Introduction

Glioblastoma (GBM) is the most common malignant primary brain tumor [1]. Despite aggressive treatment consisting of surgery and radiochemotherapy followed by temozolomide [2, 3], the median overall survival (mOS) had been 12–20 months, with a 5-year survival rate of approximately 10% [1, 3]. A recent phase 3 trial demonstrated further mOS improvement to approximately 21 months with a 5-year survival rate of 13% [4]. No standard treatment exists for recurrent disease [5]. To date, targeted agents have not demonstrated survival benefit [6], and therapeutic options for newly diagnosed and recurrent GBM are similar [3, 7, 8].

The interaction between programmed death-1 (PD-1), expressed on tumor-infiltrating immune cells (IC), and programmed death-ligand 1 (PD-L1), expressed on both tumor cells (TC) and IC, inhibits T-cell activation in the tumor microenvironment. Atezolizumab is an engineered, humanized anti-PD-L1 monoclonal antibody that targets PD-L1 and blocks its interaction with PD-1 and B7.1, reinvigorating anti-cancer immunity [9,10,11]. Additionally, direct targeting of PD-L1 with atezolizumab may preserve immune homeostasis in normal tissue by leaving the interaction between PD-1 and programmed death-ligand 2 (PD-L2) intact [12, 13]. The clinical efficacy and safety of atezolizumab have been demonstrated in many cancers [9, 14,15,16], and led to US Food and Drug Administration approval of atezolizumab as second-line therapy for patients with advanced or metastatic non-small cell lung cancer and urothelial cancer [17].

Some GBM tumor cells (TC) and tumor-infiltrating immune cells (IC) express PD-L1 [18], and higher expression is associated with higher grade and worse outcome [19]. Thus, PD-L1 could serve as a rational therapeutic target in GBM [20]. Targeting the PD-L1/PD-1 pathway has demonstrated significant activity in several cancers [21]. Data have shown clinical benefit of immune checkpoint inhibition in some patients with high tumor mutational burden (TMB) [22,23,24], including a small subset of patients with GBM [25, 26]. Several clinical trials have examined or are evaluating anti-PD-L1/PD-1 therapy in GBM [27, 28]. Preliminary results from a phase 3 study of nivolumab in recurrent GBM showed a lack of clinical benefit compared with bevacizumab in biomarker-unselected patients [28].

A phase 1a clinical trial (PCD4989g; NCT01375842) evaluated the safety and tolerability of atezolizumab monotherapy in many tumor types [9, 11]. Here we present the clinical safety and efficacy of atezolizumab in patients from the GBM cohort of this study.

Methods

PCD4989g overview

The phase 1a study followed a standard dose-escalation design and then allowed for tumor-specific expansion cohorts, including one for recurrent GBM (Figure S1).

Study objectives

The primary objective was to evaluate the safety and tolerability of single-agent atezolizumab. Key additional objectives were to assess atezolizumab activity as measured by investigator-assessed best radiographic overall response, progression-free survival (PFS), and OS, and to evaluate exploratory biomarkers.

Patients with GBM

Key eligibility criteria included initial histological confirmation of GBM, radiographic evidence of recurrent disease (first or second recurrence as GBM; if disease had progressed from grade 2 or 3 glioma, additional lines of prior therapy were allowed), measurable disease per Response Assessment in Neuro-Oncology (RANO) criteria [29], and Karnofsky performance status ≥ 70. Patients had to have prior radiation and/or temozolomide treatment. Key exclusion criteria included evidence of recent brain hemorrhage and history or risk of autoimmune disease. At enrollment, patients could be on dexamethasone (or equivalent) ≤ 4 mg daily. After enrollment, steroid dosing was at the investigator’s discretion. All Institutional Review Boards approved the study. All patients gave written informed consent.

Dosing and administration

Atezolizumab was administered intravenously every 3 weeks at 1200 mg until disease progression or unacceptable toxicity. Treatment was discontinued in patients who experienced progressive disease (PD) by RANO criteria and did not meet pre-specified criteria for continued dosing past progression. Patients could continue to receive treatment after progression until loss of clinical benefit (investigator-assessed).

Study assessments

Safety

All patients who received a dose of atezolizumab were assessed for safety. Adverse events (AEs) were recorded until 90 days after the last dose of atezolizumab or until initiation of another cancer therapy. AEs were graded according to National Cancer Institute Common Terminology Criteria for Adverse Events version 4.0. AEs of special interest (AESI) were defined in the protocol as conditions suggestive of autoimmune disorder (see Supplementary Appendix Methods).

Efficacy

The efficacy-evaluable population was defined as patients with measurable disease at baseline per RANO criteria. Patients were evaluated by imaging every 6 weeks. Best overall response and PFS were investigator-assessed. PFS was defined as the time from the start of study drug until PD or death. Best overall response was defined as the best confirmed response recorded from the start of treatment.

Exploratory biomarkers

Immunohistochemistry (IHC) for PD-L1 expression and tumor-infiltrating CD8+ Cells

PD-L1 expression on TC (percentage of PD-L1-positive TC) and IC (percentage of PD-L1–positive IC per tumor area) was centrally evaluated using the VENTANA SP142 PD-L1 IHC assay in archival and fresh tumor specimens (Ventana Medical Systems, Inc, Oro Valley, AZ, USA) [30]. Please see the Supplementary Appendix for more details. All pretreatment samples were archival, with 14 out of 16 collected at diagnosis. The median time between tissue collection and study initiation was 407 days (range 49–1291). Postprogression studies were performed on fresh tumor samples. PD-L1 expression was scored based on the percentage of TC or IC expressing PD-L1 (see Supplementary Appendix Methods). IHC was performed to detect CD8+ T cells in the tumor bed (clone C8/144B [Dako, Santa Clara, CA, USA]) (performed by HistoGeneX, Naperville, IL, USA).

Immune cell quantification in peripheral blood

Lymphocyte subsets were measured using the BD Multitest TBNK Reagent with BD Trucount tubes (BD Biosciences, San Jose, CA, USA). Total B-cell and CD4+ and CD8+ T-cell counts were quantified by flow cytometry (performed by QLAB, [Koninklijke Philips N.V.]).

Tumor mutation analyses

Somatic mutations, microsatellite instability (MSI) status, and TMB were centrally assessed using the FoundationOne DNA-based assay (Foundation Medicine, Inc, Cambridge, MA, USA) [14, 15, 31]. Briefly, after extraction from formalin-fixed paraffin-embedded biopsy or surgical specimens, 50–200 ng of DNA underwent whole-genome shotgun library construction and hybridization-based capture of 4557 exons from 287 cancer-related genes and 47 introns from 19 genes. TMB was defined as the number of somatic, coding, base substitution, and insertion/deletion mutations per megabase of DNA sequenced.

Statistical analysis

The clinical cutoff date was December 31, 2016. Sample size considerations were not made with explicit power and type I error considerations but were made to obtain preliminary safety, pharmacokinetic, and pharmacodynamic information. The Kaplan–Meier method was used to estimate the OS and PFS survival curves, including the median. The Brookmeyer–Crowley method was used to construct 95% confidence intervals for the median OS and PFS.

Study oversight

F. Hoffmann-La Roche, Ltd./Genentech, Inc. sponsored the study, provided study drugs, and collaborated with academic authors regarding study design and the collection, analysis and interpretation of data. All drafts of the manuscript were prepared by the authors, with editorial assistance from a professional medical writer funded by the sponsor. All authors approved submission. All authors verified that the study was conducted per protocol (available with the statistical analysis plan online) and vouched for data accuracy and completeness.

Results

Patients

Sixteen patients were enrolled in the GBM cohort (Table 1); all 16 were evaluated for atezolizumab safety and efficacy. The median age was 52 years (range 31–75), and 13 patients (81%) were men. An equal number of patients had KPS 70–80 and 90–100. Seven patients (44%) were previously treated with systemic steroids; 12 patients (75%) received systemic steroids while on study treatment. Eight patients (50%) had received 1 prior line of therapy, and 8 (50%) had received two. All patients had received prior radiotherapy and temozolomide; 8 (50%) had also received bevacizumab. The median atezolizumab treatment duration was 2 months (Fig. 1a).

Table 1 Patient demographics and baseline characteristics
Full size table
Fig. 1
figure 1

Clinical activity of atezolizumab in patients with glioblastoma. a Duration of treatment and time to response. b Maximum change from baseline in the SLD for target lesions; + 20% and − 30% are marked by dashed lines. c Kaplan–Meier estimate of overall survival. NE not evaluable, PD progressive disease, PR partial response, SD stable disease, SLD sum of the largest diameters

Full size image

Safety

All 16 patients experienced ≥ 1 AE; 10 patients (63%) had a treatment-related AE (TRAE; Table 2A). Three patients (19%) had grade 3 TRAEs [asthenia, increased aspartate aminotransferase (AST) and brain edema]. One patient (6%) with PD had a serious TRAE (grade 3 brain edema), which was treated with an increased dose of steroids but did not lead to atezolizumab discontinuation. One patient (6%) interrupted atezolizumab due to treatment-related grade 3 asthenia (Table 2A). No grade 4–5 TRAEs were reported. All patients discontinued atezolizumab treatment due to PD. All deaths occurred due to PD or during follow-up.

Table 2 Safety summary and treatment-related AEs
Full size table

Consistent with what is typically observed in patients with primary brain tumors [32], the most common all-grade TRAE was fatigue (25%), followed by malaise (19%) (Table 2B). Six patients (38%) experienced an AESI; 3 of these patients (19%) had grade 3 AESIs: 2 (13%) had increased alanine aminotransferase, and 1 (6%) had increased AST (Table S1).

Efficacy

The overall response rate (ORR) was 6% (Fig. 1b). After 165 days on study, 1 patient (6%) developed a partial response (PR), with a duration of 5.3 months and a maximum 69% reduction in the sum of the largest diameters (SLD) from baseline (Fig. 1b, Figure S2). Three patients (19%) had stable disease (SD). An additional patient had a 63% reduction in SLD from baseline, but was classified as having PD due to the appearance of new lesions.

The median PFS (mPFS) was 1.2 months (range 0.7–10.7 months) (Figure S3). The mOS was 4.2 months (range 1.2 to 18.8+ months) (Fig. 1c). The landmark 12-month survival rate was 21% (95% CI 0.3–42.6). Longer-term survival (range 16.0 to 18.8+ months) was observed in three patients. At data cutoff, all patients had discontinued atezolizumab treatment, but two patients were alive; 1 had SD as best response, the other had PD.

Exploratory biomarkers

We evaluated the relationship between baseline peripheral lymphocyte levels, systemic steroid use, and outcome. Patients who were taking systemic steroids at the time of study drug initiation had fewer baseline B cells and CD4+ T cells than those who did not, while no association was observed for CD8+ T cells (Fig. 2a). Patients with lymphopenic baseline CD4+ T cells and B cells (< 0.40 × 106 cells/mL and < 0.19 × 106 cells/mL, respectively) showed a trend toward reduced PFS and OS (Fig. 2b).

Fig. 2
figure 2

Association between biomarkers and clinical outcome. a Quantitation of peripheral lymphocyte levels at baseline in populations based on systemic steroid use. The line in the middle of the box is plotted at the median. b Kaplan–Meier estimates of overall survival and progression-free survival in populations defined by baseline lymphocyte levels. c Kaplan–Meier estimate of overall survival in populations defined by IDH1 mutation status. d Kaplan–Meier estimate of progression-free survival in populations defined by IDH1 mutation status. HR hazard ratio, LLN lower limit of normal, NE not evaluable

Full size image

A non-significant trend toward improved OS was observed in patients who were not taking steroids at the time of treatment initiation (mOS, 5.0 months), versus those who were on steroids (mOS, 3.9 months; Figure S4). No difference was seen in PFS between patients with and without steroid treatment at baseline (Figure S4). A non-significant trend toward shorter PFS and OS was observed in patients who had received prior bevacizumab (mOS, 3.6 months; mPFS, 1.0 months) versus those who did not receive prior bevacizumab (mOS, 5.0 months; mPFS, 2.4 months; Figure S5).

Most patients had low or no PD-L1 expression on TC and IC (Table 1). Of note, levels of tumor-infiltrating CD8+ T cells were low (mean 0.19% cells/center of the tumor) compared with tumor types known to have high levels of immune infiltration [9, 14]. Postprogression tumor tissue was collected from the patient who experienced a PR and one patient who had PD as best response. In both, tumor-infiltrating CD8+ T cells were similarly low in pretreatment and postprogression samples (PR patient, 0.06% and 0.02%; PD patient, 0.04% and 0.05%; respectively). No notable changes in PD-L1 IHC expression on TC or IC were observed in archival versus post-progression samples.

In GBM, IDH1/2 mutations and MGMT promoter methylation are associated with increased survival and benefit from standard treatments [33]. Twelve patients had tumor tissue available for genomic analyses. Three patients had R132H mutations in IDH1 (1 PR, 2 SD; Table 3); none had IDH2 mutations. Patients with IDH1-mutant tumors had better PFS with a trend toward longer OS than patients with IDH1-wild-type tumors (mPFS, 5.5 versus 1.2 months; mOS, 16.0 versus 2.7 months; Fig. 2c, d). Of the patients with IDH1-wild-type tumors, 8 experienced PD and 1 was not evaluable due to death prior to the first tumor assessment (Fig. 1a).

Table 3 Exploratory biomarkers and overall survival
Full size table

All tumors evaluated were microsatellite stable (MSS). In the 12 patients assessed, the median TMB was 2.7 mutations per megabase (Mut/Mb) (95% CI 1.80–4.51), with 1 outlier (183 Mut/Mb). This patient’s tumor was IDH1-wild-type but had a deleterious somatic mutation (L424V) in the DNA polymerase gene POLE (Table 3), which was previously identified in hypermutated tumors [34, 35]. Despite a reduction of target lesions (63% decrease in SLD; Fig. 1b) and prolonged survival (17.7+ months), this patient had confirmed PD due to new lesions.

Discussion

In this phase 1a study, atezolizumab monotherapy was well tolerated in patients with recurrent GBM. No unexpected AEs were reported. Durable radiographic response and SD were observed in a subset of patients with GBM, with 1 patient experiencing a sustained PR and 1 exhibiting prolonged SD. The landmark 12-month OS rate (21%) was comparable to that seen with bevacizumab or cytotoxic chemotherapies [7, 36, 37].

GBM is a historically difficult disease to treat, with few advances in treatment options in the last 30 years [3, 6]. Bevacizumab-refractory patients are particularly resistant to subsequent therapies and were included in this clinical trial as additional therapeutic options are needed. Furthermore, inhibition of VEGF signaling has led to enhanced antitumor immune responses in preclinical models [38], suggesting that patients previously treated with bevacizumab may derive benefit from immune checkpoint inhibitors. Recently, immunotherapy has improved survival in patients with a variety of cancer types [21], however a number of potential issues may limit its efficacy in GBM. The first is the highly immune-suppressed tumor microenvironment, influenced by both tumor intrinsic factors such as PD-L1 expression, as well as iatrogenic factors such as steroid administration. Second, GBM has relatively low TMB, differentiating it from malignancies with higher TMB that have shown sustained responses to immunotherapy. Third, the challenge in achieving adequate doses of the therapeutic agent in the target organ. Fourth is difficulty interpreting imaging studies. Finally, there is concern for CNS toxicity from either increased intracranial pressure due to a robust immune response or a misdirected immune-mediated injury.

PD-L1 is frequently expressed on tumor infiltrating lymphocytes, and the majority of GBM tumor cells have at least some PD-L1 expression [39]. These findings, cumulatively, suggest that targeting PD-L1 may provide clinical benefit. This study used the VENTANA SP142 PD-L1 IHC assay, designed to maximize detection of PD-L1 on IC. In a recent, small lung cancer study (N = 37), the SP142 assay detected fewer PD-L1-expressing TC compared with other PD-L1 IHC assays [40]. A prior study reported that PD-L1 expression was detected on diffuse/fibrillary cells (84%) and on TC (38%) in glioblastoma [18]. The difference in PD-L1 prevalence between that study and ours may be, in part, due to different IHC assays.

Patients who were taking systemic steroids at the time of study drug initiation had lower circulating lymphocyte levels than those who were not, consistent with the known mechanism of action of steroids [41, 42], and trended toward reduced clinical benefit from atezolizumab. The three patients who had long-term survival, including 1 who experienced a PR, were not receiving steroids at baseline and did not require steroids during study treatment. This may be secondary to the influence of steroids on anti-tumor immune activity within the context of an immune checkpoint inhibitor. Alternatively, either the steroid requirement may merely be an independent prognostic factor or the need for steroids may be reduced in patients with more indolent, IDH-mutant tumors (two of three long-term survivors). Patients with normal levels of peripheral CD4+ T cells and B cells had a trend toward improved outcome versus those who had lymphopenic levels. Higher CD4+ T-cell count has been associated with longer survival in GBM [43, 44], suggesting that this biomarker may be prognostic. Patients with IDH1-mutant tumors had improved outcomes versus those with IDH1-wild-type tumors, consistent with previous studies showing that IDH1-mutant gliomas represent a distinct disease entity [45]. Preclinical models have demonstrated that IDH1 mutation in gliomas is associated with suppression of CD8+ T cell accumulation, contributing to the immunosuppressive tumor microenvironment [46]. A randomized controlled trial is required to determine whether peripheral lymphocyte levels and IDH1 mutations are merely prognostic or also predictive of atezolizumab activity in GBM.

Tumors with high MSI have increased TMB and improved responses to PD-L1/PD-1 pathway inhibitors versus MSS tumors [47]. Mutations in DNA mismatch repair (MMR) genes developed in preclinical glioblastoma models treated with temozolomide [48] and in patients with low-grade gliomas that have progressed on temozolomide [49]. Sequencing of MMR genes did not show any evidence of MMR deficiency in patients enrolled in the GBM cohort of this study. In addition, mutations in DNA polymerase components are associated with a hypermutated phenotype. POLE is the catalytic subunit of the DNA polymerase complex, and POLE defects lead to loss of DNA replication fidelity. In a recent study, germline susceptibility variants in POLE, including L424V, were identified in hypermutated MSS colorectal cancers [34]. This mutation was also previously described in a patient with hypermutated GBM who displayed evidence of a clinical and immunologic response while on pembrolizumab [35]. POLE defects may lead to increased neoantigen burden and higher anti-tumor T-cell recognition [50].

In the patient cohort described here, all patients with recurrent GBM had MSS tumors. One patient who experienced prolonged survival on atezolizumab had a POLE L424V-mutant tumor with high TMB. This patient’s response was categorized as PD due to the emergence of new lesions; however, his long-term survival and the stabilization of these lesions suggest that this may have been pseudoprogression rather than true PD. Additional tissue was not obtained to confirm pseudoprogression. Evaluation of TMB and genomic alterations that confer a hypermutation phenotype, including POLE mutations, may help guide selection of patients with recurrent GBM who are more likely to receive benefit from atezolizumab.

Interpreting radiographic endpoints in patients with GBM has been challenging. Pseudoprogression, a transient worsening of enhancement on imaging, frequently occurs in patients treated with radiation and traditional cytotoxic chemotherapy [51]. With immunotherapeutic approaches, there is substantial concern that transient radiographic worsening can be seen before radiographic improvement. It is thought that the pathophysiology driving these radiographic changes differs from that observed with chemoradiotherapy. There is also concern that radiographic findings may be independent of survival outcomes in patients treated with immunotherapy. A contemporary assessment system was established to guide imaging interpretation, particularly within the context of clinical trials. The immunotherapy Response Assessment in Neuro-Oncology (iRANO) [52] is derived from RANO [29]. A key distinguishing feature of iRANO for the determination of PD (even in the setting of new lesions) is required confirmatory radiographic assessment in 3 months, among clinically stable patients who received treatment within 6 months.

In conclusion, this phase 1a study showed that atezolizumab was well tolerated in patients with GBM. Recent results have demonstrated lack of clinical efficacy with anti-PD-1 antibody monotherapy in biomarker-unselected patients with recurrent GBM [28]. The favorable safety profile supports combination studies with atezolizumab in this population, particularly in patients who may not require concomitant steroids. Potential options for combination studies with atezolizumab in recurrent GBM include radiation therapy, chemotherapy (temozolomide, lomustine), other immunotherapies, and bevacizumab. Each one targets a different part of the antitumor immune response and, in combination with atezolizumab, may help reinvigorate anticancer immunity in recurrent GBM.