Introduction

The US Food and Drug Administration (FDA), followed by several other national regulatory agencies, granted approval of single-agent bevacizumab, a humanized monoclonal antibody (MAb) against vascular endothelial growth factor (VEGF), for patients with recurrent GBM based on a rate of radiographic response that was of noteworthy durability among many patients [1]. Others, such as the European medicinal agency, denied bevacizumab approval for several reasons including the lack of a non-bevacizumab control arm [2]. Nonetheless, the rates of overall radiographic response (ORR) and progression-free survival (PFS) achieved with bevacizumab therapy among recurrent GBM patients, 28–38% and 29–50%, respectively [3, 4], are approximately 4-fold greater than those reported historically including meta-analyses of recently completed clinical trials that did not include bevacizumab [5–7]. While these results are compelling, the increment in overall survival (OS) with bevacizumab has been disproportionally low and is only on the order of 1.5–2 fold greater than that reported historically with non-bevacizumab therapy [3–7]. The explanation underlying the discordance between the degree of ORR/PFS benefit compared to OS is not clear and a matter of ongoing controversy.

Although GBM is one of the most angiogenic of cancers, blood flow within these tumor is often paradoxically deficient due to multiple morphologic and functional abnormalities relative to non-tumor vasculature [8–11]. In preclinical models, inhibition of VEGF signaling can induce a transient window of tumor vessel normalization reflected by decreased tortuosity, dilation, and permeability, as well as thinning of thickened vessel basement membranes and increased pericyte coverage. These morphological changes have been shown to translate into functionally improved blood flow with decreased interstitial pressure, greater oxygenation, and enhanced drug delivery [12–17]. Clinical evidence supporting tumor vessel normalization has also been documented, including colorectal cancer patients treated with bevacizumab [18], as well as GBM patients treated with cediranib, aVEGFR-2 tyrosine kinase inhibitor [19].

In the current study, we hypothesized that an aggressive chemotherapy regimen incorporating agents with potentially complimentary mechanisms of anti-tumor activity, may effectively exploit a window of tumor vessel normalization induced by bevacizumab, and yield greater anti-tumor benefit, particularly with regard to OS, than has been reported with bevacizumab monotherapy or bevacizumab plus other agents. Specifically, we evaluated the DNA crosslinking cytotoxicity of carboplatin when added to irinotecan, a topoisomerase inhibitor, in combination with bevacizumab.

Patients and methods

Protocol objectives

Our primary objective was to evaluate anti-tumor activity, defined by PFS at 6 months (PFS-6) of carboplatin and irinotecan combined with bevacizumab among adults with recurrent GBM. In addition, we sought to evaluate the safety of this regimen in this patient population.

Patient eligibility

Patients were required to have histologic confirmation of WHO grade IV malignant glioma (GBM or gliosarcoma) that progressed after prior radiation and temozolomide therapy. Patients with prior low-grade glioma were eligible if histologic transformation to grade IV malignant glioma was confirmed. Eligible patients were also: at least 18 years of age; had a KPS ≥70; were on a stable corticosteroid dose for at least 1 week; and had no more than three prior episodes of tumor progression. Additional enrollment criteria included: hematocrit >29%; absolute neutrophil count > 1,000 cells/μl; platelet count > 100,000 cells/μl; and serum creatinine, aspartate aminotransferase and bilirubin within 1.5 times the institutional upper limit of normal. At least 4 weeks between surgical resection or chemotherapy, and at least 12 weeks between radiotherapy and enrollment were required. All patients provided informed consent.

Patients were excluded for: prior bevacizumab therapy; progressive disease or grade ≥3 toxicity on prior carboplatin or irinotecan; uncontrolled hypertension; therapeutic anti-coagulation use; acute hemorrhage on baseline MRI; urine protein:creatinine ratio >1; pregnancy or nursing; active infection requiring intravenous antibiotics; therapeutic anti-coagulation with warfarin; and prior stereotactic radiosurgery, radiation implants, or radiolabeled MAb therapy unless there was unequivocal disease progression (such as a new lesion or biopsy-proven recurrence).

Treatment design

Eligible patients for this open-label phase II study received bevacizumab at 10 mg/kg intravenously every 14 days. Carboplatin was administered at an AUC of 4 on day one of each 28-day cycle. Irinotecan was administered on days 1 and 14 at 340 mg/m2 for patients receiving cytochrome P450 CYP3A enzyme-inducing anti-epileptics (EIAEDs; phenytoin, phenobarbital, carbamazepine, oxcarbazepine, and primidone) and at 125 mg/m2 for those not on EIAEDs. Study therapy continued until progressive disease, unacceptable toxicity, non-compliance with study protocol guidelines or withdrawal of consent.

Response evaluation

Study investigators determined response by neurologic examination and contrast-enhanced MRI after the first two treatment cycles and then prior to every other cycle based on the recently published response assessment in neuro-oncology criteria [20]. A complete response (CR) required disappearance of all enhancing and non-enhancing tumor on consecutive MRIs at least 4 weeks apart, with corticosteroid discontinuation and neurologic stability or improvement. A partial response (PR) required ≥50% reduction in size (product of largest perpendicular diameters) of enhancing tumor with stability or improvement of neurologic status and corticosteroids. Complete and PRs also required stable or improved signal abnormality on fluid-attenuated inversion recovery (FLAIR) sequences. Progressive disease (PD) included ≥25% increase of enhancing tumor, a new enhancing lesion, significant worsening of non-enhancing tumor including that detected by FLAIR or T2 sequences, or clinical decline. Stable disease (SD) was defined as any assessment not meeting CR, PR, or PD criteria.

Dose modification and retreatment criteria

Chemotherapy doses were held for grade 3 or 4, related, non-hematologic toxicity, grade 3 thrombocytopenia, grade 4 neutropenia, and fever and neutropenia (any grade) until the event resolved to grade 1 or pre-treatment baseline. Thereafter, chemotherapy doses were reduced by 25%. Chemotherapy doses were also reduced by 25% for any related event that required >2 weeks to satisfy re-treatment criteria. Patients who required more than three chemotherapy dose reductions were allowed to remain on study and receive bevacizumab alone. Dose reductions of bevacizumab were not allowed. Bevacizumab was discontinued for uncontrollable hypertension, grade 2 or greater hemorrhage, arterial thrombosis, wound dehiscence requiring surgical intervention, intestinal perforation, or grade 4 venous thrombosis, proteinuria or congestive heart failure. Bevacizumab was held until other related grade 3 events resolved to grade ≤1.

Initiation of each cycle required: an ANC ≥ 1,000/mm3; a platelet count ≥ 100,000/mm3; creatinine ≤1.5 times the upper limit of normal (ULN×), bilirubin ≤2 × ULN and AST ≤ 2.5 × ULN; proteinuria <3+ on urinalysis or urine protein: creatinine ratio ≤1.0; and resolution of any related grade ≥3 event to grade ≤1.

Statistical consideration

The study was designed to have adequate power to compare the efficacy of the study regimen to a historical benchmark. The basis for this efficacy assessment was the proportion of patients surviving progression-free for 6-months. Given a 6-month progression-free survival rate of 46% (95% confidence interval [CI]: 32, and 66%) among patients with recurrent GBM treated with bevacizumab and irinotecan [21], a sample size of forty recurrent GBM patients was chosen to allow 90% power to differentiate between PFS-6 rates of 32 and 66% with a type I error of 0.1. If the true 6-month PFS of bevacizumab, irinotecan, and carboplatin were 46% or less, there would be limited interest in developing this combination further. However, if the true 6-month PFS were 66% or more, there would definitely be interest in further investigation of this treatment regimen. Therefore, within this patient subgroup, the study will be used to differentiate between a 46 and 66% rate of 6-month PFS. Statistically, the hypothesis that will be tested is:

H0: P < 0.46 vs. H1: P > 0.66

where P is the proportion of patients who live six or more months without disease progression.

Forty (40) patients were planned to enroll in this single-stage study. If 23 or more of these 40 patients live six or more months without disease progression, the treatment regimen will be considered worthy of further investigation. Otherwise, the treatment regimen will be determined not worthy of further investigation within this patient population. The type I and II error rates associated with this testing are 0.097 and 0.098, respectively.

An interim efficacy analysis after 20 patients were accrued was planned a priori. If 15 or more of these 20 patients progressed or died within 2 months of study initiation, further accrual would be suspended.

For this study among heavily pretreated patients with an extremely poor prognosis, rates of unacceptable toxicity, defined as grade ≥2 CNS hemorrhage or grade 4 or 5 non-hematologic toxicity, of 15% or less were considered desirable, while rates of 40% or greater were considered undesirable. Stopping rules for unacceptable toxicity based upon boundaries proposed by Pocock were used to monitor this study after each group of four patients [22]. Accrual was not suspended to formally assess the toxicity profile unless the following thresholds of unacceptable toxicity were satisfied: ≥ 3/4; ≥ 4/8; ≥ 5/12; ≥ 6/16; ≥ 7/20; ≥ 7/24. The type I and II errors associated with this monitoring were 0.01 and 0.03, respectively. These guidelines did not adjust for differential length of follow-up of accrued patients.

Progression-free survival was defined as the time between the cycle one start date and the date of disease progression or death. PFS was censored at the time of last follow-up if the patient remained alive without disease progression, or at the start of non-study treatment if initiated before disease progression. OS was calculated from the start of therapy until death or last contact if censored. PFS and OS were summarized using Kaplan–Meier estimator including 95% CIs.

Results

Patient characteristics

Characteristics of the 40 patients who enrolled on this study between September 2009 and March 2011 are summarized in Table 1. Patients were relatively young (median age, 51 years) and 40% had a KPS of at least 90. All patients had progressed after standard therapy with radiation and temozolomide chemotherapy. The median time from original diagnosis to study enrollment was 11.4 months, and 27 (68%) enrolled at first recurrence.

Table 1 Patient Characteristic
Full size table

As of June 1, 2011, five patients (13%) continue to receive study therapy. Three patients (8%) completed planned therapy. Study therapy was discontinued due to PD in 18 patients (45%), toxicity in 11 patients (28%) and withdrawal of consent in three patients (8%). Among patients who discontinued study therapy, four (10%) remain free of progression. Sixteen patients (40%) remain alive while 24 patients (60%) have died.

Study drug administration and safety

A total of 189 cycles of therapy were administered including a median of 4.7 cycles (range, 1–11) per patient. All patients were assessable for toxicity. Table 2 summarizes the frequency of grade ≥2 adverse events that were at least possibly related to the study regimen per patient. Although most adverse events were grade 2, grade ≥3 events were common and were predominantly hematologic including grade 3 anemia, thrombocytopenia and neutropenia in 3 (8), 8 (20), and 10 (25%) patients. Additional grade 3 events that occurred in three or more patients included fatigue (n = 5, 13%), infection (n = 4, 10%) and nausea (n = 3, 8%). Grade 4 events were primarily hematologic as well, and included neutropenia in eight patients (20%) and thrombocytopenia in four patients (10%). In addition, single patients (3%) developed grade 4 gastrointestinal perforation and venous thrombosis, respectively. Eleven patients (28%) discontinued study therapy due to toxicity and 17 patients (43%) required dose modification. There was one study related death attributed to an intestinal perforation on day 19 of cycle 1. Although this condition stabilized with medical management, this patient developed acute neurologic decline and the family opted to withdraw further anti-tumor therapy and enrolled the patient in hospice.

Table 2 Number of patients with adverse events (number in parentheses equals % of study population)
Full size table

Outcome

The median follow-up for all patients was 15.4 months (95% CI: 7.5, 16.8), while the median follow-up for surviving patients is 7.1 months (95% CI: 2.2, 17.4). Outcome analysis was based on the intent-to-treat population and includes the patient who died in cycle one described above. The PFS-6 rate was 46.5% (95% CI: 30.4, 61.0%) and the median OS was 8.3 months (95% CI: 5.9, 10.7 months) (Table 3; Fig. 1). Best radiographic response was PR in 13 patients (33%) (Fig. 2), SD in 21 patients (53%), and PD in five (13%) while the patient who died of toxicity during cycle one was deemed non-evaluable. Among 20 who were on dexamethasone at study initiation, 13 (65%) were able to taper, including eight patients (40%) who were able to completely discontinue dexamethasone. Five patients (25%) required a stable dexamethasone dose and two patients (10%), who had progression at first evaluation, required an increase in dexamethasone. Among patients who progressed after study therapy and underwent MRI evaluation at the time of progression, only one patient had PD at a new, non-contiguous site. All other patients had local evidence of disease progression. Although most patients had evidence of worsened enhancement as well as FLAIR signal abnormality at progression, five patients had progressive FLAIR changes without increased FLAIR at the time of tumor progression (Fig. 3).

Table 3 Summary of outcome in the current study relative to previously reported prospective studies of bevacizumab therapy among bevacizumab-naïve, recurrent GBM patients
Full size table
Fig. 1
figure 1

Kaplan–Meier estimate of time to progression (a) and overall survival (b) for patients treated on the current study

Full size image
Fig. 2
figure 2

Representative partial radiographic response of a recurrent GBM patient after six cycles of treatment with bevacizumab, irinotecan, and carboplatin that includes both diminished contrast uptake as well as FLAIR signal abnormality

Full size image
Fig. 3
figure 3

Representative examples of patients with PD documented by increased T2/FLAIR signal abnormality. In both cases, the contrast-enhancing disease continued to improve. In a, the T2 signal abnormality increased significantly medially (arrow) and in b, new FLAIR signal abnormality developed in the contralateral hemisphere (arrow)

Full size image

Discussion

A number of chemotherapeutic and molecularly targeted agents have been evaluated with bevacizumab for recurrent GBM and all have achieved outcomes that are no better than bevacizumab monotherapy (Table 3) [3, 21, 23–28]. Although one may argue that these data suggest chemo synergy with bevacizumab may not be possible in recurrent GBM, an alternative explanation is that the anti-tumor activity of agents evaluated in combination with bevacizumab to date is simply insufficient to enhance the benefit of bevacizumab. Of note, for all FDA-approved indications other than GBM, bevacizumab is administered in combination with either a cytotoxic or cytostatic agent, based on improved efficacy compared to bevacizumab monotherapy [29–32]. We conducted the current study to evaluate an aggressive multi-agent chemotherapy regimen in combination with bevacizumab for recurrent GBM [3, 4]. Although carboplatin and irinotecan exhibit modest anti-tumor activity when administered separately among recurrent malignant glioma patients [33–38], we reasoned that their ability to impair DNA replication via potentially complementary mechanisms including the introduction of alkyl groups by carboplatin and the inhibition of topoisomerase 1 activity by irinotecan, may lead to greater anti-tumor activity when administered in combination. Such enhanced activity has been demonstrated in preclinical GBM models [39–42]. Furthermore, based on data supporting tumor vasculature normalization associated with anti-VEGF therapy [14, 43], we hypothesized that bevacizumab may enhance delivery of these two cytotoxins, thereby further augmenting their anti-tumor activity. We predicted adequate safety with this regimen based on the non-overlapping primary toxicities of each of the agents, although the agents do share secondary toxicities. Furthermore, we did not observe unexpected toxicity with this combination among bevacizumab-naïve, recurrent malignant glioma patients treated off study at our institution as a “best clinical management” strategy, as well as among a cohort of bevacizumab-resistant patients treated on a formal phase II study [44].

Nonetheless, the results of the current study failed to demonstrate an improvement in outcome compared to the historical precedent set by single-agent bevacizumab. Several factors may have contributed to the disappointing outcome observed on our study. First, it appears that in the clinical setting of GBM that has progressed after radiotherapy and temozolomide, the chemotherapy regimen employed in the current study was insufficiently active, despite its incorporation of potentially complementary mechanisms of anti-tumor activity. Second, bevacizumab may have failed to adequately normalize tumor vasculature and therefore did not enhance delivery and associated cytotoxicity of the chemotherapy regimen. A missed opportunity of the current study is that it did not incorporate imaging assessments to determine the impact of bevacizumab on tumor blood flow and as a correlate, potential chemotherapy delivery. Specifically we did not assess whether bevacizumab was able to effectively normalize tumor vasculature. Third, it is possible that bevacizumab may actually diminish tumor blood vessel permeability to a sufficient degree to limit intratumoral chemotherapy exposure as has been demonstrated in some GBM models [45]. The rapid and marked decrease in tumor contrast uptake typically observed after bevacizumab therapy attests to its potent anti-permeability action. Fourth, potent anti-angiogenic therapy such as bevacizumab may select for resistance mechanisms within a subpopulation of the tumor that enhance the tumor’s intrinsic aggressiveness. One example of such an adaptation is the increased ability of tumor cells to invade and infiltrate widely [46–52]. Preclinical studies in orthotopic GBM models that accurately reflect the behavior of GBM in patients are critically needed to evaluate the specific mechanism of action of bevacizumab and to assess adaptive responses of the tumor to bevacizumab exposure.

An important aspect of response assessment associated with the current study was the incorporation of the RANO criteria which includes the assessment of non-enhancing as well as enhancing disease [20] rather than solely relying on measurement of enhancing disease as has been utilized traditionally via the Macdonald criteria [53]. In fact 13% of the patients treated on our study had PD documented solely by increasing T2/FLAIR abnormality without increased contrast enhancement (Fig. 3). These patients would have not been assessed as progressive using the conventional Macdonald criteria, thus highlighting the importance of the RANO criteria for the assessment of response particularly among patients treated with anti-angiogenic agents.

Although the primary toxicities of irinotecan and carboplatin differ, we observed that irinotecan and carboplatin when combined with bevacizumab is associated with significant toxicity among recurrent GBM patients. The overlapping secondary toxicities of these agents coupled with the general debility of recurrent GBM patients likely contributed to these adverse events. Although most grade ≥3 toxicity was hematologic, a spectrum of significant non-hematologic toxicities was also observed including one death attributable to intestinal perforation. In addition, eleven patients (28%) discontinued study therapy due to toxicity and 17 patients (43%) required dose modification. Our study findings are analogous to those reported for the addition of irinotecan to bevacizumab in a recently reported randomized phase II study in which patients treated with combinatorial therapy had greater toxicity than those treated with single-agent bevacizumab [3].

Our study findings do not support further evaluation of irinotecan, carboplatin, and bevacizumab for bevacizumab naïve, recurrent GBM patients because this regimen appears to have similar anti-tumor activity as bevacizumab monotherapy, but significantly greater toxicity. Future studies should incorporate imaging and circulating biomarkers to better delineate the specific mechanism of action associated with bevacizumab therapy as well as the adaptive responses by the tumor and associated microenvironment that evolve with therapeutic resistance.