Proinvasive consequences of antiangiogenic therapy had been described previously as a collateral to the antitumor response in glioblastoma multiforme and were attributed to vessel co-option by cancer cells, but with uncertain significance for therapeutic efficacy (Kunkel et al., 2001
; Rubenstein et al., 2000
). Recently, heightened invasion has been further implicated as a response to antiangiogenic therapy in another model of orthotopic GBM (Gomez-Manzano et al., 2008
), consistent with these previous studies and our observations. More recently, we reported that abrogation of VEGF signaling with a function-blocking antibody against VEGFR2 elicits, after a period of transitory response, a bimodal resistance reaction with evident revascularization and increased invasiveness; while the revascularization was demonstrably important in the development of resistance (Casanovas et al., 2005
), the significance of the heightened invasiveness was unclear. Here we demonstrate in two distinct engineered mouse models of cancer (PNET and GBM) that therapeutically efficacious antiangiogenic therapy can elicit an adaptive-evasive response involving an augmented invasive phenotype and, in some cases, increased dissemination and the emergence of distant metastasis ().
Adaptive-Evasive Responses by Tumors to Antiangiogenic Therapies
Interestingly, PNET development in the context of genetic ablation of the Vegf-A gene produced small pancreatic lesions with increased invasiveness but without evident metastasis (data not shown), in contrast to the case of VEGF pathway inhibition. We suspect that the metastasis response is somehow enabled by the initial presence of an angiogenic vasculature in established solid tumors containing hyperproliferative cancer cells that, when disrupted, fosters metastatic dissemination. In contrast, the small avascular lesions populated by VEGF-KO cancer cells may have failed to acquire other necessary capabilities for productive dissemination. Regardless, the preclinical trials in the PNET model unambiguously demonstrate the enhancement of invasion and metastasis consequent to pharmacological angiogenesis inhibition.
In the PNET model, the multitargeted inhibitor sunitinib consistently demonstrated significantly better efficacy than the VEGFR2-specific antibody, in regard to both tumor shrinkage and augmented survival benefit. But in turn, sunitinib evidently elicited a more highly invasive adaptation, with some of the most aggressively invasive tumors observed heretofore in this model. Thus, the more effective the VEGF/angiogenesis inhibition, the more pronounced the adaptive progression to heightened and altered invasiveness. Interestingly, while both angiogenesis inhibitors enhanced liver metastasis, sunitinib did not increase lymphatic metastasis, in contrast to the anti-VEGFR2 antibody. This difference could be due to the differential specificity of inhibition by the two drugs: sunitinib potently blocks not only VEGFR2 and the PDGFRs but also lymphatic vessel-related VEGFR3 (Faivre et al., 2007
; Roskoski, 2007
). An attractive hypothesis is that the specific blockade of VEGFR3 signaling by sunitinib serves to alter the structure or permeability of lymphatic vessels or lymph nodes, thereby impeding formation of LN metastasis.
An intriguing paradox involving these results is raised by previous reports in other model systems, in which VEGF-driven angiogenesis was found to positively correlate with increased tumor invasiveness (Skobe et al., 1997
) or metastasis (Warren et al., 1995
). In contrast, our results as well as those presented in the accompanying study by Ebos et al. (2009)
in this issue of Cancer Cell
support the opposite conclusion, raising the possibility that both induction and suppression of tumor angiogenesis can exert proinvasive/prometastatic effects. We envision that the two effects, while related, may have distinct physiological bases. For example, tumors that express very high levels of VEGF (as in Skobe et al. and Warren et al.) may produce a tumor vasculature with excessive sprouting, poor pericyte coverage, and collectively impaired vascular integrity. Notably, Xian et al. (2006)
demonstrated that vessels with poor pericyte coverage are prone to increased metastasis and that overexpression of VEGF can drive intense angiogenesis with low pericyte coverage and a hyperpermeable vasculature, which can be more permissive for tumor cell intravasation and dissemination. On the other hand, antiangiogenic therapy demonstrably disturbs the tumor vasculature and in some cases (e.g., with sunitinib) also disrupts pericyte coverage, and may in addition be indirectly affecting the tumor cells by inducing a more invasive phenotype in response to hypoxia, leading to increased intravasation and metastatic dissemination.
How might tumors become more invasive during antiangiogenic therapy? One possibility is that tumors may elevate the activity of a preexisting invasion program that was not previously the driving force of expansive tumor growth, given the capability for angiogenesis. Alternatively, some tumors may switch on an invasive growth program distinct from that arising spontaneously during unperturbed tumor development and progression, as is evidently the case for GBMs, in which antiangiogenic therapy induced a phenotypic change from single-cell infiltration to migration of cell clusters along normal blood vessels. These observations suggest that the invasive growth program induced in response to therapy may be qualitatively different than the pathway used in normal tumor progression.
In this regard, genetic studies have implicated the hypoxia/HIF-1α pathway as an instigator of invasion and metastasis (Du et al., 2008
; Pennacchietti et al., 2003
). Interestingly in the GBM model, hypoxia via consequent HIF-1 activation induced angiogenesis and more infiltrative tumor cell behavior, while inhibition of HIF-1 and consequent failure to respond to hypoxia led to a blockade of angiogenesis and an exacerbated proinvasive phenotype. Due to the degree of invasion and the close proximity of invading tumor cells to blood vessels, these tumors are less hypoxic (Blouw et al., 2003
; Du et al., 2008
), despite the implication that hypoxia in the primary tumor drove their switch to a hyperinvasive condition. In concordance with the GBM results, we document herein the concomitant triggering of hypoxia with increased invasion in response to antiangiogenic therapy in the PNET model. While the concurrent initiation of hypoxia and invasion could be a circumstantial coincidence, a body of published evidence linking hypoxia and invasion (Cairns et al., 2001
; Pennacchietti et al., 2003
; Young and Hill, 1990
) suggests that the hypoxia response system could be involved in regulating invasion in this tumor type as well. Additionally, other therapy-induced mechanisms could contribute to the phenotypic progression of malignancy, such as activation or upregulation of extracellular proteases (e.g., matrix metalloproteinases and cathepsins), latent signaling circuits (e.g., c-Met), or triggering of epithelial-to-mesenchymal transition programs, each of which can demonstrably lead to more invasive and/or metastatic phenotypes and has been linked to the hypoxia response in particular model systems (Higgins et al., 2007
; Munoz-Najar et al., 2006
; Pennacchietti et al., 2003
). The possible contributions of the hypoxia response system and other factors to adaptation and emergent resistance to angiogenesis inhibitors remain to be determined and are an important topic for future research.
It must be emphasized that the three VEGF pathway inhibitors currently approved for certain cancer types (variously including colorectal, renal, breast, and hepatocellular), and likely others (both new drugs and new indications) to follow, are markedly altering the standard of care for these cancers. Despite their transitory efficacy, angiogenesis inhibitors present important therapeutic options that constitute improvements in terms of both clinical benefit and reduced toxicity compared to conventional agents used to treat these diseases. Recognizing both their benefits and their limitations, a pertinent question at hand is that of the broader applicability of our findings to antiangiogenic therapies aimed at other organ-specific cancers, and in particular their extrapolation to the outcomes of antiangiogenic therapies in cancer patients. Notably, in the accompanying paper by Ebos et al. (2009)
, one of these approved angiogenesis inhibitors, sunitinib, is shown to facilitate metastatic dissemination of both human breast cancer cells and syngeneic melanoma in mice following either orthotopic or intravenous inoculation; a second approved drug, sorafenib, evidently has similar effects. Even brief week-long antiangiogenic treatment elicited increased metastasis, much as we observed following a similar week-long treatment with the anti-VEGFR2 antibody in the PNET model. Collectively, the two studies document increased invasion and/or metastasis in four distinct mouse models of organ-specific cancer and begin to establish a mechanistic pattern of increased invasion and metastasis as an adaptive response to antiangiogenic therapy, one that seems likely to be operative in certain human tumors.
In regard to the relevance of our findings in mouse models of neuroendocrine tumors and glioblastoma to antiangiogenic therapies for the cognate human cancers, there are currently insufficient data about the responses of human pancreatic neuroendocrine tumors due to both their rarity and the limited number of patients that have been treated in clinical trials with angiogenesis inhibitors. In the case of glioblastoma, there is growing experience with antiangiogenic therapies, specifically the VEGF ligand-trapping antibody bevacizumab (Fischer et al., 2008
; Narayana et al., 2009
; Norden et al., 2008
) and the VEGFR1/2/3 kinase inhibitor AZD2171 (Batchelor et al., 2007
). In both cases, there seems to be a proinvasive adaptation to antiangiogenic therapy, as suggested by magnetic resonance imaging in a subset of GBM patients that developed multifocal recurrence of tumors during the course of therapy with anti-VEGF (bevacizumab) (Fischer et al., 2008
; Narayana et al., 2009
; Norden et al., 2008
) or AZD2171 (Batchelor et al., 2007
). Much as in the mouse GBM model, we infer that the human glioblastoma cells could be co-opting blood vessels by invading along them into the surrounding normal brain tissue, thereby achieving vascular/nutrient/oxygen sufficiency while consequently promoting an invasive dispersion phenotype (perivascular tumor invasion). Thus, there is reason to predict clinical relevance of our collective observations.
How then could the improvement of survival in many clinical trials in spite of increased invasiveness and metastasis be explained? The increased progression-free survival (PFS) could likely reflect the impairment of primary tumor growth, which, while initially efficacious, is short lived, with the onset of multiple forms of evasive resistance that would inevitably produce a return to progressive disease. This tumor relapse would compromise this initial benefit such that the prolongation of PFS and even overall survival (OS) would not be particularly robust, as is the case in many clinical trials. For example, recent results from two randomized clinical trials with bevacizumab and taxanes for the treatment of metastatic breast cancer (the E2100 study and the AVADO study) demonstrated small benefits in PFS but not an OS benefit, and our results provide a possible explanation for this disconnect.
Indeed, this undesirable therapy-triggered change in the natural history of the treated tumors is convergent with emerging examples involving other targeted therapy-driven changes, such as the case of trastuzumab (Herceptin) therapy in HER2-positive breast cancer patients. Although effective in its antitumor and prosurvival effects (13 months increased life span), trastuzumab therapy changes the course of disease progression in these patients to one with a markedly increased incidence of brain metastasis (Bendell et al., 2003
; Clayton et al., 2004
; Miller et al., 2003
; Weil et al., 2005
), which is otherwise rare in untreated patients. The companion study by Ebos et al. (2009)
documents increased metastasis in multiple organs, including the brain, in response to the angiogenesis inhibitor sunitinib in a mouse model of breast cancer, illustrating that convergence.
Another clinically relevant question raised by our results is whether concomitant chemotherapy in some tumor types (e.g., in all currently approved indications for bevacizumab) could alter or even abrogate this increased malignant phenotype. Both preclinical and clinical trials investigating the effects of standard-of-care and more recent (e.g., metronomic) chemotherapy regimens on proinvasive and prometastatic responses to anti-angiogenic therapies are warranted.
Certainly, the characteristics and consequences of the exacerbated invasive-metastatic phenotype that develops in response to treatment with angiogenesis inhibitors motivate critical assessment in the clinical setting. Moreover, further preclinical studies are warranted to elucidate the mechanisms of this adaptive-evasive resistance, so as to design and test the potential of mechanism-based combination therapies aimed at impeding this insidious consequence of singular antiangiogenic therapy.