Mechanisms of evasive resistance to anti-VEGF therapy in glioblastoma

SUMMARY

Angiogenesis inhibitors targeting the VEGF signaling pathway have been US FDA approved for various cancers including glioblastoma (GBM), one of the most lethal and angiogenic tumors. This has led to the routine use of the anti-VEGF antibody bevacizumab in recurrent GBM, conveying substantial improvements in radiographic response, progression-free survival and quality of life. Despite these encouraging beneficial effects, patients inevitably develop resistance and frequently fail to demonstrate significantly better overall survival. Unlike chemotherapies, to which tumors exhibit resistance due to genetic mutation of drug targets, emerging evidence suggests that tumors bypass antiangiogenic therapy while VEGF signaling remains inhibited through a variety of mechanisms that are just beginning to be recognized. Because of the indirect nature of resistance to VEGF inhibitors there is promise that strategies combining angiogenesis inhibitors with drugs targeting such evasive resistance pathways will lead to more durable antiangiogenic efficacy and improved patient outcomes. Further identifying and understanding of evasive resistance mechanisms and their clinical importance in GBM relapse is therefore a timely and critical issue.

Practice Points.

• Bevacizumab prolongs progression-free survival and dramatically improves quality of life for glioblastoma (GBM) patients; however, nearly all patients inevitably develop resistance during treatment. Bevacizumab does not frequently extend overall survival to a significant extent and patients who fail bevacizumab respond poorly to further treatment. Tumor progression following an initial response period reflects an adaptive response by tumors to evade anti-VEGF therapy.

• At least two distinct radiographic recurrence patterns have been described in bevacizumab-resistant GBM following initial responses. Some patients exhibit contrast-enhancing regrowth at the original tumor site, similar to the typical pattern of GBM recurrence and consistent with revascularization. Other patients develop nonenhancing fluid-attenuated inversion recovery-bright tumors, indicative of enhanced infiltrative progression without the necessity to reinitiate angiogenesis.

• We propose two general modes of adaptive resistance to antiangiogenic therapy in GBM: a proangiogenic mode, in which tumors evade anti-VEGF therapy by upregulating alternate angiogenic signaling pathways to re-establish neovascularization and growth; and a proinvasive mode, in which tumors forego revascularization in favor of increased distant invasion.

• Both phenotypes are likely to be generated by a variety of mechanisms that are still not fully understood. In addition, tumors may also induce survival cues to endure antiangiogenic intervention and eventually re-initiate bulk tumor growth. Such mechanisms include survival or enrichment of glioma stem cells and autophagy.

• The identification and our understanding of the mechanisms underlying evasive resistance to VEGF pathway inhibitors is just beginning and will remain an ongoing challenge, as will the assessment of their clinical prevalence and importance. Multiple mechanisms and modes of evasion are likely to be simultaneously induced. Combining angiogenesis inhibition with drugs that target such evasive resistance mechanisms may provide more durable efficacy and improved patient outcome.

Neovascularization, the formation of new blood vessels, has long been recognized as a hallmark of cancer [1]. Indeed, vascular proliferation is one of the defining pathologic features of glioblastoma (GBM), along with high expression levels of various proangiogenic factors, making it among the most highly vascularized human tumors [2]. In particular, expression levels of VEGF in gliomas correlate with higher-grade malignancy and poor prognosis [3,4]. Recognition of the role of VEGF as an important regulator of pathologic angiogenesis and subsequent tumor growth in most tumor types, including GBM, has provided a convincing rationale for the development of VEGF and VEGF receptor (VEGFR) inhibitors, with the goal of blocking the ‘angiogenic switch’ to restrict tumor growth and progression [1,5,6]. Alongside small-molecule receptor tyrosine kinase inhibitors that block the kinase domain of VEGFRs and other kinases, bevacizumab (Avastin^®; Genentech, CA, USA) is a humanized monoclonal antibody that specifically recognizes and blocks VEGF-A, and it became the first antiangiogenic treatment to be approved for use in cancer [7]. Bevacizumab in combination with chemotherapy experienced early clinical trial success by demonstrating prolonged overall survival in Phase III trials of metastatic colorectal cancer [8] and non-small-cell lung cancer [9], as well as prolonged progression-free survival (PFS) in renal cell carcinoma.

Antiangiogenic therapy in GBM

In GBM, bevacizumab has been shown to induce dramatic reductions in tumor contrast enhancement and improve the time to progression when administered either alone or in combination with chemotherapy. Numerous studies have observed radiological response rates between 21.2 and 61% for recurrent GBM, coupled with 6-month PFS rates of 30–51% [10–12]. These results are unprecedented in other Phase II trials for recurrent GBM, where response rates of 5–20% and 6-month PFSs of 15–20% are more typical [13–15]. Encouraging results in Phase II clinical trials of bevacizumab alone or in combination with irinotecan in recurrent GBM led to its accelerated approval by the US FDA for use in recurrent GBM in 2009 [10], making bevacizumab only the third FDA-approved drug for GBM [16]. VEGFR inhibitors have also demonstrated activity against GBM, particularly the pan-VEGFR tyrosine kinase inhibitor cediranib (AZD2171 or Recentin™; AstraZeneca, London, UK), for which radiographic response rates were roughly 50% and the 6-month PFS rate 27% [17]. Unfortunately, a Phase II study of VEGF Trap (aflibercept; Regeneron, NY, USA), a soluble decoy receptor fusion protein that scavenges VEGF and PlGF, did not demonstrate improved 6-month PFS, although patient attrition due to toxicity may have contributed to the lower PFS rate [18]. To date, there have been no completed Phase III randomized trials for antiangiogenic agents in GBM, although two large randomized Phase III trials evaluating the addition of bevacizumab to the current standard of care in newly diagnosed GBM are currently ongoing ([19] and RTOG 0825 [201]).

Although not initially recognized, one of the most compelling benefits of bevacizumab and cediranib, beyond their intended roles of eliciting tumor reduction or stasis, is their ability to reduce peritumoral edema and the need for steroids. Significant improvements in the quality of life of GBM patients is thought to be largely attributable to the reduction of steroid dose it permits, which in turn results in robust improvements in patient independence [20]. It is important to note that the ability of anti-VEGF therapy to decrease tumor vascular permeability has complicated the assessment of GBM response or progression by radiologic imaging inferred from contrast uptake [21]. Often, a rapid decrease in contrast enhancement is observed as early as 1 or 2 days following initiation of anti-VEGF therapy; however, such responses are more likely due to vascular normalization and not an antitumor effect [17]. The importance of considering both T₁-weighted contrast enhancement as well as T₂- or fluid-attenuated inversion recovery (FLAIR)-weighted MRI in assessing antiangiogenic response has emerged, as nonenhancing FLAIR-bright tumors are indicative of a different, more infiltrating growth pattern [22]. Newly defined criteria by the Response Assessment in Neuro-Oncology Working Group, an international and multidisciplinary effort to develop new standardized response criteria for clinical trials in brain tumors, have provided updated guidelines that incorporate FLAIR imaging to cope with the unique imaging challenges posed by antiangiogenic agents [23].

While the primary response rate to bevacizumab in GBM patients is unparalleled compared with other drugs evaluated in Phase II trials for recurrent GBM, it has become evident that a small proportion of patients exhibit intrinsic resistance and do not respond to VEGF inhibition, while those patients who do respond experience only transitory benefits before their tumors eventually become resistant and progress during bevacizumab treatment. Intrinsic resistance may involve similar molecular and cellular mechanisms as those that mediate evasive resistance, likely due to pre-existing microenvironmental pressures during malignant progression.

For the many patients who do initially respond, the striking improvement in PFS observed in most bevacizumab-treated GBM patients has not necessarily translated into better overall survival [24], suggesting that tumors can recur in a much more rapid and aggressive manner when they overcome antiangiogenic inhibition. However, there is a seminal difference to mechanisms of acquired drug resistance conferred by mutation of the oncogenes being targeted [25], or by alterations in drug uptake and efflux [26,27], because GBM, like other angiogenic tumors, appears to adapt to the presence of VEGF inhibitors by utilizing alternative pathways to sustain tumor growth, all while VEGFR signaling remains inhibited [28–30].

Unfortunately, patients in whom bevacizumab therapy eventually fails tend to do poorly afterward and do not respond well to further treatment. The optimal strategy to treat patients progressing after first-round bevacizumab therapy has yet to be determined, but many patients continue on bevacizumab with a change in the chemotherapeutic agent. A retrospective review of 54 high-grade gliomas predominantly given bevacizumab and carboplatin as salvage therapy after progression on first-round bevacizumab treatment found no meaningful 6-month PFS or any partial response [31]. Similarly, another study in which bevacizumab-resistant GBM patients discontinued bevacizumab and instead received various salvage chemotherapies also reported minimal efficacy [32]. These results underscore the need to identify mechanisms of evasive resistance to anti-VEGF treatment so that alternate therapies can be used in conjunction to prolong antiangiogenic efficacy and improve patient outcome.

In this review, we discuss recent preclinical work that illuminates mechanisms of adaptive evasion to antiangiogenic treatment in GBM and relate them to clinical observations when available. We propose two general phenotypes of adaptive resistance in GBM that can be caused by various mechanisms: a proangiogenic mode in which tumors upregulate alternate proangiogenic signaling pathways to re-establish neovascularization and bulk growth, and a more proinvasive mode in which tumors forego revascularization in favor of increased distant invasion.

Patterns of evasive resistance to antiangiogenic agents in GBM

In general, tumors are thought to escape antiangiogenic therapy by either re-establishing neovascularization through various different means, or by altering their behavior to propagate and progress without the need to satisfactorily reinitiate angiogenesis [29]. Given the critical role of VEGF in vascular development and homeostasis, it is not altogether surprising that inhibition of its activity can lead to compensatory upregulation of other growth factors and cytokines [33]. This induction could be generated as a direct consequence of tumor cells sensing a loss of VEGF signaling, or by the effects of VEGF inhibition on the tumor microenvironment, such as induction of hypoxia and hypoxia-inducible factors (HIFs) that regulate a wide range of tumor-promoting pathways. Hypoxia can also drive the recruitment of various bone marrow-derived cells (BMDCs) that also have the capacity to express proangiogenic factors or directly contribute to neovasculature.

In addition, several preclinical mouse models have shown that tumors may recur with increased invasion or metastasis after antiangiogenic therapy. In fact, increased tumor invasion following suppression of VEGF signaling was first observed using murine models of GBM [34–37], a phenotype that has been consistently recapitulated in numerous preclinical studies of GBM [38–41]. Proinvasive or metastatic evasion has also been described in other mouse tumor models including pancreatic neuroendocrine tumors (PNETs) [42,43] and breast cancer [44]. Conceptually, one can envision that, in the absence of revascularization, tumor cells adapt by invading from unfavorable microenvironmental conditions out into normal tissue, utilizing the normal vasculature for sustenance.

Although much work remains to be done to fully establish the clinical existence and prevalence of such evasive patterns across different human cancers, there are strong clinical indications that the inevitable progression of GBM during antiangiogenic therapy presents with at least two distinct radiographic patterns, representing proangiogenic and proinvasive mechanisms of evasion (Figure 1) [16]. Following initial response to bevacizumab and the associated decrease in tumor contrast enhancement, some resistant tumors exhibit nodular contrast-enhancing regrowth at the initial site of disease concomitant with restoration of vascular density and increased proliferation, reminiscent of proangiogenic evasion. By contrast, other patients who fail bevacizumab exhibit nonenhancing, FLAIR-bright tumors that show no signs of revascularization and are indicative of a more infiltrative progression [16,45]. This proinvasive recurrence encumbers surgical resection of recurrent GBM and further challenges therapeutic options for patients.

Figure 1. Mechanisms of evasion of antiangiogenic therapy in glioblastoma.

A subset of GBM patients treated with antiangiogenic agents exhibit intrinsic resistance and fail to show any response. By contrast, many GBM do benefit from angiogenesis inhibitors; however, the response phase is transitory and followed by the induction of evasive resistance pathways, in which tumors adapt to bypass angiogenic blockade. Two main patterns of evasive recurrence following initial antiangiogenic response in GBM have been observed. In the proangiogenic pattern, relapsing tumors display contrast enhancement on MRI and activate alternate proangiogenic factors or recruit various BMDCs to reinitiate neovascularization, even while VEGF signaling remains inhibited. In the proinvasive evasion pattern, relapsing tumors are not contrast enhancing but rather show increasing FLAIR-bright volumes indicative of more infiltrating growth. Proinvasive evasion mechanisms directly affect tumor cell behavior in response to antiangiogenic treatment to facilitate tumor cell invasion. GBM stem-like cells may also survive or be enriched by antiangiogenic therapy, eventually leading to tumor regrowth or invasion.

BMDC: Bone marrow-derived cell; EPC: Endothelial progenitor cell; FLAIR: Fluid-attenuated inversion recovery; GBM: Glioblastoma; GSC: Glioblastoma stem cell; MMP: Matrix metalloproteinase; PPC: Pericyte progenitor cell; SDF-1α: Stromal-derived factor-1α.

The incidence of invasion following bevacizumab therapy in GBM has been debated, in part owing to the current lack of a standardized definition of radiographic relapse [46–48]. Nevertheless, the frequency of invasive nonenhancing tumors appears to be higher than would be expected in patients who do not receive bevacizumab [32]. Historically, 5–10% of recurrent or progressive GBMs appear outside of the defined contrast-enhancing primary tumor site [13,49,50], while a more recent study coupling MRI with size and location measurements found that up to 20% of GBMs treated with conventional chemoradiotherapy and temozolomide recur distantly [51]. Numerous retrospective studies across multiple institutions have now reported a range of incidence between 30 and 60% for diffuse or distant progression in recurrent GBM following bevacizumab resistance (Table 1) [22,32,52–54]. Similarly, early clinical trials investigating bevacizumab as a front-line therapy for newly diagnosed GBM in combination with radiotherapy and temozolomide showed promising PFS, but also demonstrated an increased proinvasive recurrence in approximately 60% of patients who eventually failed bevacizumab [55]. More importantly, downstream studies have confirmed that tissue resected from radiologically diffuse, noncontrast-enhancing tumors correlates with invasive histopathology [38,45].

Table 1. . Incidence of invasive recurrence in bevacizumab-resistant glioblastoma.

Author (year)	Treatment	Median PFS (months)	Median OS (months)	Incidence of diffuse or distant recurrence in relapsed patients (%)	Patients studied (n)	Ref.
Recurrent GBM
Historical	XRT + chemotherapy	1.8–2.3	6.8	5.00–20.00	–	[15,49–51,121]
Norden et al. (2008)	Bev + chemotherapy	4.4	N/A	31.00	55	[22]
Narayana et al. (2009)	Bev + irinotecan or carboplatin	5.0	9.0	30.00	61	[52]
Narayana et al. (2012)	Bev + irinotecan + carboplatin	7.1†	15.4†	48.10	104	[53]
Iwamoto et al. (2009)	Bev + irinotecan or XRT	N/A	N/A	51.00	37	[32]
Zuniga et al. (2009)	Bev + irinotecan	7.6	11.5	60.53	51	[54]
Newly diagnosed GBM
Narayana et al. (2012)	Bev + XRT + TMZ	7.1†	15.4†	50.00	58	[53]
Narayana et al. (2012)	Bev + XRT + TMZ	13.0	23.0	57.10	51	[55]

^†Survival data collectively includes grade III and IV recurrent gliomas as well as newly diagnosed GBM.

Bev: Bevacizumab; GBM: Glioblastoma; N/A: Data not reported; OS: Overall survival; PFS: Progression-free survival; TMZ: Temozolomide; XRT: Radiotherapy.

Interestingly, a more invasive or metastatic clinical phenotype has yet to be definitively demonstrated in other solid tumor types treated with antiangiogenic agents, despite the suggestion from preclinical animal models [42–44,56]. A recent study comparing anti-VEGF treatment using the B20 monoclonal antibody, an analog of bevacizumab, with the VEGFR inhibitor sunitinib, reported that anti-VEGF therapy did not induce tumor invasiveness or metastatic potential in genetically engineered mouse models of PNET, pancreatic ductal adenocarcinoma, non-small-cell lung cancer and small-cell lung cancer, whereas sunitinib did recapitulate increased invasion and metastasis in the PNET model [57]. The authors suggested that drugs targeting VEGF ligands, as opposed to VEGFRs and additional receptor tyrosine kinases, might differentially affect tumor progression and resistance in a context-dependent manner. However, murine GBM models treated with B20, bevacizumab, the VEGFR2-neutralizing antibody DC101, and the VEGFR inhibitors SU10944 and sunitinib all demonstrated increased invasiveness [36,38,41,42]. In the clinic, some patients treated with the VEGFR inhibitor cediranib also appeared to progress with a more infiltrative pattern [58]. Therefore, a proinvasive phenotype subsequent to antiangiogenic treatment and resistance has only been clinically confirmed in GBM at this point [56].

Mechanisms of antiangiogenic evasion & recurrence

▪ Proangiogenic mechanisms

A number of studies support the existence of a proangiogenic recurrence phenotype in human GBM treated with antiangiogenic therapies, and several different angiogenic factors that are induced following treatment have been identified (Figure 1). Immunohistochemical staining of human tumor specimens has revealed that vascular density and intratumoral hypoxia, as measured by carbonic anhydrase 9 and HIF-1, remain essentially unchanged in bevacizumab-evading nodular-enhancing GBMs compared with pretreatment samples from the same patients, while proliferation increases [16,45]. Gene-enrichment subset analysis showed that these contrast-enhancing relapsed tumors exhibited an increased expression of gene subsets associated with vascular injury, with specific transcriptional upregulation of VEGF-A, VEGF-C, MAPKs and AQP4 compared with pretreatment specimens. These observations suggest that these tumors are able to reacquire the vascularity and reduced hypoxia they had before bevacizumab treatment, consistent with proangiogenic recurrence [16,45].

U87MG GBM cells and NSC23 glioma stem cells treated with bevacizumab in vitro were found to upregulate the proangiogenic factors angiogenin, IL-1α, IL-1β, bFGF, TNF-α and TGF-α, both transcriptionally and at the protein level [59]. When U87MG cells were intracranially implanted as xenografts in mice who were then treated with bevacizumab, a significant increase in bFGF expression was observed at the time of tumor progression and at animal death, as detected by immunohistochemical staining. Interestingly, bFGF was not elevated in tumors analyzed at early time points that had only received short-term VEGF inhibition, thereby correlating bFGF induction with tumor relapse and indicating the existence of a compensatory mechanism enhancing neovascularization. Moreover, microvessel density at the early time point remained significantly reduced compared with control tumors, whereas bevacizumab-treated tumors at the late stage exhibited restoration of microvessel density above that observed in controls, suggesting that tumors could indeed reactivate angiogenesis, in part through bFGF upregulation, following long-term anti-VEGF therapy [59].

In a Phase II trial of cediranib in GBM, plasma bFGF levels increased by 59% during tumor progression of the first 16 patients treated; however, this was not confirmed in the later report of all 31 patients in the study [17,60]. Further study will be required to determine the extent to which bFGF contributes to antiangiogenic evasion in GBM at the clinical level. If it is found to be a significant contributor, it may represent a strong target in combination with anti-VEGF therapy. Upregulation of bFGF has also been speculated to contribute to antiangiogenic therapy evasion in other cancer models as well. In the Rig1–Tag2 model of pancreatic islet carcinogenesis, tumor progression and angiogenic reactivation in the face of VEGF blockade was independent of VEGF and associated with hypoxia-mediated induction of alternative proangiogenic factors including bFGF [28]. Accordingly, combined blockade of VEGF and bFGF with the dual inhibitor brivanib prolonged tumor stasis and angiogenic blockade when used as a first-line treatment or as a second-line therapy following previous antiangiogenic inhibition [61]. Interestingly, second-line brivanib treatment was more efficacious when initiated prior to first-line antiangiogenic failure and less beneficial when administered after tumors had already initiated revascularization, emphasizing the importance of recognizing impending anti-VEGF therapy failure.

PDGF-C, a key component of the PDGF receptor-α signaling pathway, has also been found to be strongly expressed in human GBM tissues during anti-VEGF therapy [62]. Overexpression of PDGF-C in U87MG cells resulted in intracranial tumors with improved vasculature stability, highlighted by reduced vascular permeability and more extensive perivascular support, while also conferring resistance to antiangiogenic inhibition with DC101. Treatment of wild-type U87MG xenografts with DC101 also led to increased murine PDGF-C expression, indicating that antiangiogenic treatment may induce the host microenvironment to increase PDGF-C production as well to sustain angiogenesis [62]. These results suggested that upregulation of PDGF-C may help tumors circumvent VEGF inhibition by promoting revascularization. In patient specimens from the Phase II clinical trial of cediranib, PDGF-C was highly expressed in post-treatment samples compared with a more focal expression pattern in untreated control cases [63]; however, a distinct correlation between PDGF-C level and recurrence was not formally reported.

Recent studies have also suggested the stromal-derived factor-1α(SDF-1α) chemokine pathway as a mechanism of tumor resistance to antiangiogenic therapy. SDF-1α is a hypoxia-regulated factor that can recruit BMDCs to and retain them within tumors, where they can indirectly facilitate neovascularization and tumor growth. SDF-1α may also directly promote tumor angiogenesis via vasculogenesis, the de novo formation of blood vessels from marrow-derived CXCR4⁺ endothelial progenitor cells [64,65]. Our laboratory demonstrated in murine models of GBM that HIF-1-dependent SDF-1α activity can induce recruitment of a variety of proangiogenic CD45⁺ BMDCs, as well as endothelial and pericyte progenitor cells to promote neovascularization [39]. In this case, the SDF-1α-stimulated influx of CD45⁺ myeloid cells provided a critical source of matrix metalloproteinase (MMP)-9, which was essential to initiate angiogenesis by increasing VEGF bioavailability, whereas endothelial and pericyte progenitors can contribute directly but sparingly to tumor vasculature. Congruent with our results, an influx of CD11b⁺ BMDCs was observed following whole-brain irradiation of mice intracranially implanted with human U251 GBM cells, due in part to induction of HIF-1-driven SDF-1α secretion [66]. This influx restored radiation-damaged vasculature, thereby allowing resumption of tumor growth. Pharmacologic inhibition of the SDF-1α/CXCR4 signaling cascade with AMD3100 effectively prevented BMDC recruitment and postirradiation restoration of functional vasculature and tumor regrowth, whereas VEGF signaling blockade with DC101 after irradiation was much less effective. This suggests that the SDF-1α pathway and downstream vasculogenesis may represent an evasive mechanism when VEGF-dependent angiogenesis is impaired [66].

Interestingly, in the Phase II clinical trial of cediranib in GBM, an initial subset of patients who experienced radiographic tumor progression during antiangiogenic treatment was found to have 12% higher plasma SDF-1α levels [17]. These studies suggest that SDF-1α might be released into the blood and could function as a potential biomarker, but further clinical study of SDF-1α in GBM refractory to antiangiogenic therapy will be needed to strengthen these observations.

In addition to direct upregulation of alternative angiogenic factors by tumor cells, several studies have also suggested that the nontumor stromal compartment can produce factors to promote resistance to antiangiogenic therapy, thus adding another layer of mechanistic complexity. As described above, infiltrating myeloid BMDCs can support angiogenesis by expressing a variety of chemokines, cytokines and proteases. Gr1⁺CD11b⁺ myeloid BMDCs were one of the first subpopulations found to sufficiently confer inherent resistance to anti-VEGF treatment in several different murine xenograft tumor models, but GBM was not represented [67]. In this study, Gr1⁺CD11b⁺ BMDCs expressed the secreted proangiogenic protein Bv8, which was in turn regulated by the myeloid growth factor granulocyte colony-stimulating factor. Bv8 could mobilize Gr1⁺CD11b⁺ cells from the bone marrow and also directly promote tumor angiogenesis [68]. Combined treatment of refractory tumors with anti-VEGF and anti-granulocyte colony-stimulating factor or anti-Bv8 therapy reduced tumor growth compared with anti-VEGF monotherapy [69]. While these studies indicated that BMDCs conferred upfront resistance to anti-VEGF treatment, it is conceivable that these cells could also be mobilized in tumors that were initially sensitive to antiangiogenic therapy, thus contributing to evasive resistance.

Similarly, a study on the efficacy of sunitinib found increased levels of HGF primarily produced by BMDCs in the stroma of sunitinib-resistant murine xenograft tumors [70]. Expression of the HGF receptor c-Met in these models was predominantly found in vascular endothelial cells and not in tumor cells, therefore suggesting a role for HGF-stimulated c-Met signaling in endothelial cells to bypass anti-VEGF therapy.

▪ Proinvasive mechanisms

The HGF receptor tyrosine kinase c-Met not only activates endothelial cells, but also affects many cancer cells by promoting proliferation, scattering, invasion and survival. In GBM c-Met activity/expression is correlated with increased tumor invasion and poorer survival [71–75]. Several recent studies, including ours, have reported increased c-Met expression and/or activity in GBM correlating with increased tumor invasion following antiangiogenic treatment. Indeed, the HGF/c-Met axis appears to be a critical component in proinvasive evasion because hypoxia-driven as well as hypoxia-independent mechanisms are employed to establish enhanced c-Met activity in GBM and other tumor types during antiangiogenic therapy (Figure 2).

Figure 2. Regulation of c-Met activity in glioblastoma during anti-VEGF therapy.

VEGFR2 exists in a complex with c-Met and, upon VEGF stimulation, recruits the phosphotyrosine phosphatase PTP1B to suppress HGF-mediated c-Met activation on tumor cells. The presence of VEGF simultaneously promotes angiogenesis and tumor growth. Following VEGF signaling blockade through either VEGF or VEGFR2-targeted therapies, the eventual development of proinvasive evasion can be facilitated by a hypoxia-induced increase in HGF or c-Met expression by tumor cells, or by ligand-independent activation of c-Met. In addition, inhibition of VEGF signaling also represses its negative regulation of c-Met by disengaging PTP1B and thereby unleashing HGF/c-Met activity on tumor cells in a hypoxia-independent manner and without the need to upregulate total c-Met levels. The resultant increase in c-Met activation leads to increased tumor cell invasion and transformation towards a more mesenchymal and aggressive phenotype. The addition of c-Met or HGF inhibitors to anti-VEGF therapy may prevent proinvasive evasion and prolong antiangiogenic efficacy.

VEGFR2: VEGF receptor 2.

Hypoxic conditions have been shown to induce c-Met expression and subsequent invasion in tumor cell lines [76]. Similarly, c-Met transcription and protein expression were found to be increased in a HIF-1α-dependent manner following in vitro exposure to hypoxia in a panel of different GBM cell lines and primary cultures, which led to enhanced tumor cell migration upon HGF stimulation [77]. Analysis of GBM patient specimens obtained before and after bevacizumab treatment revealed increased intratumoral hypoxia in invasive recurrent tumors [32], concomitant with elevated c-Met expression [16]. Consistent with these results, prospectively obtained primary tumor cultures derived from proinvasive patient tumor specimens were substantially more invasive in Matrigel™ chambers than primary cultures from contrast-enhancing, proangiogenic tumors [16]. These results indicate that, during antiangiogenic treatment, the inability to reinitiate angiogenesis may cause a rise in intratumoral hypoxia that then induces expression of c-Met, among other factors, to drive tumor invasion as an evasive response. Interestingly, upregulation of c-Met expression and subsequent invasion has also been observed in a murine model of PNET treated with a neutralizing VEGF antibody, a phenotype that was blocked by combined inhibition of c-Met and VEGF signaling [43].

Our recent studies on the other hand revealed a hypoxia-independent mechanism, by which VEGF blockade enhanced c-Met phosphorylation and subsequent invasion in murine and human GBM [41]. We had first observed that murine GBM cells moved predominantly along blood vessels away from the main tumor mass and deep into the brain parenchyma when tumor cells were unable to initiate VEGF-dependent angiogenesis, either through genetic ablation of key angiogenic factors (HIF-1α, VEGF, MMP-9 and MMP-2) [34,39,78] or by pharmacologic targeting of VEGF signaling [42,44]. Because VEGF inhibition was a common denominator among these various genetic knockout models and pharmacologic treatments, it suggested that VEGF itself might act as a regulatory switch for invasion in GBM. These studies led to an unexpected link between HGF and VEGF signaling, in which VEGF directly and negatively modulated the activity of c-Met through the interaction of a novel c-Met–VEGFR2 heterocomplex on tumor cells [41]. VEGF-stimulated activation of VEGFR2 on GBM cells recruited the cytoplasmic phosphotyrosine phosphatase PTP1B toward HGF-activated c-Met, leading to c-Met dephosphorylation and suppression of cell motility (Figure 2). Subsequent to VEGF inhibition, HGF/c-Met activity was unleashed, increasing tumor cell migration and invasion in part through the induction of a cellular program reminiscent of epithelial–mesenchymal transition. In this scenario, we found that VEGF-dependent regulation of c-Met activity was independent of hypoxia because the total c-Met expression levels did not change despite differences in c-Met phosphorylation [41]. Congruently, c-Met phosphorylation was increased in human GBM specimens after bevacizumab treatment when compared with matched pretreatment samples. Moreover, we observed c-Met phosphorylation primarily at the invasive edges of murine tumors where there is little hypoxia, rather than within the tumor mass where low oxygen tension is more severe. Furthermore, we found that extensively hypoxic and necrotic irradiated murine GBMs were not more invasive than their nonirradiated counterparts. By contrast, GBMs treated with a broad-spectrum VEGFR inhibitor were only modestly more hypoxic than untreated tumors but exhibited the highest degree of invasiveness when compared with control and irradiated tumors [Lu KV, Bergers G, Unpublished Data]. In contrast to hypoxia-driven upregulation of c-Met expression, these findings illustrate the plasticity with which tumors can adapt to evade therapy, demonstrating distinct mechanisms by which the same signaling pathway can be activated.

Importantly, the proinvasive escape mechanism could be blocked by silencing both VEGF and c-Met in tumor cells, resulting in substantial survival benefit and a significantly reduced tumor invasion and growth of murine GBM [41]. The evidence supporting c-Met activity in mediating proinvasive evasion of antiangiogenic therapy in GBM therefore provides a strong rationale to combine HGF/c-Met and VEGF inhibitors, a strategy that may minimize or prevent tumor invasion while still reaping the benefits of bevacizumab treatment. Many different c-Met inhibitors are currently under clinical development and will likely be implemented in conjunction with anti-VEGF agents in the near future. The dual VEGFR2 and c-Met inhibitor XL184 (cabozantinib; Exilexis, CA, USA) is already a promising drug and interim results from a Phase II study for recurrent GBM indicate encouraging clinical activity with relatively few instances of distant or diffuse disease at progression, although more formal conclusions on its efficacy are awaited [79].

Another potential hypoxia-independent but HIF-1-driven mechanism for tumor invasion was recently revealed in a murine PNET model [80]. Knockout of VEGF expression in these tumors led to increased invasion into the surrounding exocrine tissue, a phenotype that required HIF-1α since VEGF and HIF-1α double knockouts abolished tumor invasion. HIF-1α expression correlated with lower E-cadherin and altered N-cadherin and NCAM expression, consistent with a more invasive phenotype. Most importantly, however, the requirement for HIF-1α in the invasive phenotype was not due to a response to acute hypoxia, indicating that HIF-1α may play additional regulatory roles that are not directly governed by microenvironmental oxygen tension. Such a role for HIF-1α has yet to be determined in GBM, although HIF-1α deficiency in our murine astrocytoma model still resulted in invasive tumors similar to VEGF-knockout GBM [34,39], likely due to reduced VEGF expression. However, we have not yet investigated the effects of VEGF and HIF-1α double-knockout GBM.

Induction of MMPs has been observed concomitant with bevacizumab-induced GBM invasion in several preclinical models. MMPs, specifically MMP-9 and -2, can facilitate collagen degradation to promote tumor invasion but MMPs are implicated in various aspects of tumorigenesis – including proliferation, angiogenesis and survival – that go far beyond degradation of the extracellular matrix [81,82]. Treatment of U87MG GBM cells and NSC23 glioma stem cells with bevacizumab increased Matrigel transwell invasion in vitro and induced transcriptional upregulation of invasion-associated genes such as MMP-2, MMP-9, MMP-12, SPARC and TIMP-1. Concordantly, enhanced expression of some of these molecules was found in tumor xenografts upon bevacizumab treatment [59]. Interestingly, the same study also found that bevacizumab-treated orthotopic U87MG xenografts upregulated bFGF expression, as described above, indicating that GBM may simultaneously elicit both proangiogenic and proinvasive pathway circuits to evade and overcome angiogenic inhibition. Combined treatment with bevacizumab and the MMP inhibitor GM-6001 suppressed GBM cell invasion in vitro but was ineffective in vivo, suggesting that the proinvasive phenotype is likely regulated by more complex cellular programs. Furthermore, it appears that MMP-2 and -9 have become more critical regulators of angiogenesis in various tumor types [39,83,84]. Indeed, several years ago we found that complete genetic knockout of MMP-2 and MMP-9 in GBM-bearing mice did not decrease but rather increased the invasiveness of GBM cells because both gelatinases were found to be critical for VEGF-dependent angiogenesis. Consistent with the inverse correlation between VEGF activity and invasion, MMP-9 or MMP-2 knockout GBMs were nonangiogenic but exhibited a proinvasive growth pattern [39,78].

Recently, de Groot et al. reported induction of hypoxia and MMP-2 and reduced vascular proliferation in bevacizumab-treated invading U87MG xenografts, an observation they found to be consistent in surgical specimens of nonenhancing infiltrating human GBMs refractory to bevacizumab [38]. Interestingly, a perivascular invasive pattern was observed in the murine GBM model, but not in any of the three human GBM cases studied, which exhibited diffuse invasion into neuropil. We have observed a tendency for perivascular invasion in our murine transformed astrocytoma model, in contrast to intrafascicular infiltration along white matter tracts in a murine neural stem cell GBM model [41]. Given that there are different patterns of GBM invasion, known as ‘secondary structures of Scherer’ [85], which include periaxonal, perivascular, subpial and perineuronal accumulations, it is therefore possible that a predominant pattern of invasion following bevacizumab resistance may be induced depending on the molecular genetic subtype of each individual tumor [86]; or perhaps multiple modes of tumor invasion can be simultaneously elicited under different microenvironmental conditions or pressures. The hypothetical mechanisms governing different invasive patterns may provide further insight into proinvasive recurrence.

In addition to the activation of invasion-promoting pathways directly on tumor cells themselves, it is possible that stromal cells may also contribute to increased invasion during antiangiogenic resistance. Apart from the proangiogenic effects of BMDCs recruited by tumors as described above, bone marrow-derived tumor-associated macrophages (TAMs) have been shown to enhance breast cancer invasion in a paracrine loop in which EGF produced by TAMs stimulates the migration and invasion of neighboring EGF receptor-expressing tumor cells. The breast cancer cells in turn express colony-stimulating factor 1, which acts as a potent chemoattractant for colony stimulating factor 1 receptor-expressing TAMs [87,88]. Similarly, in a mouse model of colorectal cancer in which TGF-β signaling is blocked, tumor cells secreted CCL9 to recruit a population of bone marrow-derived immature myeloid cells expressing the CCL9 receptor CCR1, as well as MMP-2 and MMP-9, leading to tumor invasion [89]. However, these mechanisms have yet to be reported in the context of antiangiogenic resistance; furthermore, they have not yet been described in GBM. Because GBMs have been shown to recruit BMDCs during antiangiogenic evasion, it will be interesting to determine whether they do indeed contribute to increased invasion. Other stromal constituents, such as microglia, the resident macrophages of the brain, have also been shown to promote GBM invasion [90,91]; however, their role in proinvasive recurrence after angiogenic inhibition has not been determined.

▪ Prosurvival mechanisms

It is conceivable that the evasive mechanisms described above to enable revascularization or tumor invasion during antiangiogenic therapy are likely complemented by mechanisms that facilitate tumor cell survival. Specifically, GBM stem cells (GSCs), which constitute a reservoir of tumorigenic self-sustaining cells with the exclusive ability to self-renew and maintain the bulk tumor, are thought to be refractory to standard therapies and responsible in large part for tumor resistance and recurrence. GSCs reside and are enriched in the perivascular niche of tumors where instructive cues from the tumor endothelium help maintain their self-renewal, proliferation and plasticity [92,93]. This, therefore, would imply that GSCs are sensitive to antiangiogenic therapies. While antiangiogenic agents have been shown to initially reduce GSC numbers and tumor growth by compromising the vascular niche [92,94], other studies have found that blocking angiogenesis enriches the cancer stem cell population via therapy-induced hypoxia in breast cancer [95] and that GSCs can persist and propagate orthotopic brain tumors in an angiogenesis-independent environment concomitant with upregulation of proinvasive genes [96]. Moreover, the acquisition of a more mesenchymal phenotype that favors invasive growth and predicts poorer survival in GBM may also lead to tumor cells with more stem-like properties [97,98]. Indeed, we found that c-Met signaling could drive mesenchymal transformation in GBM [41] and other reports have described HGF/c-Met as a mediator of tumor cell reprogramming into a more GSC-like phenotype [99], which has led to its proposition as a tumor stem cell marker in GBM [100,101]. Therefore, the enrichment of GSCs or tumor cells that gain GSC-like traits may constitute another way in which GBM and other tumor types overcome antiangiogenic therapies.

A conceptually novel mechanism for GSC resistance to bevacizumab therapy was recently reported in which a persistent autocrine VEGF–VEGFR2–neuropilin-1 signaling loop on GSCs allowed these cells to survive, proliferate and form tumors in the face of VEGF inhibition [102]. These results further support the concept that VEGF–VEGFR2 signaling can occur on tumor cells with downstream functional effects. Because the GSCs actively recycled VEGFR2 through the endosomal pathway and could produce increasing levels of VEGF to overcome bevacizumab activity, autocrine signaling through VEFR2 was maintained. Inhibition of VEGFR2 or VEGF–VEGFR2–neuropilin-1 attenuated GSC viability, suggesting that combined inhibition of both VEGF and VEGFR2 may be a complementary therapeutic strategy to better target angiogenesis and the GSC compartment [102].

Finally, survival of GBM after antiangiogenic treatment may also occur through hypoxia-induced autophagy, a catabolic pathway activated in response to nutrient deprivation, in which degraded macromolecules and cytoplasmic organelles are recycled to regenerate metabolites for energy and growth [103]. Exposure of primary and established GBM cells to hypoxia induced an autophagic response, while treatment of these cells with autophagic inhibitors resulted in increased apoptotic cell death [104]. Treatment of subcutaneous GBM xenografts with bevacizumab revealed increased hypoxia and expression of the autophagy mediator BNIP3 in conjunction with tumor progression, whereas the addition of the autophagy inhibitor chloroquine or silencing of ATG7, an essential mediator of autophagy, potentiated the antitumor effect of bevacizumab. Histologic analysis of bevacizumab-resistant human GBM samples showed regions of hypoxia colocalizing with higher expression of BNIP3, supporting a clinical role for autophagy in bevacizumab resistance [104]. Further work will be needed to validate this role, to better understand its mechanisms and develop specific inhibitors.

Conclusion & future perspective

The encouraging response rate of bevacizumab in GBM represents a milestone in the treatment of this aggressive disease, for which mean survival remains only slightly more than a year. The initial enthusiasm for bevacizumab in GBM therapy has, however, been tempered by modest improvements in overall survival, difficulties in radiographic assessment of response and the inability to effectively treat tumors after bevacizumab failure. Nevertheless, the apparent benefit of bevacizumab in many patients still holds promise and has sparked a flurry of studies aimed towards understanding mechanisms of evasive resistance and developing strategies to better utilize and extend anti-VEGF efficacy. The field is just beginning to recognize and uncover the myriad of mechanisms underlying evasive resistance in GBM and it remains a critical challenge to continue discovering these mechanisms and understand their clinical prevalence and importance.

Angiogenesis is a complex multistep process in GBM and the mechanisms of evasive resistance to antiangiogenic therapy appear to be no different. Fortunately, antiangiogenic evasion is distinct from classic drug resistance in that evasion implies adaptive changes allowing for tumor progression even while the specific angiogenic target (i.e., VEGF) remains inhibited, as opposed to the acquisition or selection of gene mutations in the intended drug target rendering a therapy ineffective. Therefore, there is a therapeutic window to overcome such evasive mechanisms, to improve and prolong anti-VEGF therapy. It is now clear that multiple mechanisms encompassing proangiogenic, invasive and survival modes of escape can be invoked to manifest evasive resistance in GBM (Figure 1) and it is likely that they occur simultaneously to some degree. The recent finding that multiple clonal subpopulations of GBM cells carrying distinct molecular genetic alterations can coexist within the same tumor underscores the challenge of intratumoral heterogeneity in GBM [105]. The ‘mosaicism’ of these tumors may therefore lead to the simultaneous induction of multiple different evasive pathways against antiangiogenic therapy in GBM. Perhaps the overriding resistance pathway and microenvironmental nuances unique to each tumor determine the overall recurrence pattern outcome.

Many of the obstacles that have been encountered in antiangiogenic therapy have been unforeseen; however, in hindsight they are not surprising and are consistent with our understanding of VEGF biology. For example, many proangiogenic pathways are known to be driven by hypoxia; however, it is now apparent that numerous hypoxia-independent mechanisms also exist to enable escape from angiogenesis inhibition, as described above. Moreover, the finding by multiple groups that VEGF signaling can occur on tumor cells was unexpected and has led to the discovery of several novel and unanticipated regulatory mechanisms that functionally impact tumor biology [41,102].

Several other unconventional mechanisms for GBM neovascularization have been recently described and, although they have yet to be clinically confirmed at this point, it is conceivable that such mechanisms may also contribute to antiangiogenic resistance. A number of studies have suggested that tumors have the ability to undergo vasculogenic mimicry, a process in which tumor cells form their own mural-like, functional vascular channels independent of endothelial angiogenesis. In GBM cells, vasculogenic mimicry was found to require VEGFR2 activity but not VEGF [106]. Silencing or pharmacologic inhibition of VEGFR2 impaired vasculogenic mimicry and tumor growth in vitro and in vivo, therefore indicating a potential escape mechanism to antiangiogenic treatment. Other studies have suggested that the CD133⁺ population of GSCs in GBM is specifically capable of undergoing vasculogenic mimicry [107,108], thereby providing another mechanism through which GSCs can induce antiangiogenic resistance.

Similarly, three groups have recently reported that a population of stem-like GBM cells has endothelial progenitor properties with the capacity to transdifferentiate into functional endothelial cells and that a significant proportion of the tumor vasculature is composed of cells of neoplastic origin [109–111]. In contrast to vasculogenic mimicry, in which tumor cells acquire limited endothelial features while retaining most of their tumor cell characteristics, during transdifferentiation tumor cells are thought to actively differentiate into endothelial cells and partake in the formation of neovasculature, thus representing a potential means to evade anti-VEGF therapy. Interestingly, while these tumor-derived endothelial cells shared several common endothelial markers, including CD31, CD34 and CD144, they all lacked VEGFR2 expression, the main signaling receptor for VEGF, which is an essential component in endothelial cells. Therefore, the extent to which these cells reflect functional endothelial cells has been debated. The clinical significance and impact of this transdifferentiation phenotype remains to be determined. Although some studies have emerged arguing that incorporation of GBM cells into tumor vasculature is a rare event and of limited therapeutic consequence [112], one cannot exclude the possibility that such a mechanism might become more prevalent in treated tumors to escape anti-VEGF therapy.

Because salvage treatments after bevacizumab failure appear to be largely ineffective, irrespective of whether the recurrences are proangiogenic or proinvasive, several critical challenges will need to be addressed and met in the near future in order to better optimize bevacizumab therapy for GBM patients.

As mentioned above, we must first better characterize the multitude of mechanisms by which tumors adapt to manifest in the various recurrence patterns following antiangiogenic therapy. By studying and understanding these evasive mechanisms, we have the opportunity to better utilize anti-VEGF therapy and extend its therapeutic benefit. It will be imperative to follow up with the development of novel strategies to block such resistance mechanisms. Combined treatment modalities that integrate angiogenesis inhibition with drugs that prevent or minimize evasive resistance and invasion can then be employed to enable enduring efficacy.

Second, it will be advantageous to predict the pattern with which patients will respond and relapse to anti-VEGF therapies, particularly considering the highly heterogeneous nature of GBM. The ability to distinguish patients who will respond and predict the pattern with which they might relapse may afford opportunities for pre-emptive combined therapeutic intervention to specifically block proangiogenic or proinvasive evasion pathways upfront in the right patients. One potential gateway to stratifying patients for antiangiogenic therapies is the work derived from The Cancer Genome Atlas [113] and the Repository of Molecular Brain Neoplasia Data [114], which has resulted in the identification of three or four transcriptional subtypes of GBM [86,115]. Because these classification schemes represent tumors with distinct molecular characteristics, it will be very interesting to ask whether different tumor subtypes and their associated perturbed signaling pathways can be linked to distinct patterns or mechanisms of antiangiogenic evasion. Moreover, tumors with a specific molecular signature may even tend to change from one subtype to another following angiogenic inhibition and resistance. For example, among the group classifications of GBM is a mesenchymal subtype that has been associated with reduced survival and aggressive behavior [115]. Are there certain tumors that begin as one subtype but then relapse with mesenchymal characteristics after antiangiogenic treatment and does this reflect a proinvasive recurrence pattern? It remains to be seen whether GBM subtype classifications can help determine which groups benefit the most from angiogenic inhibition and whether certain subtypes tend to recur in one fashion or another following evasive resistance.

Third, reliable biomarkers for bevacizumab activity are strongly needed for monitoring response in real time, as well as informing whether or not an antiangiogenic effect and tumor stasis are being maintained, or when evasive mechanisms have been engaged during impending antiangiogenic failure. Because some patients may benefit more than others from antiangiogenic therapy, biomarkers can first help select for responsive patients. Given the relatively high cost of antiangiogenic agents, such as bevacizumab, there are economic implications around the ability to predict and identify patients who will respond. Furthermore, the identification of such biomarkers will be critical to the timely recognition of evasive resistance, thus allowing for the opportunity to begin blocking compensatory pathways before tumors fully adapt and recur. Currently, no validated biological markers exist to indicate response or resistance to antiangiogenic therapies in cancer; however, a number of preclinical animal studies and clinical trials have suggested potential physiologic, circulating, intratumoral and imaging biomarkers that are currently under clinical investigation [116–118]. For example, it was recently found that GBM patients who exhibited improved tumor blood perfusion after treatment with cediranib demonstrated longer survival [119]. Alternatively, a study on rectal cancer reported that baseline levels of soluble VEGFR1 in the plasma were correlated with poor response to bevacizumab treatment, therefore suggesting its potential as a biomarker of intrinsic resistance [120]. Many of the biomarkers for evasive resistance in GBM may simply be the very factors that have been identified to promote and enable evasion pathways in preclinical models [60]; however, significantly more work needs to be done to validate such candidates at the clinical level. Therefore, it is important for future clinical studies and trials with antiangiogenic agents to incorporate biomarker programs into their study protocols to further this effort.

The evasive mechanisms described in this review by no means represent a complete list of factors that elicit escape from antiangiogenic therapy in GBM and we anticipate that our knowledge and understanding of such resistance mechanisms will be continually evolving. As pointed out above, there is clearly a variety of other angiogenic targets besides VEGF in GBM and other tumors that might reveal better efficacy in combination with other treatment modalities. Nevertheless, despite the limitations of angiogenesis inhibitors in GBM, drugs such as bevacizumab still represent an incremental milestone in the treatment of this devastating disease. The challenge moving forward will be to overcome these limitations and expand the benefits of antiangiogenic therapy. Various clinical trials in GBM combining angiogenesis inhibitors with drugs targeting evasive resistance pathways such as c-Met are ongoing or planned in the near future and will provide important information about the beneficial efficacy of such combinatorial strategies in GBM patients.