Window-of-opportunity clinical trial of pembrolizumab in patients with recurrent glioblastoma reveals predominance of immune-suppressive macrophages

Study Design and Participants

NCT02337686 was an open-label, single-arm exploratory window-of-opportunity phase II clinical trial in patients with GBM at first or second recurrence who required reoperation for clinical indications, conducted at The University of Texas MD Anderson Cancer Center (MD Anderson). The patients had to be at least 18 years old, have histologically confirmed GBM, a Karnofsky performance scale score ≥60, prior radiation therapy and/or chemotherapy, and MRI evidence of recurrence as defined as progressive or new contrast enhancement after initial therapy of irradiation plus concurrent and adjuvant temozolomide, with a minimal volume of 1 cm³ of contrast enhancement. The patients were permitted to receive a stable dose of dexamethasone of no more than 2 mg/day at trial enrollment. Approval for the study was granted through the institutional review board at MD Anderson, and all patients provided written informed consent before study entry. This trial was performed according to the Declaration of Helsinki principles.

Procedures

Patients with recurrent GBM for whom reoperation was indicated for tumor progression were treated with up to 2 doses of pembrolizumab on day −21 and day −1 prior to surgery on day 0. During the presurgery phase, if after one dose of pembrolizumab, a patient experienced progressive clinical symptoms, then the second dose of pembrolizumab was deferred, and the patient proceeded to surgery (at a date before day 0). After recovery from surgery (~2–3 wk), patients received pembrolizumab until disease progression occurred or unacceptable toxicities developed. Treatment (200 mg) was administered every 3 weeks (Fig. 1A). Patients underwent imaging prior to treatment and approximately 24–28 days after the first dose of pembrolizumab (within 24–28 hours of surgical resection) for surgical navigation purposes and within 24–48 hours after surgical resection to assess the extent of tumor resection. Responses to treatment were assessed by MRI scans obtained every 2–3 months thereafter. Treatment continued until disease progression was seen or the 24 months of pembrolizumab administration ended.

Analysis of Cytometry by Time-of-Flight

Tumor tissue from the surgical resection and blood from the peripheral circulation were evaluated for immunological pharmacodynamic effects, including alterations in immune checkpoint ligand expression and changes in the immune cell populations (Supplementary Table 1). Postresection GBM tissue was dissociated into single cell suspensions and the lineage marker CD45 was used to gate on leukocytes. Clustering analysis was performed using the MatLab implementation of the PhenoGraph algorithm.²¹ Clusters were identified using PhenoGraph on all the samples in the space formed by these principal components, with the parameter k for the number of nearest neighbors selected uniquely for each sample using the formula k = minimum (0.002* number of cells, 10). PhenoGraph was run with parameter k = 30, resulting in 18 clusters in the CD45+ analysis of the tumors, and 22 clusters in the analysis of the blood.

Analysis by Immunohistochemistry and Multiplex Immunofluorescence

Formalin-fixed, paraffin-embedded GBM specimens were transversely sliced into 4 μm sections. Sections were stained with mouse anti-human monoclonal antibodies against CD4 (Novocastra, CD4-368-L-A), CD8 (Thermo Scientific, MS-457-S), CD45RO (Novocastra, PA0146), granzyme B (Leica Microsystems, PA0291), inducible co-stimulator (ICOS) (Spring Bioscience, M3980), CD68 (Dako, M0876), CD20 (Dako, M0755), PD-L1 (Cell Signaling, 13684S), PD-1 (Abcam, ab137132), and FoxP3 (BioLegend, 320102).

For multiplex immunofluorescence (mIF) analysis, we followed the Opal protocol staining method²² using the following markers: CD68 (Dako, cat# M0755) with subsequent visualization using fluorescein Cy5 and CD163 (Leica Microsystems, cat# NCL-L-CD163) with subsequent visualization using fluorescein Cy3. Nuclei were visualized with DAPI (4′,6′-diamidino-2-phenylindole). For each marker, the mean fluorescent intensity per sample was then determined as a base point from which positive categorization could be established. For multispectral analysis, each of the individually stained sections was used to establish the spectral library of the fluorophores.

Outcomes

A Bayesian design was used to conduct a single-arm phase II clinical trial, with an interim analysis performed after 10 patients were enrolled. Accrual would be halted based on the co-primary endpoints of 6-month progression-free survival (PFS6) of 15% and an effector:Treg ratio measured at the time of surgery as >5%. Given the sample size of 20 patients and a significance level of 10%, we had 74% power to detect an improvement of PFS6 from 15% (ie, the historical PFS rate) to 35% in a two-sided test. Because of the challenges in obtaining tumor tissue to analyze the co-primary endpoint of intratumoral immune responses, additional subjects beyond the 10 were required. Although the PFS6 rate was 40%, and there were increases in granzyme B effector T cells in the setting of pembrolizumab, there was no commensurate decrease in Tregs. Thus, the trial was halted secondary to the second co-primary endpoint not being reached. Patients who were alive without disease progression or relapse were censored at the time of last contact.

Imaging Protocol and Imaging Response Assessment Criteria

Imaging response assessment was performed on the MRI scan obtained before the first dose of pembrolizumab and after the second dose (immediately prior to surgery). The MRI scan performed immediately after surgery (<24–48 hours) was considered the new baseline scan, as per criteria of Immunotherapy Response Assessment of Neuro-Oncology (iRANO), and used to determine the patients’ best response.²³

Statistical Analysis

All analyses were performed on the intention-to-treat population. Median PFS and overall survival (OS) times from the first administration of pembrolizumab were estimated by Kaplan–Meier analyses. Adverse events were recorded and tabulated according to type and grade. Mean biomarker expression values were compared using the Mann–Whitney test. All computations were carried out in SAS v9.4. For cytometry by time-of-flight (CyTOF) and immunohistochemistry (IHC), GraphPad Prism was used and P-values on paired samples were computed using Student’s t-test.

Additional details can be found in the Supplementary Materials and Methods.

Imaging Response

A total of 15 patients had baseline MRI (pretreatment with pembrolizumab), of whom 14 had follow-up MRI for response assessment. The patient who had one infusion and no surgery was excluded from the analysis. Thus, a total of 14 patients had a baseline scan after surgery, which met iRANO criteria for assessment of response. All 14 patients underwent subtotal GBM resection, and all patients had measurable disease (bidimensional measurements >1) per our criteria. Based on iRANO criteria, 10 patients had progressive disease, 3 had a partial response, and 1 had stable disease (Supplementary Table 5).

GBM Tumors Are Poorly Infiltrated with T Cells but Are Enriched in CD68+ Macrophages

The ability to conduct immune analysis directly from a surgical GBM specimen depends on multiple surgical factors and as such, only a subset of GBM patients treated with pembrolizumab had tumor available for analysis (n = 12) (Supplementary Table 1). GBM samples from patients with recurrent GBM who were treated with pembrolizumab prior to surgical resection and for whom adequate fresh tissue was available (n = 7) were analyzed using CyTOF. We excluded 2 patients from the final analysis due to the cell numbers being below our cutoff point of >600 cells (Supplementary Table 1). All the live singlet cells obtained from pembrolizumab-treated tumors were analyzed using CyTOF, and the fraction of CD45+ leukocytes formed a small proportion of these (7.45 ± 6%) (Fig. 2A and B). Among the total number of live cells, the frequency of CD3+ T cells (including CD4+ and CD8+ T cells) was significantly lower than that of the CD68+ macrophage cell population (Fig. 2B). An analysis of the leukocyte population was performed by gating on CD45+ cells (Fig. 2C). Consistent with the finding within the total number of live cells for the CD45+ population, the majority of clusters were CD68+ macrophage cells, whereas CD3+ T cells were less abundant (Fig. 2C). As expected, major histocompatibility complex (MHC) expression (ie, human leukocyte antigen D related [HLA-DR]) was associated with the CD68+ compartment (Fig. 2D, compare CD68+ with HLA-DR). The immunosuppressive molecules PD-L1, V-domain immunoglobulin suppressor of T-cell activation (VISTA), and B7-H3 showed a strong association with the CD68+ macrophage compartment (Fig. 2D), and they have all been shown to attenuate immune responses in cancer.^24–26 Surprisingly, PD-1 was expressed at low levels within the CD4 and CD8 T-cell compartments, but at higher levels in the CD68+ subset (Fig. 2D).

To get a comprehensive overall view of the immune cells comprised by the tumor microenvironment in GBM, we analyzed the CD45+ leukocyte population using PhenoGraph. We identified 18 immune clusters in the tumors resected from GBM patients after pembrolizumab treatment (Fig. 2E and F, Table 2). As shown in Fig. 2E, the most common immune population was that of CD68+ macrophages, constituting a total of 11 clusters out of 18, which is equivalent to 72.58% of the immune cells in the tumor (11 clusters: #1, 2, 3, 5, 6, 8, 9, 10, 12, 15, and 16). The second most abundant population was that of the CD3+ T cells (22.57%) divided into CD3+CD4+ (2 clusters: #7 and 13; 9.7%), CD3+CD8+ (2 clusters: #4 and 14; 11.37%), and CD3+CD4−CD8− (1 cluster: #17; 1.5%). Lastly, 1 natural killer (NK) cluster expressing CD56 (#11) constituted 4.59% of the immune cells in the tumor.

The CD68+ macrophage clusters show a preponderance of immunosuppressive markers. Three of the 11 CD68+ macrophage cell clusters coexpress PD-L1, VISTA, and B7-H3 (#2, 10, and 15). These are 16.82% of the total leukocytes and ~20% of CD68+ cells. In addition, 3 other clusters coexpress VISTA and B7-H3 (#6, 8, and 16), and 1 cluster expresses PD-L1 (#9). All of these clusters have the potential to be immunosuppressive (Fig. 2E and F, Table 2).

Within the low frequency of CD3+ T cells infiltrating the GBM, we observed the immunosuppressive Treg cluster #13 (~3%), which expresses CD27, ICOS, B and T lymphocyte associated (BTLA), cytotoxic T lymphocyte antigen 4 (CTLA-4), and FoxP3 (Fig. 2F, Table 2). We also observed the CD8+ cluster #14 (~2.8%) expressing CD3, CD8, CD27, and low levels of FoxP3 and CD56, indicating that these cells may be activated and are early in their effector differentiation state (Fig. 2F, Table 2).

CyTOF Analysis of Immune Subsets in Peripheral Blood Shows No Major Changes in Response to Pembrolizumab

Five GBM patients had sufficient peripheral blood mononuclear cells (PBMCs) available before initiation of pembrolizumab administration, during treatment, and upon radiographically determined tumor recurrence for analysis with CyTOF (Supplementary Table 1). There were no significant changes in the frequency of T cells, monocytic cells, and NK cells (CD3+, CD4+, CD8+, CD68+, or CD56+) in the PBMCs before pembrolizumab treatment, during pembrolizumab treatment, or upon tumor recurrence (Supplementary Figure 1). Based on CyTOF analysis of live CD45+ single cells, 22 clusters were identified in the PBMCs of the profiled pembrolizumab-treated cohort (Supplementary Figure 2A). Although clusters associated with the CD68+ macrophage and CD4+ T-cell lineages predominated in the PBMCs of GBM patients (Supplementary Figure 2A–C), their level did not change after treatment. CD68+ macrophage clusters: #2, 3, and 10 showed expression of VISTA, B7-H3, and B7-H4, which are negative regulators of effector function. Treg cluster #18 seemed to have a low frequency (~1%), and the circulating numbers were not influenced by treatment or tumor recurrence (Supplementary Figure 2C). Pembrolizumab did not induce an increase in the number of circulating CD8+ T cells, although some clusters expressed markers of early differentiation (CD27 and CD28, cluster #5), cytotoxicity (CD56, cluster #8), and activation (HLA-DR, clusters #12 and 16) (Supplementary Figure 2D). The overall frequency of NK cells did not vary longitudinally in the GBM patients, and NK T cells were infrequently seen (Supplementary Figure 2E). Cumulatively, there were no changes in the systemic immune cell subsets upon pembrolizumab treatment or upon recurrence.

Tumor Microenvironment of GBM Is Poorly Infiltrated with Effector T Cells and Is Dominated by Macrophages

Immunohistochemical analysis of the formalin-fixed, paraffin-embedded GBM tissues was performed on the following two types of tumor tissues: (i) matched GBM tumors from the initial diagnosis prior to treatment and from the pembrolizumab-treated GBM tissues from the same patient (n =4); and (ii) GBM tissues obtained during pembrolizumab treatment (n =7) which lacked matched pretreatment tissues. The untreated GBM tissues were used as baseline comparators to the unmatched GBM tissues obtained during pembrolizumab treatment (Fig. 3 and Supplementary Table 1). There were no differences in the number of GBM-infiltrating CD4+ T cells (helper), CD8+ T cells (cytotoxic), CD45RO+ (memory) T cells, or CD68+ macrophages between the untreated and pembrolizumab-treated samples. Moreover, relative to other types of malignancies like melanoma,²⁷ the absolute number of tumor-associated CD8+ T cells is rather low (median 84 cells/mm² for untreated and a median of 115 cells/mm² for pembrolizumab treated) (Fig. 3). Importantly, ICOS, a biomarker of immune effector activity, was very low in both untreated and pembrolizumab-treated tumors (median 3.5 cells/mm² and 1.9 cells/mm², respectively) (Fig. 3). We observed a modest increase in granzyme B expression in GBM tumors treated with pembrolizumab (median ~7.2 cells/mm²; P = 0.0229) relative to untreated GBM (median, ~1.4 cells/mm²) (Fig. 3). As we have previously described,² Tregs were present in both the newly diagnosed and recurrent GBMs (Fig. 3, see FoxP3).

Both the IHC and CyTOF data show the extremely low frequency of T effector cells and presence of Tregs in the tumor microenvironment of GBMs both pre-pembro treatment (IHC only) and during pembro treatment. However, the IHC analysis did not show a marked increase in the CD68+ macrophages relative to CD4+ and CD8+ T cells. Yet, these CD68+ macrophages seen via IHC could be immunosuppressive, as seen in the CyTOF analysis. A deeper analysis of the CD68+ macrophages using mIF demonstrated pronounced diffuse coexpression of the M2-skewed phenotypic marker CD163 throughout the tumor microenvironment (Fig. 4A). In over half of the GBM specimens analyzed, CD68+CD163+ macrophages were markedly evident in the walls of the blood vessels (Fig. 4B).