Glioblastoma (GBM) is the most common primary brain tumor in adults and a near uniformly fatal disease, with a median survival rate of under 2 years (1). The current standard-of-care for GBM involves maximal safe surgical resection, followed by chemoradiation and adjuvant temozolomide, but the malignancy is surgically incurable due to its invasive nature (2). Additional challenges for this paradigm include patients with significant residual disease following surgery, especially in the setting of multifocal disease or tumor in eloquent locations, mediating sensorimotor and language functions, or with variants like unmethylated O⁶-methylguanine methyltransferase (MGMT), which confers resistance to radiation therapy and temozolomide (3, 4). Immunotherapeutic approaches, such as chimeric antigen receptor (CAR) T cell therapy, may offer an additional avenue for treating GBM. Advantages of CAR T immunotherapy include the autologous nature of CAR T development, tumor specificity, and bypassing the requirements for antigen presentation and co-stimulatory signals necessary for an endogenous antitumor response. Early studies have assessed the efficacy of CAR T therapy in GBM patients against targets including the epidermal growth factor receptor (EGFR) variant III (EGFRvIII) (5, 6), ERBB2/HER2 (7), and IL13Rα2 (8).

EGFRvIII is a common tumor-specific splice variant of EGFR present in human tumors, found in 30% of newly diagnosed GBM cases and second in EGFR alteration frequency only to wild-type EGFR amplification (9, 10). EGFRvIII results in a constitutively active receptor that is resistant to EGFR inhibitors, and has been accordingly characterized as a negative prognostic marker (10, 11). Given the extracellular location of the alteration combined with the presence of a novel glycine residue as a result of the abnormal splicing, EGFRvIII has been an attractive antigen for immunotherapy, despite the lack of uniform expression on all glioblastoma cells. Early immunotherapeutic approaches have included monoclonal antibodies and rindopepimut, a peptide vaccine (12, 13). Beyond these approaches, our group recently concluded a first-in-human study of CAR T cells directed against EGFRvIII for recurrent GBM, which enrolled 10 patients (5). All ten patients had IDH wild-type GBM with an unmethylated MGMT promoter. This Phase I trial (NCT02209376) verified successful on-target activity and significant EGFRvIII antigen reduction in the brain following a single intravenous dose, with one patient exhibiting residual stable disease for 18 months and an overall survival of nearly 3 years (14). However, patients exhibited variable levels of CAR T trafficking to active tumor sites, changes in tumor antigen expression, and lymphocytic infiltration, suggesting significant interpatient heterogeneity in response.

Due to the autologous nature of CAR T cell development, CAR T cells are currently developed via T cells derived from the same patient. Accordingly, earlier work on CAR T therapy in the setting of hematologic malignancies hypothesized that differences in therapeutic success may be attributed to baseline interpatient variation in immune system deficiencies and intrinsic T cell characteristics. These variations include elevated frequencies of certain T cell populations that may be associated with higher likelihood of response (15, 16). For example, Fraietta et al. determined that remission following CD19-directed CAR T therapy for chronic lymphocytic leukemia was associated with factors such as enrichment in memory-related genes and STAT3 pathway activity at the pre-CAR T treatment baseline (15). These considerations also hold relevance in the setting of GBM, due to extensive documentation of lymphopenia even in treatment-naïve patients and GBM-induced mechanisms of T cell dysfunction, such as senescence, anergy, and exhaustion (17, 18). However, to date, no study has characterized how intrinsic characteristics and phenotypes of patient T cells and generated CAR T cells predict therapeutic efficacy for GBM or any other solid tumor. Accordingly, the aim of our study was to characterize T cell and CAR T cell characteristics in patient apheresis and transduction products predictive of peripheral engraftment and clinical response for EGFRvIII-directed CAR T therapy in recurrent GBM.

Several earlier studies in the setting of hematologic malignancies have identified genotypic and phenotypic T cell characteristics predictive of CAR T clinical response, which may inform patient selection, identification of the most effective T cell populations for clinical expansion during CAR T manufacturing, and changes to transduction and activation procedures during manufacturing that may improve CAR T efficacy. However, the impact of these characteristics is less characterized for CAR T therapy for solid tumors, including GBM. In our study, we determined that PD1 expression in patient CAR transduction product was associated with peripheral engraftment and time-on-trial for EGFRvIII-directed CAR T treatment in recurrent GBM. This positive association was also observed for PD1 co-expression with ICIs (CTLA4, TIM3, LAG3) and activation markers (GRZB, HLA-DR), but was not observed for any ICI or activation marker alone, suggesting that PD1 may be the primary driver of these correlations with our surrogates of clinical response. While these correlations in the transduction product were significant across CAR⁺ and CAR^- subpopulations as well as CD4⁺ and CD8⁺ T cells, no significant associations were observed in the apheresis product prior to CAR T generation. The demonstration of the positive association between PD1 and our primary endpoints in both the CAR⁺ and CAR^- populations raises the possibility that the potential impact of PD1 on clinical response operates on the level of several cell populations in the patient infusion product as a whole, rather than just the CAR⁺ population. Moreover, the persistence of these associations in the CAR- population suggests that these findings may be partially attributable to the general manufacturing process, independent of CAR transduction. These findings collectively suggest that PD1 expression in EGFRvIII-directed CAR T transduction products may predict improved treatment response for recurrent GBM and that the differences observed were not patient-inherent but due to the infusion product preparation.

Our analysis focused on the immune checkpoint PD1, in accordance with the focus of several trials on the blockade of PD1 signaling using agents like pembrolizumab or nivolumab as an adjunct to enhance CAR T therapy (22–28), including a recently concluded trial studying EGFRvIII-directed CAR T cells in combination with pembrolizumab (NCT03726515). Our results notably contravene earlier research documenting PD1 as a marker of T cell exhaustion and demonstrating PD1 inhibition as a means of improving T cell proliferation, activation, and effector functions (24, 29–34). Consequently, several studies have investigated PD1 blockade as an adjunct to improve the efficacy of CAR T therapy against malignancies (24, 25, 32–34). Moreover, GBM has been shown to mediate PD-L1/PD1 axis signaling via aberrant tumor expression of PD-L1, leading to attenuation of tumor infiltrating lymphocyte response and promotion of Treg expansion (35, 36). For CAR T therapy, Finney et al. determined that PD1 levels in patient apheresis product as well as PD1 acquisition during treatment were both predictive of a worse response for anti-CD19 CAR T for acute lymphoblastic leukemia (ALL) (37). Similarly, Fraietta et al. documented poorer response for chronic lymphocytic leukemia (CLL) among anti-CD19 CAR T cells expressing PD1 (15).

Nevertheless, our present study contrarily determined that in GBM, PD1 expression was associated with an improved treatment response. In accordance with our findings, some studies have argued that PD1 is not exclusively a marker of exhaustion but may also mark other physiologic conditions such as chronic antigen stimulation or stages of T cell activation (25, 38–41). Several studies have demonstrated that PD1 is not constitutively expressed on CAR T and T cells, but is rather first induced and escalated following antigen stimulation due to its role in naïve T cell priming (28, 38, 40–42), indicating that PD1 expression may also reflect the activation status of CAR T cells, rather than exhaustion alone. It is possible that the PD1-expressing cell populations may represent an activated population of CAR T cells, rather than signifying a state of exhaustion. Moreover, Wei et al. determined that PD1 blockade actually inhibited proliferation and effector differentiation for their anti-CD19 CAR construct (41). The authors posited that PD1 may carry out several T cell regulatory functions beyond exhaustion, given the presence of two signaling motifs on the PD1 receptor. In the setting of GBM, Davidson et al. determined that PD1⁺ tumor-infiltrating lymphocytes were representative of chronically activated effector T cells with characteristics of both exhaustion and activation, and that this population had higher T cell receptor diversity and IFN-γ production than its PD1^- counterpart (43). In the setting of melanoma, Gros et al. additionally found that PD1 expression was a marker for clonally expanded CD8⁺ tumor-infiltrating lymphocytes (44). Finally, in a non-cancer population, Gustafson et al. demonstrated that baseline PD1 expression levels may be high even in healthy subjects and are directly correlated with the frequency of CD45RO+ memory T cells (45). These factors may explain the present study’s findings that PD1 expression was associated with improved treatment response for anti-EGFRvIII CAR T therapy. It is also important to note that these findings for our study’s anti-EGFRvIII construct may not be generalizable to other constructs. For example, earlier research has provided evidence that the addition of pembrolizumab may enhance antitumor efficacy for CAR constructs used for B-cell malignancies and malignant pleural disease (22, 25, 46, 47). However, few studies have utilized simultaneous administration of CAR T infusion and PD1 inhibition to assess the impact of PD1 modulation during the early post-infusion period, the period of peak CAR T activity in the present trial. Consequently, further functional experiments in the setting of GBM are warranted to validate these findings.

While research into determinants of peripheral engraftment and clinical response for CAR T therapy has been conducted primarily for hematologic malignancies, this study advances the understanding of these predictors in the setting of CAR T for solid tumors. Several studies have examine genotypic and phenotypic characteristics associated with the efficacy of anti-CD19 CAR T therapy for B cell malignancies, including ALL, CLL, large B cell lymphoma (LBCL), and multiple myeloma (15, 16, 37, 48, 49). Fraietta et al. and Deng et al. identified the expression of memory signatures as a positive predictor for treatment response, while markers of exhaustion signified decreased rates of treatment response (15, 48). While Fraietta et al. did not find any patient or tumor characteristics predictive of treatment response for CLL, other studies for ALL and LBCL have suggested that variables like pretreatment tumor burden are associated with clinical response (15, 37, 49). Moreover, these analyses have facilitated the identification of mechanistically relevant T cell subpopulations in apheresis or infusion products associated with treatment response (15, 16, 49). For example, enrichment of a CD27⁺CD45RO^-CD8⁺ population in apheresis products prior to CAR T generation predicts remission for both CLL and multiple myeloma (15, 16). Nevertheless, the heterogeneity in determinants identified between studies likely emphasizes the disease- and construct-dependent nature of genotypic and phenotypic characteristics influencing CAR T therapeutic success.

The findings of the present study and the earlier body of research on predictors of treatment response raise several implications for CAR T development and utilization. Identification of T cell subpopulations associated with improved treatment response may facilitate patient selection and identification of optimal T cell populations for clinical expansion during the manufacturing process (15). These considerations are important given the immunosuppressive nature of GBM itself, with extensive documentation of T cell dysfunction in GBM patients, even in treatment-naïve status (17, 18, 36, 50). In the present study, a positive but nonsignificant correlation between baseline PD1 expression in a patient’s apheresis product and post-manufacturing PD1 expression in the infusion product was documented. However, an assessment of the association between these two variables was limited by the small sample size of the trial. Characteristics negatively influencing CAR T proliferation, engraftment, or response may also represent promising targets for modulation (15, 51). In the setting of anti-CD19 CAR T for CLL, pharmacologic inhibition of glycolysis, a pathway negatively associated with treatment response, improved CAR T proliferation and effector differentiation (15). For PD1, the phenotypic marker of focus in this study, when comparing patient apheresis and infusion products, there was a post-manufacturing increase in PD1 expression for CD4⁺ cells selectively that was not observed for the CD8⁺ population. Moreover, this change was documented across CD4⁺CAR⁺ and CD4⁺CAR^- cells, suggesting that the increase in PD1 expression occurred in a CAR transduction-independent manner. However, no correlations between manufacturing characteristics and change in PD1 expression with manufacturing were elucidated, which is an area warranting further study. Finally, these studies may inform clinical decisions including the optimal time during a malignancy’s course to initiate CAR T therapy. For example, Garfall et al. determined that the CD27⁺CD45RO^-CD8⁺ T cell subpopulation predicting treatment success was significantly less enriched in relapsed cases of multiple myeloma, relative to primary cases (16). These findings hold relevance due to the present study’s focus on recurrent GBM cases, with all nine patients having received systemic therapy, including both temozolomide and radiotherapy prior to CAR T generation. Using our more recent cohort of primary GBM patients treated with EGFRvIII-directed CAR T (NCT03726515), we aim to elucidate differences in clinical response determinants between these two clinically distinct GBM subpopulations.

There are several limitations to the present study. First, this analysis is limited by the small sample size of the phase 1 study (n=9), constraining the statistical power and range of hypothesis testing that could be performed. Nevertheless, PD1 exhibited significant associations with peripheral engraftment and time-on-trial, both alone and with ICIs and activation markers, despite the small sample size of the trial. Moreover, in contrast to similar earlier analyses conducted in the setting of hematologic malignancies (15, 16, 48), there is no binary marker of clinical response for GBM that could be studied as an endpoint, leading to the use of peripheral blood engraftment and time-on-trial, as determined by clinical and radiographical findings of progression, as clinical outcomes. Nevertheless, earlier studies have determined that peripheral CAR T engraftment is a significant correlate for treatment response (49, 52), and peripheral engraftment was predictive of time until progression for our study. Additionally, this study was conducted for recurrent GBM patients, and the present findings are not generalizable to primary GBM patients, due to factors like differences in treatment history and antigen burden. Given earlier research demonstrating that systemic therapy alters patient T cell attributes (16, 53), we aim to replicate the present analyses for our follow-up trial on EGFRvIII-directed therapy for primary GBM to characterize determinants of response in this patient population. Finally, the correlations elucidated in this study alone do not ensure a mechanistic or functional link between PD1 expression and anti-EGFRvIII CAR T efficacy. Modulating PD1 activity in our CAR T construct for functional assays will be the subject of future research.

In conclusion, our study demonstrated that PD1 expression, both alone and when co-expressed with ICIs and activation markers, in patient transduction products was associated with increased peripheral engraftment and time-on-trial for recurrent GBM.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by University of Pennsylvania Institutional Review Board. The patients/participants provided their written informed consent to participate in this study.

OT, LT, JM, SL, and ZB designed the research studies. OT, LT, TY, and RX conducted the experiments. OT, LT, TY, RX, and ZB analyzed data. IK and MG provided reagents. OT, LT, and ZB wrote the manuscript. All authors edited final version of the manuscript. All authors contributed to the article and approved the submitted version.

This work was supported by funding from the Neurosurgery Research & Education Foundation (OT), the GBM Translational Center of Excellence (DO’R, ZB), The Templeton Family Initiative in Neuro-Oncology (DO’R), and The Maria and Gabriele Troiano Brain Cancer Immunotherapy Fund (DO’R).

DO’R receives laboratory support from Tmunity Therapeutics. DO’R and ZB are on patents relating to CAR T cell therapy for GBM.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.