Tumour progression requires the activation of a tumour and stromal cell-driven angiogenic programme, and the targeting of this process demonstrates an impact on tumour growth and progression. The results of preclinical studies have demonstrated a proinvasive/metastatic effect of antiangiogenic treatments with recent evidence supporting a contribution of the stroma to tumour aggressiveness and the short-term effects of antivascular endothelial growth factor therapy. Furthermore, hypoxia-dependent and -independent factors are considered as driving forces for tumour cell escape by altering both the tumour cells themselves and the stroma. This tumour–stromal cell alliance should be taken into consideration for the development of innovative therapeutic options targeting both tumour components to improve clinical benefits for cancer patients.
Growing tumours require increasing amounts of oxygen and nutrients to sustain the deregulated proliferation of tumour cells; therefore, switching on the angiogenic programme is a limiting step during neoplastic progression. Vascular endothelial growth factor (VEGF) and its receptors (VEGFRs) are essential for tumour angiogenesis  and recently VEGF-signalling inhibitors have been introduced into clinical protocols for the treatment of several types of tumours . Nevertheless, transitory effects of these agents and the promotion of tumour aggressiveness have been reported [3-6]. Mounting evidence supports the notion that both neoplastic cells and tumour stromal cells within the tumour contribute to the formation and the characteristics of the tumour mass [7, 8]. In this review, we will consider the mechanisms involved in tumour malignization following antiangiogenic therapy, and the contribution of tumour stromal cells in response to these agents. Next, we will present new therapeutic options to counteract the undesirable side effects that emerge after anti-VEGF therapy. Finally, we will discuss the possible underlying mechanisms of the contradictory results obtained after treatment of cancer patients with VEGF-signalling inhibitors.
Interaction between angiogenesis and metastasis
Angiogenesis, whilst providing vascular support to the developing mass, not only influences the growth of the primary tumour but is inevitably associated with the systemic dissemination of tumour cells and the consequent development of distant metastases. The tumour vasculature enables tumour cell invasion and dissemination to occur via several mechanisms. The high levels of VEGF in the tumour microenvironment and the consequent activation of its signalling pathway stimulates the growth and migration of blood and lymphatic endothelial cells, thus providing large vascular areas for tumour cell intravasation. Furthermore, VEGF can facilitate tumour cell intravasation by increasing vascular permeability. Of importance, abnormalities of the tumour vasculature are not only dependent on endothelial cells but also on other stromal cells such as pericytes that are also affected by factors secreted by the tumour cells. Tumour pericytes have a loose association with the endothelial cells and this could affect the survival of endothelial cells as well as contribute to the presence of intercellular gaps or openings that provide relatively easy access for tumour cells to intravasate [9, 10]. The increased vascular leakiness causes an increase in interstitial fluid pressure, which promotes ‘passive’ escape of tumour cells through the stroma . In tumour endothelial cells, VEGF upregulates the secretion of proteases that degrade the basement membrane, and increases expression of adhesion molecules that mediate interaction with tumour cells . As described by Kaplan et al., VEGF can also promote the formation of the premetastatic niche by mobilizing VEGFR1-positive bone marrow-derived cells (BMDCs) and sustaining the growth of micrometastases to macrometastases by stimulating the vascularization of early metastatic disease through the incorporation of VEGFR2-positive endothelial cell progenitors . In addition, VEGF can directly affect tumour cell survival, invasion and migration through an autocrine vascular-independent process .
Together, these observations suggest that the inhibition of VEGF pathways should not only reduce tumour growth but also impede the formation of metastases. However, the benefits of antiangiogenic agents for the treatment of cancer are controversial; increased formation of metastases was recently demonstrated with these therapies [4-6], and progression-free survival (PFS) benefits are often not translated into long-term overall survival (OS) gain. One well-accepted explanation for this apparent contradiction is that tumours can develop resistance to anti-VEGF therapy. Indeed, a complete lack of effect in refractory tumours or tumour rebound after an initial effect have been described . Thus, the failure of these treatments in the long term suggests that there may be a connection between stromal and cancer cells. It is clearly possible that the modified microenvironment provides a greater contribution to resistance following antiangiogenic therapy, where the microenvironment is the primary target, compared with cytotoxic therapy, where the main drive towards resistance is provided by the tumour cells per se . Each of these two compartments can contribute to the intrinsic or acquired resistance to anti-VEGF therapy through production of proangiogenic factors other than VEGF, thus sustaining the formation of new blood vessels [12-15]. Other stromal alterations in the production of such factors in the stroma inevitably modify the behaviour of tumour cells leading to acquired resistance and thus the initial antitumour effects can be overcome. An example of this is cancer-associated fibroblasts (CAFs) which limit the benefits of VEGF-signalling inhibition by producing Platelet-derived growth factor C (PDGF-C) [14, 16]. Alternatively, tumours can be refractory to anti-VEGF therapies because they do not have sprouting angiogenesis driven by VEGF. Other mechanisms of tumour vascularization have been described, including co-option of existing normal vessels of the tissue or vascular mimicry where neoplastic cells can directly form vessel walls .
Alliance between tumour and stromal cells
Cells in the tumour microenvironment are normally able to manage with deficient oxygen and nutrient supplies relative to the high demand of the growing mass, due to the characteristics of the tumour vasculature such as disorganized endothelial and pericyte layer, chaotic vessel network and dilatations. Nevertheless, blocking the formation of new blood vessels in the tumour mass or destabilizing the existing vessels, with VEGF/VEGFR inhibitory therapies, produce an even greater decrease in nutrient and oxygen levels within the tumour . One of the crucial steps in the cell response to hypoxia is the stabilization of hypoxia-inducible factor (HIF)-1 in low-oxygen conditions and the consequent upregulation of genes directly controlled by this transcription factor. As HIF-1 modulates the transcription of genes involved in glycolytic metabolism, oxygen consumption, survival, angiogenesis, migration and invasion, its stabilization has a dramatic impact on the gene expression profile and ultimately on the behaviour of the cells [16, 19]. It is interesting that HIF-1 is deregulated by mutation of several genes, such as von Hippel–Lindau (VHL) and PTEN , as well as by other hypoxia-independent mechanisms. Thus, HIF-1 can accumulate in circulating tumour cells that are exposed to normoxia in the circulation . Tumour cells react to the hypoxia caused by antiangiogenic agents by reprogramming their metabolism and thus increasing the uptake of glucose to sustain energy production through glycolysis  (Table 1). Moreover, they start to produce alternative compensatory proangiogenic factors for the formation of tumour blood vessels such as FGFs, ephrins and angiopoietins which are upregulated after VEGF-pathway inhibition [3, 22]. Furthermore, hypoxia promotes modification of the cellular composition and signalling cues of the tumour stroma. Typically, both tumour and stromal cells are stimulated to produce cytokines (e.g. Stromal cell-derivated factor-1 (SDF-1), Granulocyte-colony stimulating factor (G-SCF) and interleukin-6) that recruit BMDCs such as Gr1+CD11b+ myeloid suppressor-type cells, Endothelial TEK Tyrosine Kinase dene (TIE2)-expressing monocytes and tumour-associated macrophages (TAMs), which will actively contribute to the production of proangiogenic factors [3, 14, 16].
Table 1. Mechanisms of tumour response to hypoxia induced by antiangiogenic therapy
Adaptation: Tolerance of hypoxic conditions and Angiogenesis
Neoplastic cells can become tolerant to hypoxia and modify their features to grow in poorly oxygenated areas. In addition, several observations suggest that they can actively escape from oxygen-deprived regions. Indeed, whilst the connection between antiangiogenic therapy and the invasive/metastatic phenotype needs further validation, the evidence linking hypoxia to a more aggressive metastatic behaviour of cells is well established . Different tumour-dependent mechanisms, mainly driven by HIF-1, could be involved in this reaction, including the production of prometastatic proteins, the secretion of proteolytic enzymes or alteration of the adhesion molecule pattern (Table 1). Indeed, HIF-1 has a role in several steps of the metastatic programme controlling the production of proteins involved in cell contact with other cells or with components of the extracellular matrix (ECM). Thus, HIF-1 deregulation has been associated with the malignant epithelial–mesenchymal transition programme through downregulation of E-cadherin at adherens junctions and with disruption of the basement membrane through upregulation of matrix metalloproteinases (MMPs) or urokinase-type plasminogen activator (uPAR)  (Fig. 1). Amongst the genes directly regulated by HIF-1, c-Met is undoubtedly involved in the invasive and metastatic behaviour of tumour cells after exposure to hypoxia. The receptor c-Met is typically expressed on tumour cells and binding of the receptor to hepatocyte growth factor (HGF; also known as scatter factor), produced by mesenchymal cells, activates the signalling cascade, promoting tumour growth and invasive/metastatic behaviour [24, 25] (Fig. 1). Recent studies have demonstrated that the activation of c-MET signalling is tightly implicated in the increase in invasion and metastasis observed after anti-VEGF therapy. In addition, Lu et al. recently described a proinvasive effect of VEGF-signalling inhibition directly related to the expression of VEGFR2 in glioblastoma multiforme (GBM) cells that was not associated with the typical antiangiogenic effect on endothelial cells. In particular, VEGFR2 forms a complex with c-MET and the stimulation of VEGFR2 by VEGF results in inactivation of c-MET-associated proinvasive signalling. Consequently, the inhibition of VEGF/VEGFR2 restores c-MET-mediated invasiveness  (Fig. 1). Thus, the concomitant inhibition of c-MET can reduce the invasive and metastatic capabilities promoted by VEGF-pathway inhibition [6, 27]. In addition, hypoxic tumour cells can directly increase vascular co-option and mimicry thus promoting dissemination  (Fig. 1).
The c-MET-mediated tumour escape from hypoxia highlights the crucial interactions between tumour cell and stroma in the response to low levels of oxygenation. The hypoxic tumour microenvironment is rich in HGF either due to the elevated levels produced by fibroblasts under hypoxia  or due to the hypoxia-induced activation of proteases converting pro-HGF into the mature HGF . Both tumour cells and normal cells are motile and can move into the tissue activating the same molecular mechanisms or changing shape, generating force and remodelling the ECM, alone or within a group of cells . The difference is that neoplastic escape lacks the physiological ‘stop signals’ of normal cells, immobilizing and anchoring the cells through for example integrin-mediated mechanisms . Furthermore, tumour cells do not act alone to escape from hypoxia, and examples of stromal cells that directly assist the neoplastic cells in the development of metastases have been described. Therefore, although nonmetastatic tumour cells move randomly in the stroma, their metastatic counterparts migrate in a directional polarized manner. In the mesenchymal/amoeboid single cell movement, the polarity is regulated by the presence of chemokines in the stroma and by the interaction of the cell cytoskeleton with components of the ECM. In the case of collective invasion, movement is lead by one or several tumour cells  and it is interesting that tumour cell ‘leaders’ acquire the characteristics of mesenchymal cells [32, 33]. It has also been suggested that stromal cells may directly aid tumour cells movement. For example, stromal fibroblasts or myeloid cells migrate at the leading-edge of the tumour cell strand thus directing tumour migration  (Fig. 1). In addition, CAFs assist tumour cells at several steps during metastatic dissemination by the production of chemo-attractive factors such as SDF-1 and HGF or proteases. Moreover, CAFs can promote the growth of metastases by providing an adequate microenvironment and by stimulating invasive growth in several types of tumours . Other stromal cells involved in assisting the movement of the tumour cell are the inflammatory cell repertoire, directly attracted to the tumour microenvironment as a result of tissue and endothelial injury or alterations. This is likely to occur during chemotherapy or radiotherapy (both of which cause a high degree of tissue damage), but could also represent a feature of anti-VEGF/VEGFR antiangiogenic therapies  (Fig. 1). In particular, an important immune mediator of increased metastatic capabilities after therapy is represented by TAMs. Attracted to the hypoxic areas by CSF-1 secreted by the tumour cells, they can stimulate invasiveness of epidermal growth factor (EGF) receptor (EGFR)-expressing cells by the production of EGF. Moreover, TAMs contribute to the alteration of blood vessel permeability to the tumour cells through the production of proangiogenic factors and the consequent formation of more highly fenestrated vessel walls . Furthermore, other BMDCs such as the Gr1+CD11b+ myeloid suppressor-type cells infiltrate the tumour microenvironment in response to inflammation and may have a role in promoting metastasization [35, 36].
Prevention of tumour invasiveness
The possibility of restoring tumour oxygenation represents an appealing option to prevent tumour invasiveness as hypoxia is one of the drivers of neoplastic metastasization (Table 2). As postulated by Jain, a direct effect of VEGF-signalling inhibition would be a ‘window of normalization’ with a more organized vascular structure and better distribution of oxygen associated with a decrease in intratumoral oedema . However, this suggested period of normalization is temporary and the duration is not well defined . Mazzone et al.  demonstrated that controlled normalization of tumour oxygenation can be obtained by modulating the endothelial cell response to hypoxia. In particular, they demonstrated that the endothelial cells with reduced activity of the oxygen sensor prolyl hydroxylase domain protein 2 (PHD2) form vessels with normal regular lining and less leakiness, thus restoring the oxygen supply to the tumour. This normal-like oxygenation does not prevent neoplastic progression, but tumours generated with a PHD2+/− background are better perfused, less invasive and develop fewer metastases than the same tumours with wild-type vessels [38, 39]. From studies in a transgenic model of pancreatic neuroendocrine tumours, Maione and colleagues arrived at the same conclusion that restoration of the oxygen level in the tumour through the normalization of endothelial cells impairs neoplastic invasion and dissemination. They proposed that semaphorin-3A (Sema3A) should be considered as a therapeutic agent due to a physiological role in the inhibition of angiogenesis and the ability to inhibit the invasive/metastatic behaviour promoted by VEGF-signalling inhibitors [40, 41]. Moreover, as a ‘normalizer’ of the tumour vasculature, Sema3A could represent a good candidate for the prolongation of the window of normalization. Nevertheless, other evidence suggests that intermediate levels of oxygen could increase the metastatic capability of the cells . In addition, exposure of cells to a chronic low level of hypoxia represents a recognized experimental method to induce adaptation to growth in hypoxic conditions and subsequent selection of resistant clones. Therefore, the levels of normalization of the tumour vasculature and of reoxygenation should be carefully evaluated to avoid undesirable side effects.
Table 2. Vascular-targeted approaches with antimetastatic effects
Because antiangiogenic therapy affects oxygen delivery, hypoxia and HIF activation are the main causes of tumour invasiveness. But is therapy-induced malignization driven only by hypoxia? Some evidence suggests that other mechanisms, alone or in addition to hypoxia, could be involved in the malignant reaction to these therapies . Preclinical data are consistent with the notion that the same treatment could have different and often contrasting effects depending on the tumour cell type. As described by Shojaei and colleagues, sunitinib increases metastasis in orthotopic mouse models of breast and colon cancer, whereas it does not promote metastatic behaviour in lung cancer, suggesting a role of intrinsic tumour cell characteristics in response to therapy . When considering the tumour as an organ, the microenvironment has a crucial influence on the response of tumour cells to therapy. Indeed, astrocytomas and GBM tumours can become more or less invasive depending on the site of injection in mice upon genetic inhibition of HIF-1 and VEGF. The original vascularization of normal host tissue could affect the invasive behaviour of tumour cells with or without the ability to co-opt the pre-existing vessels after inhibition of angiogenesis . It is interesting that other antiangiogenic molecules that do not influence the VEGF pathway can exert different effects on stromal cells and tumour cells. For example, depending on the molecules recruited by its receptor PlexinD1 complex, Sema3E is proapoptotic in endothelial cells but proinvasive and prometastatic in tumour cells [44, 45].
The impact of therapy on different cellular compartments could vary depending on the mode of action of the agents used for VEGF-pathway inhibition, e.g. ligand binding with bevacizumab or blockade of intracellular signalling pathways with small-molecule tyrosine kinase inhibitors (TKIs). It has been proposed that TKIs targeting pericytes could modify the structure of blood vessels producing a leakier barrier, thus allowing for enhanced tumour cell intra- and extravasation [46, 47]. On the other hand, several studies have demonstrated the increased efficacy of targeting of both pericytes and endothelial cells as endothelial cells become more sensitive to apoptosis in the absence of pericyte coverage [48-50]. Therefore, further investigation of the role of hypoxia and endothelial cell death in the tumour response to anti-VEGF therapy or tumour growth inhibition is warranted before new clinical protocols can be applied.
The results of preclinical studies are often contradictory with regard to the efficacy of anti-VEGF therapy; similarly those of clinical trials also vary depending on cancer type and the specific antiangiogenic therapy. Phase III studies have shown the benefits of VEGF-targeted therapies, including bevacizumab and sunitinib, either as single agents or in combination with chemotherapy. Nevertheless there is general agreement that antiangiogenic treatments are more effective in terms of increasing PFS than prolonging OS. Based on clear clinical benefit, although in the absence of a robust statistically significant increase in OS, VEGF-pathway inhibitors are currently the mainstay of therapy for renal cell carcinoma (RCC) [51, 52]. Because of this discrepancy between PFS and OS, and because antiangiogenic therapies typically exert their effects through increased tumour necrosis, how best to measure the clinical benefit of treatment remains controversial,. As reported, anti-VEGF agents induce cavitation and loss of viable tumour mass which could translate into an impact on tumour growth without significant alteration of tumour size measurements as recommended by RECIST (Response Evaluation Criteria in Solid Tumours) [13, 53]. The short-term effects of anti-VEGF therapies further indicate that the lack of OS prolongation might be due to at least in part to tumour malignization.
The pattern of growth and the modifications induced at the site of tumour development are greatly influenced by the tumour type and in particular by its angiogenic features and proangiogenic capacity due to specific tumour–stroma interaction. This is an important observation, as single-agent VEGF-targeted therapy is efficacious in certain cancers, such as RCC and hepatocellular carcinoma (HCC), but shows considerably less clinical benefit in others, for example colorectal cancer (CRC), for which this therapy is administered in combination with chemotherapy. In RCC, angiogenesis is likely to be highly VEGF dependent, in part due to a high rate of inactivation of the VHL tumour suppressor . The same dependence on angiogenesis is thought to be important for the efficacy of antiangiogenic therapy in HCC; these tumours are highly angiogenic when growing in liver and displace the normal parenchyma. In contrast, the metastatic foci of CRC, which typically grow in the liver, often replace rather than displace the liver parenchyma, through FAS ligand-induced death in the hepatocytes. This leads to co-option of existing blood vessels instead of dependence on sprouting angiogenesis [13, 55].
Could the different angiogenic features of each type of tumour influence the escaping behaviour of tumour cells after antiangiogenic therapy? Does reduction of the oxygen and nutrient supplies due to vessel growth inhibition more efficiently mobilize the angiogenic-dependent tumour cells? Of note, the development of astrocytomas, which are highly oxygen dependent, is associated with a change in the manner in which the tumours acquire their blood supply. Thus, low-grade astrocytomas grow by co-opting pre-existing normal brain vessels whereas an increased demand for oxygen and nutrients during progression from grade III to grade IV (i.e GBM tumours) activates the angiogenic programme . Recently, the US Food and Drug Administration (FDA) approved bevacizumab for the treatment of recurrent GBM based on several studies demonstrating efficacy in terms of increased PFS and OS in combination with conventional chemotherapy. However, tumour resistance occurs with new distant foci of progression or diffuse in situ infiltration with or without local tumour recurrence as shown by fluid-attenuated inversion recovery magnetic resonance imaging analysis [57, 58]. On the other hand, in metastatic RCC, several clinical trials confirmed the positive impact of antiangiogenic therapies in controlling the growth of this typically highly VEGF-secreting cancer [59-62]. Thus, with a remarkable increase in PFS and an acceptable prolongation of OS, sunitinib gained FDA approval for the treatment of advanced RCC [51, 52]. Unexpectedly, a retrospective analysis by Plimack and colleagues of RCC patients treated with antiangiogenic therapy compared with interferon immunotherapy alone demonstrated that those treated with antiangiogenic agents were more likely to progress with metastasis at new sites in the setting of controlled disease at the original site . Because of these contrasting and somewhat paradoxical effects on primary and metastatic tumour growth, a number of questions should be addressed. Is the potency of antiangiogenic therapy different in terms of controlling primary tumour growth compared with metastatic foci? Is it possible that metastatic spread could be a consequence of anti-VEGF treatment, as suggested by preclinical studies [4, 5]? In an effort to determine the pro-malignant effect of anti-VEGF therapy on tumour growth, Miles and colleagues conducted an interesting meta-analysis of the outcomes of different Phase III clinical trials in a variety of tumour types after discontinuation of combined treatment with bevacizumab as a result of adverse events. The authors concluded that the disease develops with the same pattern of progression in bevacizumab- and placebo-treated patients thus excluding an acceleration of tumour progression and an increase in mortality rates due to treatment. Nevertheless, this analysis excludes a number of clinical trials that reported somewhat discordant results [64, 65]. Therefore, it will be crucial to evaluate tumour aggressiveness in the future trials with antiangiogenic therapies.
Concluding remarks: disrupting the alliance
Far from the original belief that the characteristics of a cancer were dependent solely on tumour cells, a tumour mass is now thought as a heterogenous mixture of tumour and stromal cells in tight functional association. The interplay between tumour cells and the tumour microenvironment is crucial for the development of the neoplastic lesion which requires a combination of the two components to generate a vascularized growing mass. Indeed, the fundamental role played by the stroma as a co-director of the angiogenic programme is the rationale for cancer treatment with antiangiogenic therapies. It is interesting that the same tumour–stromal cell collaboration that leads to tumour growth is also involved in the tumour response to therapeutic inhibition of the VEGF pathway. Thus, together with tumour cell characteristics, the stroma contributes to the lack of efficacy of anti-VEGF therapy in terms of both intrinsic as well as acquired resistance events. Furthermore, many of the tumour cell-dependent mechanisms of resistance are implemented through modification of the stroma as the recruitment of infiltrating cells, such as CAFs and TAMs, or the production of alternative proangiogenic factors. One of the main modifications induced by antiangiogenic therapy in tumours is an increase in hypoxia and HIF-1 stabilization. Tumour cells can respond to hypoxia by modifying their metabolic characteristics to resist low levels of oxygenation. Alternatively, tumour cells can escape from hypoxic conditions alone or supported by their stromal neighbours. Furthermore, HIF-1 upregulation influences tumour cell-dependent invasive behaviour, but also mediates the interaction between tumour and stromal cells by the expression of receptors such as c-MET on neoplastic cells. In addition, intratumoral hypoxia directly modifies stromal cells which become active promoters of the metastatic dissemination of tumour cells.
The implication that the stroma has a role in the emergence of resistance to antiangiogenic therapies has important clinical implications. First, the stroma represents another relevant cause of the short-term effects of antiangiogenic treatment in cancer patients, thus removing the focus from tumour cells alone. Secondly, modulation of the prometastatic stroma provides innovative possibilities for the prevention of resistance to the effects of antiangiogenic therapy. Therefore, as tumour cells, stromal cells and their interactions initiate tumourigenesis, sustain neoplastic growth and promote metastasization and therapeutic resistance, these two cell types should be considered together in the development of new therapeutic approaches. We envision that eventually combinatorial or multitargeted therapies against several molecules in these different compartments could lead to more effective cancer treatment by disrupting the tumour–stroma alliance.
Conflict of interest statement
The authors have no conflicts of interest to declare.
The authors would like to thank Francesc Viñals and Mariona Graupera for critical reading of the manuscript and helpful suggestions. Research conducted in the authors' laboratory is supported by grants from European Union-Framework Programme 7 (EU-FP7) (ERC-StG-281830), MinEco-Spain (SAF2009-08375, RTICC-RD2006-0092) and AGAUR-Generalitat (SGR681).