This single-center, open-label phase I clinical trial was conducted at the Universitair Ziekenhuis Brussel (UZ Brussel, Brussels, Belgium).

Patients were eligible if they were aged ≥18 years and had a recurrence of a previously histologically confirmed glioblastoma (or a lower grade glioma that had transformed to a WHO grade 4 glioma), and that was amenable for a gross-total resection (with an acceptable risk for postoperative neurological deficits).

Progression of disease needed to be documented following prior treatment that at least included radiation therapy and temozolomide chemotherapy. At baseline, a measurable tumor recurrence on gadolinium-enhanced T1 magnetic resonance imaging (MRI) and on 18-fluoroethyl-L-tyrosine positron-emission tomography/computed tomography (¹⁸F-FET-PET/CT) was required.

Patients requiring oral corticosteroids exceeding a maximal daily equivalent dose of methylprednisolone (MPS) ≥8 mg or dexamethasone ≥1.5 mg per day for at least 7 days prior to enrolment were considered ineligible. Eastern Cooperative Oncology Group (ECOG) Performance Status had to be ≤2 and adequate hepatic, renal and bone marrow function was required. Patients were excluded if they had received prior immunotherapy with an anti-PD-1, -PD-L1, or -CTLA-4 mAb or any other drug specifically targeting immune checkpoints or if they were treated with immunosuppressive drugs (eg, in case of organ transplant) with the exception of topical steroids or oral steroids at the maximum daily dose mentioned previously in this paragraph. Pre-existing active autoimmune disease, prior immunodeficiency syndromes, persisting toxicities from prior therapies, diagnosis of any other malignancy within 5 years prior to enrolment, bleeding/thrombotic disorders or problematic wound healing also excluded the patient from participation. All participants provided written informed consent.

Twenty-four hours prior to surgical intervention, 10 mg of nivolumab (Opdivo®, Bristol Myers Squibb (BMS)) was administered by intravenous infusion over 15 min. Surgical resection was performed under 5-Amino-Levulinic-Acid (5-ALA) fluorescence. At the end of the surgical intervention, after careful hemostasis was achieved, the brain tissue lining the resection cavity was injected manually (by multiple 100 µL administrations) with 2 mL of ipilimumab (Yervoy®, BMS, 10 mg) in cohort-1, and 1 mL of nivolumab (10 mg) plus 1 mL of ipilimumab (5 mg) in cohort-2. The 20 injections were evenly distributed to cover the entire resection cavity wall but avoiding any nearby functional areas; tractography of the white matter tracts loaded into the navigation system was used to evaluate this. A tuberculin needle inserted to a depth of 3–5 mm was used, and the goal on injection was to obtain a slight swelling of the tissue and no evacuation of the compound into the resection cavity. Thereafter, intravenous administrations of nivolumab (at a fixed dose of 10 mg) were repeated every 2 weeks for a total maximum of five administrations. Treatment was discontinued in case of confirmed progression of disease, unacceptable toxicity or patient refusal to continue study treatment. Patients could continue study treatment following the first documentation of disease progression if the investigator considered this to be in the best interest of the patient.

Throughout the study, patients were evaluated every 2 weeks with a clinical examination and blood analysis. Gadolinium-enhanced MRI was performed on the first day postoperatively and every 6 weeks thereafter. Follow-up imaging with ¹⁸F-FET-PET/CT was scheduled as clinically indicated to complement MRI results. Tumor responses and progression of disease were defined according to the immunotherapy response assessment for neuro-oncology criteria.

Tumor tissue was collected from all patients at the time of their surgical intervention. H&E staining and immunohistochemistry (IHC) analyses were performed using formalin-fixed, paraffin-embedded (FFPE) tumor tissue. CD8 SP57 (Roche, Basel, Switzerland) and PD-L1 22C3 (Agilent, CA, USA) antibodies were used for biomarker analysis. Tumor PD-L1 expression was scored by a pathologist for the frequency of tumor cells with membrane staining of PD-L1 (ie, patients with PD-L1 expression levels ≥1% were defined as those with ≥1 PD-L1-positive tumor cells within a field of 100 evaluable cells). For the purpose of IHC staining for B7-H3, FFPE sections were deparaffinized and stained with a validated chromogenic IHC assay based on an anti-B7-H3 primary Rabbit monoclonal antibody (SP-265, Spring Biosciences). The detection was done by a Dako EnVision+System HRP Labeled Polymer Anti-Rabbit on a DAKO Autostainer Link 48. MGMT promoter methylation analysis was performed by MethyLight as previously described.36 37 Next-generation sequencing (NGS) to reveal somatic mutations was performed on FFPE-extracted-DNA using an accredited in-house capture-based comprehensive gene panel (ACVR1, ATRX, BRAF, CDKN2A, CTNNB1, EGFR, ERBB3, H3F3A, H3F3B, HIST1H3B, HIST1H3C, IDH1, IDH2, MET, NF1, NTRK1, NTRK2, NTRK3, PDGFRA, PIK3CA, PTEN, SMO, TERT, TP53) on Illumina NovaSeq6000. A validated homebrew bioinformatics pipeline provided variants that were biologically and clinically classified according to the Belgian ComPerMed (v1) guidelines.38

Total RNA was isolated from FFPE-pretreated tumor biopsies. Initially, tumor samples from 21 patients were selected; total RNA extraction was sufficient for 18 tumor samples. Gene expression profiling (GEP) of 770 genes was performed using the NanoString PanCancer IO 360 Panel (NanoString Technologies) according to the manufacturer’s instructions. Released tags were quantified in a standard nCounter analysis system. Biological signatures and scores based on the GEP were provided by NanoString Technologies.

The primary objectives of the trial were to establish safety and feasibility of the experimental treatment regimen. Primary endpoints were TRAE according to the National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events (CTCAE) version 5 and treatment disposition for nivolumab and ipilimumab.

Secondary endpoint was OS analyzed by Kaplan-Meier estimator.

Initially, the sample size for cohort-1 was determined according to a classical phase I 3+3 design. Following the first three patients treated in cohort 2, the trial was amended to initially expand this cohort to 12 patients and by a second amendment to recruit a final of 24 patients. Statistical analyses were performed using SPSS Statistics V.26.0 (IBM, Chicago, USA).

Between November 16, 2016 and August 8, 2019, 27 patients were enrolled and initiated study treatment, 3 and 24 patients in cohort-1 and cohort-2, respectively. The median age was 55 years (range 38–74). Most patients (81%) were initially diagnosed with a WHO grade 4 glioma (glioblastoma) and (85%) had been treated at first diagnosis by a surgical intervention followed by adjuvant radiation therapy (30×2 Gy) with concomitant and adjuvant temozolomide chemotherapy. Six patients had progressed on a prior systemic chemotherapy administered at first recurrence. At baseline, most patients (89%) had an ECOG PS of 0–1 without need for high doses of corticosteroids (<8 mg of MPS per day); a waver was given to four patients who were treated with more than 8 mg MPS daily (16 mg/day in one patient, 2×16 mg/day in two patients, and 2×32 mg/day in one patient) at the time of enrolment because it was anticipated that resection of the recurrent glioblastoma would alleviate the need for corticosteroid treatment. The majority of gliomas were known to be IDH1 wild type (n=25; 93%). No loss of heterozygosity (LOH) for 1 p/19q was found in the eight gliomas analyzed (LOH for 1 p/19q was not analyzed in 19 patients (70%)). All patients presented with a progressive gadolinium-enhancing tumor mass that was also visualized by ¹⁸F-FET-PET/CT of the brain. Baseline patient characteristics and prior therapies are summarized in table 1.

All patients received a 10 mg dose of intravenous nivolumab 24 hours before the surgical intervention during which they underwent resection of the gadolinium-enhanced tumor mass guided by 5-ALA fluorescence. Gross total resection was achieved in 10 of 27 patients (37%), 17 patients (63%) underwent a subtotal resection (figure 1). All patients received the predefined dose of intracerebral ipilimumab (cohort-1), or ipilimumab plus nivolumab (cohort-2), administered within the brain tissue lining the resection cavity at the end of the surgical intervention.

Thirteen patients (including all three patients treated in cohort-1; and 10 patients in cohort-2 (41%)) received all five planned postoperative intravenous doses of nivolumab. In 14 patients (52%) from cohort-2, study treatment was stopped at an earlier point because of symptomatic tumor progression; in these patients, a median number of 3 (range 1–4) intravenous doses were administered (figure 1 and online supplemental table 1).

The peroperative injections of ipilimumab (cohort-1), or ipilimumab plus nivolumab (cohort-2) could be performed as planned without occurrence of any peroperative complications.

Postoperative fever (n=14; grade 1 (n=12; including all three patients from cohort-1), and grade 2 (n=2)) was the only adverse event (AE) that was suspected of being related to the administration of ipilimumab and nivolumab. There was a low incidence of clinically significant grade 1/2 AE that were suspected to be immune-related: pruritus and rash (n=7, and n=4 respectively), hypothyroidism (n=2), a sarcoid-like reaction (n=1), and fatigue (n=17). These AE did not require systemic corticosteroid treatment or other immunosuppressive therapy. The only grade 5 AE was an infectious pneumonia that was considered not to be related to study treatment. All-cause AE occurring in more than two patients as well as AE of special interest are listed in online supplemental table 2. Lymphopenia was the most frequent abnormal blood value during study treatment and was observed in 25 of 27 patients (>grade 3, n=10) (online supplemental table 3). In all patients, lymphopenia was present at baseline and was not considered to be related to the study treatment.

Confirmed tumor progression was documented in all except one male patient from cohort-2 (who remained free from progression 158 weeks after initiating study treatment). The initial postoperative evolution of this patient was characterized by a pseudoprogression on MRI of the brain (figure 2).

At the time of radiological progression, a total of three study patients underwent a new neurosurgical resection of the suspected lesion. In a first female patient, a growing cystic lesion located at the resection site was surgically explored. Anatomopathological analysis of the removed inflammatory membrane lining the cavity as well as biopsies taken of the surrounding brain tissue revealed no glioblastoma tumor cells, instead infiltration by inflammatory cells mainly composed of lymphocytes was found (figure 3). In an additional female patient from cohort-2 who underwent a complete resection of a suspected tumor recurrence 60 weeks after the initiation of study treatment, the resected tissue revealed no evidence of recurrent glioblastoma, but post-treatment changes characterized by inflammatory changes enriched with lymphocytes and histiocytes as well as siderophages with multinuclear giant cells (figure 3). This patient subsequently resumed immunotherapy and remains free from progression 98 weeks after initiating study treatment. In a third, male patient, the resected tissue was found to be composed of glioblastoma with focal infiltration by lymphocytes and histiocytes.

At the time of data-base lock, 21 patients (78%) have died (all caused by glioblastoma progression). The median follow-up of the six (22%) surviving patients is 108 weeks (range 67–158). The median PFS for the total study population was 11.7 weeks (95% CI: 10 to 12; range 2–152) (figure 4A). Median OS is 38 weeks (95% CI: 27 to 49) with a 6-month, 1-year, and 2-year OS-rate of, respectively, 74.1% (95% CI: 57 to 90), 40.7% (95% CI: 22 to 59), and 27% (95% CI: 9 to 44) (figure 4B). For the purpose of comparing survival with a historical control group, a cohort of 469 Belgian patients with recurrent glioblastoma who were treated in three prospective phase II clinical trials (investigating axitinib, avelumab, and lomustine as investigational drugs), and a medical need program for bevacizumab was used.39–42 No significant difference was found with respect to PFS (descriptive Log-Rank test p>0.05). OS compared favorably with the OS of the historical control group (n=469), with a marked tendency for superior OS in patients with the longest survival and improved 1-year, 2-year, and 3-year survival estimates (descriptive Log-Rank test p>0.003). Six-months, 1-year, and 2-year OS-rates in this historical control group were, respectively, 50.2% (95% CI: 46 to 55), 18.3% (95% CI: 15 to 22), and 5.3% (95% CI: 3 to 8). When compared with the OS of a subpopulation of the historical control cohort composed of patients that were not in need of corticosteroid treatment (n=27; cohort-1 of the GliAvAx study), our study population also had a significantly better OS (descriptive Log-Rank p>0.048).42

All resected glioma tumor tissues were graded WHO grade 4, at the exception of one patient in whom, despite the MRI features of a high-grade gadolinium-enhancing glioma recurrence, the tumor biopsy revealed a grade 2 glioma (this patient nevertheless had a relatively short PFS of 22 weeks).

NGS analysis of DNA extracted from the resected tumor tissue detected pathogenic mutations in 20 patients (74%); 12 (44%) had a TERT c.-124C>T mutation, 9 (33%) a TP53 mutation, 6 (22%) a NF1 mutation, and 2 (7%) an IDH1 R132H mutation (table 2).

None of the recurrent glioblastomas with a IDH1 mutation was characterized by a 1p19q co-deletion. Seven out of the 20 glioblastomas (35%) that could be analyzed were characterized by MGMT-promotor hypermethylation. No significant correlation was found between the somatic mutation genotype, or MGMT-promotor methylation status and survival.

The immunophenotype was assessed by PD-L1 (n=27), and CD8 score (n=27), and NanoString IO 360 GEP scores (n=18; figure 5A). Immunohistochemical analysis has shown PD-L1 expression in seven patients. We did not observe a correlation between neither PD-L1 IHC and PD-L1 GEP score nor CD8 IHC score and CD8 GEP score.

Significant correlations between OS and the GEP score for NOS2, B7-H3, proliferation, CD45, JAK/STAT-loss, and immunoproteasome-score were found in univariate analysis. In a multivariate analysis, a higher tumor mRNA expression level of B7-H3 was the only GEP score associated with a significantly worse survival (multivariate Cox logistic regression, p>0.029)(figure 4C). B7-H3 gene expression correlated well with immunohistochemical staining intensity of glioblastoma cells for B7-H3 (figure 5).