Angiopoietin-2 May Be Involved in the Resistance to Bevacizumab in Recurrent Glioblastoma.

Despite encouraging response rate of bevacizumab (BVZ) in recurrent glioblastoma, many patients do not respond to this schedule and most of the responders develop an early relapse. Plasma concentrations of VEGF, PlGF, Ang2, and sTie2 were assessed by ELISA before and during BVZ treatment in seventy patients. Baseline levels of VEGF-A, and PlGF were higher in patients than in healthy volunteers, whereas no difference was found for Ang2, and sTie2. No biomarker at baseline was associated with response, PFS or OS. At recurrence, the authors observed an increase of Ang2 suggesting that Ang2/sTie2 could be involved in the resistance to BVZ.


Introduction
Microvascular proliferation and increased vascular permeability are both hallmarks of glioblastoma, making this tumor an attractive candidate for antiangiogenic therapies (1,2). Glioblastomas express high level of Vascular Endothelium Growth Factor (VEGF-A) (3,4). Overexpression of VEGF/VEGFR is associated with a poor prognostic and a short survival (5,4).
Anti-VEGFA antibody bevacizumab (BVZ, Avastin; Genentech, San Francisco, CA) initially associated with irinotecan, a topoisomerase I inhibitor (Camptosar, Pfizer, New York, NY) led to a high response rate (30-40%) never obtained before in recurrent glioblastomas (6,7). Subsequent phase II study suggested that BVZ alone may be also efficient and less toxic than the association BVZ-Irinotecan (8). Despite a high response rate, the impact of BVZ on overall survival remains unclear. Two randomized trials showed that upfront BVZ associated with radio-temozolomide regimen do not prolong survival (9,10). In recurrent GBMs, several patients exhibit a primary resistance to BVZ  secondary resistant to BVZ, suggesting the activation of alternative angiogenic pathways distinct from VEGF. To date, there is no validated biomarker to predict the response or resistance to bevacizumab in cancer patients (11). High MMP2 plasma levels have been associated with prolonged tumor control and survival in GBM (12). Effectors able to activate alternative pathways and promote angiogenesis include angiopoietin (Ang) 1 and 2 and their Tie-2 receptor. In order to identify potential predictive markers, we measured plasmatic concentration of VEGF, placental growth factor (PlGF), Ang2, and Tie2 before but also during BVZ treatment and at recurrence.

Patients and plasma collection
We obtained sequential blood samples from 70 patients diagnosed with recurrent glioblastoma and treated with bevacizumab (10 mg/kg) every 14 days until progression. All patients gave a written informed consent before enrollment. Peripheral blood was collected in an EDTA-containing tube before bevacizumab treatment, then every month until progression. Plasma was obtained after blood centrifugation at 1500 r.p.m. for 10 min at 4°C. Samples were aliquoted and immediately stored at -80°C. Plasmas were also collected from 23 healthy volunteers.

Efficacy assessment
Magnetic resonance imaging (MRI) scans were performed every 28 days (i.e., every two cycles of bevacizumab). MRI sequences included T1 pre/postcontrast, T2, and fluid-attenuated inversion recovery. All radiological data were reviewed by two independent investigators (C.C., J.G.P.L.). Response to treatment was evaluated according to RANO criteria. In addition to response rate, we considered the duration of response and stabilization.

Statistical analysis
Mann-Whitney test was used to compare the concentrations of biomarkers between subgroups of patients. The variations of biomarkers levels during treatment were analyzed using paired t-test (Wilcoxon test). Progression-free survival (PFS) was defined as the time between the first BVZ treatment and progression. Overall survival (OS) was defined as the time between the BVZ treatment and death or last follow-up. Survival curves were calculated according to the Kaplan-Meier method and differences between curves were assessed using the log-rank test. Multivariate analyses were performed using the Cox model. Statistical analysis was performed using GraphPad 5.0 and R softwares. Differences were considered significant at p < .05.

Patients characteristics and response to BVZ
The characteristics of the 70 enrolled patients with recurrent glioblastoma are indicated in Supplementary  Table S1: 47 primary glioblastoma (pGBM) patients and 23 secondary glioblastomas (sGBM) patients were included.
Median duration of bevacizumab therapy was 18 weeks (range 1.9-69.5). At the time of analysis, all patients experienced tumor progression based on MRI data. According to RANO criteria, we observed a global response rate (CR + PR) of 51.4% in the whole cohort: 4 patients had a complete response and 32 had a partial response, 15 patients had stable disease, and 19 had progressive disease. In our series, BVZ response rate tended to be higher for secondary glioblastoma patients, as compared to pGBM patients (response rate 60.9% vs. 46.8%, respectively ( Table 1). The other clinical parameters did not differ between responder (CR + PR) and nonresponder (SD + PD) patients (Supplementary Table S2).

Progression free survival and overall survival
Median PFS was 25.4 weeks for CR + PR patients and 10.1 weeks for SD + PD patients (p < .0001) (Figure 1(A) and (B). The 6-month-PFS rate was 28.6% in the whole cohort, 8.8% for SD + PD vs. 50.0% for CR + PR patients (p = .0002), 31.9% for pGBM vs. 26.1% for sGBM (p = NS). Median OS from the first BVZ cycle was 7.8 months in the whole cohort, 10.8 months for CR+PR patients, vs. 6.4 months for SD + PD patients (p = .0003). For PFS, and OS, differences between CR + PR and SD + PD patients remained statistically significant in both pGBM and sGBM subgroups (Figure 1(C)-(D).
We also investigated the impact of clinical characteristics (age at BVZ initiation, Karnofsky performance status, and number of prior chemotherapies) on BVZ In pGBM, median PFS were . and . weeks for R and NR patients, respectively (p = .). In sGBM, median PFS reached . and . weeks for R and NR patients, respectively (p = .). (C) In the whole cohort, median OS was . months in R patients, whereas it reached . months for NR patients (p = .). In pGBM, median OS were . and . months for R and NR patients, respectively (p = .). In sGBM, median OS were . and . months for R and NR patients, respectively (p = .).
response parameters, i.e., RANO response rate, duration of PFS, OS, and OS from the first BVZ cycle (Supplementary Table S3). We observed that PFS was not dependent on the number of previous lines of treatment (0, 1, 2, or more) (13).

Baseline VEGF-A and PlGF concentrations differ between patients and controls
We investigated the baseline plasma levels of VEGF-A, PlGF, sTie2, and Ang2 in patients and healthy volunteers. Our results showed that baseline levels of VEGF-A and PlGF were significantly higher in patients than in healthy volunteers ( Table 2). For example, median plasma level of VEGF-A was 3.6-fold increased in patients, as compared to healthy controls (median VEGF-A level, 54.08 vs. 15.07 pg/mL, respectively, p < .0001). There was no difference in median baseline levels of Ang2 and sTie2 between patients and healthy volunteers. None of the biomarkers evaluated here presented a statistical difference between responder and nonresponder patients.

Baseline plasma levels of biomarkers did not correlate with PFS and OS
To identify a predictive factor of BVZ response, we compared the PFS and OS of patients with low marker concentration vs. high concentration (< and > median). We find no correlation between baseline plasma biomarkers concentrations and PFS or overall survival (Supplementary Tables S4 and S5).

Biomarkers concentration evolution during bevacizumab treatment
We next evaluated the impact of bevacizumab treatment on plasma biomarkers levels. Mean plasma VEGF-A levels were reduced by 2.2-fold as compared to baseline values after two and four cycles of treatment ( Figure 2 and Supplementary Table S6). Plasma levels of sTie2 and Ang2 also significantly decreased during the first 2 months of bevacizumab treatment, but in lower proportions (decrease in mean biomarker level between 14.2 and 30.7% of baseline value). In contrast, plasma PlGF level strongly increased during the first 2 months of bevacizumab treatment (mean PlGF level: 10.8 +/-0.8 pg/mL at baseline vs. 26.4 +/-1.6 pg/mL after two cycles and 29.9 +/-2.1 pg/mL; p < .0007). We found no difference in evolution of plasma concentrations between responder (CR + PR) and nonresponder (SD + PD) patients (data not shown).

Biomarkers evolution at relapse
We then investigated plasma biomarkers levels at tumor recurrence ( Figure 2, Supplementary Table S6).
VEGF concentrations were still reduced as compared to baseline levels, indicating that bevacizumab was still effective at the time of relapse. Both PlGF and sTie2 levels were similar at those observed during BVZ treatment. In contrast, we observed an increase of Ang2 levels at relapse, as compared to levels after two and four cycles of treatments (p = .0137 and p = .0123, respectively).

Discussion
In order to identify plasmatic biomarkers that predict benefit of BVZ, we measured the plasma levels of various angiogenic related factors in recurrent glioblastoma patients treated with BVZ, associated for most of the patients with Irinotecan. Bevacizumab has been widely used in recurrent glioblastomas, either in combination with Irinotecan or alone since there is no evidence that the association is superior to BVZ alone (8).
We first compared plasma levels of our potential biomarkers in the patient population versus healthy volunteers. The high levels of VEGF-A and PlGF reflect the presence of an active angiogenesis in glioblastoma patients, and the high level of VEGF support the use of bevacizumab in glioblastomas. However, none of the markers was correlated with the response to BVZ. Several plasma biomarkers, including VEGF, VEGFR1, PlGF, SDF1α, or MMP-2, have been evaluated as predictive candidate factor after anti angiogenic treatments in glioblastoma and other tumor types. To date, none of them has been validated in independent series (11,12,14,15). We then evaluated the plasma biomarkers variations induced by bevacizumab treatment. First, our data showed that plasma VEGF-A levels significantly decreased after two cycles of bevacizumab, demonstrating an efficient targeting of VEGF-A by the drug. In contrast, we observed a significant raise in circulating PlGF, the VEGFR1 ligand. PlGF increase during bevacizumab has been reported in various tumors (16,17) and during cediranib treatment in glioblastoma (18). PlGF increase may reflect a shift induced by BVZ, from VEGFR2 toward VEGFR1 angiogenic signaling. In the same way, plasma levels of Ang2 and sTie2 decreased. This finding suggests that targeting VEGF pathway modulates the Ang2-sTie2 axis. Indeed, blockade of VEGF signaling with anti-VEGF receptor cediranib reduced levels of plasma Ang-2 in several patients (18). Conversely, Ang-2 destabilizes mature vessels, enabling VEGF to promote angiogenesis.
We observed an increase of Ang2 at recurrence, while patients were still under BVZ treatment, suggesting that Ang-2 could be involved in the resistance to anti-VEGF therapy and could represent an additional target for improving the efficacy of anti-VEGF therapies in glioblastoma patients. Whereas Ang1 results in vessel stabilization and pericyte coverage, Ang2 has an opposite effect and destabilizes vessels. Ectopic expression of Ang2 in an orthotropic glioma mouse model results in higher vascular density, and compromises the survival benefits of anti-VEGFR2 treatment, probably by impeding vascular normalization by anti-VEGFR2 treatment (19). Our results are in line with those of Rigamonti et coll. (20) and Daly and coll. (21), which suggest that tumor resistance to anti-VEGFA therapy involve an upregulation of the Ang2/Tie2-dependant angiogenesis, and that dual blockade of VEGFA and Ang2 suppress tumor progression. Moreover, high baseline levels of Ang2 have been associated with poor outcome in colorectal cancer patients treated by bevacizumab (22). Our findings suggest that resistance to anti-VEGF therapy may be due to Ang2/Tie2 pathway activation. Thus, targeting both systems may be beneficial in maximizing the efficacy of anti/proangiogenic therapies.