Challenges and potential of PD-1/PD-L1 checkpoint blockade immunotherapy for glioblastoma

The PD-1/PD-L1 pathway plays an important role in suppressing the function of T cells in eradicating tumour cells [18‐20]. PD-L1 is upregulated in several types of solid tumours, and high expression levels of PD-L1 often indicate better clinical efficacy of PD-1/PD-L1 checkpoint blockades [21‐24]. Although this phenomenon is not ubiquitous, it still drives us to explore the connection between PD-L1 expression and the clinical efficacy of PD-1/PD-L1 checkpoint blockades in glioblastoma patients.

Berghoff et al. observed PD-L1 expression in 88% of newly diagnosed and 72.2% of recurrent glioblastoma specimens [13]. Similarly, Nduom et al. measured PD-L1 expression in 94 glioblastoma patients and found that 61% of patients had tumours with PD-L1 positive cells [14]. A phase I cohort of Checkmate 143 evaluated the safety, tolerability and clinical effects of nivolumab with or without ipilimumab (which targets CTLA-4) in recurrent glioblastoma patients. Among all 40 patients, 10 patients were randomized to receive nivolumab 3 mg/kg every 2 weeks, 10 patients received nivolumab 1 mg/kg + ipilimumab 3 mg/kg every 3 weeks for 4 doses, and the other 20 patients received nivolumab 3 mg/kg + ipilimumab 1 mg/kg every 3 weeks for 4 doses. Omuro et al. demonstrated that the subgroup receiving nivolumab 3 mg/kg tolerated the treatment better than other subgroups that received the other combinations of nivolumab 1 mg/kg + ipilimumab 3 mg/kg and nivolumab 3 mg/kg + ipilimumab 1 mg/kg (90% vs 70% vs 80%). Other than fatigue and diarrhoea, which were the most common treatment-related adverse events (AEs) (30% vs 80% vs 55%; and 10% vs 70% vs 30%, respectively), no other side effects were observed. Nivolumab monotherapy was better tolerated than combination therapy. Simultaneously, we found that the dose of the ipilimumab monoclonal antibody was negatively correlated with patient tolerance, which may be explained by the critical role of the ipilimumab antibody in the earlier phase of T cell activation that can cause an extensive impact in the immune network [25]. The phase III clinical trial Checkmate 143 reported that PD-1 monoclonal antibody (nivolumab) monotherapy does not improve overall survival (OS) time compared with bavacizumab therapy in recurrent glioblastoma patients who were previously treated with chemotherapy and radiotherapy. The median PFS was 1.5 months for nivolumab vs 3.5 months for bavacizumab, the median OS was 9.8 months for nivolumab vs 10.0 months for bavacizumab, and the objective response rate (ORR) was 8% months for nivolumab vs 23% months for bavacizumab [26]. One possible reason for the failure of nivolumab monotherapy may be lymphopenia caused by radiotherapy. Yovino et al. found that after RT with 30 conventional fractions of 2 Gy, the circulating lymphocytes received a 2.2 Gy mean dose, and 99% of the circulating lymphocytes received mean doses ≥0.5 Gy [27]. However, nivolumab monotherapy exerts an immune activation effect through competitive binding with the PD-1 receptor on lymphocytes. Another possible reason may be the anergic nature of effector T cells to tumour-specific antigens in the TME. Furthermore, Wherry et al. examined the phenotypes of tumour infiltrating lymphocytes (TILs) in glioma specimens and found phenotypes rich in CD95, PD-1, PD-L1, CTLA-4, LAG3, and TIM-3, which obviously indicated the immune exhaustion of T cells [28]. In addition, Reardon et al. [17] also found that TILs express immunoinhibitory molecules, including CTLA-4 and PD-1, or coexpress PD-1 and TIM-3. However, the PD-1+/TIM-3+ phenotype represents an exhausted CD8+ T cell population in tumours [29]. Considering the low immunogenic characteristics and complicated immunosuppressive networks in glioblastoma, PD-1 checkpoint blockades are unlikely to overcome the factors leading to T cell anergy.

Based on the phase III clinical trial results, Checkmate 143 reported that nivolumab did not exhibit increased survival benefits over bavacizumab, researchers then explored the clinical efficacy of nivolumab + RT ± TMZ in newly diagnosed glioblastoma patients in ongoing phase III clinical trials, including Checkmate 498 (NCT02617589) and Checkmate 548 (NCT02667587). Checkmate 498 is comparing the efficacy of nivolumab + RT versus the standard treatment of TMZ + RT in newly diagnosed glioblastoma patients with unmethylated MGMT. Checkmate 548 is comparing the efficacy of nivolumab + RT + TMZ versus RT + TMZ in newly diagnosed glioblastoma patients with methylated MGMT. The clinical trials on glioblastoma are summarized in Table 1.

Glioblastoma is highly heterogeneous with intratumoural heterogeneity and intertumoural heterogeneity. According to the 2016 CNS WHO classification, glioblastomas are divided into glioblastoma, IDH-wild type and glioblastoma, IDH-mutant type based on molecular pathology [30]. Approximately 90% of glioblastomas are IDH-wild type, which indicates a worse prognosis, and approximately 10% of glioblastomas are IDH-mutant type, which indicates a better prognosis [31]. In addition, glioblastoma has been divided into four major subtypes based on genomic discrepancies: (1) neural, (2) pro-neural (PN), (3) classical (CL), and (4) mesenchymal (MES) [32]. These four subtypes have distinct cellular and molecular characteristics in their respective microenvironments. For example, NF1 and TP53 deletions and mutations were found in classical type, PDGFRA amplification and IDH1 point mutation were found in pro-neuronal type and EGFR overexpression was found in neuronal type [32]. Thus, finding therapeutically targetable genes that are expressed by all four subtypes is challenging. For example, Wang et al. analysed immune cell types in human PN, CL, and MES samples and found that CD4+ memory T cells, type-2 polarized macrophages (M2), and neutrophils were commonly increased in the MES subtype but not in the other subtypes [33]. Furthermore, Berghoff et al. demonstrated that the MES subtype of glioblastoma has higher PD-L1 expression [13]. Despite the genomic discrepancies and distinct cellular and molecular characteristics in the four subtypes, glioblastoma ubiquitously exhibited an immunosuppressive microenvironment that involves a number of tumour-cell-intrinsic and tumour-cell-extrinsic factors [34]. In contrast to NSCLC and melanoma, which have higher levels of tumour mutational load (TML) [35, 36], glioblastoma exhibits a lower TML in most instances and infrequently shows a high TML when it is deficient in MMR protein and there is an exonuclease proof-reading domain of the DNA polymerase epsilon gene (POLE) mutation. Thus, varying sensitivities to PD-1/PD-L1 checkpoint blockades may also be observed in glioblastoma. Furthermore, neoantigens represent tumour-specific mutant antigens encoded by somatic mutations in the cancer genome. The low neoantigen burden in glioblastoma reduced the chances of the immune system overcoming central tolerance to recognize tumour cells [37]. In addition, some specific gene mutations in glioblastoma induced an immunosuppressive microenvironment through regulating the crosstalk between cytokines and immune cells [14, 33, 38‐46]. The immunosuppressive microenvironment of glioblastoma is composed of a variety of immunosuppressive cells and cytokines. The effective immune cells mainly include CD4+ T cells, CD8+ T cells, NK cells, and tumour-inhibiting M1-TAMs, which are in a state of exhaustion or suppression in the microenvironment. The immunosuppressive cells mainly include Tregs, tumourigenic M2-TAMs, myeloid cells, and MDSCs. Tumour cells express high levels of PD-L1 and IDO, downregulate MHC and costimulatory molecules, express/activate STAT3, cause PTEN loss, then reduce the immunogenicity and induce recruitment of Tregs. Tumour cells secrete MICA/B, IL-10, TGF-β, and HLA-E to recruit Tregs and inhibit both T cell and NK cell activity. Through the secretion of diverse chemokines and other factors, such as CCL2, CSF1, MCP-3, CXCL12, CX3CL1, GDNF, ATP, and GM-CSF, the paracrine network signalling between glioblastoma and the TAMs attracts myeloid cells and infiltrates Tregs. Furthermore, tumour cells secrete immunomodulatory cytokines that polarize TAMs to the immunosuppressive M2 phenotype. Immunosuppressive cells, including M2-TAMs, myeloid cells, and MDSCs, secrete a variety of cytokines (IL-6, IL-10, IL-4Ra, FasL, CCL2, PGE2, EGF, VEGF, and MMP9) to suppress the function of cytotoxic T lymphocytes (CTLs) and promote the progression of tumour cells. In addition, Tregs downregulate IL-2 production, inhibit IFN-γ production, and upregulate T_H2 cytokine secretion to inhibit T cell function [34, 47‐51]. The molecular characteristics of the TME in glioblastoma patients are depicted in Fig. 1. Thus, the exploration of combination therapy based on PD-1/PD-L1 checkpoint blockades is important to the study of glioblastoma.

T cells in the TME normally exhibit an exhausted phenotype with reduced effector function. Negative regulatory signals resulting from the activation of multiple inhibitory checkpoints that block T cells are the primary mechanism leading to effector T cell dysfunction [52]. Studies have shown that inhibitory checkpoints could reverse the exhausted phenotype of effector T cells [53, 54]. Although PD-1 receptor expression is an important factor for the degree of T cell exhaustion, many patients with tumours are still unable to benefit from PD-1/PD-L1 checkpoint blockades. Exhausted T cells in the TME typically express multiple checkpoints, and blockade of a single checkpoint is not sufficient to activate the suppressed immune response [55], this is especially true in glioblastoma, which has a higher degree of immunosuppression in the TME. Therefore, it is urgent to explore a combination treatment regimen of checkpoint blockades and other regimens with a higher response rate. Since PD-1 checkpoint blockades do not significantly benefit patients with relapsed glioblastoma in OS compared with bevacizumab, exploring candidate targets related to the immune response may provide new strategies associated with promoting the clinical efficacy of PD-1/PD-L1 checkpoint blockades.

Several clinical studies have confirmed the predictive effect of PD-L1 expression on the response rate of ICIs in patients with NSCLC, melanoma, colorectal cancer, renal-cell carcinoma, and prostate cancer [95]. Although PD-L1 is highly expressed on GBM [13, 14], the predictive effect of PD-L1 expression on the efficacy of ICIs in glioblastoma remains unclear. Furthermore, the prognostic value of PD-L1 expression in glioblastoma on survival outcomes also demonstrated contradictory results [13, 14, 96, 97]. It can be inferred that the PD-1/PD-L1 signalling pathway does not play a critical role in the development and progression of glioblastoma and may be affected by other factors. Therefore, it is difficult to obtain satisfactory results by simply blocking the PD-1/PD-L1 pathway.

The correlation between MMR protein deficiency or POLE mutations and ideal therapeutic efficacy to PD-1 checkpoint blockades in patients with glioblastoma was first reported in two case reports [98, 99]. The ideal therapeutic effect of PD-1 checkpoint blockades benefits from the patients’ high mutation load. Furthermore, as a tumour-specific neoantigen, the EGFRvIII mutation occurs in 31–50% of patients with glioblastoma, and 37–86% of tumour cells express the mutated protein [100‐102]. EGFRvIII promotes tumour cell growth and invasion and plays a negative prognostic role in glioblastoma patient survival [103‐106]. Considering the high expression rate and oncogenic characteristics of EGFRvIII, it may be an ideal target and biomarker for glioblastoma immunotherapy. In addition to EGFRvIII, IDH1/2 mutations also play an important role in glioblastoma. The incidence of IDH1/2 mutations in primary glioblastoma is approximately 5%, but in recurrent glioblastoma, the incidence is approximately 84.6%. The efficacy of PD-1/PD-L1 checkpoint blockades depends on the effective infiltration of the activated T lymphocytes in tumours. Kohanbash et al. confirmed that the activation mutation of IDH1/2 inhibited the accumulation of effector T cells in glioma tumours and that treatment with IDH1 inhibitors significantly enhanced the infiltration of effector T cells [107]. Thus, the activation mutation of IDH1/2 in glioma provides a new angle to promote the clinical efficacy of PD-1/PD-L1 checkpoint blockades.

In addition to these molecular features, TILs and NK cells are also considered predictors of PD-1/PD-L1 checkpoint blockade immunotherapy efficacy. Although TILs usually represent an exhausted status and show sparse-to-moderate density infiltration in glioblastoma, a certain amount of TILs in the TME is still the basis for the efficacy of checkpoint blockade immunotherapy. Thus, the prognostic and predictive role of TILs requires further exploration. In addition to TILs, NK cells have been shown to play an indispensable role in PD-1/PD-L1 checkpoint blockades. Hsu et al. demonstrated that NK cells were inhibited by PD-1/PD-L1 interactions and recovered with PD-1/PD-L1 checkpoint blockades. This result indicated that NK cells could directly respond to PD-1/PD-L1 checkpoint blockades [108]. In addition to a single molecular marker, Cheng et al. also analysed genetic data from 297 glioblastoma samples from a bioinformatic perspective. They identified 8 genes (FOXO3, IL6, IL10, ZBTB16, CCL18, AIMP1, FCGR2B, and MMP9) with significant prognostic value in glioblastoma. A local immune-related risk signature was adopted to divide patients into two groups: low-risk patients with high expression levels of protective genes (FOXO3, AIMP1, and ZBTB16) and high-risk patients with high expression levels of risky genes (IL6, IL10, CCL18, FCGR2B, and MMP9) [109]. Thus, it is worth exploring which group of patients is more likely to benefit from PD-1/PD-L1 checkpoint blockades. The available studies presenting candidate biomarkers are summarized in Table 2.

PD-1/PD-L1 checkpoint blockades are gradually becoming an effective therapeutic strategy for several types of tumours. Whereas its therapeutic efficacy in glioblastoma remains to be elucidated, several preclinical studies have demonstrated optimal outcomes. The main challenges are as follows: First, there are discrepancies between different genomic subtypes or molecular profiles in the response to PD-1/PD-L1 checkpoint blockades. An analysis of PD-L1 expression in glioblastoma samples has revealed that the MES subtype of glioblastoma has higher PD-L1 expression [13]. In addition to PD-L1, the MES subtype displayed an immunogenic status with gene mutations along with a high neoantigen burden, which increased the response to PD-1/PD-L1 checkpoint blockades. Thus, a computational characterization of the ability of each subtype to respond to PD-1/PD-L1 checkpoint blockades is urgently needed.

Second, there is a challenge in the selection of PD-1/PD-L1 checkpoint blockades for primary versus recurrent glioblastoma. The phase III clinical trial Checkmate 143 reported that PD-1 monoclonal antibody (nivolumab) monotherapy does not significantly improve overall survival time compared with bavacizumab in recurrent glioblastoma patients who were previously treated with chemotherapy and radiotherapy [26]. A study including 22 patients with recurrent glioblastoma also demonstrated no clinical response to pembrolizumab, which targets PD-1 [110]. Then, the efficacy of nivolumab, which targets PD-1, was explored in newly diagnosed glioblastoma patients. Lim et al. [111] assessed the safety and tolerability of nivolumab in combination with RT ± temozolomide (TMZ) in two cohorts. The cohort with TMZ (1c) enrolled 55 patients, including 12 patients with methylated MGMT and 43 patients with unmethylated MGMT. The cohort without TMZ (1d) enrolled 58 patients with unmethylated MGMT. The discontinuation of treatment in the 1c and 1d groups was mostly due to radiographic progression (1c: 50% in the methylated subgroup, 37% in the unmethylated subgroup; 1d: 64%), study drug toxicity (8, 9%; 10%), or patient decision (8, 14%; 0%). The most common (≥30% of patients) AEs were headaches (42, 47%; 41%) and seizures (25, 16%; 31%), which demonstrated that nivolumab was well-tolerated in newly diagnosed glioblastoma patients, and the rate of AEs were consistent with the neurological AE frequency in other reports. Additionally, no drug toxicity-induced deaths were reported. However, the survival data need to be followed up further. All of these data support the continued exploration of nivolumab + RT ± TMZ in newly diagnosed glioblastoma patients in ongoing clinical trials, including Checkmate 498 (NCT02617589) and Checkmate 548 (NCT02667587). From Checkmate 143, we found discrepancies in tolerability and drug toxicity between newly diagnosed glioblastoma patients and recurrent glioblastoma patients. Thus, the clinical outcomes are also worth looking into.

Third, the identification of the optimal combination and sequence for combination therapy is challenging work. Although several preclinical studies have achieved optimal ORR in glioblastoma mouse models with antibodies targeting PD-1/PD-L1, CTLA-4, TIM-3 LAG-3, IDO, or OX-40 [20, 34, 74, 77], there is still a long time period before these strategies are approved for clinical use. In addition, the optimal combination strategy and the sequence of combination therapy for primary glioblastoma versus recurrent glioblastoma also needs to be identified. Considering that different antibodies, which targeted PD-1, CTLA-4, LAG3, etc., and different vaccines triggered different alterations in immune cells and the secretion of key chemotactic factors in the TME, the optimal combination strategy should be able to synergize with PD-1/PD-L1 checkpoint blockades to induce tumour cell immunogenicity and stimulate effective antitumour responses. Furthermore, if clinical trials confirm discrepancies in response rates for PD-1/PD-L1 checkpoint blockades between primary and recurrent glioblastoma tumours, a panel describing the landscape of molecular characteristics of the TME in both types of glioblastoma patients is worth investigating.