and compare the expression of each marker between the CD8⁺ and CD4⁺ T-cell populations (Fig. 3B). As suggested in Fig. 2, PD-1 was the most prominently expressed marker on TILs, both for CD8⁺ and CD4⁺ cells, and was detected on 70.3% of the CD8⁺ cells present in pGBM and 79.9% in rGBM, as well as on 67.1% of CD4⁺ TILs in pGBM and 71.6% in rGBM (Fig. 3A and B). Also, PD-1 was expressed on a higher proportion of PBLs in patients with rGBM compared with pGBM (CD8⁺: P = 0.032, mean 33.7% ± 3.7 vs. 48.4% ± 5.6 SEM; CD4⁺: P = 0.04, mean 31.1% ± 6.0 vs. 51.4% ± 7.6 SEM), while the LTS group had low percentages of PD-1⁺ cells (CD8⁺: P = 0.049–0.0003, mean 27.2% ± 6.7; CD4⁺ mean 12.4% ± 2.9 SEM). Another exhaustion marker, the KLRG1 molecule, was expressed on approximately 49.2% of CD8⁺ and 17.4% of CD4⁺ PBL T cells, respectively, with lower expression on LTS PBLs (CD8⁺: P = 0.03, mean 36.0% ± 1.9; Fig. 3A and B). However, KLRG1 was expressed on significantly fewer TILs, and almost completely absent on CD8⁺ TILs in recurrent tumors (P = 0.04, mean 4.99 ± 0.9 vs. 9.29 ± 1.8). CD57, a marker for impaired proliferation and presumably for terminally differentiated T cells, was reduced in the CD8⁺ TIL compartment compared with PBLs, while it was significantly increased on CD4⁺ TILs (P = 0.01–0.001, pGBM mean 17.0% ± 3.5; rGBM mean 13.0% ± 2.7 SEM) (Fig. 3A and B). In addition, LTS PBLs also contained fewer CD8⁺ CD57⁺ T cells compared with the GBM PBL cohorts (P = 0.034–0.010). T-cell immunoglobulin-3 (Tim-3) is a negative regulator of cytokine production in CD4⁺ and cytotoxic CD8⁺ T cells and has been implicated as potential immune checkpoint in antitumor immunity (36). Tim-3 was only expressed on a small fraction of peripheral T cells and elevated on TILs (mean CD8⁺: PBLs 4.21%; TILs 8.0%; CD4⁺: PBLs 1.4%; TILs 5.3%; Fig. 3A and B). Finally, the HLA-DR⁺ and CD45RO⁺ marker combination, describing activated memory T cells, was particularly prevalent in CD8⁺ rGBM TILs (pGBM vs. rGBM, P = 0.017). Within the peripheral blood, an increasing expression was also noted in the CD4⁺ compartment when comparing rGBM and LTS blood to HDs (P = 0.012–0.001).

ATP is abundantly released into the extracellular space in solid tumors, and adenosine generated by the ectonucleotidases CD39 and CD73 (Fig. 4A) interferes with antitumor immune responses (24). CD39, as expected, was most frequently expressed on Treg cells, and was moderately reduced on peripheral Tregs from patients with pGBM compared with HDs (P = 0.029, mean 40.8% ± 4.9 vs. 54.3% ± 2.8). Strikingly, CD39 was significantly upregulated on CD8⁺, CD4⁺, and Treg TILs compared with PBLs (P = 0.006–0.0001; Fig. 4B). CD73, in turn, was expressed by a smaller percentage of CD8⁺ in TILs versus PBLs, concomitant with the reduction in naïve T cells (T_NV; Fig. 1B). Furthermore, CD73 was detected more often in CD4⁺ cells and Treg in LTS and TIL samples, here reflecting more naïve cells, and a special population of CD4⁺ memory cells, respectively. Thus, the upregulation of CD39 and the loss of CD73 confirm the distinct differentiated/activated and exhaustion signature in TILs.

As previous studies suggested that GBM-mediated immunosuppression may not be limited to the local tumor microenvironment, we assessed the activation capacity of PBLs from pGBM and patients with rGBM. PBLs (1 × 10⁶) were stimulated with PHA-L for 24 hours and the production of key T-cell cytokines was measured (Fig. 5A; Supplementary Fig. S3). Given the potential influence of dexamethasone treatment on this assay, we additionally segregated all patients with GBM into dexamethasone and non-dexamethasone GBM groups (Fig. 5A). Independent of dexamethasone, patients with GBM produced less cytokines compared with the HD group, and this trend was more pronounced in patients with pGBM. Particularly relevant is the reduction in the production of IL2 and IFNγ, both crucial mediators of a successful antitumor response (Fig. 5B). The low production of not only Th1 cytokines, but also of the Th2 cytokines IL4, IL13, and of cytokines characteristic of the Th17 responses indicates a peripheral hyporesponsiveness of T cells to PHA stimulation in patients with GBM (Supplementary Fig. S3A). The main difference between patients with pGBM and rGBM was a tendency toward lower Th1 cytokine production in patients with pGBM, which reached significance for IFNγ (P = 0.034). To assess whether GBM TILs show a dysfunctional cytokine response in comparison with PBLs, we purified fresh CD3⁺ GBM TILs and compared their cytokine profile to expanded TILs as well as purified peripheral T cells (Fig. 5C and D). Despite the higher proportion of memory cells present in the tumor, fresh and expanded GBM TILs showed equally strong cytokine expression for IFNγ and TNFα compared with HD peripheral blood. The production of IL17A was significantly elevated (P = 0.048) in freshly isolated CD3⁺ T cells from patients with GBM compared with HD. Similarly, both fresh and expanded TILs show higher level of IL5 expression compared with the peripheral CD3, either HD or GBM. Interestingly, although the production of IL2 was reduced in expanded GBM TILs (HD vs. GBM expanded TILs; P = 0.028), their cytokine profile demonstrated a wide range of additionally expressed cytokines compared with the other groups, such as IL4, IL9, and IL21 (Fig. 5D).

After comparative immunophenotyping, we investigated the immune repertoire composition using next-generation immunosequencing, to dissect the clonal architecture of the PBL and TIL compartment. The absolute number of identified clonotypes was highest in PBMCs from HD and lowest in tumor samples (Fig. 6A). To compare the derived clonotype distributions, irrespective of the absolute number of identified sequences, the Shannon–Wiener and inverse Simpson diversity indices were calculated. In agreement with the flow cytometry data, suggesting that the TIL compartment is composed of specific T-cell subpopulations, clonal diversity was significantly lower among TILs compared with the corresponding PBL compartment. In addition, a trend toward reduced diversity in the recurrent situation was observed for both compartments, although, only the comparison of TILs using the inverse Simpson index reached significance (P = 0.026; Fig. 6A). Subdivision of the observed clonotypes by frequency indicated that highly abundant clonotypes were significantly enriched in the TIL compartment in comparison with PBL, and hyperexpanded clonotypes accounting for more than 5% of the total clonal space were specifically enriched in rGBM TILs (P = 0.001–0.016; Fig. 6B). In line with the observation that a loss of diversity in the TIL compartment is due to oligoclonal expansion of frequent clonotypes, we found that the single most expanded clonotype occupied a significantly greater space in the rGBM TIL sample (Fig. 6C). Furthermore, CDR3 length distribution deviated from the expected Gaussian distribution, showing the strongest enrichment of clonotypes with a specific CDR3 length in TILs from rGBM samples (Fig. 6D). To identify whether the oligoclonal expansion in the rGBM TIL compartment was due to previously present/overlapping or new T-cell clones, matched pGBM and rGBM samples were compared on the basis of the CDR3 amino acid sequence. The proportion of TILs in rGBM samples that was already previously identified in the corresponding pGBM samples was comparatively low (mean 14% ± 5.1 SEM; Fig. 6E). In contrast, previously identified clonotypes accounted for up to one third in rGBM blood samples (mean 31% ± 6.2). Interestingly, the frequency of shared clonotypes in blood increased with age in the recurrent setting, which was not observed in the associated TIL samples (Fig. 6E). These differences in clonal overlap and the observed age-dependency of repopulating the immune repertoire in the recurrent GBM setting, indicate that during the time between primary tumor occurrence and recurrence, the composition of the immune repertoire is altered, that is, due to radio-/chemotherapy, and elderly patients might lack mechanisms to recreate diversity in the PBLs. In summary, we report a restricted clonal architecture of TILs in GBM that, in line with our immunophenotyping results, is aggravated in the recurrent situation. As the T-cell repertoire is shaped through the presented antigens, these results suggest a plasticity of the tumor microenvironment with different and/or reduced antigen diversity presented in the recurrent situation.

The profound molecular and cellular heterogeneity is a hallmark of GBM (1). During tumor evolution, this heterogeneity provides the molecular basis for the development of immune escape mechanisms (37), rendering established GBM highly efficient at suppressing a tumor-specific immune response (2). With the introduction of immunotherapeutic treatment approaches, which directly target and promote immune cells, investigations of treatment response and failure increasingly focused on characterizing the tumor-specific immune response. Yet, the heterogeneity of the antitumor response and the GBM-specific immune exhaustion profile during initial diagnosis and disease progression is poorly understood.

In our study, we demonstrate that TILs exhibit a distinct immune signature and clonal architecture, which markedly differs from circulating T cells. Our observations confirm that infiltrating T cells represent a specialized population which either adapts to or is modified by the tumor microenvironment. The predominant transitional memory (T_TM) phenotype in pGBM TILs, and the even more advanced effector memory (T_EM) TILs in rGBM, is indicative of a highly differentiated and antigen-experienced T-cell population in tumors. In addition, these T cells showed a clonal restriction due to oligoclonal expansion of few T cells. We therefore conclude that the T cells have undergone antigen-induced activation and are likely to have selectively infiltrated the tumor to trace and eliminate tumor cells (3). Compared with their circulating, naïve, and less experienced T_CM counterparts, TILs should be able to mount a robust IFNγ-supported, granzyme B and perforin-mediated cytotoxic antitumor response (38). However, tumors can counteract these activated and intralesional T cells by several mechanisms, including loss of expression of tumor-specific antigens and induction of premature T-cell exhaustion, defined by nonresponsiveness, loss of effector functions, low IL2 production, and subsequent stable expression of exhaustion markers in the immune environment (3, 39).

To characterize the immune exhaustion profile of GBM TILs in comparison with PBLs, we analyzed the following markers: PD-1, Tim-3, KLRG1, CD57, CD39, and CD73. PD-1 proved to be the most strikingly upregulated exhaustion marker on both CD8⁺ and CD4⁺ TILs, which is in line with a recently published study (40). Expanding on those previous findings, we further demonstrate a significant increase of PD-1 expression on PBLs in patients with rGBM versus patients with pGBM, suggesting an even more pronounced systemic exhaustion in the relapse setting. Beyond PD-1, we observed an elevated expression of Tim-3 in TILs in pGBM and rGBM. Tim-3 is an established coinhibitory T-cell receptor, which was discovered as mediator for the induction of immune tolerance in CD4⁺-mediated immune responses (36). However, the mean frequency of Tim-3⁺ TILs was generally only low in TILs as well as in PBLs, arguing against a central role of Tim-3 in the exhaustion phenotype of GBM T cells. Rather, Tim-3 expression might mark a small subpopulation with pronounced exhaustion among TILs. More dominant was the striking prevalence of CD39⁺ TILs, and especially CD8⁺ T cells. So far, CD39 has primarily been described together with CD73 in the context of local metabolic immunomodulation in glioma and other tumor entities (41). Here, results suggested that CD39⁺ on CD4⁺ T cells acts in synergy with CD73 on glioma cells, entailing an increase in the local adenosine level that inhibits T-cell activation via A2A receptor (41). While CD39 and CD73 were expected to be upregulated simultaneously, our findings of an isolated increase of CD39 on TILs indicate that CD39 and CD73 are in part differentially regulated on TILs in GBM. Interestingly, a recent investigation in patients with chronic viral infections has identified CD39, particularly when expressed together with PD-1, as novel marker for terminal exhaustion in CD8⁺ T cells (42). CD39 could therefore be suitable as potential target for checkpoint inhibition. Extending the characterization of exhaustion markers, we further investigated CD57 and KLRG1. Unexpectedly, the senescence marker KLRG1 was markedly reduced and only presented by a fraction of T cells in the tumor compartment, while CD57 showed an inverse expression pattern, with reduced prevalence in CD8⁺ TILs. In one of the first studies providing evidence of alterations in T-cell activation in GBM, Fornara and colleagues (43), demonstrated that a high prevalence of CD4⁺ CD57⁺ T cells in the blood of patients with GBM correlated with immunosenescence and decreased survival. Extending this work, we observed that CD57 was primarily expressed on CD4⁺ TILs, while being reduced on CD8⁺ T cells. Interestingly, CD57 has been identified as marker for highly cytolytic CD8⁺ T cells apart from also defining terminal differentiated T cells (44). Thus, the low prevalence of CD57⁺ on CD8⁺ GBM TILs points toward a functionally impaired cytotoxic T-cell response in the tumor microenvironment. Taken together, although infiltrating T cells display an advanced transitional- (T_TM) or even effector memory phenotype (T_EM), the distinct PD-1⁺ KLRG1⁻, CD57⁻, CD39⁺, Tim-3⁺ marker profile of CD8⁺ TILs indicate an inherent immune exhaustion signature, characteristic of impaired functionality due to continuous (tumor)–antigen exposure, rather than terminally senescent T cells (45).

The protumorigenic role of Tregs was emphasized in glioma and many other cancer entities (11). Interestingly, although we did not observe a predominant Treg infiltration in the tumor compartment, we found an increase of Tregs in PBLs of pGBM patients compared with HD. PD-1 was strongly expressed on a large proportion of tumor-infiltrating Tregs compared with circulating Tregs. Whether the expression of PD-1 on functionally opposing T-cell populations, such as CD8⁺ and Tregs in tumor microenvironment has different functional implications is not yet clear. As suggested by Lowther and colleagues, the PD-1/PD-L1 axis may be crucial to induce Treg differentiation through a not yet fully understood regulatory pathway with IFNγ as a key player of tumor-mediated immune exhaustion secreted by TILs (46). Collectively, our observation of an overall high prevalence of PD-1⁺ on CD4⁺, CD8⁺, and Treg cells indicates a major role of PD-1 for both, the cytotoxic as well as immunosuppressive T-cell subpopulations.

While the brain has been described as a privileged immunologic compartment, the functional state of the peripheral immune system in patients with GBM is discussed controversially (47). The fact that GBMs rarely metastasize has been, among other hypotheses, attributed to a well-functioning peripheral antitumor immune surveillance that prevents potential circulating tumor cells (CTCs) from forming distant metastases (47). However, our observations of an increased proportion of circulating Tregs in patients with pGBM and the impaired production of key cytokines by T cells are in line with recent and ongoing studies by others, demonstrating profound GBM-mediated suppression also of the peripheral immune response. For example, a study by Fecci and colleagues revealed that GBMs impair peripheral immune activation by forcing T cells to home into the bone marrow through induction of the sphingosine-1-receptor type 1 axis (48). In fact, this observation might explain why we did not observe alterations in markers of immune exhaustion on circulating T cells in patients with GBM, although we found a strong suppression of functionality in cytokine production. Our observation of a general decline in the CD4⁺/CD8⁺ ratio in patients with GBM versus HD might be another indicator of profound alterations in immune composition in GBM and reflect the shift in the cytotoxic T-cell and Th-cell equilibrium. Interestingly, the analysis of PBLs from long-term survivors demonstrated a distinct marker expression profile, characterized by a more naïve CD8⁺ T-cell compartment with increased CD73 expression and a decreased expression of immune exhaustion markers such as PD-1. Together with our findings of a functionally impaired peripheral T-cell compartment in patients with GBM, we believe that immune profiling of the peripheral immune repertoire might serve as important prognosticator for immunotherapeutic treatment response and clinical outcome.

TCR sequencing of matched blood and tumor samples demonstrated a clonal restriction of the TIL repertoire due to oligoclonal expansion of few T-cell clones. The immunophenotypic uniqueness of the TIL compartment dominated by T_TM and T_EM phenotypic T cells is highly suggestive for tumor-specific T cells that predominately reside in the local tumor microenvironment. The diversity of these tumor-specific TCRs is determined by either few tumor antigens or different epitopes of the same antigens in contrast to the plethora of antigens recognized by PBLs. Thus, a restricted repertoire in comparison with the corresponding PBLs supports the observation of an immunophenotypic distinct TIL population. Interestingly, the clonal restriction was even more pronounced in rGBM samples. As CD8⁺ TILs with an activated memory phenotype increased in rGBM samples, the observed increase of highly abundant clonotypes might be particularly driven by CD45RO⁺ and HLA-DR⁺ TILs, also showing a high PD-1 expression. Although the restricted clonality in TILs points toward a reduced set of tumor-specific or -associated antigens in the recurrent situation, reduced clonality has been reported to correlate with enhanced response to immunotherapy directed against PD-1 in melanoma (49). Together with our finding of a highly prevalent expression of PD-1 and its role as a discriminating marker of TILs in GBM, the restricted clonality supports the rational of ongoing clinical trials with anti-PD-1/PD-L1–directed therapeutics (50).

Consistent with interlesional comparison of TIL repertoire diversity in melanoma (51), we detected a relatively low clonal overlap between TILs from paired tumor samples in comparison with blood samples. Assuming that TILs are composed of tumor antigen–specific T cells, this observation might be due to the high clonal heterogeneity in GBM and potentially a diverse antigen expression of the primary versus recurrent tumor. In addition, TILs might indeed reside in the local tumor microenvironment resulting in a spatial heterogeneity within the same tumor. Our findings of a higher clonal overlap with increasing patient age in the blood samples is in line with previous reports of an increasingly stable and restricted T-cell repertoire with advancing age (52). The observation of an age-dependent replenishment of the peripheral T-cell repertoire during disease progression also opens up the question, whether the adjuvant radio-/chemotherapy may affect T-cell diversity and immunocompetence at the time point of tumor recurrence (53). Future studies will have to address the question how immunotherapeutic approaches can be combined with adjuvant therapy and more specifically investigate the optimal time-point to evoke a lasting tumor-specific immune response. These topics can potentially be addressed by flow cytometric profiling of the exhaustion phenotype and next-generation repertoire sequencing of tumor-specific T cells after primary diagnosis and during adjuvant therapy.

Immunophenotyping of T cells in patients with GBM revealed a TIL-specific immune exhaustion profile, defined by an advanced transitional- or effector memory phenotype and the distinct PD-1⁺ KLRG1⁻, CD57⁻, CD39⁺, Tim-3⁺ marker profile of CD8⁺ TILs. Our study demonstrates that increased immune exhaustion can be found primarily in TILs, even though they are able to react to a strong stimulus when taken out of the tumor microenvironment. However, we also demonstrate an impaired cytokine response to PHA stimulation of PBLs in GBM, indicating a certain degree of immune dysfunction in peripheral blood. The clonal restriction of tumor-specific T cells further highlights the narrowing of the TCR repertoire in rGBM TILs, pointing toward a reduced antigen diversity available for the tumor-specific T-cell response during disease progression. Collectively, by combining detailed immunophenotyping of the immune exhaustion profile with TCR repertoire sequencing, our study provides valuable data to improve the design of upcoming immunotherapeutic studies and overcome the challenges of immunosuppression in newly diagnosed and recurrent GBM.

M. Westphal has received honoraria for advisory board participation for Bristol Myers Squibb.

Conception and design: M. Mohme, S. Schliffke, M. Binder, E. Tolosa

Development of methodology: M. Mohme, S. Schliffke, C.L. Maire, C. Gehbauer, N. Akyüz, M. Holz, M. Binder, E. Tolosa

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Mohme, S. Schliffke, J. Matschke, N. Akyüz, M. Schaper, T. Martens, N.O. Schmidt, S. Peine, M. Westphal, M. Binder, K. Lamszus

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Mohme, S. Schliffke, C.L. Maire, A. Rünger, L. Glau, K.C. Mende, J. Matschke, N. Akyüz, M. Binder, E. Tolosa, K. Lamszus

Writing, review, and/or revision of the manuscript: M. Mohme, S. Schliffke, C.L. Maire, L. Glau, K.C. Mende, J. Matschke, C. Gehbauer, N. Akyüz, N.O. Schmidt, S. Peine, M. Westphal, M. Binder, E. Tolosa, K. Lamszus

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Mohme, S. Schliffke, J. Matschke, N. Akyüz, S. Zapf, M. Holz, S. Peine, M. Binder

Study supervision: M. Binder, K. Lamszus

This work is supported by grants from the Else Kröner-Fresenius-Stiftung (to M. Mohme), from the German Research Foundation (DFG, KFO296/S1; to E. Tolosa), the Johannes-Bauer Stiftung (to M. Westphal and K. Lamszus), the Annemarie Hilgemann and Hubertus Wald Stiftung (to M. Binder). We also thank the technicians from the FACS core facility who continuously supported us during flow cytometric measurements.

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.