Macropinocytosis of Bevacizumab by Glioblastoma Cells in the Perivascular Niche Affects their Survival

This article is featured in Highlights of This Issue, p. 6757

The microenvironment provided in the perivascular space is emerging as a critical determinant of tumor cell growth and survival (reviewed in ref. 1). Although humanized IgG mAbs are now used in cancer therapy, surprisingly little is known concerning their access to the perivascular space in the brain and the mechanisms by which they affect tumor cells within this microenvironment. Bevacizumab, a humanized bivalent IgG1 mAb, was developed as an anti-angiogenic therapeutic agent. Currently, bevacizumab has FDA approval for use in the treatment of recurrent glioblastoma (GBM) and renal, lung, and colon cancer (2). Bevacizumab has been reported to extend the 6-month progression-free survival in patients with recurrent GBM (3, 4). Patients with recurrent GBM are now routinely treated with bevacizumab, which induces normalization of some of the tumor vasculature (2) and enhances quality-of-life. In common with other humanized mAbs, the response shows interindividual variability and approximately 20% of patients with recurrent GBM do not show any response to bevacizumab (5). The reasons for this lack of response have not been elucidated and the effects of bevacizumab on the cells in the perivascular niche have not been determined.

It is well established that populations of cells with a cancer stem cell-like phenotype (e.g., expression of the CD133 cell surface marker and of the transcription factor Sox2) are resistant to radiation therapy (6). Thus, alterations in the perivascular microenvironment could provide a niche that promotes and then sustains the development of this population of cells in various cancers, including GBM. Bevacizumab was developed based on its binding of vascular endothelial growth factor-A (VEGF-A) at the site recognized by the VEGF receptor (VEGFR) thereby preventing VEGF-A signaling (reviewed in ref. 2). As GBM cells with the CD133⁺ phenotype produce VEGF-A and express VEGFR2, VEGF-A can act in an autocrine or paracrine manner to promote survival of these cells as well as acting systemically (7, 8).

Large molecules can gain access to the perivascular tissue through transcytosis across the endothelial cells (EC) in the brain, and through alterations in the blood–brain barrier (BBB; refs. 9, 10). Although it is a reasonable assumption that bevacizumab gains access to the perivascular space in GBM, to our knowledge there are no published reports demonstrating this. One report indicated that human IgG (hIgG) was detectable in tumor lysates of orthotopic GBM xenograft tumors that had been propagated in nude mice administered bevacizumab (11). However, this was based on Western blotting of whole tumor lysates and, thus, did not differentiate between the presence of the bevacizumab in the perivascular space and the tumor-associated vessels. Similarly, the possibility that bevacizumab is internalized by GBM tumor cells has not been addressed. Recently, fluorescent-labeled bevacizumab was shown to gain access to invasive breast cancers when injected intravenously prior to mastectomy (12). In ECs, hIgG is thought to be taken up by a pinocytotic process and to bind the embryonic Fc receptor (FcRn), which recycles hIgG and is localized at the early endosome (13, 14). FcRn is expressed in brain ECs (13, 14), but was found recently either to be undetectable or present at very low levels in a large number of tumor cell lines and that when present it acts as a tumor suppressor by recycling albumin (15). Large molecules also can be internalized through the nonselective process of macropinocytosis. Macropinocytosis is initiated from plasma membrane ruffles that give rise to actin-containing plasma membrane extensions, resulting in large endocytic vacuoles called macropinosomes that ultimately become early endosomes (reviewed in refs. 16, 17). The cargo within early endosomes can be trafficked to a recycling compartment or the lysosome (16, 17). Macropinocytosis has been shown to be a mechanism by which Ras-transformed cells and starving cells engulf extracellular proteins and then lysosomally degrade the proteins, thereby promoting their own proliferation and survival (18, 19).

Here, we report that bevacizumab and hIgG gain access to the perivascular space in orthotopic mouse models of GBM and that this access is associated with defects in the BBB rather than transcytosis across the tumor-associated vasculature. We found that perivascular Sox2⁺/CD44⁺ tumor cells are largely responsible for internalization of bevacizumab within the niche. In CD133⁺/Sox2⁺ cells, propagated in media without addition of growth factors, this internalization occurs through a mechanism consistent with macropinocytosis and that a fraction is localized to a “fast” recycling compartment independent of FcRn and another fraction is localized to the lysosome. The macropinocytosis of bevacizumab and its trafficking to the lysosome promotes survival of the CD133⁺ cells, as does the autophagy induced by the bevacizumab-mediated depletion of VEGF-A.

Normal brain ECs (isolates 422 and 376) were purchased from Cell Systems; tumor-associated ECs (TEC) were isolated and propagated from primary human GBM tumors (20); CD133⁺ cells and the matched (paired) CD133⁻ (non-stem) tumor cells were isolated from two human GBM xenograft tumors (08-387 and 4121; ref. 21); and MM.1R human myeloma cells were purchased from the ATCC. Propagation of the cells and their authentication was as described in the Supplementary Methods.

Described in the Supplementary Methods.

FcγRIIB cDNA was purchased (EX-ZO435-Lv105, GeneCopoeia) and a lentivirus vector created and used to transduce MM.1R myeloma cells (22).

All studies were performed with the approval of the Animal Use and Care Committee of the Cleveland Clinic, Mayo Clinic or Emory University School of Medicine. The CD133⁺ orthotopic GBM, the PDX orthotopic GBM, and the syngeneic somatic gene transfer GBM models were generated as described (21, 23, 24), with details provided in the Supplementary Methods. The CD133⁺ and PDX models were established and treated with bevacizumab as described in the Supplemental Methods and Figure Legends. The somatic gene transfer model was generated and treated with rat anti-mouse-VEGF-A (25) as described in the Supplementary Methods and Figure Legends.

CD133⁺ GBM cells were plated overnight on chamber slides coated with 20 μg/mL laminin (R&D Systems #3400-00-01) in neural basal media (NBM) without addition of EGF or basic-FGF (bFGF). Maintenance of the stem-cell phenotype was verified by Sox2 expression (26, 27). The following day, a clinically relevant concentration (250 μg/mL; ref. 28) of bevacizumab was added to the media. After incubation (37°C, 5% CO₂) for 5 minutes, the cells were washed and fixed, or fresh media was added and the cells incubated for the times indicated. To detect internalized bevacizumab, cells were fixed with 4% buffered-paraformaldehyde, blocked, and reacted with Alexa Fluor 488-anti-human IgG, then mounted in media containing DAPI nuclear stain (Vector Lab #280772). Double-label immunofluorescent analysis was performed as described previously (20, 29). Additional details are found in the Supplementary Methods.

This assay was performed according to the manufacturer's instructions (CytoTox-Fluor Cytotoxicity Assay, Promega).

VEGF-A, VEGF-C, and placental growth factor (PGF) were measured using VEGF-A&C (RayBiotech) and PGF (R&D-Systems) ELISA assay kits as recommended by the manufacturer.

All statistical analyses were performed or overseen by the biostatistician (ASN). The test used is stated in the Figure Legends. A P-value < 0.05 was considered significant.

To determine whether bevacizumab can gain access to the perivascular space in GBM, we used a mouse model in which bevacizumab or a control hIgG was administered (i.v., 5 d) to mice with established GBM tumors. Bevacizumab and hIgG were detected in frozen sections of the brain by labeling with an Alexa Fluor 488-anti-human-IgG (green) and their location assessed by costaining with anti-von Willebrand factor (vWf), a marker of ECs, and Alexa Fluor 594-conjugated secondary antibody (red) together with nuclear staining using DAPI (blue). Both bevacizumab and the control hIgG were detectable in the perivascular space of the orthotopic GBM tumors (Supplementary Fig. S1A and S1B). Very little bevacizumab or hIgG was detected in the perivascular space of uninvolved brain from the mouse model (Supplementary Fig. S1A and S1B), consistent with prior studies indicating that minimal amounts of hIgG gain access to the perivascular tissue when the BBB is intact (30).

This analysis further indicated that bevacizumab is internalized by perivascular tumor cells in the GBM xenografts (Supplementary Fig. S1A and S1B). Using serial sections of xenograft tumors, we verified that the perivascular cells were human tumor cells by double-labeling with mAb anti-Sox2 and a rabbit antibody to human nuclei (Supplementary Fig. S1C). The relative distribution of bevacizumab in the Sox2⁺ cells versus activated microglia/macrophages was determined using intracerebral GBM xenograft tumors obtained from mice in which the tumor was established and then bevacizumab administered until euthanasia. Sections of the tumors were stained with antibodies to Sox2 and Iba1 (27, 31), and the fluorescent intensity of bevacizumab quantified. Although both perivascular Sox2⁺ cells and Iba1⁺ cells internalized bevacizumab, the bevacizumab-intensity was ∼2.5-fold higher in the perivascular Sox2⁺ cells than perivascular Iba1⁺ cells (Fig. 1A and B). On measuring the distance of Sox2⁺ and Iba1⁺ cells from the nearest blood vessel, we did not find a difference in proximity (Fig. 1C), consistent with the known localization of both cell types to the perivascular niche (1), although we did find that the mean number of Sox2⁺ cells was higher than the mean number of Iba1⁺ cells within a 25-μm-radius of the nearest blood vessel (Fig. 1D).

These observations were confirmed using an established immunocompetent mouse model in which GBM tumors were induced by PDGF-B and then treated for 2 weeks with rat anti-mouse-VEGF-A IgG followed by euthanasia. CD44⁺ marks cancer stem-like cells in this model (32) and we found that they also express Sox2 (Supplementary Fig. S1D). Tumor sections were stained with antibodies to CD44 or Iba1, as well as vWf, and the fluorescent intensity of rat anti-mouse-VEGF-A IgG quantified. We found that both perivascular CD44⁺ and Iba1⁺ cells within a 25-μm radius of blood vessels internalized anti-mouse-VEGF-A. The rat anti-mouse-VEGF-A-intensity was nearly five-fold higher in the CD44⁺ perivascular cells as compared to that in the Iba1⁺ perivascular cells (Fig. 1E and F). The distance of CD44⁺ perivascular cells from the nearest blood vessel was slightly less than that for Iba1⁺ cells (Fig. 1G).

We then determined whether there were differences in apoptosis of the perivascular tumor cells in the bevacizumab- and placebo-treated xenograft tumors by double labeling for cleaved caspase-3 and Sox2. We did not find a significant difference in the numbers of apoptotic Sox2⁺ cells in bevacizumab-treated as compared to placebo-treated tumors (P = 0.50; Fig. 2A and B). To determine whether bevacizumab induced autophagy, we examined the colocalization of LC3 puncta with LAMP2 (33). We found a significantly higher number and greater area of LC3 puncta colocalized with LAMP2 in the bevacizumab-treated as compared to the placebo-treated tumors (Fig. 2C–E), suggesting that bevacizumab therapy of GBM xenografts induces autophagy in vivo.

To confirm that CD133⁺ cells are able to take up bevacizumab, we analyzed CD133⁺ cells isolated from xenograft tumors from two different patients with GBM. The CD133⁺ cells are also Sox2⁺ (Supplementary Fig. S1E). The cells were seeded on laminin in NBM without EGF and bFGF to replicate the nutrient-starved condition in GBM tumors. We found that, under these conditions, bevacizumab was internalized rapidly (5 minutes) by the CD133⁺ cell isolates from both patients (Fig. 3A).

Double-label immunofluorescence of the CD133⁺ cells for actin (Alexa Fluor 488-phalloidin; green) and bevacizumab (Alexa Fluor 647-anti-human IgG; magenta) showed that bevacizumab was internalized into membrane ruffles surrounded by actin (Fig. 3B). This was suggestive of macropinocytosis, which is differentiated from other types of endocytosis by the requirement for remodeling of the actin cytoskeleton with polymerized actin in membrane ruffles. Macropinocytosis is blocked by inhibitors of the Na(+)/H(+) ion exchange pump, such as 5-(N-ethyl-N-isopropyl)amiloride (EIPA; refs. 16, 18, 34), and tetramethylrhodamine (TMR)-70-kDa-Dextran is known to be internalized by macropinocytosis (18). The macropinocytic capabilities of the CD133⁺ cells were confirmed by their rapid internalization of TMR-70-kDa-Dextran (5 minutes) through a mechanism that was inhibited by EIPA (Fig. 3C and D). The internalization of bevacizumab by the two isolates of CD133⁺ cells was inhibited significantly by pretreatment with EIPA (Fig. 3E and F). In contrast, the internalization of bevacizumab was not inhibited significantly by EIPA pretreatment of the paired nonstem tumor cells (Fig. 3G and Supplementary Fig. S2). Taken together, these data suggest that CD133⁺ cells, but not the paired CD133⁻ (nonstem) tumor cells, utilize macropinocytosis to internalize bevacizumab.

Internalized receptors or proteins can be recycled to the cell surface or targeted for degradation. The routes are differentiated based on their kinetics and the markers expressed on the intracellular vesicles. Rab4 marks a “fast” recycling compartment (t_1/2 ∼5 minutes) and also is involved in “slower” recycling through the endosomal recycling complex (t_1/2 ∼15–30 minutes; refs. 17, 35). LAMP1 is a marker of the late endosome/lysosome (17). We therefore examined the colocalization of bevacizumab with Rab4 and LAMP1 (17, 35). CD133⁺ cells were seeded as described above on laminin and colocalization determined at various time points after treatment with bevacizumab. Rapid, transient colocalization with Rab4 was observed. At 5 minutes bevacizumab was predominantly (68%) colocalized with Rab4, with this colocalization dissipating at 30 min (20%; Fig. 4A and B). At 3 hours, a large fraction of bevacizumab was colocalized with LAMP1 (∼55%; Fig. 4C and D). Similar results were observed on analysis of CD133⁺ cells isolated from a different GBM tumor (Supplementary Fig. S3A and S3B). Analysis of hIgG by CD133⁺ cells using the same protocol revealed a similar pattern of trafficking (Supplementary Fig. S3C and S3D). In sections from three different tumor-bearing mice administered bevacizumab, we found colocalization of bevacizumab with Rab4 or LAMP1 in perivascular tumor cells (Fig. 4E and F), suggesting similar trafficking of bevacizumab in a xenograft mouse model of GBM as found in CD133⁺ cells in vitro. Similar trafficking of rat IgG was found in perivascular tumor cells in the established immunocompetent model of GBM treated with rat anti-mouse-VEGF-A (Supplementary Fig. S3E and S3F).

FcRn is known to recycle human IgG in ECs (13, 14). We confirmed that FcRn expression was detectable in isolates of normal brain ECs (376 and 422) by Western blotting using two different FcRn-specific antibodies. However, FcRn expression was not detectable in CD133⁺ cell isolates on Western blotting using either of these antibodies (Fig. 4G and Supplementary Fig. S3G and S3H). FcRn was detectable in the isolates of the paired CD133⁻ nonstem tumor cells (Fig. 4G). On stripping of the membrane and reprobing for Sox2, Sox2 expression was detected in the CD133⁺ cells, whereas a faint band representing Sox2 was detectable in one isolate of the CD133⁻ nonstem tumor cells (Fig. 4G). Sections from five xenograft tumors were double-labeled for FcRn and vWf, or FcRn and Sox2, and we found 72% of the FcRn fluorescence (adjusted mean) colocalized with vWf in endothelial cells and 16% of the FcRn fluorescence (adjusted mean) colocalized with Sox2⁺ cells (Supplementary Fig. S4), suggesting that the large majority of Sox2⁺ cells do not express FcRn in vivo. As it has been suggested that FcγRIIB can recycle immune complexes in dendritic cells (36), we also examined expression of FcγRIIB on the CD133⁺ cells. Myeloma cells (MM.1R) transduced with FcγRIIB were used as a positive control. FcγRIIB was not detected on the CD133⁺ cells (Supplementary Fig. S3G and S3H).

Bevacizumab binds VEGF-A with high affinity and CD133⁺ cells from GBM synthesize and secrete VEGF-A. Potentially, the bevacizumab-VEGF-A complex could be trafficked differently from free bevacizumab. To characterize the trafficking of bevacizumab-VEGF complexes in CD133⁺ cells, biotinylated-VEGF-A was incubated with bevacizumab for 1 hour and the mixture added to the CD133⁺ cells for 5 minutes, and either fixed or fresh media added for 3 hours before fixing. In this experiment, the Mander's coefficient was calculated as a measure of the percentage of bevacizumab (green) colocalized with VEGF (magenta) that was colocalized with Rab4 or LAMP1 (red; ImageJ with JACoP-plug-In). At 5 minutes, 76% of the bevacizumab-VEGF complex was colocalized with Rab4 (Supplementary Fig. S5A and S5C), and at 3 hours, 47% was colocalized with LAMP1 (Supplementary Fig. S5B and S5D). These data suggest that in the GBM CD133⁺ cells, the unbound bevacizumab and the bevacizumab–VEGF-A complex are trafficked similarly.

Bevacizumab neutralizes VEGF-A and, thus, could induce VEGF-A-depletion in treated cells. VEGF-C also binds VEGFR2 and activates its signaling (37). Bevacizumab can induce expression of PGF (a VEGF family member), which binds and activates VEGFR1 and may indirectly activate VEGFR2 through several mechanisms (38, 39). Therefore, we examined the expression of VEGF-A&C and PGF in the CD133⁺ cells plated as above (in the absence of added growth factors) and treated with bevacizumab. VEGF-A, VEGF-C, and PGF were detected in the media of CD133⁺ cells, and the levels increased from 24 to 48 hours (Fig. 5A–C). With bevacizumab treatment, the levels of VEGF-A were markedly reduced whereas the levels of VEGF-C and PGF were unchanged (Fig. 5A–C). Although there was a reduction in phospho-VEGFR2 in the CD133⁺ cells on bevacizumab depletion of VEGF-A, phospho-VEGFR2 was still detectable (Supplementary Fig. S6A and S6B).

To determine whether bevacizumab depletion of VEGF-A induced autophagy in the CD133⁺ cells, we examined the colocalization of LC3 puncta with the lysosomal marker LAMP2 (33), as well as examining the effect of bevacizumab on cell cytotoxicity/death and viability/proliferation. The concentration of bevacizumab we used had been reported previously to induce autophagy in GBM and cancer stem cell-like GBM cells (40). We found significantly higher autophagy and no change in cell death in bevacizumab-treated CD133⁺ cells as compared to vehicle at 24 hours (Fig. 5D and E). Treatment with hIgG at the same concentration decreased autophagy and had no effect on cell death at 24 hours (Fig. 5D and E). Downregulation of VEGF-A with siRNA in the CD133⁺ cells increased the colocalization of LC3 puncta with LAMP2 as compared to the control siRNA at 48 hours (Supplementary Fig. S6C), suggesting bevacizumab depletion of VEGF-A in the media causes a growth factor-starvation-induced autophagy. To optimally assess viability/proliferation, an MTT assay was performed at 48 hours, and indicated that bevacizumab treatment resulted in higher viability/proliferation as compared to the vehicle control (Fig. 5F). These results suggest macropinocytosis of IgG, followed by its lysosomal trafficking and nonspecific degradation provided basic building blocks that promoted viability/proliferation. Blocking macropinocytosis of bevacizumab with EIPA treatment at 48 hours significantly increased autophagy and cell death (Fig. 5G and H). EIPA treatment also increased autophagy and cell death in vehicle-treated CD133⁺ cells, but the increase in cell death was greater in the absence of bevacizumab treatment (Fig. 5G). Treatment with BafilomycinA1, which blocks lysosomal acidification and autophagy (33), caused a significant reduction in colocalization of LC3 puncta with LAMP2 in the bevacizumab-treated CD133⁺ cells and an increase in cell death (Fig. 5G and H). Taken together, these data suggest that macropinocytosis and lysosomal trafficking of bevacizumab and other proteins promotes CD133⁺ cell survival and proliferation in vitro, and that autophagy induced by bevacizumab depletion of VEGF-A also promotes survival.

Transcytosis (reviewed in ref. 41) across ECs is regulated in normal brain, but potentially could be abnormal in TECs. Using a standard in vitro transcytosis assay for normal brain ECs (41), we compared transcytosis of bevacizumab across monolayers of normal brain ECs and TECs. Quantitation of bevacizumab in the lower chamber by ELISA assay showed that ∼30% of bevacizumab was transcytosed across both normal brain ECs and TECs over 2 hours (Supplementary Fig. S7A). There was a two-fold larger permeability coefficient for 70-kDa-FITC-Dextran in TECs as compared to the normal brain ECs (Supplementary Fig. S7B). Both normal brain ECs and TECs internalized bevacizumab over 30 minutes (Supplementary Fig. S7C and S7D). Collectively, these data suggest that transcytosis of bevacizumab is not enhanced in TECs, supporting the concept that bevacizumab gains access to the perivascular tumor space in GBM due to alterations in the BBB.

We demonstrate that bevacizumab gains access to the perivascular tumor niche in established orthotopic mouse models of GBM through the well-described alterations in the BBB, suggesting that vascular normalization by bevacizumab does not occur in 100% of tumor vessels. We found that the CD133⁺/Sox2⁺ cells and the paired nonstem tumor cells can internalize bevacizumab but do so through different mechanisms and that, in vivo, Sox2⁺/CD44⁺ perivascular tumor cells appear to largely be responsible for anti-VEGF-A IgG internalization. We further show that the intracellular trafficking of anti-VEGF-A IgG in GBM tumors affects the survival of CD133⁺/Sox2⁺ cells.

Under the conditions used in these experiments, the CD133⁺ cells internalized bevacizumab and hIgG through the process of macropinocytosis, based on our observation of actin-enriched plasma membrane ruffles surrounding and engulfing bevacizumab into large vesicles and significant inhibition of internalization by EIPA. Although macropinocytosis appeared to be the primary mechanism underlying the internalization of bevacizumab in the CD133⁺ cells, we cannot rule out the possibility of a minor role for another clathrin-independent mechanism. The paired nonstem tumor cells internalized much less bevacizumab and, as the internalization was not affected significantly by treatment with EIPA, macropinocytosis did not appear to contribute to the internalization in these cells.

Analysis of the intracellular trafficking of bevacizumab and hIgG in vitro showed colocalization with established markers of endocytic compartments in CD133⁺ cells, providing clues to the potential fate of the internalized bevacizumab. Under the experimental conditions without added growth factors, bevacizumab was largely colocalized with a marker of a “fast” recycling compartment (Rab4) at 5 minutes. This would suggest that a considerable amount of the internalized bevacizumab is recycled rapidly to the extracellular environment where it would be available to bind and neutralize VEGF-A. FcRn has been shown to be responsible for the recycling of endogenous IgG in ECs and several other cell types (reviewed in ref. 13). The time course for bevacizumab and human IgG recycling by the CD133⁺ cells was faster than has been described for FcRn (14). Moreover, we found that the CD133⁺/Sox2⁺ cells do not express FcRn by Western blot analysis and that the large majority of Sox2⁺ perivascular tumor cells do not express FcRn in vivo. Thus, FcRn was unlikely to play a role in the recycling of bevacizumab and hIgG in the CD133⁺ cells. Similarly, the inhibitory FcγRIIB has been shown to endocytose and recycle immune complexes back to the cell surface on dendritic cells (36), but we did not detect FcγRIIB on the CD133⁺ cells by Western blotting. The isoelectric point (pI) of humanized IgG mAbs has been shown to affect clearance with a higher pI promoting clearance in a manner that was FcRn independent and presumed to be due to an increased rate of fluid phase endocytosis (reviewed in ref. 42). Bevacizumab has a higher pI (pI = 8.3; http://www.drugbank.ca) than endogenous IgG (pI = 7.2). Although we did not compare the rates of internalization of bevacizumab and hIgG, we did find similar trafficking patterns in the CD133⁺ cells.

Delivery of the endosome cargo to the late endosome/lysosome is slower than its delivery to the recycling compartment (17). We found that in the CD133⁺ cells a large amount of bevacizumab colocalized with a marker of the late endosome/lysosome (LAMP1) at 3 hours. This is consistent with the report of localization of bevacizumab to the lysosome in retinal ECs at 72 hours posttreatment in vitro (43). This localization to the lysosome suggests that the CD133⁺ cells in the perivascular tumor space also degrade bevacizumab. The percentage of bevacizumab localized to the LAMP1 compartment is probably an underestimate as that pool of bevacizumab is undergoing degradation. In the early endosome, differential sorting of cargo protein to a recycling compartment or the lysosome can be based on specific cargo protein signals (17), including ubiquitination that signals sorting of membrane proteins to the lysosome (44).

The trafficking of the bevacizumab–VEGF-A complex was similar to the trafficking of bevacizumab alone in the CD133⁺ cells; a fraction colocalized with a marker of the “fast” recycling compartment (Rab4) at 5 minutes (75%) and a fraction colocalized with a marker of the late endosome/lysosome (LAMP1; 47%) at 3 hours. In ECs, the VEGF–VEGFR2 complex is known to be internalized and recycled by different mechanisms, including clathrin-coated pit and endophilin-dependent mechanisms (45, 46), but in tumor cells the internalization and trafficking of the VEGF–VEGFR2 complex is not as well-defined.

We show that bevacizumab trafficking directly affects two survival pathways in the CD133⁺ cells. First, bevacizumab in the perivascular space depletes CD133⁺ cell-secreted VEGF-A causing a growth factor-deprivation-induced autophagy. This is demonstrated by the significantly higher number and area of LC3-LAMP2 puncta when the CD133⁺ cells are treated with bevacizumab as compared to vehicle; an increase in numbers of LC3-LAMP2 puncta when VEGF-A is downregulated with siRNA but not when the cells are treated with control siRNA; and a decrease in LC3-LAMP2 puncta on treatment of the cells with hIgG. BafilomycinA1 treatment (inhibits lysosomal acidification) blocked the increase in number and area of LC3-LAMP2 puncta in CD133⁺ cells treated with bevacizumab. The significant decrease in cell death with bevacizumab treatment at 48 hours supports the concept that bevacizumab induces a prosurvival form of autophagy in the CD133⁺ cells. Autophagy is generally a survival mechanism by which cells that are stressed due to nutrient deprivation, absence of growth factors, or hypoxia metabolize organelles, misfolded proteins and aggregated proteins, thereby generating molecules that allow the cell to survive (33, 47). Previously, others have reported an increase in autophagy on bevacizumab treatment of GBM cells cultured in media with growth factors or under hypoxic conditions (40, 48), but the effect of bevacizumab on VEGF-A secreted by tumor cells was not examined. Second, our data suggest that the macropinocytosis of bevacizumab followed by its targeting to the lysosome promotes the survival of CD133⁺ cells due to the increased availability of basic building blocks (amino acids) for the cell. This is supported by the increased viability/proliferation we observed with bevacizumab treatment, and the significant increase in cell death we observed on inhibition of macropinocytosis using the inhibitor EIPA. EIPA treatment also significantly increased autophagy presumably by blocking the macropinocytosis of nutrients and bevacizumab by the cell. The increase in cell death and autophagy taken together point to the possibility that EIPA induces autophagic cell death. The increase in cell death associated with EIPA treatment was significantly lower in bevacizumab-treated CD133⁺ cells than vehicle-treated CD133⁺ cells, and this is likely due to the protective growth factor-deprivation-induced autophagy that bevacizumab promotes in the CD133⁺ cells.

Macropinocytosis is highly regulated and thought to be stimulated by growth factors (16, 17). The macropinocytosis of bevacizumab by CD133⁺ cells occurred in the absence of exogenous growth factors, suggesting that an autocrine or paracrine growth factor pathway may be promoting this process, such as the VEGF/VEGFR2 pathway or another growth factor signaling pathway. VEGF-A has been reported to be a pro-angiogenic growth factor secreted by CD133⁺ cells from GBM (7). Our data support this concept and further suggest that the CD133⁺ cells are addicted to VEGF-A. When VEGF-A is depleted by bevacizumab treatment, the cells undergo autophagy. Despite depletion of VEGF-A with bevacizumab, phosphorylated-VEGFR2 was still detectable in CD133⁺ cells, although the level of phosphorylation was reduced. This could be due to CD133⁺ cell secretion of VEGF-C, and/or to a ligand-independent VEGFR2 signaling mechanism. Growth factor receptors also can be macropinocytosed by cells, for example, the macropinocytosis and lysosomal targeting of the EGF receptor (EGFR) when bound by a combination of mAb anti-EGFR (cetuximab) and anti-human IgG has been reported in large vessel ECs transfected with EGFR (49).

Our studies of the trafficking of anti-VEGF antibody when administered to established xenograft and syngeneic mouse models of GBM indicate the IgG gains access to the perivascular tumor tissue and is internalized predominantly by perivascular Sox2⁺/CD44⁺-tumor cells. Furthermore, in these cells, a fraction of bevacizumab is colocalized with Rab4 and a fraction with LAMP1, suggesting similar trafficking of bevacizumab in vivo as we found in CD133⁺/Sox2⁺ cells in vitro. Supporting our in vitro data, in bevacizumab-treated xenograft tumors there was increased colocalization of LC3 puncta with LAMP2 in perivascular tumor cells as compared to placebo-treated tumors, suggesting bevacizumab induced autophagy in vivo in the xenograft model. Other investigators have shown that bevacizumab treatment induces autophagy in tumor cells in GBM xenograft models and this was attributed to upregulation of interferon regulatory factor-1 or hypoxia in tumor cells (40, 48).

In summary (Fig. 6), our data suggest that (i) CD133⁺/Sox2⁺ cells in media without addition of growth factors macropinocytose bevacizumab, and a fraction is trafficked to a “fast” recycling compartment independent of FcRn and a fraction to the lysosome; (ii) bevacizumab trafficking promotes CD133⁺ cell survival through two pathways: the macropinocytosis and lysosomal trafficking of bevacizumab that generates basic building blocks for the cell and growth factor deprivation-induced autophagy; and (iii) in vivo, anti-VEGF-A antibody localizes to the perivascular tumor tissue suggesting it can target tumor-secreted VEGF, is similarly trafficked in perivascular tumor cells and induces a prosurvival autophagy. It should be noted that we are not suggesting bulk effects on the levels of bevacizumab. Rather, we demonstrate a significant effect on the survival of a population of tumor cells in the perivascular microenvironment that are resistant to radiation and chemotherapy and have been suggested to be cancer stem cell-like cells. We have not addressed whether the uptake of bevacizumab could be affecting other clinically relevant processes, such as the expression of GLUT3 (50), the consequences of the survival of the perivascular cancer stem-like cells after bevacizumab treatment, or the mechanisms that trigger their subsequent expansion and potential aggressive invasiveness (2). As bevacizumab also is used to treat renal, colon, and lung cancer and as we observed similar trafficking of bevacizumab and hIgG, these studies may have applicability to other cancers in which FcRn is not expressed on cancer stem cell-like cells or tumor cells and to other mAb therapeutics.

M.S. Ahluwalia is a consultant/advisory board member for Abbvie, Astra Zeneca, Bristol-Myers Squibb, CBT Pharmaceuticals, Caris Lifesciences, Incyte, and Monteris Medical, and reports receiving commercial research support from Bristol-Myers Squibb, Novartis and Novocure. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G. Müller-Greven, M.S. Ahluwalia, C.J. Herting, M. Bredel, J.N. Sarkaria, P. Hamerlik, C.L. Gladson

Development of methodology: G. Müller-Greven, C.R. Carlin, C.L. Gladson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Müller-Greven, M.E. Burgett, A. Lauko, A.S. Nowacki, M.A. Qadan, S.A. Toms, J.D. Lathia, D. Hambardzumyan, J.N. Sarkaria, P. Hamerlik, C.L. Gladson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Müller-Greven, M.E. Burgett, A.S. Nowacki, M. Bredel, J.D. Lathia, C.L. Gladson

Writing, review, and/or revision of the manuscript: G. Müller-Greven, C.R. Carlin, M.S. Ahluwalia, A. Lauko, A.S. Nowacki, M. Bredel, S.A. Toms, J.N. Sarkaria, C.L. Gladson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.L. Gladson

Study supervision: C.L. Gladson

We thank Dr. Judy Drazba (Imaging Core) for assistance, Dr. Fiona Hunter for critically reading the manuscript, and Drs. Qing Yi and Qiang Wang for the gift of MM.1R cells transduced with FcγRIIB.

This research was supported by NIH grants R01CA175120 (to C.L. Gladson), U01 CA160882 (to D. Hambardzumyan), R01CA176830 and Mayo Brain Tumor SPORE P50 CA108961 (to J.N. Sarkaria), and Velosano Foundation funding (to C.L. Gladson). The Leica-SP5-confocal/multi-photon microscope utilized was purchased with partial funding from NIH SIG-grant 1S10RR026820-01.

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.