Phase II Study of Radiotherapy and Temsirolimus versus Radiochemotherapy with Temozolomide in Patients with Newly Diagnosed Glioblastoma without MGMT Promoter Hypermethylation (EORTC 26082)

The serine/threonine kinase, mTOR serves as a hub integrating multiple intra- and extracellular cues in cancer cells (1). mTOR is involved in the formation of two multiprotein complexes, mTORC1 and mTORC2, that direct cell metabolism, growth, proliferation, survival, and angiogenesis.

Preclinical studies suggested an enhanced activity of mTOR inhibition in PTEN-deficient tumor models (2, 3).

Activation of the PI3K/AKT/mTOR pathway has been associated with reduced survival of glioma patients (4). This signaling pathway has been subjected to a number of single- or multitargeted therapies, in trials for recurrent high grade glioma or GBM, including the mTOR inhibitor rapamycin or its derivatives, the “rapalogs” everolimus (RAD001), deforolimus (AP23573), and temsirolimus (CCI-779), all with negative results (5–9).

The experience with temozolomide (TMZ) teaches that limited activity at recurrence (10) may still relevantly modify the disease in patients with newly diagnosed glioblastoma when combined with radiotherapy (RT; ref. 11). Accordingly, mTOR inhibition has been considered an option for patients with treatment-naïve glioblastomas that likely lack some of the mechanisms of resistance acquired at recurrence.

Temsirolimus (Torisel) has been approved for advanced renal cell carcinoma (12) and relapsed or refractory mantle cell lymphoma (13). Additive effects of temsirolimus plus radiotherapy in preclinical models demonstrate that temsirolimus could complement the genotoxic activity of radiotherapy (RT) in the treatment of newly diagnosed glioblastoma. However, combination of temozolomide and temsirolimus plus radiotherapy was too toxic (14).

Therefore, the rationale of this study was to test the biological effects of mTOR inhibition when combined with ionizing radiation in patients in whom temozolomide could be safely omitted. To this end patients with tumors with an unmethylated O6 methylguanine-DNA-methyltransferase (MGMT) gene promoter were selected for the trial, as they derive little if any benefit from the addition of temozolomide (15). Another aim was to identify biologic factors, that is biomarkers linked to benefit from mTOR inhibition. Temsirolimus may counteract therapy-induced angiogenesis and invasion (16, 17).

Patients for EORTC 26082 (NCT01019434) were recruited at 14 study sites in 10 countries in Europe. First, patients were registered after consenting for independent pathology review and central testing of the MGMT promoter methylation status by licensed laboratories of MDxHealth using quantitative methylation-specific polymerase chain reaction of DNA isolated from macrodissected formalin-fixed paraffin-embedded tumor sections (18). Patients were considered MGMT unmethylated, applying a safety margin, when the ratio of MGMT to the control gene ACTB was < 0.6, calculated as (methylated MGMT/ACTB) × 1,000. This corresponds to the lower bound of the 95% confidence interval (CI) established in a cohort of 602 glioblastoma samples screened in the CENTRIC trial where the cutoff corresponding to the established nadir was at a ratio of 2 that separates methylated from unmethylated (19), as visualized in Supplementary Fig. S1. A minimum of 1,250 copies of ACTB were required for a valid result, unless the copy number for methylated MGMT was 10 or more, which was scored as MGMT methylated.

Eligible patients (Supplementary Information) were randomly assigned to receive either standard chemoradiotherapy (TMZ/RT→TMZ; ref. 11), or standard fractionated radiotherapy with concomitant temsirolimus (standard dose of 25 mg i.v. weekly beginning at day −7 from the start of radiotherapy, to be continued until disease progression; Fig. 1 and Supplementary Information). The study was conducted according to the Declaration of Helsinki, the International Conference on Harmonization note for good clinical practice (Topic E6, 1996), and regulatory requirements.

This study was funded by a grant from Pfizer (details on the Role of the Funding Source in the Supplementary Information).

Randomization was performed centrally using an interactive voice response system. Patients were stratified according to age, WHO performance status, and baseline steroids. As this was an open-label study, no blinding procedures were applied.

The primary endpoint was overall survival at 12 months (OS12) to avoid issues around pseudoprogression and generate a timely signal. Secondary endpoints included progression-free survival (PFS), OS, safety and assessment of prognostic, and predictive biomarkers.

OS12 was defined as the fraction of patients alive at 12 months from randomization; PFS was defined as duration from randomization until first observation of PD or death from any cause or censored at last disease assessment without progression or start of second anticancer therapy; OS was defined as time from randomization until death or last visit.

PFS was assessed locally by investigators according to the Macdonald criteria (20), in case of suspected pseudoprogression investigators were advised to continue treatment per protocol and repeat imaging after 1 to 2 months. If progression was confirmed, the date of first observation of tumor progress was used for the analyses.

Adverse events (AE) were coded according to the Medical Dictionary for Regulatory Activities version 15.0, and their severity was graded according to National Cancer Institute Common Terminology Criteria for Adverse Events version 3.0.

A Fleming one-sample one-stage testing procedure was used in each arm. It was assumed that with OS12 lower or equal to 60% (P0), the therapeutic activity of temsirolimus (CCI-779) was too low (11). Although a OS12 greater or equal to 80% (P1) implied that the therapeutic activity of temsirolimus (CCI-779) was adequate Type I (α) and II (β) errors were both equal to 5%. Under these hypotheses, a sample size of 54 eligible patients in each arm was required. The decision rule was that if >38 eligible patients were alive at 1 year, it was concluded that the therapeutic activity of temsirolimus was adequate.

All statistical analyses were performed on mature data (median follow-up 32 months) by Thierry Gorlia. The concept of a noncomparative control arm allows for adjustment of the initial assumptions based on contemporary control treatment. The trial would be insufficient to confirmatory declare efficacy. However, statistical comparisons are still valid and useful for hypothesis-generation and exploratory analyses.

The OS12 was also computed in the TMZ/RT→TMZ arm to assess the consistency with P0.

Tissue micro arrays (TMA) were constructed using recipient paraffin blocks with an agarose matrix (21). Immunohistochemical analyses and FISH were performed in duplicate on sections from two replicate TMAs basically as recommended by the manufacturers (see Supplementary Methods, for antibody description, conditions, and dilutions; FISH probes). Markers for post hoc analyzes of the mTOR pathway were prespecified in the protocol [phosphorylated S6 ribosomal protein, p-S6RP^Ser235/236; phosphorylated AKT, p-AKT^Ser473; PTEN; phosphorylated AKT1 Substrate 1 (proline-rich), p-PRAS40^Thr246; phosphorylated extracellular signal-regulated linase, ERK1/2^{Thr202/Tyr204}] or based on a more recent study (phosphorylated p-mTOR^Ser2448; refs. 22 and 23). Scoring and definition of dichotomization is detailed in the Supplementary Methods.

The centered score table of the markers containing missing values was analyzed by principal component analysis. Nonlinear iterative partial least squares (NIPALS) algorithm (24) was used to perform singular-value decomposition with missing value and to complete the data. A consensus hierarchical clustering analysis (25) based on Euclidean distance and Ward's algorithm was used to investigate the optimal number of clusters. The association among marker scores was illustrated by network representation based on Spearman correlation. Analyses and graphical representations were performed using R-3.2.0 and the R packages mixOmics, qgraphs (26), and ConsensusClusterPlus.

The scores of the P-markers were dichotomized into negative (scores 0, 1, corresponding to 0% to 10%) versus positive (scores 2–5, >10%). Study stratification factors (age, WHO performance status, baseline steroids) and molecular markers were correlated to OS.

Treatment arms were compared with a log-rank test at 5% significance. For each of them, PFS and OS were estimated using the Kaplan–Meier (KM) method. Associations of marker profiles with treatment efficacy were presented by Forest Plot and significance was assessed with the test for interaction computed from a Cox model including the treatment, the marker, and their interaction term. A 5% significance was used for screening predictive markers. For each factor, univariable survival estimates were calculated using the KM technique in the temozolomide and temsirolimus arms. HRs obtained from univariable Cox models were presented with 95% CI (details in the Supplementary Information).

Overall, 257 patients were registered, screened for eligibility, and assessed for MGMT promoter methylation status, whereof 28 patients were registered after screening through the CENTRIC trial that selected MGMT methylated patients only (19); 190 patients were found to have glioblastoma with an unmethylated MGMT promoter applying the cutoff with a safety margin (Supplementary Fig. S1). The primary reasons for initially registered patients not to continue to randomization were hypermethylated MGMT status (n = 67), withdrawal of consent (n = 24), and other reasons (n = 55), including insufficient tumor material (n = 30), and AEs after surgery (n = 8; Fig. 1). A total of 111 patients were randomized from December 2009 to September 2012 and constituted the ITT population: 56 patients were scheduled to receive weekly temsirolimus in addition to standard radiotherapy (temsirolimus arm) and 55 were to receive TMZ/RT→TMZ alone (control arm). In the safety population, that is, patients with at least one dose of drug, there were 53 patients in the temsirolimus and 51 patients in the temozolomide arm.

Median follow-up was 33 (95% CI, 23–37) months in the temsirolimus and 32 (95% CI, 22–40) months in the temozolomide arm. The median duration from operation to randomization was 2.6 weeks (range 0.4–6.1 weeks). Patient baseline and demographic characteristics were well balanced between treatment arms except for the WHO performance status between PS0 and PS1, which favored the control arm. This is explained because the stratification was PS 0–1 versus PS2 (Table 1).

In the biomarker cohort (n = 88), only one patient sample displayed positive staining for the IDH1-R132H mutant (1/78; 1.3%), an expected low frequency, because 75% of the few IDH1-mutant glioblastoma are MGMT hypermethylated (27). The frequency of EGFR amplification was in the expected range (54%, 44/82). There was no difference in baseline characteristics and outcome in patients with versus without markers assessment (Supplementary Fig. S2 and Supplementary Table S1).

The median duration of radiotherapy was 6.1 weeks in both arms. Main reason for interrupting radiotherapy was technical or administrative (28%). In median, radiotherapy was interrupted 2 days. radiotherapy was completed by >90% of patients. Concomitant treatment was delivered as planned per protocol by >90% of patients in both arms. Patients in the temsirolimus arm received the drug for a median (95% CI) of 16 weeks post radiotherapy (4.0–84.3), with a mean dose intensity of 21.4 (6.3–25) mg/week.

Maintenance temsirolimus was administered per protocol at a median of 13 weekly cycles. Median relative dose intensity was 85.6%. Twelve patients had a reduction in dose intensity below 70%, because of dose reduction (19.1%: 6.4% for hematologic toxicity, 10.6% for AE, 2.1% for other reasons), dose not given during at least one cycle (68%: 6.3% for hematologic toxicity, 34% for nonhematologic toxicity, 58% for other reasons) or treatment delay (58%: 2.1% for hematologic toxicity, 17% for nonhematologic toxicity, 43% for other reasons).

Median OS was 14.8 (13.3–16.4) months in the temsirolimus arm and 16.0 (13.8–18.2) in the control arm (90 deaths; HR 1.2; 95% CI, 0.8–1.8; P = 0.47; Fig. 2A). The OS12 and OS24 rates did not differ between arms (70%, 72% and 15%, 16%, respectively). Median PFS as assessed by the investigator was 5.4 (95% CI, 3.7–6.1) months in the temsirolimus arm and 6.0 (95% CI, 2.8–8.0) months in the control arm (54 PFS events; HR 1.26; 95% CI, 0.86–1.86; P = 0.24; Fig. 2B). In the per protocol population (Supplementary Information), 38 patients treated with temsirolimus had survived ≥1 year. At least 39 patients were needed to reach the targeted drug activity.

In the temsirolimus arm, severe hematological toxicity was neutropenia (G3: n = 1, 1.9%) and lymphocytopenia (G3: n = 9, 16.4%, G4: n = 1, 1.8%). In the temozolomide arm, severe hematologic toxicity was leukopenia G3 (n = 2, 3.8%), neutropenia G4 (n = 2, 3.8%), lymphocytopenia (G3: n = 14, 26.4%, G4: n = 2, 3.8%), and thrombocytopenia (G3: n = 1, 1.9%, G4: n = 1, 1.9%). There was no other severe (G3/4) treatment-related AE with an incidence >5% in either arm.

Markers interrogated for their relevance of targeting the mTOR signaling pathway (22, 23) are visualized in the mTOR KEGG pathway (ref. 28; Supplementary Fig. S3). Phosphorylated mTOR^Ser2448 was associated with prolonged OS as evidenced by the significant interaction term between treatment and p-mTOR^Ser2448 (P = 0.047; Fig. 3). Tumors of 37.6% of the patients scored positive for p-mTOR^Ser2448. There was a nonsignificant trend for longer OS when p-mTOR^Ser2448 positive patients received temsirolimus as compared with controls (HR 0.62; 95% CI, 0.26–1·47; P = 0.27). When nonphosphorylated mTOR^Ser2448 patients received temsirolimus, a nonsignificant decrease in survival was observed compared with controls (HR 1.77; 95% CI; 0.95–3.29; P = 0.07; Fig. 3). The median OS in the temsirolimus group was 17.8 months (95% CI, 14.1–28.0) for patients with p-mTOR^Ser2448 positive tumors and 13.1 months (95% CI, 9.7–15.1) in the negative subgroup (P = 0.007; Fig. 3A). In the RT/TMZ→TMZ control arm, the median OS in the p-mTOR^Ser2448 positive group was 14.0 months (95% CI, 9.6–19.6) and 16.5 months (95% CI, 9.5–18.8) in the p-mTOR^Ser2448 negative subgroup (P = 0.999). For p-PRAS40^Thr246, the interaction test with treatment was borderline nonsignificant (P = 0.07). The impact of all other markers on survival is illustrated in a forest plot for all other markers in Supplementary Fig. S4.

A multidimensional analysis used the full range of the scores of the mTOR-associated markers integrating information for the identification of clinically relevant molecular subgroups and to gain further insights on pathway interactions (Fig. 4). The two first axes obtained by PCA explained 57.8% of the total inertia. The first axis was mainly explained by p-mTOR^Ser2448 and p-PRAS40^Thr246. The p-S6RP^Ser235/236 mainly contributed to the construction of the second axis (Fig. 4E and F). PTEN expression played a minor role in the structure of the score table (Fig. 4F). Subgroups were determined by consensus clustering. We kept the cluster based on two groups (k = 2) by default, as no strong indication for the optimal number of clusters was obtained and the sample size is limited (Supplementary Fig. S5). Cluster 2, highly enriched for p-mTOR^Ser2448–positive cases, revealed a strong association with outcome in the temsirolimus treatment group and no difference in the TMZ/RT→TMZ group (Fig. 4). Significant interaction was observed with treatment (P = 0.009): in Cluster 2, the HR was 0.42 (95% CI, 0.15–1.13; P = 0.08), and in Cluster 1, the HR was 1.77 (95% CI, 0.96–3.25; P = 0.06).

In multivariable prognostic analyses of clinical and molecular factors (Supplementary Table S1), p-mTOR^Ser2448 (HR 0.13; 95% CI, 0.04–0.47; P = 0.002), p-PRAS40^Thr246 (HR 0.50; 95% CI, 0.21–1.18; P = 0.12), p-ERK^{Thr202/Tyr204} (HR 2.81; 95% CI, 0.97–8.09; P = 0.06), but no clinical factor was associated with OS in the temsirolimus arm. The PEV was equal to 14.9% In the temozolomide arm, there was a trend for decreased survival in p-AKT^Ser473–positive patients (HR 3.21; 95% CI, 0.89–11.56; P = 0.07; PEV = 4.5%). None of the models had a PEV larger than 20%.

This randomized, open label phase II trial investigating the mTOR inhibitor temsirolimus in combination with radiotherapy for patients with low probability of benefit from the temozolomide-based radiochemotherapy failed to demonstrate the targeted outcome. Neither PFS nor OS demonstrated a signal of relevant activity in the total trial population (Fig. 2). Safety and tolerability of temsirolimus in combination with standard radiotherapy were nonconcerning and the trial is an example that temozolomide can be safely omitted in patients with MGMT unmethylated glioblastoma. The trial proposes mTOR^Ser2448 phosphorylation as a biomarker for benefit from mTOR inhibition. These results need further confirmation, and a trial to prospectively assess the relevance of this putative biomarker is underway (NCT Neuro Master Match, EudraCT 2015–002752-27).

The good outcome data in both arms of the trial prompted a comparison with the EORTC26981-22981/NCIC CE3 trial. The comparison with our pivotal TMZ/RT→TMZ versus RT trial (EORTC26981-22981/NCIC CE3; ref. 29) was favorable in all aspects, supporting the principal rational to design trials for patients with MGMT unmethylated glioblastoma and withhold temozolomide in the experimental arm (Supplementary Results). Biases in favor of EORTC 26082 may have been patient selection, and the lower number of patients on steroids (30). Bevacizumab was administered in about 45% of the patients in both arms of EORTC 26082. The OS of the EORTC 26082 arms is comparable with the outcome in the control arms of trials with selection of MGMT unmethylated patients, with 13.4 months in the CORE trial (95% CI, 12.2–14.3) with a bevacizumab use at recurrence of 22% (31) and 17.3 months (95% CI, 14.8–20.4 months) in the GLARIUS trial with crossover to bevacizumab of 60% (32).

The EORTC 26082 trial aimed at not withholding temozolomide from any patient with an equivocally methylated MGMT promoter by applying a MGMT cut-off with a safety margin. This prompted an adaption also in the GLARIUS trial (32) with similar design and therefore demarcates an evolution from the S039 trial with enzastaurin (33). Two randomized phase III trials in elderly patients with newly diagnosed glioblastoma further support a strictly predictive effect of the MGMT status for benefit from temozolomide (34, 35). However, we cannot completely exclude a small baseline effect of temozolomide despite the MGMT unmethylated state (11). Hence, withholding temozolomide outside trials and elderly patients with unmethlylated MGMT promoter is not advocated by the present data. In the temsirolimus arm 59% (n = 33) of the patients received temozolomide after treatment discontinuation, and 26% of temozolomide patients (n = 14) were rechallenged with temozolomide, not being aware of the recent data from the DIRECTOR trial that rechallenge with temozolomide might be relevant only for patients with a methylated MGMT promoter (36).

The choice of temsirolimus for patients with unmethylated glioblastoma was based on preclinical data already highlighting that not every tumor responds to the treatment (37) and a response may be only transient because of the overt feedback resistance mechanisms (22, 38).

Molecular analyses of prespecified principal components of the EGFR-PI3-K/mTOR/AKT pathway were performed. EORTC 26082 provides first evidence that p-mTOR^Ser2448 and—to a lesser extent—p-PRAS40^Thr246 may serve as decisive biomarkers for the treatment of patients with newly diagnosed glioblastoma with an unmethylated MGMT promoter. Phosphorylation of mTOR^Ser2448 has been shown to be targeted and blocked by rapamycin, a major metabolite of temsirolimus (39), whereas phosphorylated PRAS40^Thr246 (substrate of AKT1) relieves inhibitory function on mTORC1 (40). The survival curves may even suggest that there is a detrimental effect of temsirolimus in p-mTOR^Ser2448 negative tumors (Figs. 3 and 4). Previous trials testing temsirolimus at recurrence had focused on the PTEN status with a PTEN deficiency as a prerequisite for response (22) or on other downstream mTOR targets, for example p-S6RP^Ser235/236, which was neither associated with outcome in biomarker analyses of patients with recurrent glioblastoma receiving temsirolimus (6, 38) nor in this study. It cannot be excluded that glioblastomas treated at recurrence may have changed mTOR pathway activity as compared to tumor specimen used for marker analyses obtained at the first resection (41). Also, “paradoxical” activation of AKT by elimination of negative feedback downregulating survival signaling has been postulated as potential resistance mechanism to mTOR inhibition in previous trials, based on the analyzes of paired tumor specimen taken before and after treatment (22, 38). Interestingly, trials in other diseases did not provide predictive biomarkers (12, 13).

The limitations of EORTC 26082 are the relatively small sample size of this noncomparative phase II trial. For the biomarker analyses using IHC, only a limited number of tumor tissue samples from the ITT cohort were available. The findings should be validated by evaluation of previous trials in particular in those treating newly diagnosed glioblastoma patients (42) and the randomized phase II study RTOG-0913. Ongoing trials using mTOR inhibitors may need to take into account a potentially detrimental effect in patients with an unphosphorylated mTOR^Ser2448. Given the ongoing efforts of biomarker-driven basket trials for patients with newly diagnosed glioblastoma, the concept of mTOR inhibition using the marker predictive in this study, p-mTOR^Ser2448 is incorporated into the design of a future study.

W. Wick reports receiving speakers bureau honoraria from Bristol-Myers Squibb, MSD, and Roche; and is a consultant/advisory board member for MSD, Novocure, and Roche. M.J. van den Bent reports receiving other commercial research support from Abbvie, and Roche; and is a consultant/advisory board member for Abbvie, Actelion, Blue earth Diagnostics, Bristol-Myers Squibb, Cavion, Merck Ag, Novartis, and Roche. M.J.B. Taphoorn is a consultant/advisory board member for Hoffmann-La Roche. P. Roth reports receiving speakers bureau honoraria from Novartis, and is a consultant/advisory board member for MSD, Molecular Partners, and Roche. K. Homicsko reports receiving commercial research grants from Roche, and is a consultant/advisory board member for Amgen, and Roche. G.A. Pesce is a consultant/advisory board member for MSD. M.E. Hegi is a consultant/advisory board member for MDxHealth. No potential conflicts of interest were disclosed by the other authors.

Conception and design: W. Wick, T. Gorlia, A.A. Brandes, M. Kosch, R. Stupp, G. Pesce, M.E. Hegi

Development of methodology: W. Wick, T. Gorlia, A.A. Brandes, G. Pesce

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W. Wick, M. Platten, M.J. van den Bent, M.J.B. Taphoorn, A.A. Brandes, M.-F. Hamou, A. Wick, M. Weller, P. Roth, J.-S. Frenel, M. Campone, D. Ricard, C. Marosi, S. Villa, A. Weyerbrock, K. Hopkins, K. Homicsko, B. Lhermitte, G. Pesce, M.E. Hegi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Wick, T. Gorlia, P. Bady, M.J. van den Bent, A.A. Brandes, A. Wick, M. Weller, R. Stupp, P. Roth, M. Campone, B. Lhermitte, G. Pesce, M.E. Hegi

Writing, review, and/or revision of the manuscript: W. Wick, T. Gorlia, M. Platten, M.J. van den Bent, M.J.B. Taphoorn, J. Steuve, A.A. Brandes, M.-F. Hamou, A. Wick, M. Kosch, M. Weller, R. Stupp, P. Roth, V. Golfinopoulos, J.-S. Frenel, M. Campone, C. Marosi, S. Villa, A. Weyerbrock, B. Lhermitte, G. Pesce, M.E. Hegi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Gorlia, J. Steuve, A.A. Brandes, M.-F. Hamou, A. Wick, M. Weller, V. Golfinopoulos, D. Ricard, M.E. Hegi

Study supervision: W. Wick, T. Gorlia, R. Stupp, V. Golfinopoulos, G. Pesce

The authors thank the patients and their families for agreeing to participate in this trial, as well as to the nurses and data managers for their collaboration. A list of the participating investigators is provided in the Supplementary Information.

This work was supported by Pfizer who provided an unrestricted academic grant. Swiss National Science Foundation (FN31003A-138116 to M.E. Hegi) and the Swiss Cancer Ligue (KFS 02670-08-2010 to M.E. Hegi) supported the biomarker analyses.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.