Prognostic Value of CD109+ Circulating Endothelial Cells in Recurrent Glioblastomas Treated with Bevacizumab and Irinotecan

Background Recent data suggest that circulating endothelial and progenitor cells (CECs and CEPs, respectively) may have predictive potential in cancer patients treated with bevacizumab, the antibody recognizing vascular endothelial growth factor (VEGF). Here we report on CECs and CEPs investigated in 68 patients affected by recurrent glioblastoma (rGBM) treated with bevacizumab and irinotecan and two Independent Datasets of rGBM patients respectively treated with bevacizumab alone (n=32, independent dataset A: IDA) and classical antiblastic chemotherapy (n=14, independent dataset B: IDB). Methods rGBM patients with KPS ≥50 were treated until progression, as defined by MRI with RANO criteria. CECs expressing CD109, a marker of tumor endothelial cells, as well as other CEC and CEP subtypes, were investigated by six-color flow cytometry. Results A baseline count of CD109+ CEC higher than 41.1/ml (1st quartile) was associated with increased progression free survival (PFS; 20 versus 9 weeks, P=0.008) and overall survival (OS; 32 versus 23 weeks, P=0.03). Longer PFS (25 versus 8 weeks, P=0.02) and OS (27 versus 17 weeks, P=0.03) were also confirmed in IDA with CD109+ CECs higher than 41.1/ml but not in IDB. Patients treated with bevacizumab with or without irinotecan that were free from MRI progression after two months of treatment had significant decrease of CD109+ CECs: median PFS was 19 weeks; median OS 29 weeks. The presence of two non-contiguous lesions (distant disease) at baseline was an independent predictor of shorter PFS and OS (P<0.001). Conclusions Data encourage further studies on the predictive potential of CD109+ CECs in GBM patients treated with bevacizumab.


Introduction
Glioblastomas (GBM) are highly vascularized tumors: several antiangiogenic drugs including bevacizumab (Avastin ® , F. Hoffmann-La Roche Ltd, Basel, Switzerland), a monoclonal antibody targeting the vascular endothelial growth factor increase in vascular disorders and tumors due to vascular damage. CEPs are mobilized from the bone marrow to complement local angiogenesis. VEGF is highly expressed in GBM and may mobilize endothelial precursors from the bone marrow [6][7][8]. In particular, levels of CEPs, defined as CD34+ CD133+ VEGFR2+ cells, were higher in GBM patients than in patients with brain metastases or in controls and correlated with increased density of tumor blood vessels [9]. Variations in the number of pericyte precursors (progenitor perivascular cells, PPCs) could also provide relevant prognostic and possibly predictive information. Progenitor or committed pericytes expressing the platelet-derived growth factor receptor-beta (PDGFR beta) [10] may play a role in shaping the architecture of the vascular niche of an experimental model of glioma [11] and promote endothelial cell survival through induction of autocrine VEGF signaling [12]. Recent data suggest that high pericyte coverage has a negative prognostic impact in clear cell renal cell carcinoma [13].
CECs and CEPs have been considered for their predictive value in patients with colorectal cancer [14][15][16][17][18], breast cancer [19] and non-small cell lung cancer [20]. Although preliminary data are conflicting due to the differences among cancers and the variety of methodologies used to detect cells [3,4], the study of CECs and CEPs also in central nervous system tumors may be relevant for the identification of new insights of pathogenesis. Seaman et al, comparing the vascular trascriptome of normal resting, normal proliferating and tumor endothelial cells, identified CD109 as one of the membrane proteins that are selectively overexpressed on blood vessels during tumor angiogenesis [21]. Thus, we included the analysis of CD109 expressing CEC in our study.
Here, we report on the potential predictive value of CEC, CEP and PPC counts in patients with recurrent GBM treated with bevacizumab and irinotecan, as well as on their clinical and radiological follow-up. The results suggest that CD109+ CECs, in particular, deserve further investigation as potential predictive marker.

Ethics Statement
The study was carried out according to the Italian Decree Law May 8 th , 2003 allowing treatment of patients with no therapeutic option, with drugs not yet approved by the Italian Regulatory Agency, but with evidence of efficacy in phase II clinical trials. The protocol was approved by the Ethics Committee of the Neurological Institute "Carlo Besta" of Milan and registered in the Institute database (#1/08). All patients gave written informed consent before inclusion in the therapeutic protocol. All clinical investigation were conducted according to the principles expressed in the Declaration of Helsinki.

Patients
Sixty-eight GBM patients, who previously underwent prior surgery and radiochemotherapy according to the Stupp's protocol [22], followed by second or third line chemotherapy, were consecutively enrolled from January 14 th , 2009 and June 1 st , 2011 (Table 1). Exclusion criteria were as proposed by Vredenburgh et al. [23]. Follow up of patients was carried out until June 2012.

MRI and response evaluation
Patients underwent conventional contrast-enhanced MRI using a 1.5T MR system (Siemens, Avanto) with an 8 channels head coil at baseline, every 8 weeks or in case of neurological worsening, until tumor progression.
MRI evaluation was performed in agreement with RANO criteria [24], by two blinded radiologists. To assess changes of FLAIR hyperintensity a threshold of 25% or more of the maximal cross-sectional area was used. Baseline tumor volumes were determined on 3D post-gadolinium T1 weighted images by manually outlining the enhancing portion of the lesion using MRIcro (http://www.mccauslandcenter.sc.edu/ mricro/). To calculate the total enhancing volume of the tumor the number of enhancing voxels was multiplied by the voxel volume.
Disease patterns were characterized as local, distant, and diffuse [25]. Scans showing an increased area of abnormal FLAIR signal in the absence of increased or new enhancement were classified as diffuse recurrence [26].

Circulating endothelial cells and progenitors analysis
Number and viability of CECs and CEPs were measured on fresh samples from 68 patients at baseline and every 8 weeks by six-color flow cytometry. Blood samples were collected in EDTA discarding the first 3 ml of blood to avoid contamination with endothelial cells from venipuncture. The samples were kept at room temperature (22±2 °C) and processed as described in Mancuso et al. [27] within 24 hours after collection. CECs were defined as Syto16(DNA) +CD45-CD31+CD146+ [27]. The combination of Syto16 and 7-AAD was used to discriminate between nucleated viable (Syto16bright/7-AAD-) and apoptotic/necrotic (Syto16dim/7-AAD+) endothelial cells, and to exclude from analysis platelets and endothelial macroparticles. The expression of CD109 in CECs was also investigated. CD109+ CECs were enumerated as Syto16(DNA) +CD45-CD31+CD146+ CD109+ cells and investigated for viability by 7-AAD. Figure 1 shows the CD109+ CEC enumeration procedure.
To define reference values, cell populations were investigated also in age-and sex-matched healthy controls. Biological markers of the patient population treated with bevacizumab and irinotecan were compared with two independent datasets. Independent Dataset A (IDA) was formed by 32 age-and sex-matched patients with recurrent GBM and similar clinical characteristics, treated with bevacizumab alone (10 mg/kg i.v., every 2 weeks). Independent Dataset B (IDB) was formed by 14 patients with recurrent GBM, treated with temozolomide (50 mg/mq/die, according to RESCUE study [32]) or fotemustine (75 mg/mq/die on day 1, 8 and 15 followed, after a 35 days interval, by 100 mg/mq on day 1 of a 21 days cycle) constituted a second control group.

Statistical analysis
Progression Free Survival (PFS) was calculated from treatment onset until disease progression or death/last followup, if censored. Overall Survival (OS) was calculated from treatment onset until death/last follow-up, if censored. Kaplan Meier analysis estimated PFS and OS. The log rank test assessed differences in progression or survival in patients with different clinical, radiological or biological parameters. These parameters were set at the 25°, 50°, 75°, 90° percentile and separately evaluated in all patients.
Correlations between biological markers and clinical parameters or treatment response were assessed using the Mann-Whitney exact U test. The Wilcoxon rank sum test evaluated differences among biological markers levels at baseline, week 8 or progression. All P values were two-sided.
A multivariate analysis and a Cox proportional hazard regression model analysis were performed on variables showing statistically significant differences at univariate analysis to investigate their independent prognostic role. In particular, CD109+ CECs was used as a dichotomic parameter. All statistical analyses were performed using STATA software 10.0.

Treatment
Clinical characteristics of the patients are described in Table  1. None of them was previously treated with bevacizumab or other anti-angiogenic drugs. Fifteen (22%) experienced progression during the first six cycles of adjuvant TMZ therapy according to the Stupp regimen [22]. Two patients progressed <12 weeks after radiation therapy and performed a second MRI CD109+ CECs and Bevacizumab in Glioblastomas PLOS ONE | www.plosone.org confirming progression after 6 weeks; no patient had pseudoprogression.
Fifty-five patients were treated with both bevacizumab and irinotecan until progression; 13 patients interrupted irinotecan before progression due to low tolerance.
IDA and IDB patient characteristics are outlined in Table S1A and B.

Toxicity
Six patients stopped treatment before disease progression due to: intra-tumoral bleeding (n=2), sub-galeal infection after surgery (n=2), thrombosis of the cerebral sinus (n=1) and consent withdrawal (n=1). Dates of their disease progression and death were included in the statistical analysis.
Five patients died before disease progression due to: pancreatic neoplasia (n=1); ischemic heart failure (n=1); cerebral sinus thrombosis (n=1); unknown reasons (n=2). A detailed list of adverse events with relative grades is provided in Table S2.

Response rate and patterns of radiological progression
Using RANO criteria [24] we found that 14 patients had a partial response, 40 stable disease and 14 disease progression at week 8; no other partial or complete response was observed later. Patients with partial response at week 8 had longer PFS and OS than the others (39 versus 15 weeks, P=0.002; 58 versus 26 weeks, P=0.002). Figures S1 and S2, respectively, show one example of tumor progression and one example of partial response 8 weeks after treatment onset.
The analysis of progression patterns excluded patients who died or interrupted treatment before progression and patients with incomplete neuroimaging. Eleven patients converted to a diffuse pattern of disease (9 patients starting from local pattern and 2 from leptomeningeal dissemination), one patient with local disease converted to leptomeningeal dissemination and 36 patients did not show changes when compared to baseline.
Clinical features potentially affecting PFS and OS are outlined in Table 2. Although tumor volumes in patients with leptomeningeal dissemination or distant tumors were not significantly larger than in other subjects (median volumes 40.1 cc and 44.0 cc, respectively), both subgroups had shorter PFS and OS. Patients assuming <8 mg dexamethasone at baseline had longer PFS and OS than the others.

Circulating Endothelial or Progenitor cells
Significantly higher levels of CECs (P=0.01), CEPs (P=0.0001), viable CECs (P=0.0001), CD109+ CECs (P=0.0001), CD45dimCD34+CD133+ (P=0.001) and CD45dimCD34+ (P=0.006) hematopoietic committed progenitors were found at baseline in patients treated with bevacizumab and irinotecan, compared to healthy controls, even if P values were adjusted for multiple comparisons (Figure 3). Lower levels of CD140b+ PPCs were detected in patients when compared to healthy controls (P=0.03, data not shown). No correlation was observed between counts of each cell subpopulation and clinical or radiological parameters (data not shown).
Baseline levels of all cell subpopulations were not significantly different among patients treated with bevacizumab and irinotecan, bevacizumab alone (IDA) or chemotherapy alone (IDB) (data not shown).
Patients treated with bevacizumab and irinotecan showing baseline counts of CD109+ CECs higher than 41.1/ml (1 st quartile) had increased PFS and OS (Table 2 and Figure 4A); these patients did not show differences of clinical and   (Table  S3). PFS and OS were significantly increased also in patients treated with bevacizumab alone and baseline counts of CD109+ CEC higher than 41.1/ml ( Figure 4B), but not in patients treated with antiblastic chemotherapy ( Figure S3). In patients treated with bevacizumab and irinotecan, different changes in cell counts were detected in patients who progressed at 2 months, here defined as non-responders (n=14), compared to others (responders, n=42): counts of CECs, viable CECs, CD109+ CECs, CD45dimCD34+VEGFR2+ hematopoietic progenitor cells and CD140b+ PPCs decreased significantly after treatment in responders only ( Figure 5A). A reduction of CD109+ CECs was also observed in 17 patients responding to treatment with bevacizumab alone (IDA, Figure 5B), whereas in patients treated with chemotherapy (IDB) no reduction of CD109+
When PFS and OS of patients with baseline counts of CD109+ CECs higher or lower than 41.1/ml were adjusted for dexamethasone dosage (≥8 or <8 mg/die), the difference still remained significant (P=0.009 and P=0.04, respectively).

Discussion
Our data show that CECs and CEPs at baseline are higher in GBM patients than in healthy controls, in agreement with previous results on other cancers [4,33]. Treatment with bevacizumab and irinotecan caused a general decrease of CECs and CEPs: however, such decrease was only significant in patients that did not show progression after two months. Our report suggests for the first time a potential link between CD109 expression in CECs and anti-VEGF treatment: baseline CD109+ CEC count >41.1/ml identified a subgroup of patients  with longer PFS and OS and were significantly higher in longterm responders than in other patients. CD109, a glycosylphosphatidylinositol-anchored cell surface glycoprotein [34], is highly expressed in several solid tumors [35] and is one of 12 endothelial markers more expressed in tumor than normal endothelial cells [21]. Treatment with bevacizumab and irinotecan also decreased viable PPC counts, maybe due to the block of PPCs recruitment from tissue reservoirs or to PPCs inclusion into blood vessels to repair the damage induced by chemotherapy [27,36,37].
The observation that patients undergoing progression did not display changes of CECs, CD140b+ PPCs and CD45dimCD34+VEGFR2+ hematopoietic progenitor cells suggest that other biological programs, possibly favoring migration and invasion, are involved in progression, as proposed by preclinical studies [38,39].
Median and 6-month PFS and OS were slightly lower than previously reported [40,41], a likely consequence of the enrolment of patients with poor prognostic factors (e.g. KPS > 50; Table 1). OS of our patients is similar to that referred by Desjardins et al. who treated seriously impaired recurrent GBM patients with bevacizumab and metronomic TMZ [42]. PFS and OS in patients treated with bevacizumab and irinotecan were higher than in patients treated with bevacizumab alone, but the difference was not significant, as also found by Friedman et al. [43].
RANO radiological criteria were used for assessment of disease response to treatment [24], as they are better suited to study the effects of anti-angiogenic factors. The observation that distant disease at baseline is associated to a worst prognosis is of interest. Distant disease did not imply a larger tumor volume: thus, this result may allude to an increased resistance to bevacizumab due to the activation of migration/ invasion programs that appear somehow alternative to progression programs based on increased angiogenesis [44]. The pattern of recurrence observed in our patients is partially different from those previously reported; at baseline local disease was less frequent than in other reports (68% versus 80% or 72% reported by Chamberlain and Pope, respectively [25,45]) and leptomeningeal dissemination was present in 14% of our patients (6.2% in Chamberlain et al. [25]). At recurrence 24% of our patients converted to a diffuse pattern (30.3% in Pérez-Larraya [46]); moreover, half of patients with leptomeningeal dissemination at baseline developed a diffuse pattern (6% of all patients).
CD109 is a monomeric cell surface glycoprotein of 170 kD that is expressed on endothelial cells, activated T-lymphocytes and platelets and a subpopulation of bone marrow CD34+ cells [47]. Its expression is also high in different cancers and cancer lines, including glioblastoma [35]. CD109 is a TGF-beta coreceptor that regulates TGF-β receptor endocytosis and degradation, thus inhibiting TGF-β signaling [48]. Both VEGF and TGF-beta are involved in the regulation of endothelial cell stability [49]: whether this relationship is relevant to the results we obtained remains to be investigated.
This preliminary study was not powered to adhere to all the criteria for marker identification included in the REMARK checklist [50]. However we believe that the data encourage a larger study on the predictive potential of CD109+ CEC in GBM and possibly other cancer patients treated with bevacizumab. It is conceivable that the combination other potential markers of interest, like baseline plasma levels of VEGF-A, possibly associated with dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) [51] may help profiling with increased precision patients that may benefit from bevacizumab.